WO2024032606A1 - 信息传输方法、装置、设备、***及存储介质 - Google Patents

信息传输方法、装置、设备、***及存储介质 Download PDF

Info

Publication number
WO2024032606A1
WO2024032606A1 PCT/CN2023/111732 CN2023111732W WO2024032606A1 WO 2024032606 A1 WO2024032606 A1 WO 2024032606A1 CN 2023111732 W CN2023111732 W CN 2023111732W WO 2024032606 A1 WO2024032606 A1 WO 2024032606A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
module
model
quantization
following
Prior art date
Application number
PCT/CN2023/111732
Other languages
English (en)
French (fr)
Inventor
杨昂
吴昊
谢天
孙鹏
Original Assignee
维沃移动通信有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 维沃移动通信有限公司 filed Critical 维沃移动通信有限公司
Publication of WO2024032606A1 publication Critical patent/WO2024032606A1/zh

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04WWIRELESS COMMUNICATION NETWORKS
    • H04W24/00Supervisory, monitoring or testing arrangements
    • H04W24/02Arrangements for optimising operational condition

Definitions

  • This application belongs to the field of communication technology, and specifically relates to an information transmission method, device, equipment, system and storage medium.
  • the network side can send a channel state information reference signal (Channel State Information-Reference Signal, CSI-RS) to the user equipment (User Equipment, UE) for channel estimation by the UE.
  • CSI-RS Channel State Information-Reference Signal
  • the UE performs channel estimation based on CSI-RS, calculates the corresponding channel information, and feeds back the Precoding Matrix Index (PMI) to the network side through the codebook.
  • PMI Precoding Matrix Index
  • the network side combines the channel information based on the codebook information fed back by the UE. As follows Before a CSI is reported, the network side uses this to perform data precoding and multi-user scheduling.
  • Embodiments of the present application provide an information transmission method, device, equipment, system and storage medium, which can solve the problem of model information leakage due to the need to inform target nodes of all details of the model when deploying models on different network nodes.
  • an information transmission method includes: a first device inputs first information to a first artificial intelligence (Artificial Intelligence, AI) module to obtain second information; the first device transmits information to a second The device sends the second information, and the second information is used by the second device to input the second information to the second AI module to obtain the first information and/or related information of the first information; wherein, between the first AI module and the second AI Before the module performs the first action, the first device and the second device align the third information; the third information includes model information of the first AI module and/or the second AI module; the first action includes at least one of the following: training, Update, reason.
  • AI Artificial Intelligence
  • an information transmission device which is applied to the first device.
  • the information transmission device includes: a processing module and a sending module.
  • the processing module is used to input the first information to the first AI module to obtain the second information.
  • the sending module is used to send the second information obtained by the processing module to the second device.
  • the second information is used by the second device to input the second information to the second AI module to obtain the first information and/or the correlation of the first information. information.
  • the first device and the second device align the third information; the third information includes model information of the first AI module and/or the second AI module;
  • the first action includes at least one of the following: training, updating, and inference.
  • an information transmission method includes: the second device receives second information from the first device.
  • the second information is obtained by the first device inputting the first information into the first AI module.
  • Information the second device inputs the second information to the second AI module to obtain the first information and/or related information of the first information;
  • the first device and the second device align the third information;
  • the third information includes model information of the first AI module and/or the second AI module;
  • the first action includes at least one of the following: training, updating, and inference.
  • an information transmission device applied to a second device, and the information transmission device includes: a receiving module and a processing module.
  • the receiving module is configured to receive second information from the first device, where the second information is the information obtained by the first device inputting the first information into the first AI module.
  • the processing module is configured to input the second information received by the receiving module into the second AI module to obtain the first information and/or related information of the first information.
  • the first device and the second device align the third information; the third information includes model information of the first AI module and/or the second AI module;
  • the first action includes at least one of the following: training, updating, and inference.
  • a communication device in a fifth aspect, includes a processor and a memory.
  • the memory stores a program or instructions that can be run on the processor.
  • the program or instructions are implemented when executed by the processor. The steps of the method as described in the first aspect.
  • a communication device including a processor and a communication interface, wherein the processor is configured to input the first information to the first AI module to obtain the second information.
  • the communication interface is used to send second information to the second device, and the second information is used by the second device to input the second information to the second AI module to obtain the first information and/or related information of the first information.
  • the first device and the second device align third information, where the third information includes model information of the first AI module and/or the second AI module.
  • the first action includes at least one of the following: training, updating, and inference.
  • a communication device in a seventh aspect, includes a processor and a memory.
  • the memory stores programs or instructions that can be run on the processor.
  • the program or instructions are implemented when executed by the processor. The steps of the method as described in the third aspect.
  • a communication device including a processor and a communication interface, wherein the communication interface is used to receive second information from the first device, and the second information is for the first device to input the first information to Information obtained from the first AI module.
  • the processor is configured to input the second information to the second AI module to obtain the first information and/or related information of the first information.
  • the first device and the second device align the third information, and the third information includes model information of the first AI module and/or the second AI module;
  • the first action includes at least one of the following: training, updating, and inference.
  • a ninth aspect provides a communication system, including: a first device and a second device.
  • the first device can be used to perform the steps of the information transmission method as described in the first aspect.
  • the second device can be used to perform The steps of the information transmission method described in the third aspect.
  • a readable storage medium is provided. Programs or instructions are stored on the readable storage medium. When the programs or instructions are executed by a processor, the steps of the method described in the first aspect are implemented, or the steps of the method are implemented as described in the first aspect. The steps of the method described in the third aspect.
  • a chip in an eleventh aspect, includes a processor and a communication interface.
  • the communication interface is coupled to the processor.
  • the processor is used to run programs or instructions to implement the method described in the first aspect. method, or implement a method as described in the third aspect.
  • a computer program/program product is provided, the computer program/program product is stored in a storage medium, and the computer program/program product is executed by at least one processor to implement as described in the first aspect
  • the multi-node problem is solved by aligning the model-related information of the first AI module and/or the second AI module in advance.
  • the model pairing problem when deploying models for inference enables models distributed on different nodes to perform joint inference on information, that is, the first device infers the first information through the first AI module, and the second device infers the first information through the first AI module.
  • the second information through the second AI module there is no need to inform the target node of all the details of the model in order to perform joint inference. This ensures the inference performance of the model and avoids the leakage of model information.
  • Figure 1 is a schematic architectural diagram of a wireless communication system provided by an embodiment of the present application.
  • Figure 2 is a schematic structural diagram of a neural network provided by related technologies
  • Figure 3 is a schematic structural diagram of a neuron provided by related technologies
  • Figure 4 is a flow chart of an information transmission method provided by an embodiment of the present application.
  • Figure 5 is one of the structural schematic diagrams of an information transmission device provided by an embodiment of the present application.
  • Figure 6 is a second structural schematic diagram of an information transmission device provided by an embodiment of the present application.
  • Figure 7 is a schematic diagram of the hardware structure of a communication device provided by an embodiment of the present application.
  • Figure 8 is a schematic diagram of the hardware structure of a UE provided by an embodiment of the present application.
  • Figure 9 is one of the hardware structural schematic diagrams of a network side device provided by an embodiment of the present application.
  • Figure 10 is the second schematic diagram of the hardware structure of a network-side device provided by an embodiment of the present application.
  • first, second, etc. in the description and claims of this application are used to distinguish similar objects and are not used to describe a specific order or sequence. It is to be understood that the terms so used are interchangeable under appropriate circumstances so that the embodiments of the present application can be practiced in sequences other than those illustrated or described herein, and that "first" and “second” are distinguished objects It is usually one type, and the number of objects is not limited.
  • the first object can be one or multiple.
  • “and/or” in the description and claims indicates at least one of the connected objects, and the character “/" generally indicates that the related objects are in an "or” relationship.
  • LTE Long Term Evolution
  • LTE-Advanced, LTE-A Long Term Evolution
  • CDMA Code Division Multiple Access
  • TDMA Time Division Multiple Access
  • FDMA Frequency Division Multiple Access
  • OFDMA Orthogonal Frequency Division Multiple Access
  • SC-FDMA Single-carrier Frequency Division Multiple Access
  • system and “network” in the embodiments of this application are often used interchangeably, and the described technology can be used not only for the above-mentioned systems and radio technologies, but also for other systems and radio technologies.
  • NR New Radio
  • the following description describes a New Radio (NR) system for example purposes, and uses NR terminology in much of the following description, but these techniques can also be applied to applications other than NR system applications, such as 6th Generation , 6G) communication system.
  • NR New Radio
  • FIG. 1 shows a block diagram of a wireless communication system to which embodiments of the present application are applicable.
  • the wireless communication system includes UE11 and network side device 12.
  • UE11 can be a mobile phone, a tablet computer (Tablet Personal Computer), a laptop computer (Laptop Computer), or a notebook computer, a personal digital assistant (Personal Digital Assistant, PDA), a palmtop computer, a netbook, or a super mobile personal computer ( ultra-mobile personal computer (UMPC), mobile Internet device (MID), augmented reality (AR)/virtual reality (VR) equipment, robots, wearable devices (Wearable Device), Vehicle-mounted equipment (VUE), pedestrian terminal (PUE), smart home (home equipment with wireless communication functions, such as refrigerators, TVs, washing machines or furniture, etc.), games Terminal devices such as computers, personal computers (PCs), teller machines or self-service machines.
  • PCs personal computers
  • teller machines or self-service machines such as computers, personal computers (PCs), teller machines or self-service machines.
  • the network side device 12 may include an access network device or a core network device, where the access network device 12 may also be called a radio access network device, a radio access network (Radio Access Network, RAN), a radio access network function or Wireless access network unit.
  • the access network device 12 may include a base station, a WLAN access point or a WiFi node, etc.
  • the base station may be called a Node B, an evolved Node B (eNB), an access point, a Base Transceiver Station (BTS), a radio Base station, radio transceiver, Basic Service Set (BSS), Extended Service Set (ESS), Home Node B, Home Evolved Node B, Transmitting Receiving Point (TRP) or all
  • eNB evolved Node B
  • BTS Base Transceiver Station
  • BSS Basic Service Set
  • ESS Extended Service Set
  • Home Node B Home Evolved Node B
  • TRP Transmitting Receiving Point
  • Core network equipment may include but is not limited to at least one of the following: core network nodes, core network functions, mobility management entities (Mobility Management Entity, MME), access mobility management functions (Access and Mobility Management Function, AMF), session management functions (Session Management Function, SMF), User Plane Function (UPF), Policy Control Function (PCF), Policy and Charging Rules Function (PCRF), Edge Application Service Discovery function (Edge Application Server Discovery Function, EASDF), Unified Data Management (UDM), Unified Data Repository (UDR), Home Subscriber Server (HSS), centralized network configuration ( Centralized network configuration (CNC), Network Repository Function (NRF), Network Exposure Function (NEF), Local NEF (Local NEF, or L-NEF), Binding Support Function (Binding Support Function, BSF), application function (Application Function, AF), etc.
  • MME mobility management entities
  • AMF Access and Mobility Management Function
  • SMF Session Management Function
  • UPF User Plane Function
  • PCF Policy Control Function
  • AI is the integration of artificial intelligence into wireless communication networks. Significantly improving technical indicators such as throughput, delay, and user capacity are important tasks for future wireless communication networks.
  • AI modules such as neural networks, decision trees, support vector machines, Bayesian classifiers, etc.
  • the embodiment of this application takes a neural network as an example for explanation, but does not limit the specific type of AI module.
  • FIG. 2 it is a schematic diagram of a neural network.
  • the neural network includes an input layer, a hidden layer and an output layer.
  • X1, X2,...,Xn are inputs and Y is output.
  • the neural network is composed of neurons, as shown in Figure 3, which is a schematic diagram of neurons.
  • a 1 , a 2 ,...a K are inputs
  • w is weight (multiplicative coefficient)
  • b is bias (additive coefficient)
  • z a 1 w 1 +...+a k w k +...+ a K w K +b
  • ⁇ (z) is the activation function.
  • activation functions include Sigmoid, tanh, ReLU (Rectified Linear Unit, linear rectification function, modified linear unit), etc.
  • the parameters of the neural network are optimized through the gradient optimization algorithm.
  • Gradient optimization algorithm is a type of algorithm that minimizes or maximizes an objective function (also called a loss function), and the objective function is a mathematical combination of model parameters and data.
  • an objective function also called a loss function
  • the objective function is a mathematical combination of model parameters and data.
  • a neural network model f(.) can be constructed. With the model, the predicted output f(x) can be obtained based on the input The difference between the values (f(x)-Y), this is the loss function. Find appropriate W and b so that the value of the above loss function reaches Minimum, the smaller the loss value, the closer the model is to the real situation.
  • the currently common optimization algorithm is based on the BP (error Back Propagation) algorithm.
  • BP error Back Propagation
  • the basic idea of BP algorithm is that the learning process consists of two processes: forward propagation of signals and back propagation of errors.
  • the input sample is passed in from the input layer, processed layer by layer by each hidden layer, and then transmitted to the output layer. If the actual output of the output layer does not match the expected output, it will enter the error backpropagation stage.
  • Error backpropagation is to backpropagate the output error in some form to the input layer layer by layer through the hidden layer, and allocate the error to all units in each layer, thereby obtaining the error signal of each layer unit. This error signal is used as a correction for each unit.
  • the basis for the weight is used as a correction for each unit.
  • This process of adjusting the weights of each layer in forward signal propagation and error back propagation is carried out over and over again.
  • the process of continuous adjustment of weights is the learning and training process of the network. This process continues until the error of the network output is reduced to an acceptable level, or until a preset number of learning times.
  • Common optimization algorithms include gradient descent (Gradient Descent), stochastic gradient descent (SGD), mini-batch gradient descent (small batch gradient descent), momentum method (Momentum), Nesterov (the name of the inventor, specifically Stochastic gradient descent with momentum), Adagrad (ADAptive GRADient descent, adaptive gradient descent), Adadelta, RMSprop (root mean square prop, root mean square error reduction), Adam (Adaptive Moment Estimation, adaptive momentum estimation), etc.
  • CQI Channel Quality Indicator
  • PMI Precoding Matrix Indicator
  • Eigen beamforming Eigen beamforming
  • the base station sends CSI-RS on certain time-frequency resources in a certain time slot (slot).
  • the UE performs channel estimation based on the CSI-RS, calculates the channel information on this slot, and feeds back the PMI to the base station through the codebook.
  • the base station combines the channel information based on the codebook information fed back by the UE, and uses this to perform data precoding and multi-user scheduling before the next CSI report.
  • the UE can change the PMI reporting on each subband to reporting PMI according to delay. Since the channels in the delay domain are more concentrated, the PMI with fewer delays can approximately represent the PMI of all subbands. , that is, the delay field information is compressed before reporting.
  • the base station can precode the CSI-RS in advance and send the coded CSI-RS to the terminal. What the UE sees is the channel corresponding to the coded CSI-RS. The UE only needs to Just select several ports with greater strength among the indicated ports and report the coefficients corresponding to these ports.
  • the model pairing problem when deploying models on multiple nodes for inference (such as joint inference), so that when models distributed on different nodes perform joint inference on information, there is no need to inform the target node of all the details of the model in order to perform joint inference. This ensures that While improving the model's inference performance, it also avoids the leakage of model information.
  • FIG. 4 shows a flow chart of an information transmission method provided by an embodiment of the present application.
  • the information transmission method provided by the embodiment of the present application may include the following steps 201 to 204.
  • Step 201 The first device inputs the first information to the first AI module to obtain the second information.
  • the first device can infer the first information through the first AI module to obtain the second information.
  • the above-mentioned first information includes at least one of the following: channel information (such as CSI) and beam quality information.
  • the above-mentioned second information includes at least one of the following: PMI, predicted beam information, or beam indication.
  • the above-mentioned first information includes channel information, and the second information is PMI.
  • the above-mentioned first information includes beam quality, and the second information is predicted beam information or beam indication.
  • Step 202 The first device sends the second information to the second device.
  • the above-mentioned second information is used by the second device to input the second information to the second AI module to obtain the first information and/or related information of the first information. That is, the second device can reason about the second information through the second AI module, obtain relevant information of the first information and/or recover the first information.
  • the first device and the second device align the third information; the third information includes the first AI module and/or the second AI module model information; the first action includes at least one of the following: training, updating, and inference.
  • the above-mentioned first device may be a network-side device or a UE; the above-mentioned second device may be a network-side device or a UE.
  • the first device is a network-side device
  • the second device is a UE.
  • the above-mentioned first device is a UE
  • the second device is a network-side device.
  • the first device and the second device are different nodes on the network side (for example, a base station, a network element).
  • the first device and the second device are different UE nodes.
  • the relevant information of the above-mentioned first information may include at least one of the following: a precoding matrix, a decomposition matrix or vector of a channel, an inverse matrix of a channel matrix, an inverse matrix of a decomposition matrix or a vector of a channel Or inverse vector, channel information in the transform domain, rank, rank index, layer, layer index, channel quality, channel signal-to-noise ratio, optional beam identifier, and beam quality of the optional beam.
  • the specific decomposition method is any of the following: singular value decomposition, eigenvalue decomposition, or triangular decomposition.
  • the above-mentioned transform domain includes at least one of the following: spatial domain, frequency domain, time domain, delay domain, Doppler domain, etc.
  • the above transform domain includes a combination domain of at least two domains among the spatial domain, the frequency domain, the time domain, the delay domain and the Doppler domain.
  • the delay domain and the Doppler domain are combined into the delay Doppler domain.
  • the above-mentioned first AI module and/or second AI module are obtained according to at least one of the following:
  • the first device is trained based on target information from the second device or other network elements
  • the second device is trained based on target information from the first device or other network elements.
  • the above target information includes at least one first information related to the first action of the AI module and at least one second information corresponding to the at least one first information.
  • the first device or other network element sends (a large amount of) first information and corresponding second information to the second device, or the second device or Other network elements send (a large amount of) first information and corresponding second information to the first device.
  • the first information refers to a certain type of information
  • at least one first information refers to at least one value or at least one parameter of this type of information.
  • the second information refers to the first information.
  • the above-mentioned first AI module and/or second AI module are updated or adjusted according to at least one of the following:
  • the first device updates or adjusts based on the target information from the second device or other network elements
  • the second device updates or adjusts based on the target information from the first device or other network elements.
  • the first device or other network element sends (a large amount of) first information and corresponding second information to the second device, or the second device or other network elements send (a large amount of) first information and corresponding second information to the first device.
  • the first device and the second device can input and output data through the interactive model to train the first AI module and/or the second AI module, or to perform training on the first AI module and/or the second AI module. Updates/adjustments to modules for inferring information with models distributed across different nodes.
  • the above-mentioned third information specifically includes at least one of the following: structural characteristics of the model, load quantification method of the model, estimation accuracy or output accuracy of the model.
  • first device and the second device may align all or part of the structural features of the first AI module and/or the second AI module. And/or, the first device and the second device may align the load quantification method of the first AI module and/or the second AI module. And/or, the first device and the second device may align the estimation accuracy/output accuracy of the first AI module and/or the second AI module.
  • the structural characteristics of the above-mentioned model include at least one of the following: model structure, basic structural characteristics of the model, structural characteristics of the model sub-modules, number of model layers, number of model neurons, model size, and model complexity. Degree, quantitative parameters of model parameters.
  • the first device and the second device align the structural features of the first AI module and/or the second AI module so that the first device and the second device use the same model structure.
  • the UE and the base station have the same model structure for generating uplink control information (UCI), that is, the third information directly indicates that the model structures of the two are the same.
  • UCI uplink control information
  • the basic structural characteristics of the above-mentioned model include at least one of the following: whether it contains a fully connected structure, whether it contains a convolutional structure, whether it contains a Long-Short Term Memory model (Long-Short Term Memory, LSTM) structure, Whether to include attention structure and whether to include residual structure.
  • LSTM Long-Short Term Memory
  • the number of model neurons includes at least one of the following: the number of fully connected neurons, the number of convolution neurons, the number of memory neurons, the number of attention neurons, and the number of residual neurons.
  • the number of neurons in the model includes at least one of the following: the number of all types of neurons, the number of a single type of neurons, the number of neurons in the entire model, and the number of neurons in a single layer or several layers. quantity.
  • the number of neurons of all types and the number of a single type of neurons can be understood as the number of neurons of one type, and the number of neurons of the entire model and the number of neurons of a single layer or several layers can be understood as the number of neurons of one type.
  • Number of neurons For example, the combination of the number of neurons of all types and the number of neurons of the entire model, that is, the first device and the second device need to align the number of neurons of all types and the entire model. Or, for example, a single type of god The number of neurons is combined with the number of neurons in a single layer or several layers, that is, the first device and the second device need to align a single type of neurons in a single layer, such as fully connected neurons in layer 3.
  • the quantified parameters of the above-mentioned model parameters include at least one of the following: the quantization method of the model parameters and the number of quantized bits of a single neuron parameter.
  • the quantization method of the model parameters includes at least one of the following: uniform quantization method, non-uniform quantization method, weight sharing quantization method or group quantization method, parameter coding quantization method, transform domain quantization method, and product quantization method.
  • weight sharing quantification method or the group quantification method can be understood as: dividing the AI parameters into multiple sets, and the elements in each set share a value.
  • the quantization method of parameter encoding can be understood as: encoding floating point numbers.
  • encoding floating point numbers include at least one of the following: lossy coding, lossless coding, etc., such as Huffman coding.
  • transform domain quantization method can be understood as: transforming the floating point number into another domain, such as the frequency domain, S domain, Z domain, etc., performing at least one of the above quantization operations, and then inversely transforming it back.
  • the Product Quantization method can be understood as: dividing floating-point numbers into multiple subspaces, and performing at least one of the above quantization operations on each subspace.
  • the above-mentioned load quantification method includes at least one of the following: quantization method, dimensionality of features before and after quantization, and quantization method used during quantization.
  • the payload quantification method here refers to how the model converts the output floating point type features into binary type feedback information starting from transmission, which is different from the quantification of model parameters in the structural characteristics of the model.
  • the above-mentioned quantization method can be configured through the third information, or can also be configured by the codebook (that is, whichever quantization method is used in the associated codebook, whichever quantization method is used for training here), Or it can also be determined based on the CSI report configuration (that is, which quantization method is used in the CSI report configuration, which quantization method is used for training here).
  • the codebook or CSI report configuration belongs to the third information.
  • the quantization method used in the above-mentioned quantization includes at least one of the following: synchronizing the codebook content and codebook usage method is required when using the codebook for quantization, and synchronizing the quantization rules is required when using specific rules for quantization. .
  • the codebook content is the matrix itself.
  • 5 floating point numbers are converted into 10 bits, and a codebook of [5,2 ⁇ 10] is constructed, that is, a matrix with 5 rows and 2 ⁇ 10 columns.
  • the content of the codebook is the matrix itself.
  • codebook quantization for a floating-point number vector with a length of 10 and a value range of [0,1], the first 5 floating-point numbers are selected from a codebook of size [5,2 ⁇ 10] and The column of vectors with the smallest error of itself is used as the quantization result; the last five floating point numbers select the column of vectors with the smallest error of itself from a codebook of size [5,2 ⁇ 15] as the quantization result. Finally, the column number corresponding to the selected quantization result in the codebook is used as the feedback binary payload information.
  • the above-mentioned quantization rule includes at least one of the following: N quantization intervals and quantization methods, where N is a positive integer.
  • the quantization method includes at least one of the following: uniform quantization method, non-uniform quantization method, weight sharing quantization method or group quantization method, parameter coding quantization method, transform domain quantization method, and product quantization method.
  • N quantization intervals N1 floating point numbers in a single quantization interval are quantized into N2 bits.
  • the quantization rule for a floating-point number vector with a length of 10 and a value range of [0,1], the first 5 floating-point numbers are uniformly quantized using 2 bits each, and the last 5 floating-point numbers are quantized using 3 bits each. Uniform quantification. Finally, the serial number of the selected interval is used as the feedback binary payload information.
  • the synchronization method includes any of the following: selecting from a predefined method set, feedback the set serial number representing the selected method during synchronization, and directly sending the codebook content.
  • the method for the first device and the second device to align the third information includes at least one of the following:
  • the first device or other network element When the first device or other network element sends the target information to the second device, it also sends the third information;
  • the second device or other network element When the second device or other network element sends the target information to the first device, it also sends the third information;
  • the first device or other network element Before the first device or other network element sends the target information to the second device, the first device or other network element sends the third information;
  • the second device or other network element Before the second device or other network element sends the target information to the first device, the second device or other network element sends the third information;
  • the second device When the second device requests target information, the second device sends the third information
  • the first device When the first device requests target information, the first device sends the third information;
  • the first device or other network element sends consent information and sends third information.
  • the consent information is used to indicate approval of the request of the second device;
  • the second device or other network element sends consent information and sends third information.
  • the consent information is used to indicate approval of the first device's request
  • the target information includes at least one first information related to the first action of the AI module and at least one second information corresponding to the at least one first information.
  • the second device or other network element sends the target information.
  • the first device or other network element sends the target information.
  • the input and output data of the interactive model and part of the model structure information are used to solve the model pairing problem when multi-node deployment models are used for joint inference, and this process does not involve interacting with all model implementation details, so the model can be avoided The problem of information leakage.
  • the method for the first device and the second device to align the third information includes at least one of the following:
  • the first AI module and/or the second AI module may use the model associated with the third information
  • the first AI module and/or the second AI module may use the model associated with the third information;
  • the first AI module and/or the second AI module may use the model associated with the third information.
  • the above-mentioned first duration is determined by any of the following: carried by the third information, carried by the confirmation information of the third information, carried by other associated information or signaling of the third information, carried by The agreement stipulates that it is determined by the capabilities of the first device or the second device.
  • Step 203 The second device receives the second information from the first device.
  • the above-mentioned second information is information obtained by the first device inputting the first information into the first AI module.
  • Step 204 The second device inputs the second information to the second AI module to obtain the first information and/or related information of the first information.
  • Embodiments of the present application provide an information transmission method by aligning model-related information of the first AI module and/or the second AI module in advance before the first AI module and the second AI module perform training, updating, and/or inference. It solves the model pairing problem when deploying models on multiple nodes for inference (such as joint inference), so that the model can be distributed on different nodes.
  • models on the same node perform joint inference on information, that is, when the first device infers the first information through the first AI module, and the second device infers the second information through the second AI module, there is no need to inform all the details of the model. Only the target nodes can perform joint inference, which ensures the inference performance of the model and avoids the leakage of model information.
  • the execution subject may be an information transmission device.
  • the information transmission method performed by the first device and the second device is used as an example to illustrate the information transmission device provided by the embodiment of the present application.
  • Figure 5 shows a possible structural diagram of the information transmission device involved in the embodiment of the present application.
  • the information transmission device is applied to the first device.
  • the information transmission device 70 may include: a processing module 71 and a sending module 72 .
  • the processing module 71 is used to input the first information to the first AI module to obtain the second information.
  • the sending module 72 is used to send the second information obtained by the processing module 71 to the second device.
  • the second information is used by the second device to input the second information to the second AI module to obtain the first information and/or the first information. related information.
  • the first device and the second device align the third information;
  • the third information includes model information of the first AI module and/or the second AI module;
  • the first action includes at least one of the following: training, updating, and inference.
  • Embodiments of the present application provide an information transmission device that aligns model-related information of the first AI module and/or the second AI module in advance before the first AI module and the second AI module perform training, updating, and/or inference. It solves the problem of model pairing when deploying models on multiple nodes for reasoning (such as joint reasoning), so that models distributed in different nodes can jointly reason about information, that is, the information transmission device infers the first information through the first AI module, and the third When the second device infers the second information through the second AI module, it does not need to inform the target node of all the details of the model to perform joint inference. This ensures the inference performance of the model and avoids the leakage of model information.
  • the above-mentioned first information includes at least one of the following: channel information and beam quality information; the above-mentioned second information includes at least one of the following: PMI, predicted beam information or beam indication.
  • the above-mentioned first AI module and/or second AI module are obtained according to at least one of the following:
  • the first device is trained based on target information from the second device or other network elements
  • the second device is trained based on target information from the first device or other network elements
  • first AI module and/or second AI module are updated or adjusted according to at least one of the following:
  • the first device updates or adjusts based on the target information from the second device or other network elements
  • the second device updates or adjusts based on the target information from the first device or other network elements
  • the target information includes at least one first information related to the first action of the AI module and at least one second information corresponding to the at least one first information.
  • the above third information specifically includes at least one of the following: structural characteristics of the model, load quantification method of the model, estimation accuracy or output accuracy of the model.
  • the structural characteristics of the above model include at least one of the following: model structure, basic structural characteristics of the model, structural characteristics of the model sub-modules, number of model layers, number of model neurons, model size, and model complexity. , quantitative parameters of model parameters.
  • the basic structural characteristics of the above model include at least one of the following: whether it contains a fully connected structure, whether it contains a convolution structure, whether it contains an LSTM structure, whether it contains an attention structure, and whether it contains a residual structure.
  • the number of neurons in the above model includes at least one of the following: the number of fully connected neurons, the number of convolutional neurons, the number of memory neurons, the number of attention neurons, and the number of residual neurons; And/or, the number of neurons in the above-mentioned model includes at least one of the following: the number of neurons of all types, the number of neurons of a single type, the number of neurons of the entire model, the number of neurons of a single layer or several layers.
  • the quantization parameters of the above model parameters include at least one of the following: a quantization method of the model parameters, and the number of quantization bits of a single neuron parameter; wherein the quantization method of the model parameters includes at least one of the following: uniform Quantization method, non-uniform quantization method, weight sharing quantization method or group quantization method, parameter coding quantization method, transform domain quantization method, product quantization method.
  • the above load quantification method includes at least one of the following: quantization method, dimensionality of features before and after quantization, and quantization method used during quantization.
  • the quantization method used in the above quantization includes at least one of the following: synchronizing the codebook content and codebook usage method is required when using the codebook for quantization, and synchronizing the quantization rules is required when using specific rules for quantization.
  • the above quantization rule includes at least one of the following: N quantization intervals and quantization methods, where N is a positive integer; wherein the quantization method includes at least one of the following: uniform quantization method, non-uniform quantization method, Weight sharing quantization method or group quantization method, parameter coding quantization method, transform domain quantization method, product quantization method.
  • the synchronization method when synchronizing the codebook content and codebook usage method, and/or synchronizing the quantization rules, includes any of the following: selecting the synchronization feedback representative from a predefined method set. Select the collection sequence number of the method and directly send the codebook content.
  • the method for the first device and the second device to align the third information includes at least one of the following:
  • the first device or other network element When the first device or other network element sends the target information to the second device, it also sends the third information;
  • the second device or other network element When the second device or other network element sends the target information to the first device, it also sends the third information;
  • the first device or other network element Before the first device or other network element sends the target information to the second device, the first device or other network element sends the third information;
  • the second device or other network element Before the second device or other network element sends the target information to the first device, the second device or other network element sends the third information;
  • the second device When the second device requests target information, the second device sends the third information
  • the first device When the first device requests target information, the first device sends the third information;
  • the first device or other network element sends consent information and sends third information.
  • the consent information is used to indicate approval of the request of the second device;
  • the second device or other network element sends consent information and sends third information.
  • the consent information is used to indicate approval of the first device's request
  • the target information includes at least one first information related to the first action of the AI module and at least one second information corresponding to the at least one first information.
  • the second device or other network element after the first device sends acknowledgment information for the third information, the second device or other network element sends the target information; and/or, after the second device sends acknowledgment information for the third information Afterwards, the first device or other network element sends the target information.
  • the method for the first device and the second device to align the third information includes at least one of the following:
  • the first AI module and/or the second AI module may use the model associated with the third information
  • the first AI module and/or the second AI module may use the model associated with the third information;
  • the first AI module and/or the second AI may use a third information association model.
  • the above-mentioned first duration is determined by any of the following: carried by the third information, carried by the confirmation information of the third information, carried by other related information or signaling of the third information, carried by the protocol Agreed, determined by the capabilities of the first device or the second device.
  • the information transmission device provided by the embodiment of the present application can implement each process implemented by the first device in the above method embodiment, and achieve the same technical effect. To avoid duplication, the details will not be described here.
  • Figure 6 shows a possible structural schematic diagram of the information transmission device involved in the embodiment of the present application.
  • the information transmission device is applied to the second device.
  • the information transmission device 80 may include: a receiving module 81 and a processing module 82 .
  • the receiving module 81 is used to receive second information from the first device.
  • the second information is the information obtained by the first device inputting the first information into the first AI module.
  • the processing module 82 is configured to input the second information received by the receiving module 81 into the second AI module to obtain the first information and/or related information of the first information.
  • the first device and the second device align the third information;
  • the third information includes model information of the first AI module and/or the second AI module;
  • the first action includes at least one of the following: training, updating, and inference.
  • Embodiments of the present application provide an information transmission device that aligns model-related information of the first AI module and/or the second AI module in advance before the first AI module and the second AI module perform training, updating, and/or inference. It solves the model pairing problem when deploying models on multiple nodes for inference (such as joint inference), allowing models distributed on different nodes to perform joint inference on information, that is, the first device infers the first information through the first AI module, and the information When the transmission device infers the second information through the second AI module, it is not necessary to inform the target node of all the details of the model in order to perform joint inference. This ensures the inference performance of the model and avoids the leakage of model information.
  • inference such as joint inference
  • the above-mentioned first information includes at least one of the following: channel information and beam quality information; the above-mentioned second information includes at least one of the following: PMI, predicted beam information or beam indication.
  • the above-mentioned first AI module and/or second AI module are obtained according to at least one of the following:
  • the first device is trained based on target information from the second device or other network elements
  • the second device is trained based on target information from the first device or other network elements
  • first AI module and/or second AI module are updated or adjusted according to at least one of the following:
  • the first device updates or adjusts based on the target information from the second device or other network elements
  • the second device updates or adjusts based on the target information from the first device or other network elements
  • the target information includes at least one first information related to the first action of the AI module and at least one second information corresponding to the at least one first information.
  • the above third information specifically includes at least one of the following: structural characteristics of the model, load quantification method of the model, estimation accuracy or output accuracy of the model.
  • the structural characteristics of the above model include at least one of the following: model structure, basic structural characteristics of the model, structural characteristics of the model sub-modules, number of model layers, number of model neurons, model size, and model complexity. , quantitative parameters of model parameters.
  • the basic structural characteristics of the above model include at least one of the following: whether it contains a fully connected structure, whether it contains a convolution structure, whether it contains an LSTM structure, whether it contains an attention structure, and whether it contains a residual structure.
  • the number of neurons in the above model includes at least one of the following: the number of fully connected neurons, the number of convolutional neurons, the number of memory neurons, the number of attention neurons, and the number of residual neurons; and /Or, the number of neurons in the above model includes at least one of the following: the number of neurons of all types, a single type The number of neurons, the number of neurons in the entire model, the number of neurons in a single layer or several layers.
  • the quantization parameters of the above model parameters include at least one of the following: a quantization method of the model parameters, and the number of quantization bits of a single neuron parameter; wherein the quantization method of the model parameters includes at least one of the following: uniform Quantization method, non-uniform quantization method, weight sharing quantization method or group quantization method, parameter coding quantization method, transform domain quantization method, product quantization method.
  • the above load quantification method includes at least one of the following: quantization method, dimensionality of features before and after quantization, and quantization method used during quantization.
  • the quantization method used in the above quantization includes at least one of the following: synchronizing the codebook content and codebook usage method is required when using the codebook for quantization, and synchronizing the quantization rules is required when using specific rules for quantization.
  • the above quantization rule includes at least one of the following: N quantization intervals and quantization methods, where N is a positive integer; wherein the quantization method includes at least one of the following: uniform quantization method, non-uniform quantization method, Weight sharing quantization method or group quantization method, parameter coding quantization method, transform domain quantization method, product quantization method.
  • the synchronization method when synchronizing the codebook content and codebook usage method, and/or synchronizing the quantization rules, includes any of the following: selecting the synchronization feedback representative from a predefined method set. Select the collection sequence number of the method and directly send the codebook content.
  • the method for the first device and the second device to align the third information includes at least one of the following:
  • the first device or other network element When the first device or other network element sends the target information to the second device, it also sends the third information;
  • the second device or other network element When the second device or other network element sends the target information to the first device, it also sends the third information;
  • the first device or other network element Before the first device or other network element sends the target information to the second device, the first device or other network element sends the third information;
  • the second device or other network element Before the second device or other network element sends the target information to the first device, the second device or other network element sends the third information;
  • the second device When the second device requests target information, the second device sends the third information
  • the first device When the first device requests target information, the first device sends the third information;
  • the first device or other network element sends consent information and sends third information.
  • the consent information is used to indicate approval of the request of the second device;
  • the second device or other network element sends consent information and sends third information.
  • the consent information is used to indicate approval of the first device's request
  • the target information includes at least one first information related to the first action of the AI module and at least one second information corresponding to the at least one first information.
  • the second device or other network element after the first device sends acknowledgment information for the third information, the second device or other network element sends the target information; and/or, after the second device sends acknowledgment information for the third information Afterwards, the first device or other network element sends the target information.
  • the method for the first device and the second device to align the third information includes at least one of the following:
  • the first AI module and/or the second AI module may use the model associated with the third information
  • the first AI module and/or the second AI module may use the model associated with the third information;
  • the first AI module and/or the second AI module may use the model associated with the third information.
  • the above-mentioned first duration is determined by any of the following: carried by the third information, carried by the confirmation information of the third information, carried by other related information or signaling of the third information, carried by the protocol Agreed, determined by the capabilities of the first device or the second device.
  • the information transmission device provided by the embodiment of the present application can implement each process implemented by the second device in the above method embodiment, and achieve the same technical effect. To avoid duplication, the details will not be described here.
  • the information transmission device in the embodiment of the present application may be a UE, such as a UE with an operating system, or may be a component in the UE, such as an integrated circuit or chip.
  • the UE may be a terminal or other equipment other than the terminal.
  • the UE may include but is not limited to the types of UE 11 listed above, and other devices may be servers, network attached storage (Network Attached Storage, NAS), etc., which are not specifically limited in the embodiment of this application.
  • NAS Network Attached Storage
  • this embodiment of the present application also provides a communication device 5000, which includes a processor 5001 and a memory 5002.
  • the memory 5002 stores programs or instructions that can be run on the processor 5001, such as , when the communication device 5000 is the first device, when the program or instruction is executed by the processor 5001, the steps of the above-mentioned first device-side method embodiment are implemented, and the same technical effect can be achieved. To avoid duplication, they will not be described again here. .
  • the communication device 5000 is a second device, when the program or instruction is executed by the processor 5001, each step of the above-mentioned second device-side method embodiment is implemented, and the same technical effect can be achieved. To avoid duplication, the details are not repeated here.
  • the above-mentioned first device may be a UE or a network-side device; the above-mentioned second device may be a network-side device or a UE.
  • the following embodiments illustrate the hardware structures of the UE and the network side equipment.
  • An embodiment of the present application also provides a UE, which includes a processor and a communication interface.
  • the processor is configured to input the first information to the first AI module to obtain the second information.
  • the communication interface is used to send the second information to the second device, and the second information is used by the second device to input the second information to the second AI module to obtain the first information and/or related information of the first information.
  • the first device and the second device align third information, where the third information includes model information of the first AI module and/or the second AI module.
  • the first action includes at least one of the following: training, updating, and inference.
  • This UE embodiment corresponds to the above-mentioned first device-side method embodiment.
  • Each implementation process and implementation manner of the above-mentioned method embodiment can be applied to this UE embodiment, and can achieve the same technical effect.
  • Embodiments of the present application also provide a UE, including a processor and a communication interface.
  • the communication interface is used to receive second information from a first device.
  • the second information is used by the first device to input the first information into the first AI module. information obtained.
  • the processor is configured to input the second information to the second AI module to obtain the first information and/or related information of the first information.
  • the first device and the second device align the third information, and the third information includes model information of the first AI module and/or the second AI module
  • the first action includes at least one of the following: training, updating, and inference.
  • This UE embodiment corresponds to the above-mentioned second device-side method embodiment.
  • Each implementation process and implementation manner of the above-mentioned method embodiment can be applied to this UE embodiment, and can achieve the same technical effect.
  • FIG. 8 is a schematic diagram of the hardware structure of a UE that implements an embodiment of the present application.
  • the UE 7000 includes but is not limited to: radio frequency unit 7001, network module 7002, audio output unit 7003, input unit 7004, sensor 7005, display unit 7006, user input unit 7007, interface unit 7008, memory 7009, processor 7010, etc. At least some parts.
  • the UE 7000 may also include a power supply (such as a battery) that supplies power to various components.
  • the power supply may be logically connected to the processor 7010 through a power management system, thereby managing charging, discharging, and power consumption through the power management system. Management and other functions.
  • the structure of the UE shown in Figure 8 does not constitute a limitation on the UE.
  • the UE may include more or less components than those shown in the figure, or some components may be combined, or different components may be used. The layout of the components will not be described in detail here.
  • the input unit 7004 may include a graphics processing unit (Graphics Processing Unit, GPU) 70041 and a microphone 70042.
  • the graphics processor 70041 is responsible for the image capture device (GPU) in the video capture mode or the image capture mode. Process the image data of still pictures or videos obtained by cameras (such as cameras).
  • the display unit 7006 may include a display panel 70061, which may be configured in the form of a liquid crystal display, an organic light emitting diode, or the like.
  • the user input unit 7007 includes at least one of a touch panel 70071 and other input devices 70072. Touch panel 70071, also known as touch screen.
  • the touch panel 70071 may include two parts: a touch detection device and a touch controller.
  • Other input devices 70072 may include but are not limited to physical keyboards, function keys (such as volume control keys, switch keys, etc.), trackballs, mice, and joysticks, which will not be described again here.
  • the radio frequency unit 7001 after receiving downlink data from the network side device, can transmit it to the processor 7010 for processing; in addition, the radio frequency unit 7001 can send uplink data to the network side device.
  • the radio frequency unit 7001 includes, but is not limited to, an antenna, amplifier, transceiver, coupler, low noise amplifier, duplexer, etc.
  • Memory 7009 may be used to store software programs or instructions as well as various data.
  • the memory 7009 may mainly include a first storage area for storing programs or instructions and a second storage area for storing data, wherein the first storage area may store an operating system, an application program or instructions required for at least one function (such as a sound playback function, Image playback function, etc.) etc.
  • memory 7009 may include volatile memory or nonvolatile memory, or memory 7009 may include both volatile and nonvolatile memory.
  • non-volatile memory can be read-only memory (Read-Only Memory, ROM), programmable read-only memory (Programmable ROM, PROM), erasable programmable read-only memory (Erasable PROM, EPROM), electrically removable memory.
  • Volatile memory can be random access memory (Random Access Memory, RAM), 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, DDRSDRAM), enhanced synchronous dynamic random access memory (Enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (Synch link DRAM) , SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DRRAM).
  • RAM Random Access Memory
  • SRAM static random access memory
  • DRAM dynamic random access memory
  • synchronous dynamic random access memory Synchronous DRAM, SDRAM
  • Double data rate synchronous dynamic random access memory Double Data Rate SDRAM, DDRSDRAM
  • enhanced SDRAM synchronous dynamic random access memory
  • Synch link DRAM synchronous link dynamic random access memory
  • SLDRAM direct memory bus random access memory
  • Direct Rambus RAM Direct Rambus RAM
  • the processor 7010 may include one or more processing units; optionally, the processor 7010 integrates an application processor and a modem processor, where the application processor mainly handles operations related to the operating system, user interface, application programs, etc., Modem processors mainly process wireless communication signals, such as baseband processors. It can be understood that the above modem processor may not be integrated into the processor 7010.
  • the processor 7010 is used to input the first information to the first AI module to obtain the second information.
  • the radio frequency unit 7001 is used to send second information to the second device.
  • the second information is used by the second device to input the second information to the second AI module to obtain the first information and/or related information of the first information.
  • the first device and the second device align third information, where the third information includes model information of the first AI module and/or the second AI module.
  • the first action includes at least one of the following: training, updating, and inference.
  • Embodiments of the present application provide a UE that solves the problem by aligning model-related information of the first AI module and/or the second AI module in advance before the first AI module and the second AI module perform training, updating and/or inference.
  • the model pairing problem when deploying models on multiple nodes for inference (such as joint inference) enables models distributed on different nodes to perform joint inference on information, that is, the UE infers the first information through the first AI module, and the second When the device infers the second information through the second AI module, it does not need to inform the target node of all the details of the model to perform joint inference. This ensures the inference performance of the model and avoids the leakage of model information.
  • the UE provided by the embodiment of the present application can implement each process implemented by the first device in the above method embodiment, and achieve the same technical effect. To avoid duplication, the details will not be described here.
  • the radio frequency unit 7001 is configured to receive second information from the first device, where the second information is the information obtained by the first device inputting the first information into the first AI module.
  • the processor 7010 is configured to input the second information to the second AI module to obtain the first information and/or related information of the first information.
  • the first device and the second device align the third information;
  • the third information includes model information of the first AI module and/or the second AI module;
  • the first action includes at least one of the following: training, updating, and inference.
  • Embodiments of the present application provide a UE that solves the problem by aligning model-related information of the first AI module and/or the second AI module in advance before the first AI module and the second AI module perform training, updating and/or inference.
  • the model pairing problem when deploying models on multiple nodes for inference (such as joint inference) enables models distributed on different nodes to jointly infer information, that is, the first device infers the first information through the first AI module, and the UE infers the first information through the first AI module.
  • the second AI module infers the second information, it is not necessary to inform the target node of all the details of the model in order to perform joint inference. This ensures the inference performance of the model and avoids the leakage of model information.
  • the UE provided by the embodiments of this application can implement each process implemented by the second device in the above method embodiment, and achieve the same technical effect. To avoid duplication, details will not be described here.
  • An embodiment of the present application also provides a network side device, including a processor and a communication interface.
  • the processor is configured to input the first information to the first AI module to obtain the second information.
  • the communication interface is used to send the second information to the second device, and the second information is used by the second device to input the second information to the second AI module to obtain the first information and/or related information of the first information.
  • the first device and the second device align third information, where the third information includes model information of the first AI module and/or the second AI module.
  • the first action includes at least one of the following: training, updating, and inference.
  • This network-side device embodiment corresponds to the above-mentioned first device method embodiment.
  • Each implementation process and implementation manner of the above-mentioned method embodiment can be applied to this network-side device embodiment, and can achieve the same technical effect.
  • An embodiment of the present application also provides a network side device, including a processor and a communication interface.
  • the communication interface is used to receive second information from the first device.
  • the second information is for the first device to input the first information to the first AI. information obtained from the module.
  • the processor is configured to input the second information to the second AI module to obtain the first information and/or related information of the first information.
  • the first device and the second device align the third information, and the third information includes model information of the first AI module and/or the second AI module
  • the first action includes at least one of the following: training, updating, and inference.
  • This network-side device embodiment corresponds to the above-mentioned second device method embodiment.
  • Each implementation process and implementation manner of the above-mentioned method embodiment can be applied to this network-side device embodiment, and can achieve the same technical effect.
  • the embodiment of the present application also provides a network side device.
  • the network side device 600 includes: an antenna 601 , a radio frequency device 602 , a baseband device 603 , a processor 604 and a memory 605 .
  • the antenna 601 is connected to the radio frequency device 602.
  • the radio frequency device 602 receives information through the antenna 601 and sends the received information to the baseband device 603 for processing.
  • the baseband device 603 processes the information to be sent and sends it to the radio frequency device 602.
  • the radio frequency device 602 processes the received information and then sends it out through the antenna 601.
  • the method performed by the network side device in the above embodiment can be implemented in the baseband device 603, which includes a baseband processor.
  • the processor 604 is used to input the first information to the first AI module to obtain the second information.
  • Radio frequency device 602 is used to send second information to the second device, and the second information is used for the second device to input the second information to The second AI module obtains the first information and/or related information of the first information.
  • the first device and the second device align third information, where the third information includes model information of the first AI module and/or the second AI module.
  • the first action includes at least one of the following: training, updating, and inference.
  • Embodiments of the present application provide a network-side device that aligns model-related information of the first AI module and/or the second AI module in advance before the first AI module and the second AI module perform training, updating, and/or inference. It solves the model pairing problem when deploying models on multiple nodes for inference (such as joint inference), allowing models distributed on different nodes to perform joint inference on information, that is, the network side device infers the first information through the first AI module, and the third When the second device infers the second information through the second AI module, it does not need to inform the target node of all the details of the model to perform joint inference. This ensures the inference performance of the model and avoids the leakage of model information.
  • inference such as joint inference
  • the network side device provided by the embodiment of the present application can implement each process implemented by the first device in the above method embodiment, and achieve the same technical effect. To avoid duplication, the details will not be described here.
  • the radio frequency device 602 is configured to receive second information from the first device, where the second information is the information obtained by the first device inputting the first information into the first AI module.
  • the processor 604 is used to input the second information to the second AI module to obtain the first information and/or related information of the first information.
  • the first device and the second device align the third information, and the third information includes model information of the first AI module and/or the second AI module;
  • the first action includes at least one of the following: training, updating, and inference.
  • Embodiments of the present application provide a network-side device that aligns model-related information of the first AI module and/or the second AI module in advance before the first AI module and the second AI module perform training, updating, and/or inference. It solves the model pairing problem when deploying models on multiple nodes for inference (such as joint inference), allowing models distributed on different nodes to perform joint inference on information, that is, the first device performs inference on the first information through the first AI module, and the network
  • the side device infers the second information through the second AI module, it does not need to inform the target node of all the details of the model to perform joint inference. This ensures the inference performance of the model and avoids the leakage of model information.
  • the network side device provided by the embodiment of the present application can implement each process implemented by the second device in the above method embodiment, and achieve the same technical effect. To avoid duplication, the details will not be described here.
  • the baseband device 603 may include, for example, at least one baseband board on which multiple chips are disposed, as shown in FIG. Program to perform the network device operations shown in the above method embodiments.
  • the network side device may also include a network interface 606, which is, for example, a common public radio interface (CPRI).
  • a network interface 606 which is, for example, a common public radio interface (CPRI).
  • CPRI common public radio interface
  • the network-side device 600 in this embodiment of the present application also includes: instructions or programs stored in the memory 605 and executable on the processor 604.
  • the processor 604 calls the instructions or programs in the memory 605 to execute the above-mentioned modules. method and achieve the same technical effect. To avoid repetition, we will not repeat it here.
  • the embodiment of the present application also provides a network side device.
  • the network side device 800 includes: a processor 801, a network interface 802 and a memory 803.
  • the network interface 802 is, for example, a common public radio interface (CPRI).
  • CPRI common public radio interface
  • the network side device 800 in this embodiment of the present invention also includes: instructions or programs stored in the memory 803 and executable on the processor 801.
  • the processor 801 calls the instructions or programs in the memory 803 to execute the above-mentioned modules. method and achieve the same technical effect. To avoid repetition, we will not repeat it here.
  • Embodiments of the present application also provide a readable storage medium.
  • Programs or instructions are stored on the readable storage medium.
  • the program or instructions are executed by a processor, each process of the above information transmission method embodiment is implemented, and can achieve To achieve the same technical effect, to avoid repetition, we will not repeat them here.
  • the processor is the processor in the communication device described in the above embodiment.
  • the readable storage medium includes computer readable storage media, such as computer read-only memory ROM, random access memory RAM, magnetic disk or optical disk, etc.
  • An embodiment of the present application further provides a chip.
  • the chip includes a processor and a communication interface.
  • the communication interface is coupled to the processor.
  • the processor is used to run programs or instructions to implement various processes of the above method embodiments. , and can achieve the same technical effect, so to avoid repetition, they will not be described again here.
  • chips mentioned in the embodiments of this application may also be called system-on-chip, system-on-a-chip, system-on-chip or system-on-chip, etc.
  • Embodiments of the present application further provide a computer program/program product.
  • the computer program/program product is stored in a storage medium.
  • the computer program/program product is executed by at least one processor to implement each of the above method embodiments.
  • the process can achieve the same technical effect. To avoid repetition, it will not be described again here.
  • An embodiment of the present application also provides a communication system, including: a first device and a second device.
  • the first device can be used to perform the steps of the information transmission method as described above.
  • the second device can be used to perform the steps of the above information transmission method. The steps of the information transmission method described above.
  • the methods of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better. implementation.
  • the technical solution of the present application can be embodied in the form of a computer software product that is essentially or contributes to the existing technology.
  • the computer software product is stored in a storage medium (such as ROM/RAM, disk , CD), including several instructions to cause a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in various embodiments of this application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本申请公开了一种信息传输方法、装置、设备、***及存储介质,属于通信技术领域,本申请实施例的信息传输方法包括:第一设备将第一信息输入到第一AI模块,得到第二信息;第一设备向第二设备发送第二信息,第二信息用于第二设备将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息;其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息;第三信息包括第一AI模块和/或第二AI模块的模型信息;第一动作包括以下至少一项:训练、更新、推理。

Description

信息传输方法、装置、设备、***及存储介质
相关申请的交叉引用
本申请主张在2022年08月12日在中国提交的申请号为202210970370.2的中国专利的优先权,其全部内容通过引用包含于此。
技术领域
本申请属于通信技术领域,具体涉及一种信息传输方法、装置、设备、***及存储介质。
背景技术
网络侧可以向用户设备(User Equipment,UE)发送信道状态信息参考信号(Channel State Information-Reference Signal,CSI-RS),以用于UE进行信道估计。UE根据CSI-RS进行信道估计,计算对应的信道信息,通过码本将预编码矩阵指示(Precoding Matrix Index,PMI)反馈给网络侧,网络侧根据UE反馈的码本信息组合出信道信息,在下一次CSI上报之前,网络侧以此进行数据预编码以及多用户调度。
目前,可以利用人工智能模型或机器学习模型增加CSI反馈,具体过程为:在某个网络节点联合训练/独立训练模型的所有模块(例如编码器和解码器);并将不同模块分别部署在多个不同的网络节点;以及所部署的各模型模块进行联合推理。然而,由于不同网络节点可能会来自不同厂商,而在不同网络节点部署模型时需要将模型的全部细节告知目标节点,因此上述过程会导致模型信息泄露的问题。
发明内容
本申请实施例提供一种信息传输方法、装置、设备、***及存储介质,能够解决在不同网络节点部署模型时需要将模型的全部细节告知目标节点,导致模型信息泄露的问题。
第一方面,提供了一种信息传输方法,该信息传输方法包括:第一设备将第一信息输入到第一人工智能(Artificial Intelligence,AI)模块,得到第二信息;第一设备向第二设备发送第二信息,第二信息用于第二设备将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息;其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息;第三信息包括第一AI模块和/或第二AI模块的模型信息;第一动作包括以下至少一项:训练、更新、推理。
第二方面,提供了一种信息传输装置,应用于第一设备,该信息传输装置包括:处理模块和发送模块。处理模块,用于将第一信息输入到第一AI模块,得到第二信息。发送模块,用于向第二设备发送处理模块得到的第二信息,该第二信息用于第二设备将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息。其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息;第三信息包括第一AI模块和/或第二AI模块的模型信息;第一动作包括以下至少一项:训练、更新、推理。
第三方面,提供了一种信息传输方法,该方法包括:第二设备接收来自第一设备的第二信息,该第二信息为第一设备将第一信息输入到第一AI模块中得到的信息;第二设备将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息; 其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息;第三信息包括第一AI模块和/或第二AI模块的模型信息;第一动作包括以下至少一项:训练、更新、推理。
第四方面,提供了一种信息传输装置,应用于第二设备,该信息传输装置包括:接收模块和处理模块。接收模块,用于接收来自第一设备的第二信息,该第二信息为第一设备将第一信息输入到第一AI模块中得到的信息。处理模块,用于将接收模块接收的第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息。其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息;第三信息包括第一AI模块和/或第二AI模块的模型信息;第一动作包括以下至少一项:训练、更新、推理。
第五方面,提供了一种通信设备,该通信设备包括处理器和存储器,所述存储器存储可在所述处理器上运行的程序或指令,所述程序或指令被所述处理器执行时实现如第一方面所述的方法的步骤。
第六方面,提供了一种通信设备,包括处理器及通信接口,其中,所述处理器用于将第一信息输入到第一AI模块,得到第二信息。所述通信接口用于向第二设备发送第二信息,第二信息用于第二设备将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息。其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息,该第三信息包括第一AI模块和/或第二AI模块的模型信息;第一动作包括以下至少一项:训练、更新、推理。
第七方面,提供了一种通信设备,该通信设备包括处理器和存储器,所述存储器存储可在所述处理器上运行的程序或指令,所述程序或指令被所述处理器执行时实现如第三方面所述的方法的步骤。
第八方面,提供了一种通信设备,包括处理器及通信接口,其中,所述通信接口用于接收来自第一设备的第二信息,该第二信息为第一设备将第一信息输入到第一AI模块中得到的信息。所述处理器用于将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息。其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息,第三信息包括第一AI模块和/或第二AI模块的模型信息;第一动作包括以下至少一项:训练、更新、推理。
第九方面,提供了一种通信***,包括:第一设备及第二设备,所述第一设备可用于执行如第一方面所述的信息传输方法的步骤,所述第二设备可用于执行如第三方面所述的信息传输方法的步骤。
第十方面,提供了一种可读存储介质,所述可读存储介质上存储程序或指令,所述程序或指令被处理器执行时实现如第一方面所述的方法的步骤,或者实现如第三方面所述的方法的步骤。
第十一方面,提供了一种芯片,所述芯片包括处理器和通信接口,所述通信接口和所述处理器耦合,所述处理器用于运行程序或指令,实现如第一方面所述的方法,或实现如第三方面所述的方法。
第十二方面,提供了一种计算机程序/程序产品,所述计算机程序/程序产品被存储在存储介质中,所述计算机程序/程序产品被至少一个处理器执行以实现如第一方面所述的信息传输方法的步骤,或者实现如第三方面所述的信息传输方法的步骤。
在本申请实施例中,在第一AI模块和第二AI模块进行训练、更新和/或推理之前,通过提前对齐第一AI模块和/或第二AI模块的模型相关信息,解决了多节点部署模型进行推理(例如联合推理)时的模型配对问题,使得分布在不同节点的模型对信息进行联合推理,即第一设备通过第一AI模块对第一信息进行推理、以及第二设 备通过第二AI模块对第二信息进行推理时,无需将模型的全部细节告知目标节点才能进行联合推理,如此在保证模型的推理性能的同时,也避免了模型信息的泄露。
附图说明
图1是本申请实施例提供的一种无线通信***的架构示意图;
图2是相关技术提供的一种神经网络的结构示意图;
图3是相关技术提供的一种神经元的结构示意图;
图4是本申请实施例提供的一种信息传输方法的流程图;
图5是本申请实施例提供的一种信息传输装置的结构示意图之一;
图6是本申请实施例提供的一种信息传输装置的结构示意图之二;
图7是本申请实施例提供的一种通信设备的硬件结构示意图;
图8是本申请实施例提供的一种UE的硬件结构示意图;
图9是本申请实施例提供的一种网络侧设备的硬件结构示意图之一;
图10是本申请实施例提供的一种网络侧设备的硬件结构示意图之二。
具体实施方式
下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行清楚描述,显然,所描述的实施例是本申请一部分实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员所获得的所有其他实施例,都属于本申请保护的范围。
本申请的说明书和权利要求书中的术语“第一”、“第二”等是用于区别类似的对象,而不用于描述特定的顺序或先后次序。应该理解这样使用的术语在适当情况下可以互换,以便本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施,且“第一”、“第二”所区别的对象通常为一类,并不限定对象的个数,例如第一对象可以是一个,也可以是多个。此外,说明书以及权利要求中“和/或”表示所连接对象的至少其中之一,字符“/”一般表示前后关联对象是一种“或”的关系。
值得指出的是,本申请实施例所描述的技术不限于长期演进型(Long Term Evolution,LTE)/LTE的演进(LTE-Advanced,LTE-A)***,还可用于其他无线通信***,诸如码分多址(Code Division Multiple Access,CDMA)、时分多址(Time Division Multiple Access,TDMA)、频分多址(Frequency Division Multiple Access,FDMA)、正交频分多址(Orthogonal Frequency Division Multiple Access,OFDMA)、单载波频分多址(Single-carrier Frequency Division Multiple Access,SC-FDMA)和其他***。本申请实施例中的术语“***”和“网络”常被可互换地使用,所描述的技术既可用于以上提及的***和无线电技术,也可用于其他***和无线电技术。以下描述出于示例目的描述了新空口(New Radio,NR)***,并且在以下大部分描述中使用NR术语,但是这些技术也可应用于NR***应用以外的应用,如第6代(6th Generation,6G)通信***。
图1示出本申请实施例可应用的一种无线通信***的框图。无线通信***包括UE11和网络侧设备12。其中,UE11可以是手机、平板电脑(Tablet Personal Computer)、膝上型电脑(Laptop Computer)或称为笔记本电脑、个人数字助理(Personal Digital Assistant,PDA)、掌上电脑、上网本、超级移动个人计算机(ultra-mobile personal computer,UMPC)、移动上网装置(Mobile Internet Device,MID)、增强现实(augmented reality,AR)/虚拟现实(virtual reality,VR)设备、机器人、可穿戴式设备(Wearable Device)、车载设备(VUE)、行人终端(PUE)、智能家居(具有无线通信功能的家居设备,如冰箱、电视、洗衣机或者家具等)、游戏 机、个人计算机(personal computer,PC)、柜员机或者自助机等终端侧设备,可穿戴式设备包括:智能手表、智能手环、智能耳机、智能眼镜、智能首饰(智能手镯、智能手链、智能戒指、智能项链、智能脚镯、智能脚链等)、智能腕带、智能服装等。需要说明的是,在本申请实施例并不限定UE11的具体类型。网络侧设备12可以包括接入网设备或核心网设备,其中,接入网设备12也可以称为无线接入网设备、无线接入网(Radio Access Network,RAN)、无线接入网功能或无线接入网单元。接入网设备12可以包括基站、WLAN接入点或WiFi节点等,基站可被称为节点B、演进节点B(eNB)、接入点、基收发机站(Base Transceiver Station,BTS)、无线电基站、无线电收发机、基本服务集(Basic Service Set,BSS)、扩展服务集(Extended Service Set,ESS)、家用B节点、家用演进型B节点、发送接收点(Transmitting Receiving Point,TRP)或所述领域中其他某个合适的术语,只要达到相同的技术效果,所述基站不限于特定技术词汇,需要说明的是,在本申请实施例中仅以NR***中的基站为例进行介绍,并不限定基站的具体类型。核心网设备可以包含但不限于如下至少一项:核心网节点、核心网功能、移动管理实体(Mobility Management Entity,MME)、接入移动管理功能(Access and Mobility Management Function,AMF)、会话管理功能(Session Management Function,SMF)、用户平面功能(User Plane Function,UPF)、策略控制功能(Policy Control Function,PCF)、策略与计费规则功能单元(Policy and Charging Rules Function,PCRF)、边缘应用服务发现功能(Edge Application Server Discovery Function,EASDF)、统一数据管理(Unified Data Management,UDM),统一数据仓储(Unified Data Repository,UDR)、归属用户服务器(Home Subscriber Server,HSS)、集中式网络配置(Centralized network configuration,CNC)、网络存储功能(Network Repository Function,NRF),网络开放功能(Network Exposure Function,NEF)、本地NEF(Local NEF,或L-NEF)、绑定支持功能(Binding Support Function,BSF)、应用功能(Application Function,AF)等。需要说明的是,在本申请实施例中仅以NR***中的核心网设备为例进行介绍,并不限定核心网设备的具体类型。
下面对本申请实施例提供的一种信息传输方法、装置、设备、***及存储介质中涉及的一些概念和/或术语做一下解释说明。
一、人工智能(AI)
AI是将人工智能融入无线通信网络,显著提升吞吐量、时延以及用户容量等技术指标是未来的无线通信网络的重要任务。AI模块有多种实现方式,例如神经网络、决策树、支持向量机、贝叶斯分类器等。本申请实施例以神经网络为例进行说明,但是并不限定AI模块的具体类型。
示例性地,如图2所示,为一个神经网络的示意图。该神经网络包括输入层、隐层和输出层,X1,X2,…,Xn为输入,Y为输出。
其中,神经网络由神经元组成,如图3所示,为神经元的示意图。其中,a1,a2,…aK为输入,w为权值(乘性系数),b为偏置(加性系数),z=a1w1+…+akwk+…+aKwK+b,σ(z)为激活函数。通常,激活函数包括Sigmoid、tanh、ReLU(Rectified Linear Unit、线性整流函数、修正线性单元)等。
神经网络的参数通过梯度优化算法进行优化。梯度优化算法是一类最小化或者最大化目标函数(也可以称为损失函数)的算法,而目标函数是模型参数和数据的数学组合。例如给定数据X和其对应的标签Y,可以构建一个神经网络模型f(.),有了模型后,可以根据输入x就可以得到预测输出f(x),并且可以计算出预测值和真实值之间的差距(f(x)-Y),这个就是损失函数。找到合适的W,b使上述的损失函数的值达到 最小,损失值越小,则说明模型越接近于真实情况。
目前常见的优化算法,为基于BP(error Back Propagation,误差反向传播)算法。BP算法的基本思想是,学习过程由信号的正向传播与误差的反向传播两个过程组成。正向传播时,输入样本从输入层传入,经各隐层逐层处理后,传向输出层。若输出层的实际输出与期望的输出不符,则转入误差的反向传播阶段。误差反传是将输出误差以某种形式通过隐层向输入层逐层反传,并将误差分摊给各层的所有单元,从而获得各层单元的误差信号,此误差信号即作为修正各单元权值的依据。这种信号正向传播与误差反向传播的各层权值调整过程,是周而复始地进行的。权值不断调整的过程,也就是网络的学习训练过程。此过程一直进行到网络输出的误差减少到可接受的程度,或进行到预先设定的学习次数为止。
常见的优化算法有梯度下降(Gradient Descent)、随机梯度下降(Stochastic Gradient Descent,SGD)、mini-batch gradient descent(小批量梯度下降)、动量法(Momentum)、Nesterov(发明者的名字,具体为带动量的随机梯度下降)、Adagrad(ADAptive GRADient descent,自适应梯度下降)、Adadelta、RMSprop(root mean square prop,均方根误差降速)、Adam(Adaptive Moment Estimation,自适应动量估计)等。
这些优化算法在误差反向传播时,都是根据损失函数得到的误差/损失,对当前神经元求导数/偏导,加上学习速率、之前的梯度/导数/偏导等影响,得到梯度,将梯度传给上一层。
二、信道状态信息(CSI)反馈
准确的CSI对信道容量的至关重要。尤其是对于多天线***来讲,发送端可以根据CSI优化信号的发送,使其更加匹配信道的状态。如:信道质量指示(Channel Quality Indicator,CQI)可以用来选择合适的调制编码方案(Modulation And Coding Scheme,MCS)实现链路自适应;预编码矩阵指示(Precoding Matrix Indicator,PMI)可以用来实现特征波束成形(eigen beamforming)从而最大化接收信号的强度,或者用来抑制干扰(例如小区间干扰、多用户之间干扰等)。因此,自从多天线技术(Multi-Input Multi-Output,MIMO)被提出以来,CSI获取一直都是研究热点。
通常,基站在在某个时隙(slot)的某些时频资源上发送CSI-RS,UE根据CSI-RS进行信道估计,计算这个slot上的信道信息,通过码本将PMI反馈给基站,基站根据UE反馈的码本信息组合出信道信息,在下一次CSI上报之前,基站以此进行数据预编码及多用户调度。
为了进一步减少CSI反馈开销,UE可以将每个子带上报PMI改成按照delay上报PMI,由于时延(delay)域的信道更集中,用更少的delay的PMI就可以近似表示全部子带的PMI,即将delay域信息压缩之后再上报。
同样,为了减少开销,基站可以事先对CSI-RS进行预编码,将编码后的CSI-RS发送个终端,UE看到的是经过编码之后的CSI-RS对应的信道,UE只需要在网络侧指示的端口中选择若干个强度较大的端口,并上报这些端口对应的系数即可。
下面结合附图,通过一些实施例及其应用场景对本申请实施例提供的信息传输方法进行详细地说明。
目前,空口AI/ML中涉及在多个网络节点分别训练模型,同时又要求联合使用训练得到的模型进行推理。利用AI/ML增强CSI反馈时需要按如下步骤进行:1)在某个网络结点联合训练模型的所有模块(即编码器与解码器);2)将不同模块分别部署在多个不同的网络节点;3)所部署的各模型模块进行联合推理。
然而,由于不同网络节点可能会来自不同厂商(例如基站、UE通常是属于不同 厂商的产品),而部分厂商不愿意向其他厂商暴露模型细节,但某些用例(如CSI压缩等)要求分布在多个网络节点的模型进行联合推理,因此在不同网络节点部署模型时需要将模型的全部细节告知目标节点,如此上述过程面临模型信息泄露的问题。
为了解决上述问题,本申请实施例中,在第一AI模块和第二AI模块进行训练、更新和/或推理之前,通过提前对齐第一AI模块和/或第二AI模块的模型相关信息,解决了多节点部署模型进行推理(例如联合推理)时的模型配对问题,使得分布在不同节点的模型对信息进行联合推理时,无需将模型的全部细节告知目标节点才能进行联合推理,如此在保证模型的推理性能的同时,也避免了模型信息的泄露。
本申请实施例提供一种信息传输方法,图4示出了本申请实施例提供的一种信息传输方法的流程图。如图4所示,本申请实施例提供的信息传输方法可以包括下述的步骤201至步骤204。
步骤201、第一设备将第一信息输入到第一AI模块,得到第二信息。
本申请实施例中,第一设备可以通过第一AI模块对第一信息进行推理,得到第二信息。
可选地,本申请实施例中,上述第一信息包括以下至少一项:信道信息(例如CSI)、波束质量信息。
可选地,本申请实施例中,上述第二信息包括以下至少一项:PMI、预测的波束信息或波束指示。
可选地,本申请实施例中,上述第一信息包括信道信息,第二信息为PMI。或者,上述第一信息包括波束质量,第二信息为预测的波束信息或波束指示。
步骤202、第一设备向第二设备发送第二信息。
本申请实施例中,上述第二信息用于第二设备将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息。即第二设备可以通过第二AI模块对第二信息进行推理,得到第一信息的相关信息和/或恢复出第一信息。
本申请实施例中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息;第三信息包括第一AI模块和/或第二AI模块的模型信息;第一动作包括以下至少一项:训练、更新、推理。
可选地,本申请实施例中,上述第一设备可以为网络侧设备或UE;上述第二设备可以为网络侧设备或UE。
可选地,本申请实施例中,上述第一设备为网络侧设备,第二设备为UE。或者,上述第一设备为UE,第二设备为网络侧设备。或者,上述第一设备和第二设备为网络侧的不同节点(例如基站、网元)。或者,上述第一设备和第二设备是不同的UE节点。
可选地,本申请实施例中,上述第一信息的相关信息可以包括以下至少一项:预编码矩阵、信道的分解矩阵或向量、信道矩阵的逆矩阵、信道的分解矩阵或向量的逆矩阵或逆向量、变换域的信道信息、秩、秩标识(Rank Index)、层、层标识(Layer Index)、信道质量、信道信噪比、可选波束标识、可选波束的波束质量。
可选地,本申请实施例中,上述信道的分解矩阵或向量中,分解的具体方法为以下任一项:奇异值分解、特征值分解、三角分解。
可选地,本申请实施例中,上述变换域包括以下至少一项:空域、频域、时域、时延域、多普勒域等。或者,上述变换域包括空域、频域、时域、时延域和多普勒域中的至少两个域的组合域,例如时延域和多普勒域组合为时延多普勒域。
可选地,本申请实施例中,上述第一AI模块和/或第二AI模块根据以下至少一项得到:
第一设备根据来自第二设备或其它网元的目标信息训练得到;
第二设备根据来自第一设备或其它网元的目标信息训练得到。
本申请实施例中,上述目标信息包括与AI模块的第一动作相关的至少一个第一信息和与至少一个第一信息对应的至少一个第二信息。
可以理解,在第一AI模块与第二AI模块的训练阶段,第一设备或其它网元将(大量的)第一信息和对应的第二信息发送给第二设备,或者,第二设备或其它网元将(大量的)第一信息与对应的第二信息发送给第一设备。
需要说明的是,第一信息是指某一类信息,至少一个第一信息是指这类信息的至少一个取值或至少一个参数。第二信息同理。
可选地,本申请实施例中,上述第一AI模块和/或第二AI模块根据以下至少一项进行更新或调整:
第一设备根据来自第二设备或其它网元的目标信息进行更新或调整;
第二设备根据来自第一设备或其它网元的目标信息进行更新或调整。
可以理解,对于第一AI模块与第二AI模块的更新或调整,第一设备或其它网元将(大量的)第一信息和对应的第二信息发送给第二设备,或者,第二设备或其它网元将(大量的)第一信息与对应的第二信息发送给第一设备。
本申请实施例中,第一设备和第二设备可以通过交互模型输入输出的数据,以进行第一AI模块和/或第二AI模块的训练,或者进行第一AI模块和/或第二AI模块的更新/调整,以用于分布在不同节点的模型对信息进行推理。
可选地,本申请实施例中,上述第三信息具体包括以下至少一项:模型的结构特征、模型的载荷量化方法、模型的估计精度或输出精度。
可以理解,第一设备和第二设备可以对齐第一AI模块和/或第二AI模块的全部结构特征或部分结构特征。和/或,第一设备和第二设备可以对齐第一AI模块和/或第二AI模块的载荷量化方法。和/或,第一设备和第二设备可以对齐第一AI模块和/或第二AI模块的估计精度/输出精度。
可选地,本申请实施例中,上述模型的结构特征包括以下至少一项:模型结构、模型基本结构特征、模型子模块的结构特征、模型层数、模型神经元数量、模型大小、模型复杂度、模型参数的量化参数。
可以理解,第一设备和第二设备通过对齐第一AI模块和/或第二AI模块的结构特征,以使得第一设备和第二设备使用相同的模型结构。例如,UE和基站对生成上行控制信息(Uplink Control Information,UCI)的模型结构一致,即第三信息直接指示二者的模型结构相同。
可选地,本申请实施例中,上述模型基本结构特征包括以下至少一项:是否包含全连接结构、是否包含卷积结构、是否包含长短期记忆模型(Long-Short Term Memory,LSTM)结构、是否包含注意力结构、是否包含残差结构。
可选地,本申请实施例中,上述模型神经元数量包括以下至少一项:全连接神经元数量、卷积神经元数量、记忆神经元数量、注意力神经元数量、残差神经元数量。
可选地,本申请实施例中,上述模型神经元数量包括以下至少一项:所有类型的神经元数量、单个类型的神经元数量、整个模型的神经元数量、单层或数层的神经元数量。
需要说明的是,所有类型的神经元数量和单个类型的神经元数量可以理解为是一类神经元数量,整个模型的神经元数量和单层或数层的神经元数量可以理解为是一类神经元数量。例如,所有类型的神经元数量和整个模型的神经元数量组合,即第一设备和第二设备需要对齐所有类型、整个模型的神经元数量。或者,例如单个类型的神 经元数量和单层或数层的神经元数量组合,即第一设备和第二设备需要对齐单个类型、单层的神经元,例如第3层的全连接神经元。
可选地,本申请实施例中,上述模型参数的量化参数包括以下至少一项:模型参数的量化方式、单个神经元参数的量化比特数。其中,模型参数的量化方式包括以下至少一项:均匀量化方式、非均匀量化方式、权值共享量化方式或分组量化方式、参数编码的量化方式、变换域量化方式、乘积量化方式。
需要说明的是,权值共享量化方式或分组量化方式可以理解为:将AI参数划分到多个集合,每个集合中的元素共享一个值。
参数编码的量化方式(参数编码法)可以理解为:对浮点数进行编码。例如,包括以下至少一项:有损编码、无损编码等,例如霍夫曼编码。
变换域量化方式(变换域量化法)可以理解为:将浮点数变换到另一个域,例如频域、S域、Z域等,进行上述至少之一的量化操作,然后再反变换回来。
乘积量化(Product Quantization)方式可以理解为:将浮点数划分成多个子空间,并在每个子空间上进行上述至少之一的量化操作。
可选地,本申请实施例中,上述载荷量化方法包括以下至少一项:量化方式、量化前后特征的维数、量化时使用的量化方法。
需要说明的是,此处的载荷(payload)量化方法是指模型如何将输出的浮点数类型特征转化为始于传输的二进制类型反馈信息,与模型的结构特征中模型参数的量化不同。
可选地,本申请实施例中,上述量化方式可以通过第三信息配置,或者也可以由码本配置(即关联的码本使用的是什么量化方式,这里的训练就用什么量化方式),或者还可以根据CSI报告配置确定(即CSI报告配置使用的是什么量化方式,这里的训练就用什么量化方式)。换句话说,码本或CSI报告配置属于第三信息。
可选地,本申请实施例中,上述量化时使用的量化方法包括以下至少一项:使用码本进行量化时需要同步码本内容和码本使用方法、使用特定规则进行量化时需要同步量化规则。
示例性地,码本内容为矩阵本身。例如,5个浮点数量化为10个比特,构建一个[5,2^10]的码本,即行数5、列数2^10的矩阵,码本内容即这个矩阵本身。
又示例性地,码本量化:对于长度为10且取值区间为[0,1]的浮点数向量,前5个浮点数从一个大小为[5,2^10]的码本中选取与自身误差最小的一列向量作为量化结果;后5个浮点数从一个大小为[5,2^15]的码本中选取与自身误差最小的一列向量作为量化结果。最后将所选量化结果在码本中对应的列序号作为反馈的二进制payload信息。
可选地,本申请实施例中,上述量化规则包括以下至少一项:N个量化区间、量化方式,N为正整数。其中,量化方式包括以下至少一项:均匀量化方式、非均匀量化方式、权值共享量化方式或分组量化方式、参数编码的量化方式、变换域量化方式、乘积量化方式。
需要说明的是,针对此处的各种量化方式的说明,可以参见上述实施例中的描述,此处不再赘述。
示例性地,N个量化区间:单个量化区间内N1个浮点数量化为N2个比特。
又示例性地,量化规则:对于长度为10且取值区间为[0,1]的浮点数向量,前5个浮点数每个使用2bit进行均匀量化,后5个浮点数每个使用3bit进行均匀量化。最后以所选区间的序号作为反馈的二进制payload信息。
可选地,本申请实施例中,在同步码本内容和码本使用方法,和/或,同步量化 规则时,同步方法包括以下任一项:从预定义的方法集合中选取同步时反馈代表所选方法的集合序号、直接发送码本内容。
可选地,本申请实施例中,上述第一设备和第二设备对齐第三信息的方式包括以下至少一项:
第一设备或其它网元将目标信息发送给第二设备时,同时发送第三信息;
第二设备或其它网元将目标信息发送给第一设备时,同时发送第三信息;
第一设备或其它网元将目标信息发送给第二设备前,第一设备或其它网元发送第三信息;
第二设备或其它网元将目标信息发送给第一设备前,第二设备或其它网元发送第三信息;
第二设备在请求目标信息时,第二设备发送第三信息;
第一设备在请求目标信息时,第一设备发送第三信息;
第二设备在请求目标信息时,第一设备或其它网元发送同意信息并发送第三信息,同意信息用于指示同意第二设备的请求;
第一设备在请求目标信息时,第二设备或其它网元发送同意信息并发送第三信息,同意信息用于指示同意第一设备的请求;
其中,目标信息包括与AI模块的第一动作相关的至少一个第一信息和与至少一个第一信息对应的至少一个第二信息。
可选地,本申请实施例中,在第一设备发送对第三信息的确认信息之后,第二设备或其它网元发送目标信息。
可选地,本申请实施例中,在第二设备发送对第三信息的确认信息后,第一设备或其它网元发送目标信息。
本申请实施例中,通过交互模型的输入输出数据以及部分模型结构信息,以解决多节点部署模型进行联合推理时的模型配对问题,并且该过程不涉及交互所有的模型实现细节,因此可以避免模型信息泄露的问题。
可选地,本申请实施例中,上述第一设备和第二设备对齐第三信息的方式包括以下至少一项:
在接收到第三信息的一个设备发送第三信息的确认信息后,第一AI模块和/或第二AI模块可使用第三信息关联的模型;
在接收到第三信息的一个设备发送第三信息的确认信息、且经过第一时长后,第一AI模块和/或第二AI模块可使用第三信息关联的模型;
在第三信息的发送时间或接收时间经过第一时长后,第一AI模块和/或第二AI模块可使用第三信息关联的模型。
可选地,本申请实施例中,上述第一时长由以下任一项确定:由第三信息携带、由第三信息的确认信息携带、由第三信息的其它关联信息或信令携带、由协议约定,由第一设备或第二设备的能力确定。
步骤203、第二设备接收来自第一设备的第二信息。
本申请实施例中,上述第二信息为第一设备将第一信息输入到第一AI模块中得到的信息。
步骤204、第二设备将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息。
本申请实施例提供一种信息传输方法,在第一AI模块和第二AI模块进行训练、更新和/或推理之前,通过提前对齐第一AI模块和/或第二AI模块的模型相关信息,解决了多节点部署模型进行推理(例如联合推理)时的模型配对问题,使得分布在不 同节点的模型对信息进行联合推理,即第一设备通过第一AI模块对第一信息进行推理、以及第二设备通过第二AI模块对第二信息进行推理时,无需将模型的全部细节告知目标节点才能进行联合推理,如此在保证模型的推理性能的同时,也避免了模型信息的泄露。
本申请实施例提供的信息传输方法,执行主体可以为信息传输装置。本申请实施例中以第一设备和第二设备执行信息传输方法为例,说明本申请实施例提供的信息传输装置。
图5出了本申请实施例中涉及的信息传输装置的一种可能的结构示意图,该信息传输装置应用于第一设备。如图5所示,信息传输装置70可以包括:处理模块71和发送模块72。
其中,处理模块71,用于将第一信息输入到第一AI模块,得到第二信息。发送模块72,用于向第二设备发送处理模块71得到的第二信息,该第二信息用于第二设备将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息。其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息;第三信息包括第一AI模块和/或第二AI模块的模型信息;第一动作包括以下至少一项:训练、更新、推理。
本申请实施例提供一种信息传输装置,在第一AI模块和第二AI模块进行训练、更新和/或推理之前,通过提前对齐第一AI模块和/或第二AI模块的模型相关信息,解决了多节点部署模型进行推理(例如联合推理)时的模型配对问题,使得分布在不同节点的模型对信息进行联合推理,即信息传输装置通过第一AI模块对第一信息进行推理、以及第二设备通过第二AI模块对第二信息进行推理时,无需将模型的全部细节告知目标节点才能进行联合推理,如此在保证模型的推理性能的同时,也避免了模型信息的泄露。
在一种可能的实现方式中,上述第一信息包括以下至少一项:信道信息、波束质量信息;上述第二信息包括以下至少一项:PMI、预测的波束信息或波束指示。
在一种可能的实现方式中,上述第一AI模块和/或第二AI模块根据以下至少一项得到:
第一设备根据来自第二设备或其它网元的目标信息训练得到;
第二设备根据来自第一设备或其它网元的目标信息训练得到;
或者,上述第一AI模块和/或第二AI模块根据以下至少一项进行更新或调整:
第一设备根据来自第二设备或其它网元的目标信息进行更新或调整;
第二设备根据来自第一设备或其它网元的目标信息进行更新或调整;
其中,目标信息包括与AI模块的第一动作相关的至少一个第一信息和与至少一个第一信息对应的至少一个第二信息。
在一种可能的实现方式中,上述第三信息具体包括以下至少一项:模型的结构特征、模型的载荷量化方法、模型的估计精度或输出精度。
在一种可能的实现方式中,上述模型的结构特征包括以下至少一项:模型结构、模型基本结构特征、模型子模块的结构特征、模型层数、模型神经元数量、模型大小、模型复杂度、模型参数的量化参数。
在一种可能的实现方式中,上述模型基本结构特征包括以下至少一项:是否包含全连接结构、是否包含卷积结构、是否包含LSTM结构、是否包含注意力结构、是否包含残差结构。
在一种可能的实现方式中,上述模型神经元数量包括以下至少一项:全连接神经元数量、卷积神经元数量、记忆神经元数量、注意力神经元数量、残差神经元数量; 和/或,上述模型神经元数量包括以下至少一项:所有类型的神经元数量、单个类型的神经元数量、整个模型的神经元数量、单层或数层的神经元数量。
在一种可能的实现方式中,上述模型参数的量化参数包括以下至少一项:模型参数的量化方式、单个神经元参数的量化比特数;其中,模型参数的量化方式包括以下至少一项:均匀量化方式、非均匀量化方式、权值共享量化方式或分组量化方式、参数编码的量化方式、变换域量化方式、乘积量化方式。
在一种可能的实现方式中,上述载荷量化方法包括以下至少一项:量化方式、量化前后特征的维数、量化时使用的量化方法。
在一种可能的实现方式中,上述量化时使用的量化方法包括以下至少一项:使用码本进行量化时需要同步码本内容和码本使用方法、使用特定规则进行量化时需要同步量化规则。
在一种可能的实现方式中,上述量化规则包括以下至少一项:N个量化区间、量化方式,N为正整数;其中,量化方式包括以下至少一项:均匀量化方式、非均匀量化方式、权值共享量化方式或分组量化方式、参数编码的量化方式、变换域量化方式、乘积量化方式。
在一种可能的实现方式中,在同步码本内容和码本使用方法,和/或,同步量化规则时,同步方法包括以下任一项:从预定义的方法集合中选取同步时反馈代表所选方法的集合序号、直接发送码本内容。
在一种可能的实现方式中,上述第一设备和第二设备对齐第三信息的方式包括以下至少一项:
第一设备或其它网元将目标信息发送给第二设备时,同时发送第三信息;
第二设备或其它网元将目标信息发送给第一设备时,同时发送第三信息;
第一设备或其它网元将目标信息发送给第二设备前,第一设备或其它网元发送第三信息;
第二设备或其它网元将目标信息发送给第一设备前,第二设备或其它网元发送第三信息;
第二设备在请求目标信息时,第二设备发送第三信息;
第一设备在请求目标信息时,第一设备发送第三信息;
第二设备在请求目标信息时,第一设备或其它网元发送同意信息并发送第三信息,同意信息用于指示同意第二设备的请求;
第一设备在请求目标信息时,第二设备或其它网元发送同意信息并发送第三信息,同意信息用于指示同意第一设备的请求;
其中,目标信息包括与AI模块的第一动作相关的至少一个第一信息和与至少一个第一信息对应的至少一个第二信息。
在一种可能的实现方式中,在第一设备发送对第三信息的确认信息之后,第二设备或其它网元发送目标信息;和/或,在第二设备发送对第三信息的确认信息后,第一设备或其它网元发送目标信息。
在一种可能的实现方式中,上述第一设备和第二设备对齐第三信息的方式包括以下至少一项:
在接收到第三信息的一个设备发送第三信息的确认信息后,第一AI模块和/或第二AI模块可使用第三信息关联的模型;
在接收到第三信息的一个设备发送第三信息的确认信息、且经过第一时长后,第一AI模块和/或第二AI模块可使用第三信息关联的模型;
在第三信息的发送时间或接收时间经过第一时长后,第一AI模块和/或第二AI 模块可使用第三信息关联的模型。
在一种可能的实现方式中,上述第一时长由以下任一项确定:由第三信息携带、由第三信息的确认信息携带、由第三信息的其它关联信息或信令携带、由协议约定,由第一设备或第二设备的能力确定。
本申请实施例提供的信息传输装置能够实现上述方法实施例中第一设备实现的各个过程,并达到相同的技术效果,为避免重复,这里不再赘述。
图6出了本申请实施例中涉及的信息传输装置的一种可能的结构示意图,该信息传输装置应用于第二设备。如图6所示,信息传输装置80可以包括:接收模块81和处理模块82。
其中,接收模块81,用于接收来自第一设备的第二信息,该第二信息为第一设备将第一信息输入到第一AI模块中得到的信息。处理模块82,用于将接收模块81接收的第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息。其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息;第三信息包括第一AI模块和/或第二AI模块的模型信息;第一动作包括以下至少一项:训练、更新、推理。
本申请实施例提供一种信息传输装置,在第一AI模块和第二AI模块进行训练、更新和/或推理之前,通过提前对齐第一AI模块和/或第二AI模块的模型相关信息,解决了多节点部署模型进行推理(例如联合推理)时的模型配对问题,使得分布在不同节点的模型对信息进行联合推理,即第一设备通过第一AI模块对第一信息进行推理、以及信息传输装置通过第二AI模块对第二信息进行推理时,无需将模型的全部细节告知目标节点才能进行联合推理,如此在保证模型的推理性能的同时,也避免了模型信息的泄露。
在一种可能的实现方式中,上述第一信息包括以下至少一项:信道信息、波束质量信息;上述第二信息包括以下至少一项:PMI、预测的波束信息或波束指示。
在一种可能的实现方式中,上述第一AI模块和/或第二AI模块根据以下至少一项得到:
第一设备根据来自第二设备或其它网元的目标信息训练得到;
第二设备根据来自第一设备或其它网元的目标信息训练得到;
或者,上述第一AI模块和/或第二AI模块根据以下至少一项进行更新或调整:
第一设备根据来自第二设备或其它网元的目标信息进行更新或调整;
第二设备根据来自第一设备或其它网元的目标信息进行更新或调整;
其中,目标信息包括与AI模块的第一动作相关的至少一个第一信息和与至少一个第一信息对应的至少一个第二信息。
在一种可能的实现方式中,上述第三信息具体包括以下至少一项:模型的结构特征、模型的载荷量化方法、模型的估计精度或输出精度。
在一种可能的实现方式中,上述模型的结构特征包括以下至少一项:模型结构、模型基本结构特征、模型子模块的结构特征、模型层数、模型神经元数量、模型大小、模型复杂度、模型参数的量化参数。
在一种可能的实现方式中,上述模型基本结构特征包括以下至少一项:是否包含全连接结构、是否包含卷积结构、是否包含LSTM结构、是否包含注意力结构、是否包含残差结构。
在一种可能的实现方式中,上述模型神经元数量包括以下至少一项:全连接神经元数量、卷积神经元数量、记忆神经元数量、注意力神经元数量、残差神经元数量;和/或,上述模型神经元数量包括以下至少一项:所有类型的神经元数量、单个类型 的神经元数量、整个模型的神经元数量、单层或数层的神经元数量。
在一种可能的实现方式中,上述模型参数的量化参数包括以下至少一项:模型参数的量化方式、单个神经元参数的量化比特数;其中,模型参数的量化方式包括以下至少一项:均匀量化方式、非均匀量化方式、权值共享量化方式或分组量化方式、参数编码的量化方式、变换域量化方式、乘积量化方式。
在一种可能的实现方式中,上述载荷量化方法包括以下至少一项:量化方式、量化前后特征的维数、量化时使用的量化方法。
在一种可能的实现方式中,上述量化时使用的量化方法包括以下至少一项:使用码本进行量化时需要同步码本内容和码本使用方法、使用特定规则进行量化时需要同步量化规则。
在一种可能的实现方式中,上述量化规则包括以下至少一项:N个量化区间、量化方式,N为正整数;其中,量化方式包括以下至少一项:均匀量化方式、非均匀量化方式、权值共享量化方式或分组量化方式、参数编码的量化方式、变换域量化方式、乘积量化方式。
在一种可能的实现方式中,在同步码本内容和码本使用方法,和/或,同步量化规则时,同步方法包括以下任一项:从预定义的方法集合中选取同步时反馈代表所选方法的集合序号、直接发送码本内容。
在一种可能的实现方式中,上述第一设备和第二设备对齐第三信息的方式包括以下至少一项:
第一设备或其它网元将目标信息发送给第二设备时,同时发送第三信息;
第二设备或其它网元将目标信息发送给第一设备时,同时发送第三信息;
第一设备或其它网元将目标信息发送给第二设备前,第一设备或其它网元发送第三信息;
第二设备或其它网元将目标信息发送给第一设备前,第二设备或其它网元发送第三信息;
第二设备在请求目标信息时,第二设备发送第三信息;
第一设备在请求目标信息时,第一设备发送第三信息;
第二设备在请求目标信息时,第一设备或其它网元发送同意信息并发送第三信息,同意信息用于指示同意第二设备的请求;
第一设备在请求目标信息时,第二设备或其它网元发送同意信息并发送第三信息,同意信息用于指示同意第一设备的请求;
其中,目标信息包括与AI模块的第一动作相关的至少一个第一信息和与至少一个第一信息对应的至少一个第二信息。
在一种可能的实现方式中,在第一设备发送对第三信息的确认信息之后,第二设备或其它网元发送目标信息;和/或,在第二设备发送对第三信息的确认信息后,第一设备或其它网元发送目标信息。
在一种可能的实现方式中,上述第一设备和第二设备对齐第三信息的方式包括以下至少一项:
在接收到第三信息的一个设备发送第三信息的确认信息后,第一AI模块和/或第二AI模块可使用第三信息关联的模型;
在接收到第三信息的一个设备发送第三信息的确认信息、且经过第一时长后,第一AI模块和/或第二AI模块可使用第三信息关联的模型;
在第三信息的发送时间或接收时间经过第一时长后,第一AI模块和/或第二AI模块可使用第三信息关联的模型。
在一种可能的实现方式中,上述第一时长由以下任一项确定:由第三信息携带、由第三信息的确认信息携带、由第三信息的其它关联信息或信令携带、由协议约定,由第一设备或第二设备的能力确定。
本申请实施例提供的信息传输装置能够实现上述方法实施例中第二设备实现的各个过程,并达到相同的技术效果,为避免重复,这里不再赘述。
本申请实施例中的信息传输装置可以是UE,例如具有操作***的UE,也可以是UE中的部件,例如集成电路或芯片。该UE可以是终端,也可以为除终端之外的其他设备。示例性的,UE可以包括但不限于上述所列举的UE 11的类型,其他设备可以为服务器、网络附属存储器(Network Attached Storage,NAS)等,本申请实施例不作具体限定。
可选地,如图7所示,本申请实施例还提供一种通信设备5000,包括处理器5001和存储器5002,存储器5002上存储有可在所述处理器5001上运行的程序或指令,例如,该通信设备5000为第一设备时,该程序或指令被处理器5001执行时实现上述第一设备侧方法实施例的各个步骤,且能达到相同的技术效果,为避免重复,这里不再赘述。该通信设备5000为第二设备时,该程序或指令被处理器5001执行时实现上述第二设备侧方法实施例的各个步骤,且能达到相同的技术效果,为避免重复,这里不再赘述。
需要说明的是,本申请实施例中,上述第一设备可以为UE或网络侧设备;上述第二设备可以为网络侧设备或UE。下述实施例中对UE和网络侧设备的硬件结构进行示意。
本申请实施例还提供一种UE,包括处理器和通信接口,处理器用于将第一信息输入到第一AI模块,得到第二信息。通信接口用于向第二设备发送第二信息,第二信息用于第二设备将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息。其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息,该第三信息包括第一AI模块和/或第二AI模块的模型信息,第一动作包括以下至少一项:训练、更新、推理。该UE实施例与上述第一设备侧方法实施例对应,上述方法实施例的各个实施过程和实现方式均可适用于该UE实施例中,且能达到相同的技术效果。
本申请实施例还提供一种UE,包括处理器和通信接口,通信接口用于接收来自第一设备的第二信息,该第二信息为第一设备将第一信息输入到第一AI模块中得到的信息。处理器用于将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息。其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息,第三信息包括第一AI模块和/或第二AI模块的模型信息,第一动作包括以下至少一项:训练、更新、推理。该UE实施例与上述第二设备侧方法实施例对应,上述方法实施例的各个实施过程和实现方式均可适用于该UE实施例中,且能达到相同的技术效果。
具体地,图8为实现本申请实施例的一种UE的硬件结构示意图。
该UE 7000包括但不限于:射频单元7001、网络模块7002、音频输出单元7003、输入单元7004、传感器7005、显示单元7006、用户输入单元7007、接口单元7008、存储器7009以及处理器7010等中的至少部分部件。
本领域技术人员可以理解,UE 7000还可以包括给各个部件供电的电源(比如电池),电源可以通过电源管理***与处理器7010逻辑相连,从而通过电源管理***实现管理充电、放电、以及功耗管理等功能。图8中示出的UE结构并不构成对UE的限定,UE可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部 件布置,在此不再赘述。
应理解的是,本申请实施例中,输入单元7004可以包括图形处理单元(Graphics Processing Unit,GPU)70041和麦克风70042,图形处理器70041对在视频捕获模式或图像捕获模式中由图像捕获装置(如摄像头)获得的静态图片或视频的图像数据进行处理。显示单元7006可包括显示面板70061,可以采用液晶显示器、有机发光二极管等形式来配置显示面板70061。用户输入单元7007包括触控面板70071以及其他输入设备70072中的至少一种。触控面板70071,也称为触摸屏。触控面板70071可包括触摸检测装置和触摸控制器两个部分。其他输入设备70072可以包括但不限于物理键盘、功能键(比如音量控制按键、开关按键等)、轨迹球、鼠标、操作杆,在此不再赘述。
本申请实施例中,射频单元7001接收来自网络侧设备的下行数据后,可以传输给处理器7010进行处理;另外,射频单元7001可以向网络侧设备发送上行数据。通常,射频单元7001包括但不限于天线、放大器、收发信机、耦合器、低噪声放大器、双工器等。
存储器7009可用于存储软件程序或指令以及各种数据。存储器7009可主要包括存储程序或指令的第一存储区和存储数据的第二存储区,其中,第一存储区可存储操作***、至少一个功能所需的应用程序或指令(比如声音播放功能、图像播放功能等)等。此外,存储器7009可以包括易失性存储器或非易失性存储器,或者,存储器7009可以包括易失性和非易失性存储器两者。其中,非易失性存储器可以是只读存储器(Read-Only Memory,ROM)、可编程只读存储器(Programmable ROM,PROM)、可擦除可编程只读存储器(Erasable PROM,EPROM)、电可擦除可编程只读存储器(Electrically EPROM,EEPROM)或闪存。易失性存储器可以是随机存取存储器(Random Access Memory,RAM),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,DDRSDRAM)、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(Synch link DRAM,SLDRAM)和直接内存总线随机存取存储器(Direct Rambus RAM,DRRAM)。本申请实施例中的存储器7009包括但不限于这些和任意其它适合类型的存储器。
处理器7010可包括一个或多个处理单元;可选的,处理器7010集成应用处理器和调制解调处理器,其中,应用处理器主要处理涉及操作***、用户界面和应用程序等的操作,调制解调处理器主要处理无线通信信号,如基带处理器。可以理解的是,上述调制解调处理器也可以不集成到处理器7010中。
其中,处理器7010,用于将第一信息输入到第一AI模块,得到第二信息。射频单元7001,用于向第二设备发送第二信息,第二信息用于第二设备将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息。其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息,该第三信息包括第一AI模块和/或第二AI模块的模型信息;第一动作包括以下至少一项:训练、更新、推理。
本申请实施例提供一种UE,在第一AI模块和第二AI模块进行训练、更新和/或推理之前,通过提前对齐第一AI模块和/或第二AI模块的模型相关信息,解决了多节点部署模型进行推理(例如联合推理)时的模型配对问题,使得分布在不同节点的模型对信息进行联合推理,即UE通过第一AI模块对第一信息进行推理、以及第二 设备通过第二AI模块对第二信息进行推理时,无需将模型的全部细节告知目标节点才能进行联合推理,如此在保证模型的推理性能的同时,也避免了模型信息的泄露。
本申请实施例提供的UE能够实现上述方法实施例中第一设备实现的各个过程,并达到相同的技术效果,为避免重复,这里不再赘述。
或者,射频单元7001,用于接收来自第一设备的第二信息,该第二信息为第一设备将第一信息输入到第一AI模块中得到的信息。处理器7010,用于将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息。其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息;第三信息包括第一AI模块和/或第二AI模块的模型信息;第一动作包括以下至少一项:训练、更新、推理。
本申请实施例提供一种UE,在第一AI模块和第二AI模块进行训练、更新和/或推理之前,通过提前对齐第一AI模块和/或第二AI模块的模型相关信息,解决了多节点部署模型进行推理(例如联合推理)时的模型配对问题,使得分布在不同节点的模型对信息进行联合推理,即第一设备通过第一AI模块对第一信息进行推理、以及UE通过第二AI模块对第二信息进行推理时,无需将模型的全部细节告知目标节点才能进行联合推理,如此在保证模型的推理性能的同时,也避免了模型信息的泄露。
本申请实施例提供的UE能够实现上述方法实施例中第二设备实现的各个过程,并达到相同的技术效果,为避免重复,这里不再赘述。
本申请实施例还提供一种网络侧设备,包括处理器和通信接口,处理器用于将第一信息输入到第一AI模块,得到第二信息。通信接口用于向第二设备发送第二信息,第二信息用于第二设备将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息。其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息,该第三信息包括第一AI模块和/或第二AI模块的模型信息,第一动作包括以下至少一项:训练、更新、推理。该网络侧设备实施例与上述第一设备方法实施例对应,上述方法实施例的各个实施过程和实现方式均可适用于该网络侧设备实施例中,且能达到相同的技术效果。
本申请实施例还提供一种网络侧设备,包括处理器和通信接口,通信接口用于接收来自第一设备的第二信息,该第二信息为第一设备将第一信息输入到第一AI模块中得到的信息。处理器用于将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息。其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息,第三信息包括第一AI模块和/或第二AI模块的模型信息,第一动作包括以下至少一项:训练、更新、推理。该网络侧设备实施例与上述第二设备方法实施例对应,上述方法实施例的各个实施过程和实现方式均可适用于该网络侧设备实施例中,且能达到相同的技术效果。
具体地,本申请实施例还提供了一种网络侧设备。如图9所示,该网络侧设备600包括:天线601、射频装置602、基带装置603、处理器604和存储器605。天线601与射频装置602连接。在上行方向上,射频装置602通过天线601接收信息,将接收的信息发送给基带装置603进行处理。在下行方向上,基带装置603对要发送的信息进行处理,并发送给射频装置602,射频装置602对收到的信息进行处理后经过天线601发送出去。
以上实施例中网络侧设备执行的方法可以在基带装置603中实现,该基带装置603包括基带处理器。
其中,处理器604,用于将第一信息输入到第一AI模块,得到第二信息。射频装置602,用于向第二设备发送第二信息,第二信息用于第二设备将第二信息输入到 第二AI模块,得到第一信息和/或第一信息的相关信息。其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息,该第三信息包括第一AI模块和/或第二AI模块的模型信息,第一动作包括以下至少一项:训练、更新、推理。
本申请实施例提供一种网络侧设备,在第一AI模块和第二AI模块进行训练、更新和/或推理之前,通过提前对齐第一AI模块和/或第二AI模块的模型相关信息,解决了多节点部署模型进行推理(例如联合推理)时的模型配对问题,使得分布在不同节点的模型对信息进行联合推理,即网络侧设备通过第一AI模块对第一信息进行推理、以及第二设备通过第二AI模块对第二信息进行推理时,无需将模型的全部细节告知目标节点才能进行联合推理,如此在保证模型的推理性能的同时,也避免了模型信息的泄露。
本申请实施例提供的网络侧设备能够实现上述方法实施例中第一设备实现的各个过程,并达到相同的技术效果,为避免重复,这里不再赘述。
或者,射频装置602,用于接收来自第一设备的第二信息,该第二信息为第一设备将第一信息输入到第一AI模块中得到的信息。处理器604,用于将第二信息输入到第二AI模块,得到第一信息和/或第一信息的相关信息。其中,在第一AI模块和第二AI模块执行第一动作之前,由第一设备和第二设备对齐第三信息,第三信息包括第一AI模块和/或第二AI模块的模型信息;第一动作包括以下至少一项:训练、更新、推理。
本申请实施例提供一种网络侧设备,在第一AI模块和第二AI模块进行训练、更新和/或推理之前,通过提前对齐第一AI模块和/或第二AI模块的模型相关信息,解决了多节点部署模型进行推理(例如联合推理)时的模型配对问题,使得分布在不同节点的模型对信息进行联合推理,即第一设备通过第一AI模块对第一信息进行推理、以及网络侧设备通过第二AI模块对第二信息进行推理时,无需将模型的全部细节告知目标节点才能进行联合推理,如此在保证模型的推理性能的同时,也避免了模型信息的泄露。
本申请实施例提供的网络侧设备能够实现上述方法实施例中第二设备实现的各个过程,并达到相同的技术效果,为避免重复,这里不再赘述。
基带装置603例如可以包括至少一个基带板,该基带板上设置有多个芯片,如图9所示,其中一个芯片例如为基带处理器,通过总线接口与存储器605连接,以调用存储器605中的程序,执行以上方法实施例中所示的网络设备操作。
该网络侧设备还可以包括网络接口606,该接口例如为通用公共无线接口(common public radio interface,CPRI)。
具体地,本申请实施例的网络侧设备600还包括:存储在存储器605上并可在处理器604上运行的指令或程序,处理器604调用存储器605中的指令或程序执行上述各模块执行的方法,并达到相同的技术效果,为避免重复,故不在此赘述。
具体地,本申请实施例还提供了一种网络侧设备。如图10所示,该网络侧设备800包括:处理器801、网络接口802和存储器803。其中,网络接口802例如为通用公共无线接口(common public radio interface,CPRI)。
具体地,本发明实施例的网络侧设备800还包括:存储在存储器803上并可在处理器801上运行的指令或程序,处理器801调用存储器803中的指令或程序执行上述各模块执行的方法,并达到相同的技术效果,为避免重复,故不在此赘述。
本申请实施例还提供一种可读存储介质,所述可读存储介质上存储有程序或指令,该程序或指令被处理器执行时实现上述信息传输方法实施例的各个过程,且能达 到相同的技术效果,为避免重复,这里不再赘述。
其中,所述处理器为上述实施例中所述的通信设备中的处理器。所述可读存储介质,包括计算机可读存储介质,如计算机只读存储器ROM、随机存取存储器RAM、磁碟或者光盘等。
本申请实施例另提供了一种芯片,所述芯片包括处理器和通信接口,所述通信接口和所述处理器耦合,所述处理器用于运行程序或指令,实现上述方法实施例的各个过程,且能达到相同的技术效果,为避免重复,这里不再赘述。
应理解,本申请实施例提到的芯片还可以称为***级芯片,***芯片,芯片***或片上***芯片等。
本申请实施例另提供了一种计算机程序/程序产品,所述计算机程序/程序产品被存储在存储介质中,所述计算机程序/程序产品被至少一个处理器执行以实现上述方法实施例的各个过程,且能达到相同的技术效果,为避免重复,这里不再赘述。
本申请实施例还提供了一种通信***,包括:第一设备及第二设备,所述第一设备可用于执行如上所述的信息传输方法的步骤,所述第二设备可用于执行如上所述的信息传输方法的步骤。
需要说明的是,在本文中,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、物品或者装置不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、物品或者装置所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、方法、物品或者装置中还存在另外的相同要素。此外,需要指出的是,本申请实施方式中的方法和装置的范围不限按示出或讨论的顺序来执行功能,还可包括根据所涉及的功能按基本同时的方式或按相反的顺序来执行功能,例如,可以按不同于所描述的次序来执行所描述的方法,并且还可以添加、省去、或组合各种步骤。另外,参照某些示例所描述的特征可在其他示例中被组合。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以计算机软件产品的形式体现出来,该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中,包括若干指令用以使得一台终端(可以是手机,计算机,服务器,空调器,或者网络设备等)执行本申请各个实施例所述的方法。
上面结合附图对本申请的实施例进行了描述,但是本申请并不局限于上述的具体实施方式,上述的具体实施方式仅仅是示意性的,而不是限制性的,本领域的普通技术人员在本申请的启示下,在不脱离本申请宗旨和权利要求所保护的范围情况下,还可做出很多形式,均属于本申请的保护之内。

Claims (30)

  1. 一种信息传输方法,包括:
    第一设备将第一信息输入到第一人工智能AI模块,得到第二信息;
    所述第一设备向第二设备发送所述第二信息,所述第二信息用于所述第二设备将所述第二信息输入到第二AI模块,得到所述第一信息和/或所述第一信息的相关信息;
    其中,在所述第一AI模块和所述第二AI模块执行第一动作之前,由所述第一设备和所述第二设备对齐第三信息;所述第三信息包括所述第一AI模块和/或所述第二AI模块的模型信息;所述第一动作包括以下至少一项:训练、更新、推理。
  2. 根据权利要求1所述的方法,其中,所述第一信息包括以下至少一项:信道信息、波束质量信息;
    所述第二信息包括以下至少一项:预编码矩阵指示PMI、预测的波束信息或波束指示。
  3. 根据权利要求1或2所述的方法,其中,所述第一AI模块和/或所述第二AI模块根据以下至少一项得到:
    所述第一设备根据来自所述第二设备或其它网元的目标信息训练得到;
    所述第二设备根据来自所述第一设备或其它网元的目标信息训练得到;
    或者,所述第一AI模块和/或所述第二AI模块根据以下至少一项进行更新或调整:
    所述第一设备根据来自所述第二设备或其它网元的目标信息进行更新或调整;
    所述第二设备根据来自所述第一设备或其它网元的目标信息进行更新或调整;
    其中,所述目标信息包括与AI模块的所述第一动作相关的至少一个第一信息和与所述至少一个第一信息对应的至少一个第二信息。
  4. 根据权利要求1所述的方法,其中,所述第三信息具体包括以下至少一项:模型的结构特征、模型的载荷量化方法、模型的估计精度或输出精度。
  5. 根据权利要求4所述的方法,其中,所述模型的结构特征包括以下至少一项:模型结构、模型基本结构特征、模型子模块的结构特征、模型层数、模型神经元数量、模型大小、模型复杂度、模型参数的量化参数。
  6. 根据权利要求5所述的方法,其中,所述模型基本结构特征包括以下至少一项:是否包含全连接结构、是否包含卷积结构、是否包含长短期记忆模型LSTM结构、是否包含注意力结构、是否包含残差结构。
  7. 根据权利要求5所述的方法,其中,所述模型神经元数量包括以下至少一项:全连接神经元数量、卷积神经元数量、记忆神经元数量、注意力神经元数量、残差神经元数量;
    和/或,
    所述模型神经元数量包括以下至少一项:所有类型的神经元数量、单个类型的神经元数量、整个模型的神经元数量、单层或数层的神经元数量。
  8. 根据权利要求5所述的方法,其中,所述模型参数的量化参数包括以下至少一项:模型参数的量化方式、单个神经元参数的量化比特数;
    其中,所述模型参数的量化方式包括以下至少一项:均匀量化方式、非均匀量化方式、权值共享量化方式或分组量化方式、参数编码的量化方式、变换域量化方式、乘积量化方式。
  9. 根据权利要求4所述的方法,其中,所述载荷量化方法包括以下至少一项:量化方式、量化前后特征的维数、量化时使用的量化方法。
  10. 根据权利要求9所述的方法,其中,量化时使用的量化方法包括以下至少一项:使用码本进行量化时需要同步码本内容和码本使用方法、使用特定规则进行量化时需要同步量化规则。
  11. 根据权利要求10所述的方法,其中,所述量化规则包括以下至少一项:N个量化区间、量化方式,N为正整数;
    其中,所述量化方式包括以下至少一项:均匀量化方式、非均匀量化方式、权值共享量化方式或分组量化方式、参数编码的量化方式、变换域量化方式、乘积量化方式。
  12. 根据权利要求10所述的方法,其中,在同步所述码本内容和所述码本使用方法,和/或,同步所述量化规则时,同步方法包括以下任一项:从预定义的方法集合中选取同步时反馈代表所选方法的集合序号、直接发送码本内容。
  13. 根据权利要求1所述的方法,其中,所述第一设备和所述第二设备对齐所述第三信息的方式包括以下至少一项:
    所述第一设备或其它网元将目标信息发送给所述第二设备时,同时发送所述第三信息;
    所述第二设备或其它网元将目标信息发送给所述第一设备时,同时发送所述第三信息;
    所述第一设备或其它网元将目标信息发送给所述第二设备前,所述第一设备或所述其它网元发送所述第三信息;
    所述第二设备或其它网元将目标信息发送给所述第一设备前,所述第二设备或所述其它网元发送所述第三信息;
    所述第二设备在请求目标信息时,所述第二设备发送所述第三信息;
    所述第一设备在请求目标信息时,所述第一设备发送所述第三信息;
    所述第二设备在请求目标信息时,所述第一设备或其它网元发送同意信息并发送所述第三信息,所述同意信息用于指示同意所述第二设备的请求;
    所述第一设备在请求目标信息时,所述第二设备或其它网元发送同意信息并发送所述第三信息,所述同意信息用于指示同意所述第一设备的请求;
    其中,所述目标信息包括与AI模块的所述第一动作相关的至少一个第一信息和与所述至少一个第一信息对应的至少一个第二信息。
  14. 根据权利要求13所述的方法,其中,在所述第一设备发送对所述第三信息的确认信息之后,所述第二设备或所述其它网元发送所述目标信息;
    和/或,
    在所述第二设备发送对所述第三信息的确认信息后,所述第一设备或所述其它网元发送所述目标信息。
  15. 根据权利要求1所述的方法,其中,所述第一设备和所述第二设备对齐所述第三信息的方式包括以下至少一项:
    在接收到所述第三信息的一个设备发送所述第三信息的确认信息后,所述第一AI模块和/或所述第二AI模块可使用所述第三信息关联的模型;
    在接收到所述第三信息的一个设备发送所述第三信息的确认信息、且经过第一时长后,所述第一AI模块和/或所述第二AI模块可使用所述第三信息关联的模型;
    在所述第三信息的发送时间或接收时间经过第一时长后,所述第一AI模块和/或所述第二AI模块可使用所述第三信息关联的模型。
  16. 根据权利要求15所述的方法,其中,所述第一时长由以下任一项确定:由所述第三信息携带、由所述第三信息的确认信息携带、由所述第三信息的其它关联信息 或信令携带、由协议约定,由所述第一设备或所述第二设备的能力确定。
  17. 一种信息传输方法,包括:
    第二设备接收来自第一设备的第二信息,所述第二信息为所述第一设备将第一信息输入到第一人工智能AI模块中得到的信息;
    所述第二设备将所述第二信息输入到第二AI模块,得到所述第一信息和/或所述第一信息的相关信息;
    其中,在所述第一AI模块和所述第二AI模块执行第一动作之前,由所述第一设备和所述第二设备对齐第三信息;所述第三信息包括所述第一AI模块和/或所述第二AI模块的模型信息;所述第一动作包括以下至少一项:训练、更新、推理。
  18. 根据权利要求17所述的方法,其中,所述第一信息包括以下至少一项:信道信息、波束质量信息;
    所述第二信息包括以下至少一项:预编码矩阵指示PMI、预测的波束信息或波束指示。
  19. 根据权利要求17或18所述的方法,其中,所述第一AI模块和/或所述第二AI模块根据以下至少一项得到:
    所述第一设备根据来自所述第二设备或其它网元的目标信息训练得到;
    所述第二设备根据来自所述第一设备或其它网元的目标信息训练得到;
    或者,所述第一AI模块和/或所述第二AI模块根据以下至少一项进行更新或调整:
    所述第一设备根据来自所述第二设备或其它网元的目标信息进行更新或调整;
    所述第二设备根据来自所述第一设备或其它网元的目标信息进行更新或调整;
    其中,所述目标信息包括与AI模块的所述第一动作相关的至少一个第一信息和与所述至少一个第一信息对应的至少一个第二信息。
  20. 根据权利要求17所述的方法,其中,所述第三信息具体包括以下至少一项:模型的结构特征、模型的载荷量化方法、模型的估计精度或输出精度。
  21. 根据权利要求20所述的方法,其中,所述模型的结构特征包括以下至少一项:模型结构、模型基本结构特征、模型子模块的结构特征、模型层数、模型神经元数量、模型大小、模型复杂度、模型参数的量化参数。
  22. 根据权利要求20所述的方法,其中,所述载荷量化方法包括以下至少一项:量化方式、量化前后特征的维数、量化时使用的量化方法。
  23. 根据权利要求17所述的方法,其中,所述第一设备和所述第二设备对齐所述第三信息的方式包括以下至少一项:
    所述第一设备或其它网元将目标信息发送给所述第二设备时,同时发送所述第三信息;
    所述第二设备或其它网元将目标信息发送给所述第一设备时,同时发送所述第三信息;
    所述第一设备或其它网元将目标信息发送给所述第二设备前,所述第一设备或所述其它网元发送所述第三信息;
    所述第二设备或其它网元将目标信息发送给所述第一设备前,所述第二设备或所述其它网元发送所述第三信息;
    所述第二设备在请求目标信息时,所述第二设备发送所述第三信息;
    所述第一设备在请求目标信息时,所述第一设备发送所述第三信息;
    所述第二设备在请求目标信息时,所述第一设备或其它网元发送同意信息并发送所述第三信息,所述同意信息用于指示同意所述第二设备的请求;
    所述第一设备在请求目标信息时,所述第二设备或其它网元发送同意信息并发送所述第三信息,所述同意信息用于指示同意所述第一设备的请求;
    其中,所述目标信息包括与AI模块的所述第一动作相关的至少一个第一信息和与所述至少一个第一信息对应的至少一个第二信息。
  24. 根据权利要求17所述的方法,其中,所述第一设备和所述第二设备对齐所述第三信息的方式包括以下至少一项:
    在接收到所述第三信息的一个设备发送所述第三信息的确认信息后,所述第一AI模块和/或所述第二AI模块可使用所述第三信息关联的模型;
    在接收到所述第三信息的一个设备发送所述第三信息的确认信息、且经过第一时长后,所述第一AI模块和/或所述第二AI模块可使用所述第三信息关联的模型;
    在所述第三信息的发送时间或接收时间经过第一时长后,所述第一AI模块和/或所述第二AI模块可使用所述第三信息关联的模型。
  25. 一种信息传输装置,应用于第一设备,包括:处理模块和发送模块;
    所述处理模块,用于将第一信息输入到第一人工智能AI模块,得到第二信息;
    所述发送模块,用于向第二设备发送所述处理模块得到的所述第二信息,所述第二信息用于所述第二设备将所述第二信息输入到第二AI模块,得到所述第一信息和/或所述第一信息的相关信息;
    其中,在所述第一AI模块和所述第二AI模块执行第一动作之前,由所述第一设备和所述第二设备对齐第三信息;所述第三信息包括所述第一AI模块和/或所述第二AI模块的模型信息;所述第一动作包括以下至少一项:训练、更新、推理。
  26. 一种信息传输装置,应用于第二设备,包括:接收模块和处理模块;
    所述接收模块,用于接收来自第一设备的第二信息,所述第二信息为所述第一设备将第一信息输入到第一人工智能AI模块中得到的信息;
    所述处理模块,用于将所述接收模块接收的所述第二信息输入到第二AI模块,得到所述第一信息和/或所述第一信息的相关信息;
    其中,在所述第一AI模块和所述第二AI模块执行第一动作之前,由所述第一设备和所述第二设备对齐第三信息;所述第三信息包括所述第一AI模块和/或所述第二AI模块的模型信息;所述第一动作包括以下至少一项:训练、更新、推理。
  27. 一种通信设备,包括处理器,存储器及存储在所述存储器上并可在所述处理器上运行的程序或指令,所述程序或指令被所述处理器执行时实现如权利要求1至16中任一项所述的信息传输方法的步骤。
  28. 一种通信设备,包括处理器,存储器及存储在所述存储器上并可在所述处理器上运行的程序或指令,所述程序或指令被所述处理器执行时实现如权利要求17至24中任一项所述的信息传输方法的步骤。
  29. 一种通信***,所述通信***包括如权利要求25所述的信息传输装置以及如权利要求26所述的信息传输装置;或者,
    所述通信***包括如权利要求27所述的通信设备和如权利要求28所述的通信设备。
  30. 一种可读存储介质,所述可读存储介质上存储程序或指令,所述程序或指令被处理器执行时实现如权利要求1至16中任一项所述的信息传输方法的步骤,或者实现如权利要求17至24中任一项所述的信息传输方法的步骤。
PCT/CN2023/111732 2022-08-12 2023-08-08 信息传输方法、装置、设备、***及存储介质 WO2024032606A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210970370.2 2022-08-12
CN202210970370.2A CN117692032A (zh) 2022-08-12 2022-08-12 信息传输方法、装置、设备、***及存储介质

Publications (1)

Publication Number Publication Date
WO2024032606A1 true WO2024032606A1 (zh) 2024-02-15

Family

ID=89850945

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/111732 WO2024032606A1 (zh) 2022-08-12 2023-08-08 信息传输方法、装置、设备、***及存储介质

Country Status (2)

Country Link
CN (1) CN117692032A (zh)
WO (1) WO2024032606A1 (zh)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114390580A (zh) * 2020-10-20 2022-04-22 维沃移动通信有限公司 波束上报方法、波束信息确定方法及相关设备
WO2022141397A1 (zh) * 2020-12-31 2022-07-07 华为技术有限公司 一种神经网络的训练方法以及相关装置
CN114765771A (zh) * 2021-01-08 2022-07-19 展讯通信(上海)有限公司 模型更新方法及装置、存储介质、终端、网络侧设备

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114390580A (zh) * 2020-10-20 2022-04-22 维沃移动通信有限公司 波束上报方法、波束信息确定方法及相关设备
WO2022141397A1 (zh) * 2020-12-31 2022-07-07 华为技术有限公司 一种神经网络的训练方法以及相关装置
CN114765771A (zh) * 2021-01-08 2022-07-19 展讯通信(上海)有限公司 模型更新方法及装置、存储介质、终端、网络侧设备

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
HUAWEI, HISILICON: "Evaluation on AI/ML for CSI feedback enhancement", 3GPP DRAFT; R1-2203140, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. RAN WG1, no. e-Meeting; 20220509 - 20220520, 29 April 2022 (2022-04-29), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France, XP052143958 *
VIVO: "Evaluation on AI/ML for CSI feedback enhancement", 3GPP DRAFT; R1-2203550, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. RAN WG1, no. e-Meeting; 20220509 - 20220520, 29 April 2022 (2022-04-29), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France, XP052153025 *
ZTE CORPORATION: "Evaluation assumptions on AI/ML for CSI feedback", 3GPP DRAFT; R1-2203248, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. RAN WG1, no. e-Meeting; 20220509 - 20220520, 29 April 2022 (2022-04-29), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France, XP052152890 *

Also Published As

Publication number Publication date
CN117692032A (zh) 2024-03-12

Similar Documents

Publication Publication Date Title
US20230412430A1 (en) Inforamtion reporting method and apparatus, first device, and second device
WO2023246618A1 (zh) 信道矩阵处理方法、装置、终端及网络侧设备
WO2024032606A1 (zh) 信息传输方法、装置、设备、***及存储介质
WO2024037380A1 (zh) 信道信息处理方法、装置、通信设备及存储介质
WO2023185978A1 (zh) 信道特征信息上报及恢复方法、终端和网络侧设备
WO2023179476A1 (zh) 信道特征信息上报及恢复方法、终端和网络侧设备
WO2023179473A1 (zh) 信道特征信息上报及恢复方法、终端和网络侧设备
WO2023134628A1 (zh) 传输方法、装置和设备
WO2024055974A1 (zh) Cqi传输方法、装置、终端及网络侧设备
WO2024104126A1 (zh) 更新ai网络模型的方法、装置和通信设备
WO2023179474A1 (zh) 信道特征信息辅助上报及恢复方法、终端和网络侧设备
WO2024055993A1 (zh) Cqi传输方法、装置、终端及网络侧设备
WO2024088162A1 (zh) 信息传输方法、信息处理方法、装置和通信设备
WO2024027683A1 (zh) 模型匹配方法、装置、通信设备及可读存储介质
WO2024088161A1 (zh) 信息传输方法、信息处理方法、装置和通信设备
CN117411527A (zh) 信道特征信息上报及恢复方法、终端和网络侧设备
WO2024041421A1 (zh) 测量反馈处理方法、装置、终端及网络侧设备
WO2024032695A1 (zh) Csi预测处理方法、装置、通信设备及可读存储介质
WO2023179570A1 (zh) 信道特征信息传输方法、装置、终端及网络侧设备
WO2024041420A1 (zh) 测量反馈处理方法、装置、终端及网络侧设备
WO2024067280A1 (zh) 更新ai模型参数的方法、装置及通信设备
WO2024046202A1 (zh) 发送方法、用户设备、网络侧设备及可读存储介质
WO2023185980A1 (zh) 信道特征信息传输方法、装置、终端及网络侧设备
WO2024093999A1 (zh) 信道信息的上报和接收方法、终端及网络侧设备
CN117335849A (zh) 信道特征信息上报及恢复方法、终端和网络侧设备

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: 23851827

Country of ref document: EP

Kind code of ref document: A1