WO2022135288A1 - 信息处理的方法和装置 - Google Patents
信息处理的方法和装置 Download PDFInfo
- Publication number
- WO2022135288A1 WO2022135288A1 PCT/CN2021/139077 CN2021139077W WO2022135288A1 WO 2022135288 A1 WO2022135288 A1 WO 2022135288A1 CN 2021139077 W CN2021139077 W CN 2021139077W WO 2022135288 A1 WO2022135288 A1 WO 2022135288A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- network
- sub
- information
- communication
- communication devices
- Prior art date
Links
- 230000010365 information processing Effects 0.000 title claims abstract description 34
- 238000003672 processing method Methods 0.000 title abstract description 12
- 238000004891 communication Methods 0.000 claims abstract description 616
- 238000000034 method Methods 0.000 claims abstract description 214
- 238000012545 processing Methods 0.000 claims abstract description 141
- 230000008569 process Effects 0.000 claims description 98
- 238000012549 training Methods 0.000 claims description 78
- 238000009826 distribution Methods 0.000 claims description 45
- 238000004590 computer program Methods 0.000 claims description 17
- 230000011664 signaling Effects 0.000 claims description 16
- 230000007613 environmental effect Effects 0.000 claims description 12
- 230000008447 perception Effects 0.000 claims description 9
- 230000006872 improvement Effects 0.000 abstract description 6
- 230000009467 reduction Effects 0.000 abstract description 5
- 238000013528 artificial neural network Methods 0.000 description 35
- 238000010586 diagram Methods 0.000 description 31
- 230000006870 function Effects 0.000 description 27
- 238000013461 design Methods 0.000 description 17
- 238000013473 artificial intelligence Methods 0.000 description 13
- 230000005540 biological transmission Effects 0.000 description 13
- 238000005516 engineering process Methods 0.000 description 12
- 238000000605 extraction Methods 0.000 description 9
- 230000007246 mechanism Effects 0.000 description 9
- 230000008859 change Effects 0.000 description 8
- 230000008878 coupling Effects 0.000 description 6
- 238000010168 coupling process Methods 0.000 description 6
- 238000005859 coupling reaction Methods 0.000 description 6
- 230000003993 interaction Effects 0.000 description 6
- 230000006835 compression Effects 0.000 description 5
- 238000007906 compression Methods 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 5
- 230000002829 reductive effect Effects 0.000 description 5
- 230000001360 synchronised effect Effects 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 4
- 239000000284 extract Substances 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000001514 detection method Methods 0.000 description 3
- 238000007726 management method Methods 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 230000000737 periodic effect Effects 0.000 description 3
- 230000000306 recurrent effect Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 230000009471 action Effects 0.000 description 2
- 210000004027 cell Anatomy 0.000 description 2
- 239000003795 chemical substances by application Substances 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 2
- 230000001934 delay Effects 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 238000010295 mobile communication Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000036961 partial effect Effects 0.000 description 2
- 230000002787 reinforcement Effects 0.000 description 2
- 230000001953 sensory effect Effects 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 230000010267 cellular communication Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 239000013256 coordination polymer Substances 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000006837 decompression Effects 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 210000000653 nervous system Anatomy 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
- 230000001960 triggered effect Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/02—Arrangements for optimising operational condition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5061—Partitioning or combining of resources
- G06F9/5077—Logical partitioning of resources; Management or configuration of virtualized resources
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
Definitions
- the present application relates to the field of wireless communication, in particular, to a wireless communication technology applying an intelligent network, and in particular, to a method and apparatus for information processing.
- AI technology covers implementation technologies such as machine learning and deep learning.
- the realization of AI technology is mainly to enable the computer to find information from the data, so as to learn some laws, and through model training and optimization, the model can have a deeper understanding of the data.
- DNNs deep neural networks
- MIMO multi-input multi-output
- the present application provides an information processing method and device, which can well balance performance and complexity, is more suitable for the special needs of wireless communication, and can improve the realizability of intelligent networks used in communication networks.
- an information processing method may include: acquiring information in a communication network; and processing the information with a target intelligent network, where the target intelligent network includes a plurality of sub-networks, and at least two of the sub-networks correspond to different types of features.
- the target intelligent network which can be called intelligent modules, models, “black boxes", can represent networks with machine learning capabilities, such as artificial intelligence networks.
- the method can be executed by a communication device, and can also be executed by a chip, a chip system or a circuit applied to the communication device.
- the network architecture of the hierarchical mechanism is designed by utilizing the characteristic grouping of the wireless communication network and the classification processing characteristics of the intelligent network.
- flexible training and use can be performed based on the characteristics corresponding to each sub-network. For example, for multiple objects, similar features among the multiple objects can be processed by using the same sub-network, thereby reducing the complexity and saving training overhead.
- different sub-networks can be used for processing, such as training and updating of each object or designing sub-networks corresponding to this type of features, so as to ensure the performance as much as possible. Therefore, based on this scheme, through feature layering, features at each level can be effectively grasped and extracted to ensure training performance, and training overhead can be minimized to improve training efficiency.
- an information processing method may include: acquiring information in a communication network; processing the information with a target intelligent network, where the target intelligent network includes multiple sub-networks, some of the sub-networks correspond to multiple objects, and some of the sub-networks correspond to one object.
- the method can be executed by a communication device, and can also be executed by a chip, a chip system or a circuit applied to the communication device.
- an "object” may represent a communication device (such as a terminal device or a network device), or may also represent a communication scene, or may also represent other objects.
- the sub-network corresponds to an object, which means that the sub-network can be used for the object, or the sub-network can participate in processing the information of the object, or the sub-network is used to process some features of the object.
- the meaning of "part of the sub-network” includes one or more sub-networks.
- a network architecture with a hierarchical mechanism is designed to ensure performance.
- the network overhead can be greatly reduced.
- the object's target intelligent network includes multiple sub-networks.
- a part of the sub-networks in the plurality of sub-networks can be used not only for the object, but also for one or more other objects, and another part of the sub-networks in the plurality of sub-networks is only used for the object.
- an information processing method may include: acquiring information in a communication network; processing the information with a target intelligent network, the target intelligent network comprising a first sub-network and a second sub-network, the first sub-network is used for processing common features in the information, the second Subnets are used to handle non-public features in information.
- the first sub-network is used to represent the sub-network in the target intelligent network for processing common features in the information. It should be understood that the number of the first sub-network is not limited to one, for example, the first sub-network may include one or more sub-networks.
- the second sub-network is used to represent the sub-network in the target intelligent network for processing non-common features in the information. It should be understood that the number of the second sub-network is not limited to one, for example, the second sub-network may include one or more sub-networks.
- the first sub-network is used to process common features in the information, that is, the first sub-network corresponds to common features, such as features corresponding to the public environment, and different objects (such as communication equipment or communication scenarios, etc.) With this first sub-network, the frequency of change does not need to be too high, so complexity and cost can be reduced.
- the second sub-network is used to process non-public features in the information, that is, the second sub-network corresponds to non-public features, and different objects (such as communication devices or communication scenarios) can be trained and updated separately for the second sub-network, that is, the second sub-network can be updated in a shorter Training updates are performed periodically, so performance is guaranteed as much as possible. In this way, performance and complexity can be effectively balanced, opening up a whole new space for performance improvement and complexity reduction.
- the first sub-network is used to process common features of one or more communication devices; or, the first sub-network is used to process one or more communication scenarios public features.
- the communication device may be a terminal device, or the communication device may also be a network device.
- multiple communication devices or multiple communication scenarios can share one first sub-network, thereby reducing complexity.
- the second sub-network is used to process non-public features of a communication device; or, the second sub-network is used to process non-public features in a communication scenario.
- each communication device or each communication scenario may have its own corresponding second sub-network, so that the second sub-network can be trained and updated according to the actual situation, so that the performance can be guaranteed as much as possible.
- an information processing method may include: acquiring information in a communication network; processing the information with a target intelligent network; wherein the target intelligent network includes a first sub-network and a second sub-network, the first sub-network corresponds to a plurality of communication devices, and the second sub-network The network corresponds to one communication device; or, the first sub-network corresponds to multiple communication scenarios, and the second sub-network corresponds to one communication scenario.
- the first sub-network is used to represent a sub-network in the target intelligent network that can correspond to multiple communication devices or can correspond to multiple communication scenarios. It should be understood that the number of the first sub-network is not limited to one, for example, the first sub-network may include one or more sub-networks.
- the second sub-network is used to represent a sub-network in the target intelligent network that corresponds to a communication device or can correspond to a communication scenario. It should be understood that the number of the second sub-network is not limited to one, for example, the second sub-network may include one or more sub-networks.
- the target intelligent network may include a first sub-network and a second sub-network, the first sub-network corresponds to multiple objects (such as multiple communication devices or multiple communication scenarios), and the second sub-network corresponds to an object (such as a communication device or a communication scenario). Therefore, some communication devices or some communication scenarios can reuse the first sub-network, that is, the frequency of change does not need to be too high, which can reduce complexity and cost. Different communication devices or different communication scenarios may have respective corresponding second sub-networks, that is, the second sub-networks may be trained and updated respectively, so that the performance can be guaranteed as much as possible. In this way, performance and complexity can be effectively balanced, opening up a whole new space for performance improvement and complexity reduction.
- the first sub-network and the second sub-network are divided based on one or more of the following: environmental data information, perception signal information, communication device identification information, Service type information, feedback suggestion information from terminal equipment.
- the information related to the first sub-network and/or the second sub-network is carried in one or more of the following signaling: radio resource control, media intervention Control - Control elements, downlink control information, uplink control information.
- the first sub-network corresponds to one or more indexes.
- the correspondence between the first sub-network and the index is predefined or pre-configured.
- the information of the first sub-network can be obtained.
- the communication device can directly read the first sub-network corresponding to the ID according to the ID.
- the communication device can apply to other devices (such as network devices) for the first sub-network corresponding to the ID according to the ID, so that other devices can send the information of the first sub-network corresponding to the ID to the communication device.
- the update period of the first sub-network is different from the update period of the second sub-network.
- the update period of the first sub-network is longer than the update period of the second sub-network.
- the first sub-network corresponds to multiple communication devices; the first sub-network is obtained by training some or all of the multiple communication devices.
- the training of the first sub-network can be flexibly designed.
- the first sub-network corresponds to multiple terminal devices; the first sub-network is obtained by training some or all of the multiple terminal devices.
- some or all of the communication devices are communication devices indicated by the network device; or, some or all of the communication devices are communication devices determined based on preset information.
- some or all of the terminal devices are terminal devices indicated by the network device; or, some or all of the terminal devices are terminal devices determined based on preset information.
- the network device may indicate the terminal device participating in generating the first sub-network in the form of an x-bit switch, where x is an integer greater than 1 or equal to 1.
- the network device may send a list of terminal devices, indicating the terminal devices participating in generating the first sub-network.
- the first sub-network is obtained by training some of the plurality of communication devices, and some of the communication devices share the first sub-network with the rest of the plurality of communication devices. network.
- the training of the first sub-network can be flexibly designed.
- the first sub-network is obtained by training some terminal devices in the multiple terminal devices, and some terminal devices share the first sub-network with the remaining terminal devices in the multiple terminal devices.
- the method further includes: acquiring information of the first sub-network according to an index corresponding to the first sub-network; or, receiving indication information, and acquiring the first sub-network according to the indication information Sub-network information, wherein the indication information is used to indicate the information of the first sub-network.
- the common features include one or more of the following: geographic location information, delay spread information, Doppler distribution information, and angle distribution information.
- a method for generating an intelligent network may include: acquiring information of a plurality of sub-networks, at least two sub-networks corresponding to different types of features; and generating a target intelligent network for the communication network based on the information of the plurality of sub-networks.
- the method may include: acquiring information of a first sub-network and a second sub-network, the first sub-network is used to process common features in the information, and the second sub-network is used to process non-common features in the information;
- the information of a sub-network and a second sub-network generates a target intelligent network for the communication network.
- a method for generating an intelligent network may include: acquiring information of multiple sub-networks, some of the sub-networks corresponding to multiple objects, and some of the sub-networks corresponding to one object; and generating a target intelligent network for the communication network based on the information of the multiple sub-networks.
- the method may include: acquiring information of a first sub-network and a second sub-network, where the first sub-network corresponds to multiple objects, and the second sub-network corresponds to one object; based on the information of the first sub-network and the second sub-network , generating the target intelligent network for the communication network.
- an information processing apparatus configured to execute the methods provided in the above-mentioned first to sixth aspects.
- the apparatus may include units and/or modules for performing the methods provided in the first to sixth aspects, such as a processing unit and/or a communication unit.
- the apparatus is a communication device.
- the communication unit may be a transceiver, or an input/output interface;
- the processing unit may be a processor.
- the apparatus is a chip, a system of chips, or a circuit for use in a communication device.
- the communication unit may be an input/output interface, interface circuit, output circuit, input circuit, pin or Related circuits, etc.;
- the processing unit may be a processor, a processing circuit, a logic circuit, or the like.
- the above transceiver may be a transceiver circuit.
- the above-mentioned input/output interface may be an input/output circuit.
- an apparatus for information processing including a processor.
- the processor is coupled to the memory and can be used to execute instructions in the memory to implement the methods of the first to sixth aspects above.
- the apparatus further includes a communication interface, the processor is coupled to the communication interface, and the communication interface is used to transmit data and/or instructions with the outside world.
- the apparatus further includes a memory.
- the apparatus is a communication device.
- the communication unit may be a transceiver, or an input/output interface;
- the processing unit may be a processor.
- the apparatus is a chip, a system of chips, or a circuit for use in a communication device.
- the communication unit may be an input/output interface, interface circuit, output circuit, input circuit, pin or Related circuits, etc.;
- the processing unit may be a processor, a processing circuit, a logic circuit, or the like.
- the above transceiver may be a transceiver circuit.
- the above-mentioned input/output interface may be an input/output circuit.
- an information processing device comprising: a memory for storing a program; a processor for executing the program stored in the memory, and when the program stored in the memory is executed, the processor is used for executing the above-mentioned first Aspects to the methods provided in the sixth aspect.
- the apparatus is a communication device.
- the apparatus is a chip, a system of chips, or a circuit for use in a communication device.
- the present application provides a processor for executing the methods provided by the above aspects.
- the process of sending the above-mentioned information and obtaining/receiving the above-mentioned information in the above-mentioned methods can be understood as the process of outputting the above-mentioned information by the processor and the process of receiving the above-mentioned input information by the processor.
- the processor When outputting the above-mentioned information, the processor outputs the above-mentioned information to the transceiver for transmission by the transceiver. After the above-mentioned information is output by the processor, other processing may be required before reaching the transceiver.
- the transceiver obtains/receives the above-mentioned information, and inputs it into the processor. Furthermore, after the transceiver receives the above-mentioned information, the above-mentioned information may need to perform other processing before being input to the processor.
- obtaining the information in the communication network mentioned in the foregoing method may be understood as the processor receiving the input indication information.
- the above-mentioned processor may be a processor specially used to execute these methods, or may be a processor that executes computer instructions in a memory to execute these methods, such as a general-purpose processor.
- the above-mentioned memory can be a non-transitory (non-transitory) memory, such as a read-only memory (Read Only Memory, ROM), which can be integrated with the processor on the same chip, or can be set on different chips respectively.
- ROM read-only memory
- the embodiment does not limit the type of the memory and the setting manner of the memory and the processor.
- a computer-readable storage medium stores program codes for device execution, the program codes including methods for executing the above-mentioned first to sixth aspects.
- a twelfth aspect provides a computer program product comprising instructions, which, when the computer program product runs on a computer, causes the computer to execute the methods provided in the first to sixth aspects above.
- a thirteenth aspect provides a chip, the chip includes a processor and a communication interface, the processor reads an instruction stored in a memory through the communication interface, and executes the methods provided in the first to sixth aspects.
- the chip may further include a memory, in which instructions are stored, the processor is configured to execute the instructions stored in the memory, and when the instructions are executed, the The processor is configured to execute the methods provided in the first to sixth aspects above.
- FIG. 1 is a simplified schematic diagram of a communication system applicable to an embodiment of the present application.
- FIG. 2 is a simplified schematic structural diagram of a communication system applicable to an embodiment of the present application.
- FIG. 3 is a simplified schematic block diagram of a communication system applicable to an embodiment of the present application.
- Figure 4 shows a schematic diagram of the ELAA structure.
- Fig. 5 shows a schematic structural diagram of the deep network model.
- FIG. 6 shows a schematic diagram of CSI reconstruction using the traditional scheme and the neural network scheme.
- FIG. 7 shows a schematic diagram comparing the performance of traditional methods for CSI acquisition and DNN for CSI acquisition under a specific channel distribution.
- Figure 8 shows a schematic diagram comparing the performance of the schemes for multi-distribution training-oriented networks and uni-distribution training-oriented networks.
- FIG. 9 is a schematic diagram of an information processing method provided by an embodiment of the present application.
- FIG. 10 is a schematic diagram of a target intelligent network provided according to an embodiment of the present application.
- FIG. 11 is a schematic diagram of an information processing method provided by another embodiment of the present application.
- FIG. 12 is a schematic diagram of an information processing method provided by another embodiment of the present application.
- FIG. 13 to FIG. 16 show schematic diagrams of possible use cases of sub-networks applicable to the embodiments of the present application.
- FIG. 17 shows a schematic diagram of geographic location-based grouping applicable to an embodiment of the present application.
- FIG. 18 shows a schematic diagram of target intelligent network processing suitable for the embodiment of the present application.
- FIG. 19 is a schematic block diagram of an apparatus for information processing provided according to an embodiment of the present application.
- FIG. 20 is another schematic block diagram of an apparatus for information processing provided according to an embodiment of the present application.
- FIG. 21 is another schematic block diagram of an apparatus for information processing provided according to an embodiment of the present application.
- FIG. 22 is a schematic diagram of a chip hardware structure provided by an embodiment of the present application.
- the present application is not limited thereto, and a simplified schematic diagram of a communication system to which the embodiments of the present application are applied is provided.
- the communication system includes a radio access network 100 .
- the radio access network 100 may be a next-generation (eg, 6th generation (6G) or later) radio access network, or a legacy (eg, 5th generation (5G), fourth generation ( 4th generation (4G), third generation (3G) or second generation (2th generation, 2G)) radio access network.
- One or more communication devices 120a-120j, collectively 120 may be connected to each other or to one or more network nodes (110a, 110b, collectively 110) in the radio access network 100.
- the communication system includes a core network (not shown).
- the wireless access network equipment is connected to the core network in a wireless or wired manner.
- the core network may depend on or be independent of the radio access technology used in the radio access network 100 .
- the core network device and the radio access network device can be independent and different physical devices, or the functions of the core network device and the logical functions of the radio access network device can be integrated on the same physical device, or they can be one physical device. It integrates the functions of some core network equipment and some functions of the wireless access network equipment. Terminals and terminals and wireless access network devices and wireless access network devices may be connected to each other in a wired or wireless manner.
- the communication system may further include other network devices, for example, a wireless relay device, or a wireless backhaul device.
- One or more of the network nodes 110 in the radio access network 100 may be next generation nodes, legacy nodes, or a combination thereof.
- Network nodes are used to communicate with communication devices and/or other network nodes.
- network nodes are also sometimes referred to as network devices.
- a non-limiting example of a network node is a base station (BS).
- BS can be referred to by any of various names in a broad sense, such as: gNodeB/gNB, eNodeB/eNB, Node B, next-generation base station in 6G mobile communication system), future mobile communication system Base station, access point in Wifi system, base transceiver station / base transceiver station (Base Transceiver Station, BTS), transmission reception point (transmission reception point, TRP), macro base station (MacroeNB, MeNB), micro base station (PicoeNB) , SeNB), Multi-Standard Radio (MSR) wireless node, home base station, network controller, access node, wireless node, access point (AP), transmission node, transceiver node, baseband unit (Base Band Unit, BBU), remote radio unit (Remote Radio Unit, RRU), active antenna unit (Active Antenna Unit, AAU), radio head (Remote Radio Head, RRH), centralized unit (centralized unit, CU) , distributed unit (DU),
- a base station may be a macro base station, a micro base station, a relay node, a donor node, or the like, or a combination thereof.
- the base station may also refer to a device built in the above-mentioned equipment, for example, a communication module, a modem or a chip in the above-mentioned equipment.
- the embodiments of the present application do not limit the specific technology and specific device form adopted by the wireless access network device.
- Base stations can support networks of the same or different access technologies. For ease of description, the network node is described below by taking a base station (BS) as an example.
- BS base station
- Network nodes can be fixed or mobile.
- the network nodes 110a, 110b are stationary and are responsible for wireless transmission and reception in one or more cells from the communication device 120.
- the aircraft (eg, helicopter or drone) 120i shown in FIG. 1 may be configured to act as a mobile BS, and one or more cells may move according to the location of the aircraft 120i.
- a helicopter or drone (120i) may be configured to function as a communication device with network node 110a.
- the communication device 120 is used to connect people, things, machines, etc.
- the communication device 120 can be widely used in various scenarios, such as cellular communication, device-to-device (D2D), vehicle to everything (V2X) ), Peer to Peer (P2P), Machine to Machine (M2M), Machine-type Communications (MTC), Internet of Things (IoT), Virtual Reality (virtual reality) reality, VR), augmented reality (AR), industrial control, autonomous driving, telemedicine, smart grid, smart furniture, smart office, smart wear, smart transportation, smart city drone, robot, remote sensing, passive transmission Sensing, positioning, navigation and tracking, autonomous delivery and other scenarios.
- D2D device-to-device
- V2X vehicle to everything
- P2P Peer to Peer
- M2M Machine to Machine
- MTC Machine-type Communications
- IoT Internet of Things
- VR Virtual Reality
- AR augmented reality
- industrial control autonomous driving, telemedicine, smart grid, smart furniture, smart office, smart wear, smart transportation, smart city
- the communication apparatus 120 may be a user equipment (user equipment, UE), fixed equipment, mobile equipment, handheld equipment, wearable equipment, terminal equipment, cellular phone, Smartphones, SIP phones, tablets, laptops, computers with wireless transceiver capabilities, smart books, vehicles, satellites, Global Positioning System (GPS) devices, target tracking devices, aircraft (e.g.
- UE user equipment
- Fixed equipment mobile equipment
- handheld equipment wearable equipment
- terminal equipment cellular phone
- Smartphones SIP phones
- tablets laptops
- computers with wireless transceiver capabilities smart books, vehicles, satellites, Global Positioning System (GPS) devices
- GPS Global Positioning System
- target tracking devices e.g.
- the communication apparatus 120 is described below by taking a terminal, a terminal device or a UE as an example.
- the UE may be used to act as a base station.
- a UE may act as a scheduling entity that provides sidelink signals between UEs in scenarios such as V2X, D2D or P2P.
- UE 120a and UE 120b may communicate using sidelink signals.
- the UE 120a and the UE 120d communicate without relaying the communication signal through the network node 110a.
- the function of the base station may also be performed by a module (eg, a chip) in the base station, or may be performed by a control subsystem including the function of the base station.
- the control subsystem including the base station function may be the control center in the application scenarios of the above-mentioned terminals such as smart grid, industrial control, intelligent transportation, and smart city.
- the functions of the terminal can also be performed by a module (such as a chip or a modem) in the terminal, and can also be performed by a device including the terminal functions.
- FIG. 2 a simplified schematic structural diagram of a communication system is provided.
- FIG. 2 only shows network node 110 (eg, BS 110 ), communication device 120 (eg, UE 120 ) and network 130 .
- BS 110 includes interface 111 and processor 112.
- Processor 112 may optionally store program 114 .
- BS 110 may optionally include memory 113 .
- Memory 113 may optionally store program 115 .
- UE 120 includes interface 121 and processor 122.
- Processor 122 may optionally store program 124 .
- UE 120 may optionally include memory 123.
- Memory 123 may optionally store program 125 .
- These components work together to provide the various functions described in this application.
- processor 112 and interface 121 work together to provide a wireless connection between BS 110 and UE 220.
- the processor 122 and the interface 121 work together to realize downlink transmission and/or uplink transmission of the UE 120.
- the network 130 may include one or more network nodes 130a, 130b to provide core network functions.
- the network nodes 130a, 130b may be next generation (eg, 6G or later) core network nodes, or legacy (eg, 5G, 4G, 3G or 2G) core network nodes.
- the networks 130a, 130b may be an Access and Mobility Management Function (AMF), a mobility management entity (mobility management entity,
- AMF Access and Mobility Management Function
- mobility management entity mobility management entity
- the network 130 may also include a Public Switched Telephone Network (PSTN), a packet data network, an optical network, one or more network nodes in an IP network, a Wide Area Network (WAN), a Local Area Network (Local Area Network) , LAN), Wireless Local Area Network (WLAN), wired network, wireless network, metropolitan area network, and other networks to enable communication between UE 120 and/or BS 110.
- PSTN Public Switched Telephone Network
- WAN Wide Area Network
- Local Area Network Local Area Network
- LAN Local Area Network
- WLAN Wireless Local Area Network
- wired network wireless network
- wireless network metropolitan area network
- BS 110 metropolitan area network
- a processor may include one or more processors and be implemented as a combination of computing devices.
- a processor eg, processor 112 and/or processor 122
- DSP digital signal processor
- DSPD digital signal processing device
- ASIC application specific integrated circuit
- FPGA field programmable gate array
- PLD programmable logic device
- the processor may be a general purpose processor or a special purpose processor.
- processor 112 and/or processor 122 may be a baseband processor or a central processing unit.
- a baseband processor may be used to process communication protocols and communication data.
- the central processing unit may be used to cause the BS 110 and/or the UE 120 to execute software programs and process data in the software programs.
- a portion of the processor may also include non-volatile random access memory.
- the processor may also store device type information.
- Interfaces may include wires for enabling communication with one or more computer devices (eg, for example, in some embodiments, interfaces may include wires for coupling wired connections or for coupling wireless transceivers with A terminal and/or pin for wireless connection.
- the interface can include a transmitter, a receiver, a transceiver and/or an antenna.
- the interface can be configured to use any available protocol (for example, the 3GPP standard), For communication between computer devices (eg, UEs, BSs and/or network nodes).
- a program in this application is used in the broadest sense to mean software.
- Non-limiting examples of software include: program code, programs, subroutines, instructions, sets of instructions, codes, code segments, software modules, applications, or software applications, and the like.
- Programs can run on processors and/or computers. to cause the BS 110 and/or the UE 120 to perform various functions and/or procedures described in this application.
- Memory may store data required by processors 112, 122 when executing software.
- Memory 113, 123 may be implemented using any suitable storage technology.
- memory may be the processor and/or Or any available storage medium that a computer can access.
- Non-limiting examples of storage media include: random access memory (RAM), read-only memory (ROM), electrically erasable programmable only memory Read memory (electrically EPROM, EEPROM), CD-ROM (Compact Disc-ROM, CD-ROM), static random access memory (static RAM, SRAM), dynamic random access memory (dynamic RAM, DRAM), synchronous dynamic Random access memory (synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (double data rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous connection dynamic random access memory Access memory (synchlink DRAM, SLDRAM) and direct memory bus random access memory (direct rambus RAM, DR RAM), removable media, optical disk storage, magnetic disk storage media, magnetic storage devices, flash memory, registers, state memory, remote storage On-board memory, local or remote memory components, or any other medium that can carry or store software, data or information and can be accessed by a processor/computer. It should be noted that the memory described herein is intended to include
- the memory eg, the memory 113 and/or the memory 123) and the processor (eg, the processor 112 and/or the processor 122) may be provided separately or integrated.
- the memory may be used in connection with the processor such that the processor can access the memory from the memory Read information in, store and/or write information in memory.
- Memory 113 may be integrated in processor 112.
- Memory 123 may be integrated in processor 122.
- the processor eg, processor 112 and/or processor 122) may be provided in an integrated circuit (eg, the integrated circuit may be provided in a UE or BS or other network node.
- the various functions described in this application may be implemented by an apparatus (eg, a UE, a BS, and/or any network node) that includes one or more modules, components, circuits, software, components, etc. (collectively referred to as components). These elements may be implemented using hardware, software, firmware, and/or combinations thereof.
- the wireless communication system may include at least one network device, such as the network device 311 shown in FIG. 3 (which may be a network node (such as BS 110) in FIG. 1 and FIG. 2 ), and the wireless communication system may also include at least one terminal device,
- the terminal device 321 shown in FIG. 3 can be the communication device (such as UE 120) in FIGS. 1 and 2 ).
- Both the network device and the terminal device can be configured with multiple antennas, and the network device and the terminal device can communicate using the multi-antenna technology.
- the network device and the terminal device may send and receive RRC signaling through a radio resource control (radio resource control, RRC) signaling interaction module.
- RRC radio resource control
- network equipment and terminal equipment may send and receive media access control-control elements (media access control-control element, MAC) through a media access control (media access control, MAC) signaling interaction module - CE) signaling.
- network equipment and terminal equipment may send and receive signaling and data through a physical layer (PHY) layer, for example, uplink control information (UCI) (such as physical layer, PHY).
- PHY physical layer
- UCI uplink control information
- Uplink control channel Physical uplink control channel, PUCCH
- downlink control signaling downlink control information, DCI
- DCI downlink control information
- PDCCH physical downlink control channel
- uplink data such as physical uplink shared channel (physical uplink shared channel) uplink shared channel, PUSCH)
- PDSCH physical downlink shared channel
- the technical solutions of the embodiments of the present application may be applied to a waveform system in a wireless communication network, for example, a single carrier, orthogonal frequency division multiplexing (orthogonal frequency division multiplexing, OFDM) or other waveform systems.
- the architecture of the embodiment of the present application may be the same as the communication architecture in the current standard, such as a traditional single-carrier or OFDM system, or may be other different communication architectures, which is not limited.
- a series of processing is performed on the signal at the transmitting end (such as the network device 311), such as performing discrete Fourier transformation (DFT) operations on the signal, and the signal after DFT is processed.
- DFT discrete Fourier transformation
- Subcarrier mapping inverse fast Fourier transform (IFFT), and cyclic prefix (CP) are added.
- IFFT inverse fast Fourier transform
- CP cyclic prefix
- inverse operations are performed at the receiving end (eg, the terminal device 321 ), such as the processes of de-CP, DFT, subcarrier demapping/equalization, and IFFT.
- the technical solutions of the embodiments of the present application can be applied to any learning based on intelligent network modules in the network architecture.
- intelligent network such as neural network, decision tree, deep forest and other networks
- channel estimation channel estimation, signal detection, precoding design, coding and decoding, modulation and demodulation, etc. are realized.
- AGI Artificial general intelligence
- AGI Artificial intelligence (AI) with generality.
- AI Artificial intelligence
- AGI not only has some functions of AI, such as the existing key functions such as knowledge expression, reasoning and decision-making, but also has the ability to quickly learn and generalize, that is, the agent can effectively build its own knowledge system based on the logical mapping relationship between behaviors and actions. So as to form experience to adapt to the environment and survive.
- Perception is the ability to feel and understand the characteristics of the environment. Decision-making, i.e. being able to take action based on a goal to maximize the probability of success.
- Machine agents spontaneously build a unique knowledge system, and then form experience to adapt to the environment. In other words, trying to make artificial intelligence able to learn to handle various tasks based on the same set (or the same class) of algorithms, rather than any single task being trained from scratch to survive. This is the main purpose of AGI research.
- Generalized Reinforcement Learning The form of generalized reinforcement reflects the process of how humans acquire information and transform it into knowledge and skills, and finally form a solidified and preserved experience and common sense, and then form high-level semantic information to be reused and disseminated.
- Meta learning that is, learning to learn. Meta-learning essentially provides a way to extract the interconnectedness of different pieces of information, not just a way to learn to learn.
- Transfer learning The time-consuming and power consumption of artificial intelligence network training in real network deployment will be one of its application bottlenecks, so network transfer and reuse will be one of the keys in the application process. This requires the network module to have the function of self-migration, which can enable rapid deployment and application in similar scenarios.
- Neural network modularization is a trend.
- the expected neural network is a neural network that can process various types of input such as speech, vision, and control at the same time. Modularization is similar to different functional areas of the brain, during which a lot of information can be shared and transferred.
- Massive MIMO massive MIMO
- Massive MIMO as a larger-dimensional multi-antenna system, utilizes the resources of the spatial dimension to enable the signal to obtain array gain, multiplexing, and diversity gain in space without increasing the system bandwidth.
- Interference cancellation gain multiplying the capacity and spectral efficiency of the communication system.
- FIG. 4 shows a schematic diagram of the ELAA structure.
- the antenna viewing area of the terminal is no longer the entire ELAA, and the terminal may only see a certain part of the sky, that is, a multipath cluster, and the multipath clusters experienced by different terminal groups are different .
- neural network algorithm has become a very potential technical method due to its ability to infinitely approximate any continuous function endowed by its universal approximation theorem.
- a deep neural network is generally a multi-layer structure. Increasing the depth and width of a neural network can improve its expressiveness, providing more powerful information extraction and abstract modeling capabilities for complex systems.
- Classic deep neural networks include: fully connected network (FCN), recurrent neural network (RNN) designed to model changes in time series, and convolution proposed to simulate the human visual nervous system Deep neural network (convolutional neural network, CNN).
- FCN fully connected network
- RNN recurrent neural network
- CNN convolutional neural network
- FIG. 5 shows a schematic structural diagram of a deep network model.
- a deep network model can include an input layer, a hidden layer, and an output layer.
- the circle marked with 1 represents the input layer
- the circle marked with 2 represents the hidden layer
- the circle marked with 3 represents the output layer.
- the role of the input layer includes inputting the information to be processed.
- DNN generally contains more than one hidden layer
- the role of the hidden layer includes extracting information features to varying degrees.
- the role of the output layer includes mapping the required output information from the extracted feature information.
- the circle marks represent the neurons of each layer, and the connection method of neurons in each layer and the activation function used will determine the expression function of the neural network.
- FIG. 5 is only an exemplary illustration, which is not limited thereto.
- different deep network models are applicable to different problems.
- the appropriate deep network model can be selected based on the model performance and complexity.
- DNNs are now widely used in image, speech and video processing. Based on its advantages in feature extraction and mining, many academic studies have turned their attention to the application of deep learning networks in the physical layer of wireless communication (such as MIMO). For example, deep learning networks have done a lot of research on channel estimation, MIMO channel acquisition, MIMO detection, precoding design, and reference signals.
- MIMO physical layer of wireless communication
- FIG. 6 shows a schematic diagram of CSI reconstruction using the traditional scheme and the neural network scheme.
- the traditional scheme if the traditional scheme is adopted, it mainly processes some typical angle-delay pairs of the channel in the angle-delay domain, such as compression (compress) and feedback processing.
- a neural network scheme such as DNN, can perform high-dimensional feature transformation on the angle delay information, and compress and process it in a higher dimension.
- the neural network algorithm does not understand the model (that is, it is similar to a black box), that is, except for the explicit requirements of the input and output, the feature transformation of each hidden layer inside the network cannot be known.
- changing features such as changing the scene, changing the compression rate, changing the quantization algorithm, etc., it is necessary to re-train the overall network to match the latest feature requirements.
- Using DNN in wireless communication scenarios can include the following two schemes:
- FIG. 7 shows a schematic diagram comparing the performance of traditional methods for CSI acquisition and DNN for CSI acquisition under a specific channel distribution.
- the performance of traditional methods such as 3GPP standard version 16 (Release 16, R16)
- DNN-based CSI DNN-based CSI
- SNR signal-noise ratio
- the other is a multi-distribution training-oriented network, that is, a generalized network.
- the network characteristics are based on a preset network and are configured on each terminal in the existing network. At this time, user-specific network training is no longer required.
- the cost and power consumption will be greatly reduced, and it can also meet the low latency requirements. But at this time, the network needs to face various channel scenarios. The inference performance of the network will deteriorate sharply due to the diversity of samples.
- FIG. 8 shows the performance of the above two schemes.
- (1) in FIG. 8 corresponds to a scheme for a multi-distribution training-oriented network (ie, a general network)
- (2) in FIG. 8 corresponds to a scheme for a single-distribution training network.
- Table 1 the characteristics of various types in FIG. 8 are shown in Table 1.
- samples are generally divided into three parts: training set, validation set and test set. Among them, the training set is used to build the model, and the test set is used to test the performance of the final optimal model.
- CDL in Table 1 represents the cluster delay line (cluster delay line, CDL).
- A represents the channel type, and B also represents the channel type.
- the network for multi-distribution training (that is, the general network) and the network for single-distribution training
- the former that is, the network for multi-distribution training
- the latter that is, the network for single-distribution training
- the latter that is, the network for single-distribution training
- the embodiments of the present application provide a solution, which can balance performance and complexity well, is more suitable for special needs of wireless communication (such as complex and diverse scenarios, low latency, high reliability, etc.), and can improve the use of intelligent networks in communication networks. achievability.
- FIG. 9 is a schematic diagram of an information processing method 900 provided by an embodiment of the present application.
- Method 900 may include the following steps.
- the target intelligent network uses the target intelligent network to process the information, where the target intelligent network includes a plurality of sub-networks, and at least two of the sub-networks correspond to different types of features.
- the information in the communication network is input to the target intelligent network, and the target intelligent network processes the information.
- the target intelligence network uses the target intelligence network to capture or process or capture features in the communication network.
- the information in the communication network is regarded as an image, and feature extraction processing is performed on the image according to the extraction capability of the target intelligent network.
- Intelligent networks which can be called intelligent modules, models, “black boxes", can represent networks with machine learning capabilities, such as artificial intelligence networks.
- the intelligent network used in the communication network provided by the embodiments of the present application is recorded as the target intelligent network.
- FIG. 10 shows a schematic diagram of a target intelligent network provided according to an embodiment of the present application.
- the target intelligent network includes multiple sub-networks, as shown in Figure 10 (1), the target intelligent network may include sub-networks L1, L2, ..., Li, and as shown in Figure 10 (2), the target intelligent network may include sub-networks L1, L2, ..., Lk.
- At least two sub-networks correspond to different types of features, or it can also be expressed that at least two types of sub-networks correspond to different types of features, or at least two sub-networks correspond to different types of features. At least two sub-networks correspond to different kinds of features, which means that the sub-networks capture or process their corresponding features, or in other words, different kinds of features are processed through their corresponding sub-networks.
- L1 and Li correspond to different kinds of features
- L2 and Li correspond to different kinds of features
- L1 and L2 may correspond to the same kind of features.
- L1 and L2 are the same type of sub-network
- Li is another type of sub-network
- the two types of sub-networks correspond to different types of features.
- L1 and L2 process the same type of features, Li processes another type of features; or, one type of features is processed by L1 and L2, and the other type of features is processed by Li; or, L1 and L2 process some features, Li processes another part of the features .
- L1 and Lk correspond to different types of features
- L2 and Lk correspond to different types of features
- L1 and L2 may correspond to the same type of features.
- L1 and L2 are the same type of sub-network
- Lk is another type of sub-network
- the two types of sub-networks correspond to different types of features.
- L1 and L2 process the same type of features
- Lk processes another type of features; or, one type of features is processed by L1 and L2, and the other type of features is processed by Lk; or, L1 and L2 process some features, Lk processes another part of the features .
- the target intelligent network provided by this application can be considered as a hierarchical network architecture.
- the target intelligent network includes a plurality of sub-networks, and at least two sub-networks correspond to different kinds of features.
- the capabilities of terminal devices are uneven, the service characteristics of the terminal devices may be similar, and even some terminal devices have different characteristics of grouping or clustering.
- the feature gap is relatively large; for terminal devices within a group, the feature gap is relatively small.
- intelligent networks such as neural networks themselves, intelligent networks can implement feature classification processing.
- the lower convolutional layer of the deep neural network model in computer vision captures low-level image features, such as edge information, color, etc., such features are almost invariable in different classification tasks, and the real difference is the high-level features.
- Higher convolutional layers capture increasingly complex details such as body parts, faces, and other compositional features.
- the network architecture of the hierarchical mechanism is designed by utilizing the characteristic grouping of the wireless communication network and the classification processing characteristics of the intelligent network.
- flexible training and use can be performed based on the characteristics corresponding to each sub-network.
- similar features among the multiple objects can be processed by the same sub-network, thereby reducing the complexity and saving training overhead.
- the specific features of each object can be processed by different sub-networks, for example, each object is trained and updated or a sub-network corresponding to this type of features is designed, so as to ensure the performance as much as possible. Therefore, based on this scheme, through feature layering, features at each level can be effectively grasped and extracted to ensure training performance, and training overhead can be minimized to improve training efficiency.
- Different types of features may include features with different degrees of similarity between different objects.
- different types of features may include features with different degrees of similarity between different communication devices (such as terminal devices or network devices), or may also include features with different degrees of similarity in different communication scenarios (such as different channel scenarios or different distributions). , or may also include features with different degrees of similarity under different processing objects.
- the processing objects are described below.
- the angle and time delay of the CSI information can be regarded as a type of feature, and this type of feature corresponds to a sub-network, that is, the same sub-network
- the network processes the angle and time delay information; or the CSI information may be classified into two types of features, that is, one sub-network processes the angle information, and another sub-network processes the time delay information.
- the target intelligent network may include multiple sub-networks, a part of the sub-networks corresponds to public features, and another part of the sub-networks corresponds to non-public features.
- Common features may indicate that such features do not vary with different communication devices, or that such features do not change significantly or differ little between different communication devices. For example, for multiple communication devices in a certain scenario, the same or similar features among the multiple communication devices, or the features with insignificant changes or minor differences, may be considered as common features.
- the common feature may indicate that such features do not vary with different communication scenarios (eg, different channel distributions), or that such features do not change significantly or differ little between different communication scenarios.
- a communication device is in multiple channel distributions, and the same or similar features of the communication device under the multiple channel distributions, or features with insignificant changes or minor differences may be considered as common features.
- the common feature can indicate that the feature of this class does not vary with different processing objects, or that the feature of this class does not change significantly or differs little between different processing objects. For example, for multiple processing objects in a certain scene, the same or similar features among the multiple processing objects, or the features with insignificant changes or minor differences, may be considered as common features.
- Non-public features may also be called private features or specific features.
- Non-common features may indicate that such features vary with different communication devices, or that such features vary significantly or differ greatly between different communication devices. For example, for multiple communication devices in a certain scenario, different features among the multiple communication devices, or specific features (or detailed features) dedicated to communication devices, or relatively obvious or different features Features, which can be considered non-public features.
- the non-common features may indicate that such features vary with different communication scenarios, or that such features have obvious changes or great differences between different communication scenarios.
- a communication device is in multiple channel distributions, and for the different characteristics of the communication device under the multiple channel distributions, or a specific feature (or detailed feature) dedicated to the channel distribution, or a feature with obvious changes or large differences , which can be considered a non-public feature.
- the non-common features may indicate that the features of this type vary with different processing objects, or that the changes of the features of this type between different processing objects are relatively obvious or different. For example, for multiple processing objects in a certain scene, the features that are different among the multiple processing objects, or the specific features (or detailed features) that are exclusive to the processing objects, or those with obvious changes or large differences Features, which can be considered non-public features.
- the same feature (or the same parameter for information, such as angle delay) can be classified differently in different scenarios.
- some features in the information of communication device 1 and communication device 2 can be considered as common features, and some features can be considered as non-public features; in scenario 2, when processing communication For the information of device 1 and communication device 2, the features belonging to the common features may be different from the features belonging to the common features under the scenario, and the features belonging to the non-public features may be different from the features belonging to the non-public features under the scenario.
- the frequency domain features corresponding to the communication devices can be considered as common features, and the time domain features corresponding to the communication devices. Can be considered a non-public feature.
- the frequency domain changes corresponding to different communication devices are obvious, but the time domain changes are not obvious, then in this scenario, the time domain features corresponding to the communication devices can be considered as common features, and the frequency domain features corresponding to the communication devices. Can be considered a non-public feature.
- the same feature is classified differently in the same scene.
- some features in the information of communication device 1 and communication device 2 can be considered as common features, and some features can be considered as non-public features; processing in the same scenario
- the common features and non-common features of communication device 3 and communication device 4 may be different from the common features and non-public features of communication device 1 and communication device 2.
- the angular delay can be considered as a common feature, that is, The angle delay information of the multiple UEs can be processed by the first sub-network.
- the angular delays can be considered as non-common features, that is, using a
- the second sub-network processes the angular delay information of the UEs in the UE group 1, and uses another second sub-network to process the angular delay information of the UEs in the UE group 2.
- sub-network is only an exemplary illustration for description, and its naming does not limit the protection scope of the embodiments of the present application.
- a sub-network may also be referred to as a network
- a sub-network may be referred to as a module
- a sub-network may be referred to as a component
- a sub-network may be referred to as a layer, and so on.
- the names that can represent the same or similar meanings as the subnet are all applicable to the embodiments of the present application.
- a sub-network is taken as an example for exemplary description.
- the different types of characteristics may include, for example, three types of characteristics, which are respectively recorded as the first type of characteristics, the second type of characteristics, and the third type of characteristics.
- the first type of features have the largest degree of similarity, for example, the features with the smallest difference between the characteristics of different communication devices belong to the first type of features;
- the third type of features have the smallest degree of similarity, such as the largest difference between the characteristics of different communication devices.
- the features belong to the third category of features; the similarity of the second category of features lies between the first category of features and the second category of features.
- the first type of feature corresponds to a sub-network
- the second type of feature corresponds to another sub-network
- the third type of feature corresponds to another sub-network.
- the update periods of the sub-networks corresponding to various features may be different.
- the update period of the sub-network corresponding to the first type of feature may be the longest
- the update period of the sub-network corresponding to the third type of feature may be the shortest.
- the above-mentioned division into multiple types of features based on the similarity degree of the features is only an exemplary description, and is not limited thereto. In the future, any manner of dividing features based on other factors will fall within the protection scope of the embodiments of the present application.
- the features may be divided into features in at least one of the time domain, the spatial domain, the frequency domain, and the time delay domain.
- feature division or definition can be performed by one or more modules independent of the sub-network.
- the module can automatically classify and input features of different classes to the corresponding sub-network.
- the sub-networks can also extract their corresponding features respectively.
- the information is input to the neural network as a whole, and each sub-network is capable of extracting its corresponding features for processing. This is not limited.
- the target intelligent network may include two or more sub-networks.
- each sub-network can be considered as an independent network, that is, the target intelligent network can be a concept defined for description, that is to say, each sub-network is considered as an independent sub-network, and the multiple independent sub-networks are considered as A target intelligent network is formed.
- each sub-network can also be considered as a partial network of the target intelligent network.
- the target intelligent network can be designed for different processing objects, such as being defined, constructed and trained for different processing objects.
- the processing object may be, for example, channel state information processing, codec processing, modulation and demodulation processing, channel codec, and so on.
- a processing object can correspond to one or more target intelligent networks.
- the terminal side and the network side may have respective corresponding target intelligent networks, and the terminal side and the network side may also have a common target intelligent network.
- the target intelligent network provided in this application can correspond to or be used for any processing objects in current or future wireless communication networks.
- the definition of processing objects may be different from existing classification methods.
- a processing object may be any combination of CSI processing, coding and decoding processing, modulation and demodulation processing, channel coding and decoding, channel estimation, and data detection.
- the receiving end may process all the processes from signal reception to decoding through one module.
- the target intelligent network provided by this application can correspond to the new processing object represented by the above process. .
- the construction form of the target intelligent network may be a neural network or other intelligent learning architecture, which is not limited thereto.
- the target intelligent network can be constructed through feedforward neural network, recurrent network and other architectures, including one or more of neural networks such as perceptron, convolutional neural network, recurrent neural network, etc. (for example, combined in some combination) to fulfill.
- the target intelligent network can also be implemented through non-neural networks, such as decision trees, deep forests and other networks.
- the above method 900 is mainly described from the point of view of features.
- the description can also be made from the perspective of the object to which the sub-network is applied, and an exemplary description will be given below in conjunction with the method 1100 shown in FIG. 11 .
- 1120 Use the target intelligent network to process the information, where the target intelligent network includes multiple sub-networks, some of the sub-networks correspond to multiple objects, and some of the sub-networks correspond to one object.
- corresponding may be understood as association or binding.
- a sub-network for an object then the object can be considered to correspond to the sub-network.
- a "partial sub-network" may include one or more sub-networks.
- an "object” may represent a communication device (such as a terminal device or a network device), or may also represent a communication scene, or may also represent other objects, which is not limited thereto.
- the object can also be the processing object described above when introducing the target intelligent network.
- target intelligent networks corresponding to each processing object are designed respectively, and some sub-networks may be the same in the multiple target intelligent networks corresponding to the multiple processing objects. It can be understood that the part of the sub-network can participate in processing part of the characteristics of the respective processing objects.
- the target intelligent network designed for the first processing object and the target intelligent network designed for the second processing object some sub-networks can be shared, and this part of the sub-network can not only participate in the processing.
- Part of the features of the first processing object may also participate in processing part of the features of the second processing object.
- the sub-network corresponds to an object, which means that the sub-network can be used for the object, or the sub-network can participate in processing the information of the object, or the sub-network is used to process some features of the object.
- a sub-network corresponds to multiple communication devices, which can mean that the multiple communication devices can all use the sub-network, or the sub-network can be shared by multiple communication devices, or the sub-network can participate in processing the multiple communication devices information, or that the sub-network can be used to process certain characteristics of the plurality of communication devices.
- the sub-network corresponds to one communication device, which means that the sub-network is only used for one communication device, or the sub-network is associated with one communication device, or the sub-network is dedicated to one communication device, or the sub-network is only used for processing the sub-network. Certain characteristics of a communication device.
- L1 and L2 correspond to multiple objects
- Lk and Li correspond to one object respectively.
- L1 and L2 can be shared by multiple communication devices, Lk corresponds to one communication device, and Li corresponds to another communication device.
- the target intelligent network shown in FIG. 10(1) is the target intelligent network corresponding to a communication device 1, and the target intelligent network shown in FIG. target intelligent network.
- Some of the sub-networks in the two target intelligent networks are the same, such as sub-networks L1 and L2, that is, the communication device 1 and the communication device 2 can share the sub-networks L1 and L2.
- L1 and L2 can be shared by multiple processing objects, Lk corresponds to one processing object, and Li corresponds to another processing object.
- the target intelligent network shown in Figure 10(1) is the target intelligent network corresponding to the codec processing
- the target intelligent network shown in Figure 10(2) is the corresponding modulation and demodulation processing. target intelligent network.
- Some of the sub-networks in the two target intelligent networks are the same, such as sub-networks L1 and L2, that is, the decoding process and the modulation and demodulation process can share the sub-networks L1 and L2.
- a hierarchical network architecture is proposed.
- a network architecture with a hierarchical mechanism is designed to ensure performance.
- the network overhead can be greatly reduced.
- the object's target intelligent network includes multiple sub-networks. A part of the sub-networks in the plurality of sub-networks can be used not only for the object, but also for one or more other objects, and another part of the sub-networks in the plurality of sub-networks is only used for the object.
- method 1100 and method 900 may be used independently or in combination.
- some sub-networks correspond to different objects, which can indicate that the part of the sub-network can be used to process the common features of the different objects; non-public features.
- first sub-network two types of sub-networks, respectively denoted as a first sub-network and a second sub-network, are mainly used as examples for illustrative description below.
- first sub-network and the second sub-network are only exemplary descriptions for distinction, and their names do not limit the protection scope of the embodiments of the present application.
- the first sub-network may also be referred to as a public sub-network
- the second sub-network may also be referred to as a private sub-network or a specific sub-network.
- the names that can represent the same or similar meanings as the first sub-network and the second sub-network are all applicable to the embodiments of the present application.
- the first sub-network and the second sub-network are taken as examples for exemplary description.
- FIG. 12 is a schematic diagram of an information processing method 1200 provided by an embodiment of the present application.
- Method 1200 may include the following steps.
- the target intelligent network includes a first sub-network and a second sub-network.
- the first sub-network is used for processing common features in information
- the second sub-network is used for processing non-public features in information; and/or, the first sub-network corresponds to multiple objects, and the second sub-network corresponds to one object.
- the first sub-network is used to represent the sub-network used to process common features in the information in the target intelligent network, and/or the first sub-network is used to represent the sub-network corresponding to an object. It should be understood that the number of the first sub-network is not limited to one, for example, the first sub-network may include one or more sub-networks. As shown in FIG. 10 , both L1 and L2 can be considered as the first sub-network.
- the second sub-network is used to represent the sub-network in the target intelligent network for processing non-common features in the information, and/or the second sub-network is used to represent the sub-network corresponding to a plurality of objects. It should be understood that the number of the second sub-network is not limited to one, for example, the second sub-network may include one or more sub-networks.
- Li can be regarded as a second sub-network
- Lk can be regarded as a second sub-network.
- the target intelligent network may include a first sub-network and a second sub-network, the first sub-network corresponds to public features, and the second sub-network corresponds to non-public features; and/or, the first sub-network corresponds to multiple objects, and the second sub-network corresponds to multiple objects
- a network corresponds to an object.
- the first sub-network corresponds to a common feature, such as a feature corresponding to a public environment, and different objects can reuse the first sub-network, that is, its frequency of change does not need to be too high, so complexity and cost can be reduced.
- the second sub-network corresponds to non-common features, and different objects can train and update the second sub-network, that is, it can perform training and updating in a shorter period, so the performance can be guaranteed as much as possible. In this way, performance and complexity can be effectively balanced, opening up a whole new space for performance improvement and complexity reduction.
- the object is mainly a communication device and a communication scene as an example for illustrative description.
- the first sub-network and the second sub-network may belong to any of the following situations.
- the first sub-network corresponds to multiple communication devices
- the second sub-network corresponds to one communication device
- the target intelligent network corresponding to the communication device includes a first sub-network and a second sub-network, and the communication device can reuse the first sub-network corresponding to other communication devices, It is sufficient to update and train the second sub-network corresponding to the communication device.
- FIG. 13 shows a specific example. It is assumed that the first sub-network corresponds to the first communication device and the second communication device. As shown in FIG. 13 , the target intelligent network corresponding to the first communication device includes a first sub-network and a second sub-network #1, and the target intelligent network corresponding to the second communication device includes a first sub-network and a second sub-network #2, It can be seen that the first communication device and the second communication device may share the first sub-network.
- the first sub-network may be used to handle features common to the first communication device and the second communication device.
- the first communication device and the second communication device belong to terminal devices in UE group 1 as shown in FIG. 4
- the first sub-network can be used to process the angular delay in the CSI information corresponding to the first communication device
- the first sub-network may also be used to process the angular delay in the CSI information corresponding to the second communication device.
- each communication device can share one first sub-network, thereby reducing the complexity.
- each communication device can have its own corresponding second sub-network, so that the second sub-network can be trained and updated according to the actual situation, so that the performance can be guaranteed as much as possible.
- the first sub-network corresponds to one communication device in different communication scenarios
- the second sub-network corresponds to one communication device in one communication scenario
- the target intelligent network corresponding to the communication device includes a first sub-network and a second sub-network, and the communication device can reuse the communication device In the first sub-network in other communication scenarios, it is sufficient to update and train the second sub-network in this communication scenario.
- FIG. 14 shows a specific example. It is assumed that the first sub-network corresponds to the first communication device in the first communication scenario and the second communication scenario. As shown in FIG. 14 , the target intelligent network corresponding to the first communication device in the first communication scenario includes a first sub-network and a second sub-network #1, and the target intelligent network corresponding to the first communication device in the second communication scenario includes It can be seen from the first sub-network and the second sub-network #2 that the first communication device in the first communication scenario and the first communication device in the second communication scenario can share the first sub-network.
- the first sub-network may be used to process common features for the first communication device in the first communication scenario and the communication device in the second communication scenario.
- the environmental feature ie the conference room
- the first sub-network can use the In order to extract the environmental characteristics of the first communication device in the first communication scenario, the first sub-network can also be used to extract the environmental characteristics of the first communication device in the second communication scenario.
- the communication devices can share a first sub-network in different communication scenarios, thereby reducing the complexity.
- the communication devices may have respective corresponding second sub-networks in each communication scenario, so that the second sub-network can be trained and updated according to the actual situation, so that the performance can be guaranteed as much as possible.
- the first sub-network corresponds to multiple communication scenarios
- the second sub-network corresponds to one communication scenario
- the target intelligent network corresponding to the communication scenario includes a first sub-network and a second sub-network, and the communication scenario can reuse the first sub-network corresponding to other communication scenarios, It is sufficient to update and train the second sub-network in the communication scenario.
- FIG. 15 shows a specific example. It is assumed that the first sub-network corresponds to the first communication scenario and the second communication scenario. As shown in FIG. 15 , the target intelligent network corresponding to the first communication scenario includes the first sub-network and the second sub-network #1, and the target intelligent network corresponding to the second communication scenario includes the first sub-network and the second sub-network #2, It can be seen that the first communication scenario and the second communication scenario may share the first sub-network.
- the first sub-network may be used to handle common features for the first communication scenario and the second communication scenario.
- the geographic feature ie the school
- the first sub-network can use
- the first sub-network can also be used to extract the geographic features of any communication device in the second communication scenario.
- each communication scenario can have its own corresponding second sub-network, so that the second sub-network can be trained and updated according to the actual situation, so that the performance can be guaranteed as much as possible.
- the first sub-network corresponds to multiple communication devices in one communication scenario
- the second sub-network corresponds to one communication device in one communication scenario
- the target intelligent network corresponding to the communication device includes a first sub-network and a second sub-network, and the communication device can multiplex other communication devices In the first sub-network in the communication scenario, the second sub-network for training the communication device in the communication scenario can be updated.
- FIG. 16 shows a specific example. It is assumed that the first sub-network corresponds to the first communication device in the first communication scenario and the second communication device in the first communication scenario. As shown in FIG. 14 , the target intelligent network corresponding to the first communication device in the first communication scenario includes a first sub-network and a second sub-network #1, and the target intelligent network corresponding to the second communication device in the first communication scenario includes The first sub-network and the second sub-network #2, it can be seen that the first communication device and the second communication device can share the first sub-network in the first communication scenario.
- the first sub-network may be used to process common features for the first communication device in the first communication scenario and the second communication device in the first communication scenario.
- the first communication device in the first communication scenario and the second communication device in the first communication scenario are terminal devices in UE group 1 as shown in FIG. 4 , in this case, the angular delay can be considered as is a common feature
- the first sub-network can be used to process the angle delay information of the first communication device in the first communication scenario
- the first sub-network can also be used to process the second communication device in the first communication scenario angle delay information.
- different communication devices in the same communication scenario can share one first sub-network, thereby reducing the complexity.
- different communication devices in the same communication scenario may have respective corresponding second sub-networks, so that the second sub-network can be trained and updated according to the actual situation, so that the performance can be guaranteed as much as possible.
- the first sub-network may correspond to information about an object group, and the group may be a communication device group (such as a terminal device group), or the group may also be a communication scene group, or the group may also be a feature group, etc. .
- the first object may correspond to multiple processing objects, that is, the communication devices or communication scenarios in the foregoing cases 1 to 4 may be replaced with processing objects.
- the first sub-network corresponds to one or more indexes (index, ID).
- the communication device can learn the information of the first sub-network by learning the ID corresponding to the first sub-network.
- the correspondence between the first sub-network and the ID is predefined or pre-configured.
- the communication device can directly read the first sub-network corresponding to the ID according to the ID.
- the first sub-network may be pre-trained, such as information that is pre-trained and broadcast by the network.
- the ID corresponding to the first sub-network may be the ID of the object corresponding to the first sub-network.
- the first sub-network corresponds to three IDs, and the three IDs are an ID that identifies the communication device 1 , an ID that identifies the communication device 2 , and an ID that identifies the communication device 3 .
- the ID corresponding to the first sub-network may be a group ID.
- the ID corresponding to the first sub-network is the ID of the object group, or any information form that can reflect the group.
- the ID corresponding to the first sub-network may be the group ID of the communication device.
- the ID corresponding to the first sub-network may be the group ID of the communication device, and the ID corresponding to the first sub-network may be the communication device.
- the ID corresponding to the first sub-network may also be an ID defined for the first sub-network.
- IDs may be defined or configured for each of the first sub-networks.
- the first sub-network has a group attribute.
- the first sub-network corresponds to multiple objects, and the multiple objects may be considered as one object group or multiple object groups.
- a first sub-network may correspond to one or more object groups.
- the object group may be a communication device group, or a communication scene group, or a processing object group.
- the first sub-network may correspond to one or more communication device groups, that is, the communication devices in the one or more communication device groups may share the corresponding first sub-network.
- the first sub-network may correspond to one or more communication scenario groups, that is, the communication scenarios in the one or more communication scenario groups may share the corresponding first sub-network.
- the first sub-network may correspond to one or more processing object groups, that is, the corresponding first sub-network may be shared when processing the processing objects in the one or more processing object groups.
- object group is not limited, for example, it may also be a feature group, and so on.
- the update periods of the first sub-network and the second sub-network may be different.
- the first sub-network is used to process common features in information, or the first sub-network corresponds to multiple objects, and the frequency of change does not need to be too high;
- the second sub-network corresponds to non-public features, or the second sub-network corresponds to an object , the training update can be performed in a shorter period in order to adapt to changes in equipment or changes in scenarios.
- the information related to the first sub-network and/or the second sub-network may be carried in one or more of the following signaling: RRC, MAC-CE, DCI, UCI.
- RRC Radio Resource Control
- MAC-CE Packet Control Entity
- DCI Downlink Control
- UCI User Information
- These signaling contents can be sent by means of channels such as PDCCH, PUCCH, PDSCH, PUSCH, and physical broadcast channel (PBCH).
- PBCH physical broadcast channel
- the interaction of information related to the first sub-network and/or the second sub-network may be: aperiodic, semi-permanent or periodic.
- the information related to the first sub-network and/or the second sub-network may include relevant information for dividing the first sub-network and the second sub-network, and may also include relevant information for training the first sub-network and the second sub-network, or Information about using the first sub-network and the second sub-network may be included, and so on.
- Information related to the first sub-network and/or the second sub-network may include, but is not limited to: information referenced when dividing the first sub-network and the second sub-network, data set information for training the first sub-network, training Data set information of the second sub-network, parameter information of the first sub-network, model structure information of the first sub-network, parameter information of the second sub-network, model structure information of the second sub-network, multiple data corresponding to the first sub-network Information about an object or object group, information about an object corresponding to the second sub-network, information about the correspondence between the first sub-network and a plurality of corresponding objects or object groups, and so on.
- the parameter information of the sub-network may include, but is not limited to, weight and/or bias information, for example.
- the model structure information of the sub-network may include, but is not limited to, the network type, scale and/or connection relationship between network nodes adopted by the intelligent network.
- a possible implementation could be based on some common characteristic or some kind of common characteristic.
- the common features may include but are not limited to one or more of the following: geographic location information, statistical delay spread information, Doppler distribution information, and angle distribution information.
- grouping can be based on geographic location. Taking the object group as a communication device group as an example, FIG. 17 shows a schematic diagram of grouping based on geographic location. As shown in FIG. 17 , grouping can be performed based on the geographic location of the communication device (or based on the area to which the communication device belongs), and the divided groups can be denoted as group 1 (G1), group 2 (G2), group 3 ( G3), group 4 (G4).
- the first sub-network corresponds to one or more communication device groups, the communication devices in the one or more communication device groups may share the first sub-network, and different communication device groups may correspond to different first sub-networks. Taking group 1 as an example, group 1 corresponds to a first sub-network, then all communication devices in this group 1 can share the first sub-network corresponding to this group 1.
- grouping may be based on Doppler distribution.
- UE group 1 can be regarded as an object group, the UE group 1 can correspond to a first sub-network, and each UE in the UE group 1 can respectively correspond to A second sub-network;
- UE group 2 may be regarded as an object group, the UE group 2 may correspond to another first sub-network, and each UE in the UE group 2 may correspond to a second sub-network respectively.
- the embodiment of the present application does not limit the grouping manner of the object groups.
- the grouping method can also be different.
- communication devices they may be grouped according to geographic locations; for communication scenarios, they may be grouped according to characteristics.
- the network device may notify the terminal device, or the terminal device may judge by itself.
- the first sub-network corresponds to one or more group IDs
- the network device may indicate the group ID to the terminal device.
- the targeted intelligent network design of the hierarchical mechanism can be aided by the way the network device indicates the group ID.
- the network device is responsible for dividing group information (eg, area information), and the terminal device determines the belonging by itself (eg, determines the group according to its location information and group reference location), and reports the group ID (eg, area ID).
- the target intelligent network design of the hierarchical mechanism can be assisted by means of the terminal device feeding back the group ID.
- the differences between different object groups are relatively obvious, the differences between the objects in the object groups are relatively small. Therefore, by designing the first sub-network for the object group, it can not only reduce the impact on performance as much as possible, but also reduce the complexity.
- Aspect 4 the division of the first sub-network and the second sub-network.
- the division of the first sub-network and the second sub-network may be divided by reference to one or more of the following information: environmental data information, geographic location information, delay spread information, Doppler distribution information, sensory signal information, communication equipment Identification information, service type information, terminal equipment feedback suggestion information, and so on.
- the network device divides the first sub-network and the second sub-network according to one or more of the following information: environmental data information, geographic location information, delay spread information, Doppler distribution information, perception signal information, communication equipment identification information, service type information, terminal equipment feedback suggestion information, and so on.
- the network device may notify the terminal device of the definition or division information of the first sub-network and the second sub-network.
- the interaction of the definition or division information of the first sub-network and the second sub-network may be: aperiodic, semi-permanent, or periodic.
- the definition or division information of the first sub-network and the second sub-network may be carried in one or more of the following signaling: RRC, MAC-CE, DCI, UCI.
- RRC Radio Resource Control
- MAC-CE MAC-CE
- DCI User Information Code Division Multiple Access
- UCI User Information Code Division Multiple Access
- These signaling contents can be sent by means of channels such as PDCCH, PUCCH, PDSCH, PUSCH, and PBCH.
- the terminal device feeds back definitions/division information suggestions for the first sub-network and the second sub-network according to one or more of the following information: environmental data information, geographic location information, delay extension information, multiple Puller distribution information, perception signal information, terminal equipment identification information, service type information, and so on.
- the feedback suggestion from the terminal device may also be: aperiodic, semi-permanent or periodic.
- a possible implementation is obtained by training a communication device.
- the communication device trains part or all of the network content in the first sub-network, and obtains the training result, that is, the first sub-network.
- Another possible implementation is obtained by training multiple communication devices.
- multiple communication devices train part or all of the network content in the first sub-network respectively, share and combine to obtain a joint training result, that is, the first sub-network.
- the manner of combining network sharing may include, but is not limited to, the D2D manner. The manner of sharing by terminal devices is described in detail below with reference to the information on sharing the first sub-network between terminal devices.
- the communication device participating in the training may be a default communication device, or may also be a contracted communication device (eg, a communication device contracted based on preset information). Or when the communication device is a terminal device, it may also be a terminal device indicated by the network device, which is not limited.
- the preset information may include: selecting a certain number of communication devices in sequence (eg, from large to small, or from small to large) according to the size of the communication device identifiers.
- the communication device participating in the training can be a communication device with strong capabilities in the communication device group, or it can be a communication device in the communication device group that has participated in the training, or it can be any communication device in the communication device group. limited.
- the communication devices participating in the training may be some or all of the communication devices in the communication device group.
- the network device may instruct the terminal device participating in generating the first sub-network in any of the following manners.
- the network device may indicate the terminal device participating in generating the first sub-network in the form of an x-bit switch, where x is an integer greater than 1 or equal to 1.
- whether to participate in generating the first sub-network may be indicated by a 1-bit field.
- 0 corresponds to participating in the generation of the first sub-network
- 1 corresponds to not participating in the generation of the first sub-network.
- 1 corresponds to not participating in generating the first sub-network
- 0 corresponds to participating in generating the first sub-network. It should be understood that how to indicate specifically, or how many bits are used to indicate, is not limited in this embodiment of the present application.
- the network device may send a list of terminal devices, indicating the terminal devices participating in generating the first sub-network.
- the network device may send a list of terminal devices participating in generating the first sub-network, such as UEi-UEj, all of which participate in generating the first sub-network.
- the network device may send a list of terminal devices that do not participate in generating the first sub-network, such as UEi'-UEj', and terminal devices other than UEi'-UEj' participate in generating the first sub-network.
- the network device can be pre-trained and sent to the terminal device.
- the first sub-network can also be predefined in other ways.
- the generation of the first sub-network and the second sub-network is described above.
- the specific process of generating the first sub-network and the second sub-network, or the training process, is not strictly limited.
- the above training process (the training process of the first sub-network and the second sub-network) may be offline training or online training, which is not limited.
- offline training can be performed, and the convergence is fast.
- offline training can realize the reasoning process of the intelligent network in the process of real-time communication, and the learning or training process of the sub-networks (such as the first sub-network and the second sub-network) is completed before the real-time communication.
- Online training completes the update of the sub-network at the same time in the process of real-time communication between the sender and the receiver, including the learning and reasoning process.
- training set allocation information and constraint information may be determined through configuration (such as an indication of a network device) or a pre-agreed agreement (such as according to a standard table agreement).
- the data set can also be selected according to a certain operation method (such as proportional) according to the information of the scale of the sub-network or the information of the proportion of the whole network.
- a convergence condition ie, a network training termination condition
- a constraint condition can be configured or pre-agreed as a constraint condition.
- the communication device may generate the target intelligent network according to the first sub-network and the second sub-network. It should be understood that the fact that the communication device generates the target intelligent network according to the first sub-network and the second sub-network is only a deterministic description, that is, the first sub-network and the second sub-network jointly process the information of the communication device, which generates the target intelligent network.
- the network may not necessarily act.
- the network formed by connecting the first sub-network and the second sub-network is the target intelligent network.
- connection between the first sub-network and the second sub-network is not limited.
- the connection mode of each neural unit in the neural network may be referred to, that is, the output of the first sub-network may be the input of the second sub-network, or the output of the second sub-network may be the input of the first sub-network.
- Aspect 6 the way for the communication device to know the sub-network.
- the communication device can train and update the second sub-network by itself, so it can learn the second sub-network by itself.
- the communication device may acquire the information of the first sub-network in the following multiple ways.
- the information of the first sub-network may include, but is not limited to: data set information for training the first sub-network, parameter information of the first sub-network, model structure information of the first sub-network, multiple data corresponding to the first sub-network Information about an object or object group, information about the correspondence between the first sub-network and a plurality of corresponding objects or object groups, and so on.
- the parameter information of the first sub-network may include, but is not limited to, weight and/or bias information, for example.
- the model structure information of the first sub-network may include, but is not limited to, the type of network adopted by the intelligent network, the scale, and the connection relationship between network nodes, and the like.
- the terminal device may acquire the information of the first sub-network in any of the following manners.
- the network device indicates the information of the first sub-network to the terminal device.
- the network device sends the information of the first sub-network to the terminal device.
- the network device may directly indicate the parameters and model structure information of the first sub-network to the terminal device.
- the first sub-network is bound to an ID (eg, a group ID)
- the network device can indicate the ID corresponding to the first sub-network to the terminal device, and the terminal device can learn the information of the first sub-network according to the ID.
- the terminal device can determine the belonging by itself, and report the group ID to apply for the information of the first sub-network corresponding to the group ID.
- the network device sends the information of the first sub-network corresponding to the group ID to the terminal device according to the group ID reported by the terminal device.
- the terminal device can learn the information of the first sub-network by itself.
- the terminal device participates in training the first sub-network and saves the information of the first sub-network.
- the correspondence between the terminal device group and the first sub-network may be stored in advance.
- the terminal device inherits the first sub-network corresponding to the terminal device group according to the terminal device group to which it belongs, or directly reads the information of the first sub-network corresponding to the terminal device group.
- the terminal devices share the information of the first sub-network.
- the terminal devices in the terminal device group share the information of the first sub-network. That is, the terminal devices corresponding to the same sub-network can share the first sub-network.
- the terminal devices in the terminal device group may share the information of the first sub-network in a D2D manner.
- Terminal devices may share relevant information of the sub-network (eg, information of the first sub-network, or network content trained by the terminal device, etc.) through D2D (eg, sidelink).
- the process may include at least three steps of synchronization (including the transmission of primary and secondary D2D synchronization signals), device discovery and communication.
- the sending of relevant information in this process may be implemented by the PC5 interface or other interfaces, which is not limited.
- the following describes the steps that D2D communication may include by taking information sharing between terminal devices in a terminal device group in a D2D manner as an example.
- synchronization For multiple terminal devices in a terminal device group to perform D2D communication, synchronization is generally required first. For example, a synchronization signal is received before a plurality of terminal devices communicate through D2D.
- the synchronization source can be a network device or a terminal device.
- the terminal device can be a terminal device participating in the training of the first sub-network (such as a sending terminal device), or a terminal device not participating in the training of the first sub-network (such as a receiving terminal device). ).
- the device discovery process can be triggered by the sending terminal device in the terminal device group or by the receiving terminal device.
- the sending terminal device notifies the sub-network to update related information, and other terminal devices in the terminal device group can monitor and receive it.
- the receiving terminal device sends request information to request the sub-network to update related information, and the transmitting terminal device responds to the sub-network related information based on the request information.
- the sending terminal device and the receiving terminal device implement D2D direct communication through D2D communication resources.
- the group ID refers to the group ID corresponding to the first sub-network used for the terminal device group.
- the group ID may be the ID of the terminal equipment group, or may be any information form that can reflect the terminal equipment group.
- the scheduling manner of the D2D communication resources is not limited.
- the D2D communication resource may be indicated by the network device, or may be selected by the terminal device from the resource pool by itself.
- the network device schedules designated resources to the terminal device for the terminal device to send direct data (direct data) and direct communication control information (direct control information).
- a resource pool is pre-allocated or configured or demarcated, and when it needs to be used, the terminal device selects resources from the resource pool by itself to send direct data and direct communication control information.
- a variety of optional interaction modes for sharing the relevant information of the public sub-network through the D2D method can be formed according to one or more of the following: the device discovery process requester, the synchronization signal sender, Communication resource scheduling, as well as group division and index configuration. That is to say, for a given terminal device group, according to different device discovery process requests, or different synchronization signal senders, or different communication resource scheduling methods, or different group divisions, or different index configuration methods, or any combination of the above, form A variety of optional interactive ways to share the relevant information of the public sub-network through D2D.
- the sending terminal device can start the device discovery process through the PCx interface according to the training situation of its first sub-network (for example, after the training is completed, or observe that the first network is quite different from the previous version), and after the receiving terminal device gives a response, The connection is established based on the configuration group index and the communication resources configured by the network device. And, the sending terminal device sends part or all of the first sub-network to the receiving terminal device.
- the communication device group 1 and the communication device group N respectively correspond to a first sub-network.
- the communication device group 1 includes: a communication device 1, a communication device 2, ..., a communication device n.
- the first sub-network corresponding to the communication device group 1 may have the attributes described in aspect 2.
- the communication device group 1 may correspond to an ID, such as the group ID of the communication device group 1 .
- the communication devices in the communication device group 1 may be grouped according to the solution described in aspect 3. For example, based on the geographic location of the communication devices, the communication device 1 , the communication device 2 , . . .
- the target intelligent network corresponding to the communication device in the communication device 1 includes a first sub-network and a second sub-network, and the first sub-network and the second sub-network may be divided based on the solution described in aspect 4.
- the target intelligent networks corresponding to the communication devices in the communication device group 1 all include a first sub-network and a second sub-network, and the generation of the first sub-network and the second sub-network may be generated based on the solution described in aspect 5.
- the communication device 1 is trained to generate a first sub-network, and the communication device 1 can share the first sub-network with the rest of the communication devices in the communication device group 1 (eg, communication device 2, . . . , communication device n).
- the communication devices in the communication device group 1 may obtain the information of the first sub-network according to the solution described in aspect 6.
- the communication device n directly reads the information of the first sub-network according to the ID corresponding to the first sub-network. The content of each aspect will not be repeated here.
- the information of the communication device 1 is input to the target intelligent network, the information is jointly processed by the first sub-network and the corresponding second sub-network in the target intelligent network, and the result is finally output.
- the result is input into a decoder (decoder), such as a decoder corresponding to an encoder (encoder).
- decoder such as a decoder corresponding to an encoder (encoder).
- the first sub-network processes the common features of the information, or in other words, the common features in the information are processed by the first sub-network.
- the second sub-network corresponding to the communication device 1 processes the non-public features of the information (for example, it can be denoted as distinct feature), or in other words, the non-public features in the information are processed by the second sub-network corresponding to the communication device 1.
- the information of the communication device n is input to the target intelligent network, the information is jointly processed by the first sub-network and the corresponding second sub-network in the target intelligent network, and the result is finally output.
- the result is input into the decoder, such as the decoder corresponding to the encoder.
- the first sub-network processes the common features of the information, or in other words, the common features in the information are processed by the first sub-network.
- the second sub-network corresponding to the communication device n processes the non-public features of the information, or in other words, the non-public features in the information are processed by the second sub-network corresponding to the communication device n.
- the communication device 1 and the communication device n correspond to the same first sub-network, and respectively correspond to a second sub-network.
- the angular delay can be regarded as a common feature for communication device 1 and communication device n.
- the angle delay information of the communication device 1 and the angle delay information of the communication device n can be processed through the same first sub-network, and other information can be processed through their corresponding second sub-networks respectively.
- the communication device sends information, and the target intelligent network processes the information, or in other words, the target intelligent network captures or processes or captures features in the information.
- one or more modules may be added between the communication device group and the target intelligent network, or in the target intelligent network, and the modules are used to perform feature division.
- the module can automatically classify it, input the angle delay information to the first sub-network, and input other information to the second sub-network.
- each sub-network has the ability to extract its corresponding features.
- the information of the communication device 1 is input to the target intelligent network, the first sub-network extracts the angle delay information for processing, and the second sub-network extracts other information for processing.
- each sub-network is not strictly limited.
- the output of the first sub-network may be the input of the second sub-network, as shown in FIG. 18 ; alternatively, the output of the second sub-network may be the input of the first sub-network.
- each sub-network may input its own result into the decoder.
- the combined result is input to the decoder.
- a communication device group may correspond to one decoder, as shown in FIG. 18 ; or a communication device group may also correspond to multiple decoders, which is not limited.
- FIG. 18 is only an exemplary illustration, which does not limit the protection scope of the embodiments of the present application.
- the network architecture of the first sub-network shown in FIG. 18 mainly refers to the neural network architecture, and the network architecture in the first sub-network may also be other intelligent learning architectures.
- the target intelligent network includes the first sub-network and the second sub-network as an example for description, but this does not limit the present application.
- a target intelligent network may include two or more sub-networks.
- the sub-network is mainly used as an example for illustration.
- the above sub-networks can all be replaced by modules, or components, or layers, or networks, and so on.
- the type here is not limited to the type to which the feature belongs.
- all of them may belong to public features, or all of them may belong to non-public features, or some of them may belong to public features, and some of them may belong to non-public features, which is not limited. Specifically, it needs to be determined in combination with other factors (such as the scene in which it is located).
- the target intelligent network corresponding to the modulation and demodulation processing includes the second sub-network.
- the second sub-network may include a plurality of sub-networks, and the plurality of sub-networks are respectively used for processing a plurality of non-common features. That is to say, for different non-public features, the sub-networks corresponding to each non-public feature can be trained and updated separately.
- the second sub-network may include a sub-network that handles non-public features. That is, for different non-common features, a sub-network can be trained and updated for processing.
- target intelligent network in the embodiments of the present application can be replaced with a target neural network (eg, DNN).
- DNN target neural network
- a network used in a communication network may be referred to as an AI network, or may also be referred to as an AI-MIMO network, and so on.
- target intelligent network ie, hierarchical network architecture
- implementation architecture which can be used in any complex network capable of feature classification in a communication network to improve the overall performance of the network.
- a communication device is used as an example for illustration, wherein the communication device can be replaced by a terminal device.
- a correspondence relationship is mentioned, such as a correspondence relationship between the first sub-network and an object (such as a communication device or a communication scene, etc.).
- the first sub-network has a corresponding relationship with the object, that is, it is used to indicate that the first sub-network is associated with the object, or the first sub-network can be used for the object.
- the specific form of the corresponding relationship can be, for example, the form of a table, or it can be related information about the corresponding object in the first sub-network, or it can be the related information of the corresponding first sub-network in the object, or it can also be other forms, which are not strictly limited.
- an effective hierarchical network architecture is realized through the design idea of combining the first sub-network with the second sub-network (or "public sub-network + private sub-network"). It can not only solve the problem that the network for multi-distribution training is difficult to grasp and extract the characteristics of diverse samples, but also solve the problem of insufficient generalization of the network for single-distribution training.
- the multi-dimensional optimization goal of greatly improving the performance of the communication system, efficiently balancing the air interface load and realizing the cost can be achieved, so as to alleviate the contradiction between the learning accuracy and the learning efficiency.
- the hierarchical network architecture (or AI architecture that can also be referred to as a hierarchical mechanism) proposed in this application can be used as a general framework to support various system solutions involving complex neural networks in commercial communication systems, opening up opportunities for performance improvement and complexity reduction. A new space that can improve the realizability of intelligent networks used in communication networks.
- the first sub-network can be trained by a few communication devices, and other communication devices can share the first sub-network, thereby minimizing training overhead and improving training efficiency.
- the training strategy can be flexibly adjusted for different problems, that is, targeted learning and updating can be carried out in the face of practical problems.
- each communication device can train the second sub-network independently to improve the speed and training efficiency, and to ensure the network performance as much as possible. and real-time requirements.
- the first sub-network can be trained based on a small number of terminal devices, and the network device may indicate a group ID (eg, terminal device group ID) to assist the target intelligent network design of the hierarchical mechanism.
- the network device may indicate a group ID (eg, terminal device group ID) to assist the target intelligent network design of the hierarchical mechanism.
- FIGS. 9 to 18 may be independent solutions, or may be combined according to internal logic, and these solutions all fall within the protection scope of the present application.
- the operations implemented by the terminal device may also be implemented by components (such as chips or circuits) that can be used in the terminal device
- the operations implemented by the network device may also be implemented by the network device. components (such as chips or circuits) are implemented.
- an embodiment of the present application provides an apparatus 1900 for information processing.
- the apparatus 1900 may be used to perform the method 900, the method 1100, and the method 1200 in the above embodiments.
- the apparatus 1900 includes a communication unit 1910 and a processing unit 1920 .
- the communication unit 1910 is used to obtain the information in the communication network; the processing unit 1920 is used to process the information with the target intelligent network, the target intelligent network includes a first sub-network and a second sub-network, the first sub-network One sub-network is used to process common features in the information, and a second sub-network is used to process non-common features in the information.
- the first sub-network is used to process common features of one or more communication devices; or, the first sub-network is used to process common features in one or more communication scenarios.
- the second sub-network is used to process non-public features of a communication device; or, the second sub-network is used to process non-public features in a communication scenario.
- the communication unit 1910 is used to obtain information in the communication network; the processing unit 1920 is used to process the information with the target intelligent network, where the target intelligent network includes a first sub-network and a second sub-network, The first sub-network corresponds to multiple communication devices, and the second sub-network corresponds to one communication device; or, the first sub-network corresponds to multiple communication scenarios, and the second sub-network corresponds to one communication scenario.
- the first sub-network and the second sub-network are divided based on one or more of the following: environmental data information, sensory signal information, communication device identification information, service type information, and feedback suggestion information from terminal devices.
- the information related to the first sub-network and/or the second sub-network is carried in one or more of the following signaling: radio resource control, media intervention control-control element, downlink control information, uplink control information .
- the first sub-network corresponds to one or more indices.
- the update period of the first sub-network is different from the update period of the second sub-network.
- the first sub-network corresponds to multiple terminal devices; the first sub-network is obtained by training some or all of the multiple terminal devices.
- some or all of the terminal devices are terminal devices indicated by the network device; or, some or all of the terminal devices are terminal devices determined based on preset information.
- the first sub-network is obtained by training some terminal devices in the multiple terminal devices, and some terminal devices share the first sub-network with the remaining terminal devices in the multiple terminal devices.
- the communication unit 1910 is further configured to acquire information of the first sub-network according to an index corresponding to the first sub-network; or, receive indication information, and acquire information of the first sub-network according to the indication information, wherein the indication information Information used to indicate the first subnet.
- the common features include one or more of the following: geographic location information, delay spread information, Doppler distribution information, and angle distribution information.
- the product realization form of the apparatus 1900 for information processing is a program code that can be run on a computer.
- the apparatus 1900 for information processing may be a communication device, or may be a chip, a chip system, or a circuit applied to the communication device.
- the communication unit 1910 may be a transceiver, or an input/output interface
- the processing unit 1920 may be a processor.
- the communication unit 1910 may be an input/output interface, interface circuit, output circuit, input circuit, pin on the chip, chip system or circuit or related circuits, etc.
- the processing unit 1920 may be a processor, a processing circuit, a logic circuit, or the like.
- an embodiment of the present application further provides an apparatus 2000 for information processing.
- the apparatus 2000 includes a processor 2010, the processor 2010 is coupled with a memory 2020, the memory 2020 is used for storing computer programs or instructions, and the processor 2010 is used for executing the computer programs or instructions stored in the memory 2020, so that the methods in the above method embodiments are implemented be executed.
- the apparatus 2000 may further include a memory 2020 .
- the apparatus 2000 may further include a communication interface 2030, and the communication interface 2030 is used for data transmission with the outside world.
- the apparatus 2000 is used to implement the method 900 in the embodiment shown in FIG. 9 .
- the apparatus 2000 is used to implement the method 1100 in the embodiment shown in FIG. 11 .
- the apparatus 2000 is used to implement the method 1200 in the embodiment shown in FIG. 12 .
- each step of the above-mentioned method can be completed by an integrated logic circuit of hardware in the processor 2010 or an instruction in the form of software.
- the methods disclosed in conjunction with the embodiments of the present application may be directly embodied as executed by a hardware processor, or executed by a combination of hardware and software modules in the processor.
- the software module may be located in random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, registers and other storage media mature in the art.
- the storage medium is located in the memory 2020, and the processor 2010 reads the information in the memory 2020, and completes the steps of the above method in combination with its hardware. To avoid repetition, detailed description is omitted here.
- the processor may be one or more integrated circuits, and is configured to execute related programs to execute the method embodiments of the present application.
- an embodiment of the present application further provides an apparatus 2100 for information processing.
- the apparatus 2100 can be used to implement the methods in the embodiments shown in FIGS. 9 to 18 .
- the apparatus 2100 includes a public module 2110 and a non-public module 2120 .
- the common module 2110 may be used to execute the relevant steps of the first sub-network in the above method embodiments.
- the common module 2110 can be used to process the common features of each communication device in the communication device group 1 as shown in FIG. 18 .
- the non-public module 2120 may be configured to execute the relevant steps of the second sub-network in the foregoing method embodiments.
- the common module 2110 may be used to process non-public features of the information of the communication device 1 as shown in FIG. 18 .
- the product implementation form of the public module 2110 and the non-public module 2120 is program code that can be executed on a computer.
- the common module 2110 and the non-common module 2120 may be integrated on the same chip, or may be provided on different chips respectively.
- Embodiments of the present application further provide a computer-readable storage medium, where the computer-readable storage medium stores program codes for device execution, where the program codes include the methods for executing the foregoing embodiments.
- the embodiments of the present application also provide a computer program product containing instructions, when the computer program product is run on a computer, the computer is made to execute the method of the above embodiment.
- An embodiment of the present application further provides a chip, the chip includes a processor and a communication interface, and the processor reads an instruction stored in a memory through the communication interface, and executes the method of the above embodiment.
- the chip may further include a memory, the memory stores instructions, the processor is configured to execute the instructions stored in the memory, and when the instructions are executed, the processor is configured to execute the methods in the foregoing embodiments.
- FIG. 22 is a schematic diagram of a chip system according to an embodiment of the present application.
- the method 900 in the above method embodiment, the method 1100 in the embodiment shown in FIG. 11 or the method 1200 in the embodiment shown in FIG. 12 can all be implemented in the chip shown in FIG. 22 .
- the chip system 2200 shown in FIG. 22 includes: a logic circuit 2210 and an input/output interface (input/output interface) 2220, the logic circuit is used for coupling with the input interface, and data is transmitted through the input/output interface (for example, a second at least part of the channel model) parameters to perform the methods described in FIGS. 9 to 18 .
- a logic circuit 2210 and an input/output interface (input/output interface) 2220
- the logic circuit is used for coupling with the input interface
- data is transmitted through the input/output interface (for example, a second at least part of the channel model) parameters to perform the methods described in FIGS. 9 to 18 .
- the disclosed apparatus and method may be implemented in other manners.
- the apparatus embodiments described above are only illustrative.
- the division of the above-mentioned units is only a logical function division.
- multiple units or components may be combined or may be Integration into another system, or some features can be ignored, or not implemented.
- the shown or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of devices or units, which may be in electrical, mechanical or other forms.
- the above-mentioned units described as separate components may or may not be physically separated, and components shown as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to implement the solution provided in this application.
- each functional unit in each embodiment of the present application may be integrated into one unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.
- the computer program product includes one or more computer instructions.
- the computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable device.
- the computer may be a personal computer, a server, or a network device or the like.
- Computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server, or data center over a wire (e.g.
- coaxial cable fiber optic, digital subscriber line (DSL)) or wireless (eg, infrared, wireless, microwave, etc.) to another website site, computer, server, or data center.
- DSL digital subscriber line
- wireless eg, infrared, wireless, microwave, etc.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
本申请提供了一种信息处理的方法和装置。该方法可以包括:获取通信网络中的信息;用智能网络对该信息进行处理。其中,智能网络包括第一子网络和第二子网络,第一子网络用于处理信息中的公共特征,第二子网络用于处理信息中的非公共特征。第一子网络可以用于多个通信设备,并且可以用于处理该多个通信设备的公共特征,如公共环境对应的特征。通过该方式,不同通信设备可以复用第一子网络,其变化频率不需要太高,因此可以降低复杂度和成本;每个通信设备可以各自训练各自对应的第二子网络,因此可以尽可能地保证性能。因此,通过本申请,可以有效地平衡性能和复杂度,为性能提升和复杂度降低打开全新空间。
Description
本申请要求于2020年12月24日提交中国专利局、申请号为202011556785.2、申请名称为“信息处理的方法和装置”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
本申请涉及无线通信领域,具体地,涉及应用了智能网络的无线通信技术,尤其涉及一种信息处理的方法和装置。
近些年人工智能(artificial intelligence,AI)技术的发展迅猛。AI技术涵盖机器学习、深度学习等实现技术。AI技术的实现主要在于使得计算机能够从数据中找到信息,从而学习一些规律,通过模型训练和优化,使模型对数据的理解更加深入。
AI技术中的智能网络,例如,深度神经网络(deep neural network,DNN)现在已广泛应用于图像、语音和视频处理等领域。基于DNN在特征提取和挖掘上的优势,有不少学术研究将目光转移到深度学习网络这样的智能网络在无线通信场景(如多输入多输出(multi-input multi-output,MIMO)***)中的应用上。
考虑到无线通信场景的特殊性,如低时延要求,更严格的功耗和成本要求,以及用户终端的运算能力受限等,如何为无线通信场景提供高效的神经网络学习架构是研究的热点。
发明内容
本申请提供一种信息处理的方法和装置,可以很好的平衡性能和复杂度,更加适合无线通信特殊需求,可以提升智能网络用于通信网络中的可实现性。
第一方面,提供了一种信息处理的方法。该方法可以包括:获取通信网络中的信息;用目标智能网络对信息进行处理,目标智能网络包括多个子网络,至少两个子网络对应不同种类的特征。
目标智能网络,可以称为智能模块、模型、“黑盒子”,可以表示具有机器学习功能的网络,如人工智能网络。
可以理解,该方法可以由通信设备执行,也可以由应用于通信设备上的芯片、芯片***或电路执行。
基于上述技术方案,利用无线通信网络的特征分组以及智能网络的分类处理特性,设计分级机制的网络架构。通过该方案,可以基于各个子网络对应的特征进行灵活的训练和使用。例如,对于多个对象来说,对于该多个对象之间相似的特征,可以使用相同的子网络进行处理,从而可以降低复杂度,节省训练开销。对于各个对象的特定特征,可以使用 不同的子网络进行处理,如各个对象各自训练更新或设计该类特征对应的子网络,从而可以尽可能地保证性能。因此,基于该方案,通过特征分层,既能有效把握和提取各个级别的特征,以保证训练性能,还能实现训练开销的最小化,提升训练效率。
第二方面,提供了一种信息处理的方法。该方法可以包括:获取通信网络中的信息;用目标智能网络对信息进行处理,目标智能网络包括多个子网络,部分子网络对应多个对象,部分子网络对应一个对象。
可以理解,该方法可以由通信设备执行,也可以由应用于通信设备上的芯片、芯片***或电路执行。
示例地,“对象”,可以表示通信设备(如终端设备或者网络设备),或者也可以表示通信场景,或者也可以表示其他对象。
子网络对应一对象,即表示该子网络可以用于该对象,或者说该子网络可以参与处理该对象的信息,或者说该子网络用于处理该对象的某些特征。
可以理解,其中“部分子网络”的含义,包括一个或多个子网络。基于上述技术方案,考虑到对象组之间的对象有一定相似或相近的特征,以及智能网络(如神经网络)内在的分类处理特性,因此,设计分级机制的网络架构,从而在保证性能的前提下可以大幅降低网络开销。例如,对于某一对象的目标智能网络来说,该对象的目标智能网络包括多个子网络。该多个子网络中的一部分子网络不仅可以用于该对象,还可以用于其他一个或多个对象,该多个子网络中的又一部分子网络仅用于该对象。因此,针对每个对象,不需要单独去设计完整的目标智能网络,可以节省训练成本和开销。此外,针对每个对象,有特定的子网络,那么可以基于每个对象的特点,单独训练特定的子网络,从而也可以尽可能保证网络性能和实时性需求。
第三方面,提供了一种信息处理的方法。该方法可以包括:获取通信网络中的信息;用目标智能网络对信息进行处理,目标智能网络包括第一子网络和第二子网络,第一子网络用于处理信息中的公共特征,第二子网络用于处理信息中的非公共特征。
第一子网络用于表示目标智能网络中用来处理信息中的公共特征的子网络。应理解,第一子网络并不限定子网络的数量为一个,例如第一子网络可以包括一个或多个子网络。
第二子网络用于表示目标智能网络中用来处理信息中的非公共特征的子网络。应理解,第二子网络并不限定子网络的数量为一个,例如第二子网络可以包括一个或多个子网络。
本申请提供的技术方案中提及的公共特征和非公共特征,是一个相对的概念,或者说是在某一特定情景下的划分,其并不限定特定的特征(或者说特定的参数)。
基于上述技术方案,第一子网络用于处理信息中的公共特征,即第一子网络对应的为公共特征,如公共环境所对应的特征,不同对象(如通信设备或者通信场景等)可以复用该第一子网络,其变化频率不需要太高,因此可以降低复杂度和成本。第二子网络用于处理信息中的非公共特征,即第二子网络对应非公共特征,不同对象(如通信设备或者通信场景等)可以各自训练更新第二子网络,即其可以以更短周期执行训练更新,因此可以尽可能地保证性能。通过该方式,可以有效地平衡性能和复杂度,为性能提升和复杂度降低打开全新空间。
结合第三方面,在第三方面的某些实现方式中,第一子网络用于处理一个或多个通信 设备的公共特征;或者,第一子网络用于处理一种或多种通信场景下的公共特征。
示例地,通信设备可以为终端设备,或者通信设备也可以为网络设备。
基于上述技术方案,多个通信设备或者多种通信场景可以共用一个第一子网络,进而降低复杂度。
结合第三方面,在第三方面的某些实现方式中,第二子网络用于处理一个通信设备的非公共特征;或者,第二子网络用于处理一种通信场景下的非公共特征。
基于上述技术方案,每个通信设备或者每种通信场景下可以有各自对应的第二子网络,从而可以根据实际情况训练更新第二子网络,以便可以尽可能地保证性能。
第四方面,提供了一种信息处理的方法。该方法可以包括:获取通信网络中的信息;用目标智能网络对信息进行处理;其中,目标智能网络包括第一子网络和第二子网络,第一子网络对应多个通信设备,第二子网络对应一个通信设备;或者,第一子网络对应多种通信场景,第二子网络对应一种通信场景。
第一子网络用于表示目标智能网络中可以对应多个通信设备或者可以对应多种通信场景的子网络。应理解,第一子网络并不限定子网络的数量为一个,例如第一子网络可以包括一个或多个子网络。
第二子网络用于表示目标智能网络中对应一个通信设备或者可以对应一种通信场景的子网络。应理解,第二子网络并不限定子网络的数量为一个,例如第二子网络可以包括一个或多个子网络。
基于上述技术方案,目标智能网络可以包括第一子网络和第二子网络,第一子网络对应多个对象(如多个通信设备或多种通信场景),第二子网络对应一个对象(如一个通信设备或一种通信场景)。因此,有些通信设备或者有些通信场景可以复用第一子网络,即其变化频率不需要太高,可以降低复杂度和成本。不同通信设备或不同通信场景下可以有各自对应的第二子网络,即可以各自训练更新第二子网络,因此可以尽可能地保证性能。通过该方式,可以有效地平衡性能和复杂度,为性能提升和复杂度降低打开全新空间。
结合第三方面或第四方面,在某些实现方式中,第一子网络和第二子网络,是基于以下一项或多项划分的:环境数据信息、感知信号信息、通信设备标识信息、业务类型信息、终端设备的反馈建议信息。
结合第三方面或第四方面,在某些实现方式中,与第一子网络和/或第二子网络相关的信息,承载于以下一项或多项信令中:无线资源控制、媒体介入控制-控制元素、下行控制信息、上行控制信息。
结合第三方面或第四方面,在某些实现方式中,第一子网络对应一个或多个索引。
示例地,预定义或预先配置第一子网络和索引之间的对应关系。
基于上述技术方案,通过获知第一子网络对应的索引,便可以获知第一子网络的信息。例如,通信设备可以根据ID,直接读取该ID所对应的第一子网络。又如,通信设备可以根据ID,向其他设备(如网络设备)申请该ID对应的第一子网络,从而其他设备可以向通信设备发送该ID对应的第一子网络的信息。
结合第三方面或第四方面,在某些实现方式中,第一子网络的更新周期与第二子网络的更新周期不同。
通过该方式,可以提升更新效率,节省更新开销。
一种可能的设计,第一子网络的更新周期比第二子网络的更新周期长。
结合第三方面或第四方面,在某些实现方式中,第一子网络对应多个通信设备;第一子网络由多个通信设备中的部分或全部通信设备训练得到。
通过该方式,可以灵活设计第一子网络的训练。
示例地,第一子网络对应多个终端设备;第一子网络由多个终端设备中的部分或全部终端设备训练得到。
结合第三方面或第四方面,在某些实现方式中,部分或全部通信设备为网络设备指示的通信设备;或者,部分或全部通信设备为基于预设信息确定的通信设备。
示例地,部分或全部终端设备为网络设备指示的终端设备;或者,部分或全部终端设备为基于预设信息确定的终端设备。
一方式,网络设备可以通过x比特(bit)开关形式,指示参与生成第一子网络的终端设备,x为大于1或等于1的整数。
又一方式,网络设备可以发送终端设备列表,指示参与生成第一子网络的终端设备。
结合第三方面或第四方面,在某些实现方式中,第一子网络由多个通信设备中的部分通信设备训练得到,部分通信设备与多个通信设备中的其余通信设备共享第一子网络。
通过该方式,可以灵活设计第一子网络的训练。
示例地,第一子网络由多个终端设备中的部分终端设备训练得到,部分终端设备与多个终端设备中的其余终端设备共享第一子网络。
结合第三方面或第四方面,在某些实现方式中,方法还包括:根据第一子网络对应的索引,获取第一子网络的信息;或者,接收指示信息,根据指示信息,获取第一子网络的信息,其中,指示信息用于指示第一子网络的信息。
结合第三方面或第四方面,在某些实现方式中,公共特征包括以下一项或多项:地理位置信息、时延扩展信息、多普勒分布信息、角度分布信息。
第五方面,提供了一种智能网络的生成方法。该方法可以包括:获取多个子网络的信息,至少两个子网络对应不同种类的特征;基于多个子网络的信息,生成用于通信网络的目标智能网络。
一示例,该方法可以包括:获取第一子网络和第二子网络的信息,第一子网络用于处理信息中的公共特征,第二子网络用于处理信息中的非公共特征;基于第一子网络和第二子网络的信息,生成用于通信网络的目标智能网络。
第六方面,提供了一种智能网络的生成方法。该方法可以包括:获取多个子网络的信息,部分子网络对应多个对象,部分子网络对应一个对象;基于多个子网络的信息,生成用于通信网络的目标智能网络。
一示例,该方法可以包括:获取第一子网络和第二子网络的信息,第一子网络对应多个对象,第二子网络对应一个对象;基于第一子网络和第二子网络的信息,生成用于通信网络的目标智能网络。
第七方面,提供一种信息处理的装置,该装置用于执行上述第一方面至第六方面提供的方法。具体地,该装置可以包括用于执行第一方面至第六方面提供的方法的单元和/或模块,如处理单元和/或通信单元。
在一种实现方式中,该装置为通信设备。当该装置为通信设备时,所述通信单元可以 是收发器,或,输入/输出接口;所述处理单元可以是处理器。
在另一种实现方式中,该装置为用于通信设备中的芯片、芯片***或电路。当该装置为用于通信设备中的芯片、芯片***或电路时,所述通信单元可以是该芯片、芯片***或电路上的输入/输出接口、接口电路、输出电路、输入电路、管脚或相关电路等;所述处理单元可以是处理器、处理电路或逻辑电路等。
可选地,上述收发器可以为收发电路。可选地,上述输入/输出接口可以为输入/输出电路。
第八方面,提供一种信息处理的装置,包括处理器。该处理器与存储器耦合,可用于执行存储器中的指令,以实现上述第一方面至第六方面中的方法。可选地,该装置还包括通信接口,处理器与通信接口耦合,通信接口用于与外界进行数据和/或指令的传输。可选地,该装置还包括存储器。
在一种实现方式中,该装置为通信设备。当该装置为通信设备时,所述通信单元可以是收发器,或,输入/输出接口;所述处理单元可以是处理器。
在另一种实现方式中,该装置为用于通信设备中的芯片、芯片***或电路。当该装置为用于通信设备中的芯片、芯片***或电路时,所述通信单元可以是该芯片、芯片***或电路上的输入/输出接口、接口电路、输出电路、输入电路、管脚或相关电路等;所述处理单元可以是处理器、处理电路或逻辑电路等。
可选地,上述收发器可以为收发电路。可选地,上述输入/输出接口可以为输入/输出电路。
第九方面,提供一种信息处理的装置,该装置包括:存储器,用于存储程序;处理器,用于执行存储器存储的程序,当存储器存储的程序被执行时,处理器用于执行上述第一方面至第六方面提供的方法。
在一种实现方式中,该装置为通信设备。
在另一种实现方式中,该装置为用于通信设备中的芯片、芯片***或电路。
第十方面,本申请提供一种处理器,用于执行上述各方面提供的方法。在执行这些方法的过程中,上述方法中有关发送上述信息和获取/接收上述信息的过程,可以理解为由处理器输出上述信息的过程,以及处理器接收输入的上述信息的过程。在输出上述信息时,处理器将该上述信息输出给收发器,以便由收发器进行发射。该上述信息在由处理器输出之后,还可能需要进行其他的处理,然后才到达收发器。类似的,处理器接收输入的上述信息时,收发器获取/接收该上述信息,并将其输入处理器。更进一步的,在收发器收到该上述信息之后,该上述信息可能需要进行其他的处理,然后才输入处理器。
基于上述原理,举例来说,前述方法中提及的获取通信网络中的信息可以理解为处理器接收输入的指示信息。
对于处理器所涉及的发射、发送和获取/接收等操作,如果没有特殊说明,或者,如果未与其在相关描述中的实际作用或者内在逻辑相抵触,则均可以更加一般性的理解为处理器输出和接收、输入等操作,而不是直接由射频电路和天线所进行的发射、发送和接收操作。
在实现过程中,上述处理器可以是专门用于执行这些方法的处理器,也可以是执行存储器中的计算机指令来执行这些方法的处理器,例如通用处理器。上述存储器可以为非瞬 时性(non-transitory)存储器,例如只读存储器(Read Only Memory,ROM),其可以与处理器集成在同一块芯片上,也可以分别设置在不同的芯片上,本申请实施例对存储器的类型以及存储器与处理器的设置方式不做限定。
第十一方面,提供一种计算机可读存储介质,该计算机可读介质存储用于设备执行的程序代码,该程序代码包括用于执行上述第一方面至第六方面提供的方法。
第十二方面,提供一种包含指令的计算机程序产品,当该计算机程序产品在计算机上运行时,使得计算机执行上述第一方面至第六方面提供的方法。
第十三方面,提供一种芯片,所述芯片包括处理器与通信接口,所述处理器通过所述通信接口读取存储器上存储的指令,执行上述第一方面至第六方面提供的方法。
可选地,作为一种实现方式,所述芯片还可以包括存储器,所述存储器中存储有指令,所述处理器用于执行所述存储器上存储的指令,当所述指令被执行时,所述处理器用于执行上述第一方面至第六方面提供的方法。
图1是适用于本申请实施例的一种通信***的简化示意图。
图2是适用于本申请实施例的一种通信***的简化结构示意图。
图3是适用于本申请实施例的一种通信***的简化示意框图。
图4示出了ELAA结构的一示意图。
图5示出了深度网络模型的一示意性结构图。
图6示出了采用传统方案和神经网络方案对CSI重构的一示意图。
图7示出了特定信道分布下,比较传统方法用于CSI获取与DNN用于CSI获取的性能的示意图。
图8示出了比较面向多分布训练的网络和面向单分布训练的网络的方案的性能的示意图。
图9是本申请一实施例提供的一种信息处理的方法的示意图。
图10是根据本申请实施例提供的目标智能网络的示意图。
图11是本申请另一实施例提供的信息处理的方法的示意图。
图12是本申请又一实施例提供的信息处理的方法的示意图。
图13至图16示出了适用于本申请实施例的子网络可能使用的情况的示意图。
图17示出了适用于本申请实施例的基于地理位置分组的示意图。
图18示出了适用于本申请实施例的目标智能网络处理的示意图。
图19是根据本申请实施例提供的信息处理的装置的示意性框图。
图20是根据本申请实施例提供的信息处理的装置的另一示意性框图。
图21是根据本申请实施例提供的信息处理的装置的又一示意性框图。
图22是本申请实施例提供的一种芯片硬件结构示意图。
下面将结合附图,对本申请中的技术方案进行描述。
为便于理解本申请实施例,首先结合图1至图3详细说明适用于本申请实施例的通信 ***。
如图1所示,作为示例性说明,本申请并不限于此,提供了本申请的实施例应用的一种通信***的简化示意图。该通信***包括无线接入网100。无线接入网100可以是下一代(例如,第六代(6th generation,6G)或更高版本)无线接入网,或传统(例如,第五代(5th generation,5G)、***(4th generation,4G)、第三代(3th generation,3G)或第二代(2th generation,2G))无线接入网。一个或多个通信装置(120a-120j,统称为120)可以相互连接或连接到无线接入网100中的一个或多个网络节点(110a,110b,统称为110)。可选的,通信***包括核心网(未示出)。所述无线接入网设备通过无线或有线方式与核心网连接。所述核心网可以依赖于或独立于无线接入网100中使用的无线接入技术。核心网设备与无线接入网设备可以是独立的不同的物理设备,也可以是将核心网设备的功能与无线接入网设备的逻辑功能集成在同一个物理设备上,还可以是一个物理设备上集成了部分核心网设备的功能和部分的无线接入网设备的功能。终端和终端之间以及无线接入网设备和无线接入网设备之间可以通过有线或无线的方式相互连接。可选的,通信***还可以包括其他网络设备,例如,无线中继设备,或无线回传设备等。
无线接入网100中的一个或多个网络节点110可以是下一代节点、传统节点或其组合。网络节点用于与通信装置和/或其他网络节点通信。本申请中,网络节点有时也称为网络设备。网络节点的非限制性示例是基站(base staion,BS)。本申请中,BS可以广义地使用各种名称中的任一种来引用,例如:gNodeB/gNB、eNodeB/eNB、Node B、6G移动通信***中的下一代基站),未来移动通信***中的基站,Wifi***中的接入点,基地收发机站/基站收发信台(Base Transceiver Station,BTS)、发送接收点(transmission reception point,TRP)、宏基站(MacroeNB,MeNB)、微基站(PicoeNB,SeNB)、多制式无线(Multi-Standard Radio,MSR)无线节点、家庭基站、网络控制器、接入节点、无线节点、接入点(access point,AP)、传输节点、收发节点、基带单元(Base Band Unit,BBU)、射频拉远单元(Remote Radio Unit,RRU)、有源天线单元(Active Antenna Unit,AAU)、射频头(Remote Radio Head,RRH)、集中单元(centralized unit,CU)、分布单元(distributed unit,DU)、定位节点等。基站可以是宏基站、微基站、中继节点、施主节点或类似物,或其组合。基站也可以指内置在上述设备中的装置,例如,上述设备中的通信模块、调制解调器或芯片。本申请的实施例对无线接入网设备所采用的具体技术和具体设备形态不做限定。基站可以支持相同或不同接入技术的网络。为便于描述,下文将网络节点以基站(BS)为例进行说明。
网络节点可以是固定的,也可以是移动的。例如,网络节点110a、110b是静止的,并负责来自通信装置120的一个或多个小区中的无线传输和接收。图1中示出的飞行器(如直升机或无人机)120i可以被配置成充当移动BS,并且一个或多个小区可以根据飞行器120i的位置移动。在其他示例中,直升机或无人机(120i)可以被配置成用作与网络节点110a通信的通信装置。
通信装置120用于连接人、物、机器等,通信装置120可以广泛应用于各种场景,例如蜂窝通信、设备到设备(device-to-device,D2D)、车到物(vehicle to everything,V2X)、端到端(Peer to Peer,P2P)、机器到机器(Machine to Machine,M2M)、机器类型通信(Machine-type Communications,MTC)、物联网(internet of things,IoT)、虚拟现实(virtual reality,VR)、增强现实(augmented reality,AR)、工业控制、自动驾驶、远程医疗、智能电网、智能家具、智能办公、智能穿戴、智能交通、智慧城市无人机、机器人、遥感、被动传感、定位、导航与跟踪、自主交付等场景。通信装置120可以是第三代合作伙伴项目(The 3rd Generation Partnership Project,3GPP)标准的用户设备(user equipment,UE)、固定设备、移动设备、手持设备、可穿戴设备、终端设备,蜂窝电话、智能电话、SIP电话、平板电脑、笔记本电脑、具有无线收发功能的计算机、智能书、车辆、卫星、全球定位***(Global Positioning System,GPS)设备、目标跟踪设备、飞行器(例如无人机、直升机、多直升机、四直升机、或飞机等)、船只、遥控设备智能家居设备、工业设备,或者内置于上述设备中的装置(例如,上述设备中的通信模块、调制解调器或芯片等,为了描述方便,下文将通信装置120以终端,终端设备或UE为例来描述。
在一些实施例中,UE可以用于充当基站。例如,UE可以充当调度实体,其在V2X、D2D或P2P等场景中的UE之间提供侧行链路信号。如图1所示,UE120a和UE120b可以利用侧行链路信号进行通信。UE120a和UE120d之间通信,而无需通过网络节点110a中继通信信号。
在本申请的实施例中,基站的功能也可以由基站中的模块(如芯片)来执行,也可以由包含有基站功能的控制子***来执行。例如,包含有基站功能的控制子***可以是智能电网、工业控制、智能交通、智慧城市等上述终端的应用场景中的控制中心。终端的功能也可以由终端中的模块(如芯片或调制解调器)来执行,也可以由包含有终端功能的装置来执行。
参见图2,作为示例性说明,本申请并不限于此,提供了一种通信***的简化结构示意图。为了简单起见,图2仅示出了网络节点110(例如,BS 110)、通信装置120(例如,UE 120)和网络130。BS 110包括接口111和处理器112。处理器112可选地可以存储程序114。BS 110可选地可以包括存储器113。存储器113可选地可以存储程序115。UE 120包括接口121和处理器122。处理器122可选地可以存储程序124。UE 120可选地可以包括存储器123。存储器123可选地可以存储程序125。这些组件一起工作,以提供本申请中描述的各种功能。例如,处理器112和接口121一起工作以提供BS 110与UE 220之间的无线连接。处理器122和接口121共同作用,实现UE 120的下行传输和/或上行传输。
网络130可以包括一个或多个网络节点130a、130b,以提供核心网功能。网络节点130a、130b可以是下一代(例如,6G或更高版本)核心网节点,或传统(例如,5G、4G、3G或2G)核心网节点。例如,网络130a、130b可以是接入和移动性管理功能(Access and Mobility Management Function,AMF)、移动性管理实体(mobility management entity,
MME)等。网络130还可以包括公共交换电话网络(Public Switched Telephone Network,PSTN)、分组数据网络、光网络、IP网络中的一个或多个网络节点、广域网(Wide Area Network,WAN)、局域网(Local Area Network,LAN)、无线局域网(Wireless Local Area Network,WLAN)、有线网络、无线网络、城域网和其他网络,以便使UE 120和/或BS110之间能够进行通信。
处理器(例如,处理器112和/或处理器122)可包括一个或多个处理器并实现为计算设备的组合。处理器(例如,处理器112和/或处理器122)可分别包括以下一种或多种: 微处理器、微控制器、数字信号处理器(digital signal processor,DSP)、数字信号处理设备(digital signal processing device,DSPD)、专用集成电路(application specific integrated circuit,ASIC)、现场可编程门阵列(field programmable gate array,FPGA)、可编程逻辑器件(programmable logic device,PLD)、选通逻辑、晶体管逻辑、分立硬件电路、处理电路或其它合适的硬件、固件和/或硬件和软件的组合,用于执行本公开中所描述的各种功能。处理器(例如,存储器112和/或存储器122)可以是通用处理器或专用处理器。例如,处理器112和/或处理器122可以是基带处理器或中央处理器。基带处理器可用于处理通信协议和通信数据。中央处理器可用于使BS 110和/或UE 120执行软件程序,并处理软件程序中的数据。此外,处理器的一部分还可以包括非易失性随机存取存储器。例如,处理器还可以存储设备类型的信息。
接口(例如,接口111和/或121可包括用于实现与一个或多个计算机设备(例如,例如,在一些实施例中,接口可以包括用于耦合有线连接的电线或用于耦合无线收发器用于无线连接的终端和/或管脚。在一些实施例中,接口可以包括发射器、接收器、收发器和/或天线。接口可以被配置为使用任何可用的协议(例如,3GPP标准),用于计算机设备之间的通信(例如,UE、BS和/或网络节点)。
本申请中的程序在广义上用于表示软件。软件的非限制性示例包括:程序代码、程序、子程序、指令、指令集、代码、代码段、软件模块、应用程序、或软件应用程序等。程序可以在处理器和/或计算机中运行。以使得BS 110和/或UE 120执行本申请中描述的各种功能和/或过程。
存储器(例如,存储器113和/或存储器123可存储供处理器112、122在执行软件时所需的数据。存储器113、123可以使用任何合适的存储技术实现。例如,存储器可以是处理器和/或计算机能够访问的任何可用存储介质。存储介质的非限制性示例包括:随机存取存储器(random access memory,RAM)、只读存储器(read-only memory,ROM)、电可擦除可编程只读存储器(electrically EPROM,EEPROM)、光盘只读存储器(Compact Disc-ROM,CD-ROM)、静态随机存取存储器(static RAM,SRAM)、动态随机存取存储器(dynamic RAM,DRAM)、同步动态随机存取存储器(synchronous DRAM,SDRAM)、双倍数据速率同步动态随机存取存储器(double data rate SDRAM,DDR SDRAM)、增强型同步动态随机存取存储器(enhanced SDRAM,ESDRAM)、同步连接动态随机存取存储器(synchlink DRAM,SLDRAM)和直接内存总线随机存取存储器(direct rambus RAM,DR RAM)、可移动介质、光盘存储器、磁盘存储介质、磁存储设备、闪存、寄存器、状态存储器、远程挂载存储器、本地或远程存储器组件,或能够携带或存储软件、数据或信息并可由处理器/计算机访问的任何其它介质。需要说明的是,本文描述的存储器旨在包括但不限于这些和任意其它适合类型的存储器。
存储器(例如,存储器113和/或存储器123和处理器(例如,处理器112和/或处理器122)可以分开设置或集成在一起。存储器可以用于与处理器连接,使得处理器能够从存储器中读取信息,在存储器中存储和/或写入信息。存储器113可以集成在处理器112中。存储器123可以集成在处理器122中。存储器(例如,存储器113和/或存储器123)和处理器(例如,处理器112和/或处理器122)可以设置在集成电路中(例如,该集成电路可以设置在UE或BS或其他网络节点中。
根据一些实施例,本申请中描述的各种功能可以由装置(例如,UE、BS和/或任意网络节点),其包括用于执行各个功能的一个或多个模块、组件、电路、软件、元件等(统称为元件)。这些元件可以使用硬件、软件、固件和/或它们的组合来实现。
参见图3,作为示例性说明,本申请并不限于此,提供了一种通信***的简化示意框图。该无线通信***可以包括至少一个网络设备,例如图3所示的网络设备311(可以是图1和图2中的网络节点(如BS 110),该无线通信***还可以包括至少一个终端设备,例如图3所示的终端设备321(可以是图1和2中的通信装置(如UE 120)。网络设备和终端设备均可配置多个天线,网络设备与终端设备可使用多天线技术通信。
示例地,网络设备和终端设备(如网络设备311和终端设备321)可以通过无线资源控制(radio resource control,RRC)信令交互模块收发RRC信令。示例地,网络设备和终端设备(如网络设备311和终端设备321)可以通过媒体介入控制(media access control,MAC)信令交互模块收发媒体介入控制-控制元素(media access control-control element,MAC-CE)信令。示例地,网络设备和终端设备(如网络设备311和终端设备321)可以通过物理(physical layer,PHY)层收发信令和数据,例如,上行控制信令(uplink control information,UCI)(如物理上行控制信道(physical uplink control channel,PUCCH))、下行控制信令(downlink control information,DCI)(如物理下行控制信道(physical downlink control channel,PDCCH))、上行数据(如物理上行共享信道(physical uplink shared channel,PUSCH))、以及下行数据(如物理下行共享信道(physical downlink shared channel,PDSCH))。
本申请实施例的技术方案可以应用于无线通信网络中的波形***中,例如:单载波、正交频分复用(orthogonal frequency division multiplexing,OFDM)或其他波形***中。本申请实施例架构可与现在标准中的通信架构相同,如传统的单载波或OFDM***,也可以为其他不同的通信架构,对此不作限定。例如,以现有通信架构为例,在发送端(如网络设备311)对信号做一系列处理,如对信号做离散傅里叶变换(discrete fourier transformation,DFT)操作,对DFT后的信号做子载波映射,并进行快速傅里叶逆变换(inverse fast fourier transform,IFFT),添加循环前缀(cyclic prefix,CP)。相应地,在接收端(如终端设备321)进行逆操作处理,如去CP、DFT、子载波解映射/均衡、以及IFFT的过程。
此外,本申请实施例的技术方案可以适用于网络架构中的任意基于智能网络模块的学习。如基于智能网络(如神经网络,决策树、深度森林等网络)实现信道估计、信号检测、预编码设计、编码解码、调制解调等。
为便于理解本申请实施例,对本申请实施例中涉及的几个基本概念做简单说明。应理解,下文中所介绍的基本概念是以目前协议中规定的基本概念或者目前所使用的基本概念为例进行简单说明,其名称及具体概念,不对本申请实施例的保护范围造成限定。
1、通用人工智能(artificial general intelligence,AGI)
AGI:具有通用性的人工智能(artificial intelligence,AI)。AGI不仅具有AI的一些功能,如既有的知识表达、推理和决策等关键功能,还具有快速学习泛化能力,即智能体基于行为动作间的逻辑映射关系,能够有效构建自己的知识体系,从而形成经验以适应环境进而生存。
传统的人工智能具备两个关键能力:感知和决策。感知,即能够感受并了解环境的特征。决策,即能够基于目标最大化成功概率而采取行动。但随着神经网络的应用领域日益广泛,人们越来越关注神经网络的普适性以及泛化能力。机器智能体自发构建特有的知识体系,进而形成经验以适应环境。换句话说,试图让人工智能能够基于同一套(或者说同一类)算法,学习处理各式任务,而不是任何单一任务都从头训练进而生存。这便是AGI课题研究的主要目的。
目前,可能实现AGI的几个可行方向简单分类如下。
广义强化学习:广义强化的形式体现了人类如何获取信息并转化为知识和技能的过程,并最终形成一种被固化和保留下来的经验常识,进而形成高级语义信息被重复利用和传播。
元学习(meta learning):即学会学习(learning to learn)。元学习在本质上提供了一种提取不同信息相互联系的方法,而不仅仅是提供一种方式来学会学习。
迁移学习(transfer learning):人工智能在真实布网中网络训练的耗时和功耗将是其应用瓶颈之一,因而网络迁移重用将会是应用过程中的关键之一。这就要求网络模块具有自迁移的功能,可以使能近似场景中进行快速部署和应用。
分级结构模型(hierarchy model):针对神经网络在网络结构层面进行改进,其分层不仅仅是在时间尺度上。神经网络模块化作为一种趋势,期望的神经网络为一个神经网络能够同时处理语音、视觉、控制等各种类型输入,模块化类似于大脑不同功能区域,其间很多信息可以共用和迁移。
2、极大规模天线阵列
伴随着广覆盖和大容量已成为现代通信***的主要特征,能够显著提高***容量的多输入多输出(multi-input multi-output,MIMO)技术在多种无线解决方案中脱颖而出。大规模的多输入多输出(massive MIMO)作为更大维度的多天线***,利用空间维度的资源,可以在不增加***带宽的前提下,使信号在空间获得阵列增益、复用和分集增益以及干扰抵消增益,成倍地提升通信***的容量和频谱效率。
极大规模天线阵列(extremely large antenna array,ELAA)是天线规模进一步演进的重要方向。在该ELAA***中,区别于传统多天线形态,ELAA天线布置在极大一块区域内,如商场和体育馆的外墙等。此时,ELAA与传统MIMO最大的区别之一便是终端的天线可视域不再是整个天面,部分终端可能看到的只是某一部分天面,进而不同终端组所经历的多径簇可能也不再相同。作为示例性说明,图4示出了ELAA结构的一示意图。在图4所示的示例中,终端的天线可视域不再是整个ELAA,终端可能看到的只是某一部分天面,即多径簇(cluster),不同终端组所经历的多径簇不同。
3、深度神经网络
机器学习的众多研究方向中,神经网络算法由于其具有的通用近似定理(universal approximation theorem)所赋予的无限逼近任意连续函数的能力,成为极有潜力的一种技术方法。
深度神经网络(deep neural network,DNN)一般为多层结构。增加神经网络的深度和宽度都可以提高它的表达能力,为复杂***提供更强大的信息提取和抽象建模能力。
经典的深度神经网络包括:全连接网络(fully connected network,FCN)、旨在对时 间序列变化进行建模的递归神经网络(recurrent neural network,RNN)、以及模拟人的视觉神经***提出的卷积深度神经网络(convolutional neural network,CNN)。作为示例性说明,图5示出了深度网络模型的一示意性结构图。
深度网络模型可以包括输入层、隐含层、以及输出层。如图5所示,标1的圆框表示输入层,标2的圆框表示隐含层,标3的圆框表示输出层。输入层的作用包括输入待处理的信息。DNN中一般含有多于1个隐含层,隐含层的作用包括不同程度地提取信息特征。输出层的作用包括从提取的特征信息中映射出所需要的输出信息。圆形标志为各层神经元示意,每层的神经元连接方式和采用的激活函数将决定神经网络的表达函数。
应理解,图5仅是一种示例性说明,对此不作限定。实际操作中,不同问题适用的深度网络模型也不尽相同。可以综合模型性能和复杂度选取合适的深度网络模型。
DNN现在已广泛应用于图像、语音和视频处理领域。基于其在特征提取和挖掘上的优势,有不少学术研究将目光转移到深度学习网络在无线通信物理层(如MIMO)中的应用上。如深度学习网络在信道估计、MIMO信道获取、MIMO检测、预编码设计以及参考信号上,都有不少研究。
以MIMO信道获取为例,最近几年中,学术界已尝试多种基于特征提取、压缩、解压缩、特征恢复等步骤实现的信道状态信息(channel state information,CSI)的压缩重构方案。其主要思路为:把CSI信息视做图像,依据神经网络提取能力对该图像进行特征提取压缩以及传递。作为示例性说明,图6示出了采用传统方案和神经网络方案对CSI重构的一示意图。
如图6所示,在CSI重构过程中,如采用传统方案,主要是对信道在角度时延域的部分典型角度时延对(particular angle-delay pairs)进行处理,如包括压缩(compress)和反馈(feedback)处理。采用神经网络方案,如DNN,可以对角度时延信息进行高维特征转化,在更高维度上进行压缩和处理。
然而,如果神经网络算法对模型不做理解(即类似于黑盒子),也就是说,除了输入和输出有明确的要求外,网络内部各隐层的特征转换无法获知。在更换特征的情况下,如更换场景、更换压缩率、更换量化算法等,需要重新进行整体网络训练以匹配最新特征需求。
将DNN用于无线通信场景中,可以包括以下两种方案:
一种是面向单分布训练的网络(如记为UE特定(UE-specific)网络),即针对某一种分布进行精确的特征提取。作为示例性说明,图7示出了特定信道分布下,比较传统方法用于CSI获取与DNN用于CSI获取的性能的示意图。如图7所示,在某种特定信道分布下,比较传统方法(如3GPP标准版本16(Release 16,R16))用于CSI获取与DNN用于CSI获取(DNN-based CSI)的性能。图7中的横坐标表示信噪比(signal-noise ratio,SNR),纵坐标表示吞吐量(throughput)。从图7可知,在某种特定信道分布下,DNN-based CSI获取相比于传统方法,有极大优势。
另一种是面向多分布训练的网络,即通用(generalized)网络,该网络特征是基于预设的一个网络,在现网中配置于每个终端,此时不再需要用户特定的网络训练。
然而,关于面向单分布训练的网络,意味着无线通信中,若不计代价,可以针对某一UE的特定分布(如某一时刻某一种信道场景)做针对性的网络和算法设计,且能获得较 优性能。此外,脱离特定分布后性能可能会急剧恶化。然而综合考虑无线通信场景的特殊性,如低时延要求,更严格的功耗和成本要求,以及用户终端的运算能力受限等因素。UE特定的DNN网络设计思路难以成行。
关于面向多分布训练的网络,即通用网络,成本功耗都会大幅降低,且也能满足低时延要求。但是此时网络需要面对的是各式各样的信道场景。该网络的推理性能将会因为面对样本的多样性而急剧恶化。
以DNN用于CSI获取为例(或者也可以记为AI-CSI),图8示出了上述两种方案的性能。图8中的(1)对应面向多分布训练的网络(即通用网络)的方案,图8中的(2)对应面向单分布训练的网络的方案。其中,图8中的各种类型(type)的特征如表1所示。在机器学习中,一般将样本分成三部分:训练集(train set)、验证集(validation set)与测试集(test set)。其中,训练集用于建立模型,测试集用来检验最终选择最优的模型的性能如何。
表1
其中,表1中的CDL表示集群延迟线(cluster delay line,CDL)。A表示信道类型,B也表示信道类型。数字“100”或“300”均表示平均时延大小。
从图8可知,面向多分布训练的网络(即通用网络)和面向单分布训练的网络,前者(即面向多分布训练的网络)实现简单,但相比于后者(即面向单分布训练的网络)有近20%的性能损失;后者性能很好但难以实现。对于终端侧需要参与学习的通信问题,尤其是端到端学习问题,现网布置困难较大,瓶颈之一是用户终端运算能力。
上述两种方案均很难平衡性能和复杂度。
本申请实施例提供一种方案,可以很好的平衡性能和复杂度,更加适合无线通信特殊需求(如场景复杂多样,低时延、高可靠等),可以提升智能网络用于通信网络中的可实现性。
下面将结合附图详细说明本申请提供的各个实施例。
图9是本申请实施例提供的一种信息处理的方法900的示意图。方法900可以包括如下步骤。
910,获取通信网络中的信息。
920,用目标智能网络对信息进行处理,目标智能网络包括多个子网络,至少两个子 网络对应不同种类的特征。
在本申请实施例中,将通信网络中的信息,输入至目标智能网络,由目标智能网络对这些信息进行处理。或者说,使用目标智能网络来捕获或处理或抓取通信网络中的特征。例如,将通信网络中的信息视做图像,依据目标智能网络提取能力对该图像进行特征提取处理。
智能网络,可以称为智能模块、模型、“黑盒子”,可以表示具有机器学习功能的网络,如人工智能网络。为区分,将本申请实施例提供的用于通信网络中的智能网络记为目标智能网络。
作为示例性说明,图10示出了提供了根据本申请实施例提供的目标智能网络的示意图。
目标智能网络包括多个子网络,如图10(1)所示,目标智能网络可以包括子网络L1、L2、……、Li,又如图10(2)所示,目标智能网络可以包括子网络L1、L2、……、Lk。
至少两个子网络对应不同种类的特征,或者也可以表示为,至少两类子网络对应不同种类的特征,或者说至少两部分子网络对应不同种类的特征。至少两个子网络对应不同种类的特征,其表示,子网络捕获或处理其对应的特征,或者说不同种类的特征通过其相对应的子网络进行处理。
例如,如图10(1)所示,L1和Li对应不同种类的特征,L2和Li对应不同种类的特征,L1和L2可以对应同一种类的特征。或者,也可以理解为,L1和L2作为同一类子网络,Li作为另一类子网络,这两类子网络对应不同种类的特征。L1和L2处理同一类特征,Li处理另一类特征;或者说,一类特征由L1和L2处理,另一类特征由Li处理;或者说,L1和L2处理部分特征,Li处理另一部分特征。
又如,如图10(2)所示,L1和Lk对应不同种类的特征,L2和Lk对应不同种类的特征,L1和L2可以对应同一种类的特征。或者,也可以理解为,L1和L2作为同一类子网络,Lk作为另一类子网络,这两类子网络对应不同种类的特征。L1和L2处理同一类特征,Lk处理另一类特征;或者说,一类特征由L1和L2处理,另一类特征由Lk处理;或者说,L1和L2处理部分特征,Lk处理另一部分特征。
本申请提供的目标智能网络,可以认为是一种分级网络架构。目标智能网络包括多个子网络,至少两个子网络对应不同种类的特征。一方面,在无线通信网络中,尽管终端设备能力参差不齐,然而终端设备的业务特点可能类似,甚至有些终端设备之间不同特征具有分组或者分簇特性。对于不同组的终端设备来说,特征差距较大;对于组内的终端设备来说,特征差距相对较小。另一方面,对智能网络(如神经网络)本身而言,智能网络可以实现对特征分类处理。如计算机视觉中深度神经网络模型较低的卷积层捕获的是低级图像特征,例如边缘信息,颜色等,这样的特征在不同的分类任务中几乎是不变的,真正区别的是高层特征,较高的卷积层捕获越来越复杂的细节,例如身体部位、面部和其他组合性特征。
因此,在本申请中,利用无线通信网络的特征分组以及智能网络的分类处理特性,设计分级机制的网络架构。通过该方案,可以基于各个子网络对应的特征进行灵活的训练和使用。例如,对于多个对象来说,该多个对象之间相似的特征,可以使用相同的子网络进行处理,从而可以降低复杂度,节省训练开销。对于多个对象来说,各个对象的特定特征, 可以使用不同的子网络进行处理,如各个对象各自训练更新或设计该类特征对应的子网络,从而可以尽可能地保证性能。因此,基于该方案,通过特征分层,既能有效把握和提取各个级别的特征,以保证训练性能,还能实现训练开销的最小化,提升训练效率。
不同种类的特征,可以包括不同对象之间相似程度不同的特征。如不同种类的特征,可以包括不同通信设备(如终端设备或网络设备)之间相似程度不同的特征,或者也可以包括不同通信场景下(如不同信道场景或者不同分布下)相似程度不同的特征,或者也可以包括不同处理对象下相似程度不同的特征。关于处理对象,下文介绍。
例如,以图4所示的示例为例,通过目标智能网络对CSI信息进行处理时,可以将该CSI信息的角度和时延作为一类特征,该类特征对应一子网络,即由同一子网络对角度和时延信息进行处理;或者也可以将该CSI信息的为两类特征,即由一子网络对角度信息进行处理,由另一子网络对时延信息进行处理。
应理解,当通过目标智能网络处理通信网络中的信息时,对该信息的特征如何进行分类,不作严格限定。在不同场景下,特征的分类依据可能不同,同样的特征在不同场景下也可能不同。
作为示例而非限定,不同种类的特征可以包括公共特征和非公共特征。在该示例下,目标智能网络中可以包括多个子网络,一部分子网络对应公共特征,另一部分子网络对应非公共特征。
下面介绍公共特征和非公共特征。
1、公共特征
公共特征,例如也可以称为共有的特征。
公共特征,可以表示该类特征不随通信设备的不同而不同,或者说不同通信设备之间该类特征的变化不明显或区别不大。例如,对于某一场景中的多个通信设备来说,该多个通信设备之间相同或相似的特征,或者变化不明显或区别不大的特征,可以认为是公共特征。
或者,公共特征,可以表示该类特征不随通信场景的不同(如信道分布的不同)而不同,或者说不同通信场景之间该类特征的变化不明显或区别不大。例如,通信设备处于多种信道分布中,该通信设备在该多种信道分布下相同或相似的特征,或者变化不明显或区别不大的特征,可以认为是公共特征。
或者,公共特征,可以表示该类特征不随处理对象的不同而不同,或者说不同处理对象之间该类特征的变化不明显或区别不大。例如,对于某一场景中的多个处理对象来说,该多个处理对象之间相同或相似的特征,或者变化不明显或区别不大的特征,可以认为是公共特征。
2、非公共特征
非公共特征,例如也可以称为私有特征或特定特征。
非公共特征,可以表示该类特征随通信设备的不同而不同,或者说不同通信设备之间该类特征的变化比较明显或区别较大。例如,对于某一场景中的多个通信设备来说,对于该多个通信设备之间不同的特征,或者专属于通信设备的特定特征(或细节特征),或者变化比较明显或区别较大的特征,可以认为是非公共特征。
或者,非公共特征,可以表示该类特征随通信场景的不同而不同,或者说不同通信场 景之间该类特征的变化比较明显或区别较大。例如,通信设备处于多种信道分布中,对于该通信设备在该多种信道分布下不同的特征,或者专属于信道分布的特定特征(或细节特征),或者变化比较明显或区别较大的特征,可以认为是非公共特征。
或者,非公共特征,可以表示该类特征随处理对象的不同而不同,或者说不同处理对象之间该类特征的变化比较明显或区别较大。例如,对于某一场景中的多个处理对象来说,对于该多个处理对象之间不同的特征,或者专属于处理对象的特定特征(或细节特征),或者变化比较明显或区别较大的特征,可以认为是非公共特征。
应理解,上述公共特征和非公共特征仅是一种为区分做的命名,其名称不对本申请实施例的保护范围造成限定。
此外,还应理解,本申请中提及的公共特征和非公共特征,是一个相对的概念,或者说是在某一特定情景下的划分,在不同情景下还会相对变化,其并不限定特定的特征(或者说特定的参数)。
一可能的情况,同样的特征(或者说对于信息来说同一个参数,如角度时延)在不同场景下的分类不同。例如,在场景一下处理通信设备1和通信设备2的信息时,通信设备1和通信设备2的信息中的有些特征可以认为是公共特征,有些特征可以认为是非公共特征;在场景二下处理通信设备1和通信设备2的信息时,属于公共特征的特征可能不同于场景一下属于公共特征的特征,属于非公共特征的特征可能不同于场景一下属于非公共特征的特征。假设在某一场景下,不同通信设备对应的时域变化很明显,频域变化不明显,那么在该场景下,通信设备对应的频域特征可以认为是公共特征,通信设备对应的时域特征可以认为是非公共特征。假设在另一场景下,不同通信设备对应的频域变化很明显,时域变化不明显,那么在该场景下,通信设备对应的时域特征可以认为是公共特征,通信设备对应的频域特征可以认为是非公共特征。
又一可能的情况,对于不同的通信设备,同样的特征在同一场景下的分类不同。例如,在某一场景下处理通信设备1和通信设备2的信息时,通信设备1和通信设备2的信息中有些特征可以认为是公共特征,有些特征可以认为是非公共特征;在同样场景下处理通信设备3和通信设备4的信息时,通信设备3和通信设备4的公共特征和非公共特征,可能不同于通信设备1和通信设备2的公共特征和非公共特征。
以图4为例,在某种场景下,对UE组1中的多个UE的信息进行处理时,由于组内的角度时延分布差距较小,故角度时延可以认为是公共特征,即用第一子网络可以处理该多个UE的角度时延信息。在某种场景下,对UE组1中的UE和UE组2中的UE的信息进行处理时,由于不同组角度时延分布差距较大,故角度时延可以认为是非公共特征,即用一第二子网络处理UE组1中的UE的角度时延信息,用另一个第二子网络处理UE组2中的UE的角度时延信息。
应理解,上述仅是示例性说明,对此不作限定。当使用目标智能网络处理无线通信中的信息时,通过多个子网络分工处理该信息的不同特征的方案,都落入本申请实施例的保护范围。
还应理解,子网络,仅是为描述做的示例性说明,其命名不对本申请实施例的保护范围造成限定。例如,子网络也可以称为网络,子网络也可以称为模块,或者子网络也可以称为组件,或者子网络也可以称为层等等。可以理解,能够表示与子网络相同或相似含义 的命名,都适用于本申请实施例。下文,为统一,以子网络为例进行示例性说明。
还应理解,在实际应用中,可以基于特征的相似程度划分成多类特征。以不同通信设备之间的特征为例,不同种类的特征例如可以包括三类特征,分别记为第一类特征、第二类特征、第三类特征。其中,第一类特征的相似程度最大,如不同通信设备之间的特征的区别最小的特征属于第一类特征;第三类特征的相似程度最小,如不同通信设备之间的特征的区别最大的特征属于第三类特征;第二类特征的相似程度位于第一类特征和第二类特征之间。第一类特征对应一子网络,第二类特征对应又一子网络,第三类特征对应另一子网络。在实际使用时,各类特征对应的子网络的更新周期可以不同。例如,第一类特征对应的子网络更新周期可能最长,第三类特征对应的子网络更新周期可能最短。
还应理解,上述基于特征的相似程度划分成多类特征,如划分为公共特征和非公共特征,仅是示例性说明,对此不作限定。在未来,基于其他因素将特征进行划分的方式都落入本申请实施例的保护范围。例如,可以将特征划分为在时域、空域、频域以及时延域中至少一个域的特征。
还应理解,本申请实施例对于特征的划分不作严格限定。例如,可以通过独立于子网络的一个或多个模块来执行特征划分或定义,如在信息输入时,通过该模块可以自动分类,并将不同类的特征,分别输入至对应的子网络。又如,也可以是子网络分别提取各自对应的特征,如信息整体输入神经网络,各个子网络有能力提取各自对应的特征进行处理。对此不作限定。
还应理解,在实际应用中,目标智能网络可以包括两个或两个以上的子网络。
还应理解,各个子网络可以认为是一个独立的网络,即目标智能网络可以是一个为描述定义的概念,也就是说将各个子网络认为是独立的子网络,该多个独立的子网络认为组成了一个目标智能网络。或者,各个子网络也可以认为是目标智能网络的部分网络。
还应理解,目标智能网络可以针对不同处理对象进行设计,如针对不同处理对象定义、构建和训练。处理对象例如可以是:信道状态信息处理、编解码处理、调制解调处理、信道编解码,等等。那么,相应地,可以有对应CSI处理的目标智能网络,也可以有对应编解码处理的目标智能网络,也可以有对应调制解调处理的目标智能网络,等等。一个处理对象可以对应一个或多个目标智能网络。终端侧和网络侧可以分别有各自对应的目标智能网络,终端侧和网络侧也可以有共同的目标智能网络。
此外,上述关于不同处理对象,主要是结合传统无线通信网络进行了示例性说明,对此不作限定,本申请提供的目标智能网络可以对应或者说用于当前或未来无线通信网络中的任何处理对象。示例地,处理对象的定义可能不同于现有分类方式,如某处理对象可能是CSI处理、编解码处理、调制解调处理、信道编解码、信道估计、数据检测等对象的任意组合。例如,在未来无线通信网络中,接收端可能会通过一个模块处理从信号接收到译码的所有过程,在该情况下,本申请提供的目标智能网络可以对应上述过程所代表的新的处理对象。
还应理解,目标智能网络的构建形式可以是神经网络或其他智能学习架构,对此不作限定。例如,目标智能网络可以通过前馈神经网络、递归网络等架构所构建的包括感知机、卷积神经网络、循环神经网络等神经网络中的一种或多种(如以某种组合方式组合)来实现。又如,目标智能网络也可以通过非神经网络,如决策树、深度森林等网络来实现。
上述方法900主要是从特征的角度进行了描述。此外,也可以用子网络所应用的对象的角度进行描述,下面结合图11所示的方法1100进行示例性说明。
1110,获取通信网络中的信息。
1120,用目标智能网络对信息进行处理,目标智能网络包括多个子网络,部分子网络对应多个对象,部分子网络对应一个对象。
在本申请实施例,“对应”,可以理解为关联或者说绑定。例如,用于某一对象的子网络,那么可以认为该对象和该子网络对应。
在本申请实施例中,“部分子网络”可以包括一个或多个子网络。
示例地,“对象”,可以表示通信设备(如终端设备或者也可以是网络设备),或者也可以表示通信场景,或者也可以表示其他对象,对此不作限定。例如,对象也可以为上文介绍目标智能网络时所述的处理对象。当对象为处理对象时,其表示的是,对于不同处理对象,分别设计各个处理对象对应的目标智能网络,该多个处理对象对应的多个目标智能网络中可以有部分子网络是相同的,可以理解,该部分子网络可以参与处理该各个处理对象的部分特征。以第一处理对象和第二处理对象为例,针对第一处理对象设计的目标智能网络与针对第二处理对象设计的目标智能网络中,部分子网络可以共用,该部分子网络不仅可以参与处理第一处理对象的部分特征,还可以参与处理第二处理对象的部分特征。
子网络对应一对象,即表示该子网络可以用于该对象,或者说该子网络可以参与处理该对象的信息,或者说该子网络用于处理该对象的某些特征。
以对象表示通信设备为例。子网络对应多个通信设备,其可以表示,该多个通信设备均可以使用该子网络,或者说该子网络可以被多个通信设备共享,或者说该子网络可以参与处理该多个通信设备的信息,或者说该子网络可以用来处理该多个通信设备的某些特征。子网络对应一个通信设备,即表示该子网络仅用于一个通信设备,或者说该子网络关联一个通信设备,或者说该子网络专属于一个通信设备,或者说该子网络仅用于处理该一个通信设备的某些特征。
以图10为例,假设L1和L2对应多个对象,Lk和Li分别对应一个对象。
以通信设备为例,L1和L2可以被多个通信设备共用,Lk对应一个通信设备,Li对应另一个通信设备。以通信设备1和通信设备2为例,图10(1)所示的目标智能网络为对应一个通信设备1的目标智能网络,图10(2)所示的目标智能网络为对应通信设备2的目标智能网络。该两个目标智能网络中部分子网络相同,如子网络L1和L2,即通信设备1和通信设备2可以共享子网络L1和L2。针对不同的通信设备,可以有特定的子网络。如针对通信设备1,单独训练特定子网络Li;针对通信设备2,单独训练特定子网络Lk。
以处理对象为例,L1和L2可以被多个处理对象共用,Lk对应一个处理对象,Li对应另一个处理对象。以编解码处理和调制解调处理为例,图10(1)所示的目标智能网络为对应编解码处理的目标智能网络,图10(2)所示的目标智能网络为对应调制解调处理的目标智能网络。该两个目标智能网络中部分子网络相同,如子网络L1和L2,即解码处理和调制解调处理可以共享子网络L1和L2。针对不同的处理对象,可以有特定的子网络。如针对编解码处理,单独训练特定子网络Li;针对调制解调处理,单独训练特定子网络Lk。
在本申请实施例中,提出了一种分级网络架构。在本申请实施例中,考虑到对象组之 间的对象有一定相似或相近的特征,以及智能网络(如神经网络)内在的分类处理特性,因此,设计分级机制的网络架构,从而在保证性能的前提下可以大幅降低网络开销。例如,对于某一对象的目标智能网络来说,该对象的目标智能网络包括多个子网络。该多个子网络中的一部分子网络不仅可以用于该对象,还可以用于其他一个或多个对象,该多个子网络中的又一部分子网络仅用于该对象。因此,针对每个对象,不需要单独去设计完整的目标智能网络,可以节省训练成本和开销。此外,针对每个对象,有特定的子网络,那么可以基于每个对象的特点,单独训练特定的子网络,从而也可以尽可能保证网络性能和实时性需求。
关于如何划分子网络以及如何划分不同对象的方式,下文详细介绍。
应理解,方法1100和方法900可以独立使用,也可以结合使用。例如,在结合使用时,部分子网络对应不同对象,可以表示该部分子网络可以用于处理该不同对象的公共特征;部分子网络对应一个对象,可以表示该部分子网络可以用于处理该对象的非公共特征。
为便于理解,下文主要以两类子网络,分别记为第一子网络和第二子网络,为例进行示例性说明。
应理解,第一子网络和第二子网络,仅是为区分做的示例性说明,其命名不对本申请实施例的保护范围造成限定。例如,第一子网络还可以称为公共子网络,第二子网络还可以称为私有子网络或特定子网络。可以理解,能够表示与第一子网络和第二子网络相同或相似含义的命名,都适用于本申请实施例。下文,为描述,以第一子网络和第二子网络为例进行示例性说明。
图12是本申请实施例提供的一种信息处理的方法1200的示意图。方法1200可以包括如下步骤。
1210,获取通信网络中的信息。
1220,用目标智能网络对信息进行处理,目标智能网络包括第一子网络和第二子网络。第一子网络用于处理信息中的公共特征,第二子网络用于处理信息中的非公共特征;和/或,第一子网络对应多个对象,第二子网络对应一个对象。
在本申请中,第一子网络用于表示目标智能网络中用来处理信息中的公共特征的子网络,和/或,第一子网络用于表示对应一个对象的子网络。应理解,第一子网络并不限定子网络的数量为一个,例如第一子网络可以包括一个或多个子网络。如图10所示,L1和L2均可以认为是第一子网络。
类似地,第二子网络用于表示目标智能网络中用来处理信息中的非公共特征的子网络,和/或,第二子网络用于表示对应多个对象的子网络。应理解,第二子网络并不限定子网络的数量为一个,例如第二子网络可以包括一个或多个子网络。如图10(1)所示的目标智能网络中,Li可以认为是一个第二子网络,如图10(2)所示的目标智能网络中,Lk可以认为是一个第二子网络。
关于公共特征和非公共特征、子网络、以及对象的相关描述,参考方法900和方法1100中的描述,此处不再赘述。
在本申请实施例中,提出了一种分级网络架构。例如,目标智能网络可以包括第一子网络和第二子网络,第一子网络对应公共特征,第二子网络对应非公共特征;和/或,第一子网络对应多个对象,第二子网络对应一个对象。第一子网络对应的为公共特征,如公 共环境所对应的特征,不同对象可以复用该第一子网络,即其变化频率不需要太高,因此可以降低复杂度和成本。第二子网络对应非公共特征,不同对象可以训练更新第二子网络,即其可以以更短周期执行训练更新,因此可以尽可能地保证性能。通过该方式,可以有效地平衡性能和复杂度,为性能提升和复杂度降低打开全新空间。
下面从几个方面详细介绍本申请实施例。下文,为便于理解,主要以对象为通信设备和通信场景为例进行示例性说明。
方面1,子网络可能使用的情况。
以第一子网络和第二子网络为例,第一子网络和第二子网络,例如可以属于以下任一情况。
情况1,第一子网络对应多个通信设备,第二子网络对应一个通信设备。
在该情况下,例如,对于某一通信设备来说,该通信设备对应的目标智能网络包括第一子网络和第二子网络,该通信设备可以复用其他通信设备对应的第一子网络,更新训练该通信设备对应的第二子网络即可。
图13示出了一具体示例。假设第一子网络对应第一通信设备和第二通信设备。如图13所示,第一通信设备对应的目标智能网络包括第一子网络和第二子网络#1,第二通信设备对应的目标智能网络包括第一子网络和第二子网络#2,可以看出,第一通信设备和第二通信设备可以共用第一子网络。
可选地,该第一子网络可以用于处理对于第一通信设备和第二通信设备来说的公共特征。例如,如果第一通信设备和第二通信设备属于如图4所示的UE组1中的终端设备,该第一子网络可以用于处理第一通信设备对应的CSI信息中的角度时延,该第一子网络还可以用于处理第二通信设备对应的CSI信息中的角度时延。
基于该情况1,多个通信设备可以共用一个第一子网络,进而降低复杂度。此外,每个通信设备可以有各自对应的第二子网络,从而可以根据实际情况训练更新第二子网络,以便可以尽可能地保证性能。
情况2,第一子网络对应不同通信场景下的一个通信设备,第二子网络对应一种通信场景下的一个通信设备。
在该情况下,例如,对于处于某一通信场景下的某一通信设备来说,该通信设备对应的目标智能网络包括第一子网络和第二子网络,该通信设备可以复用该通信设备在其他通信场景下的第一子网络,更新训练在该通信场景下的第二子网络即可。
图14示出了一具体示例。假设第一子网络对应在第一通信场景和第二通信场景下的第一通信设备。如图14所示,第一通信设备在第一通信场景下对应的目标智能网络包括第一子网络和第二子网络#1,第一通信设备在第二通信场景下对应的目标智能网络包括第一子网络和第二子网络#2,可以看出,第一通信设备在第一通信场景下和第一通信设备在第二通信场景下可以共用第一子网络。
可选地,该第一子网络可以用于处理对于第一通信设备在第一通信场景下和通信设备在第二通信场景下来说的公共特征。例如,如果第一通信场景和第二通信场景均为同一个会议室中的通信场景,那么在该情况下,环境特征(即会议室)可以认为是公共特征,那么该第一子网络可以用于提取第一通信设备在第一通信场景下的环境特征,该第一子网络还可以用于提取第一通信设备在第二通信场景下的环境特征。
基于该情况2,通信设备在不同通信场景下可以共用一个第一子网络,进而降低复杂度。此外,通信设备在各个通信场景下可以有各自对应的第二子网络,从而可以根据实际情况训练更新第二子网络,以便可以尽可能地保证性能。
情况3,第一子网络对应多种通信场景,第二子网络对应一种通信场景。
在该情况下,例如,对于某一通信场景来说,该通信场景对应的目标智能网络包括第一子网络和第二子网络,该通信场景可以复用其他通信场景对应的第一子网络,更新训练该通信场景下的第二子网络即可。
图15示出了一具体示例。假设第一子网络对应第一通信场景和第二通信场景。如图15所示,第一通信场景对应的目标智能网络包括第一子网络和第二子网络#1,第二通信场景对应的目标智能网络包括第一子网络和第二子网络#2,可以看出,第一通信场景和第二通信场景可以共用第一子网络。
可选地,该第一子网络可以用于处理对于第一通信场景和第二通信场景来说的公共特征。例如,例如,如果第一通信场景和第二通信场景均为同一个学校中的通信场景,那么在该情况下,地理特征(即学校)可以认为是公共特征,那么该第一子网络可以用于提取任一通信设备在第一通信场景下的地理特征,该第一子网络还可以用于提取任一通信设备在第二通信场景下的地理特征。
基于该情况3,多种通信场景可以共用一个第一子网络,进而降低复杂度。此外,每种通信场景可以有各自对应的第二子网络,从而可以根据实际情况训练更新第二子网络,以便可以尽可能地保证性能。
情况4,第一子网络对应一种通信场景下的多个通信设备,第二子网络对应一种通信场景下的一个通信设备。
在该情况下,例如,对于处于某一通信场景下的某一通信设备来说,该通信设备对应的目标智能网络包括第一子网络和第二子网络,该通信设备可以复用其他通信设备在该通信场景下的第一子网络,更新训练该通信设备在该通信场景下的第二子网络即可。
图16示出了一具体示例。假设第一子网络对应在第一通信场景的第一通信设备和在第一通信场景的第二通信设备。如图14所示,在第一通信场景的第一通信设备对应的目标智能网络包括第一子网络和第二子网络#1,在第一通信场景的第二通信设备对应的目标智能网络包括第一子网络和第二子网络#2,可以看出,第一通信设备和第二通信设备在第一通信场景下可以共用第一子网络。
可选地,该第一子网络可以用于处理对于在第一通信场景的第一通信设备和在第一通信场景的第二通信设备来说的公共特征。例如,如果第一通信场景下的第一通信设备和第第一通信场景的第二通信设备为如图4所示的UE组1中的终端设备,那么在该情况下,角度时延可以认为是公共特征,那么该第一子网络可以用于处理在第一通信场景的第一通信设备的角度时延信息,该第一子网络还可以用于处理在第一通信场景的第二通信设备的角度时延信息。
基于该情况4,在同一通信场景下的不同通信设备可以共用一个第一子网络,进而降低复杂度。此外,在同一通信场景下的不同通信设备可以有各自对应的第二子网络,从而可以根据实际情况训练更新第二子网络,以便可以尽可能地保证性能。
应理解,上文主要从子网络所应用的对象的角度,结合4种情况进行了示例性说明, 对此不作限定。任何属于上述4种情况的变形方案,都落入本申请实施例的保护范围。例如,第一子网络可以对应对象组的信息,该组可以是通信设备组(如终端设备组),或者,该组也可以是通信场景组,或者,该组也可以是特征组,等等。又如,第一对象可以对应多个处理对象,即上述情况1至情况4中的通信设备或者通信场景可以替换为处理对象。
方面2,子网络的属性。
可选地,第一子网络对应一个或多个索引(index,ID)。
通过该设计,通信设备通过获知第一子网络对应的ID,便可以获知第一子网络的信息。例如,预定义或预先配置第一子网络和ID之间的对应关系,在实际应用中,通信设备可以根据ID,直接读取该ID所对应的第一子网络。其中,第一子网络可以是预先训练好的,如网络前期训练好并广播下去的信息。
一可能的设计,第一子网络对应的ID,可以是第一子网络对应的对象的ID。例如,第一子网络对应三个ID,该三个ID分别是标识通信设备1的ID、标识通信设备2的ID、标识通信设备3的ID。
又一可能的设计,第一子网络对应的ID可以是组ID。如第一子网络对应的ID为对象组的ID,或者任意能够体现组的信息形式。以通信设备组为例,第一子网络对应的ID可以为通信设备的组ID,如第一子网络对应的ID可以是通信设备的组ID,又如第一子网络对应的ID可以是通信设备组所在的区域ID,等等。
又一可能的设计,第一子网络对应的ID也可以是为第一子网络定义的ID。例如,有多个第一子网络,可以为各个第一子网络分别定义或配置ID。
应理解,上述仅是示例性说明,对此不作限定,只要可以用ID标识第一子网络,都适用于本申请实施例。
可选地,第一子网络具有组属性。
如前所述,第一子网络对应多个对象,该多个对象可以认为是一个对象组或者多个对象组。一个第一子网络可以对应一个或多个对象组。其中,对象组可以是通信设备组,或者也可以是通信场景组,或者也可以是处理对象组。一示例,第一子网络可以对应一个或多个通信设备组,即该一个或多个通信设备组中的通信设备可以共用其对应的第一子网络。又一示例,第一子网络可以对应一种或多种通信场景组,即该一种或多种通信场景组中的通信场景可以共用其对应的第一子网络。又一示例,第一子网络可以对应一个或多个处理对象组,即在处理该一个或多个处理对象组中的处理对象时可以共用其对应的第一子网络。
应理解,关于对象组的具体形式,不作限定,例如也可以是特征组,等等。
可选地,第一子网络和第二子网络的更新周期可以不同。例如,第一子网络用于处理信息中的公共特征,或者第一子网络对应多个对象,其变化频率不需要太高;第二子网络对应非公共特征,或者第二子网络对应一个对象,可以以更短周期执行训练更新,以便适应设备的变化或者场景的变化。
可选地,与第一子网络和/或第二子网络相关的信息,可以承载于以下一项或多项信令中:RRC、MAC-CE、DCI、UCI。这些信令内容可以借助PDCCH、PUCCH、PDSCH、PUSCH、物理广播信道(physical broadcast channel,PBCH)等信道发送。此外,与第一子网络和/或第二子网络相关的信息的交互可以是:非周期性、半永久或者周期性。
与第一子网络和/或第二子网络相关的信息,可以包括划分第一子网络和第二子网络的相关信息,也可以包括训练第一子网络和第二子网络的相关信息,也可以包括使用第一子网络和第二子网络的相关信息,等等。
与第一子网络和/或第二子网络相关的信息,例如可以包括但不限于:划分第一子网络和第二子网络时参考的信息、训练第一子网络的数据集信息、训练第二子网络的数据集信息、第一子网络的参数信息、第一子网络的模型结构信息、第二子网络的参数信息、第二子网络的模型结构信息、第一子网络对应的多个对象或者对象组的信息、第二子网络对应的一个对象的信息、第一子网络与对应的多个对象或者对象组之间对应关系的信息,等等。
其中,子网络的参数信息(如第一子网络的参数信息或第二子网络的参数信息)例如可以包括但不限于:权值和/或偏置信息。
子网络的模型结构信息(如第一子网络的模型结构或第二子网络的模型结构)例如可以包括但不限于:智能网络所采用的网络类型,规模和/或网络节点间连接关系等。
关于划分第一子网络和第二子网络时参考的信息,下文结合方面4介绍。
方面3,对象组的分组方式。
一种可能的实现方式,可以基于某种公共特征或某类公共特征进行分组。其中,公共特征,例如可以包括不限于以下一项或多项:地理位置信息、统计时延扩展信息、多普勒分布信息、角度分布信息。
一示例,可以根据地理位置进行分组。以对象组为通信设备组为例,图17示出了基于地理位置分组的示意图。如图17所示,可以基于通信设备的地理位置(或者基于通信设备所属的区域)进行分组,分成的组(group)例如可以记为组1(G1)、组2(G2)、组3(G3)、组4(G4)。第一子网络对应一个或多个通信设备组,该一个或多个通信设备组中的通信设备可以共用第一子网络,不同的通信设备组可以对应不同的第一子网络。以组1为例,组1对应一第一子网络,那么该组1内的所有通信设备均可以共用该组1对应的第一子网络。
又一示例,可以根据多普勒分布进行分组。以对象组为通信设备组为例,如图4所示,UE组1可以认为是一个对象组,该UE组1可以对应一个第一子网络,该UE组1中的每个UE可以分别对应一个第二子网络;UE组2可以认为是一个对象组,该UE组2可以对应又一个第一子网络,该UE组2中的每个UE可以分别对应一个第二子网络。
应理解,关于对象组的分组方式,本申请实施例不作限定。此外,针对不同对象时,分组方式也可以不同。作为示例而非限定,针对通信设备,可以按照地理位置分组;针对通信场景,可以按照特征分组。
可选地,关于分组的情况,网络设备可以通知终端设备,也可以由终端设备自己判断。一示例,第一子网络对应一个或多个组ID,那么网络设备可以向终端设备指示组ID。在该示例下,可以通过网络设备指示组ID的方式来辅助该分级机制的目标智能网络设计。又一示例,网络设备负责划分组信息(如区域信息),终端设备自己判断归属(如依据其位置信息和分组参考位置确定所在分组),并上报组ID(如区域ID)。在该示例下,可以通过终端设备反馈组ID的方式来辅助该分级机制的目标智能网络设计。
通过本申请实施例,考虑到不同对象组之间区别较明显,对象组内的对象之间区别较 小。因此,通过为对象组设计第一子网络,这样不仅可以尽可能地降低对性能的影响,也可以降低复杂度。
方面4,第一子网络和第二子网络的划分。
第一子网络和第二子网络的划分,可以是参考以下一项或多项信息划分的:环境数据信息、地理位置信息、时延扩展信息、多普勒分布信息、感知信号信息、通信设备标识信息、业务类型信息、终端设备反馈建议信息,等等。
一种可能的实现方式,网络设备根据以下一项或多项信息,划分第一子网络和第二子网络:环境数据信息、地理位置信息、时延扩展信息、多普勒分布信息、感知信号信息、通信设备标识信息、业务类型信息、终端设备反馈建议信息,等等。
在该实现方式下,网络设备可以通知终端设备第一子网络和第二子网络的定义或划分信息。第一子网络和第二子网络的定义或划分信息的交互可以是:非周期性、半永久或者周期性。
第一子网络和第二子网络的定义或划分信息可以承载于以下一项或多项信令中:RRC、MAC-CE、DCI、UCI。这些信令内容可以借助PDCCH、PUCCH、PDSCH、PUSCH、PBCH等信道发送。
又一种可能的实现方式,终端设备根据以下一项或多项信息,反馈第一子网络和第二子网络的定义/划分信息建议:环境数据信息、地理位置信息、时延扩展信息、多普勒分布信息、感知信号信息、终端设备标识信息、业务类型信息,等等。
终端设备的反馈建议也可以是:非周期性、半永久或者周期性。
方面5,第一子网络和第二子网络的生成。
1、关于第一子网络的生成,至少可以通过以下任一方案实现。
方案1,由通信设备训练获得。
一种可能的实现方式,由一个通信设备训练获得。
例如,通信设备训练第一子网络中部分或者全部网络内容,获得训练结果,即第一子网络。
又一种可能的实现方式,由多个通信设备训练获得。
例如,多个通信设备分别训练第一子网络中部分或者全部网络内容,共享合并获得联合训练结果,即第一子网络。其中,通信设备为终端设备时,网络共享合并的方式可以包括但不限于D2D方式。关于终端设备共享的方式,下文结合终端设备之间共享第一子网络的信息详细介绍。
在方案1中,参与训练的通信设备可以是默认的通信设备,或者也可以是约定的通信设备(如基于预设信息约定的通信设备)。或者通信设备为终端设备时,也可以是网络设备指示的终端设备,对此不作限定。其中,预设信息,例如可以包括:按照通信设备标识的大小,依次(如标识从大到小,或者从小到大)选择一定数量的通信设备。
参与训练的通信设备,可以是通信设备组中能力较强的通信设备,或者可以是通信设备组中曾经参与过训练的通信设备,或者也可以是通信设备组中的任意通信设备,对此不作限定。参与训练的通信设备,可以是通信设备组中的部分通信设备,也可以是全部通信设备。
以通信设备为终端设备为例,可选地,网络设备可以通过以下任一方式,指示参与生 成第一子网络的终端设备。
一方式,网络设备可以通过x比特(bit)开关形式,指示参与生成第一子网络的终端设备,x为大于1或等于1的整数。
例如,可以通过一个1比特的字段来指示是否参与生成第一子网络。其中,0对应参与生成第一子网络,1对应不参与生成第一子网络。或者,1对应不参与生成第一子网络,0对应参与生成第一子网络。应理解,具体如何指示,或者使用多少比特来指示,本申请实施例对此不作限定。
又一方式,网络设备可以发送终端设备列表,指示参与生成第一子网络的终端设备。
例如,网络设备可以发送参与生成第一子网络的终端设备的列表,如UEi~UEj,UEi~UEj均参与生成第一子网络。
又如,网络设备可以发送不参与生成第一子网络的终端设备的列表,如UEi’~UEj’,除UEi’~UEj’以外的终端设备,参与生成第一子网络。
方案2,由网络设备下发。
一种可能的实现方式,网络设备可以预先训练好,并发送给终端设备。
应理解,上述两种方案仅是示例性说明,对此不作限定。例如也可以通过其他方式预先定义第一子网络。
2、关于第二子网络,所有通信设备可以自己训练更新第二子网络。
上文介绍了第一子网络和第二子网络的生成。关于生成第一子网络和第二子网络的具体过程,或者说训练过程,不作严格限定。
可选地,上述训练过程(第一子网络和第二子网络的训练过程),可以是离线训练,或者也可以是在线训练,对此不作限定。例如,更新层数量少的情况下,可以在线训练,收敛快。
其中,离线训练在实时通信的过程中可以实现智能网络的推理过程,子网络(如第一子网络和第二子网络)的学习或训练过程在实时通信之前完成。在线训练在收发端实时通信的过程中同时完成子网络的更新,如包括学习和推理过程。
可选地,可以通过配置(如网络设备指示)或者预先约定(如按照标准表格约定)确定训练集分配信息、约束条件信息(如迭代次数等)。此外,也可以按照子网络规模信息或其占全网络的比例信息,按一定运算方式(如按比例)选择数据集。
上述迭代次数,可以作为用于约束迭代过程的约束条件的一种可能的实现方式,当然,还可以其他方式设置约束条件。例如,可以通过配置或者预先约定收敛条件(即网络训练终止条件),作为约束条件。
通信设备可以根据第一子网络和第二子网络,生成目标智能网络。应理解,通信设备根据第一子网络和第二子网络生成目标智能网络,只是一种确定性的描述,即第一子网络和第二子网络共同来处理通信设备的信息,其生成目标智能网络可以不是必要动作。
例如,第一子网络和第二子网络联结在一起形成的网络为目标智能网络。
此外,关于第一子网络和第二子网络的连接,也不作限定。例如,可以参考神经网络中个神经单元的连接方式,即第一子网络的输出可以是第二子网络的输入,或者,第二子网络的输出可以是第一子网络的输入。
方面6,通信设备获知子网络的方式。
通信设备可以自己训练更新第二子网络,故自身可以获知第二子网络。
在本申请实施例中,通信设备可以通过以下多种方式获知第一子网络的信息。其中,第一子网络的信息,例如可以包括但不限于:训练第一子网络的数据集信息、第一子网络的参数信息、第一子网络的模型结构信息、第一子网络对应的多个对象或者对象组的信息、第一子网络与对应的多个对象或者对象组之间对应关系的信息,等等。
其中,第一子网络的参数信息例如可以包括但不限于:权值和/或偏置信息。
第一子网络的模型结构信息例如可以包括但不限于:智能网络所采用的网络类型,规模和网络节点间连接关系等。
以通信设备为终端设备为例,终端设备可以通过以下任一方式,获取第一子网络的信息。
一可能的实现方式,网络设备向终端设备指示第一子网络的信息。
例如,网络设备向终端设备发送第一子网络的信息。如,网络设备可以直接向终端设备指示第一子网络的参数和模型结构信息。
又如,第一子网络与ID(如组ID)绑定,网络设备可以向终端设备指示第一子网络对应的ID,终端设备根据该ID可以获知第一子网络的信息。
又如,终端设备可以自己判断归属,并上报组ID,以申请该组ID对应的第一子网络的信息。网络设备根据终端设备上报的组ID,向终端设备发送该组ID对应的第一子网络的信息。
又一种可能的实现方式,终端设备可以自己获知第一子网络的信息。
例如,终端设备参与训练第一子网络,并保存第一子网络的信息。
又如,可以预先保存终端设备组与第一子网络的对应关系。终端设备根据自己所属的终端设备组,继承该终端设备组对应的第一子网络,或者直接读取该终端设备组对应的第一子网络的信息。
又一种可能的实现方式,终端设备之间共享第一子网络的信息。
如终端设备组中的终端设备之间共享第一子网络的信息。即对应同一子网络的终端设备之间,可以共享第一子网络。
例如,终端设备组中终端设备可以通过D2D的方式共享第一子网络的信息。
终端设备之间可以通过D2D(如侧行链路(sidelink))方式共享子网络的相关信息(如第一子网络的信息,或者终端设备训练的网络内容等)。该过程至少可以包括同步(包括主辅D2D同步信号发送)、设备发现以及通信三个步骤。在该过程中的相关信息发送可以由PC5接口或者其他接口实现,对此不作限定。
下面以一个终端设备组内的终端设备之间通过D2D方式共享信息为例,介绍D2D通信可能包括的步骤。
1)同步。终端设备组内多个终端设备要进行D2D通信,一般要先进行同步。例如,在多个终端设备通过D2D通信之前先接收同步信号。同步源可以是网络设备,也可以是终端设备。同步源为终端设备时,该终端设备可以是参与训练第一子网络的终端设备(如记为发送终端设备),也可以是未参与训练第一子网络的终端设备(如记为接收终端设备)。
2)设备发现过程。设备发现过程可以由终端设备组内的发送终端设备触发,也可以由接收终端设备触发。例如,发送终端设备通知子网络更新相关信息,终端设备组内的其 他终端设备可以监测并接收。又如,接收终端设备发送请求信息,请求子网络更新相关信息,发送终端设备基于请求信息回应子网络相关信息。
3)通信阶段。在直接通信阶段,发送终端设备和接收终端设备通过D2D通信资源,实现D2D直接通信。例如,发送终端设备和接收终端设备在被配置组ID之后(如自行获取组ID或者被通知组ID),接入或检测组通信空口资源,实现D2D直接通信。其中,组ID即表示用于终端设备组的第一子网络对应的组ID。该组ID可以是终端设备组的ID,也可以是任意能够体现终端设备组的信息形式。
其中,关于D2D通信资源的调度方式不作限定。示例地,D2D通信资源可以是网络设备指示的,也可以是终端设备自行从资源池中选择的。例如,网络设备给终端设备调度指定资源,用于终端设备发送直接数据(direct data)和直接通信控制信息(direct control information)。又如,预先分配或配置或划定资源池,在需要使用时,终端设备自行从资源池中选择资源来发送直接数据和直接通信控制信息。
关于终端设备之间通过D2D方式共享子网络的相关信息的具体交互方式,不作严格限定。例如,对于给定的终端设备组,可以依据以下一项或多项,形成多种可选的通过D2D方式共享公共子网络的相关信息的交互方式:设备发现流程请求方,同步信号发送方,通信资源调度方式,以及组划分和索引配置方式。也就是说,对于给定的终端设备组,依据设备发现流程请求不同,或者同步信号发送方不同,或者通信资源调度方式不同,或者组划分不同,或者索引配置方式不同,或者上述任意组合,形成多种可选的通过D2D方式共享公共子网络的相关信息的交互方式。
举个示例,发送终端设备可以依据其第一子网络训练情况(如训练完成后,或观察第一网络与之前版本差异较大)通过PCx接口启动设备发现流程,接收终端设备给出响应后,基于配置组索引和网络设备配置的通信资源建立连接。并且,发送终端设备向接收终端设备发送第一子网络的部分或全部。
应理解,上述仅是示例性说明,任何属于上述任一步骤的变形,或者任何可以实现上述各步骤所要实现的功能的方式,都适用于本申请实施例。
还应理解,只要可以实现终端设备之间共享信息的方式(如可以实现点到点通信的方式),都落入本申请实施例的保护范围。
上文结合方面1至方面6详细描述了本申请实施例的内容。应理解,上述各个方面的内容可以独立使用,也可以结合使用,对此不作严格限定。以方面5的方案和方面6的方案结合使用为例。通信设备组中的一个或多个通信设备训练得到第一子网络后,可以与该通信设备组中的其他通信设备共享该第一子网络。
上文主要是从不同方面介绍了本申请实施例的方案,下面为便于理解,结合图18给出一具体示例。
如图18所示,通信设备组1和通信设备组N分别对应一个第一子网络。以通信设备组1为例,通信设备组1中包括:通信设备1,通信设备2,……,通信设备n。
以通信设备组1为例。通信设备组1对应的第一子网络可以具有如方面2所述的属性。例如,通信设备组1可以对应一个ID,如通信设备组1的组ID。通信设备组1中的通信设备,可以是按照如方面3所述的方案进行的分组。例如,基于通信设备的地理位置,将通信设备1,通信设备2,……,通信设备n划分为一个通信设备组,即通信设备组1。 通信设备1中的通信设备对应的目标智能网络均包括第一子网络和第二子网络,第一子网络和第二子网络可以是基于方面4所述的方案划分的。通信设备组1中的通信设备对应的目标智能网络均包括第一子网络和第二子网络,第一子网络和第二子网络的生成可以是基于方面5所述的方案生成的。例如,通信设备1训练生成第一子网络,并且该通信设备1可以与通信设备组1中的其余通信设备(如通信设备2,……,通信设备n)共享该第一子网络。通信设备组1中的通信设备可以按照方面6所述的方案,获知第一子网络的信息。例如,通信设备n根据第一子网络对应的ID,直接读取第一子网络的信息。关于各个方面的内容,此处不再赘述。
如图18所示,通信设备1的信息输入至目标智能网络,由目标智能网络中的第一子网络和对应的第二子网络共同处理该信息,最终输出结果。例如,目标智能网络处理完后,结果输入至解码器(decoder)中,如与编码器(encoder)对应的解码器中。其中,第一子网络处理该信息的公共特征(common feature),或者说,该信息中的公共特征由第一子网络处理。通信设备1对应的第二子网络处理该信息的非公共特征(如可以记为distinct feature),或者说,该信息中的非公共特征由通信设备1对应的第二子网络处理。
如图18所示,通信设备n的信息输入至目标智能网络,由目标智能网络中的第一子网络和对应的第二子网络共同处理该信息,最终输出结果。例如,目标智能网络处理完后,结果输入至解码器中,如与编码器对应的解码器中。其中,第一子网络处理该信息的公共特征,或者说,该信息中的公共特征由第一子网络处理。通信设备n对应的第二子网络处理该信息的非公共特征,或者说,该信息中的非公共特征由通信设备n对应的第二子网络处理。
由图18可知,通信设备1和通信设备n对应同一个第一子网络,且分别对应一个第二子网络。假设通信设备1和通信设备n为图4所示分布中的UE组1内的终端设备,那么在该情况下,对于通信设备1和通信设备n来说,角度时延可以认为是公共特征。如图18所示,通信设备1的角度时延信息和通信设备n的角度时延信息可以通过同一个第一子网络处理,其他信息可以分别通过各自对应的第二子网络处理。
应理解,关于通信设备的信息如何输入目标智能网络,不作严格限定。例如,通信设备发送信息,由目标智能网络来对该信息进行处理,或者说,由目标智能网络来捕获或处理或抓取该信息中的特征。
还应理解,对于特征如何输入至各自对应的子网络,不作严格限定。
例如,可以在通信设备组和目标智能网络之间,或者目标智能网络中,新增一个或多个模块,该模块用来执行特征划分。如在通信设备1的信息输入目标智能网络时,通过该模块可以自动分类,将角度时延信息输入至第一子网络,其他信息输入至第二子网络。
又如,各个子网络有能力提取各自对应的特征。如通信设备1的信息输入目标智能网络,第一子网络提取角度时延信息进行处理,第二子网络提取其他信息进行处理。
还应理解,关于各个子网络处理的先后顺序不作严格限定。如第一子网络的输出可以是第二子网络的输入,如图18所示;或者,第二子网络的输出可以是第一子网络的输入。
还应理解,关于结果输出的方式不作严格限定。例如,可以是各个子网络将各自的结果输入至解码器中。又如,也可以是各个子网络的结果合并后,将合并的结果输入至解码器中。
还应理解,通信设备组可以对应一个解码器,如图18所示;或者通信设备组也可以对应多个解码器,对此不作限定。
还应理解,图18仅是一种示例性说明,其不对本申请实施例的保护范围造成限定。例如图18所示的第一子网络的网络架构主要参考了神经网络架构,第一子网络中的网络架构还可以是其他智能学习架构。
以上,结合图1至图18详细说明了本申请实施例提供的方法。
应理解,在上述一些实施例中,以目标智能网络包括第一子网络和第二子网络为例进行描述,但这并不对本申请造成限定。例如,目标智能网络可以包括两个或两个以上的子网络。
还应理解,在上述一些实施例中,主要以子网络为例进行示例性说明。上述子网络均可以替换为模块,或者组件,或者层,或者网络,等等。
还应理解,在上述一些实施例中提到了不同种类的特征。此处的种类,并不限定是特征所属的种类。例如,对于属于时域的特征,可以均属于公共特征,或者也可以均属于非公共特征,或者也可以部分属于公共特征,部分属于非公共特征,对此不作限定。具体地,需要结合其他因素(如所处的场景)确定。
还应理解,本申请中多次提及公共特征和非公共特征,应理解,公共特征和非公共特征是一个相对的概念,或者说是在某一特定情景下的划分,其并不限定特定的特征(或者说特定的参数)。
还应理解,对于非公共特征对应的子网络的数量不作限定。以调制解调处理为例,对应调制解调处理的目标智能网络中包括第二子网络。一示例,该第二子网络可以包括多个子网络,该多个子网络分别用于处理多个非公共特征。也就是说,对于不同的非公共特征,可以分别训练更新各个非公共特征对应的子网络。又一示例,该第二子网络可以包括一个子网络,该一个子网络处理非公共特征。也就是说,对于不同的非公共特征,可以训练更新一个子网络来进行处理。
还应理解,本申请实施例中的目标智能网络,例如均可以替换为目标神经网络(如DNN)。
还应理解,本申请中多次提及智能网络,其具体命名不对本申请实施例保护范围造成限定。例如,用于通信网络中的网络可以记为AI网络,或者也可以记为AI-MIMO网络,等等。
还应理解,本申请提供的目标智能网络(即分级网络架构),可以认为是一种实施架构,可用于通信网络中任意可进行特征分类的复杂网络,用以提升网络整体性能。
还应理解,在上述一些实施例中,以通信设备为例进行了示例性说明,其中,通信设备均可以替换为终端设备。
还应理解,在上述一些实施例中,提到了对应关系,如第一子网络与对象(如通信设备或者通信场景等)的对应关系。此处,第一子网络与对象具有对应关系,即用于表示第一子网络与对象关联,或者说第一子网络可以用于该对象。关于对应关系的具体形式例如可以是表格的形式,或者也可以是第一子网络中有对应的对象的相关信息,或者也可以是对象中有对应的第一子网络的相关信息,或者也可以是其他形式,对此,不作严格限定。
基于上述技术方案,通过第一子网络结合第二子网络(或者说“公共子网络+私有子 网络”)的设计思路,实现一种有效的分级网络架构。不仅可以解决面向多分布训练的网络难以把握和提取具有多样性的样本特征带来性能恶化的问题,还可以解决面向单分布训练的网络泛化性不足的问题。基于本申请技术方案,可以达到极大提高通信***性能、高效平衡空口负载和实现代价的多维优化目标,以缓解学习精度和学习效率的矛盾。本申请提出的分级网络架构(或者也可以称为分级机制的AI架构),可以作为一种通用框架支撑商用通信***中的各种涉及复杂神经网络的***方案,为性能提升和复杂度降低打开全新空间,可以提升智能网络用于通信网络中的可实现性。
基于上述技术方案,可以通过少数通信设备训练第一子网络,其他通信设备可以共享该第一子网络,从而可以实现训练开销的最小化,提升训练效率。此外,针对不同问题灵活调整训练策略,即面对实际问题时可以进行针对性的学习更新,如可以各个通信设备可以各自训练第二子网络,提升速度和训练效率,可以尽可能地保证网络性能和实时性需求。
此外,在上述一些实施例中,可以是基于少数终端设备训练第一子网络,以及网络设备指示组ID(如终端设备组的ID)的方式,辅助该分级机制的目标智能网络设计。或者,也可以是基于少数终端设备训练第一子网络,以及终端设备反馈组ID的方式,辅助该分级机制的目标智能网络设计。
上述图9至图18描述的各个实施例可以为独立的方案,也可以根据内在逻辑进行组合,这些方案都落入本申请的保护范围中。
可以理解的是,上述各个方法实施例中,由终端设备实现的操作,也可以由可用于终端设备的部件(例如芯片或者电路)实现,由网络设备实现的操作,也可以由可用于网络设备的部件(例如芯片或者电路)实现。
以下,结合图19至图22详细说明本申请实施例提供的装置。应理解,装置实施例的描述与方法实施例的描述相互对应,因此,未详细描述的内容可以参见上文方法实施例,为了简洁,这里不再赘述。
如图19所示,本申请实施例提供一种信息处理的装置1900。装置1900可以用于执行上文实施例中的方法900、方法1100、方法1200。该装置1900包括通信单元1910、处理单元1920。
一种可能的实现方式,通信单元1910,用于获取通信网络中的信息;处理单元1920,用于用目标智能网络对信息进行处理,目标智能网络包括第一子网络和第二子网络,第一子网络用于处理信息中的公共特征,第二子网络用于处理信息中的非公共特征。
一示例,第一子网络用于处理一个或多个通信设备的公共特征;或者,第一子网络用于处理一种或多种通信场景下的公共特征。
又一示例,第二子网络用于处理一个通信设备的非公共特征;或者,第二子网络用于处理一种通信场景下的非公共特征。
又一种可能的实现方式,通信单元1910,用于获取通信网络中的信息;处理单元1920,用于用目标智能网络对信息进行处理,目标智能网络包括第一子网络和第二子网络,第一子网络对应多个通信设备,第二子网络对应一个通信设备;或者,第一子网络对应多种通信场景,第二子网络对应一种通信场景。
一示例,第一子网络和第二子网络,是基于以下一项或多项划分的:环境数据信息、感知信号信息、通信设备标识信息、业务类型信息、终端设备的反馈建议信息。
又一示例,与第一子网络和/或第二子网络相关的信息,承载于以下一项或多项信令中:无线资源控制、媒体介入控制-控制元素、下行控制信息、上行控制信息。
又一示例,第一子网络对应一个或多个索引。
又一示例,第一子网络的更新周期与第二子网络的更新周期不同。
又一示例,第一子网络对应多个终端设备;第一子网络由多个终端设备中的部分或全部终端设备训练得到。
又一示例,部分或全部终端设备为网络设备指示的终端设备;或者,部分或全部终端设备为基于预设信息确定的终端设备。
又一示例,第一子网络由多个终端设备中的部分终端设备训练得到,部分终端设备与多个终端设备中的其余终端设备共享第一子网络。
又一示例,通信单元1910,还用于根据第一子网络对应的索引,获取第一子网络的信息;或者,接收指示信息,根据指示信息,获取第一子网络的信息,其中,指示信息用于指示第一子网络的信息。
又一示例,公共特征包括以下一项或多项:地理位置信息、时延扩展信息、多普勒分布信息、角度分布信息。
示例地,本申请实施例提供的信息处理的装置1900的产品实现形态是可以在计算机上运行的程序代码。
示例地,本申请实施例提供的信息处理的装置1900可以是通信设备,也可以是应用于通信设备上的芯片、芯片***或电路。当该装置1900为通信设备时,通信单元1910可以是收发器,或,输入/输出接口;处理单元1920可以是处理器。当该装置1900为为用于通信设备中的芯片、芯片***或电路时,通信单元1910可以是该芯片、芯片***或电路上的输入/输出接口、接口电路、输出电路、输入电路、管脚或相关电路等;处理单元1920可以是处理器、处理电路或逻辑电路等。
如图20所示,本申请实施例还提供一种信息处理的装置2000。该装置2000包括处理器2010,处理器2010与存储器2020耦合,存储器2020用于存储计算机程序或指令,处理器2010用于执行存储器2020存储的计算机程序或指令,使得上文方法实施例中的方法被执行。
可选地,如图20所示,该装置2000还可以包括存储器2020。
可选地,如图20所示,该装置2000还可以包括通信接口2030,通信接口2030用于与外界进行数据的传输。
例如,该装置2000用于实现图9所示实施例中的方法900。
又例如,该装置2000用于实现图11所示实施例中的方法1100。
再例如,该装置2000用于实现图12所示实施例中的方法1200。
在实现过程中,上述方法的各步骤可以通过处理器2010中的硬件的集成逻辑电路或者软件形式的指令完成。结合本申请实施例所公开的方法可以直接体现为硬件处理器执行完成,或者用处理器中的硬件及软件模块组合执行完成。软件模块可以位于随机存储器,闪存、只读存储器,可编程只读存储器或者电可擦写可编程存储器、寄存器等本领域成熟的存储介质中。该存储介质位于存储器2020,处理器2010读取存储器2020中的信息,结合其硬件完成上述方法的步骤。为避免重复,这里不再详细描述。
应理解,本申请实施例中,处理器可以为一个或多个集成电路,用于执行相关程序,以执行本申请方法实施例。
关于存储器和处理器的描述见上文,此处不再赘述。
如图21所示,本申请实施例还提供一种信息处理的装置2100。该装置2100可以用于实现图9至图18所示实施例中的方法。该装置2100包括公共模块2110和非公共模块2120。
例如,公共模块2110可以用于执行上述方法实施例中第一子网络的相关步骤。如公共模块2110可以用于处理如图18所示的通信设备组1中各个通信设备的公共特征。
又如,非公共模块2120可以用于执行上述方法实施例中第二子网络的相关步骤。如公共模块2110可以用于处理如图18所示的通信设备1的信息的非公共特征。
示例地,公共模块2110和非公共模块2120的产品实现形态是可以在计算机上运行的程序代码。
示例地,公共模块2110和非公共模块2120可以集成在同一块芯片上,也可以分别设置在不同的芯片上。
本申请实施例还提供一种计算机可读存储介质,该计算机可读存储介质存储用于设备执行的程序代码,该程序代码包括用于执行上述实施例的方法。
本申请实施例还提供一种包含指令的计算机程序产品,当该计算机程序产品在计算机上运行时,使得计算机执行上述实施例的方法。
本申请实施例还提供一种芯片,该芯片包括处理器与通信接口,处理器通过通信接口读取存储器上存储的指令,执行上述实施例的方法。
可选地,作为一种实现方式,该芯片还可以包括存储器,存储器中存储有指令,处理器用于执行存储器上存储的指令,当指令被执行时,处理器用于执行上述实施例中的方法。
图22是本申请实施例的一种芯片***的示意图。上文方法实施例中的方法900、图11所示实施例中的方法1100或图12所示实施例中的方法1200均可在如图22所示的芯片中得以实现。
图22所示的芯片***2200包括:逻辑电路2210以及输入/输出接口(input/output interface)2220,所述逻辑电路用于与输入接口耦合,通过所述输入/输出接口传输数据(例如第二信道模型的至少部分)参数,以执行图9至图18所述的方法。
应理解,本文中术语“和/或”,仅仅是一种描述关联对象的关联关系,表示可以存在三种关系,例如,A和/或B,可以表示:单独存在A,同时存在A和B,单独存在B这三种情况。另外,本文中字符“/”,一般表示前后关联对象是一种“或”的关系。
在本申请所提供的几个实施例中,应该理解到,所揭露的装置和方法,可以通过其它的方式实现。例如,以上所描述的装置实施例仅是示意性的,例如,上述单元的划分,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式,例如多个单元或组件可以结合或者可以集成到另一个***,或一些特征可以忽略,或不执行。此外,所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口,装置或单元的间接耦合或通信连接,可以是电性,机械或其它的形式。
上述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络 单元上。可以根据实际的需要选择其中的部分或者全部单元实现本申请提供的方案。
另外,在本申请各个实施例中的各功能单元可以集成在一个单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。
本领域普通技术人员可以意识到,结合本文中所公开的实施例描述的各示例的单元及算法步骤,能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟以硬件还是软件方式来执行,取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能,但是这种实现不应认为超出本申请的范围。
当使用软件实现时,可以全部或部分地以计算机程序产品的形式实现。该计算机程序产品包括一个或多个计算机指令。在计算机上加载和执行计算机程序指令时,全部或部分地产生按照本申请实施例所述的流程或功能。计算机可以是通用计算机、专用计算机、计算机网络、或者其他可编程装置。例如,计算机可以是个人计算机,服务器,或者网络设备等。计算机指令可以存储在计算机可读存储介质中,或者从一个计算机可读存储介质向另一个计算机可读存储介质传输,例如,计算机指令可以从一个网站站点、计算机、服务器或数据中心通过有线(例如同轴电缆、光纤、数字用户线(DSL))或无线(例如红外、无线、微波等)方式向另一个网站站点、计算机、服务器或数据中心进行传输。关于计算机可读存储介质,可以参考上文描述。
以上所述,仅为本申请的具体实施方式,但本申请的保护范围并不局限于此,任何熟悉本技术领域的技术人员在本申请揭露的技术范围内,可轻易想到的变化或替换,都应涵盖在本申请的保护范围之内。因此,本申请的保护范围应以权利要求和说明书的保护范围为准。
Claims (62)
- 一种信息处理的方法,其特征在于,包括:获取通信网络中的信息;用目标智能网络对所述信息进行处理,所述目标智能网络包括第一子网络和第二子网络,所述第一子网络用于处理所述信息中的公共特征,所述第二子网络用于处理所述信息中的非公共特征。
- 根据权利要求1所述的方法,其特征在于,所述第一子网络用于处理一个或多个通信设备的公共特征;或者,所述第一子网络用于处理一种或多种通信场景下的公共特征。
- 根据权利要求1或2所述的方法,其特征在于,所述第二子网络用于处理一个通信设备的非公共特征;或者,所述第二子网络用于处理一种通信场景下的非公共特征。
- 根据权利要求1至3中任一项所述的方法,其特征在于,所述公共特征包括以下一项或多项:地理位置信息、时延扩展信息、多普勒分布信息、角度分布信息。
- 根据权利要求1至4中任一项所述的方法,其特征在于,所述第一子网络和所述第二子网络,是基于以下一项或多项划分的:环境数据信息、感知信号信息、通信设备标识信息、业务类型信息、终端设备的反馈建议信息。
- 根据权利要求1至5中任一项所述的方法,其特征在于,与所述第一子网络和/或所述第二子网络相关的信息,承载于以下一项或多项信令中:无线资源控制、媒体介入控制-控制元素、下行控制信息、上行控制信息。
- 根据权利要求1至6中任一项所述的方法,其特征在于,所述第一子网络对应一个或多个索引。
- 根据权利要求7所述的方法,其特征在于,所述第一子网络对应多个通信设备,所述索引为所述多个通信设备的组标识。
- 根据权利要求1至8中任一项所述的方法,其特征在于,所述第一子网络的更新周期与所述第二子网络的更新周期不同。
- 根据权利要求1至9中任一项所述的方法,其特征在于,所述第一子网络对应多个通信设备;所述第一子网络由所述多个通信设备中的部分或全部通信设备训练得到。
- 根据权利要求10所述的方法,其特征在于,所述部分或全部通信设备为网络设备指示的通信设备;或者,所述部分或全部通信设备为基于预设信息确定的通信设备。
- 根据权利要求10或11所述的方法,其特征在于,所述第一子网络由所述多个通信设备中的部分通信设备训练得到,所述部分通信设备与所述多个通信设备中的其余通信设备共享所述第一子网络。
- 根据权利要求1至12中任一项所述的方法,其特征在于,所述方法还包括:根据所述第一子网络对应的索引,获取所述第一子网络的信息;或者,接收指示信息,根据所述指示信息,获取所述第一子网络的信息,其中,所述指示信息用于指示所述第一子网络的信息。
- 一种信息处理的方法,其特征在于,包括:获取通信网络中的信息;用目标智能网络对所述信息进行处理;其中,所述目标智能网络包括第一子网络和第二子网络,所述第一子网络对应多个通信设备,所述第二子网络对应一个通信设备;或者,所述第一子网络对应多种通信场景,所述第二子网络对应一种通信场景。
- 根据权利要求14所述的方法,其特征在于,所述第一子网络和所述第二子网络,是基于以下一项或多项划分的:环境数据信息、感知信号信息、通信设备标识信息、业务类型信息、终端设备的反馈建议信息。
- 根据权利要求14或15所述的方法,其特征在于,与所述第一子网络和/或所述第二子网络相关的信息,承载于以下一项或多项信令中:无线资源控制、媒体介入控制-控制元素、下行控制信息、上行控制信息。
- 根据权利要求14至16中任一项所述的方法,其特征在于,所述第一子网络对应一个或多个索引。
- 根据权利要求17所述的方法,其特征在于,所述第一子网络对应多个通信设备,所述索引为所述多个通信设备的组标识。
- 根据权利要求14至18中任一项所述的方法,其特征在于,所述第一子网络的更新周期与所述第二子网络的更新周期不同。
- 根据权利要求14至19中任一项所述的方法,其特征在于,所述第一子网络对应多个通信设备;所述第一子网络由所述多个通信设备中的部分或全部通信设备训练得到。
- 根据权利要求20所述的方法,其特征在于,所述部分或全部通信设备为网络设备指示的通信设备;或者,所述部分或全部通信设备为基于预设信息确定的通信设备。
- 根据权利要求20或21所述的方法,其特征在于,所述第一子网络由所述多个通信设备中的部分通信设备训练得到,所述部分通信设备与所述多个通信设备中的其余通信设备共享所述第一子网络。
- 根据权利要求14至22中任一项所述的方法,其特征在于,所述方法还包括:根据所述第一子网络对应的索引,获取所述第一子网络的信息;或者,接收指示信息,根据所述指示信息,获取所述第一子网络的信息,其中,所述指示信息用于指示所述第一子网络的信息。
- 一种信息处理的装置,其特征在于,包括:通信单元和处理单元,所述通信单元,用于获取通信网络中的信息;所述处理单元,用于用目标智能网络对所述信息进行处理,所述目标智能网络包括第一子网络和第二子网络,所述第一子网络用于处理所述信息中的公共特征,所述第二子网 络用于处理所述信息中的非公共特征。
- 根据权利要求24所述的装置,其特征在于,所述第一子网络用于处理一个或多个通信设备的公共特征;或者,所述第一子网络用于处理一种或多种通信场景下的公共特征。
- 根据权利要求24或25所述的装置,其特征在于,所述第二子网络用于处理一个通信设备的非公共特征;或者,所述第二子网络用于处理一种通信场景下的非公共特征。
- 根据权利要求24至26中任一项所述的装置,其特征在于,所述公共特征包括以下一项或多项:地理位置信息、时延扩展信息、多普勒分布信息、角度分布信息。
- 根据权利要求24至27中任一项所述的装置,其特征在于,所述第一子网络和所述第二子网络,是基于以下一项或多项划分的:环境数据信息、感知信号信息、通信设备标识信息、业务类型信息、终端设备的反馈建议信息。
- 根据权利要求24至28中任一项所述的装置,其特征在于,与所述第一子网络和/或所述第二子网络相关的信息,承载于以下一项或多项信令中:无线资源控制、媒体介入控制-控制元素、下行控制信息、上行控制信息。
- 根据权利要求24至29中任一项所述的装置,其特征在于,所述第一子网络对应一个或多个索引。
- 根据权利要求30所述的装置,其特征在于,所述第一子网络对应多个通信设备,所述索引为所述多个通信设备的组标识。
- 根据权利要求24至31中任一项所述的装置,其特征在于,所述第一子网络的更新周期与所述第二子网络的更新周期不同。
- 根据权利要求24至32中任一项所述的装置,其特征在于,所述第一子网络对应多个通信设备;所述第一子网络由所述多个通信设备中的部分或全部通信设备训练得到。
- 根据权利要求33所述的装置,其特征在于,所述部分或全部通信设备为网络设备指示的通信设备;或者,所述部分或全部通信设备为基于预设信息确定的通信设备。
- 根据权利要求33或34所述的装置,其特征在于,所述第一子网络由所述多个通信设备中的部分通信设备训练得到,所述部分通信设备与所述多个通信设备中的其余通信设备共享所述第一子网络。
- 根据权利要求24至35中任一项所述的装置,其特征在于,所述通信单元,具体用于:根据所述第一子网络对应的索引,获取所述第一子网络的信息;或者,接收指示信息,根据所述指示信息,获取所述第一子网络的信息,其中,所述指示信息用于指示所述第一子网络的信息。
- 根据权利要求24至36中任一项所述的装置,其特征在于,所述处理单元为处理器。
- 根据权利要求24至37中任一项所述的装置,其特征在于,所述装置为以下任一 项:通信设备、芯片或芯片***。
- 一种信息处理的装置,其特征在于,包括:通信单元和处理单元,所述通信单元,用于获取通信网络中的信息;所述处理单元,用于用目标智能网络对所述信息进行处理;其中,所述目标智能网络包括第一子网络和第二子网络,所述第一子网络对应多个通信设备,所述第二子网络对应一个通信设备;或者,所述第一子网络对应多种通信场景,所述第二子网络对应一种通信场景。
- 根据权利要求39所述的装置,其特征在于,所述第一子网络和所述第二子网络,是基于以下一项或多项划分的:环境数据信息、感知信号信息、通信设备标识信息、业务类型信息、终端设备的反馈建议信息。
- 根据权利要求39或40所述的装置,其特征在于,与所述第一子网络和/或所述第二子网络相关的信息,承载于以下一项或多项信令中:无线资源控制、媒体介入控制-控制元素、下行控制信息、上行控制信息。
- 根据权利要求39至41中任一项所述的装置,其特征在于,所述第一子网络对应一个或多个索引。
- 根据权利要求42所述的装置,其特征在于,所述第一子网络对应多个通信设备,所述索引为所述多个通信设备的组标识。
- 根据权利要求39至43中任一项所述的装置,其特征在于,所述第一子网络的更新周期与所述第二子网络的更新周期不同。
- 根据权利要求39至44中任一项所述的装置,其特征在于,所述第一子网络对应多个通信设备;所述第一子网络由所述多个通信设备中的部分或全部通信设备训练得到。
- 根据权利要求45所述的装置,其特征在于,所述部分或全部通信设备为网络设备指示的通信设备;或者,所述部分或全部通信设备为基于预设信息确定的通信设备。
- 根据权利要求45或46所述的装置,其特征在于,所述第一子网络由所述多个通信设备中的部分通信设备训练得到,所述部分通信设备与所述多个通信设备中的其余通信设备共享所述第一子网络。
- 根据权利要求39至47中任一项所述的装置,其特征在于,所述通信单元,具体用于:根据所述第一子网络对应的索引,获取所述第一子网络的信息;或者,接收指示信息,根据所述指示信息,获取所述第一子网络的信息,其中,所述指示信息用于指示所述第一子网络的信息。
- 根据权利要求39至48中任一项所述的装置,其特征在于,所述处理单元为处理器。
- 根据权利要求39至49中任一项所述的装置,其特征在于,所述装置为以下任一项:通信设备、芯片或芯片***。
- 一种信息处理的装置,其特征在于,包括:通信接口,用于输入和/或输出信息;处理器,用于执行计算机程序,以使得所述装置实现如权利要求1至13中任一项所述的方法。
- 根据权利要求51所述的装置,其特征在于,所述装置为以下任一项:通信设备、芯片或芯片***。
- 一种信息处理的装置,其特征在于,包括:通信接口,用于输入和/或输出信息;处理器,用于执行计算机程序,以使得所述装置实现如权利要求14至23中任一项所述的方法。
- 根据权利要求53所述的装置,其特征在于,所述装置为以下任一项:通信设备、芯片或芯片***。
- 一种信息处理的装置,其特征在于,包括:存储器,用于存储可执行指令;处理器,用于调用并运行所述存储器中的所述可执行指令,以执行权利要求1至13中任一项所述的方法。
- 根据权利要求55所述的装置,其特征在于,所述装置为以下任一项:通信设备、芯片或芯片***。
- 一种信息处理的装置,其特征在于,包括:存储器,用于存储可执行指令;处理器,用于调用并运行所述存储器中的所述可执行指令,以执行权利要求14至23中任一项所述的方法。
- 根据权利要求57所述的装置,其特征在于,所述装置为以下任一项:通信设备、芯片或芯片***。
- 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质中存储有程序指令,当所述程序指令由处理器运行时,实现权利要求1至13中任一项所述的方法。
- 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质中存储有程序指令,当所述程序指令由处理器运行时,实现权利要求14至23中任一项所述的方法。
- 一种计算机程序产品,其特征在于,所述计算机程序产品包括计算机程序代码,当所述计算机程序代码在计算机上运行时,实现权利要求1至13中任一项所述的方法。
- 一种计算机程序产品,其特征在于,所述计算机程序产品包括计算机程序代码,当所述计算机程序代码在计算机上运行时,实现权利要求14至23中任一项所述的方法。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP21909284.8A EP4255011A4 (en) | 2020-12-24 | 2021-12-17 | INFORMATION PROCESSING METHOD AND DEVICE |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011556785.2 | 2020-12-24 | ||
CN202011556785.2A CN114666812A (zh) | 2020-12-24 | 2020-12-24 | 信息处理的方法和装置 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022135288A1 true WO2022135288A1 (zh) | 2022-06-30 |
Family
ID=82024549
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2021/139077 WO2022135288A1 (zh) | 2020-12-24 | 2021-12-17 | 信息处理的方法和装置 |
Country Status (3)
Country | Link |
---|---|
EP (1) | EP4255011A4 (zh) |
CN (1) | CN114666812A (zh) |
WO (1) | WO2022135288A1 (zh) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN118282879A (zh) * | 2022-12-30 | 2024-07-02 | 华为技术有限公司 | 一种通信方法、装置、可读存储介质及芯片*** |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105072045A (zh) * | 2015-08-10 | 2015-11-18 | 济南大学 | 一种具有恶意软件网络行为发现能力的无线路由器 |
CN106095983A (zh) * | 2016-06-20 | 2016-11-09 | 北京百度网讯科技有限公司 | 一种基于个性化深度神经网络的相似度确定方法及装置 |
CN109362092A (zh) * | 2018-12-07 | 2019-02-19 | ***通信集团福建有限公司 | 网络参数调整方法、装置、设备及介质 |
US20190130019A1 (en) * | 2017-10-30 | 2019-05-02 | Qualcomm Incorporated | System and method for compact storage and efficient retrieval of access point information for detecting rogue access points |
CN111818139A (zh) * | 2020-06-28 | 2020-10-23 | 盾钰(上海)互联网科技有限公司 | 一种基于神经网络的无线异构控制计算*** |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10460255B2 (en) * | 2016-07-29 | 2019-10-29 | Splunk Inc. | Machine learning in edge analytics |
-
2020
- 2020-12-24 CN CN202011556785.2A patent/CN114666812A/zh active Pending
-
2021
- 2021-12-17 EP EP21909284.8A patent/EP4255011A4/en active Pending
- 2021-12-17 WO PCT/CN2021/139077 patent/WO2022135288A1/zh unknown
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105072045A (zh) * | 2015-08-10 | 2015-11-18 | 济南大学 | 一种具有恶意软件网络行为发现能力的无线路由器 |
CN106095983A (zh) * | 2016-06-20 | 2016-11-09 | 北京百度网讯科技有限公司 | 一种基于个性化深度神经网络的相似度确定方法及装置 |
US20190130019A1 (en) * | 2017-10-30 | 2019-05-02 | Qualcomm Incorporated | System and method for compact storage and efficient retrieval of access point information for detecting rogue access points |
CN109362092A (zh) * | 2018-12-07 | 2019-02-19 | ***通信集团福建有限公司 | 网络参数调整方法、装置、设备及介质 |
CN111818139A (zh) * | 2020-06-28 | 2020-10-23 | 盾钰(上海)互联网科技有限公司 | 一种基于神经网络的无线异构控制计算*** |
Non-Patent Citations (1)
Title |
---|
See also references of EP4255011A4 |
Also Published As
Publication number | Publication date |
---|---|
EP4255011A4 (en) | 2024-05-22 |
CN114666812A (zh) | 2022-06-24 |
EP4255011A1 (en) | 2023-10-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Singh et al. | The evolution of radio access network towards open-RAN: Challenges and opportunities | |
WO2022033456A1 (zh) | 一种信道状态信息测量反馈方法及相关装置 | |
WO2022206513A1 (zh) | 模型处理的方法、通信装置和*** | |
CN112929223B (zh) | 一种基于联邦学习方式训练神经网络模型的方法及*** | |
CN113302900A (zh) | 混合传输方案确定 | |
US20240211770A1 (en) | Communication method and apparatus | |
WO2023125660A1 (zh) | 一种通信方法及装置 | |
WO2022135288A1 (zh) | 信息处理的方法和装置 | |
US20240179603A1 (en) | Communication method and apparatus | |
WO2023040971A1 (zh) | 一种人工智能模型下载方法、装置及*** | |
WO2021179165A1 (zh) | 模型协调方法及装置 | |
CN116151348A (zh) | 一种模型训练方法及相关装置 | |
Chayon et al. | Key challenges and potential applications of LTE-Advanced | |
WO2024114686A1 (zh) | 获取训练数据的方法和通信装置 | |
WO2024067245A1 (zh) | 模型匹配的方法和通信装置 | |
WO2023246267A1 (zh) | 通信方法、通信装置和*** | |
WO2024130506A1 (zh) | 通信方法、装置及存储介质 | |
Wang et al. | Airborne Computing Platform Based Joint Offload and Deployment Optimization Algorithm | |
US20240054351A1 (en) | Device and method for signal transmission in wireless communication system | |
WO2024067248A1 (zh) | 一种获取训练数据集的方法和装置 | |
WO2024046215A1 (zh) | 一种通信方法及装置 | |
WO2023125598A1 (zh) | 一种通信方法及通信装置 | |
US20230325654A1 (en) | Scalable deep learning design for missing input features | |
WO2024061125A1 (zh) | 一种通信方法及装置 | |
WO2023044904A1 (zh) | 通信方法、终端设备及网络设备 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 21909284 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2021909284 Country of ref document: EP Effective date: 20230630 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |