WO2023276382A1 - 通信装置、通信方法、および通信システム - Google Patents
通信装置、通信方法、および通信システム Download PDFInfo
- Publication number
- WO2023276382A1 WO2023276382A1 PCT/JP2022/015609 JP2022015609W WO2023276382A1 WO 2023276382 A1 WO2023276382 A1 WO 2023276382A1 JP 2022015609 W JP2022015609 W JP 2022015609W WO 2023276382 A1 WO2023276382 A1 WO 2023276382A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- calculations
- calculation
- range
- communication
- information processing
- Prior art date
Links
- 238000004891 communication Methods 0.000 title claims abstract description 899
- 238000000034 method Methods 0.000 title description 36
- 238000004364 calculation method Methods 0.000 claims abstract description 995
- 230000010365 information processing Effects 0.000 claims abstract description 167
- 238000013528 artificial neural network Methods 0.000 claims abstract description 79
- 230000005540 biological transmission Effects 0.000 claims description 53
- 238000003672 processing method Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 description 118
- 238000010586 diagram Methods 0.000 description 44
- 230000006870 function Effects 0.000 description 40
- 230000008859 change Effects 0.000 description 19
- 230000033001 locomotion Effects 0.000 description 17
- 230000008569 process Effects 0.000 description 16
- 238000005516 engineering process Methods 0.000 description 13
- 238000003860 storage Methods 0.000 description 13
- 230000009977 dual effect Effects 0.000 description 12
- 230000001934 delay Effects 0.000 description 10
- 230000000694 effects Effects 0.000 description 9
- 238000005259 measurement Methods 0.000 description 8
- 238000009826 distribution Methods 0.000 description 7
- 230000011664 signaling Effects 0.000 description 7
- 230000004913 activation Effects 0.000 description 6
- 238000013527 convolutional neural network Methods 0.000 description 6
- 238000007726 management method Methods 0.000 description 6
- 238000004088 simulation Methods 0.000 description 6
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 6
- 238000006243 chemical reaction Methods 0.000 description 5
- 230000010287 polarization Effects 0.000 description 5
- 238000001514 detection method Methods 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 230000015654 memory Effects 0.000 description 4
- 230000007423 decrease Effects 0.000 description 3
- 238000007667 floating Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 241000282326 Felis catus Species 0.000 description 2
- 229910000831 Steel Inorganic materials 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 2
- 230000003321 amplification Effects 0.000 description 2
- 239000000872 buffer Substances 0.000 description 2
- 239000003795 chemical substances by application Substances 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000003058 natural language processing Methods 0.000 description 2
- 238000003199 nucleic acid amplification method Methods 0.000 description 2
- 230000000737 periodic effect Effects 0.000 description 2
- 230000010363 phase shift Effects 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 239000010959 steel Substances 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 102100022734 Acyl carrier protein, mitochondrial Human genes 0.000 description 1
- PGLIUCLTXOYQMV-UHFFFAOYSA-N Cetirizine hydrochloride Chemical compound Cl.Cl.C1CN(CCOCC(=O)O)CCN1C(C=1C=CC(Cl)=CC=1)C1=CC=CC=C1 PGLIUCLTXOYQMV-UHFFFAOYSA-N 0.000 description 1
- 101000678845 Homo sapiens Acyl carrier protein, mitochondrial Proteins 0.000 description 1
- 235000008694 Humulus lupulus Nutrition 0.000 description 1
- 101150119040 Nsmf gene Proteins 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 238000005452 bending Methods 0.000 description 1
- 230000001364 causal effect Effects 0.000 description 1
- 230000010267 cellular communication Effects 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 238000013523 data management Methods 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 230000001627 detrimental effect Effects 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 239000000523 sample Substances 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 238000012421 spiking Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L45/00—Routing or path finding of packets in data switching networks
Definitions
- the present disclosure relates to communication devices, communication methods, and communication systems.
- This application mainly performs calculations based on a neural network (DNN: Deep Neural Network) with multiple layers, whose internal parameters are optimized by machine learning. Such calculations are more expensive than other common applications. Therefore, when the application is executed on a general-purpose wireless communication terminal such as a smart phone, there arises a problem that calculation time, power consumption, and the like increase.
- a method in which a cloud server performs the calculation on behalf of the user is also conceivable.
- the wireless communication terminal transmits information necessary for the calculation to the cloud server and receives the calculation result from the cloud server, resulting in a large amount of communication.
- delays are likely to occur due to unstable communication quality. Therefore, this method may exceed the amount of delay that the application can tolerate.
- the communication network that relays the communication between the communication terminal and the cloud server is also being considered to share part of the DNN calculations. That is, at least one of a plurality of communication nodes forming a communication network can take charge of part of a series of calculations from the input layer to the output layer of the DNN.
- distributing the DNN computations over multiple devices can also have detrimental effects. For example, the load on the communication node responsible for calculation, the communication quality of the network, etc. are not always stable. Therefore, depending on the situation, the calculation may take longer than expected.
- the present disclosure provides an information processing device and the like that reduces the time required to return the calculation results while distributing the calculations based on the DNN. Specifically, it is realized by dynamically changing the device in charge of the calculation, terminating the calculation in the middle, and feeding back the result of the calculation in the middle. Also, the problems presented here are only part of the problems to be solved by the present invention, and the present disclosure may be used for other problems that can be solved by the present disclosure.
- An information processing device is an information processing device in charge of part of a series of calculations of a deep neural network, wherein a result of an intermediate calculation in the series of calculations of the deep neural network is obtained by a first If it is not determined to transmit to the first communication device, at least part of the calculations included in the first range of the series of calculations The result of the calculation is transmitted to a second communication device that is in charge of calculation of the second range following the first range, and if it is determined to be transmitted to the first communication device, the series of deep neural networks and transmitting the result of the executed calculation to the first communication device.
- the information processing device performs at least the calculations included in the first range.
- a part may be executed and the result of the executed calculation may be used as the result of the calculation to be transmitted to the second communication device, or the result of an intermediate calculation in a series of calculations of the deep neural network may be used as the result of the calculation. If it is determined to transmit to the first communication device, perform at least a portion of the calculations included in the first range, and transmit the results of the performed calculations to the first communication device. may be used in the above calculation for
- the information processing device performs at least part of the calculations included in the first range, the determination is performed while the calculations included in the first range are performed, and If it is determined to transmit to the first communication device, the results of said computations performed prior to said determination may be used in said computations for transmission to said first communication device.
- the information processing device performs at least part of the calculations included in the first range, the determination is performed while the calculations included in the first range are performed, and When it is determined to transmit to the first communication device, the result of the calculation performed prior to the determination may be used as the result of the calculation to be transmitted to the second communication device.
- the information processing device performs at least part of the calculations included in the first range, the determination is performed while the calculations included in the first range are performed, and If it is determined to transmit to the first communication device, continue the calculations included in the first range, and use the results of the calculations in the first range as the results of the calculations to be transmitted to the second communication device.
- the information processing device when it is determined that the information processing device transmits intermediate calculation results in the series of calculations of the deep neural network to the first communication device, the information processing device further performs may be transmitted to the second communication device.
- the information processing device executes all of the calculations included in the first range when it is determined that intermediate calculation results in a series of calculations of the deep neural network are to be transmitted to the first communication device. If not, the results of the performed calculations may be sent to the second communication device along with information indicating where in the series of calculations the results of the performed calculations are located.
- the information processing device sends the results of the executed calculations and the results of the executed calculations to the first communication device. to a third communication device capable of executing at least part of the series of calculations.
- the information processing device when it is determined to transmit the results of intermediate calculations in a series of calculations of the deep neural network to the first communication device, performs the calculation up to a predetermined position included in the first range. and the results of the performed calculations may be used in said calculations for transmission to said first communication device.
- the information processing device determines which position of the calculation included in the first range. Calculations included in the first range are performed up to the determined position, and the result of the performed calculation and information indicating the determined position are sent to the second communication device You may send.
- the information processing device includes information on the computational spare capacity of a device that performs at least part of a series of calculations of the deep neural network, and a device that performs at least a part of the series of calculations of the deep neural network. information about the amount of traffic in the .
- the information processing device may make the determination based on at least one of quality of communication with the first communication device and mobility information of the first communication device.
- the information processing device may include information instructing to terminate the series of calculations halfway through, and information instructing to transmit a result of an intermediate calculation in the series of calculations to the first communication device. The determination may be made based on at least one of
- the information processing device may acquire request information from the first communication device and make the determination based on the request information.
- the information processing device sends the results of intermediate calculations in a series of calculations of the deep neural network to the first communication. It may be determined to transmit to the device.
- the information processing device determines whether or not to halfway finish the calculations included in the first range in the series of calculations of the deep neural network, and determines to halfway finish the calculations included in the first range. if performed, a part of the calculations included in the first range is executed, and information indicating the result of the executed calculation and the position of the result of the executed calculation in the series of calculations; , to the device responsible for computing a second range following said first range.
- the information processing device transmits information indicating whether or not the device in charge of calculation of the second range may terminate the calculation included in the second range in the middle, to the calculation in charge of the second range. It may be further transmitted to a device.
- An information processing method which is another aspect of the present disclosure, is an information processing method executed by an information processing device in charge of part of a series of calculations of a deep neural network, the method comprising: a step of determining whether or not to transmit a result of an intermediate computation in a series of computations to a first communication device; transmitting the result of at least a portion of the calculations included in the first range of the first range to a second communication device in charge of calculation of the second range following the first range; If it is determined to transmit to the device, performing calculation for transmission to the first communication device using intermediate calculation results in a series of calculations of the deep neural network, and executing calculation results to the first communication device.
- Another information processing method of the present disclosure includes a step of determining whether or not to end a calculation included in a first range in a series of calculations of a deep neural network, and performing a calculation included in the first range. If it is determined to end midway, a part of the calculations included in the first range is executed, and the result of the executed calculation and the position of the result of the executed calculation in the series of calculations and sending to a device responsible for computing a second range following said first range.
- An information processing system that is another aspect of the present disclosure includes at least a first information processing device and a second information processing device that are in charge of part of a series of calculations of a deep neural network, and the first information
- the processing device determines whether or not to transmit intermediate calculation results in a series of calculations of the deep neural network to the first communication device, and if it is not determined to transmit to the first communication device in the case where it is determined to transmit the results of at least part of the calculations included in the first range in the series of calculations to the second information processing device and to the first communication device; performs calculations to be transmitted to the first communication device using intermediate calculation results in a series of calculations of the deep neural network, and transmits the results of the performed calculations to the first communication device; and the second information processing device performs calculations subsequent to the calculation performed by the first information processing device in the series of calculations based on the results of the calculations performed by the first information processing device An information processing system that executes.
- the information processing system may further include a third information processing device that determines the first range.
- Another information processing system of the present disclosure includes at least a first information processing device that performs a first range of calculations in a series of deep neural network calculations; a second information processing device that performs calculation of a second range following the range; When it is determined that the calculations included in one range are to be terminated halfway, a part of the calculations included in the first range is executed, and the results of the executed calculations and the results of the executed calculations are combined into the series of calculations. to the second information processing device, and when the second information processing device receives the information, the second information processing device performs the first information processing of the series of calculations. A continuation of the computations performed by the device is performed based on the results of the computations performed by the first information processing device.
- the another information processing system further includes a third information processing device that determines whether or not to terminate the calculations included in the first range, wherein the third information processing device is included in the first range. is transmitted to the first information processing apparatus, and the first information processing apparatus determines to terminate the calculations included in the first range when receiving the instruction. good.
- FIG. 1 is a diagram showing a configuration example of an information processing system according to an embodiment of the present disclosure
- FIG. 4 is a diagram for explaining distribution of DNN calculations
- FIG. 5 is a diagram for explaining differences in delays and output data amounts according to splitting points
- FIG. 2 illustrates an example architecture of an IAB network
- Fig. 3 shows the effect of variance on DNN computations
- FIG. 2 illustrates the effect of communication network resources on execution delay
- FIG. 4 is a schematic sequence diagram showing the flow of overall processing of the embodiment
- FIG. 4 is a diagram for explaining Splitting mode set for each communication route;
- FIG. 4 is a diagram showing an example of Splitting mode for each communication route; Sequence diagrams before and after switching the computation manager. The figure which shows an example of the conditions for determining the range in charge of a communication terminal.
- FIG. 5 is a diagram showing an example of a calculation result transmitted from a communication terminal when the communication terminal determines its own range;
- FIG. 3 is a schematic sequence diagram showing the flow of overall processing when a communication terminal determines its own range;
- 1 is a diagram showing a configuration example of a 5GS (5G System) network architecture including a core network;
- FIG. 4 is a diagram showing an example of a calculation range when performing early-exiting;
- FIG. 5GS 5G System
- FIG. 3 is a schematic sequence diagram showing a first example of the flow of processing relating to early-exiting
- FIG. 10 is a schematic sequence diagram showing a second example of the flow of processing relating to early-exiting
- FIG. 11 is a schematic sequence diagram showing a third example of the flow of processing relating to early-exiting
- Conceptual diagram for explaining multi-feedback
- FIG. 1 is a diagram illustrating a configuration example of an information processing system according to the first embodiment of the present disclosure.
- An information processing system 1 according to the first embodiment includes a communication terminal 11 , a cloud system (Cloud) 12 , and a communication network 13 .
- the same numbers are assigned to individuals of the same type, and individual individuals are distinguished by alphabets.
- alphabetical symbols will not be described.
- the information processing system 1 is a system for operating applications using a deep neural network (DNN: Deep Neural Network) learned by machine learning (ML: Machine Learning).
- DNN Deep Neural Network
- ML Machine Learning
- the communication terminal 11 is also an information processing device capable of starting ML applications, and corresponds to smartphones, laptops, and the like. For example, it is assumed that an ML application is installed in a smart phone, and the ML application is activated by the user of the smart phone. Also, the shape of the communication terminal 11 is not particularly limited. For example, it may be a wearable terminal like glasses. A robot whose motion is controlled by the ML application also corresponds to the communication terminal 11 .
- the cloud system 12 includes one or more information processing devices called cloud servers that have higher performance than the communication terminal 11 and provides services that can be used by the communication terminal 11 .
- the communication network 13 is composed of a plurality of communication nodes and relays communication between the communication terminals 11 and the cloud system 12 . A communication node is also called a communication base station.
- FIG. 1 shows an example in which the communication network 13 includes a wireless communication network.
- the example of FIG. 1 shows an example using an IAB (Integrated Access and Backhaul) network used for wireless communication of the fifth generation mobile communication system (5G), and the communication terminal 11 is shown as a wireless communication terminal.
- the communication network 13 includes a wireless communication node 131 that can be wirelessly connected to the communication terminal 11, a donor node 132 that is an upper node of one or more wireless communication nodes 131, and wired communication between the donor node 132 and the cloud system 12. and a core network 133 that performs As shown in FIG.
- the communication network 13 it is preferable for the communication network 13 to include a wireless communication network whose communication quality is more unstable than that for wired communication, because the effects described later are greater than in the conventional case.
- all communication of the information processing system 1 may be wired communication, and the wireless communication network that can be included in the communication network 13 is not limited to the IAB network.
- FIG. 2 is a diagram explaining the DNN.
- a network surrounded by a dotted line frame 2 in FIG. 2 corresponds to the DNN.
- the DNN is composed of a plurality of nodes 21 and links 22 connecting the nodes 21 together. Further, as shown in FIG. 2, the plurality of nodes 21 are divided into node groups arranged in a vertical line, and the node groups are called layers. Although the DNN has seven layers in the example of FIG. 2, the DNN may have three or more layers.
- the DNN calculation is performed for each node 21.
- image information is input to each node 21 of the first layer called the input layer, and calculations are performed in each node 21 of the first layer. These calculation results are sent to each node 21 of the second layer via the link 22, and the calculation is also performed in each node 21 of the second layer. In this way, calculations are performed from the input layer, and the final calculation results are output from the nodes of the final layer called the output layer. Then, based on the output calculation result, the object shown in the input image is determined to be a cat.
- FIG. 2 shows an example of image recognition
- the use of the ML application is not particularly limited.
- AR Augmented Reality
- VR Virtual Realty
- MR Magnetic Realty
- xR Augmented Reality
- Autonomous driving, robotics, speech recognition, etc. may also be possible using DNNs, and ML applications may relate to such uses.
- the implementation environment of the ML application is not particularly limited, and may be used in systems such as digital twins and tactile internet.
- Smartphones which correspond to the communication terminal 11, generally have lower specs than cloud servers. Therefore, if the communication terminal 11 is caused to perform all the processing of the ML application, especially the calculation of the DNN, as it is (In device learning), the calculation time until completion will be long. In other words, a large computational delay occurs. However, due to the specifications of the ML application, it may be required to keep the time required to execute the ML application within a predetermined allowable limit. It is likely to exceed.
- the cloud system 12 instead of the communication terminal 11 executes the DNN calculation (cloud learning)
- the time required for communication in other words, communication delay becomes a problem.
- the cloud server which increases the communication delay, and the sum of the communication delay and the calculation delay may exceed the allowable limit of the ML application.
- this data transmission puts pressure on the band, which may affect other communications.
- the information processing system 1 determines a plurality of persons in charge of calculation from among the communication terminal 11, the cloud system 12, and the communication network 13, and distributes a series of calculations based on the DNN to the plurality of persons in charge of calculation. let it be processed. Such processing is also called distributed learning.
- computational responsibility refers to the entity responsible for at least part of the computation of the DNN.
- FIG. 3 is a diagram explaining the distribution of DNN calculations.
- FIGS. 3A and 3B show an example of federated learning that is not distributed learning
- FIG. 3C shows an example of distributed learning.
- the communication terminal 11, the cloud system 12, and the communication network 13 are each in charge of part of the DNN calculation.
- the communication network 13 also provides computational power for the ML applications running on the communication terminal 11 . Since the cloud server of the cloud system 12 with high computing power takes charge of part of the DNN calculation, the calculation delay can be suppressed as compared with the case where only the communication terminal 11 performs the DNN calculation.
- the transmission data from the communication terminal 11 is received and processed by the communication network 13 and then transmitted to the cloud system 12 . If the data transmitted from the communication network 13 to the cloud system 12 can be made smaller than the data transmitted from the communication terminal 11, the communication time will be shortened.
- the sum of the calculation delay, which is the time required for each communication staff to calculate the DNN, and the communication delay, which is the time required for each communication staff to communicate the information necessary for the DNN calculation, is shown in FIG. It may be shorter than it should be.
- the time required to execute the ML application by distributing and processing a series of calculations of the DNN, the time required to execute the ML application, more specifically, the output from the DNN after the input to the DNN is performed To keep the time-to-obtain within a given acceptable limit.
- execution delay the time required to execute the ML application
- one or more communication nodes in the communication network 13 are further determined to be in charge of calculation.
- the wireless communication node 131 and the donor node 132 described above correspond to communication nodes.
- the DNN in FIG. 2 has seven layers. It is shown that the cloud system 12 is in charge of 7 layers.
- the communication terminal 11 transmits the calculation result in the second layer to the wireless communication node 131, and the wireless communication node 131 performs the calculation in the third layer and the fourth layer from the calculation result in the second layer.
- the calculation results of the four layers are transmitted to the cloud system 12, and the cloud system 12 performs calculations of the fifth to seventh layers based on the calculation results of the fourth layer.
- the cloud system 12 may return the calculation result of the seventh layer to the communication terminal 11, and the communication terminal 11 may determine that the input is an image cat based on the calculation result of the seventh layer.
- the cloud system 12 may determine that the input is an image cat based on the calculation result of the seventh layer and return the determination result to the communication terminal 11 .
- DNNs convolutional neural networks
- CNNs convolutional neural networks
- the DNN parameters may or may not be updated. That is, the DNN has already completed learning and the parameters of the DNN may not be updated.
- a correct answer may be received from the user of the communication terminal 11 via the ML application, and learning may be performed based on the correct answer.
- a new updated DNN shall be distributed to the calculation staff in order to prevent a situation in which different DNNs are used depending on the calculation staff.
- the infrastructure for communication within the communication node may actually perform the calculation.
- a server that performs calculations may be provided within the communication node.
- An information processing device, such as a communication node, that performs part of the cloud service from a location closer to the user than the cloud service (also called an edge) is generally called an edge server.
- the DNN calculations are not necessarily distributed to each of the communication terminal 11, the cloud system 12, and the communication network 13. Depending on the application, it is possible to complete the DNN calculation within the communication network 13 without using the cloud system 12 . In this case, since the total distance of the communication route is shortened, it is possible to further reduce the communication delay. Further, there may be a case where the communication terminal 11 does not perform DNN calculation, and the DNN calculation is distributed to the cloud system 12 and the communication network 13 . Alternatively, if a communication terminal 11 connected to the communication network 13 and having a large computational capacity is found apart from the communication terminal 11 that has executed the ML application, the The communication terminal 11 may be in charge of part of the DNN calculation. Alternatively, it may be determined in advance that at least one of the communication nodes existing between the communication route between the communication terminal 11 and the cloud system 12 is in charge of calculation.
- cloud system 12 is not necessarily in charge of the final calculation.
- cloud system 12 may perform computation first, and communication network 13 may take over computation of cloud system 12 .
- FIG. 4 is a diagram explaining the difference in delay and output data amount according to the splitting point.
- the dot-patterned bar graph in FIG. 4 indicates the amount of data output when calculation is performed from the input layer to the layer corresponding to the bar graph. According to FIG. 4, the amount of data output from each layer is not uniform, and it is preferable not to separate the DNN in a layer that outputs a large amount of data, because the communication delay does not increase.
- the white bar graph in FIG. 4 indicates the calculation delay in the layer corresponding to the bar graph. For example, the white bar graph corresponding to the layer named "fc6" is high, so we know that the calculation of the fc6 layer is time consuming. Therefore, it can be seen that it is preferable to let a device with high computational power take charge of the calculation of the fc6 layer.
- the delay of the ML application may increase due to changes in the status of the information processing system 1 .
- computational delays fluctuate because the computational spare capacity of a computational person is not always constant.
- the quality of the wireless communication link frequently fluctuates, so the communication delay tends to fluctuate.
- the communication terminal 11 is portable, the movement of the communication terminal 11 changes the communication route and the like. There may also be variations in network topology. Such a change in circumstances can cause an application's execution delay to exceed acceptable limits, even though it was initially within acceptable limits.
- the aforementioned IAB network aims to integrate backhaul links and access links, and not only access links but also backhaul links are wireless lines. Therefore, the status of the communication link is likely to change. Therefore, when the IAB network is included in the communication network 13 of the present embodiment, the communication delay is likely to fluctuate. Delays can get worse.
- the distribution is dynamically changed based on the status of the information processing system 1 . More specifically, based on the situation of the candidate in charge of calculation who can be in charge of calculation and the situation of the communication link between the candidates in charge of calculation, the person in charge of calculation, the range in charge of calculation, the communication route between the candidates in charge of calculation, etc. are changed.
- communication nodes within the network perform relay communication. This makes it possible to secure a communicable area (coverage) even in millimeter wave communication.
- TDM Time Division Multiplexing
- FDM Frequency Division Multiplexing
- SDM Space Division Multiplexing
- Efficient communication can be performed as compared with relay communication in a highly effective communication layer.
- communication using millimeter waves is assumed, and the coverage problem in millimeter wave communication can be improved by using relay communication like the IAB network, and coverage can be expanded efficiently. be able to do it effectively.
- Multi-hop communication is also assumed in the IAB network, and deployment to a mesh type is also assumed in the future.
- the IAB network is not limited to millimeter wave communications.
- it can be applied to vehicle tethering with IAB nodes installed in cars, moving cells installed in trains, and drone cells installed in drones.
- IoT Internet of things
- wearable devices that connects smartphones and wearable devices.
- fields such as medicine and factory automation. The same is true when applying the IAB network to this embodiment.
- FIG. 5 is a diagram illustrating an example architecture of an IAB network.
- the IAB-donor corresponding to the donor node 132 is assumed to be a communication node such as gNB (Next Generation Node B).
- gNB Next Generation Node B
- IAB-node which corresponds to the wireless communication node 131 and is a relay node, and these nodes are wirelessly connected while configuring a plurality of multi-hops.
- Each IAB-node connects UE (User Equipment) corresponding to the communication terminal 11 by an access link.
- An IAB-node may connect with multiple IAB-nodes to improve the redundancy of backhaul links.
- the IAB-node includes a function as a UE (MT) and a function as a communication node (DU). That is, it operates as MT when performing downlink (DL) reception and uplink (UL) transmission using the backhaul link, and operates as DU during DL transmission and UL reception. Since the IAB-node looks like a normal base station to the UE, even if the UE is a legacy terminal, it is possible to connect to the IAB network as shown in FIG. 5(B). It should be noted that not only the combination of MT and DU, but also the combination of MT and other MTs may be used.
- FIG. 6 is a diagram illustrating the effect of distributed computation of a DNN.
- the bar graph (1) shows the execution delay when the communication terminal 11 alone executes the DNN calculation.
- the bar graph (2) shows the execution delay when the communication terminal 11 and the cloud server in the cloud system 12 execute the DNN calculation.
- the bar graph of (3) shows the execution delay when the DNN calculation is executed by the communication terminal 11 and the MEC (Multi-access Edge computing) server, which is a kind of edge server, of the communication node in the communication network 13. indicates
- the bar graph (4) shows the execution delay when the DNN calculation is executed by the communication terminal 11, the MEC server, and the cloud server.
- the dot-patterned portion of the bar graph represents the calculation delay
- the white portion represents the communication delay.
- the communication terminal 11 uses a commercially available laptop
- the MEC server uses a Ryzen (registered trademark) 3800X CPU (Central Processing Unit) and a memory of 32 GB (Gigabyte)
- the cloud server is
- the CPU is Intel (registered trademark) Core i9-9900 and the memory is 128 GB, so that the cloud server has less calculation delay than the MEC server.
- the communication capacity between the communication terminal 11 and the MEC server is set to 100 Mbps (Mega bits per second), and the communication capacity between the MEC server and the cloud server is set to 30 Mbs.
- the execution delay is the largest at 212ms (milliseconds).
- communication delay increases.
- the communication delay is suppressed because the MEC server is close to the communication terminal 11, but the computational delay increases because the MEC server has lower computing power than the cloud server. It is larger than in the case of 2).
- the calculation delay is larger than when the cloud server executes all DNN calculations, and the communication delay is larger than when the MEC server executes all DNN calculations.
- the computational delay is lower than when the server performs all of the DNN computations, and the communication delay is less than when the cloud server performs all of the DNN computations.
- the execution delay is the minimum of 53 ms.
- the execution delay can be reduced by also using the communication nodes in the communication network 13 to distribute the DNN calculations.
- the calculation delay and the communication delay differ depending on the scope of responsibility, so the simulation results in FIG. 6 may also vary depending on the scope of responsibility.
- the execution delay in the case of (4) above may be greater than in the cases of (1) to (3). It is possible to reduce the execution delay in the case of (1) to (3) above.
- FIG. 7 shows a diagram of the network topology in the IAB network used for the simulation.
- the example network of FIG. 7 consists of 10 nodes: wireless communication nodes 131A through 131F of the IAB network, donor node 132 of the IAB network, communication node 1331A of the core network 133 and 1331B and the cloud server 121.
- the specifications of the wireless communication nodes 131A to 131F are the same as the MEC server used when showing the effect of distributed calculation of the DNN in the example of FIG.
- the specifications of the cloud server 121 are the same as those of the cloud server used to show the effect of distributed computation of the DNN.
- the access link and backhaul link of the IAB network shall share a communication capacity of 4 Gbps (Gigabit per second).
- the communication link between donor node 132 and communication node 1331A is a 1 Gbps wired link
- the communication link between communication nodes 1331A and 1331B is a 400 Mbps wired link
- the communication link between communication node 1331B and cloud server 121 is a 400 Mbps wired link.
- the communication link is a 100 Mbps wired link.
- the communication terminal 11 uses the above-mentioned commercially available laptop and moves as indicated by the arrow in FIG.
- the communication terminal 11 first connects to the nearest wireless communication node 131F, but the connected wireless communication node 131 is switched as it moves. Therefore, the communication route to the cloud server 121 is also switched. Therefore, every time the communication route is switched, the person in charge of calculation and the range in charge are determined, and the DNN is calculated.
- FIG. 8 is a diagram showing variations in communication capacity for simulation.
- FIG. 8A shows fluctuations in the communication capacity of the access link between the communication terminal 11 and the wireless communication node 131 in FIG.
- FIG. 8B shows variations in communication capacity between wireless communication nodes 131 in FIG.
- the communication capacity fluctuates from 200 Mbps to 800 Mbps over time.
- FIG. 9 is a diagram showing the influence of resources of the communication network 13 on execution delay.
- FIG. 9A shows the relationship between communication capacity and execution delay between the communication terminal 11 and the IAB node.
- the bar graphs in FIG. 9A represent execution delays, and the bar graph on the left where the communication capacity of the wireless communication link is large is smaller. That is, the execution delay improves as the communication capacity of the wireless communication link increases. Conversely, when the quality of the wireless communication link deteriorates and the communication capacity decreases, the execution delay also increases at the same time. Since the quality of the wireless communication link is likely to fluctuate, when distributing the DNN computation, it is necessary to change the setting for distribution in consideration of the quality of the wireless communication link. Also, FIG.
- FIG. 9B shows the relationship between the computational spare capacity of the IAB node in charge of computation and the execution delay.
- the bar graph on the left side where the computational spare capacity is large, is smaller, and the delay amount can be expected to be reduced as the computational spare capacity of the IAB node increases.
- the calculation delays in each layer of the DNN are also different, it is necessary to determine the range in charge according to fluctuations in the calculation spare capacity of the IAB node.
- KPIs Key Performance Indicators
- control targets control targets
- information to be used when implementing DNN distribution are shown below.
- the KPI is the execution delay of the ML application.
- the execution delay of the ML application includes at least a computation delay in each computational charge and a communication delay between each computational charge.
- the sum of each calculation delay and each communication delay is not taken into account after receiving the calculation result from the previous calculation person and before starting the calculation of the range of one's own responsibility. You can think of it as an execution delay.
- Control targets include routing, DL or UL configuration at each communication node (DL/UL configuration), DNN splitting points, and the like.
- the information to be used is assumed to be the processing capability of each computational candidate, the status of each wireless communication link, the required specifications of the ML application, the required specifications of the communication network 13, the movement status (mobility) of the communication terminal 11, and the like.
- the candidates in charge of calculation are the communication terminal 11, the cloud system 12, and the communication nodes in the communication network 13, but it may be determined in advance whether the communication terminal 11 and the cloud system 12 are in charge of calculation. , in that case, it may be removed from the candidate in charge of calculation.
- the processing capacity of each candidate in charge of calculation is assumed to be the calculation capacity, the current calculation spare capacity, etc. For example, first, among the candidates in charge of calculation belonging to the communication network 13, the candidate in charge of calculation having the highest calculation ability is appointed to be in charge of calculation. The responsibility may be changed to another candidate who has sufficient computation responsibility. In this way, the person in charge of calculation may be changed based on the available calculation capacity of the person in charge of calculation.
- the status of the communication link may include communication capacity and communication quality.
- the status of the backhaul link and access link is included.
- the allowable limit of the execution delay of the ML application in other words, the upper limit of the execution delay allowed by the ML application is assumed.
- an upper limit value that is allowed for the communication delay and the calculation delay may be set individually.
- the upper limit of traffic on each link is assumed as the required specification for communication. Also, an upper limit of traffic on a route set between the communication terminal 11 and the cloud system 12 may be set. These upper limits may be determined based on the requirements of the ML application and the splitting point of the DNN.
- the movement status of the communication terminal 11 may be any information related to movement such as movement speed, movement direction, and movement pattern.
- the determination of the person in charge of calculation and the range in charge may be performed by any device belonging to the information processing system 1, and is not particularly limited. In other words, it is possible to appropriately determine the entity that determines the charge of calculation and the range of charge. Note that when devices belonging to the information processing system 1, such as the communication terminal 11, communication node, and cloud server, are not distinguished, they are described as entities, and the subject responsible for calculation and the scope of responsibility is described as a logical entity. .
- a communication node of the communication network 13 or a server for making the determination may be implemented in the cloud system 12 and used as a logical entity.
- a logical entity may implement a module responsible for determining
- a device present in a position suitable for communication be a logical entity.
- one logical entity may determine both the responsibility for calculation and the scope of responsibility, or the logical entity that determines the responsibility for calculation and the logical entity that determines the scope of responsibility may be separated.
- the resources of the information processing system 1 include the computational spare capacity of computational candidates belonging to the information processing system 1, the communication capacity of the communication link within the communication network 13, the communication quality, and the like.
- Fluctuations in the communication environment include, for example, fluctuations in the quality of communication links, computational capacity of communication nodes, network topology, and communication routes.
- FIG. 10 is a schematic sequence diagram showing the flow of the overall processing of this embodiment.
- the communication node and the cloud server are shown as a set for convenience of explanation.
- the entity of the information processing system 1 is assumed to consist of components in charge of each process.
- a logical entity comprises a receiver, a transmitter, and a determiner.
- the candidate in charge of calculation such as the communication terminal 11, communication node, and cloud server, includes a receiving unit, a transmitting unit, an acquiring unit (measuring unit), a setting unit, and a calculating unit.
- the subject of each process in FIG. 10 is assumed to be the above components.
- the transmission unit of the logical entity transmits settings related to acquisition and transmission of information such as resources of the information processing system 1 used for determining who is in charge of calculation to each entity such as the communication terminal 11, the communication node, and the cloud server (T101, Measurement configuration).
- the receiving unit of each entity receives the acquisition setting from the logical entity (T102), the acquiring unit of each entity acquires resource-related information based on the setting (T103), and the transmitting unit of each entity responds to the setting. to the logical entity (T104).
- the receiving unit of the logical entity receives information on resources from each entity (T105), and the determining unit of the logical entity determines the control content of each entity in order to keep the execution delay of the ML application within the allowable limit ( T106). As will be described later, whether or not to be appointed as the person in charge of calculation is determined as the content of this control. Further, the determining unit of the logical entity determines parameter values, in other words, setting values, to be set in the communication terminal 11 and the communication node in order to implement the determined control content (T107, Parameter configuration). The determined setting value is transmitted to the communication terminal 11 and the communication node by the transmission section of the logic entity (T108).
- the receiving unit of each entity receives the setting value from the logical entity (T109), and the setting unit of each entity sets the parameter for operating each entity to the setting value (T110).
- T109 the setting value from the logical entity
- T110 the setting unit of each entity sets the parameter for operating each entity to the setting value (T110).
- the ML application is executed in the communication terminal 11 (T111).
- the calculation unit of the communication terminal 11 performs the calculation of the calculation range in charge.
- the transmission unit of the communication terminal 11 transmits information necessary for DNN calculation to the specified destination (T112). If the communication terminal 11 is designated to be in charge of calculation, the calculation results up to the middle of the series of calculations of the DNN are included in the information. is included in the information. Also, the designated destination is the person in charge of the following calculations.
- the receiving unit in charge of the next calculation receives the information necessary for calculating the DNN (T113), the calculation unit in charge of the next calculation calculates its own range (T114), and the transmission unit in charge of the next calculation
- the calculation result is further transmitted to the next person in charge of calculation (T115).
- the processing from T113 to T115 is performed by each person in charge of calculation. An entity that is not designated to be in charge of calculation does not calculate the DNN.
- the transmission unit in charge of the final calculation returns the calculation result to the communication terminal 11 .
- the receiving unit of the communication terminal 11 receives the final DNN calculation result (T116), and the ML application process is executed based on the final calculation result (T117). Thus, the processing of the ML application is completed.
- each entity acquires and transmits resources based on the acquisition settings, and each time the logical entity receives a resource, it determines whether the execution delay exceeds the allowable upper limit value. It may be determined whether or not, and if it is determined that it exceeds, the content of control may be changed. In this way, it may be prepared in case the ML application is run again. It should be noted that acquisition and transmission of resources may be stopped, and acquisition and transmission of resources may be resumed when activation of the ML application is detected.
- the information instructed to be obtained from the logical entity may be information regarding computational power.
- the information about the computational power includes, for example, the maximum computational capability (Capability), the computational spare capacity, the computational load (computational amount), the computational delay amount assumed from the computational load, and the like.
- the number of GPUs (Graphical Processor Units) included in each entity may be set as the maximum computational capacity.
- the number of currently unused GPUs may be used as the computational spare capacity.
- the information may be information about the status of the connected communication link.
- it may be information related to wireless communication link connection such as Radio link failure, or communication quality information of wireless communication link such as RSRP (Reference Signal Received Power), RSRQ (Reference Signal Received Quality), RSSI (Reference Signal Strength Indication). It's okay.
- RSRP Reference Signal Received Power
- RSRQ Reference Signal Received Quality
- RSSI Reference Signal Strength Indication
- the information may be information on the required specifications of the ML application. For example, there is an upper limit of delay allowed by the ML application. Note that the required specifications of the ML application may differ from one communication terminal 11 to another.
- the information may be information related to the traffic of the communication network 13. For example, traffic limits, traffic buffer status, etc. can be raised. Note that estimates may be used instead of actual traffic measurements.
- the information may be information related to movement (mobility) of the communication terminal 11 .
- the communication terminal 11 may move while the ML application is running. Since movement affects communication quality, information such as movement speed and movement direction may be acquired.
- the information may be information related to DNN calculation.
- each entity may estimate the computational delay for each layer of the DNN.
- a plurality of candidate areas of responsibility may be defined in advance, and the logical entity may instruct each entity to estimate the calculation delay for each candidate area of responsibility.
- the load for example, the GPU usage rate
- the calculation delay may be calculated based on the past calculation history, or calculated as a theoretical time when calculating the data size shown in FIG. 4 assuming that the current calculation spare capacity continues. good too.
- the entity may actually measure the indicated information and transmit the measured value to the logical entity.
- future estimates calculated based on actual measurements may be sent to the logic entity. For example, if the scheduled execution time of the ML application is 10 seconds later, the predicted position of the communication terminal 11 after 10 seconds may be transmitted to the logical entity.
- the communication terminal 11 and the communication node may quantize the actual measurement value, determine which of the predetermined classification items the actual measurement value corresponds to, and store the information of the corresponding classification item as a logical entity. You may send. The estimation may be made based on past records.
- Publicly known techniques may be used to acquire information on resources. For example, information about the performance of an entity, such as computing power and computing power, may be acquired using functions such as tools provided by an OS (operating system) installed in the entity. Also, information about the quality of the communication link, eg, communication quality such as RSRQ, may be ascertained using known techniques.
- information about the performance of an entity such as computing power and computing power
- functions such as tools provided by an OS (operating system) installed in the entity.
- information about the quality of the communication link eg, communication quality such as RSRQ, may be ascertained using known techniques.
- a communication node that acts like a representative by collecting information to be sent to the logical entity and sending it to the logical entity as a representative.
- information such as the traffic of each link and the movement of the communication terminal 11 may be transmitted to the logical entity after summing the information from a plurality of communication nodes.
- the timing of information acquisition may be specified.
- a periodic measurement may be indicated.
- the logical entity may determine the acquisition start time, acquisition end time, and acquisition period and instruct each entity, and each entity may acquire according to the instruction.
- the number of acquisitions, the repetition waiting period, and the like may be indicated.
- dynamic acquisition Trigger based measurement
- Trigger conditions for each entity to start acquisition dynamically may be defined as appropriate. For example, acquisition may be initiated upon detection of a wireless communication link failure. Alternatively, the acquisition may be started when the node processing load, ML application delay, communication delay, or the like exceeds a predetermined threshold. Note that these thresholds may be adjusted by the logical entity. Alternatively, acquisition may be initiated upon receipt of a request for acquisition. The request may be transmitted from the logical entity, or may be transmitted from a higher node different from the logical entity.
- the backhaul link in order to measure the quality of the backhaul link, it may be specified to periodically measure the RSRQ of the backhaul link in a period of 10 ms, for example, at intervals of 100 ms.
- the transmission of this information to the logical entity can be done as appropriate, and the timing of transmission and the format of the data to be transmitted are not particularly limited.
- the transmission may also be performed periodically.
- it may be performed when conditions such as when the value of the RSRQ of the communication link becomes equal to or less than a predetermined threshold, or when the processing load of the node becomes equal to or greater than a predetermined threshold.
- it may be transmitted immediately after acquisition, or may be transmitted after an offset time from acquisition has elapsed.
- it may be sent when the acquired value satisfies the conditions. For example, if there is a change that requires a change in the person in charge of calculation, the range in charge, etc., a report may be issued, and if not, the report may not be issued.
- each entity does not have to send all of the acquired information to the logical entity.
- information may be acquired with fine granularity, and only information that satisfies a predetermined condition, such as information with large fluctuations or information exceeding a threshold, may be transmitted to the logical entity. That is, the logical entity may separately indicate the information to be obtained and the information to be reported. Also, the acquired information may be processed as appropriate for reporting to the logical entity.
- the settings may be different for each entity. For example, since the communication link connected to the cloud system 12 is assumed to be wired and stable, the cloud system 12 may not need to acquire information about the communication link.
- the determined control contents include communication links and wireless communication parameters.
- the person in charge of calculation, the range of responsibility, etc. are determined.
- Controls related to communication links include, for example, determination of communication routes.
- the communication network 13 includes a relay-type network such as an IAB network
- a relay route is determined. Even if an attempt is made to select a person in charge of calculation from among the communication nodes on the communication route between the communication terminal 11 and the cloud system 12, the selection cannot be made unless there is a communication node on the communication route that has calculation capacity. Therefore, the logical entity may determine the communication route using not only the quality of the communication link, but also the computing power, computing power, etc. of the communication node. Also, the IAB node to be passed through and the number of hops may be changed in the same manner.
- Controlling communication parameters includes, for example, improving the quality of communication links on communication routes.
- the logical entity may transmit a setting value that increases the intensity (transmission power) of radio waves to be transmitted to the wireless communication node 131 on the communication route.
- the wireless communication node 131 may reduce the communication capacity of the wireless communication link that is not on the communication route so as not to cause interference. In this manner, settings may be determined that improve the quality of the communication link.
- the correspondence relationship between the downlink (DL) and the uplink (UL) in the wireless communication link may be changed.
- Wireless communication links can be adjusted to increase one of the DL and UL communication bands and decrease the other. Therefore, the correspondence between DL and UL may be adjusted so as to reduce communication delay.
- the communication delay may be calculated from the size of the data to be transmitted and the communication capacity of the communication link through which the data flows. A delay due to communication quality may also be considered.
- the computational responsibility and the responsible range are determined in consideration of the computational capacity of each wireless communication node 131, the amount of data output in each responsible range, the quality of the communication link on the communication route, and the like. At least the wireless communication node 131, which becomes a delay bottleneck, should not be in charge of calculation.
- FIG. 11 is a diagram explaining the Splitting mode.
- FIG. 11 shows four Splitting modes.
- a table showing a plurality of splitting modes as shown in FIG. 11 is also referred to as a splitting mode table.
- the responsible range of each calculation is determined.
- the communication terminal 11, the communication node in the communication network 13, and the cloud system 12 are in charge of calculation, but a splitting mode in charge of calculation may be prepared.
- a specific Splitting mode may be selected by default, and then switched to another Splitting mode.
- the splitting mode in the second line in which the range of calculations other than the communication terminal 11 is small, is selected to increase the load on the communication terminal 11. It is conceivable to have them take over. In this way, when there is a load on a particular person in charge of calculation, it can be easily improved by switching to the Splitting mode in which the area in charge of the person in charge of calculation is small.
- the consideration of whether or not to switch the splitting mode may be performed periodically or dynamically.
- Both the splitting mode to be used normally and the temporary splitting mode to be used when it is determined that the splitting mode used normally cannot meet the requirements of the ML application may be determined. Then, the splitting mode can be quickly switched without performing the process of selecting an appropriate splitting mode when it is determined that the request of the ML application cannot be satisfied.
- the content of the Splitting mode that is, the range of responsibility of each person in charge of calculation may be appropriately updated by the logical entity.
- the updated splitting mode is sequentially notified to each entity so that each calculation operator does not perform calculations based on the splitting mode before updating.
- FIG. 12 is a diagram explaining the Splitting mode set for each communication route.
- FIG. 12 shows three communication routes Route_A, Route_B, and Route_C. A plurality of splitting modes as shown in FIG. 11 are set for each of these three communication routes.
- the communication node of the core network 133, the donor node 132, the wireless communication node 131C, the wireless communication node 131A, and the communication terminal 11A are present on the communication route Route_A.
- DNN layers are assigned to these calculation candidates to create a Splitting mode table.
- candidates for calculation are selected and a Splitting mode table is created.
- FIG. 13 is a diagram showing an example of Splitting mode for each communication route.
- FIG. 13A shows the splitting mode table for the communication route Route_A
- FIG. 13B shows the splitting mode table for the communication route Route_B.
- the number of DNN layers is assumed to be 40
- the numerical value of each cell of the Splitting mode table indicates the number of layers that the corresponding candidate in charge of calculation is in charge of. If "0" is written in a cell, it means that there is no layer for which the corresponding candidate in charge of calculation is in charge. In other words, it means that you will not be in charge of calculation.
- the logical entity determines the range of responsibility, that is, the Splitting mode.
- the communication terminal 11 selects the Splitting mode from the Splitting mode table of the communication route after the change, and the selected Splitting mode is changed.
- Splitting mode can be reset by notifying each person in charge of calculation.
- the range in charge of calculations using wired links may be fixed.
- the cloud system 12 and the edge server of the core network 133 do not perform wireless communication, it is considered that there is little change in the state of the communication link.
- the splitting mode of the communication route Route_A shown in FIG. 13(A) includes seven candidates in charge of calculation.
- the logical entity may change the Splitting mode table based on the anchor point.
- An anchor point is a communication node that always exists on a communication route set for communication terminal 11 as long as communication terminal 11 is within the assumed movement area. Although the communication route changes as the communication terminal 11 moves, the anchor point is a communication node common to all communication routes that can be set within the assumed movement area of the communication terminal 11 . For example, in the example of FIG. 12, if the communication terminal 11 wirelessly connects to any one of the wireless communication nodes 131A to 131D, the donor node 132 always exists on the communication route with the cloud system 12. FIG. Thus, in the example of Figure 12, donor node 132 is the anchor point.
- the logical entity determines the Splitting mode from the Splitting mode table as long as the anchor point does not disappear from the communication route, and restarts the Splitting mode table itself when it detects that the anchor point has disappeared from the communication route. You can make settings.
- the splitting mode table may be recreated when a predetermined communication node no longer exists on the communication route.
- the logical entity may change the scope of responsibility of each calculation on a layer-by-layer basis. For example, if the load on the communication terminal 11 rises slightly after the range in charge of the communication terminal 11 is determined to be layers 1 to 4, the range in charge of the communication terminal 11 is changed to layers 1 to 3. An adjustment may be made such that the layer 4 that has become the next calculation is in charge. In this case, control is performed at a finer granularity than the Splitting mode level, and although the load on the logical entity increases, the risk of not satisfying the requirements of the ML application can be reduced.
- the communication terminals 11 and communication nodes update the values of parameters related to communication links, areas of responsibility, etc., according to what is determined by the logic entity.
- the parameter setting instruction may be given directly from the logical entity, or may be given indirectly via the representative wireless communication node 131 that bundles the plurality of wireless communication nodes 131 .
- the notification method is not particularly limited, and may be signaling notification in the application layer or signaling notification in the physical layer. Semi-static notification such as RRC (Radio Resource Control) signaling, or dynamic notification such as DCI (Downlink control information) or UCI (Uplink control information) may be used.
- sequence diagram when the calculation charge is switched is also illustrated.
- 14A and 14B are sequence diagrams before and after the switching of the person in charge of calculation.
- the blocks in FIG. 14 are labeled with the reference numerals of the processes shown in FIG.
- FIG. 14 shows the case where the donor node 132 has implemented the logical entity. Also, initially, the communication terminal 11, the wireless communication node 131A, the wireless communication node 131C, and the cloud system 12 were in charge of calculation, but the quality of the backhaul link between the wireless communication node 131A and the wireless communication node 131C deteriorated. and the scope of responsibility is switched. It should be noted that the processing up to the parameter setting (T110) shown in FIG. 10 has already been executed, and the processing from T111 is shown.
- T110 parameter setting
- the ML application of the communication terminal 11 is executed (T111), and the communication terminal 11 transmits information necessary for DNN calculation to the next person in charge of calculation (T112).
- the wireless communication node 131A in charge of the next calculation receives the information (T113), calculates the range in which it is in charge (T114), and transmits the calculation result to the wireless communication node 131C in charge of the next calculation ( T115).
- the wireless communication node 131C similarly executes the processing from T113 to T115, and the calculation result of the wireless communication node 131C is transmitted to the cloud system 12 in charge of the next calculation.
- the cloud system 12, which is in charge of the next calculation similarly executes the processing from T113 to T115, and since the cloud system 12 is in charge of the final calculation, the cloud system 12 transmits the final calculation result of DNN to the communication terminal 11. be done.
- each entity periodically acquires resources (T103), and the wireless communication node 131A that detected the problem reports to the donor node 132, which is a logical entity (T104).
- T104 a logical entity
- the core network 133 and the cloud system 12 are set so as not to report to the logical entity, and the core network 133 and the cloud system 12 do not show the block T104.
- other entities are also set not to report to the logical entity when no deterioration is detected. Therefore, since no entity other than wireless communication node 131A that detected the problem has reported, the block of T104 is not shown.
- each entity performs measurements on the backhaul link. Then, it is assumed that the radio communication node 131A detects that the RSRQ value of the backhaul link with the radio communication node 131C has become equal to or less than a predetermined value, and transmits it to the logical entity.
- the donor node 132 which is a logical entity, receives the report of the wireless communication node 131A, determines from the reporting result that it is not enough to increase the bandwidth of the backhaul link in question, and makes new settings such as changing the calculation charge. is determined and transmitted to each entity (T105 to T108). Note that in the example of FIG. 14 , the logical entity transmits settings only to entities that require new settings, so no arrow indicating transmission is shown in the core network 133 and the cloud system 12 . Note that the settings may be sent to entities that do not require new settings.
- the logical entity may request additional reports from each entity. For example, when receiving a report from the wireless communication node 131A that there is a problem with the backhaul link, communication nodes around the wireless communication node 131A in order to examine whether it is possible to respond by increasing the band of the backhaul link may be requested to send reports such as traffic buffers.
- Each entity that has received new settings from the logical entity receives the new settings and sets them as parameters (T109, T110).
- T109, T110 the new settings and sets them as parameters.
- the backhaul link from the wireless communication node 131A to the wireless communication node 131C disappears and the backhaul link from the wireless communication node 131A to the wireless communication node 131D is newly installed.
- the communication route is changed, the wireless communication node 131C not on the communication route is removed from the calculation, and the wireless communication node 131D is added to the calculation.
- the ML application is executed again (T111), and the communication terminal 11 transmits information necessary for DNN calculation to the wireless communication node 131A in charge of the next calculation (T112).
- the wireless communication node 131A receives the information (T113) and calculates its own range (T114). It transmits to the wireless communication node 131D (T115). Accordingly, unlike the previous time, the processing from T113 to T115 is not executed in the wireless communication node 131C.
- the wireless communication node 131D similarly executes the processing from T113 to T115, and the calculation result of the wireless communication node 131D is transmitted to the cloud system 12 in charge of the next calculation.
- the cloud system 12, which is in charge of the next calculation similarly executes the processing from T113 to T115, and since the cloud system 12 is in charge of the final calculation, the cloud system 12 transmits the final calculation result of DNN to the communication terminal 11. be done.
- a splitting mode that reduces the coverage of the wireless communication node 131C may be selected from the splitting mode table as shown in FIG. Further, for example, until now, the radio communication node 131A was in charge of layers 20 to 25 of the DNN, and the radio communication node 131C was in charge of layers 26 to 40 of the DNN. The coverage of the node 131A may be increased from layer 20 to layer 29 of the DNN, and the coverage of the wireless communication node 131D may be from layer 30 to layer 40 of the DNN. In this way, the radio communication node 131C may continue to be in charge of calculation.
- the change in the responsible range may be made even when the responsible for the calculation is changed as in the example of FIG.
- the information processing system 1 includes the communication terminal 11, the communication network 13, and the cloud system 12, but in reality, these are assumed to have different owners. Also, it is assumed that the owner of the network for accessing the communication terminal 11 such as the IAB network and the core network 133 are different. Therefore, the range that can be specified and set by the logical entity may be part of the information processing system 1 . For example, if the logical entity is a communication node within the IAB network, the logical entity may not be able to change the calculation range of the cloud system 12, and may only perform settings for the communication node within the IAB network.
- the input to the DNN is transmitted from the communication terminal 11 to the external device.
- the communication terminal 11 is in charge of at least the calculation from the beginning to the middle of a series of DNN calculations so as not to transmit the input data itself to the outside, such a problem can be alleviated.
- the entity that decides the responsibility of calculation and the scope of responsibility is described as a logical entity, and it is assumed that the logical entity is handled by a communication node of the communication network 13, a cloud server, or the like.
- a device suitable for grasping the resource status may be a logical entity so that the person in charge of calculation and the range of responsibility can be determined according to the resource status of the information processing system 1.
- a device that issues an instruction to improve the quality of the communication link to the wireless communication node 131 on the communication route is defined as a logical entity.
- Communication terminal 11 can also be a logical entity. In other words, the communication terminal 11 may determine the charge of calculation and the range of charge.
- each entity such as the communication terminal 11 periodically transmits resources to the logical entity, and the logical entity is in charge of calculation and charge based on the resource of each entity.
- the range was decided and notified to each person in charge of calculation. Therefore, when the communication terminal 11 activates the ML application or when the ML application executes the DNN calculation, the person in charge of the calculation and the range in charge have already been determined.
- the communication terminal 11 it is also possible for the communication terminal 11 to determine its own range of responsibility by notifying the communication terminal 11 in advance of the conditions for determining the range of responsibility from a logical entity or the like. Note that the entity that transmits the conditions to the communication terminal 11 may not be a logical entity.
- the communication terminal 11 checks items such as its own calculation capacity, communication quality with the cloud system 12, etc., and determines to which position the DNN calculation is performed according to the items. You may Note that the communication terminal 11 may decide to perform calculations up to any layer of the DNN, or may decide to perform calculations up to a part of a plurality of calculations set for nodes in a certain layer. Alternatively, the logic entity may notify the communication terminal 11 of the minimum range of calculations to be performed after determining the responsibility of each calculation, and the communication terminal 11 may expand the range of responsibility according to the matter.
- the logic entity after determining each calculation responsibility, notifies the communication terminal 11 of the range of responsibility that may be performed (in other words, the upper limit of the range of responsibility), and the communication terminal 11 logically You may reduce the scope of responsibility notified by the entity.
- the communication terminal 11 When the communication terminal 11 holds the conditions for determining the range of responsibility of the communication terminal 11 and dynamically determines the range of responsibility of the communication terminal 11, the communication terminal 11 activates the ML application or the ML application Responsibilities can be determined based on the resources at which DNN performs the DNN computation. Therefore, the coverage of the communication terminal 11 can be made according to the state of the communication terminal 11 . In this case, it is possible to reduce the number of periodic transmissions of resources from the communication terminal 11 to the logical entity and the number of times the logical entity notifies the communication terminal 11 of changes in the scope of responsibility, thereby reducing the processing load of each entity and the use of communication resources. can be mitigated.
- FIG. 15 is a diagram showing an example of conditions for determining the coverage area of the communication terminal 11.
- FIG. 15A conditions for determining the DNN calculation range based on the computational capacity of the communication terminal 11 are shown.
- the range in charge is indicated as n. 11 is in charge.
- n is assumed to be an integer of 10 or more.
- the n-th layer may be the final layer of the DNN, or the final layer of the range notified by the logical entity.
- the assigned range is reduced as the computational spare capacity is reduced.
- FIG. 15 conditions for determining the DNN calculation range based on the computational capacity of the communication terminal 11 are shown.
- n is assumed to be an integer of 10 or more.
- the n-th layer may be the final layer of the DNN, or the final layer of the range notified by the logical entity.
- the range in charge when the computational capacity is less than 90% and 80% or more, the range in charge is shown to be up to the 4n/5th layer, which is lower than when the computational capacity is 90% or more. ing.
- the computational spare capacity is less than 80% and 60% or more, the range in charge is up to the 3n/5 layer, and if the computational spare capacity is less than 60% and 40% or more, the range in charge is the 3rd layer. Up to 2n/5 layers are shown, and when the computational capacity is less than 40% and 20% or more, the range in charge is shown up to the n/5th layer. In this way, the coverage area of the communication terminal 11 may be determined.
- the computational capacity when the computational capacity is other than that, that is, when it is less than 20%, the coverage range is up to the first layer, which means that the communication terminal 11 does not perform DNN computation. In other words, even if the communication terminal 11 is designated to be in charge of calculation, the communication terminal 11 may refuse to perform the calculation. By reducing the range in charge when the calculation capacity of the communication terminal 11 is small in this way, it is possible to prevent a situation in which the calculation of the range in charge takes time due to the small calculation capacity of the communication terminal 11 .
- FLOPS the product of the clock frequency and the number of operations per clock
- the responsible range is determined similarly to FIG. 15(A), but the conditions are based on delay time, which is a type of communication quality.
- delay time which is a type of communication quality.
- the delay time with any communication destination may be determined in advance, and is not particularly limited. It may be in charge of the next calculation, a logical entity, or a wireless communication node to which the communication terminal 11 is wirelessly connected.
- the main cause of the delay time is the wireless processing performed by each entity, the time associated with the wireless processing may be regarded as the delay time without considering the propagation delay in wireless (radio waves) and wired.
- the communication terminal 11 when the delay time is 500 ms or more, the communication terminal 11 is in charge of calculations from the first layer to the n-th layer of the DNN.
- the responsible range When the delay time is less than 500 ms and 250 ms or more, the responsible range is up to the 4n/5 layer, and when the delay time is less than 250 ms and 100 ms or more, the responsible range is up to the 3 n/5 layer.
- the delay time is less than 100 ms and 50 ms or more, the responsible range is up to the 3n/5th layer, and when the delay time is less than 50 ms and 10 ms or more, the responsible range is the 2nth layer. /5 layers, and the communication terminal 11 does not calculate the DNN if the delay time is other than that, that is, if it is less than 10 ms.
- the combination of the delay time and the assigned range may be determined by referring to the data shown in FIG. 4 and considering the data size of the calculation result in each layer.
- conditions may be appropriately set according to the specifications of the embodiment, and are not particularly limited.
- conditions can be changed for each type of ML application.
- it may be changed when a plurality of conditions are provided and all of them are satisfied, or may be changed according to the condition with the highest predetermined priority among the satisfied conditions.
- the degree of confidentiality is determined in advance for each type of ML application, and when the degree of confidentiality of the executed ML application is equal to or greater than a predetermined threshold value, the coverage of the communication terminal 11 is set from the first layer to the second layer or higher. can be By doing so, the communication terminal 11 does not transmit DNN input data to the outside. As a result, the risk of leakage of information with a high degree of confidentiality to other than the communication terminal 11 can be reduced.
- the communication terminal 11 when the communication terminal 11 has determined its own range, the next person in charge of calculation cannot recognize from which layer of the DNN the calculation should be started. Therefore, for example, although the logical entity has notified each person in charge of calculation of the range of responsibility, if the communication terminal 11 changes the range of responsibility notified from the logical entity, the communication terminal 11 will change the range of responsibility of the next person in charge of calculation. There is also a risk of inputting the calculation results from the communication terminal 11 to each node of the first layer of the planned area of responsibility without knowing the change. Therefore, when the communication terminal 11 decides or changes its assigned range, the communication terminal 11 needs to notify not only the calculation result but also information for recognizing the position where the next person in charge of calculation starts the calculation. .
- the information may be, for example, information indicating the last layer in the range in charge of the communication terminal 11, information indicating the first layer in the range in charge of the next calculation, or information indicating the node that outputs the calculation result. Alternatively, it may be information indicating the node to which the calculation result should be input.
- the communication terminal 11 may transmit the information directly to the next person in charge of calculation, or may transmit the information to the next person in charge of calculation via a logical entity.
- FIG. 16 is a diagram showing an example of a calculation result transmitted from the communication terminal 11 when the communication terminal 11 has determined its own area of responsibility.
- the output value of each node, which is the calculation result, and the identification information for identifying the node that output the output value are included.
- the identification information (identifier) of the node is described as "node 3_4", but the number at the end indicates the number of the layer containing the node, which is behind "node”. A number indicates the node number in the layer. That is, "node 3_4" indicates the third node included in the fourth layer.
- output 3 described in the same line as “node 3_4" indicates the output value by the third node included in the fourth layer. To which node the output of each node is input can be recognized from the structure of the DNN or the like. After receiving the information shown in FIG. 16, the next person in charge of calculation should recognize the node to which the received output value should be input from the structure of the DNN, etc., and start the calculation.
- the person in charge of calculation performs the calculation set for each node in the area in charge and sends the output value of the node belonging to the last layer in the area in charge to the next person in charge of calculation.
- a node of a DNN is set up with multiple computations. Therefore, the person in charge of calculation may perform up to a part of the plurality of calculations set in the node, and the rest may be performed by the next person in charge of calculation.
- each input data input to the node is multiplied by a weighting factor set to the link through which each input data passed, and then added. Furthermore, the bias value set for each node is added to the added value.
- the added value is input to a predetermined activation function, and the output from the activation function becomes the output value of the node. Therefore, for example, it may be determined in advance that a person in charge of calculation performs up to the calculation of the added value, and the next person in charge of calculation starts with the calculation of the activation function, and the calculations may be shared in such a manner.
- a link connected to a node is also called an edge.
- FIG. 17 is a schematic sequence diagram showing the flow of overall processing when the communication terminal 11 determines its own range of responsibility.
- the cloud system 12 manages the structure of the DNN used by the ML application, the conditions for determining the range in charge of a series of calculations of the DNN, and the like.
- the communication terminal 11 and the cloud system 12 are in charge of the DNN calculation, but the communication terminal 11 and the communication node may be in charge of the calculation.
- the functionality of the core network 133 can be implemented within the cloud system 12 . That is, the core network 133 can also manage the above conditions.
- the cloud system 12 transmits information such as the DNN used by the ML application, the settings of the DNN, and the conditions for determining the scope of responsibility (T201).
- the information is transferred via the communication node of the communication network 13, the communication terminal 11 receives the information (T202), and sets the ML application such as DNN to be used based on the information. (T203).
- the communication node is a connection request of the communication terminal 11, for example, a service request (Service Request) or a PDU (Protocol Data Unit) session establishment request (PDU Session Establishment Request), 5QI (5G QoS Identifier) or S-NSSAI (Single-Network Slice Selection Assistance Information), etc.
- a service request Service Request
- PDU Session Establishment Request PDU Session Establishment Request
- 5QI 5G QoS Identifier
- S-NSSAI Single-Network Slice Selection Assistance Information
- the communication terminal 11 decides to execute the ML application (T204).
- the communication terminal 11 confirms the processing capacity of the communication terminal 11 itself (T205), and determines the range in charge of the communication terminal 11 based on the conditions and processing capacity for determining the range in charge of the DNN calculation (T205).
- T206 For example, if the condition for determining the DNN's assigned range is the example shown in FIG. %, the communication terminal 11 determines the fourth layer as the layer for dividing the DNN.
- the communication terminal 11 executes the ML application and calculates the coverage of the communication terminal 11 (T207). In the previous example, calculations are performed for the first to fourth layers of the DNN.
- the scope of responsibility may be expanded again after the scope of responsibility is calculated. For example, it may be checked whether or not a predetermined condition is satisfied after the calculation of the range in charge is completed, and based on the confirmation result, it may be determined whether or not to continue the calculation in the next layer.
- whether or not a predetermined condition is satisfied may be determined based on computational capacity, delay time, degree of confidentiality, and the like. In this way, the range of responsibility may be determined multiple times.
- the communication terminal 11 After calculating the coverage area of the communication terminal 11, the communication terminal 11 transmits information showing the coverage area of the communication terminal 11 and the calculation result as shown in FIG. 16 to the cloud system 12 via the communication node (T208). .
- the cloud system 12 receives the information via the communication node (T209).
- the cloud system 12 Based on the received identification information of each node, the cloud system 12 identifies the node to which the received output value is input, that is, each node in the layer next to the last layer in the range handled by the communication terminal 11. Calculate the range of responsibility of the system 12 (T210). After completing the calculation, the cloud system 12 returns the calculation result of the range in charge of the cloud system 12 to the communication terminal 11 (T211). It should be noted that the cloud system 12 is assumed to cover all the remaining calculations of the DNN, but it does not have to cover all the remaining DNN calculations. For example, the communication terminal 11 may receive the calculation result of the cloud system 12 and further calculate the remaining DNN.
- the communication terminal 11 receives the calculation result of the cloud system 12 via the communication node (T212). Then, the ML application process is executed based on the final calculation result (T213). Thus, the processing of the ML application is completed.
- An entity other than the communication terminal 11, such as the cloud system 12, may calculate up to the processing result of the ML application.
- the communication terminal when distributed learning of the DNN is performed among the entities, the communication terminal holds the conditions for determining the range in charge of the DNN, and determines the range in charge of the communication terminal. can be distributed more appropriately according to the situation.
- the communication terminal by causing the communication terminal to perform DNN calculation up to at least the second layer according to the degree of secrecy of the ML application, it is possible to prevent the occurrence of a situation such as leakage of input data.
- CNN convolutional neural networks
- RNN recurrent neural networks
- LSTM Long Short-Term Memory
- hidden layers are composed of layers called convolution layers and pooling layers.
- filtering is performed by a convolution operation to extract data called a feature map.
- the pooling layer compresses the information of the feature map output from the convolutional layer and performs downsampling.
- the RNN has a network structure in which hidden layer values are recursively input to the hidden layer, and for example, short-term time series data are processed.
- LSTM processes time-series data of a longer period than RNN.
- image recognition is used for applications such as tagging people in SNS (Social Network Service) and automatic driving. Speech recognition is applied to smart speakers and the like.
- Natural language processing is applied to search by browser and automatic translation. Anomaly detection by robots is used in airports, railways, manufacturing sites, and the like.
- a communication node is called a communication base station (also simply called a base station) and includes an infrastructure for communication, and the infrastructure is also called a base station device.
- a base station device is a type of communication device, and can also be said to be an information processing device.
- the base station device may be a device that causes a communication node to function as a wireless base station (Base Station, Node B, eNB, gNB, etc.), wireless access point (Access Point), or the like.
- the base station device may be a device that causes a communication node to function as a donor station or relay station.
- the base station device may be an optical extension device called RRH (Remote Radio Head).
- the base station device may be a device that causes the communication node to function as a receiving station such as an FPU (Field Pickup Unit).
- the base station device serves as an IAB (Integrated Access and Backhaul) donor node or IAB relay node that provides radio access lines and radio backhaul lines by time division multiplexing, frequency division multiplexing, or space division multiplexing. It may be a device that functions.
- the base station apparatus may be composed of a plurality of apparatuses, and may be, for example, a combination of an antenna installed in a structure such as a building and a signal processing apparatus connected to the antenna.
- the radio access technology used by the base station apparatus may be cellular communication technology or wireless LAN technology.
- the radio access technologies used by the base station apparatus are not limited to these, and other radio access technologies may be used.
- the radio access technology used by the base station apparatus may be LPWA (Low Power Wide Area) communication technology.
- wireless communication used by the base station apparatus may be wireless communication using millimeter waves.
- the wireless communication used by the base station apparatus may be wireless communication using radio waves, or may be wireless communication using infrared rays or visible light (optical wireless).
- the base station device may be capable of NOMA (Non-Orthogonal Multiple Access) communication with the communication terminal 11 .
- NOMA communication refers to communication (transmission, reception, or both) using non-orthogonal resources.
- the base station apparatus may be capable of NOMA communication with other base station apparatuses.
- the base station devices may be able to communicate with each other via a base station-core network interface (eg, S1 Interface, etc.). This interface can be wired or wireless. Also, the base station devices may be able to communicate with each other via inter-base station interfaces (eg, X2 Interface, S1 Interface, etc.). This interface can be wired or wireless.
- a base station-core network interface eg, S1 Interface, etc.
- This interface can be wired or wireless.
- inter-base station interfaces eg, X2 Interface, S1 Interface, etc.
- the base station devices may be able to communicate with each other via a base station-core network interface (eg, NG Interface, S1 Interface, etc.). This interface can be wired or wireless. Also, the base station devices may be able to communicate with each other via inter-base station interfaces (eg, Xn Interface, X2 Interface, etc.). This interface can be wired or wireless.
- a base station-core network interface eg, NG Interface, S1 Interface, etc.
- This interface can be wired or wireless.
- inter-base station interfaces eg, Xn Interface, X2 Interface, etc.
- base station may also mean a structure with the functions of a base station.
- the structure is not particularly limited.
- buildings such as skyscrapers, houses, steel towers, station facilities, airport facilities, harbor facilities, office buildings, school buildings, hospitals, factories, commercial facilities, and stadiums are included in the structures.
- non-building structures such as tunnels, bridges, dams, fences, steel poles, etc., and facilities such as cranes, gates, windmills, etc. are also included in the relevant structures.
- the place where the structure is installed is not particularly limited. In other words, not only on land (in a narrow sense, above ground) or underground structures, but also on water structures such as piers and mega floats, and underwater structures such as ocean observation equipment, are structures with base station functions. can be.
- the base station may be a fixed station or a mobile station, as described above.
- a base station may become a mobile station by installing a base station apparatus in a mobile object.
- the base station apparatus may have mobility, and the base station apparatus itself may become a mobile station by moving.
- vehicles, UAVs (Unmanned Aerial Vehicles) represented by drones, which originally have the ability to move, are equipped with the functions of a base station (at least part of the functions of a base station), and mobile stations are also mobile. It can also be said that it is a base station device as a station.
- Devices such as smartphones that move by being carried by a mobile object and that have the functions of a base station (at least part of the functions of a base station) are also referred to as mobile stations and base station equipment of mobile stations. It can also be said.
- a mobile object that constitutes a mobile station may be a mobile object that moves on land (ground in a narrow sense) (for example, vehicles such as automobiles, bicycles, buses, trucks, motorcycles, trains, linear motor cars, etc.). However, it may be a mobile body (e.g., subway) that moves underground (e.g., in a tunnel), or a mobile body that moves on water (e.g., a passenger ship, a cargo ship, a ship such as a hovercraft).
- a mobile body e.g., subway
- water e.g., a passenger ship, a cargo ship, a ship such as a hovercraft
- It may be a mobile body that moves underwater (for example, a submarine, a submarine, an unmanned underwater vehicle, etc.), or a mobile body that moves in the air (for example, an airplane, an airship, a drone, etc.) aircraft, etc.), or in other words, a moving body capable of floating in outer space (for example, an artificial celestial body such as an artificial satellite, a spacecraft, a space station, or a probe).
- a base station floating outside the atmosphere is also called a satellite station.
- a base station located closer to the earth than outside the atmosphere is also called a ground station.
- a base station floating in the atmosphere, such as an aircraft is also called an aircraft station.
- the satellites that will become satellite stations are Low Earth Orbiting (LEO) satellites, Medium Earth Orbiting (MEO) satellites, Geostationary Earth Orbiting (GEO) satellites, and Highly Elliptical Orbiting (HEO) satellites. ) satellite.
- LEO Low Earth Orbit
- MEO Medium Earth Orbit
- GEO Geostationary Earth Orbit
- HEO Highly Elliptical Orbit
- unmanned aircraft control systems include unmanned aerial systems (UAS: Unmanned Aircraft Systems), tethered UAS, LTA (Lighter than Air UAS), HTA (Heavier than Air UAS), HAPs (High Altitude UAS Platforms).
- UAS Unmanned Aircraft Systems
- LTA Lighter than Air UAS
- HTA Heavier than Air UAS
- HAPs High Altitude UAS Platforms
- these control systems may control the flight of aircraft stations.
- the size of the coverage of the base station device is not particularly limited, and may be as large as a macro cell, as small as a pico cell, or as extremely small as a femto cell.
- the base station apparatus may have beamforming capability. In this case, the base station apparatus may form a cell or service area for each beam. Therefore, the base station apparatus may be equipped with an antenna array composed of a plurality of antenna elements and configured to provide advanced antenna technology represented by MIMO (Multiple Input Multiple Output) and beamforming. .
- MIMO Multiple Input Multiple Output
- FIG. 18 is a diagram showing a configuration example of a base station device.
- the base station device 50 shown in FIG. 18 is assumed to perform wireless communication, and includes a wireless communication unit 51, a storage unit 52, a control unit 53, a calculation unit 54, a network communication unit 55, an antenna 56; Note that the configuration shown in FIG. 18 is a functional configuration and may differ from the hardware configuration. Also, the components in FIG. 18 may be further distributed or aggregated with other components. 18 may exist independently as a device separate from base station device 50, and the functions of base station device 50 may be implemented by a plurality of devices.
- the wireless communication unit 51 performs signal processing for wireless communication with another wireless communication device (for example, the communication terminal 11).
- the wireless communication section 51 operates under the control of the control section 53 .
- the radio communication unit 51 supports one or more radio access schemes.
- the wireless communication unit 51 supports both NR (New Radio) and LTE (Long Term Evolution).
- the wireless communication unit 51 may support W-CDMA (Wideband Code Division Multiple Access) and CDMA2000 (Code Division Multiple Access 2000) in addition to NR and LTE.
- the wireless communication unit 51 may support automatic retransmission technology such as HARQ (Hybrid Automatic Repeat reQuest).
- HARQ Hybrid Automatic Repeat reQuest
- the wireless communication unit 51 includes a transmission processing unit 510 and a reception processing unit 515.
- the wireless communication unit 51 may include multiple transmission processing units 510 and multiple reception processing units 515 .
- each component of the wireless communication unit 51 can be individually configured for each wireless access method.
- the transmission processing unit 510 and the reception processing unit 515 may be individually configured for LTE and NR.
- the antenna 56 may be one or more, and may be composed of a plurality of antenna elements (for example, a plurality of patch antennas).
- the wireless communication unit 51 may be configured to enable beamforming.
- the wireless communication unit 51 may be configured to enable polarization beamforming using vertical polarization (V polarization) and horizontal polarization (H polarization).
- the transmission processing unit 510 performs transmission processing of downlink control information and downlink data.
- the encoding unit 511 of the transmission processing unit 510 encodes the downlink control information and downlink data input from the control unit 53 using an encoding method such as block encoding, convolutional encoding, and turbo encoding. make a change.
- the encoding may be encoding by polar code or encoding by LDPC code (Low Density Parity Check Code).
- the modulation unit 512 of the transmission processing unit 510 modulates the encoded bits with a predetermined modulation method such as BPSK (Binary Phase Shift Keying), QPSK (Quadrature Phase shift Keying), 16QAM (Quadrature Amplitude Modulation), 64QAM, 256QAM, etc. do.
- BPSK Binary Phase Shift Keying
- QPSK Quadrature Phase shift Keying
- 16QAM Quadrature Amplitude Modulation
- 64QAM Quadrature Amplitude Modulation
- 64QAM 64QAM
- 256QAM 256QAM
- the signal points on the constellation of the modulation scheme do not necessarily have to be equidistant.
- the constellation may be a non-uniform constellation (NUC).
- the multiplexing unit 513 of the transmission processing unit 510 multiplexes the modulation symbols of each channel used for transmission and the downlink reference signal, and arranges them in predetermined resource elements.
- the transmission processing unit 510 performs various signal processing on the multiplexed signal.
- the radio transmission unit 514 of the transmission processing unit 510 performs conversion to the frequency domain by fast Fourier transform, addition of a guard interval (cyclic prefix), generation of a baseband digital signal, conversion to an analog signal, quadrature modulation, Perform processing such as up-conversion, removal of unnecessary frequency components, and power amplification.
- a signal generated by the radio transmission unit 514 is transmitted from the antenna 56 .
- the reception processing unit 515 processes uplink signals received via the antenna 56 .
- the radio reception unit 516 of the reception processing unit 515 performs down-conversion, removal of unnecessary frequency components, control of amplification level, quadrature demodulation, conversion to digital signals, guard interval (cyclic prefix ), and extraction of frequency domain signals by fast Fourier transform.
- the demultiplexing unit 517 of the reception processing unit 515 extracts uplink channels such as PUSCH (Physical Uplink Shared Channel), PUCCH (Physical Uplink Control Channel) and uplink reference from the signal processed by the radio reception unit 516. Separate the signals.
- PUSCH Physical Uplink Shared Channel
- PUCCH Physical Uplink Control Channel
- the demodulation unit 518 of the reception processing unit 515 demodulates the received signal using a modulation scheme such as BPSK or QPSK for the modulation symbols of the uplink channel.
- the modulation scheme used for demodulation may be 16QAM, 64QAM, or 256QAM.
- the signal points on the constellation do not necessarily have to be equidistant.
- the constellation may be a non-uniform constellation (NUC).
- the decoding unit 519 of the reception processing unit 515 performs decoding processing on the coded bits of the demodulated uplink channel.
- the decoded uplink data and uplink control information are output to the control section 53 .
- the antenna 56 mutually converts electric current and radio waves.
- the antenna 56 may be composed of one antenna element (for example, one patch antenna), or may be composed of a plurality of antenna elements (for example, a plurality of patch antennas). If the antenna 56 is composed of a plurality of antenna elements, the wireless communication section 51 may be configured to enable beamforming. For example, the wireless communication unit 51 may be configured to generate directional beams by controlling the directivity of wireless signals using a plurality of antenna elements.
- the antenna 56 may be a dual polarized antenna. When the antenna 56 is a dual polarized antenna, the wireless communication unit 51 may use vertical polarized waves (V polarized waves) and horizontal polarized waves (H polarized waves) for transmitting wireless signals. Then, the wireless communication unit 51 may control the directivity of the wireless signal transmitted using the vertically polarized wave and the horizontally polarized wave.
- the storage unit 52 serves as storage means for the base station device 50 and stores information necessary for processing of the base station device 50, processing results, and the like. For example, various programs for performing processing of the base station device 50 may be stored.
- the control unit 53 controls each unit of the base station device 50 .
- the control unit 53 externally acquires, via the wireless communication unit 51 or the network communication unit 55, information related to the DNN used by the logical entity, conditions for determining the range in charge of a series of DNN calculations, and the like. control necessary to
- the calculation unit 54 performs calculations necessary for the processing of the base station device 50 according to instructions from the control unit 53 . For example, part of the processing performed by the transmission processing unit 510 and the reception processing unit 515, such as high-load calculations, may be performed by the calculation unit 54. Further, for example, when the base station apparatus is in charge of calculation, the calculating section 54 may calculate the coverage area of the base station apparatus. Further, for example, when the base station apparatus 50 is a logical entity, the operation unit 54 may perform processing executed by the logical entity, for example, determination of a person in charge of calculation based on resources, determination of a range of responsibility, and the like.
- the network communication unit 55 performs signal processing for wired communication with another communication device (for example, the cloud system 12).
- the network communication unit 55 is connected to AMF (Access and Mobility Management Function) and UPF (User Plane Function) of the core network to exchange information and signaling.
- AMF Access and Mobility Management Function
- UPF User Plane Function
- the base station device may consist of multiple physical or logical devices.
- the base station device may be classified into a plurality of devices such as BBU (Baseband Unit) and RU (Radio Unit).
- BBU Baseband Unit
- RU Radio Unit
- a base station device may be interpreted as a collection of these devices, in other words, a base station system.
- the base station apparatus may be either one of the BBU and the RU, or both.
- the BBU and RU may be connected by a predetermined interface such as eCPRI (enhanced Common Public Radio Interface).
- RU may also be called RRU (Remote Radio Unit) or RD (Radio DoT).
- the RU may correspond to gNB-DU (gNB Distributed Unit), which will be described later.
- the BBU may correspond to gNB-CU (gNB Central Unit), which will be described later.
- the RU may be a unit integrally formed with the antenna.
- the antennas of the base station apparatus eg, antennas formed integrally with the RU
- the antennas of the base station may have, for example, 64 transmitting antenna ports and 64 receiving antenna ports.
- one or more antennas may be attached to the RU, and the antenna may be an antenna panel composed of one or more antenna elements.
- the RU is an antenna panel that includes two types of horizontally polarized antenna panel and vertically polarized antenna panel, or an antenna panel that includes two types of right-handed circularly polarized antenna panel and left-handed circularly polarized antenna panel.
- a panel may be mounted.
- the RU may also form and control independent beams for each antenna panel.
- a base station of a radio access network is sometimes called a RAN node, and a base station of an AN (Access Network) is sometimes called an AN node.
- RAN radio access network
- AN Access Network
- the RAN in LTE is sometimes called E-UTRAN (Enhanced Universal Terrestrial RAN).
- E-UTRAN Enhanced Universal Terrestrial RAN
- NG-RAN NG-RAN
- W-CDMA W-CDMA
- the LTE base station is also called eNodeB (Evolved Node B) or eNB, and at this time, E-UTRAN can be said to include one or more eNodeBs (eNB).
- eNB evolved NodeB
- NR base stations are also referred to as gNodeBs or gNBs, when the NG-RAN is said to include one or more gNBs.
- the E-UTRAN may include gNBs (en-gNBs) connected to a core network (EPC) in the LTE communication system (EPS).
- EPC LTE communication system
- the NG-RAN may include ng-eNBs connected to a core network 5GC in a 5G communication system (5GS).
- 5GS 5G communication system
- the base station is sometimes called 3GPP Access.
- the base station when the base station is a wireless access point (Access Point), the base station may be referred to as a non-3GPP access (Non-3GPP Access).
- the base station when the base station is a gNB, the base station may be a combination of the above-described gNB-CU and gNB-DU, or either gNB-CU or gNB-DU. may
- the gNB-CU hosts multiple higher layers (eg, RRC, SDAP, PDCP) of the access stratum for communication with the UE.
- the gNB-DU hosts multiple lower layers (eg, RLC, MAC, PHY) of the access stratum. That is, among messages or information such as RRC signaling, MAC Control Element (MAC Control Element), and DCI, RRC signaling (semi-static notification) is generated by gNB-CU, while MAC CE and DCI (dynamic notification) may be generated with gNB-DU.
- some configurations such as IE: cellGroupConfig are generated by gNB-DU, and the remaining configurations are generated by gNB-CU good too. These configurations may be sent and received on the F1 interface described below.
- the base station may be configured to be able to communicate with other base stations.
- the base stations may be connected via an X2 interface.
- the devices may be connected via an Xn interface.
- the devices may be connected by the F1 interface described above. Messages or information such as RRC signaling, MAC CE, DCI may be transmitted between multiple base stations, for example via the X2 interface, the Xn interface, or the F1 interface.
- a cell provided by a base station is sometimes called a serving cell.
- the concept of serving cell includes PCell (Primary Cell) and SCell (Secondary Cell).
- PCell Primary Cell
- SCell Secondary Cell
- dual connectivity When dual connectivity is configured in the UE, the PCell provided by the MN (Master Node) and zero or more SCells may be called a Master Cell Group.
- dual connectivity include E-UTRA-E-UTRA Dual Connectivity, E-UTRA-NR Dual Connectivity (ENDC), E-UTRA-NR Dual Connectivity with 5GC, NR-E-UTRA Dual Connectivity (NEDC), NR- NR Dual Connectivity can be mentioned.
- the serving cell may include a PSCell (Primary Secondary Cell or Primary SCG Cell).
- PSCell Primary Secondary Cell or Primary SCG Cell
- a PSCell provided by an SN (Secondary Node) and zero or more SCells may be called an SCG (Secondary Cell Group).
- PUCCH Physical Uplink control channel
- SCell Secondary Cell Group
- a radio link failure Radio Link Failure
- SCell SCell
- One cell may be associated with one downlink component carrier and one uplink component carrier.
- the system bandwidth corresponding to one cell may be divided into a plurality of BWPs (Bandwidth Parts).
- one or more BWPs may be set in the UE, and one BWP may be used by the UE as an active BWP.
- the radio resources for example, frequency band, numerology (subcarrier spacing), slot format (Slot configuration)
- the radio resources for example, frequency band, numerology (subcarrier spacing), slot format (Slot configuration) that can be used by the UE may differ for each cell, each component carrier, or each BWP.
- the communication terminal 11 may move by being installed in a mobile object, or may be the mobile object itself.
- the communication terminal 11 is a vehicle that moves on roads such as automobiles, buses, trucks, and motorcycles, a vehicle that moves on rails installed on a track such as a train, or a vehicle mounted on the vehicle. It may be a wireless communication device.
- the mobile object may be a mobile terminal, or a mobile object that moves on land (ground in a narrow sense), underground, on water, or in water. Further, the mobile object may be a mobile object such as a drone, a helicopter, or the like that moves in the atmosphere, or a mobile object that moves outside the atmosphere, such as an artificial satellite.
- the communication terminal 11 can be used for any main purpose as long as it is a device that has an information processing function and a communication function and is capable of executing the processing of the present disclosure.
- it may be a device such as a business camera equipped with an information processing function and a communication function, or a communication device such as an FPU (Field Pickup Unit).
- the communication terminal 11 may be an M2M (Machine to Machine) device or an IoT (Internet of Things) device.
- the communication terminal 11 may be capable of NOMA communication with the base station. Also, the communication terminal 11 may be able to use an automatic retransmission technique such as HARQ when communicating with a base station. The communication terminal 11 may be capable of sidelink communication with another communication terminal 11 . The communication terminal 11 may be able to use an automatic retransmission technique such as HARQ even when performing sidelink communication. The communication terminal 11 may be capable of NOMA communication also in communication (side link) with another communication terminal 11 . Also, the communication terminal 11 may be capable of LPWA communication with other communication devices (eg, base station, other communication terminal 11). Also, the wireless communication used by the communication terminal 11 may be wireless communication using millimeter waves. Wireless communication (including sidelink communication) used by the communication terminal 11 may be wireless communication using radio waves, or wireless communication using infrared rays or visible light (optical wireless). .
- the communication terminal 11 may be a communication device installed on a mobile object, or a communication device with mobility.
- the mobile body on which the communication terminal 11 is installed may be a vehicle such as an automobile, a bus, a truck, or a motorcycle that moves on a road, or a vehicle such as a train that moves on a rail installed on a track. good too.
- the place where the moving body moves is not particularly limited. Therefore, the mobile body may be a mobile body that moves on land (narrowly defined ground), underground, water, or water. Further, the mobile body may be a mobile body such as a drone, a helicopter, or the like that moves in the atmosphere, or a mobile body that moves outside the atmosphere, such as an artificial satellite.
- the communication terminal 11 may communicate with multiple base stations or multiple cells at the same time. For example, when one base station supports a communication area through multiple cells (for example, pCell, sCell), carrier aggregation (CA: Carrier Aggregation) technology and dual connectivity (DC: Dual Connectivity) technology, Multi-connectivity (MC) technology enables communication between the base station and the communication terminal 11 by bundling the plurality of cells.
- communication terminal 11 and a plurality of base stations can communicate with each other by CoMP (Coordinated Multi-Point Transmission and Reception) technology via cells of different base stations.
- CoMP Coordinatd Multi-Point Transmission and Reception
- FIG. 19 is a diagram showing a configuration example of the communication terminal 11. As shown in FIG. FIG. 19 shows a configuration example for wireless communication, and communication terminal 11 includes wireless communication section 111 , storage section 112 , control section 113 , calculation section 114 , and antenna 115 . Note that the configuration shown in FIG. 19 is a functional configuration and may differ from the hardware configuration. Also, the functions of the communication terminal 11 may be distributed and implemented in a plurality of physically separated components.
- the wireless communication unit 111 performs signal processing for wireless communication with other wireless communication devices (eg, base station, relay station, wireless communication node 131, donor node 132, other communication terminal 11, etc.).
- Wireless communication section 111 operates under the control of control section 113 .
- Wireless communication section 111 includes transmission processing section 1110 and reception processing section 1115 .
- Components related to wireless communication of communication terminal 11 may be the same as corresponding components related to wireless communication of base station apparatus 50 . That is, the configuration of the radio communication unit 111 and its internal components, and the antenna 115 may be the same as the radio communication unit 51 of the base station apparatus 50, its internal components, and the antenna 56, respectively.
- the wireless communication unit 111 may be configured to enable beamforming, like the wireless communication unit 51 of the base station device 50 .
- the storage unit 112 serves as storage means for the communication terminal 11 and stores information necessary for processing of the communication terminal 11, processing results, and the like. For example, various programs for processing the communication terminal 11 may be stored.
- the control unit 113 controls each unit of the communication terminal 11. For example, the control unit 113 externally acquires, via the wireless communication unit 111, information related to the DNN used by a logical entity or the like, conditions for determining a range in charge of a series of DNN calculations, and the like. control.
- the calculation unit 114 performs calculations necessary for the processing of the communication terminal 11 according to instructions from the control unit 113 . For example, part of the processing performed by the transmission processing unit 1110 and the reception processing unit 1115, such as high-load calculations, may be performed by the calculation unit 114. Further, for example, calculations required for the ML application executed by the communication terminal 11, such as calculation of DNN, are performed.
- FIG. 20 is a diagram showing a configuration example of a 5GS (5G System) network architecture including the core network 133.
- 5GS is composed of communication terminal 11 (denoted as UE in FIG. 20), RAN 134, and core network 133.
- the RAN 134 provides a network function (NF) like the radio communication node 131 and the donor node 132 in FIG.
- the core network 133 in 5GS is called NGC (Next Generation Core), 5GC (5G Core), and the like.
- the function group of the control plane of the core network 133 is AMF (Access and Mobility Management Function) 601, NEF (Network Exposure Function) 602, NRF (Network Repository Function) 603, NSSF (Network Slice Selection Function) 604, PCF (Policy Control Function) 605, SMF (Session Management Function) 606, UDM (Unified Data Management) 607, AF (Application Function) 608, and AUSF (Authentication Server Function) 609 , and UCMF (UE radio Capability Management Function) 610 .
- AMF Access and Mobility Management Function
- NEF Network Exposure Function
- NRF Network Repository Function
- NSSF Network Slice Selection Function
- PCF Policy Control Function
- SMF Session Management Function
- UDM Unified Data Management
- AF Application Function
- AUSF Authentication Server Function
- UCMF UE radio Capability Management Function
- the UDM 607 holds, manages, and processes subscriber information.
- the execution unit for holding and managing subscriber information is also called UDR (Unified Data Repository) and may be separated from FE (Front End), which is the execution unit for processing subscriber information.
- the AMF 601 performs mobility management.
- the SMF 606 performs session management.
- UCMF 610 holds UE radio capability information (UE Radio Capability Information) corresponding to all UE radio capability IDs (UE Radio Capability IDs) in PLMN (Public Land Mobile Network).
- UCMF 610 is responsible for assigning each PLMN-assigned UE Radio Capability ID.
- FIG. 20 shows the service-based interface of NF.
- Namf is a service-based interface provided by AMF 601
- Nsmf is a service-based interface provided by SMF 606
- Nnef is a service-based interface provided by NEF 602
- Npcf is a service-based interface provided by PCF 605.
- Nudm is a service-based interface provided by UDM 607
- Naf is a service-based interface provided by AF 608
- Nnrf is a service-based interface provided by NRF 603
- Nnssf is a service-based interface provided by NSSF 604.
- interface, and Nausf is a service-based interface provided by AUSF609.
- Each NF exchanges information with other NFs via each service-based interface.
- the application side can grasp the information of the communication layer in detail, and the NF can be controlled from the external application.
- a UPF User Plane Function
- DN Data Network
- MNO Mobile Network Operator
- the RAN 134 establishes communication connections with the core network 133, communication terminals 11, and the like. It should be noted that a communication connection with another communication network (not shown), for example, an access network (AN), may also be made.
- the RAN 134 includes base stations called gNBs or ng-eNBs. RAN is sometimes called NG (Next Generation)-RAN.
- Information is mutually exchanged between the UE 10 and the AMF 601 via the reference point N1.
- Information is mutually exchanged between the RAN 134 and the AMF 601 via the reference point N2.
- Information is mutually exchanged between the SMF 606 and the UPF 630 via the reference point N4.
- the communication quality may be indicated by, for example, delay time in transmission and reception, data rate, channel occupancy ratio (Channel Occupancy Ratio), and the like.
- the channel occupancy rate may be indicated by CBR (Channel Busy Ratio), resource usage rate, or congestion degree.
- CBR Channel Busy Ratio
- resource usage rate or congestion degree.
- CBR may be indicated as a ratio of used radio resources to total available radio resources.
- the degree of congestion may be indicated by the ratio of RRSI (Received Signal Strength Indicator), which is the total received power in the band, to RSRP (Reference Signal Received Power), which is the received strength of the reference signal.
- RRSI Receiveived Signal Strength Indicator
- RSRP Reference Signal Received Power
- the degree of congestion may be indicated by the reciprocal of RSRQ (Reference Signal Received Quality), which is the received quality of the reference signal.
- the ML application performs processing based on the results of a series of DNN calculations.
- recent research has revealed that even if the ML application is processed based on the intermediate calculation results of the series of calculations, there may be cases where the processing result of the ML application is less affected. In this way, causing a breakout in the middle of the calculation without performing the entire series of DNN calculations is called early-exiting (also called early termination).
- the accuracy of the ML application can be reduced, but since the calculation is not performed until the end, the time required to complete the processing of the ML application can be reduced.
- the information processing system of the second embodiment performs early-exiting depending on the situation, and returns the intermediate calculation results of a series of DNN calculations to the communication terminal 11 . This reduces the time required to complete the processing of the ML application.
- the logic entity determines who is responsible for calculation, but in the second embodiment where early-exiting is performed, who is responsible for calculation may be determined in advance. Also, in the first embodiment, the logical entity could dynamically change the person in charge of calculation according to the situation, but in the second embodiment, the person in charge of calculation may be fixed and cannot be changed. It may also be determined whether to perform dynamic change of computational responsibility or early-exiting when it is determined that the requirements of the ML application, such as delay time, cannot be met.
- one of the persons in charge of calculation transmits the calculation result in the middle of a series of calculations of the DNN to the communication terminal 11.
- a certain calculation is in charge of the 3rd and 4th layers of the DNN, it is possible to perform up to the calculation of the 3rd layer and transmit the calculation result of the 3rd layer to the communication terminal 11 .
- the logical entity determines the scope of responsibility for calculation so as not to increase the delay due to the large size of data to be transmitted. Therefore, if the person in charge of the calculation ends the calculation in the range they are in charge of, the data size of the calculation result will be large depending on the position of the end in the middle, the communication will take time, and the delay will increase even if early-exiting is performed. Such a situation can also occur. In addition, if the communication bandwidth between calculation members is large, the logical entity can determine that there is no problem even if the data size of the calculation result is large.
- the communication band to be considered is different.
- the communication terminal 11 and the communication node A, which is in charge of the first calculation are wirelessly connected, and the communication node A and the communication node B, which is in charge of the second calculation, are connected by wire. could be.
- the data size of the calculation result transmitted from communication node A to communication node B may be too large for wireless communication.
- the accuracy of the ML application will be low as a result of calculating the coverage area of the communication node A. In this way, it may not be suitable for early-exiting to have each person in charge of calculations execute all the calculations included in the range they are in charge of.
- FIG. 21 is a diagram showing an example of a calculation range when Early-exiting is performed.
- a calculation flow 71 indicated by an arrow in FIG. 21 shows the flow of calculation (calculation range) when the calculation of the range in charge is terminated halfway.
- Calculation flow 72 shows the flow of calculation when the calculation of the assigned range is terminated halfway.
- a calculation flow 73 shows the flow of calculation when the calculation of the assigned range is not terminated halfway.
- the convolution processing after the bending of calculation flow 71 and calculation flow 72 is a calculation added separately from the calculation included in the normal range of responsibility.
- the calculation for outputting the inference result is performed from the calculation flow that was terminated halfway.
- calculations are performed to transmit intermediate calculation results in a series of deep neural network calculations.
- 5x5 convolution processing Conv5x5
- Conv3x3 3x3 convolution processing
- Which of the calculation flow 71 and the calculation flow 72 is to be executed may be selected according to the data size of the calculation result, the required time, and the like.
- the calculation flow 72 may be selected.
- the person in charge of calculation executes at least part of the calculation in the range in charge, but the entire contents of the calculation executed by the person in charge of calculation are It can vary depending on the situation. Note that the Early-exiting determination may be performed at any position between the start and end of the calculation of the assigned range.
- the calculation may be performed before the determination, the calculation may not be performed before the determination, the calculation may be performed after the determination, or the calculation may be performed after the determination. It doesn't have to be. Moreover, the presence or absence of these processes may differ based on the result of the determination and other information shown in this specification.
- Various conditions are conceivable for early-exiting judgment conditions, and may be determined as appropriate.
- a plurality of conditions may be provided, and determination conditions based on a plurality of parameters may be used.
- the condition regarding the calculation result may be used as the Early-exiting determination condition.
- determination may be made based on the final calculation result of the calculation range itself or the calculation result in each layer. Further, determination may be made after further applying an activation function such as a softmax function to the calculation result.
- the cross-entropy value may be used for determination. For example, when the output value of the softmax function is equal to or greater than a predetermined threshold, the DNN processing may be terminated in that layer and early-exiting may be performed.
- a condition regarding the person in charge of calculation may be set as an Early-exiting judgment condition.
- the calculation spare capacity of the person in charge of calculation may be used as the determination condition.
- conditions related to communication may be provided as early-exiting determination conditions.
- the communication quality of the communication network such as RARP, RARQ, RSSI, and Link failure may be used as the determination condition.
- the communication link to the next computational worker is of poor quality, packet errors can add to the delay. Therefore, a determination condition may be provided such that the worse the communication quality is, the more early-exiting is performed.
- the data amount of the calculation result to be transmitted is small, it may be possible to transmit even if the communication quality is poor. Therefore, the data amount of the calculation result to be transmitted may also be included in the determination conditions.
- the path information of the communication network may be used as the determination condition.
- the determination condition may be based on traffic such as information on the amount of traffic flowing on the route or information on the amount of traffic being processed at each node. Note that the traffic may be represented by the usage rate of communication resources or the like.
- a condition regarding movement (mobility) of the communication terminal 11 may be provided as an Early-exiting determination condition.
- movement speed, moving direction, and information about link switching and handover caused by movement may be used as determination conditions.
- request information including the time (delay) until the DNN calculation result is returned to the communication terminal 11, the certainty (accuracy) of the DNN calculation result returned by Early-exiting, etc. is obtained from the ML application. If so, whether or not these requirements are met may be determined based on computational capacity, communication quality, past performance, and the like.
- the communication terminal 11 may be used as a condition for judging whether or not an instruction to execute Early-exiting has been received. For example, when the time (delay) until the DNN calculation result is returned to the communication terminal 11 is to be kept within a predetermined time, the communication terminal 11 attaches a time stamp in the application layer and counts the time until the reply is received. Then, when the time has passed the upper limit, an instruction to execute Early-exiting is sent. When the communication terminal 11 recognizes all persons in charge of calculation, the communication terminal 11 may transmit execution instructions to all persons in charge of calculation. In cases such as when the communication terminal 11 does not recognize each person in charge of calculation, the execution instruction may be relayed to each person in charge of calculation in order from the communication terminal 11 .
- the person in charge of calculation who receives the Early-exiting notification during the calculation should perform the Early-exiting.
- each person in charge of calculation subtracts the time required for the calculation from the allowable time set in advance and notifies the next person in charge of calculation, and the calculation that has used up the allowable time. It is also conceivable that the person in charge transmits the calculation result to the communication terminal 11 . Also, if the time required to calculate the range of responsibility is shorter than the allowable time, the calculation of the range of responsibility may be terminated halfway.
- the values of parameters for determining whether to perform Early-exiting such as the upper limit value of the required time and the delay margin value, may be determined by the logical entity or by the ML application.
- Early-exiting criteria may be changed according to the time period, the type of ML application, and other ML applications that are executed in parallel. For example, in the case of ML applications that require accuracy, the criteria for early-exiting are stricter than the standard, and in the case of ML applications that require less time, the criteria for early-exiting are stricter than the standard. You can take it easy.
- the information used for the determination condition may be notified from the application layer to the communication layer via NEF (Network Exposure Function) or the like.
- NEF Network Exposure Function
- FIG. 22 is a schematic sequence diagram showing a first example of the flow of processing regarding early-exiting.
- the calculations other than the communication terminal are collectively represented as one.
- the logical entity sends the Early-exiting judgment condition to the calculation manager (T301). If the person in charge of calculation is not fixed, the determination condition may be sent in advance to the entity that can be in charge of calculation, or the fact that the person in charge of calculation is changed is notified when the person in charge of calculation is changed. You may transmit the said judgment conditions with.
- Each calculator receives and sets the judgment conditions from the logical entity (T302).
- the communication terminal 11 executes the ML application (T303), and transmits information necessary for DNN calculation to the next person in charge of calculation (T304).
- the communication terminal 11 may transmit information used for determination. For example, the communication terminal 11 transmits the upper limit value (permissible time) of the time required for the calculation result to be returned to the communication terminal 11, the accuracy of the calculation result, etc. to each person in charge of calculation. may be used as the value of the parameter of
- the next person in charge of calculation receives information necessary for DNN calculation (T305), collects information necessary for Early-exiting judgment, and uses the collected information to make Early-exiting judgment (T306). If the information may be collected in advance, it may be collected in advance. DNN calculation is executed based on the determination result (T307). As described above, the DNN calculation range may differ between when Early-exiting is executed and when it is not executed.
- the DNN calculation is started after the Early-exiting determination is performed here, the Early-exiting determination may be performed during the DNN calculation.
- the person in charge of calculation sends the calculation result to the partner corresponding to the judgment result (T308). If early-exiting is not executed and there is a next person in charge of calculation, the calculation result is sent to the next person in charge of calculation. In this case, the next person in charge of calculation performs the processing from T305 to T308. Note that the arrow from T308 to T305 in FIG. 22 does not indicate that the same person in charge of calculation performs the processing from T305 to T308 again, but that another person in charge of calculation performs the processing from T305 to T308. do.
- the person in charge of calculation When there is no next person in charge of calculation, in other words, when the calculation is completed up to the output layer of the DNN, or when Early-exiting is executed, the person in charge of calculation sends the calculation result to the communication terminal 11 . In addition, you may transmit necessary information other than a calculation result. For example, when the person in charge of calculation receives information used for determination from the communication terminal 11, the information is also transmitted to the next person in charge of calculation. Information used to determine early-exiting may be sent to the computational staff in order in a bucket-brigade fashion.
- the communication terminal 11 receives the DNN calculation result (T309) and executes the ML application process based on the calculation result (T310).
- the calculation result is transmitted to the communication terminal 11 without waiting for the end of the series of calculations of the DNN, so that the processing of the ML application is prevented from being delayed. can be done.
- the early-exiting determination can also be said to be a determination as to whether or not to transmit the calculation result to the communication terminal 11 .
- FIG. 23 is a schematic sequence diagram showing a second example of the processing flow regarding early-exiting.
- the case where the person in charge of calculation executes all the calculations in the area of responsibility is also included in the Early-exiting, but in the example of FIG. 23, the person in charge of calculation executes all the calculations in the area of responsibility It means that the person in charge of calculation ends the calculation in the middle of the calculation in the range in charge.
- the communication terminal 11 can also end the calculation of the range of responsibility in the middle. Therefore, the example of FIG. 23 shows an example in which the communication terminal 11 is in charge of the first calculation.
- the example of FIG. 23 shows a case where, in addition to the communication terminal 11, there are a second person in charge of calculation, a third person in charge of calculation, and a person in charge of final calculation (fourth person in charge of calculation).
- Each person in charge of calculation receives and sets the judgment conditions from the logical entity (T302). After that, the communication terminal 11 executes the ML application (T303). The communication terminal 11 collects information necessary for Early-exiting determination, and uses the collected information to perform Early-exiting determination (T306). Then, the DNN is calculated based on the determination result (T307).
- the range of the calculation differs depending on whether the Early-exiting is executed or not, and the calculation may not be executed depending on the judgment result or timing of the judgment. In other words, the processing of T307 does not necessarily have to be performed. This is the same for any embodiment in this specification.
- the communication terminal 11 sends the calculation result to the second calculation manager, who is in charge of the next calculation, regardless of whether the early-exiting is executed (T308).
- information indicating the position may be transmitted to the next person in charge of calculation so that the next person in charge of calculation recognizes the position where the calculation is to be started.
- the information may be, for example, information indicating the last layer in the range in charge of the communication terminal 11, information indicating the first layer in the range in charge of the next calculation, or information indicating the node that outputs the calculation result. Alternatively, it may be information indicating the node to which the calculation result should be input.
- next person in charge of calculation knows the predetermined position to end halfway, the calculation will be performed up to that position, and if the calculation result is sent, the information for the next person in charge of calculation to recognize the position where the calculation will start. need not be sent.
- the second person in charge of calculation is in charge of receiving information related to the calculation result of the communication terminal 11, collecting information necessary for Early-exiting judgment, making an Early-exiting judgment, executing DNN calculation based on the judgment result, and being in charge of the next calculation.
- Send to the third person in charge of calculation (T305 to T308).
- the third person in charge of calculation similarly performs the processing from T305 to T308, and the calculation result of the third person in charge of calculation is transmitted to the final person in charge of calculation.
- the person in charge of final calculation similarly performs the processing from T305 to T307.
- Send (T311) the communication terminal 11 receives the DNN calculation result (T309) and executes the ML application process based on the calculation result (T310). By doing so, it is possible to reduce the amount of calculation in charge of calculation, which may cause a delay in the processing, and it is possible to reduce the waiting time of the communication terminal 11, though not as much as in the example of FIG.
- the accuracy of the calculation results received by the communication terminal 11 is higher than in the case where the second calculation charge sends, for example, the result of the intermediate termination of the DNN calculation. obtain.
- FIG. 24 is a schematic sequence diagram showing a third example of the processing flow regarding early-exiting.
- the example in FIG. 24 is different from the example in FIG. 22 in that the logical entity makes early-exiting determination. Note that the entity that decides who is in charge of calculation may be different from the entity that decides the early-exiting determination.
- the logic entity transmits the Early-exiting judgment condition to the calculation staff (T301), and each calculation staff receives and sets the judgment condition from the logic entity (T302). is, in the example of FIG. 24, whether or not an instruction to execute Early-exiting has been received from the logical entity.
- the communication terminal 11 executes the ML application (T303), it notifies the logical entity of requesting DNN calculation (T312).
- the logical entity collects information necessary for Early-exiting determination, and uses the collected information to perform Early-exiting determination (T306).
- the determination of logical entities is repeated periodically. For example, a logical entity may periodically check communication resources such as computing power and communication quality, and make decisions based on the communication resources. Alternatively, the logical entity may measure the elapsed time after the notification from the communication terminal 11 and check whether the upper limit notified from the communication terminal 11 is exceeded.
- the communication terminal 11 transmits information necessary for calculating the DNN to the first calculation person in charge of the next calculation, as in the example of FIG. 22 (T304).
- the first person in charge of calculation performs the processing from T305 to T308, as in the example of FIG.
- the calculation of the scope of responsibility is executed and the calculation result is sent to the second calculation manager.
- the judgment condition is satisfied and the logical entity decides to execute Early-exiting and transmits to all the computational agents (T314).
- the logical entity since the logical entity does not recognize which person in charge of calculation is in charge of calculation, it is assumed that the information is transmitted to all the persons in charge of calculation. If the logical entity is aware of which computation is in progress by, for example, each computation notifying the logical entity of the end of computation, the information may be sent only to the next computation.
- the second calculator receives the execution notification of Early-exiting before receiving the calculation result of the first calculator (T315). After that, the calculation result of the first person in charge of calculation is transmitted, and the second person in charge of calculation performs the processing from T305 to T308. Therefore, the second person in charge of calculation executes the calculation for Early-exiting and transmits the executed calculation result to the communication terminal 11 .
- the communication terminal 11 executes the processes of T309 and T310 as in the example of FIG. In this way, the logic entity may determine the execution of Early-exiting, and the execution notification from the logic entity may be used as the Early-exiting determination condition for each calculation. Note that when an Early-exiting execution notification is received during DNN calculation, the person in charge of calculation may perform Early-exiting if possible.
- the processing for not performing Early-exiting may be continued in parallel with the processing for Early-exiting. That is, the person in charge of calculation who has executed the early-exiting may transmit the intermediate calculation result to the communication terminal 11 and may also transmit the result of calculation to the next person in charge of calculation. In this way, the communication terminal 11 can receive the final calculation result of the DNN after receiving the calculation result up to the middle of the DNN.
- the processing result of the ML application can be obtained early from the intermediate result of the DNN calculation, and then whether the processing of the ML application was correct is confirmed based on the final calculation result of the DNN received later. be able to.
- the person in charge of calculation may request the next person in charge of calculation to perform the remaining calculations, as shown in the example of FIG. After executing all the calculations in the range of responsibility, the calculation results may be sent to the next person in charge of calculation.
- the communication terminal 11 may notify whether or not to perform multi-feedback.
- the calculation responsible for the Early-exiting notifies the logical entity of the Early-exiting, and upon receipt of the notification, the logical entity decides whether or not to perform multi-feedback depending on the situation.
- the multi-feedback execution determination condition may also be determined according to the calculation spare capacity of the person in charge of calculation, the communication resource, the required specifications of the ML application, and the like.
- the early-exiting calculation staff may notify the next calculation staff that the early-exiting has been performed, and the next calculation staff may not perform the early-exiting. Conversely, even if Early-exiting has already been performed, the next person in charge of calculation may also perform Early-exiting. Therefore, the communication terminal 11 may receive a plurality of early-exiting calculation results and the final DNN calculation result.
- FIG. 25 is a conceptual diagram explaining multi-feedback.
- the communication terminal and the second to fourth calculators are in charge of DNN calculations.
- the second calculator performs the early-exiting, and the feedback FD1 indicated by an arrow indicates that the calculation result by the early-exiting of the second calculator is returned to the communication terminal 11. It is also, the second calculator sends the calculation results to the third calculator for multi-feedback.
- the third calculator is also performing Early-exiting, and the feedback FD2 indicates that the calculation result by the Early-exiting of the third calculator is returned to the communication terminal 11 .
- the third calculator also sends the calculation results to the fourth calculator for multi-feedback.
- FIG. 25 the communication terminal and the second to fourth calculators are in charge of DNN calculations.
- the second calculator performs the early-exiting, and the feedback FD1 indicated by an arrow indicates that the calculation result by the early-exiting of the second calculator is returned to the communication terminal 11. It is Also, the second calculator sends the calculation results to
- the feedback FD3 indicates that the fourth calculator performs calculations up to the end of a series of DNN calculations and sends the final calculation results to the communication terminal 11.
- the communication terminal 11 may receive a plurality of early-exiting calculation results (FD1 and FD2) and the final DNN calculation result (FD3).
- the calculation of DNN after Early-exiting is performed is not performed by each person in charge of calculation, but a specific entity performs the calculation intensively.
- a specific entity performs the calculation intensively.
- the second and third calculators may not perform calculations, and the fourth communication manager may collectively perform the remaining calculations.
- the early-exiting calculator may request the logical entity to change the calculator. For example, if early-exiting is done due to a problem with your own calculation capacity, you may request that you be removed from the calculation. For example, if early-exiting is performed due to a problem in communication quality with the next person in charge of calculation, it may be requested to change the next person in charge of calculation.
- the communication terminal 11 may notify the logic entity and each calculation manager whether or not Early-exiting can be executed. For example, when the communication terminal 11 has sent a notification to reject the execution of Early-exiting, each calculation operator may not perform Early-exiting without complying with the Early-exiting determination conditions. In addition, a notification regarding whether or not multi-feedback can be executed may be made. Also, when allowing multiple feedbacks, the upper limit of the number of feedbacks shown in FIG. 25 may be specified.
- each entity in the second embodiment may be the same as in the first embodiment, so it will be omitted.
- the calculation range when Early-exiting is not performed and the calculation range when Early-exiting is performed are stored in the storage unit of each device (storage unit 52 in FIG. 18 or storage unit 112 in FIG. 19), Switch the calculation range used by the control unit of each device (control unit 53 in FIG. 18 or control unit 113 in FIG. 19), and the calculation unit of each device (calculation unit 54 in FIG. 18 or calculation unit 114 in FIG. 19) You just have to do the calculations.
- early-exiting is executed when multiple entities perform distributed learning in which a series of DNN calculations are shared.
- early-exiting it is possible to suppress the time for feedback of the DNN calculation result to the communication terminal 11 within a predetermined time even in distributed learning using communication nodes that are prone to communication delays.
- the computational spare capacity of the person in charge of calculation is less than expected, the calculation within the range in charge can be terminated halfway and the rest of the calculation can be entrusted to the next person in charge of calculation. In this way, the time required for a series of DNN calculations can also be reduced.
- the processing procedure described in the present disclosure may be regarded as a method having a series of these procedures. Alternatively, it may be regarded as a program for causing a computer to execute the series of procedures, or a recording medium for storing the program. Also, the logical entities and computational processes described above are executed by a processor, such as a CPU of a computer. Also, the type of recording medium is not particularly limited since it does not affect the embodiments of the present disclosure.
- each component shown in FIGS. 18 to 20 shown in the present disclosure may be realized by software or by hardware.
- each component may be a software module implemented by software such as a microprogram, and each component may be implemented by a processor executing the software module.
- each component may be realized by a circuit block on a semiconductor chip (die), for example, an integrated circuit such as ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array).
- ASIC Application Specific Integrated Circuit
- FPGA Field Programmable Gate Array
- the number of components and the number of hardware that implements the components need not match.
- a single processor or circuit may implement multiple components.
- one component may be implemented by multiple processors or circuits.
- processors described in the present disclosure are not limited.
- it may be a CPU, MPU (Micro Processing Unit), GPU (Graphics Processing Unit), or the like.
- DNN digital neural network
- SNN spike neural network
- a spike signal may be used to convey any of the information described above.
- causal relationships may be extracted from the time-series data, and any of the above-described judgments may be made based on the information.
- Components for storing data may be implemented by devices capable of reading and writing data, and the devices may be appropriately selected. good.
- the devices may be a DRAM, SRAM, flash memory, hard disk, or the like.
- this disclosure can also take the following configurations.
- An information processing device in charge of a part of a series of calculations of a deep neural network Determining whether to transmit intermediate calculation results in a series of calculations of the deep neural network to the first communication device, If it is not determined to transmit to the first communication device, the result of at least part of the calculations included in the first range of the series of calculations is transmitted to the second range following the first range. to the second communication device responsible for the calculation of When it is determined to transmit to the first communication device, performing a calculation for transmitting intermediate calculation results in a series of calculations of the deep neural network to the first communication device, transmitting the result of the calculation to the first communication device; Information processing equipment.
- [10] determining to which position of the calculations included in the first range is to be executed when it is determined that intermediate calculation results in a series of calculations of the deep neural network are to be transmitted to the first communication device; performing the calculations included in the first range up to the determined position, and transmitting the result of the performed calculations and information indicating the determined position to the second communication device;
- the information processing device according to any one of [1] to [7].
- [11] information on the computational spare capacity of a device in charge of at least part of a series of calculations of said deep neural network; and information on a traffic volume in a device in charge of at least part of said series of calculations of said deep neural network making the determination based on at least one of The information processing device according to any one of [1] to [10].
- [12] making the determination based on at least one of communication quality with the first communication device and mobility information of the first communication device;
- the information processing device according to any one of [1] to [11].
- [13] based on at least one of information instructing to terminate the series of calculations halfway and information instructing to transmit the result of an intermediate calculation in the series of calculations to the first communication device. make a decision,
- the information processing device according to any one of [1] to [12].
- [14] The information processing device according to any one of [1] to [13], wherein request information is acquired from the first communication device and the determination is made based on the request information.
- [17] further transmitting information indicating whether or not the device in charge of calculation of the second range may end the calculations included in the second range halfway to the device in charge of calculation of the second range;
- the information processing device according to [16].
- An information processing method comprising: [19] a step of determining whether or not to terminate the calculations included in the first range in the series of calculations of the deep neural network; When it is determined that the calculations included in the first range are to be terminated halfway, a part of the calculations included in the first range is performed, and the results of the performed calculations and the results of the performed calculations are returned to the series of calculations.
- An information processing method comprising: [20] At least a first information processing device and a second information processing device that are in charge of part of a series of calculations of the deep neural network, The first information processing device is Determining whether to transmit intermediate calculation results in a series of calculations of the deep neural network to the first communication device, If it is not determined to transmit to the first communication device, the results of at least part of the calculations included in the first range in the series of calculations are transmitted to the second information processing device. , If it is determined to transmit to the first communication device, the calculation for transmission to the first communication device is performed using intermediate calculation results in a series of calculations of the deep neural network.
- a first information processing device that performs at least a first range of calculations in a series of deep neural network calculations, and a second range of calculations that follows the first range in a series of deep neural network calculations.
- the first information processing device is Determining whether to end the calculations included in the first range halfway; When it is determined that the calculations included in the first range are to be terminated halfway, a part of the calculations included in the first range is performed, and the results of the performed calculations and the results of the performed calculations are returned to the series of calculations. and information indicating which position in the calculation of, to the second information processing device, The second information processing device is When the information is received, the continuation of the calculations performed by the first information processing device in the series of calculations is performed based on the results of the calculations performed by the first information processing device.
- Information processing system [22] The information processing system according to [20], further comprising a third information processing device that determines the first range.
- [23] further comprising a third information processing device that determines whether or not to terminate the calculations included in the first range, The third information processing device transmits to the first information processing device an instruction to prematurely terminate calculations included in the first range; The first information processing device, when receiving the instruction, determines to end the calculations included in the first range halfway.
- the information processing system according to [21].
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Neurology (AREA)
- Mobile Radio Communication Systems (AREA)
- Telephonic Communication Services (AREA)
Abstract
Description
以下に、本開示の実施形態について図面に基づいて詳細に説明する。図1は、本開示の第1実施形態に係る情報処理システムの構成例を示す図である。第1実施形態に係る情報処理システム1は、通信端末11と、クラウドシステム(Cloud)12と、通信ネットワーク13と、を備える。なお、符号は、図1に示された11Aおよび11Bのように、同じ種類の個体に対し同一の数字を振ることとし、各個体の区別はアルファベットによって行うものとする。また、本説明において、特に個体を区別する必要がない場合は、符号のアルファベットは記載しない。
第1実施形態において説明した通り、MLアプリケーションは、DNNの一連の計算の結果に基づいて処理を行う。しかし、当該一連の計算の途中の計算結果に基づいてMLアプリケーションの処理を行ったとしても、MLアプリケーションの処理結果に影響が少ない場合もあり得ることが、近年の研究により判明している。このように、DNNの一連の計算全てを行わずに計算途中でブレイクアウトさせることは、Early-exitingと称される(Early terminationとも称される)。
算を実行しなくても良い。また、それらの処理の有無は判定の結果や本明細書に示すその他の情報に基づいて異なっても良い。
[1]
ディープニューラルネットワークの一連の計算のうちの一部の計算を担当する情報処理装置であって、
前記ディープニューラルネットワークの一連の計算における途中の計算の結果を、第1の通信装置に送信するか否かの判定を行い、
前記第1の通信装置に送信すると判定されなかった場合には、前記一連の計算のうちの第1範囲に含まれる計算の少なくとも一部の計算の結果を、前記第1範囲に続く第2範囲の計算担当である第2の通信装置に送信し、
前記第1の通信装置に送信すると判定された場合には、前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信するための計算を実行し、実行された計算の結果を、前記第1の通信装置に送信する、
情報処理装置。
[2]
前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定されなかった場合には、前記第1範囲に含まれる計算の少なくとも一部を実行し、実行された計算の結果を、前記第2の通信装置に送信する前記計算の結果として用い、
前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定された場合には、前記第1範囲に含まれる計算のうちの少なくとも一部を実行し、実行された計算の結果を、前記第1の通信装置に送信するための前記計算に用いる、
[1]に記載の情報処理装置。
[3]
前記第1範囲に含まれる計算のうちの少なくとも一部を実行し、
前記判定が前記第1範囲に含まれる計算を実行している間に行われ、かつ、前記第1の通信装置に送信すると判定された場合に、前記判定の前に実行された前記計算の結果を前記第1の通信装置に送信するための前記計算に用いる、
[1]に記載の情報処理装置。
[4]
前記第1範囲に含まれる計算のうちの少なくとも一部を実行し、
前記判定が前記第1範囲に含まれる計算を実行している間に行われ、かつ、前記第1の通信装置に送信すると判定された場合に、前記判定の前に実行された前記計算の結果を前記第2の通信装置に送信する前記計算の結果として用いる、
[1]に記載の情報処理装置。
[5]
前記第1範囲に含まれる計算のうちの少なくとも一部を実行し、
前記判定が前記第1範囲に含まれる計算を実行している間に行われ、かつ、前記第1の通信装置に送信すると判定された場合に、前記第1範囲に含まれる計算を継続し、前記第1範囲の計算の結果を前記第2の通信装置に送信する前記計算の結果として用いる、
[1]ないし[4]のいずれか一項に記載の情報処理装置。
[6]
前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定された場合に、さらに、前記一連の計算のうちの第1範囲に含まれる計算の少なくとも一部の計算の結果を、前記第2の通信装置に送信する、
[1]ないし[5]のいずれか一項に記載の情報処理装置。
[7]
前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定された場合に、前記第1範囲に含まれる計算の全てが実行されなかったときは、実行された計算の結果とともに、実行された計算の結果が前記一連の計算のいずれの位置であるかを示す情報を、前記第2の通信装置に送信する、
[6]に記載の情報処理装置。
[8]
前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定された場合に、実行された計算の結果と、実行された計算の結果が前記一連の計算のいずれの位置であるかを示す情報と、を、前記一連の計算の少なくとも一部を実行可能な第3の通信装置に送信する、
[1]ないし[7]のいずれか一項に記載の情報処理装置。
[9]
前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定された場合に、前記第1範囲に含まれる計算の所定位置までを実行し、実行された計算の結果を、前記第1の通信装置に送信するための前記計算に用いる、
[1]ないし[7]のいずれか一項に記載の情報処理装置。
[10]
前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定された場合に、前記第1範囲に含まれる計算のいずれの位置まで実行するかを決定し、前記第1範囲に含まれる計算を決定された位置まで実行し、実行された計算の結果と、決定された位置を示す情報と、を前記第2の通信装置に送信する、
[1]ないし[7]のいずれか一項に記載の情報処理装置。
[11]
前記ディープニューラルネットワークの一連の計算のうちの少なくとも一部を担当する装置の計算余力に関する情報、前記ディープニューラルネットワークの一連の計算のうちの少なくとも一部を担当する装置におけるトラフィック量に関する情報、のうちの少なくとも1つに基づいて前記判定を行う、
[1]ないし[10]のいずれか一項に記載の情報処理装置。
[12]
前記第1の通信装置との間の通信品質、前記第1の通信装置のモビリティ情報、のうちの少なくとも1つに基づいて前記判定を行う、
[1]ないし[11]のいずれか一項に記載の情報処理装置。
[13]
前記一連の計算を途中終了することを指示する情報、前記一連の計算における途中の計算の結果を前記第1の通信装置に送信することを指示する情報、のうちの少なくとも1つに基づいて前記判定を行う、
[1]ないし[12]のいずれか一項に記載の情報処理装置。
[14]
前記第1の通信装置から要求情報を取得し、前記要求情報に基づいて前記判定を行う、 [1]ないし[13]のいずれか一項に記載の情報処理装置。
[15]
前記第1範囲の計算に必要な時間が与えられた許容時間よりも大きい場合に、前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定する、
[1]ないし[14]のいずれか一項に記載の情報処理装置。
[16]
ディープニューラルネットワークの一連の計算のうちの第1範囲に含まれる計算を途中終了するか否かの判定を行い、
前記第1範囲に含まれる計算を途中終了すると判定された場合に、前記第1範囲に含まれる計算の一部を実行し、実行された計算の結果と、実行された計算の結果が前記一連の計算のいずれの位置であるかを示す情報と、を、前記第1範囲に続く第2範囲の計算担当である装置に送信する、
情報処理装置。
[17]
前記第2範囲の計算担当である装置が前記第2範囲に含まれる計算を途中終了してもよいか否かを示す情報を、前記第2範囲の計算担当である装置にさらに送信する、
[16]に記載の情報処理装置。
[18]
ディープニューラルネットワークの一連の計算のうちの一部の計算を担当する情報処理装置において実行される情報処理方法であって、
前記ディープニューラルネットワークの一連の計算における途中の計算の結果を、第1の通信装置に送信するか否かの判定を行うステップと、
前記第1の通信装置に送信すると判定されなかった場合には、前記一連の計算のうちの第1範囲に含まれる計算の少なくとも一部の計算の結果を、前記第1範囲に続く第2範囲の計算担当である第2の通信装置に送信するステップと、
前記第1の通信装置に送信すると判定された場合には、前記ディープニューラルネットワークの一連の計算における途中の計算の結果を用いて前記第1の通信装置に送信するための計算を実行し、実行された計算の結果を、前記第1の通信装置に送信するステップと、
を備える情報処理方法。
[19]
ディープニューラルネットワークの一連の計算のうちの第1範囲に含まれる計算を途中終了するか否かの判定を行うステップと、
前記第1範囲に含まれる計算を途中終了すると判定された場合に、前記第1範囲に含まれる計算の一部を実行し、実行された計算の結果と、実行された計算の結果が前記一連の計算のいずれの位置であるかを示す情報と、を、前記第1範囲に続く第2範囲の計算担当である装置に送信するステップと、
を備える情報処理方法。
[20]
ディープニューラルネットワークの一連の計算のうちの一部の計算を担当する第1情報処理装置および第2情報処理装置を少なくとも備え、
前記第1情報処理装置は、
前記ディープニューラルネットワークの一連の計算における途中の計算の結果を、第1の通信装置に送信するか否かの判定を行い、
前記第1の通信装置に送信すると判定されなかった場合には、前記一連の計算のうちの第1範囲に含まれる計算の少なくとも一部の計算の結果を、前記第2情報処理装置に送信し、
前記第1の通信装置に送信すると判定された場合には、前記ディープニューラルネットワークの一連の計算における途中の計算の結果を用いて前記第1の通信装置に送信するための計算を実行し、実行された計算の結果を、前記第1の通信装置に送信し、
前記第2情報処理装置は、
前記一連の計算のうちの前記第1情報処理装置によって実行された計算以降の計算を前記第1情報処理装置によって実行された計算の結果に基づいて実行する、
情報処理システム。
[21]
少なくとも、ディープニューラルネットワークの一連の計算のうちの第1範囲の計算を行う第1情報処理装置と、前記ディープニューラルネットワークの一連の計算のうちの前記第1範囲に続く第2範囲の計算を行う第2情報処理装置と、を備え、
前記第1情報処理装置は、
前記第1範囲に含まれる計算を途中終了するか否かの判定を行い、
前記第1範囲に含まれる計算を途中終了すると判定された場合に、前記第1範囲に含まれる計算の一部を実行し、実行された計算の結果と、実行された計算の結果が前記一連の計算のいずれの位置であるかを示す情報と、を、前記第2情報処理装置に送信し、
前記第2情報処理装置は、
前記情報を受け取ったときは、前記一連の計算の前記第1情報処理装置によって実行された計算の続きを、前記第1情報処理装置によって実行された計算の結果に基づいて実行する、
情報処理システム。
[22]
前記第1範囲を決定する第3情報処理装置
をさらに備える[20]に記載の情報処理システム。
[23]
前記第1範囲に含まれる計算を途中終了するか否かの判定を行う第3情報処理装置
をさらに備え、
前記第3情報処理装置は、前記第1範囲に含まれる計算を途中終了する指示を前記第1情報処理装置に送信し、
前記第1情報処理装置は、前記指示を受け取った場合に、前記第1範囲に含まれる計算を途中終了すると判定する、
[21]に記載の情報処理システム。
Claims (23)
- ディープニューラルネットワークの一連の計算のうちの一部の計算を担当する情報処理装置であって、
前記ディープニューラルネットワークの一連の計算における途中の計算の結果を、第1の通信装置に送信するか否かの判定を行い、
前記第1の通信装置に送信すると判定されなかった場合には、前記一連の計算のうちの第1範囲に含まれる計算の少なくとも一部の計算の結果を、前記第1範囲に続く第2範囲の計算担当である第2の通信装置に送信し、
前記第1の通信装置に送信すると判定された場合には、前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信するための計算を実行し、実行された計算の結果を、前記第1の通信装置に送信する、
情報処理装置。 - 前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定されなかった場合には、前記第1範囲に含まれる計算の少なくとも一部を実行し、実行された計算の結果を、前記第2の通信装置に送信する前記計算の結果として用い、
前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定された場合には、前記第1範囲に含まれる計算のうちの少なくとも一部を実行し、実行された計算の結果を、前記第1の通信装置に送信するための前記計算に用いる、
請求項1に記載の情報処理装置。 - 前記第1範囲に含まれる計算のうちの少なくとも一部を実行し、
前記判定が前記第1範囲に含まれる計算を実行している間に行われ、かつ、前記第1の通信装置に送信すると判定された場合に、前記判定の前に実行された前記計算の結果を前記第1の通信装置に送信するための前記計算に用いる、
請求項1に記載の情報処理装置。 - 前記第1範囲に含まれる計算のうちの少なくとも一部を実行し、
前記判定が前記第1範囲に含まれる計算を実行している間に行われ、かつ、前記第1の通信装置に送信すると判定された場合に、前記判定の前に実行された前記計算の結果を前記第2の通信装置に送信する前記計算の結果として用いる、
請求項1に記載の情報処理装置。 - 前記第1範囲に含まれる計算のうちの少なくとも一部を実行し、
前記判定が前記第1範囲に含まれる計算を実行している間に行われ、かつ、前記第1の通信装置に送信すると判定された場合に、前記第1範囲に含まれる計算を継続し、前記第1範囲の計算の結果を前記第2の通信装置に送信する前記計算の結果として用いる、
請求項1に記載の情報処理装置。 - 前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定された場合に、さらに、前記一連の計算のうちの第1範囲に含まれる計算の少なくとも一部の計算の結果を、前記第2の通信装置に送信する、
請求項1に記載の情報処理装置。 - 前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定された場合に、前記第1範囲に含まれる計算の全てが実行されなかったときは、実行された計算の結果とともに、実行された計算の結果が前記一連の計算のいずれの位置であるかを示す情報を、前記第2の通信装置に送信する、
請求項6に記載の情報処理装置。 - 前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定された場合に、実行された計算の結果と、実行された計算の結果が前記一連の計算のいずれの位置であるかを示す情報と、を、前記一連の計算の少なくとも一部を実行可能な第3の通信装置に送信する、
請求項1に記載の情報処理装置。 - 前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定された場合に、前記第1範囲に含まれる計算の所定位置までを実行し、実行された計算の結果を、前記第1の通信装置に送信するための前記計算に用いる、
請求項1に記載の情報処理装置。 - 前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定された場合に、前記第1範囲に含まれる計算のいずれの位置まで実行するかを決定し、前記第1範囲に含まれる計算を決定された位置まで実行し、実行された計算の結果と、決定された位置を示す情報と、を前記第2の通信装置に送信する、
請求項1に記載の情報処理装置。 - 前記ディープニューラルネットワークの一連の計算のうちの少なくとも一部を担当する装置の計算余力に関する情報、前記ディープニューラルネットワークの一連の計算のうちの少なくとも一部を担当する装置におけるトラフィック量に関する情報、のうちの少なくとも1つに基づいて前記判定を行う、
請求項1に記載の情報処理装置。 - 前記第1の通信装置との間の通信品質、前記第1の通信装置のモビリティ情報、のうちの少なくとも1つに基づいて前記判定を行う、
請求項1に記載の情報処理装置。 - 前記一連の計算を途中終了することを指示する情報、前記一連の計算における途中の計算の結果を前記第1の通信装置に送信することを指示する情報、のうちの少なくとも1つに基づいて前記判定を行う、
請求項1に記載の情報処理装置。 - 前記第1の通信装置から要求情報を取得し、前記要求情報に基づいて前記判定を行う、
請求項1に記載の情報処理装置。 - 前記第1範囲の計算に必要な時間が与えられた許容時間よりも大きい場合に、前記ディープニューラルネットワークの一連の計算における途中の計算の結果を前記第1の通信装置に送信すると判定する、
請求項1に記載の情報処理装置。 - ディープニューラルネットワークの一連の計算のうちの第1範囲に含まれる計算を途中終了するか否かの判定を行い、
前記第1範囲に含まれる計算を途中終了すると判定された場合に、前記第1範囲に含まれる計算の一部を実行し、実行された計算の結果と、実行された計算の結果が前記一連の計算のいずれの位置であるかを示す情報と、を、前記第1範囲に続く第2範囲の計算担当である装置に送信する、
情報処理装置。 - 前記第2範囲の計算担当である装置が前記第2範囲に含まれる計算を途中終了してもよいか否かを示す情報を、前記第2範囲の計算担当である装置にさらに送信する、
請求項16に記載の情報処理装置。 - ディープニューラルネットワークの一連の計算のうちの一部の計算を担当する情報処理装置において実行される情報処理方法であって、
前記ディープニューラルネットワークの一連の計算における途中の計算の結果を、第1の通信装置に送信するか否かの判定を行うステップと、
前記第1の通信装置に送信すると判定されなかった場合には、前記一連の計算のうちの第1範囲に含まれる計算の少なくとも一部の計算の結果を、前記第1範囲に続く第2範囲の計算担当である第2の通信装置に送信するステップと、
前記第1の通信装置に送信すると判定された場合には、前記ディープニューラルネットワークの一連の計算における途中の計算の結果を用いて前記第1の通信装置に送信するための計算を実行し、実行された計算の結果を、前記第1の通信装置に送信するステップと、
を備える情報処理方法。 - ディープニューラルネットワークの一連の計算のうちの第1範囲に含まれる計算を途中終了するか否かの判定を行うステップと、
前記第1範囲に含まれる計算を途中終了すると判定された場合に、前記第1範囲に含まれる計算の一部を実行し、実行された計算の結果と、実行された計算の結果が前記一連の計算のいずれの位置であるかを示す情報と、を、前記第1範囲に続く第2範囲の計算担当である装置に送信するステップと、
を備える情報処理方法。 - ディープニューラルネットワークの一連の計算のうちの一部の計算を担当する第1情報処理装置および第2情報処理装置を少なくとも備え、
前記第1情報処理装置は、
前記ディープニューラルネットワークの一連の計算における途中の計算の結果を、第1の通信装置に送信するか否かの判定を行い、
前記第1の通信装置に送信すると判定されなかった場合には、前記一連の計算のうちの第1範囲に含まれる計算の少なくとも一部の計算の結果を、前記第2情報処理装置に送信し、
前記第1の通信装置に送信すると判定された場合には、前記ディープニューラルネットワークの一連の計算における途中の計算の結果を用いて前記第1の通信装置に送信するための計算を実行し、実行された計算の結果を、前記第1の通信装置に送信し、
前記第2情報処理装置は、
前記一連の計算のうちの前記第1情報処理装置によって実行された計算以降の計算を前記第1情報処理装置によって実行された計算の結果に基づいて実行する、
情報処理システム。 - 少なくとも、ディープニューラルネットワークの一連の計算のうちの第1範囲の計算を行う第1情報処理装置と、前記ディープニューラルネットワークの一連の計算のうちの前記第1範囲に続く第2範囲の計算を行う第2情報処理装置と、を備え、
前記第1情報処理装置は、
前記第1範囲に含まれる計算を途中終了するか否かの判定を行い、
前記第1範囲に含まれる計算を途中終了すると判定された場合に、前記第1範囲に含まれる計算の一部を実行し、実行された計算の結果と、実行された計算の結果が前記一連の計算のいずれの位置であるかを示す情報と、を、前記第2情報処理装置に送信し、
前記第2情報処理装置は、
前記情報を受け取ったときは、前記一連の計算の前記第1情報処理装置によって実行された計算の続きを、前記第1情報処理装置によって実行された計算の結果に基づいて実行する、
情報処理システム。 - 前記第1範囲を決定する第3情報処理装置
をさらに備える請求項20に記載の情報処理システム。 - 前記第1範囲に含まれる計算を途中終了するか否かの判定を行う第3情報処理装置
をさらに備え、
前記第3情報処理装置は、前記第1範囲に含まれる計算を途中終了する指示を前記第1情報処理装置に送信し、
前記第1情報処理装置は、前記指示を受け取った場合に、前記第1範囲に含まれる計算を途中終了すると判定する、
請求項21に記載の情報処理システム。
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202280045103.2A CN117546175A (zh) | 2021-07-01 | 2022-03-29 | 通信设备、通信方法和通信*** |
EP22832543.7A EP4365773A1 (en) | 2021-07-01 | 2022-03-29 | Communication device, communication method, and communication system |
JP2023531450A JPWO2023276382A1 (ja) | 2021-07-01 | 2022-03-29 |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2021-110381 | 2021-07-01 | ||
JP2021110381 | 2021-07-01 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2023276382A1 true WO2023276382A1 (ja) | 2023-01-05 |
Family
ID=84692660
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2022/015609 WO2023276382A1 (ja) | 2021-07-01 | 2022-03-29 | 通信装置、通信方法、および通信システム |
Country Status (4)
Country | Link |
---|---|
EP (1) | EP4365773A1 (ja) |
JP (1) | JPWO2023276382A1 (ja) |
CN (1) | CN117546175A (ja) |
WO (1) | WO2023276382A1 (ja) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018196631A1 (en) * | 2017-04-26 | 2018-11-01 | Midea Group Co., Ltd. | Training machine learning models on a large-scale distributed system using a job server |
JP2019148876A (ja) * | 2018-02-26 | 2019-09-05 | 日本電信電話株式会社 | 制御装置、データ処理制御方法、及びプログラム |
JP2021068393A (ja) * | 2019-10-28 | 2021-04-30 | 富士通株式会社 | 情報処理システム、情報処理装置、および情報処理プログラム |
-
2022
- 2022-03-29 CN CN202280045103.2A patent/CN117546175A/zh active Pending
- 2022-03-29 WO PCT/JP2022/015609 patent/WO2023276382A1/ja active Application Filing
- 2022-03-29 JP JP2023531450A patent/JPWO2023276382A1/ja active Pending
- 2022-03-29 EP EP22832543.7A patent/EP4365773A1/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018196631A1 (en) * | 2017-04-26 | 2018-11-01 | Midea Group Co., Ltd. | Training machine learning models on a large-scale distributed system using a job server |
JP2019148876A (ja) * | 2018-02-26 | 2019-09-05 | 日本電信電話株式会社 | 制御装置、データ処理制御方法、及びプログラム |
JP2021068393A (ja) * | 2019-10-28 | 2021-04-30 | 富士通株式会社 | 情報処理システム、情報処理装置、および情報処理プログラム |
Non-Patent Citations (3)
Title |
---|
"TR (Technical Report) 22.874, V0.1.0", 3GPP (3RD GENERATION PARTNERSHIP PROJECT |
SURAT TEERAPITTAYANON, BRANCHYNET: FAST INFERENCE VIA EARLY-EXITING FROM DEEP NEURAL NETWORKS, Retrieved from the Internet <URL:http://bradmcdanel.com/files/2016-icpr-teerapittayanon-mcdanel-kung.pdf> |
SURAT TEERAPITTAYANON, DISTRIBUTED DEEP NEURAL NETWORKS OVER THE CLOUD, THE EDGE AND END DEVICES |
Also Published As
Publication number | Publication date |
---|---|
CN117546175A (zh) | 2024-02-09 |
JPWO2023276382A1 (ja) | 2023-01-05 |
EP4365773A1 (en) | 2024-05-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Dong et al. | An edge computing empowered radio access network with UAV-mounted FSO fronthaul and backhaul: Key challenges and approaches | |
KR20200095469A (ko) | 핸드오버 관련 기술, 장치 및 방법 | |
JP7480713B2 (ja) | 情報処理装置、情報処理方法、及び情報処理プログラム | |
WO2021117373A1 (ja) | 情報処理装置、情報処理方法、端末装置、基地局装置、及びプログラム | |
JP7452437B2 (ja) | 通信装置、通信方法、情報処理装置、及び情報処理方法 | |
EP4027680A1 (en) | Communication control device, communication control method, and communication control program | |
WO2021070703A1 (ja) | 端末装置及び通信方法 | |
WO2020218183A1 (ja) | 基地局装置、基地局装置の制御方法、端末装置、及び端末装置の制御方法 | |
KR102346094B1 (ko) | 무선 통신 시스템에서 경로를 제공하기 위한 장치 및 방법 | |
WO2022138232A1 (ja) | 通信装置、通信方法、および通信システム | |
WO2023276382A1 (ja) | 通信装置、通信方法、および通信システム | |
EP4135474A1 (en) | Mobile and plug-and-play cell networks | |
Peng et al. | Fog Radio Access Networks (F-RAN) | |
US20220330334A1 (en) | Communication control device, communication device, communication control method, and communication method | |
US20240259464A1 (en) | Communication device, communication method, and communication system | |
CN117643105A (zh) | 使用移动通信节点进行数据路由的方法和装置 | |
CN116648957A (zh) | 通信设备,通信方法和通信*** | |
WO2024018780A1 (ja) | 端末装置、情報処理装置、及び通信方法 | |
WO2020001769A1 (en) | Quality of service control and mobility management for advanced vehicular users of wireless network | |
WO2022113808A1 (ja) | 通信装置、通信方法、及び通信システム | |
WO2021111751A1 (ja) | 制御装置、基地局装置、制御方法および接続方法 | |
WO2021132194A1 (ja) | 端末装置、管理装置及び通信方法 |
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: 22832543 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2023531450 Country of ref document: JP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 202280045103.2 Country of ref document: CN |
|
WWE | Wipo information: entry into national phase |
Ref document number: 18574143 Country of ref document: US |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2022832543 Country of ref document: EP |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
ENP | Entry into the national phase |
Ref document number: 2022832543 Country of ref document: EP Effective date: 20240201 |