WO2024103352A1 - 一种通信方法、装置及系统 - Google Patents

一种通信方法、装置及系统 Download PDF

Info

Publication number
WO2024103352A1
WO2024103352A1 PCT/CN2022/132654 CN2022132654W WO2024103352A1 WO 2024103352 A1 WO2024103352 A1 WO 2024103352A1 CN 2022132654 W CN2022132654 W CN 2022132654W WO 2024103352 A1 WO2024103352 A1 WO 2024103352A1
Authority
WO
WIPO (PCT)
Prior art keywords
target branch
channel
information
target
dictionary set
Prior art date
Application number
PCT/CN2022/132654
Other languages
English (en)
French (fr)
Inventor
胡斌
张公正
王坚
李榕
Original Assignee
华为技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Priority to PCT/CN2022/132654 priority Critical patent/WO2024103352A1/zh
Publication of WO2024103352A1 publication Critical patent/WO2024103352A1/zh

Links

Images

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L27/00Modulated-carrier systems

Definitions

  • the present application relates to the field of communication technology, and in particular to a communication method, device and system.
  • deep learning technology can improve the performance of wireless communication systems and has the potential to be applied in interference adjustment, channel estimation and signal detection, signal processing and other aspects at the physical layer.
  • the neural network transceiver optimizes specific performance indicators and channel models by combining the transmitter and receiver, so that customized self-evolving air interfaces can be realized without the need for prior expert knowledge, and the Shannon limit can be approached.
  • the distribution of channels changes dynamically. This requires the neural network model to be able to identify channel changes in the current environment and quickly adapt to new scenarios with less training overhead.
  • Park S et al. combined the meta-learning method to give a meta-autoencoder network structure.
  • the scheme is to use the end-to-end training under various channels as a subtask to train the initial transceiver network with potential under various channels.
  • the training goal of the main task is to obtain the optimal network initialization parameters, so that the system can converge the training of any channel with fewer stochastic gradient descent (SGD) steps, and then quickly adapt to the changes of time-varying channels.
  • SGD stochastic gradient descent
  • the present application discloses a communication method, device and system, which can solve the problems of neural network training and deduction caused by dynamic changes of channels in the environment.
  • an embodiment of the present application provides a communication method.
  • the method can be performed by a communication device, or by a component of a communication device (such as a chip (system)).
  • the method includes: a reference signal receiving end receives a first reference signal. Then, the reference signal receiving end inputs the first reference signal into a first target model for processing, and obtains K first target branch adaptive layers in the first target model and the weights of the K first target branch adaptive layers.
  • the first target model includes N first target branch adaptive layers, K is not greater than N, and K and N are both positive integers.
  • the reference signal receiving end also sends a first information.
  • the first information indicates the weights of the K first target branch adaptive layers and the K first target branch adaptive layers.
  • the weights of the K first target branch adaptive layers and the K first target branch adaptive layers are used by the transmitting end to process the coded data to be sent.
  • the reference signal receiving end determines K first target branch adaptive layers and weights of the K first target branch adaptive layers among the N first target branch adaptive layers of the first target model based on the received reference signal.
  • the actual channel distribution is characterized based on the determined K first target branch adaptive layers and K weights to cope with the dynamic changes of the channel, which can improve the system performance.
  • the method further includes: a reference signal receiving end receives first data, where the first data is obtained by the transmitting end processing the coded data to be transmitted, and then the reference signal receiving end inputs the first data into the K first target branch adaptive layers for processing to obtain processed data.
  • the method further includes: a reference signal receiving end sends a first channel dictionary set, where the first channel dictionary set is used by the transmitting end to determine a correspondence between the N first target branch adaptive layers and the M second target branch adaptive layers of the transmitting end, where M is a positive integer.
  • the transmitting end selects the optimal processing method under the corresponding channel distribution to process the encoded data, which can improve communication performance.
  • the method further includes: the reference signal receiving end receives a second channel dictionary set. Then, the reference signal receiving end performs dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set. Finally, the reference signal receiving end sends first indication information, and the first indication information is used to indicate the correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the transmitting end selects the optimal processing method under the corresponding channel distribution to process the encoded data, which can improve communication performance.
  • the first channel dictionary set includes a correspondence between a first target branch adaptation layer and a channel label.
  • the method further includes: the reference signal receiving end receives the second reference signal. Then, the reference signal receiving end sends third information, the third information is the same as the first information, or the third information indicates that the transmitting end sends the reference signal after a first time interval. The third information is obtained based on the second reference signal. Finally, the reference signal receiving end receives the data sent by the transmitting end within the first time.
  • the transmitting end sends data within the first time, which can be understood as the transmitting end not sending a reference signal within the first time. In this way, the pilot transmission overhead can be reduced and the spectrum efficiency of data transmission can be increased.
  • an embodiment of the present application provides a communication method.
  • the method can be performed by a communication device, or by a component of a communication device (e.g., a chip (system)).
  • the method includes: a reference signal receiving end receives a first reference signal. Then, the reference signal receiving end inputs the first reference signal into a first target model for processing, and obtains K first target branch adaptive layers in the first target model and the weights of the K first target branch adaptive layers.
  • the first target model includes N first target branch adaptive layers, K is not greater than N, and K and N are both positive integers.
  • the reference signal receiving end also sends fourth information, which indicates the weights of the K second target branch adaptive layers and the K second target branch adaptive layers.
  • the weights of the K second target branch adaptive layers and the K second target branch adaptive layers are used by the transmitting end to process the coded data to be sent.
  • the weights of the K second target branch adaptive layers and the K second target branch adaptive layers are determined based on the weights of the K first target branch adaptive layers and the K first target branch adaptive layers.
  • the reference signal receiving end determines K first target branch adaptive layers and the weights of the K first target branch adaptive layers among the N first target branch adaptive layers of the first target model based on the received reference signal. Then, the K second target branch adaptive layers and the weights of the K second target branch adaptive layers are indicated to the transmitting end.
  • the actual channel distribution is characterized based on the determined K first target branch adaptive layers and K weights to cope with the dynamic changes of the channel, which can improve the system performance. And by indicating to the transmitting end so that the transmitting end can process the coded data to be sent, the communication performance can be improved.
  • the method further includes: the reference signal receiving end receives first data.
  • the first data is obtained by the transmitting end processing the coded data to be transmitted.
  • the reference signal receiving end also inputs the first data into the K first target branch adaptive layers for processing to obtain processed data.
  • the method further includes: the reference signal receiving end sends a first channel dictionary set, the first channel dictionary set is used by the transmitting end to determine the correspondence between the N first target branch adaptation layers and the M second target branch adaptation layers of the transmitting end, M is a positive integer.
  • the reference signal receiving end also receives first indication information, the first indication information is used to indicate the correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the reference signal receiving end can determine the weights of the above-mentioned K second target branch adaptive layers and the K second target branch adaptive layers to instruct the transmitting end to process the encoded data according to the optimal processing method under the corresponding channel distribution, thereby improving communication performance.
  • the method further includes: the reference signal receiving end receives a second channel dictionary set. Then, the reference signal receiving end performs dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the reference signal receiving end can determine the weights of the above-mentioned K second target branch adaptive layers and the K second target branch adaptive layers to instruct the transmitting end to process the encoded data according to the optimal processing method under the corresponding channel distribution, thereby improving communication performance.
  • the first channel dictionary set includes a correspondence between a first target branch adaptation layer and a channel label.
  • the method further includes: the reference signal receiving end receives the second reference signal. Then the reference signal receiving end sends fifth information, the fifth information is the same as the fourth information, or the fifth information indicates that the transmitting end sends the reference signal after a first time interval, and the fifth information is obtained according to the second reference signal.
  • the reference signal receiving end also receives data sent by the transmitting end within the first time.
  • the transmitting end sends data within the first time, which can be understood as the transmitting end not sending a reference signal within the first time. In this way, the pilot transmission overhead can be reduced and the spectrum efficiency of data transmission can be increased.
  • an embodiment of the present application provides a communication method.
  • the method can be performed by a communication device, or by a component of a communication device (e.g., a chip (system)).
  • the method includes: a transmitting end sends a first reference signal. Then, the transmitting end receives first information, which indicates K first target branch adaptive layers and weights of the K first target branch adaptive layers in a first target model of the first reference signal receiving end. The first information is obtained based on the first reference signal, and K is a positive integer.
  • the information received by the transmitting end indicates the K first target branch adaptive layers and the weights of the K first target branch adaptive layers in the first target model of the first reference signal receiving end.
  • the actual channel distribution is characterized based on the K first target branch adaptive layers and the K weights to cope with the dynamic changes of the channel, which can improve the communication performance.
  • the method further includes: the transmitting end determines, according to the first information, K second target branch adaptive layers corresponding to the K first target branch adaptive layers and weights of the K second target branch adaptive layers. Then, the transmitting end inputs the coded data to be transmitted into the K second target branch adaptive layers of the second target model for processing to obtain the first data. Finally, the transmitting end transmits the first data.
  • the transmitting end updates the third target model to obtain the second target model, so that the second target model includes the K second target branch adaptive layers.
  • the third target model includes M second target branch adaptive layers, where M is a positive integer.
  • the corresponding target branch adaptation layer and weight can be determined based on the received information to process the encoded data, thereby improving communication performance.
  • the method further includes: the transmitting end receives a first channel dictionary set. Then, the transmitting end performs dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • both ends can align the dictionaries and then select the optimal processing method under the corresponding channel distribution to process the encoded data, thereby improving communication performance.
  • the method further includes: the transmitting end transmits the second channel dictionary set.
  • the transmitting end also receives first indication information, the first indication information being used to indicate a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the transmitting end selects the optimal processing method under the corresponding channel distribution to process the encoded data, which can improve communication performance.
  • the second channel dictionary set includes a correspondence between a second target branch adaptation layer and a channel label.
  • the method further includes: the transmitting end sends a second reference signal. Then, the transmitting end receives third information, the third information is the same as the first information, or the third information indicates that the reference signal is sent after a first time interval. The third information is obtained based on the second reference signal. Finally, the transmitting end sends data within the first time.
  • the transmitting end sends data within the first time, which can be understood as the transmitting end not sending a reference signal within the first time. In this way, the pilot transmission overhead can be reduced and the spectrum efficiency of data transmission can be increased.
  • an embodiment of the present application provides a communication method.
  • the method can be performed by a communication device, or by a component of a communication device (e.g., a chip (system)).
  • the method includes: a transmitting end sends a first reference signal. Then the transmitting end receives fourth information.
  • the fourth information indicates K second target branch adaptive layers and the weights of the K second target branch adaptive layers.
  • the fourth information is obtained based on the first reference signal, and K is a positive integer.
  • the information received by the transmitting end indicates K second target branch adaptive layers and the weights of the K second target branch adaptive layers.
  • the actual channel distribution is characterized based on the K target branch adaptive layers and the K weights to cope with the dynamic changes of the channel, which can improve the communication performance.
  • the method further includes:
  • the transmitting end inputs the coded data to be sent into the K second target branch adaptive layers of the second target model for processing according to the fourth information to obtain the first data.
  • the sending end sends the first data.
  • the transmitting end processes the coded data according to the optimal processing method under the corresponding channel distribution, which can improve the communication performance.
  • the method further includes:
  • the sending end receives the first channel dictionary set.
  • the transmitting end performs dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the transmitting end further sends first indication information, where the first indication information is used to indicate a corresponding relationship between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • both ends can align the dictionaries, and then instruct the reference signal receiving end to determine the corresponding target branch adaptation layer and weight, so that the transmitting end can process the encoded data based on the optimal processing method under the corresponding channel distribution, which can improve communication performance.
  • the method further includes:
  • the transmitting end sends a second channel dictionary set, which is used by the reference signal receiving end to determine the correspondence between the N first target branch adaptive layers of the reference signal receiving end and the M second target branch adaptive layers of the transmitting end, where M and N are both positive integers.
  • the reference signal receiving end can align the dictionaries at both ends to determine the corresponding target branch adaptation layer and weight, so that the transmitting end can process the encoded data based on the optimal processing method under the corresponding channel distribution, which can improve communication performance.
  • the second channel dictionary set includes a correspondence between a second target branch adaptation layer and a channel label.
  • the method further includes:
  • the transmitting end sends a second reference signal.
  • the transmitting end receives fifth information, where the fifth information is the same as the fourth information, or the fifth information indicates to send the reference signal after a first time interval, and the fifth information is obtained according to the second reference signal;
  • the sending end sends data within the first time.
  • the transmitting end sends data within the first time, which can be understood as the transmitting end not sending a reference signal within the first time. In this way, the pilot transmission overhead can be reduced and the spectrum efficiency of data transmission can be increased.
  • an embodiment of the present application provides a communication method.
  • the method can be executed by a communication device, or by a component of a communication device (such as a chip (system)).
  • the method includes: a reference signal receiving end receives a first reference signal.
  • the reference signal receiving end inputs the first reference signal into a first target model for processing, and obtains K first target branch adaptive layers in the first target model and the weights of the K first target branch adaptive layers.
  • the first target model includes N first target branch adaptive layers, K is not greater than N, and K and N are both positive integers.
  • the reference signal receiving end also receives first data. Then, the reference signal receiving end inputs the first data into the K first target branch adaptive layers for processing to obtain processed data.
  • the maximum weight among the weights of the K first target branch adaptive layers is not less than a preset value.
  • the reference signal receiving end determines K first target branch adaptive layers and the weights of the K first target branch adaptive layers among the N first target branch adaptive layers of the first target model based on the received reference signal.
  • the maximum weight among the weights of the K first target branch adaptive layers is not less than a preset value
  • the received first data is input into the K first target branch adaptive layers for processing to obtain processed data.
  • the actual channel distribution is characterized based on the determined K first target branch adaptive layers and K weights to cope with the dynamic changes of the channel, which can improve the system performance.
  • the first target model further includes a first target channel feature extraction network and a target sparse gating module, wherein the first target channel feature extraction network is used to process the first reference signal to obtain channel distribution information and channel category information corresponding to the channel distribution information.
  • the target sparse gating module is used to calculate the K first target branch adaptive layers and the weights of the K first target branch adaptive layers according to the channel distribution information and the channel category information.
  • the first target channel feature extraction network and the target sparse gating module can be two independent modules or integrated into one, and this solution does not impose any limitation on this.
  • the K first target branch adaptive layers are used to process the first data respectively to obtain data processed by the K first target branch adaptive layers respectively.
  • the weights of the K first target branch adaptive layers are used to perform weighted sum processing on the data processed by the K first target branch adaptive layers respectively to obtain the processed data.
  • the method further includes: when a maximum weight among the weights of the K first target branch adaptive layers is less than the preset value, the reference signal receiving end decodes the first data to obtain processed data.
  • the method further includes: a reference signal receiving end sends first information.
  • the first information indicates the K first target branch adaptive layers and the weights of the K first target branch adaptive layers.
  • the K first target branch adaptive layers and the weights of the K first target branch adaptive layers are used by the transmitting end to process the coded data to be sent to obtain the first data.
  • the transmitter selects the optimal processing method under the corresponding channel distribution to process the coded data according to the information instruction, which can improve the communication performance.
  • the method further includes: when the maximum weight among the weights of the K first target branch adaptive layers is less than the preset value, the reference signal receiving end sends second information, and the second information instructs the transmitting end to directly send the first data.
  • the method further includes: a reference signal receiving end sends a first channel dictionary set, where the first channel dictionary set is used by the transmitting end to determine a correspondence between the N first target branch adaptive layers and the M second target branch adaptive layers of the transmitting end, where M is a positive integer.
  • both ends can align the dictionaries so that the transmitting end can select the optimal processing method under the corresponding channel distribution to process the encoded data, thereby improving communication performance.
  • the method further includes: the reference signal receiving end receives a second channel dictionary set.
  • the reference signal receiving end performs dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the reference signal receiving end sends first indication information, and the first indication information is used to indicate the correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the transmitting end selects the optimal processing method under the corresponding channel distribution to process the encoded data, which can improve communication performance.
  • the method further includes:
  • the reference signal receiving end sends fourth information, and the fourth information indicates the weights of the K second target branch adaptive layers and the K second target branch adaptive layers.
  • the weights of the K second target branch adaptive layers and the K second target branch adaptive layers are used by the transmitting end to process the coded data to be sent to obtain the first data.
  • the weights of the K second target branch adaptive layers and the K second target branch adaptive layers are determined according to the weights of the K first target branch adaptive layers and the K first target branch adaptive layers.
  • the method further includes:
  • the reference signal receiving end sends a first channel dictionary set.
  • the reference signal receiving end also receives first indication information.
  • the first indication information is used to indicate a corresponding relationship between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the method further includes:
  • the reference signal receiving end receives the second channel dictionary set.
  • the reference signal receiving end also performs dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the first channel dictionary set includes a correspondence between a first target branch adaptation layer and a channel label.
  • the method further includes: the reference signal receiving end receives the second reference signal. Then, the reference signal receiving end sends third information, the third information is the same as the first information, or the third information indicates that the transmitting end sends the reference signal after a first time interval. The third information is obtained based on the second reference signal. The reference signal receiving end also receives data sent by the transmitting end within the first time.
  • the method further includes: the reference signal receiving end receives the second reference signal. Then, the reference signal receiving end sends fifth information, the fifth information is the same as the fourth information, or the fifth information indicates that the transmitting end sends the reference signal after a first time interval. The fifth information is obtained based on the second reference signal. The reference signal receiving end also receives data sent by the transmitting end within the first time.
  • the transmitting end sends data within the first time, which can be understood as the transmitting end not sending a reference signal within the first time.
  • the pilot transmission overhead can be reduced and the spectrum efficiency of data transmission can be increased.
  • This example can refer to the introduction of the third information above, which will not be repeated here.
  • an embodiment of the present application provides a communication method.
  • the method can be executed by a communication device, or by a component of a communication device (such as a chip (system)).
  • the method includes: a transmitting end sends a first reference signal.
  • the transmitting end receives the first information
  • the first information indicates the weights of the K first target branch adaptive layers and the K first target branch adaptive layers in the first target model of the first reference signal receiving end, and according to the first information, the K second target branch adaptive layers corresponding to the K first target branch adaptive layers and the weights of the K second target branch adaptive layers are determined.
  • the first information is obtained based on the first reference signal, and K is a positive integer.
  • the transmitting end inputs the coded data to be sent into the K second target branch adaptive layers of the second target model for processing to obtain the first data. Finally, the transmitting end sends the first data.
  • the transmitting end processes the coded data to be transmitted based on the received information before transmitting.
  • the actual channel distribution is characterized based on the determined K first target branch adaptive layers and K weights to cope with the dynamic changes of the channel; and the communication performance can be improved by selecting the optimal processing method under the corresponding channel distribution to process the coded data.
  • the transmitting end updates the third target model to obtain the second target model, so that the second target model includes the K second target branch adaptive layers.
  • the third target model includes M second target branch adaptive layers, where M is a positive integer.
  • the corresponding target branch adaptation layer and weight can be determined based on the received information to process the encoded data, thereby improving communication performance.
  • the K second target branch adaptive layers are used to respectively process the coded data to be sent to obtain data processed by the K second target branch adaptive layers respectively.
  • the weights of the K second target branch adaptive layers are used to perform weighted sum processing on the data processed respectively by the K second target branch adaptive layers to obtain the first data.
  • the method further includes: when the transmitting end receives second information, the second information indicates to directly send the coded data to be sent, and the transmitting end sends the coded data to be sent.
  • the second information is obtained according to the first reference signal.
  • the method further includes: the transmitting end receives a first channel dictionary set. Then, the transmitting end performs dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • both ends can align the dictionaries and then select the optimal processing method under the corresponding channel distribution to process the encoded data, thereby improving communication performance.
  • the method further includes: the transmitting end transmits a second channel dictionary set. Then, the transmitting end receives first indication information, where the first indication information is used to indicate a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the transmitting end selects the optimal processing method under the corresponding channel distribution to process the encoded data, which can improve communication performance.
  • the second channel dictionary set includes a correspondence between a second target branch adaptation layer and a channel label.
  • the method further includes: the transmitting end sends a second reference signal. Then the transmitting end receives third information, the third information is the same as the first information; or the third information is used to indicate that the reference signal is sent after a first time interval. The third information is obtained based on the second reference signal. Finally, the transmitting end sends data within the first time.
  • the transmitting end sends data within the first time, which can be understood as the transmitting end not sending a reference signal within the first time. In this way, the pilot transmission overhead can be reduced and the spectrum efficiency of data transmission can be increased.
  • an embodiment of the present application provides a communication method.
  • the method can be executed by a communication device, or by a component of a communication device (such as a chip (system)).
  • the method includes: a transmitting end sends a first reference signal.
  • the fourth information indicates the weights of the K second target branch adaptive layers and the K second target branch adaptive layers, and according to the fourth information, the coded data to be sent is input into the K second target branch adaptive layers of the second target model for processing to obtain the first data.
  • the fourth information is obtained based on the first reference signal, and K is a positive integer.
  • the transmitting end also sends the first data.
  • the information received by the transmitting end indicates K second target branch adaptive layers and the weights of the K second target branch adaptive layers.
  • the actual channel distribution is characterized based on the K target branch adaptive layers and the K weights to cope with the dynamic changes of the channel, which can improve the communication performance.
  • the transmitting end processes the encoded data according to the optimal processing method under the corresponding channel distribution, which can improve the communication performance.
  • the K second target branch adaptive layers are used to respectively process the encoded data to be sent to obtain the data respectively processed by the K second target branch adaptive layers; the weights of the K second target branch adaptive layers are used to perform weighted sum processing on the data respectively processed by the K second target branch adaptive layers to obtain the first data.
  • the method further includes:
  • the second information When receiving the second information, the second information indicates to directly send the coded data to be sent, and the transmitting end sends the coded data to be sent.
  • the second information is obtained according to the first reference signal.
  • the method further includes:
  • the transmitting end receives the first channel dictionary set. Then, the transmitting end performs dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set. The transmitting end also sends first indication information, which is used to indicate the correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • both ends can align the dictionaries, and then instruct the reference signal receiving end to determine the corresponding target branch adaptation layer and weight, so that the transmitting end can process the encoded data based on the optimal processing method under the corresponding channel distribution, which can improve communication performance.
  • the method further includes:
  • the transmitting end sends a second channel dictionary set, which is used by the reference signal receiving end to determine the correspondence between the N first target branch adaptive layers of the reference signal receiving end and the M second target branch adaptive layers of the transmitting end, where M and N are both positive integers.
  • the reference signal receiving end can align the dictionaries at both ends to determine the corresponding target branch adaptation layer and weight, so that the transmitting end can process the encoded data based on the optimal processing method under the corresponding channel distribution, which can improve communication performance.
  • the second channel dictionary set includes a correspondence between a second target branch adaptation layer and a channel label.
  • the method further includes:
  • the transmitting end sends a second reference signal.
  • the transmitting end receives fifth information.
  • the fifth information is the same as the fourth information; or the fifth information is used to indicate that the reference signal is sent after the first time interval, and the fifth information is obtained according to the second reference signal;
  • the sending end also sends data within the first time.
  • the transmitting end sends data within the first time, which can be understood as the transmitting end not sending a reference signal within the first time. In this way, the pilot transmission overhead can be reduced and the spectrum efficiency of data transmission can be increased.
  • an embodiment of the present application provides a target model training method.
  • the method can be performed by a communication device, or by a component of a communication device (such as a chip (system)).
  • the target model includes a target encoding network, a target decoding network, and a first target model, and the first target model includes a first target channel feature extraction network, N first target branch adaptive layers, and a target sparse gating module.
  • the method includes: training the initial channel feature extraction network to obtain the first target channel feature extraction network and the first channel dictionary set.
  • the N initial branch adaptive layers are trained to obtain the N first target branch adaptive layers, where N is the number of first target branch adaptive layers in the first channel dictionary set.
  • the initial sparse gating module is trained to obtain the target sparse gating module.
  • the first target channel feature extraction network is obtained based on training the initial channel feature extraction network.
  • N first target branch adaptive layers are obtained by training N initial branch adaptive layers.
  • the initial sparse gating module is trained according to the trained target encoding network, target decoding network, first target channel feature extraction network and N first target branch adaptive layers to obtain a target sparse gating module.
  • the target model obtained by this means can solve the problems of neural network training and deduction caused by dynamic changes in channels in the environment.
  • this scheme adjusts the output of the branch adaptive layer by training the weights of the branch adaptive layer instead of training the target encoding network parameters, which can reduce the overhead of network training.
  • the first channel dictionary set includes a correspondence between a first target branch adaptation layer and a channel label.
  • the initial channel feature extraction network is trained to obtain the first target channel feature extraction network and the first channel dictionary set, including: training the initial channel feature extraction network multiple times to obtain the first target channel feature extraction network and the first channel dictionary set.
  • the initial channel feature extraction network is pre-trained based on labeled historical channel information or a preset channel model.
  • both the unlabeled channel information and the labeled historical channel information are input into the channel feature extraction network U t-1 for processing, and the channel distribution information corresponding to the unlabeled channel information and the labeled historical channel information is obtained.
  • N t-1 cluster centers are obtained.
  • a predicted channel dictionary set is obtained.
  • N initial branch adaptive layers are trained to obtain the N first target branch adaptive layers, including: when the mth initial branch adaptive layer is trained for the tth time, the first sample data is input into the target encoding network and the preset channel for processing to obtain the second sample data.
  • the second sample data is input into the mth initial branch adaptive layer for processing to obtain the data processed by the mth initial branch adaptive layer, where m is a positive integer and m is not greater than N.
  • the data processed by the mth initial branch adaptive layer is input into the target decoding network for processing to obtain the third sample data.
  • the method further includes: sending second indication information, where the second indication information instructs the mth initial branch adaptive layer to perform training.
  • the initial sparse gating module is trained according to the target encoding network, the target decoding network, the first target channel feature extraction network and the N first target branch adaptive layers to obtain the target sparse gating module, including: training the initial sparse gating module multiple times to obtain the target sparse gating module.
  • the fourth sample data is input into the target encoding network and the preset channel for processing to obtain the fifth sample data.
  • the fifth sample data is input into the first target channel feature extraction network for processing to obtain channel distribution information training data and channel category information training data.
  • the channel distribution information training data and the channel category information training data are both input into the sparse gating module X t-1 for processing to obtain R first target branch adaptive layer samples and R first target branch adaptive layer weight samples corresponding to the channel distribution information training data, where R is a positive integer and R is not greater than N.
  • the fifth sample data is processed according to the R first target branch adaptive layer samples and the R first target branch adaptive layer weight samples to obtain processed training data.
  • the processed training data is input into the target decoding network for processing to obtain the sixth sample data.
  • the present application provides a communication device, comprising: a communication module, configured to receive a first reference signal; a processing module, configured to input the first reference signal into a first target model for processing, to obtain K first target branch adaptive layers in the first target model and the weights of the K first target branch adaptive layers, wherein the first target model comprises N first target branch adaptive layers, K is not greater than N, and K and N are both positive integers; the communication module is also configured to send first information, wherein the first information indicates the K first target branch adaptive layers and the weights of the K first target branch adaptive layers, and the K first target branch adaptive layers and the weights of the K first target branch adaptive layers are used by the transmitting end to process the coded data to be sent.
  • the communication module is also used to: receive first data, where the first data is obtained by the sending end processing the encoded data to be sent; the processing module is also used to input the first data into the K first target branch adaptive layers for processing to obtain processed data.
  • the communication module is further used to: send a first channel dictionary set, where the first channel dictionary set is used by the transmitting end to determine the correspondence between the N first target branch adaptive layers and the M second target branch adaptive layers of the transmitting end, where M is a positive integer.
  • the communication module is further used to: receive a second channel dictionary set; the processing module is further used to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set; the communication module is further used to send first indication information, and the first indication information is used to indicate the correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the first channel dictionary set includes a correspondence between a first target branch adaptation layer and a channel label.
  • the communication module is also used to: receive a second reference signal; send third information, the third information is the same as the first information, or the third information indicates that the transmitter sends the reference signal after a first time interval, and the third information is obtained based on the second reference signal; receive data sent by the transmitter within the first time.
  • the present application provides a communication device, including:
  • a communication module configured to receive a first reference signal
  • a processing module configured to input the first reference signal into a first target model for processing, and obtain K first target branch adaptive layers and weights of the K first target branch adaptive layers in the first target model, wherein the first target model includes N first target branch adaptive layers, K is not greater than N, and K and N are both positive integers;
  • the communication module is also used to send fourth information, which indicates the weights of K second target branch adaptive layers and the K second target branch adaptive layers.
  • the weights of the K second target branch adaptive layers and the K second target branch adaptive layers are used by the transmitting end to process the coded data to be sent, and the weights of the K second target branch adaptive layers and the K second target branch adaptive layers are determined according to the weights of the K first target branch adaptive layers and the K first target branch adaptive layers.
  • the communication module is further used to:
  • the processing module is further used to input the first data into the K first target branch adaptive layers for processing to obtain processed data.
  • the communication module is further used to:
  • First indication information is received, where the first indication information is used to indicate a corresponding relationship between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the communication module is further used to:
  • the processing module is further used to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptive layer in the second channel dictionary set and the first target branch adaptive layer in the first channel dictionary set.
  • the first channel dictionary set includes a correspondence between a first target branch adaptation layer and a channel label.
  • the communication module is further used to:
  • Sending fifth information where the fifth information is the same as the fourth information, or the fifth information instructs the transmitting end to send the reference signal after a first time interval, and the fifth information is obtained according to the second reference signal;
  • the present application provides a communication device, comprising: a communication module for sending a first reference signal; the communication module is also used to receive first information, the first information indicating K first target branch adaptive layers and the weights of the K first target branch adaptive layers in the first target model of the first reference signal receiving end, the first information is obtained based on the first reference signal, and K is a positive integer.
  • the device also includes a processing module, which is used to: determine, based on the first information, K second target branch adaptive layers corresponding to the K first target branch adaptive layers and the weights of the K second target branch adaptive layers; the processing module is also used to input the encoded data to be sent into the K second target branch adaptive layers of the second target model for processing to obtain first data; the communication module is also used to send the first data.
  • a processing module which is used to: determine, based on the first information, K second target branch adaptive layers corresponding to the K first target branch adaptive layers and the weights of the K second target branch adaptive layers; the processing module is also used to input the encoded data to be sent into the K second target branch adaptive layers of the second target model for processing to obtain first data; the communication module is also used to send the first data.
  • the processing module when M is less than K, is also used to update the third target model to obtain the second target model, so that the second target model includes the K second target branch adaptive layers, wherein the third target model includes M second target branch adaptive layers, and M is a positive integer.
  • the communication module is further used to: receive a first channel dictionary set; the processing module is further used to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the communication module is further used to: send a second channel dictionary set; receive first indication information, where the first indication information is used to indicate the correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the second channel dictionary set includes a correspondence between a second target branch adaptation layer and a channel label.
  • the communication module is also used to: send a second reference signal; receive fifth information, the fifth information is the same as the fourth information, or the fifth information indicates that the reference signal is sent after a first time interval, and the fifth information is obtained based on the second reference signal; and send data within the first time.
  • the present application provides a communication device, including:
  • a communication module configured to send a first reference signal
  • the communication module is further used to receive fourth information, where the fourth information indicates K second target branch adaptive layers and weights of the K second target branch adaptive layers, and the fourth information is obtained according to the first reference signal, where K is a positive integer.
  • the device further includes a processing module, configured to:
  • the coded data to be sent is input into the K second target branch adaptive layers of the second target model for processing to obtain first data;
  • the communication module is also used to send the first data.
  • the communication module is further used to:
  • the processing module is further used to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptive layer in the second channel dictionary set and the first target branch adaptive layer in the first channel dictionary set;
  • the communication module is further used to send first indication information, where the first indication information is used to indicate a corresponding relationship between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the communication module is further used to:
  • a second channel dictionary set is sent, where the second channel dictionary set is used by a reference signal receiving end to determine a correspondence between N first target branch adaptive layers of the reference signal receiving end and M second target branch adaptive layers of the transmitting end, where M and N are both positive integers.
  • the second channel dictionary set includes a correspondence between a second target branch adaptation layer and a channel label.
  • the communication module is further used to:
  • the fifth information is the same as the fourth information, or the fifth information indicates that the reference signal is sent after a first time interval, and the fifth information is obtained according to the second reference signal;
  • the data is sent within the first time.
  • the present application provides a communication device, including: a communication module, used to receive a first reference signal; a processing module, used to input the first reference signal into a first target model for processing, and obtain K first target branch adaptive layers in the first target model and the weights of the K first target branch adaptive layers, the first target model includes N first target branch adaptive layers, K is not greater than N, and K and N are both positive integers; the communication module is also used to receive first data; the processing module is also used to input the first data into the K first target branch adaptive layers for processing to obtain processed data, wherein the maximum weight among the weights of the K first target branch adaptive layers is not less than a preset value.
  • the first target model also includes a first target channel feature extraction network and a target sparse gating module.
  • the first target channel feature extraction network is used to process the first reference signal to obtain channel distribution information and channel category information corresponding to the channel distribution information;
  • the target sparse gating module is used to calculate the K first target branch adaptive layers and the weights of the K first target branch adaptive layers based on the channel distribution information and the channel category information.
  • the K first target branch adaptive layers are used to process the first data respectively to obtain the data processed by the K first target branch adaptive layers respectively; the weights of the K first target branch adaptive layers are used to perform weighted sum processing on the data processed by the K first target branch adaptive layers respectively to obtain the processed data.
  • the processing module is further used to: when a maximum weight among the weights of the K first target branch adaptive layers is less than the preset value, decode the first data to obtain processed data.
  • the communication module is also used to: send first information, where the first information indicates the K first target branch adaptive layers and the weights of the K first target branch adaptive layers, and the weights of the K first target branch adaptive layers and the K first target branch adaptive layers are used by the sending end to process the encoded data to be sent to obtain the first data.
  • the communication module is also used to: when the maximum weight among the weights of the K first target branch adaptive layers is less than the preset value, send second information instructing the sending end to directly send the first data.
  • the communication module is further used to: send a first channel dictionary set, where the first channel dictionary set is used by the transmitter to determine the correspondence between the N first target branch adaptive layers and the M second target branch adaptive layers of the transmitter, where M is a positive integer.
  • the communication module is further used to: receive a second channel dictionary set; the processing module is further used to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set; the communication module is further used to send first indication information, and the first indication information is used to indicate the correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the communication module is further used to:
  • Send fourth information the fourth information indicating the weights of K second target branch adaptive layers and the K second target branch adaptive layers, the weights of the K second target branch adaptive layers and the K second target branch adaptive layers are used by the transmitting end to process the coded data to be sent to obtain the first data, and the weights of the K second target branch adaptive layers and the K second target branch adaptive layers are determined according to the weights of the K first target branch adaptive layers and the K first target branch adaptive layers.
  • the communication module is further used to:
  • First indication information is received, where the first indication information is used to indicate a corresponding relationship between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the communication module is further used to:
  • the processing module is further used to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptive layer in the second channel dictionary set and the first target branch adaptive layer in the first channel dictionary set.
  • the first channel dictionary set includes a correspondence between a first target branch adaptation layer and a channel label.
  • the communication module is also used to: receive a second reference signal; send third information, the third information is the same as the first information, or the third information indicates that the sending end sends the reference signal after a first time interval, and the third information is obtained based on the second reference signal; receive data sent by the sending end within the first time.
  • the communication module is also used to: receive a second reference signal; send fifth information, the fifth information is the same as the fourth information, or the fifth information indicates that the sending end sends the reference signal after a first time interval, and the fifth information is obtained based on the second reference signal; receive data sent by the sending end within the first time.
  • the present application provides a communication device, comprising: a communication module, used to send a first reference signal; a processing module, used to, when receiving first information, indicate K first target branch adaptive layers and weights of the K first target branch adaptive layers in a first target model of a receiving end of the first reference signal, and determine, according to the first information, K second target branch adaptive layers corresponding to the K first target branch adaptive layers and weights of the K second target branch adaptive layers, wherein the first information is obtained based on the first reference signal, and K is a positive integer; the processing module is also used to input the coded data to be sent into the K second target branch adaptive layers of the second target model for processing to obtain first data; the communication module is also used to send the first data.
  • the processing module when M is less than K, is also used to update the third target model to obtain the second target model, so that the second target model includes the K second target branch adaptive layers, wherein the third target model includes M second target branch adaptive layers, and M is a positive integer.
  • the K second target branch adaptive layers are used to respectively process the encoded data to be sent to obtain the data respectively processed by the K second target branch adaptive layers; the weights of the K second target branch adaptive layers are used to perform weighted sum processing on the data respectively processed by the K second target branch adaptive layers to obtain the first data.
  • the communication module is further used to: when receiving second information, the second information indicates to directly send the coded data to be sent, then send the coded data to be sent, and the second information is obtained according to the first reference signal.
  • the communication module is further used to: receive a first channel dictionary set; the processing module is further used to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the communication module is further used to: send a second channel dictionary set; receive first indication information, where the first indication information is used to indicate the correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the second channel dictionary set includes a correspondence between a second target branch adaptation layer and a channel label.
  • the communication module is also used to: send a second reference signal; receive third information, which is the same as the first information; or, the third information is used to indicate that the reference signal is sent after a first time interval, and the third information is obtained based on the second reference signal; and send data within the first time.
  • the present application provides a communication device, including:
  • a communication module configured to send a first reference signal
  • a processing module configured to, when receiving fourth information, the fourth information indicating K second target branch adaptive layers and weights of the K second target branch adaptive layers, input the coded data to be sent into the K second target branch adaptive layers of the second target model for processing according to the fourth information, to obtain first data, the fourth information being obtained according to the first reference signal, and K being a positive integer;
  • the communication module is also used to send the first data.
  • the K second target branch adaptive layers are used to respectively process the encoded data to be sent to obtain the data respectively processed by the K second target branch adaptive layers; the weights of the K second target branch adaptive layers are used to perform weighted sum processing on the data respectively processed by the K second target branch adaptive layers to obtain the first data.
  • the communication module is further used to:
  • the second information indicates that the coded data to be sent is directly sent, then the coded data to be sent is sent, and the second information is obtained according to the first reference signal.
  • the communication module is further used to:
  • the processing module is further used to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptive layer in the second channel dictionary set and the first target branch adaptive layer in the first channel dictionary set;
  • the communication module is also used to send first indication information, where the first indication information is used to indicate the correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the communication module is further used to:
  • a second channel dictionary set is sent, where the second channel dictionary set is used by a reference signal receiving end to determine a correspondence between N first target branch adaptive layers of the reference signal receiving end and M second target branch adaptive layers of the transmitting end, where M and N are both positive integers.
  • the second channel dictionary set includes a correspondence between a second target branch adaptation layer and a channel label.
  • processing module is further configured to:
  • the fifth information is the same as the fourth information; or, the fifth information is used to indicate that the reference signal is sent after a first time interval, and the fifth information is obtained according to the second reference signal;
  • the data is sent within the first time.
  • an embodiment of the present application provides a target model training device.
  • the target model includes a target encoding network, a target decoding network and a first target model, and the first target model includes a first target channel feature extraction network, N first target branch adaptive layers and a target sparse gating module.
  • the device includes: a first training module for training the initial channel feature extraction network to obtain the first target channel feature extraction network and the first channel dictionary set; a second training module for training the N initial branch adaptive layers according to the target encoding network and the target decoding network to obtain the N first target branch adaptive layers, where N is the number of first target branch adaptive layers in the first channel dictionary set; a third training module for training the initial sparse gating module according to the target encoding network, the target decoding network, the first target channel feature extraction network and the N first target branch adaptive layers to obtain the target sparse gating module.
  • the first channel dictionary set includes a correspondence between a first target branch adaptation layer and a channel label.
  • the first training module is used to: train the initial channel feature extraction network multiple times to obtain the first target channel feature extraction network and the first channel dictionary set.
  • the initial channel feature extraction network is pre-trained based on labeled historical channel information or a preset channel model.
  • both the unlabeled channel information and the labeled historical channel information are input into the channel feature extraction network U t-1 for processing, and the channel distribution information corresponding to the unlabeled channel information and the labeled historical channel information is obtained.
  • N t-1 cluster centers are obtained.
  • a predicted channel dictionary set is obtained.
  • the second training module is used to: when the mth initial branch adaptive layer is trained for the tth time, the first sample data is input into the target coding network and the preset channel for processing to obtain the second sample data.
  • the second sample data is input into the mth initial branch adaptive layer for processing to obtain the data processed by the mth initial branch adaptive layer, where m is a positive integer and m is not greater than N.
  • the data processed by the mth initial branch adaptive layer is input into the target decoding network for processing to obtain the third sample data.
  • the device further includes a communication module, configured to send second indication information, where the second indication information instructs the mth initial branch adaptive layer to perform training.
  • the third training module is used to: perform multiple training on the initial sparse gating module to obtain the target sparse gating module.
  • the fourth sample data is input into the target encoding network and the preset channel for processing to obtain the fifth sample data.
  • the fifth sample data is input into the first target channel feature extraction network for processing to obtain channel distribution information training data and channel category information training data.
  • the channel distribution information training data and the channel category information training data are both input into the sparse gating module X t-1 for processing to obtain R first target branch adaptive layer samples and R first target branch adaptive layer weight samples corresponding to the channel distribution information training data, where R is a positive integer and R is not greater than N.
  • the fifth sample data is processed according to the R first target branch adaptive layer samples and the R first target branch adaptive layer weight samples to obtain processed training data.
  • the processed training data is input into the target decoding network for processing to obtain the sixth sample data.
  • the processing module may be a processor, and the communication module may be a transceiver module, a transceiver, or a communication interface. It is understandable that the communication module may be a transceiver in the device, for example, implemented by an antenna, a feeder, and a codec in the device, or, if the communication device is a chip provided in a device, the communication module may be an input/output interface of the chip, such as an input/output circuit, a pin, etc.
  • the first training module, the second training module and the third training module may all be processors.
  • the communication module may be a transceiver module, a transceiver or a communication interface. It is understandable that the communication module may be a transceiver in the device, for example, implemented by an antenna, a feeder and a codec in the device, or, if the communication device is a chip provided in the device, the communication module may be an input/output interface of the chip, for example, an input/output circuit, a pin, etc.
  • an embodiment of the present application provides a communication device, comprising one or more processors; wherein the one or more processors are used to execute computer programs stored in one or more memories, so that the communication device implements a method as described in any one of the first aspect, or, implements a method as described in any one of the second aspect, or, implements a method as described in any one of the third aspect, or, implements a method as described in any one of the fourth aspect, or, implements a method as described in any one of the fifth aspect, or, implements a method as described in any one of the sixth aspect, or, implements a method as described in any one of the seventh aspect.
  • the communication device further includes the one or more memories.
  • the communication device is a chip or a chip system.
  • an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, and when the instructions are executed by a processor, the method described in any one of aspect 1 is implemented, or the method described in any one of aspect 2 is implemented, or the method described in any one of aspect 3 is implemented, or the method described in any one of aspect 4 is implemented, or the method described in any one of aspect 5 is implemented, or the method described in any one of aspect 6 is implemented, or the method described in any one of aspect 7 is implemented.
  • an embodiment of the present application provides a computer program product, including a computer program.
  • the computer program When the computer program is executed, it implements the method as described in any one of the first aspect, or implements the method as described in any one of the second aspect, or implements the method as described in any one of the third aspect, or implements the method as described in any one of the fourth aspect, or implements the method as described in any one of the fifth aspect, or implements the method as described in any one of the sixth aspect, or implements the method as described in any one of the seventh aspect.
  • an embodiment of the present application provides a communication system, comprising an apparatus as described in any one of the ninth aspect or comprising an apparatus as described in any one of the tenth aspect, and also comprising an apparatus as described in any one of the twelfth aspect; or, the communication system comprises an apparatus as described in any one of the eleventh aspect, and an apparatus as described in any one of the thirteenth aspect; or, the communication system comprises an apparatus as described in any one of the fourteenth aspect, and an apparatus as described in any one of the fifteenth aspect or an apparatus as described in any one of the sixteenth aspect.
  • the apparatus described in the ninth aspect the apparatus described in the tenth aspect, the apparatus described in the eleventh aspect, the apparatus described in the twelfth aspect, the apparatus described in the thirteenth aspect, the apparatus described in the fourteenth aspect, the apparatus described in the fifteenth aspect, the apparatus described in the sixteenth aspect, the apparatus described in the seventeenth aspect, the computer storage medium described in the eighteenth aspect, or the computer program product described in the nineteenth aspect, and the system described in the twentieth aspect are all used to execute any method provided in the first aspect, any method provided in the second aspect, any method provided in the third aspect, any method provided in the fourth aspect, any method provided in the fifth aspect, any method provided in the sixth aspect, any method provided in the seventh aspect, or any method provided in the eighth aspect. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding method, which will not be repeated here.
  • FIG1a is a schematic diagram of the architecture of a communication system provided in an embodiment of the present application.
  • FIG1b is a schematic diagram of the architecture of a communication system provided in an embodiment of the present application.
  • FIG1c is a schematic diagram of the architecture of a communication system provided in an embodiment of the present application.
  • FIG2a is a flow chart of a communication method provided in an embodiment of the present application.
  • FIG2b is a schematic diagram of a target sparse gating module provided in an embodiment of the present application.
  • FIG3 is a schematic diagram of a communication system provided in an embodiment of the present application.
  • FIG4 is a flow chart of another communication method provided in an embodiment of the present application.
  • FIG5a is a schematic diagram of another communication system provided in an embodiment of the present application.
  • FIG5b is a schematic diagram of a frame structure provided in an embodiment of the present application.
  • FIG6a is a schematic diagram of a flow chart of a target model training method provided in an embodiment of the present application.
  • FIG6b is a schematic diagram of another frame structure provided in an embodiment of the present application.
  • FIG7a is a schematic diagram of the structure of a communication device provided in an embodiment of the present application.
  • FIG7b is a schematic diagram of the structure of another communication device provided in an embodiment of the present application.
  • FIG8 is a schematic diagram of the structure of another communication device provided in an embodiment of the present application.
  • the present application provides a communication method, device and system that can cope with dynamic channel changes and achieve migration adaptation of unknown channels.
  • FIG. 1a is a schematic diagram of a communication system applicable to the embodiment of the present application, and the system includes a network device 101 and a terminal 102.
  • the network device 101 may be a base station, an evolved NodeB (eNodeB), a transmission reception point (TRP), a next generation NodeB (gNB) in a fifth generation (5G) mobile communication system, a next generation NodeB in a sixth generation (6G) mobile communication system, a base station in a future mobile communication system, or an access node in a WiFi system, etc.; it may also be a module or unit that completes some functions of a base station, for example, a centralized unit (CU) or a distributed unit (DU).
  • the network device 101 may be a macro base station, a micro base station or an indoor station, a relay node or a donor node, etc.
  • the embodiments of the present application do not limit the specific technology and specific device form adopted by the network device 101. For ease of description, the following description takes a base station as an example of a network device.
  • the terminal 102 may also be referred to as a terminal device, user equipment (UE), mobile station, mobile terminal, etc.
  • the terminal can be widely used in various scenarios, for example, device-to-device (D2D), vehicle-to-everything (V2X) communication, machine-type communication (MTC), Internet of Things (IOT), virtual reality, augmented reality, industrial control, automatic driving, telemedicine, smart grid, smart furniture, smart office, smart wear, smart transportation, smart city, etc.
  • the terminal may be a mobile phone, a tablet computer, a computer with wireless transceiver function, a wearable device, a vehicle, a drone, a helicopter, an airplane, a ship, a robot, a mechanical arm, a smart home device, etc.
  • the embodiments of the present application do not limit the specific technology and specific device form adopted by the terminal.
  • Base stations and terminals can be fixed or movable. Base stations and terminals can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; they can also be deployed on the water surface; they can also be deployed on aircraft, balloons, and artificial satellites in the air. The embodiments of this application do not limit the application scenarios of base stations and terminals.
  • Base stations and terminals, base stations and base stations, and terminals and terminals can communicate through authorized spectrum, unauthorized spectrum, or both; they can communicate through spectrum below 6 gigahertz (GHz), spectrum above 6 GHz, or spectrum below 6 GHz and spectrum above 6 GHz.
  • GHz gigahertz
  • the embodiments of the present application do not limit the spectrum resources used for wireless communication.
  • the functions of the base station may also be performed by a module (such as a chip) in the base station, or by a control subsystem including the base station functions.
  • the control subsystem including the base station functions here may be a control center in the above-mentioned application scenarios such as smart grid, industrial control, and smart transportation.
  • the functions of the terminal may also be performed by a module (such as a chip or a modem) in the terminal, or by a device including the terminal functions.
  • a wireless communication system is usually composed of cells, each of which contains a base station (BS), which provides communication services to multiple mobile stations (MS).
  • the base station contains a baseband unit (BBU) and a remote radio unit (RRU).
  • BBU and RRU can be placed in different places, for example: RRU is remote and placed in an area with high traffic volume, and BBU is placed in a central computer room. BBU and RRU can also be placed in the same computer room. BBU and RRU can also be different components under one rack.
  • the wireless communication systems include but are not limited to: Narrow Band-Internet of Things (Narrow Band-Internet of Things)
  • the 5G networks include NB-IoT, Global System for Mobile Communications (GSM), Enhanced Data rate for GSM Evolution (EDGE), Wideband Code Division Multiple Access (WCDMA), Code Division Multiple Access 2000 (CDMA2000), Time Division-Synchronization Code Division Multiple Access (TD-SCDMA), Long Term Evolution (LTE), and the three application scenarios of the next generation 5G mobile communication system: Enhanced Mobile Broadband (eMBB), Ultra-reliable and Low Latency Communications (URLLC) and LTE-enhanced Machine Type Communication (eMTC).
  • GSM Global System for Mobile Communications
  • EDGE Enhanced Data rate for GSM Evolution
  • WCDMA Wideband Code Division Multiple Access
  • CDMA2000 Code Division Multiple Access 2000
  • TD-SCDMA Time Division-Synchronization Code Division Multiple Access
  • LTE Long Term Evolution
  • eMBB Enhanced Mobile Broadband
  • URLLC Ultra-reliable
  • Figure 1b is a schematic diagram of another communication system applicable to an embodiment of the present application, and the system includes a satellite base station 103 and a terminal 104.
  • the satellite base station 103 may be a drone, a hot air balloon, a low-orbit satellite, a medium-orbit satellite, or a high-orbit satellite. Alternatively, the satellite base station 103 may also refer to a non-ground base station or a non-ground device. The satellite base station 103 may be used as a network device or as a terminal device. The satellite base station 103 may not have the function of a base station, or may have some or all of the functions of a base station, which is not limited in this application.
  • terminal 104 For the introduction of the terminal 104 , please refer to the description of the terminal 102 in FIG. 1 a , which will not be repeated here.
  • the satellite base station 103 can provide communication services for the terminal 104.
  • the satellite base station 103 transmits downlink data to the terminal 104.
  • the data is encoded using channel coding.
  • the channel-coded data is transmitted to the terminal 104 after constellation modulation.
  • the terminal 104 transmits uplink data to the satellite base station 103.
  • the uplink data can also be encoded using channel coding.
  • the encoded data is transmitted to the satellite base station 103 after constellation modulation.
  • the communication method of the present application can also be applied to an inter-satellite communication system.
  • FIG1c is a schematic diagram of another communication system applicable to the embodiment of the present application, and the system includes a satellite 105 and a satellite 106.
  • the traditional satellite intersatellite link communication system can be divided into two parts: space beam acquisition, pointing and tracking (APT) subsystem and communication subsystem.
  • the communication subsystem is responsible for the transmission of intersatellite information and is the main body of the intersatellite communication system; the APT subsystem is responsible for the acquisition, pointing and tracking between satellites.
  • the APT subsystem In order to minimize the attenuation and interference effects in the channel and require high confidentiality and transmission rate, the APT subsystem must be adjusted in real time to continuously adapt to changes.
  • the existing APT systems are all optical systems, and the disadvantage is that optical alignment is difficult and requires mechanical adjustment of pointing.
  • Most of the existing communication subsystems are optical communication systems, and there are also some microwave band systems, which mostly use a single high-gain antenna.
  • the existing APT system and communication subsystem are independent systems.
  • the disadvantage is that optical communication is easily affected by vibration and the rate is unstable; the millimeter wave frequency is low, the communication capacity is low, and the antenna needs to be mechanically adjusted.
  • satellite 105 and satellite 106 both include a communication module, a transceiver antenna, an APT module, and an APT transmitter/receiver.
  • the communication method of the present application can be applied to the communication module.
  • the communication method of this solution can be deployed at the network device 101 end, or the satellite base station 103 end, or the satellite 105 end, and can also be deployed at the terminal 102, the terminal 104 or the satellite 106, etc., and this solution does not impose any restrictions on this.
  • a multi-branch adaptive layer is used to correspond to terminal networks under different channel distributions, thereby improving performance.
  • FIG. 2a it is a flow chart of a communication method provided by an embodiment of the present application.
  • the method can be applied to the aforementioned communication system, such as the communication system shown in Figure 1a.
  • the communication method shown in Figure 2a may include steps 201-206.
  • steps 201-206 It should be understood that, for the convenience of description, the present application is described in the order of 201-206, and is not intended to be limited to execution in the above order.
  • the embodiment of the present application does not limit the order of execution, execution time, number of executions, etc. of the above one or more steps.
  • the following description is based on the example that the execution subjects of steps 201 and 204 of the communication method are base stations, and the execution subjects of 202, 203, 205 and 206 are terminals. This application is also applicable to other execution subjects.
  • Steps 201-206 are as follows:
  • a base station sends a first reference signal
  • the first reference signal may be a pilot signal.
  • the pilot signal is a direct sequence spread spectrum signal that is continuously transmitted by the base station without modulation.
  • the pilot signal enables the terminal to obtain the forward code division multiple access channel time limit, provide a related demodulation phase reference, etc.
  • the terminal receives the first reference signal
  • the terminal receives the first reference signal from the base station.
  • the terminal inputs the first reference signal into a first target model for processing, and obtains K first target branch adaptive layers and weights of the K first target branch adaptive layers in the first target model, where the first target model includes N first target branch adaptive layers, K is not greater than N, and K and N are both positive integers;
  • each first target branch adaptive layer can be understood as corresponding to a different type of channel distribution, that is, N first target branch adaptive layers correspond to N types of channel distribution.
  • N first target branch adaptive layers correspond to N' types of clustered channel distributions, where N is greater than N'. That is, at least one first target branch adaptive layer may correspond to a type of clustered channel distribution.
  • the clustering can be understood as merging channel distributions with similar or identical channel distributions into one type of channel distribution.
  • K first target branch adaptive layers among the N first target branch adaptive layers of the first target model and the weights of the K first target branch adaptive layers can be determined. That is, the current channel distribution can be comprehensively characterized by the types of channel distributions corresponding to the determined K first target branch adaptive layers and their corresponding weights.
  • the first target model further includes a first target channel feature extraction network and a target sparse gating module.
  • the first target channel feature extraction network is used to process the first reference signal to obtain channel distribution information and channel category information corresponding to the channel distribution information.
  • the channel distribution information may be a feature vector used to characterize the current channel distribution in the feature space.
  • the channel category information corresponding to the channel distribution information may be a probability of predicting the channel type to which the current channel distribution belongs.
  • the target sparse gating module is used to calculate the K first target branch adaptive layers and the weights of the K first target branch adaptive layers according to the channel distribution information and the channel category information.
  • the target sparse gating module can derive the type corresponding to the current channel distribution based on the above channel distribution information and channel category information, which is comprehensively represented by K first target branch adaptive layers and their corresponding weights.
  • the target sparse gating module may calculate the weights of each first target branch adaptive layer based on the following method:
  • softmax is a normalized exponential function
  • WSG is a parameter of the target sparse gating module.
  • h is the channel distribution information.
  • pcluster is the channel category information corresponding to the channel distribution information.
  • is the weight of the branch adaptive layer.
  • the K first target branch adaptive layers are used to process the first data respectively to obtain data processed by the K first target branch adaptive layers respectively.
  • each first target branch adaptive layer is to perform an affine transformation on the first data.
  • the first data is f(x) ⁇ C 72 ⁇ 14 ⁇ 1 ⁇ 1
  • the dimension represents a complex domain of 72 subcarriers, 14 symbols, 1 antenna at the receiving end, and 1 antenna at the transmitting end.
  • Each first target branch adaptive layer corresponds to a set of parameters (W i ,b i ), W ⁇ C 72 ⁇ 14 ⁇ 1 ⁇ 1 , b ⁇ C 72 ⁇ 14 ⁇ 1 ⁇ 1 .
  • i represents the label of the K first target branch adaptive layers, and i is a positive integer.
  • each first target branch adaptive layer may also be other processing, and this solution does not impose strict restrictions on this.
  • the weights of the K first target branch adaptive layers are used to perform weighted sum processing on the data processed by the K first target branch adaptive layers respectively to obtain the processed data.
  • the weighted summation process can be expressed as ⁇ i represents the weights of the K first target branch adaptive layers.
  • the base station sends first data
  • the terminal receives first data
  • the terminal receives first data from the base station.
  • the terminal inputs the first data into the K first target branch adaptive layers for processing to obtain processed data, wherein a maximum weight among the weights of the K first target branch adaptive layers is not less than a preset value.
  • the largest weight value is found.
  • the maximum weight value is not less than the preset value, it indicates that the obtained K first target branch adaptive layers have high reliability. Therefore, the received first data is input into the K first target branch adaptive layers for processing.
  • the preset value can be any positive number not greater than 1. Of course, it can also be 0, which is not limited in this solution. Optionally, the preset value is 0.8.
  • the processing may be to perform an affine transformation on the input first data.
  • the target sparse gating module may also include multiple sparse gating submodules.
  • branch level A is trained for different delay spreads (e.g., branch A1: 30ns; branch A2: 100ns)
  • branch level B is trained for different speeds (e.g., branch B1: 3km/s; branch B2: 30km/s).
  • the final weighted summation process can be expressed as: t Bi and t Ai represent the affine transformation operations corresponding to each branch adaptive layer.
  • the value range of i is an integer from 1 to N.
  • ⁇ Ai is the weight of each branch adaptive layer of branch level A
  • ⁇ Bi is the weight of each branch adaptive layer of branch level B.
  • multiple sparse gating submodules are combined to achieve a multi-level dynamic characterization of channel distribution.
  • a nonlinear channel distribution characterization is obtained, which improves the generalization migration ability.
  • the processed data may be data after decoding the data T.
  • the decoded data may be obtained by inputting the data T into a target decoding network for decoding.
  • the first target model includes a first target channel feature extraction network, a target sparse gating module and N first target branch adaptive layers.
  • the terminal inputs the received first reference signal into the first target channel feature extraction network for processing to obtain channel distribution information and channel category information. Then, the channel distribution information and channel category information are input into the target sparse gating module for processing to obtain K first target branch adaptive layers among the N first target branch adaptive layers and the weights of the K first target branch adaptive layers.
  • the terminal When the maximum weight among the weights of the K first target branch adaptive layers is not less than the preset value, the terminal inputs the received first data into the K first target branch adaptive layers for processing, and performs weighted sum calculation, and then inputs the obtained data T into the target decoding network for decoding processing, that is, the above-mentioned processed data is obtained.
  • the terminal when the maximum weight among the weights of the K first target branch adaptive layers is less than a preset value, the terminal directly decodes the first data to obtain processed data.
  • the terminal directly decodes the received first data without being processed by the K first target branch adaptive layers.
  • the current channel data may be saved and used as training data to retrain the first target model to obtain a model with better performance.
  • the introduction to the training of the first target model can be found in the subsequent model training section, which will not be repeated here.
  • this embodiment is described by taking the base station as the transmitting end and the terminal as the receiving end as an example. It can also be the terminal as the transmitting end and the base station as the receiving end, and the processing process is the same as the above, which will not be repeated here.
  • the terminal determines K first target branch adaptive layers and the weights of the K first target branch adaptive layers among the N first target branch adaptive layers of the first target model based on the received reference signal.
  • the maximum weight among the weights of the K first target branch adaptive layers is not less than a preset value
  • the received first data is input into the K first target branch adaptive layers for processing to obtain processed data.
  • the actual channel distribution is characterized based on the determined K first target branch adaptive layers and K weights to cope with the dynamic changes of the channel, which can improve the system performance.
  • FIG. 4 it is a flow chart of another communication method provided by an embodiment of the present application.
  • the method can be applied to the aforementioned communication system, such as the communication system shown in Figure 1a.
  • the communication method shown in Figure 4 may include steps 401-409. It should be understood that for the convenience of description, the present application is described in the order of 401-409, and is not intended to be limited to execution in the above order. The embodiment of the present application does not limit the order of execution, execution time, number of executions, etc. of the above one or more steps. The following description is based on the example that the execution subject of steps 401, 405, 406 and 407 of the communication method is a base station, and the execution subject of steps 402-404, 408 and 409 is a terminal. This application is also applicable to other execution subjects. Steps 401-409 are as follows:
  • a base station sends a first reference signal
  • step 201 in the embodiment shown in FIG. 2 a , which will not be repeated here.
  • the terminal receives a first reference signal
  • step 202 for the introduction of this step, please refer to the description of step 202 in the embodiment shown in FIG. 2 a , which will not be repeated here.
  • the terminal inputs the first reference signal into a first target model for processing, and obtains K first target branch adaptive layers and weights of the K first target branch adaptive layers in the first target model, where the first target model includes N first target branch adaptive layers, K is not greater than N, and K and N are both positive integers;
  • step 203 in the embodiment shown in FIG. 2 a , which will not be repeated here.
  • the terminal sends first information, where the first information indicates the weights of the K first target branch adaptive layers and the K first target branch adaptive layers, and the weights of the K first target branch adaptive layers and the K first target branch adaptive layers are used by the base station to process the coded data to be sent to obtain first data;
  • the terminal when the maximum weight among the weights of the K first target branch adaptive layers is not less than the preset value, the terminal sends information to the base station so that the base station side processes the coded data to be sent before sending it. In this way, the base station side selects the optimal processing method under the corresponding channel distribution to process the coded data according to the information indication, which can improve the communication performance.
  • the embodiment of the present application is only introduced by taking the case where the maximum weight among the weights of the K first target branch adaptive layers is not less than the preset value and the terminal sends the first information as an example. It can also be that the terminal sends the first information under any condition. For example, if the preset value is 0, the terminal sends the first information. This solution does not impose strict restrictions on this.
  • the base station receives first information, and determines, according to the first information, K second target branch adaptive layers corresponding to the K first target branch adaptive layers and weights of the K second target branch adaptive layers;
  • a second target model is deployed on the base station side.
  • FIG5a a schematic diagram of another communication system provided in an embodiment of the present application is shown.
  • the second target model includes M second target branch adaptive layers.
  • the first target model includes the above-mentioned first target channel feature extraction network, a target sparse gating module and N first target branch adaptive layers.
  • the second target model includes the K second target branch adaptive layers, and the K second target branch adaptive layers correspond to the K first target branch adaptive layers.
  • the terminal side characterizes the current channel distribution based on the aforementioned K first target branch adaptive layers.
  • the terminal side feeds back the result to the base station side so that the base station side determines K second target branch adaptive layers that characterize the current channel distribution so as to process the coded data to be sent.
  • the following introduces a manner in which the base station side determines K second target branch adaptation layers corresponding to the K first target branch adaptation layers.
  • Example 1 The terminal sends a first channel dictionary set to the base station.
  • the base station receives the first channel dictionary set, and performs dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the base station can determine the K second target branch adaptive layers corresponding to the K first target branch adaptive layers based on the corresponding relationship, and then determine the weights of the K second target branch adaptive layers based on the weights of the K first target branch adaptive layers.
  • Example 2 The base station sends a second channel dictionary set to the terminal.
  • the terminal receives the second channel dictionary set, and performs dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the terminal sends first indication information to the base station.
  • the first indication information is used to indicate a corresponding relationship between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the base station can determine the K second target branch adaptation layers corresponding to the K first target branch adaptation layers and the weights of the K second target branch adaptation layers.
  • the first channel dictionary set includes a correspondence between a first target branch adaptation layer and a channel label.
  • the second channel dictionary set includes a correspondence between a second target branch adaptation layer and a channel label.
  • the corresponding relationship between the branch adaptation layers can be determined based on the same channel label.
  • the terminal may obtain a first channel dictionary set based on a first target channel feature extraction network in the first target model.
  • the second target model also includes a second target channel feature extraction network.
  • the base station may obtain a second channel dictionary set based on a second target channel feature extraction network in the second target model.
  • the above dictionary alignment operation is performed because the training of the channel feature extraction network is only related to the channel training set, and the transmitting and receiving ends can train the module independently. Since the channel training sets on both sides may not be exactly the same, the obtained channel dictionary sets may be different, so the matching of the adaptive branch layers at each end is achieved by dictionary alignment.
  • the sampled data ⁇ H, Y label ⁇ of the tapped delay line (TDL) or clustered delay line (CDL) channel model in the 3rd Generation Partnership Project (3GPP) protocol with labels is used as the channel data set for network training.
  • H represents the sample of the channel model
  • Y label represents the label of the channel model.
  • N is the total number of different channel labels involved in training.
  • the two ends train the channel feature extraction network based on the loss function to obtain different channel dictionary sets.
  • D UE ⁇ (1: TDL-A30v3, TDL-C30v3), (2: TDL-A300v100) ⁇ .
  • D BS ⁇ (a: CDL-A30v100), (b: TDL-A30v3, TDL-C30v3), (c: TDL-A300v100, TDL-C300v100) ⁇ .
  • TDL-A30v3, TDL-C30v3, CDL-A30v100, etc. are all channel labels.
  • TDL-A30v3, TDL-C30v3 means that the channel category (that is, the branch adaptive layer) corresponding to the channel with the channel label TDL-A30v3, TDL-C30v3 is numbered 1.
  • a:CDL-A30v100 indicates that the channel category (i.e., branch adaptation layer) corresponding to the channel with the channel label CDL-A30v100 is labeled a.
  • a, b, c, and the aforementioned 1 and 2 represent the labels of different channel categories.
  • TDL-A30v3 represents the TDL-A channel model with a delay spread of 30ns and a speed of 3km/s.
  • the terminal sends its own channel dictionary set D UE . If the channel labels in the channel dictionary set do not belong to the TDL or CDL channel model in the protocol (such as a model built by custom historical channel information), the terminal needs to additionally send the channel data h p closest to the cluster center point so that the base station can redefine the channel distribution information.
  • the base station receives the channel dictionary set D UE and matches it with the channel dictionary set D BS obtained by itself to obtain a channel matching set D m .
  • the channel matching set D m ⁇ (b:1),(c:2) ⁇ . That is, the channel adaptation layer 1 on the terminal side corresponds to the channel adaptation layer b on the base station side.
  • the channel adaptation layer 2 on the terminal side corresponds to the channel adaptation layer c on the base station side.
  • the base station can obtain channel distribution information by inputting its own channel feature extraction network, thereby determining the channel label to which it belongs. If there are some unmatched situations, it means that the multi-branch adaptive layers at both ends of the channel cannot be completely matched, and the model needs to be retrained to obtain a matching channel dictionary set.
  • the base station may also feed back the channel matching set Dm to the terminal for use in deduction.
  • the base station updates the third target model to obtain the second target model, so that the second target model includes the K second target branch adaptation layers, and the third target model includes M second target branch adaptation layers.
  • the third target model is updated to obtain the second target model.
  • the second target model can meet the condition of including the K second target branch adaptive layers.
  • the base station can determine K second target branch adaptation layers corresponding to the K first target branch adaptation layers.
  • the K second target branch adaptive layers are used to process the encoded data to be sent respectively to obtain data processed by the K second target branch adaptive layers respectively.
  • each second target branch adaptive layer is to perform an affine transformation on the coded data to be sent.
  • the coded data to be sent is f(x) ⁇ C 72 ⁇ 14 ⁇ 1 ⁇ 1
  • the dimension represents a complex domain of 72 subcarriers, 14 symbols, 1 antenna at the receiving end, and 1 antenna at the transmitting end.
  • Each second target branch adaptive layer corresponds to a set of parameters (W j , b j ), W ⁇ C 72 ⁇ 14 ⁇ 1 ⁇ 1 , b ⁇ C 72 ⁇ 14 ⁇ 1 ⁇ 1 .
  • j represents the label of the K second target branch adaptive layers, and j is a positive integer.
  • each second target branch adaptive layer may also be other processing, and this solution does not impose strict restrictions on this.
  • the weights of the K second target branch adaptive layers are used to perform weighted sum processing on the data processed respectively by the K second target branch adaptive layers to obtain first data.
  • the weighted summation process can be expressed as ⁇ j represents the weights of the K second target branch adaptive layers.
  • the base station inputs the coded data to be sent into the K second target branch adaptive layers of the second target model for processing to obtain first data.
  • step 405 For the introduction of this step, please refer to the record of step 405, which will not be repeated here.
  • the base station sends first data
  • the base station sends the data processed by the K second target branch adaptive layers to the terminal.
  • the terminal receives first data
  • the terminal inputs the first data into the K first target branch adaptive layers for processing to obtain processed data.
  • step 206 for the introduction of this step, please refer to the description of step 206 in the embodiment shown in FIG. 2 a , which will not be repeated here.
  • the terminal when the maximum weight among the weights of the K first target branch adaptive layers is less than the preset value, the terminal sends second information, where the second information instructs the base station to directly send the first data.
  • the base station does not need to process the coded data (first data) to be transmitted based on the second target branch adaptive layer. Accordingly, the terminal directly decodes and processes the received first data without going through the first target branch adaptive layer for processing.
  • the base station further sends a second reference signal.
  • the terminal receives the second reference signal. Then, the terminal sends third information based on the second reference signal.
  • the third information is the same as the first information. Alternatively, the third information indicates that the base station sends the reference signal after a first time interval.
  • the terminal determines based on the second reference signal that the distribution of the current channel is consistent with the distribution of the channel corresponding to the first reference signal, that is, the channel distribution has not changed relatively during the period from the first reference signal to the second reference signal.
  • the terminal instructs the base station to send the reference signal again after a period of time.
  • time slot #1 the transmission of the reference signal is reduced in the next time slot (time slot #1). For example, there is no reference signal in time slot #1.
  • the base station After receiving the third information, the base station sends data within the first time.
  • the sending of data within the first time may be understood as the base station no longer sending a reference signal within the first time.
  • the terminal receives the data sent within the first time. That is, the terminal no longer receives the reference signal within the first time.
  • the base station sends a second reference signal.
  • the terminal receives the second reference signal.
  • the terminal sends third information based on the second reference signal.
  • the third information is different from the first information. That is, the current channel distribution changes. Therefore, the terminal needs to feed back the newly determined first target branch adaptive layers and their weights to the base station.
  • the channel distribution information changes, for example, the weight changes, as shown in FIG5b, it is sent in the frame structure of time slot #0.
  • the feedback dimension of the information in the embodiment of the present application is related to the number N of branches in the first target branch adaptation layer, the preset value ⁇ , and the TopK value of the target sparse gating module.
  • the quantization loss is set to ⁇ . By default, the maximum weight value is sent first.
  • the remaining weight calculation formula is:
  • the weight of the last branch calculated by the weights of the first K-1 branches is B1 is the decimal representation of the maximum weight value.
  • B i is the decimal representation of the i-th weight value.
  • the number of feedback bits required for the feedback table can be calculated as: K*N index +(K-1)*N quan . This can reduce the feedback information overhead.
  • the above parameters can be configured in the channel dictionary set so that the synchronization of the transmitting and receiving ends can be achieved.
  • Branch number (3 bits) Weight (3 bits) Branch 1 (001) 0.9(100) Branch 2 (010) 0.05(010) Branch 4 (100) 0.05(/)
  • the transmitter can obtain channel estimation information through channel reciprocity, and similarly can directly estimate the branch weight through the channel feature extraction network and sparse gating module.
  • the transmitter can feedback the synchronization to the receiver through the above design to ensure the consistency of the branches at both ends.
  • the performance of the target encoding network and the target decoding network trained in the mixed channel under high signal-to-noise ratio can be improved in a specific channel.
  • the combination of clustered channels through the target sparse gating module can improve performance.
  • this scheme focuses on the expression of channel distribution categories rather than the expression of single-frame channel feedback. Through experiments, it is found that the adaptive adjustment of the channel in this scheme is robust to the change of feedback delay and the output dimension of the channel distribution information.
  • this embodiment is described by taking the base station as the transmitting end and the terminal as the receiving end as an example. It can also be the terminal as the transmitting end and the base station as the receiving end, and the processing process is the same as the above, which will not be repeated here.
  • This embodiment is introduced by taking an example in which a reference signal receiving end sends first information, where the first information indicates K first target branch adaptive layers and weights of the K first target branch adaptive layers.
  • the reference signal receiving end may also send the fourth information.
  • the fourth information indicates the weights of the K second target branch adaptive layers and the K second target branch adaptive layers. The weights of the K second target branch adaptive layers and the K second target branch adaptive layers are used by the transmitting end to process the coded data to be sent.
  • the sending end can directly process the encoded data to be sent according to the information.
  • the weights of the K second target branch adaptive layers and the K second target branch adaptive layers are determined according to the weights of the K first target branch adaptive layers and the K first target branch adaptive layers.
  • the reference signal receiving end sends a first channel dictionary set, where the first channel dictionary set is used by the transmitting end to determine a correspondence between N first target branch adaptation layers and M second target branch adaptation layers of the transmitting end, where M is a positive integer.
  • the transmitting end receives the first channel dictionary set, and performs dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the transmitting end also sends first indication information, where the first indication information is used to indicate a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the reference signal receiving end determines, based on the received first indication information, K second target branch adaptive layers corresponding to the K first target branch adaptive layers, and weights of the K second target branch adaptive layers.
  • the sending end sends a second channel dictionary set.
  • the reference signal receiving end receives the second channel dictionary set. Then, the first channel dictionary set and the second channel dictionary set are dictionary-aligned to obtain a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the reference signal receiving end determines K second target branch adaptive layers corresponding to the K first target branch adaptive layers, and weights of the K second target branch adaptive layers.
  • the base station further sends a second reference signal.
  • the terminal receives the second reference signal. Then, the terminal sends fifth information based on the second reference signal.
  • the fifth information is the same as the fourth information. Alternatively, the fifth information indicates that the base station sends the reference signal after a first time interval.
  • the terminal determines based on the second reference signal that the distribution of the current channel is consistent with the distribution of the channel corresponding to the first reference signal, that is, the channel distribution has not changed relatively during the period from the first reference signal to the second reference signal.
  • the terminal instructs the base station to send the reference signal again after a period of time.
  • the terminal determines K first target branch adaptive layers and the weights of the K first target branch adaptive layers among the N first target branch adaptive layers of the first target model based on the received reference signal.
  • the terminal sends information to the base station so that the base station inputs the coded data to be sent into the K second target branch adaptive layers for processing based on the information, and sends the processed first data.
  • the terminal obtains the processed data by inputting the received first data into the K first target branch adaptive layers for processing.
  • the terminal characterizes the actual channel distribution based on the determined K first target branch adaptive layers and K weights to cope with the dynamic changes of the channel, which can improve the system performance. Moreover, the terminal feeds back the K first target branch adaptive layers and K weights corresponding to the actual channel distribution to the base station so that the base station selects the optimal processing method under the corresponding channel distribution to process the coded data according to the information indication, which can improve the communication performance.
  • the communication method in the embodiment of the present application is introduced above.
  • the following is an introduction to the training method of the target model in the embodiment of the present application.
  • FIG 6a it is a flow chart of a target model training method provided in an embodiment of the present application.
  • the training method of the target model shown in Figure 6a may include steps 601-603. It should be understood that this application is described in the order of 601-603 for the convenience of description, and is not intended to be limited to execution in the above order.
  • the embodiment of the present application does not limit the order of execution, execution time, number of executions, etc. of the above one or more steps. Steps 601-603 are as follows:
  • step 601 may include:
  • the initial channel feature extraction network is trained multiple times to obtain the first target channel feature extraction network and the first channel dictionary set.
  • the initial channel feature extraction network is obtained by pre-training based on labeled historical channel information or a preset channel model.
  • the preset model can be a TDL model or CDL channel model in the 3GPP protocol. For example, sampling is performed from a training set (historical channel information with labels), input into a channel feature extraction network, the probability of the sample belonging to each channel category is output, and then the network parameters are trained by calculating the cross entropy loss function with the label to obtain the initial channel feature extraction network.
  • a training set historical channel information with labels
  • the probability of the sample belonging to each channel category is output
  • the network parameters are trained by calculating the cross entropy loss function with the label to obtain the initial channel feature extraction network.
  • the channel feature extraction network U t-1 when the channel feature extraction network U t-1 is trained for the tth time, the unlabeled channel information and the labeled historical channel information are both input into the channel feature extraction network U t-1 for processing to obtain the channel distribution information corresponding to the unlabeled channel information and the labeled historical channel information respectively.
  • N t-1 cluster centers are obtained according to the channel distribution information.
  • the above cluster centers are obtained by clustering using K-means clustering algorithm (K-means) or subspace K-means method.
  • a predicted channel dictionary set is obtained according to the N t-1 cluster centers and the channel distribution information.
  • the Euclidean distance between any two cluster centers and the channel distribution information is calculated, and based on the Euclidean distance, the most likely channel type label is inferred using the existing channel label.
  • a predicted channel dictionary set can be obtained.
  • the channel feature extraction network may include an extractor network and a classifier network.
  • the input of the extractor network is reference information, and of course it can also be a channel estimation result Hest .
  • Hest can be a least squares (LS) channel estimation or a minimum mean square error (MMSE) channel estimation based on the reference information, or it can be a channel estimation result obtained by a receiving neural network.
  • the output of the extractor network is the channel distribution information h, which is used to characterize the feature vector of the current channel distribution in the feature space.
  • the input of the classifier network is the channel distribution information h, and the output is the channel category information p cluster corresponding to the channel distribution information.
  • the channel category information is represented based on a channel dictionary set.
  • the loss function L1 can be expressed as:
  • ⁇ [0,1], ⁇ is a hyperparameter used to adjust the influence ratio of the two loss functions L classify and L cluster ;
  • L classify is the cross entropy loss function;
  • L cluster is the clustering loss function;
  • Q) represents the (Kullback-Leibler, KL) divergence between the auxiliary target distribution P and the soft cluster distribution Q of the sample;
  • qij represents the probability that sample i belongs to cluster j
  • hi is the channel distribution information of sample i output by the extractor network, ⁇ j is the centroid of the jth cluster, j′ is 1, 2...N, N is the number of cluster centers;
  • pij represents the auxiliary target distribution; is 1, 2...M, where M is the number of samples.
  • N is the number of first target branch adaptive layers in the first channel dictionary set
  • step 602 may include:
  • the first sample data is input into the target coding network and the preset channel for processing to obtain the second sample data.
  • the second sample data is input into the mth initial branch adaptive layer for processing to obtain data processed by the mth initial branch adaptive layer, where m is a positive integer and is not greater than N.
  • the data obtained by processing the m-th initial branch adaptive layer is input into the target decoding network for processing to obtain third sample data.
  • a second loss value is calculated based on the first sample data and the third sample data, and the mth initial branch adaptive layer parameter is adjusted based on the second loss value.
  • the method further includes:
  • the transmitting end sends the instruction information to the receiving end to instruct it to train the mth branch. Then, the transmitting end sends the training data to the receiving end.
  • the receiving end sends the indication information to the transmitting end to inform the receiving end that the transmitting end wants to train the mth branch, so that the transmitting end can send training data to the receiving end.
  • a large amount of data sets are required for training the multi-branch adaptive layer to characterize the characteristics of the channel distribution.
  • a pilot reference signal
  • the channel dictionary set number can be additionally indicated in the frame header. This is done to indicate the adaptive layer trained in the frame.
  • the initial sparse gating module is trained based on the trained target encoding network, target decoding network, first target channel feature extraction network and N first target branch adaptive layers.
  • the target model includes the target encoding network, the target decoding network and the first target model, and the first target model includes a first target channel feature extraction network, N first target branch adaptive layers and a target sparse gating module.
  • step 603 may include:
  • the initial sparse gating module is trained multiple times to obtain the target sparse gating module.
  • the fourth sample data is input into the target coding network and the preset channel for processing to obtain the fifth sample data;
  • the loss function L2 can be expressed as:
  • L llr is the loss function for joint training of the target encoding network and the target decoding network.
  • is the weight of the branch adaptive layer
  • Entropy( ⁇ ) represents the entropy of the branch adaptive layer weight
  • is a hyperparameter not less than 0.
  • the target model also includes a second target model, and the second target model includes a second target channel feature extraction network and M second target branch adaptive layers.
  • the training process for the second target channel feature extraction network can refer to the record of step 601, which will not be repeated here.
  • the training of the M second target branch adaptive layers may be obtained by training the M initial branch adaptive layers and the aforementioned N initial branch adaptive layers together for multiple times, wherein:
  • the seventh sample data is input into the target coding network for processing to obtain the eighth sample data.
  • the eighth sample data is input into the m'th initial branch adaptive layer for processing to obtain data processed by the m'th initial branch adaptive layer, where m' is a positive integer and m' is not greater than M.
  • the data processed by the m'th initial branch adaptive layer is input into the preset channel, the m'th initial branch adaptive layer and the target decoding network for processing to obtain the ninth sample data.
  • the m'th initial branch adaptive layer in the second target model corresponds to the m'th initial branch adaptive layer in the first target model.
  • the first channel dictionary set and the second channel dictionary set can be obtained respectively by training based on the aforementioned channel feature extraction network.
  • the first channel dictionary set corresponds to the first target model
  • the second channel dictionary set corresponds to the second target model. In this way, during training, the two models can align the adaptive layers and then train together.
  • a fourth loss value is calculated based on the seventh sample data and the ninth sample data, and the m'th initial branch adaptive layer parameters and the mth initial branch adaptive layer are adjusted based on the fourth loss value.
  • M N.
  • M and N may also be in other relationships, which is not limited in this solution.
  • this scheme adjusts the output of the branch adaptive layer by training the weights of the branch adaptive layer instead of training the parameters of the target encoding network and the target decoding network, which can reduce the overhead of network training.
  • the above target model of the embodiment of the present application is introduced by taking multiple network modules as an example. It should be noted that the multiple network modules can be independent network modules, or they can be integrated into one, or some network modules can be integrated into one, etc. This solution does not impose strict restrictions on this.
  • the embodiment of the present application is described by taking the first target model (including the first target channel feature extraction network, N first target branch adaptive layers, and the target sparse gating module) deployed at the receiving end and the second target model (including the second target channel feature extraction network and M second target branch adaptive layers) deployed at the transmitting end as an example. It can also be that the first target model is deployed at the transmitting end and the second target model is deployed at the receiving end. The embodiment of the present application can also deploy only the first target model, etc. This solution does not impose strict restrictions on this.
  • the division of multiple units or modules is only a logical division based on function, and is not a limitation on the specific structure of the device.
  • some functional modules may be subdivided into more small functional modules, and some functional modules may also be combined into one functional module, but no matter whether these functional modules are subdivided or combined, the general process performed by the device is the same.
  • some devices contain a receiving unit and a sending unit.
  • the sending unit and the receiving unit can also be integrated into a communication unit, which can implement the functions implemented by the receiving unit and the sending unit.
  • each unit corresponds to its own program code (or program instruction), and when the program code corresponding to each of these units is run on the processor, the unit is controlled by the processing unit to execute the corresponding process to implement the corresponding function.
  • the embodiments of the present application also provide a device for implementing any of the above methods, for example, providing a communication device including a module (or means) for implementing each step performed by the terminal in any of the above methods.
  • a communication device including a module (or means) for implementing each step performed by the terminal in any of the above methods.
  • another communication device is also provided, including a module (or means) for implementing each step performed by the base station in any of the above methods.
  • FIG. 7a it is a schematic diagram of the structure of a communication device provided in an embodiment of the present application.
  • the communication device is used to implement the aforementioned communication method, such as the communication method shown in Fig. 2a and Fig. 4.
  • the device may include a communication module 701 and a processing module 702, as follows:
  • the communication module 701 is configured to receive a first reference signal
  • a processing module 702 is used to input the first reference signal into a first target model for processing, to obtain K first target branch adaptive layers in the first target model and weights of the K first target branch adaptive layers, wherein the first target model includes N first target branch adaptive layers, K is not greater than N, and K and N are both positive integers;
  • the communication module 701 is further configured to receive first data
  • the processing module 702 is further used to input the first data into the K first target branch adaptive layers for processing to obtain processed data, wherein the maximum weight among the weights of the K first target branch adaptive layers is not less than a preset value.
  • the first target model also includes a first target channel feature extraction network and a target sparse gating module, wherein the first target channel feature extraction network is used to process the first reference signal to obtain channel distribution information and channel category information corresponding to the channel distribution information; the target sparse gating module is used to calculate the K first target branch adaptive layers and the weights of the K first target branch adaptive layers based on the channel distribution information and the channel category information.
  • the K first target branch adaptive layers are used to process the first data respectively to obtain the data processed by the K first target branch adaptive layers respectively; the weights of the K first target branch adaptive layers are used to perform weighted sum processing on the data processed by the K first target branch adaptive layers respectively to obtain the processed data.
  • the processing module 702 is further used to: when a maximum weight among the weights of the K first target branch adaptive layers is less than the preset value, decode the first data to obtain processed data.
  • the communication module 701 is also used to: send first information, where the first information indicates the K first target branch adaptive layers and the weights of the K first target branch adaptive layers, and the weights of the K first target branch adaptive layers and the K first target branch adaptive layers are used by the sending end to process the encoded data to be sent to obtain the first data.
  • the communication module 701 is further used to: when the maximum weight among the weights of the K first target branch adaptive layers is less than the preset value, send second information instructing the sender to directly send the first data.
  • the communication module 701 is further used to: send a first channel dictionary set, where the first channel dictionary set is used by the transmitting end to determine the correspondence between the N first target branch adaptation layers and the M second target branch adaptation layers of the transmitting end, where M is a positive integer.
  • the communication module 701 is further configured to:
  • the processing module 702 is further configured to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set;
  • the communication module 701 is further used to send first indication information, where the first indication information is used to indicate a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the first channel dictionary set includes a correspondence between a first target branch adaptation layer and a channel label.
  • the communication module 701 is further configured to:
  • Send fourth information indicates weights of the K second target branch adaptive layers and the K second target branch adaptive layers
  • the weights of the K second target branch adaptive layers and the K second target branch adaptive layers are used by the sending end to process the coded data to be sent to obtain the first data
  • the weights of the K second target branch adaptive layers and the K second target branch adaptive layers are determined according to the weights of the K first target branch adaptive layers and the K first target branch adaptive layers.
  • the communication module 701 is further configured to:
  • First indication information is received, where the first indication information is used to indicate a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the communication module 701 is further configured to:
  • the processing module 702 is further configured to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the communication module 701 is further configured to: receive a second reference signal
  • Sending third information where the third information is the same as the first information, or the third information instructs the transmitting end to send the reference signal after a first time interval, and the third information is obtained according to the second reference signal;
  • FIG. 7b it is a schematic diagram of the structure of another communication device provided in an embodiment of the present application.
  • the communication device is used to implement the aforementioned communication method, such as the communication method shown in Fig. 2a and Fig. 4.
  • the communication device includes a communication module 703 and a processing module 704, which are specifically as follows:
  • the communication module 703 is configured to send a first reference signal
  • the processing module 704 is configured to, when first information is received, the first information indicating K first target branch adaptation layers and weights of the K first target branch adaptation layers in the first target model of the first reference signal receiving end, determine, according to the first information, K second target branch adaptation layers corresponding to the K first target branch adaptation layers and weights of the K second target branch adaptation layers, wherein the first information is obtained according to the first reference signal, and K is a positive integer;
  • the processing module 704 is further configured to input the coded data to be sent into the K second target branch adaptive layers of the second target model for processing to obtain first data;
  • the communication module 703 is further configured to send the first data.
  • the processing module 704 is also used to update the third target model to obtain the second target model, so that the second target model includes the K second target branch adaptive layers, wherein the third target model includes M second target branch adaptive layers, and M is a positive integer.
  • the K second target branch adaptive layers are used to process the encoded data to be sent respectively to obtain the data processed by the K second target branch adaptive layers respectively; the weights of the K second target branch adaptive layers are used to perform weighted sum processing on the data processed by the K second target branch adaptive layers respectively to obtain the first data.
  • the communication module 703 is further used to: when receiving second information, the second information indicates to directly send the coded data to be sent, then send the coded data to be sent, and the second information is obtained based on the first reference signal.
  • the communication module 703 is further used to: receive a first channel dictionary set;
  • the processing module 704 is further configured to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the communication module 703 is further configured to:
  • First indication information is received, where the first indication information is used to indicate a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the second channel dictionary set includes a correspondence between a second target branch adaptation layer and a channel label.
  • the communication module 703 is further configured to: send a second reference signal
  • the third information is the same as the first information; or, the third information is used to indicate that a reference signal is sent after a first time interval, and the third information is obtained according to the second reference signal;
  • the present application also provides a communication device, including:
  • a communication module configured to receive a first reference signal
  • a processing module is used to input the first reference signal into a first target model for processing to obtain K first target branch adaptive layers and weights of the K first target branch adaptive layers in the first target model, wherein the first target model includes N first target branch adaptive layers, K is not greater than N, and K and N are both positive integers; the communication module is also used to send first information, wherein the first information indicates the K first target branch adaptive layers and the weights of the K first target branch adaptive layers, and the K first target branch adaptive layers and the weights of the K first target branch adaptive layers are used by the transmitting end to process the coded data to be sent.
  • the communication module is also used to: receive first data, where the first data is obtained by the sending end processing the encoded data to be sent; the processing module is also used to input the first data into the K first target branch adaptive layers for processing to obtain processed data.
  • the communication module is further used to: send a first channel dictionary set, where the first channel dictionary set is used by the transmitting end to determine the correspondence between the N first target branch adaptation layers and the M second target branch adaptation layers of the transmitting end, where M is a positive integer.
  • the communication module is further used to: receive a second channel dictionary set
  • the processing module is further used to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set;
  • the communication module is further used to send first indication information, where the first indication information is used to indicate a corresponding relationship between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the first channel dictionary set includes a correspondence between a first target branch adaptation layer and a channel label.
  • the communication module is also used to: receive a second reference signal; send third information, the third information is the same as the first information, or the third information indicates that the transmitter sends the reference signal after a first time interval, and the third information is obtained based on the second reference signal; receive data sent by the transmitter within the first time.
  • the present application also provides a communication device, including:
  • a communication module configured to send a first reference signal
  • the communication module is also used to receive first information, where the first information indicates K first target branch adaptive layers and weights of the K first target branch adaptive layers in the first target model of the first reference signal receiving end, and the first information is obtained based on the first reference signal, and K is a positive integer.
  • the device further includes a processing module, configured to: determine, according to the first information, K second target branch adaptive layers corresponding to the K first target branch adaptive layers and weights of the K second target branch adaptive layers;
  • the processing module is further used to input the coded data to be sent into the K second target branch adaptive layers of the second target model for processing to obtain first data;
  • the communication module is further used to send the first data.
  • the processing module when M is less than K, is also used to update the third target model to obtain the second target model, so that the second target model includes the K second target branch adaptive layers, wherein the third target model includes M second target branch adaptive layers, and M is a positive integer.
  • the communication module is further used to: receive a first channel dictionary set
  • the processing module is further used to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the communication module is further used to: send a second channel dictionary set; receive first indication information, wherein the first indication information is used to indicate the correspondence between the second target branch adaptation layer in the second channel dictionary set and the first target branch adaptation layer in the first channel dictionary set.
  • the second channel dictionary set includes a correspondence between a second target branch adaptation layer and a channel label.
  • the communication module is also used to: send a second reference signal; receive third information, the third information is the same as the first information, or the third information indicates that the reference signal is sent after a first time interval, and the third information is obtained based on the second reference signal; and send data within the first time.
  • the present application also provides a communication device, including:
  • a communication module configured to receive a first reference signal
  • a processing module configured to input the first reference signal into a first target model for processing, and obtain K first target branch adaptive layers and weights of the K first target branch adaptive layers in the first target model, wherein the first target model includes N first target branch adaptive layers, K is not greater than N, and K and N are both positive integers;
  • the communication module is also used to send fourth information, wherein the fourth information indicates weights of the K second target branch adaptive layers and the K second target branch adaptive layers, and the weights of the K second target branch adaptive layers and the K second target branch adaptive layers are used by the transmitting end to process the coded data to be sent, and the weights of the K second target branch adaptive layers and the K second target branch adaptive layers are determined according to the weights of the K first target branch adaptive layers and the K first target branch adaptive layers.
  • the communication module is further used to:
  • the processing module is further used to input the first data into the K first target branch adaptive layers for processing to obtain processed data.
  • the communication module is further used to:
  • First indication information is received, where the first indication information is used to indicate a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the communication module is further used to:
  • the processing module is further used to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the first channel dictionary set includes a correspondence between a first target branch adaptation layer and a channel label.
  • the communication module is further used to:
  • the present application also provides a communication device, including:
  • a communication module configured to send a first reference signal
  • the communication module is further used to receive fourth information, where the fourth information indicates K second target branch adaptive layers and weights of the K second target branch adaptive layers, and the fourth information is obtained according to the first reference signal, where K is a positive integer.
  • the device further includes a processing module, configured to:
  • the communication module is further used to send the first data.
  • the communication module is further used to:
  • the processing module is further used to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set;
  • the communication module is further used to send first indication information, where the first indication information is used to indicate a corresponding relationship between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the communication module is further used to:
  • a second channel dictionary set is sent, where the second channel dictionary set is used by a reference signal receiving end to determine a correspondence between N first target branch adaptation layers of the reference signal receiving end and M second target branch adaptation layers of a transmitting end, where M and N are both positive integers.
  • the second channel dictionary set includes a correspondence between a second target branch adaptation layer and a channel label.
  • the communication module is further used to:
  • third information is the same as the first information, or the third information indicates that a reference signal is sent after a first time interval, and the third information is obtained according to the second reference signal;
  • the present application also provides a communication device, including:
  • a communication module configured to send a first reference signal
  • a processing module configured to, when receiving fourth information, the fourth information indicating K second target branch adaptive layers and weights of the K second target branch adaptive layers, input the coded data to be sent into the K second target branch adaptive layers of the second target model for processing according to the fourth information, to obtain first data, wherein the fourth information is obtained according to the first reference signal, and K is a positive integer;
  • the communication module is further used to send the first data.
  • the K second target branch adaptive layers are used to process the encoded data to be sent respectively to obtain the data processed by the K second target branch adaptive layers respectively; the weights of the K second target branch adaptive layers are used to perform weighted sum processing on the data processed by the K second target branch adaptive layers respectively to obtain the first data.
  • the communication module is further used to:
  • the coded data to be sent is sent, and the second information is obtained according to the first reference signal.
  • the communication module is further used to:
  • the processing module is further used to perform dictionary alignment on the first channel dictionary set and the second channel dictionary set to obtain a correspondence between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set;
  • the communication module is further used to send first indication information, where the first indication information is used to indicate a corresponding relationship between a second target branch adaptation layer in the second channel dictionary set and a first target branch adaptation layer in the first channel dictionary set.
  • the communication module is further used to:
  • a second channel dictionary set is sent, where the second channel dictionary set is used by a reference signal receiving end to determine a correspondence between N first target branch adaptation layers of the reference signal receiving end and M second target branch adaptation layers of a transmitting end, where M and N are both positive integers.
  • the second channel dictionary set includes a correspondence between a second target branch adaptation layer and a channel label.
  • the processing module is further configured to:
  • the third information is the same as the first information; or, the third information is used to indicate that a reference signal is sent after a first time interval, and the third information is obtained according to the second reference signal;
  • the embodiment of the present application also provides a target model training device.
  • the target model includes a target encoding network, a target decoding network and a first target model, and the first target model includes a first target channel feature extraction network, N first target branch adaptive layers and a target sparse gating module.
  • the device includes:
  • a first training module is used to train the initial channel feature extraction network to obtain the first target channel feature extraction network and the first channel dictionary set;
  • a second training module is used to train the N initial branch adaptive layers according to the target encoding network and the target decoding network to obtain the N first target branch adaptive layers, where N is the number of first target branch adaptive layers in the first channel dictionary set;
  • the third training module is used to train the initial sparse gating module according to the target encoding network, the target decoding network, the first target channel feature extraction network and the N first target branch adaptive layers to obtain the target sparse gating module.
  • the first channel dictionary set includes a correspondence between a first target branch adaptation layer and a channel label.
  • the first training module is used to: train the initial channel feature extraction network multiple times to obtain the first target channel feature extraction network and the first channel dictionary set.
  • the initial channel feature extraction network is pre-trained based on labeled historical channel information or a preset channel model.
  • both the unlabeled channel information and the labeled historical channel information are input into the channel feature extraction network U t-1 for processing, and the channel distribution information corresponding to the unlabeled channel information and the labeled historical channel information is obtained.
  • N t-1 cluster centers are obtained.
  • a predicted channel dictionary set is obtained.
  • the second training module is used to: when the mth initial branch adaptive layer is trained for the tth time, the first sample data is input into the target coding network and the preset channel for processing to obtain the second sample data.
  • the second sample data is input into the mth initial branch adaptive layer for processing to obtain the data processed by the mth initial branch adaptive layer, where m is a positive integer and m is not greater than N.
  • the data processed by the mth initial branch adaptive layer is input into the target decoding network for processing to obtain the third sample data.
  • the device further includes a communication module, which is used to: send second indication information, where the second indication information instructs the mth initial branch adaptive layer to perform training.
  • the third training module is used to: perform multiple training on the initial sparse gating module to obtain the target sparse gating module.
  • the fourth sample data is input into the target encoding network and the preset channel for processing to obtain the fifth sample data.
  • the fifth sample data is input into the first target channel feature extraction network for processing to obtain channel distribution information training data and channel category information training data.
  • the channel distribution information training data and the channel category information training data are both input into the sparse gating module X t-1 for processing to obtain R first target branch adaptive layer samples and R first target branch adaptive layer weight samples corresponding to the channel distribution information training data, where R is a positive integer and R is not greater than N.
  • the fifth sample data is processed according to the R first target branch adaptive layer samples and the R first target branch adaptive layer weight samples to obtain processed training data.
  • the processed training data is input into the target decoding network for processing to obtain the sixth sample data.
  • each module in each of the above devices is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated.
  • the modules in the communication device or the training device of the target model can be implemented in the form of a processor calling software; for example, the communication device includes a processor, the processor is connected to a memory, and instructions are stored in the memory.
  • the processor calls the instructions stored in the memory to implement any of the above methods or realize the functions of each module of the device, wherein the processor is, for example, a general-purpose processor, such as a central processing unit (CPU) or a microprocessor, and the memory is a memory inside the device or a memory outside the device.
  • CPU central processing unit
  • microprocessor a microprocessor
  • the modules in the device may be implemented in the form of hardware circuits, and the functions of some or all units may be implemented by designing the hardware circuits, which may be understood as one or more processors; for example, in one implementation, the hardware circuit is an application-specific integrated circuit (ASIC), and the functions of some or all of the above units may be implemented by designing the logical relationship of the components in the circuit; for another example, in another implementation, the hardware circuit may be implemented by a programmable logic device (PLD), and a field programmable gate array (FPGA) may be used as an example, which may include a large number of logic gate circuits, and the connection relationship between the logic gate circuits may be configured by a configuration file, so as to implement the functions of some or all of the above units. All modules of the above devices may be implemented in the form of a processor calling software, or in the form of hardware circuits, or in part by a processor calling software, and the rest by hardware circuits.
  • All modules of the above devices may be implemented in the form of a processor calling software, or
  • FIG8 it is a schematic diagram of the hardware structure of another communication device provided in an embodiment of the present application.
  • the communication device 800 shown in FIG8 (the device 800 may be a computer device) includes a memory 801, a processor 802, a communication interface 803, and a bus 804. Among them, the memory 801, the processor 802, and the communication interface 803 are connected to each other through the bus 804.
  • Memory 801 can be a read-only memory (ROM), a static storage device, a dynamic storage device or a random access memory (RAM).
  • ROM read-only memory
  • RAM random access memory
  • the memory 801 can store programs. When the program stored in the memory 801 is executed by the processor 802, the processor 802 and the communication interface 803 are used to execute the various steps of the communication method or the target model training method of the embodiment of the present application.
  • the processor 802 is a circuit with signal processing capability.
  • the processor 802 may be a circuit with instruction reading and running capability, such as a central processing unit CPU, a microprocessor, a graphics processing unit (GPU) (which may be understood as a microprocessor), or a digital signal processor (DSP); in another implementation, the processor 802 may implement certain functions through the logical relationship of a hardware circuit, and the logical relationship of the hardware circuit may be fixed or reconfigurable, such as the processor 802 being a hardware circuit implemented by an ASIC or a programmable logic device PLD, such as an FPGA.
  • a programmable logic device PLD such as an FPGA.
  • the process of the processor loading a configuration document to implement the hardware circuit configuration may be understood as the process of the processor loading instructions to implement the functions of some or all of the above modules.
  • it may also be a hardware circuit designed for artificial intelligence, which may be understood as an ASIC, such as a neural network processing unit (NPU), a tensor processing unit (TPU), a deep learning processing unit (DPU), etc.
  • Processor 802 is used to execute relevant programs to implement the functions required to be performed by the units in the communication device or target model training device of the embodiment of the present application, or to execute the communication method or target model training method of the method embodiment of the present application.
  • each module in the above device can be one or more processors (or processing circuits) configured to implement the above method, such as: CPU, GPU, NPU, TPU, DPU, microprocessor, DSP, ASIC, FPGA, or a combination of at least two of these processor forms.
  • the modules in the above device can be integrated together in whole or in part, or can be implemented independently. In one implementation, these modules are integrated together and implemented in the form of a system-on-a-chip (SOC).
  • SOC may include at least one processor for implementing any of the above methods or implementing the functions of the modules of the device.
  • the type of the at least one processor may be different, for example, including a CPU and an FPGA, a CPU and an artificial intelligence processor, a CPU and a GPU, etc.
  • the communication interface 803 uses a transceiver device such as, but not limited to, a transceiver to implement communication between the device 800 and other devices or a communication network. For example, data can be obtained through the communication interface 803.
  • a transceiver device such as, but not limited to, a transceiver to implement communication between the device 800 and other devices or a communication network. For example, data can be obtained through the communication interface 803.
  • the bus 804 may include a path for transmitting information between various components of the device 800 (eg, the memory 801 , the processor 802 , and the communication interface 803 ).
  • the device 800 shown in FIG8 only shows a memory, a processor, and a communication interface, in the specific implementation process, those skilled in the art should understand that the device 800 also includes other devices necessary for normal operation. At the same time, according to specific needs, those skilled in the art should understand that the device 800 may also include hardware devices for implementing other additional functions. In addition, those skilled in the art should understand that the device 800 may also only include the devices necessary for implementing the embodiments of the present application, and does not necessarily include all the devices shown in FIG8.
  • An embodiment of the present application also provides a communication system, for example, including the above-mentioned device as shown in FIG. 7a and the device as shown in FIG. 7b.
  • An embodiment of the present application also provides a computer-readable storage medium, which stores instructions.
  • the computer-readable storage medium is executed on a computer or a processor, the computer or the processor executes one or more steps in any of the above methods.
  • the embodiment of the present application further provides a computer program product including instructions.
  • the computer program product is run on a computer or a processor, the computer or the processor executes one or more steps in any of the above methods.
  • A/B can represent A or B; wherein A and B can be singular or plural.
  • multiple refers to two or more than two.
  • At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items.
  • at least one of a, b, or c can represent: a, b, c, a-b, a-c, b-c, or a-b-c, wherein a, b, c can be single or multiple.
  • the words “first”, “second”, etc. are used to distinguish the same items or similar items with substantially the same functions and effects. Those skilled in the art can understand that the words “first”, “second”, etc. do not limit the quantity and execution order, and the words “first”, “second”, etc. do not limit them to be necessarily different. Meanwhile, in the embodiments of the present application, words such as “exemplary” or “for example” are used to indicate examples, illustrations or descriptions. Any embodiment or design described as “exemplary” or “for example” in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as “exemplary” or “for example” is intended to present related concepts in a concrete manner for ease of understanding.
  • the disclosed systems, devices and methods can be implemented in other ways.
  • the division of the unit is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.
  • the mutual coupling, direct coupling, or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • the computer program product includes one or more computer instructions.
  • the computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.
  • the computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium.
  • the computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.
  • the computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated.
  • the available medium can be a read-only memory (ROM), or a random access memory (RAM), or a magnetic medium, such as a floppy disk, a hard disk, a tape, a disk, or an optical medium, such as a digital versatile disc (DVD), or a semiconductor medium, such as a solid state disk (SSD), etc.
  • ROM read-only memory
  • RAM random access memory
  • magnetic medium such as a floppy disk, a hard disk, a tape, a disk, or an optical medium, such as a digital versatile disc (DVD), or a semiconductor medium, such as a solid state disk (SSD), etc.
  • SSD solid state disk

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Compression, Expansion, Code Conversion, And Decoders (AREA)

Abstract

本申请实施例提供一种通信方法、装置及系统。该方法可包括:参考信号接收端接收第一参考信号。参考信号接收端将该第一参考信号输入至第一目标模型中进行处理,得到该第一目标模型中的K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重。其中,该第一目标模型包括N个第一目标分支自适应层。参考信号接收端还接收第一数据。然后,参考信号接收端将该第一数据输入至该K个第一目标分支自适应层中进行处理,得到处理后的数据。其中,该K个第一目标分支自适应层的权重中最大权重不小于预设值。基于确定的K个第一目标分支自适应层和K个权重来表征实际的信道分布,以应对信道的动态变化,这样可以提高系统性能。

Description

一种通信方法、装置及系统 技术领域
本申请涉及通信技术领域,尤其涉及一种通信方法、装置及系统。
背景技术
近些年,深度学习的发展引起了学术界和工业界对基于深度学习的无线通信技术的研究,研究结果证实了深度学习技术可以提高无线通信系统的性能,并有潜力应用在物理层进行干扰调整、信道估计和信号检测、信号处理等方面。
其中,神经网络收发机通过联合发送端和接收端,实现针对特定性能指标和信道模型的优化,使得无需借助先验的专家知识,实现定制自演化空口,并逼近香农极限。而在实际信道场景应用中,信道的分布是动态变化的。这就需要神经网络模型具有判别出当前环境的信道变化,并能用较少的训练开销实现快速适配新场景的能力。
为解决信道分布变化导致的问题,ParkS等人结合元学习方法给出元-自编码meta-autoencoder网络结构。该方案是将在各种信道下的端到端训练分别作为一个子任务,从而训练对各种信道下均有潜力的初始收发端网络,该主任务的训练目标是得到最优的网络初始化参数,使得系统能够在较少随机梯度下降(Stochastic Gradient Descent,SGD)步数下对任意信道的训练收敛,进而快速适应时变信道的变化。
由于实际场景中的信道分布是动态变化的,其可能不存在在上述元学习方法中已有的子任务中,因此则需要迭代计算更多的子任务,使得训练的计算开销较大;且每一轮迭代都需要计算所有子任务中信道的情况,因此需要将每个子任务训练至拟合得到反馈,更为耗时。
发明内容
本申请公开了一种通信方法、装置及系统,可以解决由于环境中的信道动态变化而导致的神经网络训练与推演的问题。
第一方面,本申请实施例提供一种通信方法。该方法可以由通信设备执行,或由通信设备的组件(如,芯片(系统))执行。该方法包括:参考信号接收端接收第一参考信号。然后,参考信号接收端将该第一参考信号输入至第一目标模型中进行处理,得到该第一目标模型中的K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重。其中,该第一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数。参考信号接收端还发送第一信息。该第一信息指示该K个第一目标分支自适应层和该K个第一目标分支自适应层的权重。该K个第一目标分支自适应层和该K个第一目标分支自适应层的权重用于发送端对待发送的编码数据进行处理。
本申请实施例,参考信号接收端基于接收到的参考信号确定第一目标模型的N个第一目标分支自适应层中的K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重。采用该手段,基于确定的K个第一目标分支自适应层和K个权重来表征实际的信道分布,以应对信道的动态变化,这样可以提高系统性能。
在一种可能的实现方式中,该方法还包括:参考信号接收端接收第一数据,该第一数据是该发送端对待发送的编码数据进行处理得到的。然后,参考信号接收端将该第一数据输入至该K个第一目标分支自适应层中进行处理,得到处理后的数据。
在一种可能的实现方式中,该方法还包括:参考信号接收端发送第一信道字典集,该第一信道字典集用于该发送端确定该N个第一目标分支自适应层和该发送端的M个第二目标分支自适应层之间的对应关系,M为正整数。
通过进行字典对齐,使得发端选择对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
在另一种可能的实现方式中,该方法还包括:参考信号接收端接收第二信道字典集。然后,参考信号接收端将第一信道字典集和该第二信道字典集进行字典对齐,得到该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。最后,参考信号接收端发送第一指示信息,该第一指示信息用于指示该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
通过进行字典对齐,使得发端选择对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
在一种可能的实现方式中,该第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该方法还包括:参考信号接收端接收第二参考信号。然后,参考信号接收端发送第三信息,该第三信息与该第一信息相同,或者,该第三信息指示该发送端间隔第一时间后再发送参考信号。其中,该第三信息是根据该第二参考信号得到的。最后,参考信号接收端接收该发送端在该第一时间内发送的数据。
发送端在该第一时间内发送数据,可以理解为,发送端在该第一时间内不发送参考信号。这样,可以减少导频传输开销,增加数据传输的频谱效率。
第二方面,本申请实施例提供一种通信方法。该方法可以由通信设备执行,或由通信设备的组件(如,芯片(系统))执行。该方法包括:参考信号接收端接收第一参考信号。然后,参考信号接收端将该第一参考信号输入至第一目标模型中进行处理,得到该第一目标模型中的K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重。该第一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数。
参考信号接收端还发送第四信息,该第四信息指示K个第二目标分支自适应层和该K个第二目标分支自适应层的权重。该K个第二目标分支自适应层和该K个第二目标分支自适应层的权重用于发送端对待发送的编码数据进行处理。该K个第二目标分支自适应层和该K个第二目标分支自适应层的权重是根据该K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重确定的。
本申请实施例,参考信号接收端基于接收到的参考信号确定第一目标模型的N个第一目标分支自适应层中的K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重。然后向发送端指示K个第二目标分支自适应层和该K个第二目标分支自适应层的权重。采用该手段,基于确定的K个第一目标分支自适应层和K个权重来表征实际的信道分布,以应对信道的动态变化,这样可以提高系统性能。且通过向发送端指示,以便发送端对待发送的编码数据进行处理,这样可以提高通信性能。
在一种可能的实现方式中,该方法还包括:参考信号接收端接收第一数据。该第一数据是该发送端对待发送的编码数据进行处理得到的。
参考信号接收端还将该第一数据输入至该K个第一目标分支自适应层中进行处理,得到 处理后的数据。
在一种可能的实现方式中,该方法还包括:参考信号接收端发送第一信道字典集,该第一信道字典集用于该发送端确定该N个第一目标分支自适应层和该发送端的M个第二目标分支自适应层之间的对应关系,M为正整数。参考信号接收端还接收第一指示信息,该第一指示信息用于指示该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
通过进行字典对齐,进而参考信号接收端可以确定上述K个第二目标分支自适应层和该K个第二目标分支自适应层的权重,以指示发端按照对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
在另一种可能的实现方式中,该方法还包括:参考信号接收端接收第二信道字典集。然后,参考信号接收端将第一信道字典集和该第二信道字典集进行字典对齐,得到该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
通过进行字典对齐,进而参考信号接收端可以确定上述K个第二目标分支自适应层和该K个第二目标分支自适应层的权重,以指示发端按照对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
在一种可能的实现方式中,该第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该方法还包括:参考信号接收端接收第二参考信号。然后参考信号接收端发送第五信息,该第五信息与该第四信息相同,或者,该第五信息指示该发送端间隔第一时间后再发送参考信号,该第五信息是根据该第二参考信号得到的。参考信号接收端还接收该发送端在该第一时间内发送的数据。
发送端在该第一时间内发送数据,可以理解为,发送端在该第一时间内不发送参考信号。这样,可以减少导频传输开销,增加数据传输的频谱效率。
第三方面,本申请实施例提供一种通信方法。该方法可以由通信设备执行,或由通信设备的组件(如,芯片(系统))执行。该方法包括:发送端发送第一参考信号。然后,发送端接收第一信息,该第一信息指示该第一参考信号接收端的第一目标模型中的K个第一目标分支自适应层和该K个第一目标分支自适应层的权重。其中,该第一信息是根据该第一参考信号得到的,K为正整数。
本申请实施例,发送端接收到的信息指示第一参考信号接收端的第一目标模型中的K个第一目标分支自适应层和该K个第一目标分支自适应层的权重。这样做,基于K个第一目标分支自适应层和K个权重来表征实际的信道分布,以应对信道的动态变化,这样可以提高通信性能。
在一种可能的实现方式中,该方法还包括:发送端根据该第一信息,确定与该K个第一目标分支自适应层对应的K个第二目标分支自适应层和该K个第二目标分支自适应层的权重。然后,发送端将待发送的编码数据输入至第二目标模型的该K个第二目标分支自适应层中进行处理,得到第一数据。最后,发送端发送该第一数据。
在一种可能的实现方式中,当M小于K时,发送端更新第三目标模型,得到该第二目标模型,使得该第二目标模型包括该K个第二目标分支自适应层。其中该第三目标模型包括M个第二目标分支自适应层,M为正整数。
这样可以使得基于接收到的信息能够确定出对应的目标分支自适应层和权重,来对编码数据进行处理,可以提高通信性能。
在一种可能的实现方式中,该方法还包括:发送端接收第一信道字典集。然后,发送端将该第一信道字典集和第二信道字典集进行字典对齐,得到该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
通过接收信道字典集,以便两端进行字典对齐,进而可以选择对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
在另一种可能的实现方式中,该方法还包括:发送端发送第二信道字典集。发送端还接收第一指示信息,该第一指示信息用于指示该第二信道字典集中的第二目标分支自适应层与第一信道字典集中的第一目标分支自适应层之间的对应关系。
通过进行字典对齐,使得发端选择对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
在一种可能的实现方式中,该第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该方法还包括:发送端发送第二参考信号。然后,发送端接收第三信息,该第三信息与该第一信息相同,或者,该第三信息指示间隔第一时间后再发送参考信号。其中,该第三信息是根据该第二参考信号得到的。最后,发送端在该第一时间内发送数据。
发送端在该第一时间内发送数据,可以理解为,发送端在该第一时间内不发送参考信号。这样,可以减少导频传输开销,增加数据传输的频谱效率。
第四方面,本申请实施例提供一种通信方法。该方法可以由通信设备执行,或由通信设备的组件(如,芯片(系统))执行。该方法包括:发送端发送第一参考信号。然后发送端接收第四信息。该第四信息指示K个第二目标分支自适应层和该K个第二目标分支自适应层的权重。该第四信息是根据该第一参考信号得到的,K为正整数。
本申请实施例,发送端接收到的信息指示K个第二目标分支自适应层和该K个第二目标分支自适应层的权重。这样做,基于K个目标分支自适应层和K个权重来表征实际的信道分布,以应对信道的动态变化,这样可以提高通信性能。
在一种可能的实现方式中,该方法还包括:
发送端根据该第四信息,将待发送的编码数据输入至第二目标模型的该K个第二目标分支自适应层中进行处理,得到第一数据。
然后,发送端发送该第一数据。
这样,发端按照对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
在一种可能的实现方式中,该方法还包括:
发送端接收第一信道字典集。
然后,发送端将该第一信道字典集和第二信道字典集进行字典对齐,得到该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
发送端还发送第一指示信息,该第一指示信息用于指示该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
通过接收信道字典集,以便两端进行字典对齐,然后指示参考信号接收端来确定对应的目标分支自适应层及权重,以便发端基于对应信道分布下的最优处理方式对编码数据进行处理,这样可以提高通信性能。
在另一种可能的实现方式中,该方法还包括:
发送端发送第二信道字典集,该第二信道字典集用于参考信号接收端确定该参考信号接收端的N个第一目标分支自适应层和发送端的M个第二目标分支自适应层之间的对应关系,M、N均为正整数。
通过发送信道字典集,以便参考信号接收端进行两端字典对齐,来确定对应的目标分支自适应层及权重,进而以便发端基于对应信道分布下的最优处理方式对编码数据进行处理,这样可以提高通信性能。
在一种可能的实现方式中,该第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该方法还包括:
发送端发送第二参考信号。
然后发送端接收第五信息,该第五信息与该第四信息相同,或者,该第五信息指示间隔第一时间后再发送参考信号,该第五信息是根据该第二参考信号得到的;
发送端在该第一时间内发送数据。
发送端在该第一时间内发送数据,可以理解为,发送端在该第一时间内不发送参考信号。这样,可以减少导频传输开销,增加数据传输的频谱效率。
第五方面,本申请实施例提供一种通信方法。该方法可以由通信设备执行,或由通信设备的组件(如,芯片(系统))执行。该方法包括:参考信号接收端接收第一参考信号。参考信号接收端将该第一参考信号输入至第一目标模型中进行处理,得到该第一目标模型中的K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重。其中,该第一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数。参考信号接收端还接收第一数据。然后,参考信号接收端将该第一数据输入至该K个第一目标分支自适应层中进行处理,得到处理后的数据。其中,该K个第一目标分支自适应层的权重中最大权重不小于预设值。
本申请实施例,参考信号接收端基于接收到的参考信号确定第一目标模型的N个第一目标分支自适应层中的K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重。当该K个第一目标分支自适应层的权重中最大权重不小于预设值,通过将接收到的第一数据输入至该K个第一目标分支自适应层中进行处理,得到处理后的数据。采用该手段,基于确定的K个第一目标分支自适应层和K个权重来表征实际的信道分布,以应对信道的动态变化,这样可以提高系统性能。
在一种可能的实现方式中,该第一目标模型还包括第一目标信道特征提取网络和目标稀疏门控模块。其中,该第一目标信道特征提取网络用于对该第一参考信号进行处理,得到信道分布信息和该信道分布信息对应的信道类别信息。
该目标稀疏门控模块用于根据该信道分布信息和该信道类别信息计算得到该K个第一目标分支自适应层和该K个第一目标分支自适应层的权重。
该第一目标信道特征提取网络和目标稀疏门控模块可以是两个独立的模块,也可以是集成一体的,本方案对此不作限制。
在一种可能的实现方式中,该K个第一目标分支自适应层用于分别对该第一数据进行处理,得到该K个第一目标分支自适应层分别处理后的数据。
该K个第一目标分支自适应层的权重用于对该K个第一目标分支自适应层分别处理后的数据进行加权求和处理,得到该处理后的数据。
在一种可能的实现方式中,该方法还包括:当该K个第一目标分支自适应层的权重中最大权重小于该预设值时,参考信号接收端对该第一数据进行解码处理,得到处理后的数据。
在一种可能的实现方式中,该方法还包括:参考信号接收端发送第一信息。该第一信息指示该K个第一目标分支自适应层和该K个第一目标分支自适应层的权重。该K个第一目标分支自适应层和该K个第一目标分支自适应层的权重用于发送端对待发送的编码数据进行处理得到该第一数据。
当K个第一目标分支自适应层的权重中最大权重不小于预设值时,向发送端发送信息,以便发送端对待发送的编码数据进行处理后再发送。这样做,发送端按信息指示,选择对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
在一种可能的实现方式中,该方法还包括:当该K个第一目标分支自适应层的权重中最大权重小于该预设值,参考信号接收端发送第二信息,该第二信息指示发送端直接发送该第一数据。
在一种可能的实现方式中,该方法还包括:参考信号接收端发送第一信道字典集,该第一信道字典集用于发送端确定该N个第一目标分支自适应层和该发送端的M个第二目标分支自适应层之间的对应关系,M为正整数。
通过发送信道字典集,以便两端进行字典对齐,使得发端选择对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
在另一种可能的实现方式中,该方法还包括:参考信号接收端接收第二信道字典集。参考信号接收端将第一信道字典集和该第二信道字典集进行字典对齐,得到该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。然后,参考信号接收端发送第一指示信息,该第一指示信息用于指示该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
通过进行字典对齐,使得发端选择对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
在一种可能的实现方式中,该方法还包括:
参考信号接收端发送第四信息,该第四信息指示K个第二目标分支自适应层和该K个第二目标分支自适应层的权重。该K个第二目标分支自适应层和该K个第二目标分支自适应层的权重用于发送端对待发送的编码数据进行处理得到该第一数据。该K个第二目标分支自适应层和该K个第二目标分支自适应层的权重是根据该K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重确定的。
在一种可能的实现方式中,该方法还包括:
参考信号接收端发送第一信道字典集。
参考信号接收端还接收第一指示信息。该第一指示信息用于指示该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,该方法还包括:
参考信号接收端接收第二信道字典集。
参考信号接收端还将第一信道字典集和该第二信道字典集进行字典对齐,得到该第二信 道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,该第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该方法还包括:参考信号接收端接收第二参考信号。然后,参考信号接收端发送第三信息,该第三信息与该第一信息相同,或者,该第三信息指示发送端间隔第一时间后再发送参考信号。其中,该第三信息是根据该第二参考信号得到的。参考信号接收端还接收该发送端在该第一时间内发送的数据。
在另一种可能的实现方式中,该方法还包括:参考信号接收端接收第二参考信号。然后,参考信号接收端发送第五信息,该第五信息与该第四信息相同,或者,该第五信息指示发送端间隔第一时间后再发送参考信号。其中,该第五信息是根据该第二参考信号得到的。参考信号接收端还接收该发送端在该第一时间内发送的数据。
发送端在该第一时间内发送数据,可以理解为,发送端在该第一时间内不发送参考信号。这样,可以减少导频传输开销,增加数据传输的频谱效率。该示例可参阅上述第三信息的介绍,在此不再赘述。
第六方面,本申请实施例提供一种通信方法。该方法可以由通信设备执行,或由通信设备的组件(如,芯片(系统))执行。该方法包括:发送端发送第一参考信号。当发送端接收到第一信息时,该第一信息指示该第一参考信号接收端的第一目标模型中的K个第一目标分支自适应层和该K个第一目标分支自适应层的权重,根据该第一信息,确定与该K个第一目标分支自适应层对应的K个第二目标分支自适应层和该K个第二目标分支自适应层的权重。其中,该第一信息是根据该第一参考信号得到的,K为正整数。然后,发送端将待发送的编码数据输入至第二目标模型的该K个第二目标分支自适应层中进行处理,得到第一数据。最后,发送端发送该第一数据。
本申请实施例,发送端基于接收到的信息,对待发送的编码数据进行处理后再发送。这样做,基于确定的K个第一目标分支自适应层和K个权重来表征实际的信道分布,以应对信道的动态变化;且通过选择对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
在一种可能的实现方式中,当M小于K时,发送端更新第三目标模型,得到该第二目标模型,使得该第二目标模型包括该K个第二目标分支自适应层。其中该第三目标模型包括M个第二目标分支自适应层,M为正整数。
这样可以使得基于接收到的信息能够确定出对应的目标分支自适应层和权重,来对编码数据进行处理,可以提高通信性能。
在一种可能的实现方式中,该K个第二目标分支自适应层用于分别对该待发送的编码数据进行处理,得到该K个第二目标分支自适应层分别处理后的数据。
该K个第二目标分支自适应层的权重用于对该K个第二目标分支自适应层分别处理后的数据进行加权求和处理,得到该第一数据。
在一种可能的实现方式中,该方法还包括:当发送端接收到第二信息时,该第二信息指示直接发送该待发送的编码数据,则发送端发送该待发送的编码数据。其中,该第二信息是根据该第一参考信号得到的。
在一种可能的实现方式中,该方法还包括:发送端接收第一信道字典集。然后,发送端 将该第一信道字典集和第二信道字典集进行字典对齐,得到该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
通过接收信道字典集,以便两端进行字典对齐,进而可以选择对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
在另一种可能的实现方式中,该方法还包括:发送端发送第二信道字典集。然后,发送端接收第一指示信息,该第一指示信息用于指示该第二信道字典集中的第二目标分支自适应层与第一信道字典集中的第一目标分支自适应层之间的对应关系。
通过进行字典对齐,使得发端选择对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
在一种可能的实现方式中,该第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该方法还包括:发送端发送第二参考信号。然后发送端接收第三信息,该第三信息与该第一信息相同;或者,该第三信息用于指示间隔第一时间后再发送参考信号。其中,该第三信息是根据该第二参考信号得到的。最后,发送端在该第一时间内发送数据。
发送端在该第一时间内发送数据,可以理解为,发送端在该第一时间内不发送参考信号。这样,可以减少导频传输开销,增加数据传输的频谱效率。
第七方面,本申请实施例提供一种通信方法。该方法可以由通信设备执行,或由通信设备的组件(如,芯片(系统))执行。该方法包括:发送端发送第一参考信号。然后,发送端当接收到第四信息时,该第四信息指示K个第二目标分支自适应层和该K个第二目标分支自适应层的权重,根据该第四信息,将待发送的编码数据输入至第二目标模型的该K个第二目标分支自适应层中进行处理,得到第一数据。其中,该第四信息是根据该第一参考信号得到的,K为正整数。发送端还发送该第一数据。
本申请实施例,发送端接收到的信息指示K个第二目标分支自适应层和该K个第二目标分支自适应层的权重。这样做,基于K个目标分支自适应层和K个权重来表征实际的信道分布,以应对信道的动态变化,这样可以提高通信性能。且,发端按照对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
在一种可能的实现方式中,该K个第二目标分支自适应层用于分别对该待发送的编码数据进行处理,得到该K个第二目标分支自适应层分别处理后的数据;该K个第二目标分支自适应层的权重用于对该K个第二目标分支自适应层分别处理后的数据进行加权求和处理,得到该第一数据。
在一种可能的实现方式中,该方法还包括:
当接收到第二信息时,该第二信息指示直接发送该待发送的编码数据,则发送端发送该待发送的编码数据。其中,该第二信息是根据该第一参考信号得到的。
在一种可能的实现方式中,该方法还包括:
发送端接收第一信道字典集。然后,发送端将该第一信道字典集和第二信道字典集进行字典对齐,得到该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。发送端还发送第一指示信息,该第一指示信息用于指示该第二信道字典集中的第二目标分支自适应层与第一信道字典集中的第一目标分支自适应层之间的对应关系。
通过接收信道字典集,以便两端进行字典对齐,然后指示参考信号接收端来确定对应的目标分支自适应层及权重,以便发端基于对应信道分布下的最优处理方式对编码数据进行处理,这样可以提高通信性能。
在一种可能的实现方式中,该方法还包括:
发送端发送第二信道字典集,该第二信道字典集用于参考信号接收端确定该参考信号接收端的N个第一目标分支自适应层和发送端的M个第二目标分支自适应层之间的对应关系,M、N均为正整数。
通过发送信道字典集,以便参考信号接收端进行两端字典对齐,来确定对应的目标分支自适应层及权重,进而以便发端基于对应信道分布下的最优处理方式对编码数据进行处理,这样可以提高通信性能。
在一种可能的实现方式中,该第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该方法还包括:
发送端发送第二参考信号。
然后,发送端接收第五信息。该第五信息与该第四信息相同;或者,该第五信息用于指示间隔第一时间后再发送参考信号,该第五信息是根据该第二参考信号得到的;
发送端还在该第一时间内发送数据。
发送端在该第一时间内发送数据,可以理解为,发送端在该第一时间内不发送参考信号。这样,可以减少导频传输开销,增加数据传输的频谱效率。
第八方面,本申请实施例提供一种目标模型训练方法。该方法可以由通信设备执行,或由通信设备的组件(如,芯片(系统))执行。该目标模型包括目标编码网络、目标解码网络和第一目标模型,该第一目标模型包括第一目标信道特征提取网络、N个第一目标分支自适应层和目标稀疏门控模块。该方法包括:对初始信道特征提取网络进行训练,得到该第一目标信道特征提取网络以及第一信道字典集。然后,根据该目标编码网络和该目标解码网络,对N个初始分支自适应层进行训练,得到该N个第一目标分支自适应层,该N为该第一信道字典集中的第一目标分支自适应层的个数。最后,根据该目标编码网络、该目标解码网络、该第一目标信道特征提取网络和该N个第一目标分支自适应层,对初始稀疏门控模块进行训练,得到该目标稀疏门控模块。
本申请实施例,基于对初始信道特征提取网络进行训练,得到该第一目标信道特征提取网络。然后,根据对N个初始分支自适应层进行训练,得到N个第一目标分支自适应层。最后,根据训练好的目标编码网络、目标解码网络、第一目标信道特征提取网络和N个第一目标分支自适应层对初始稀疏门控模块进行训练,得到目标稀疏门控模块。采用该手段得到的目标模型,可以解决由于环境中的信道动态变化而导致的神经网络训练与推演的问题。
另一方面,本方案通过训练分支自适应层权重的方式来调整分支自适应层的输出,而非对目标编码网络参数进行训练,这样能够减少网络训练的开销。
可选的,该第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该对初始信道特征提取网络进行训练,得到该第一目标信道特征提取网络以及第一信道字典集,包括:对该初始信道特征提取网络进行多次训练,得到该第一目标信道特征提取网络以及第一信道字典集。其中,该初始信道特征提取网络是根据带标签的历史信道信息或者预设信道模型进行预训练得到的。
其中,当对信道特征提取网络U t-1进行第t次训练时,将无标签的信道信息与该带标签的历史信道信息均输入至信道特征提取网络U t-1中进行处理,得到该无标签的信道信息与该带标签的历史信道信息分别对应的信道分布信息。根据该信道分布信息得到N t-1个聚类中心。根据该N t-1个聚类中心和该信道分布信息,得到预测的信道字典集。然后,根据该预测的信道字典集与标注的信道字典集计算得到第一损失值,并根据该第一损失值调整该信道特征提取网络U t-1的参数。令t=t+1,并重复执行上述步骤,直到迭代次数达到预设次数,将该信道特征提取网络U t-1作为该第一目标信道特征提取网络,将该预测的信道字典集作为该第一信道字典集。
在一种可能的实现方式中,根据该目标编码网络和该目标解码网络,对N个初始分支自适应层进行训练,得到该N个第一目标分支自适应层,包括:对于第m个初始分支自适应层进行第t次训练时,将第一样本数据输入至该目标编码网络和预设信道中进行处理,得到第二样本数据。将该第二样本数据输入至该第m个初始分支自适应层中进行处理,得到该第m个初始分支自适应层处理得到的数据,m为正整数,m不大于N。将该第m个初始分支自适应层处理得到的数据输入至该目标解码网络中进行处理,得到第三样本数据。然后,根据该第一样本数据和该第三样本数据计算得到第二损失值,并根据该第二损失值调整该第m个初始分支自适应层参数。令t=t+1,并重复上述步骤,直到达到停止条件,将该第m个初始分支自适应层作为第m个第一目标分支自适应层。
令m=m+1,并重复执行上述步骤,直到m=N,得到该N个第一目标分支自适应层。
在一种可能的实现方式中,该方法还包括:发送第二指示信息,该第二指示信息指示该第m个初始分支自适应层进行训练。
在一种可能的实现方式中,根据该目标编码网络、该目标解码网络、该第一目标信道特征提取网络和该N个第一目标分支自适应层,对初始稀疏门控模块进行训练,得到该目标稀疏门控模块,包括:对该初始稀疏门控模块进行多次训练,得到该目标稀疏门控模块。
其中,当对稀疏门控模块X t-1进行第t次训练时,将第四样本数据输入至该目标编码网络和预设信道中进行处理,得到第五样本数据。将该第五样本数据输入至该第一目标信道特征提取网络中进行处理,得到信道分布信息训练数据和信道类别信息训练数据。将该信道分布信息训练数据和该信道类别信息训练数据均输入至该稀疏门控模块X t-1中进行处理,得到与该信道分布信息训练数据对应的R个第一目标分支自适应层样本和该R个第一目标分支自适应层的权重样本,R为正整数,R不大于N。根据该R个第一目标分支自适应层样本和该R个第一目标分支自适应层的权重样本对该第五样本数据进行处理,得到处理后的训练数据。将该处理后的训练数据输入至该目标解码网络中进行处理,得到第六样本数据。然后,根据该第四样本数据、该第六样本数据和该R个第一目标分支自适应层的权重样本计算得到第三损失值,并根据该第三损失值调整该稀疏门控模块X t-1的参数。令t=t+1,并重复执行上述步骤,直到达到停止条件,将该稀疏门控模块X t-1作为该目标稀疏门控模块。
第九方面,本申请提供一种通信装置,包括:通信模块,用于接收第一参考信号;处理模块,用于将该第一参考信号输入至第一目标模型中进行处理,得到该第一目标模型中的K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重,该第一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数;该通信模块,还用于发送第一信息,该第一信息指示该K个第一目标分支自适应层和该K个第一目标分支自适应层的权重,该K个第一目标分支自适应层和该K个第一目标分支自适应层的权重用于发送端对待发送的 编码数据进行处理。
在一种可能的实现方式中,该通信模块,还用于:接收第一数据,该第一数据是该发送端对待发送的编码数据进行处理得到的;该处理模块,还用于将该第一数据输入至该K个第一目标分支自适应层中进行处理,得到处理后的数据。
在一种可能的实现方式中,该通信模块,还用于:发送第一信道字典集,该第一信道字典集用于该发送端确定该N个第一目标分支自适应层和该发送端的M个第二目标分支自适应层之间的对应关系,M为正整数。
在另一种可能的实现方式中,该通信模块,还用于:接收第二信道字典集;该处理模块,还用于将第一信道字典集和该第二信道字典集进行字典对齐,得到该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系;该通信模块,还用于发送第一指示信息,该第一指示信息用于指示该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,该第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该通信模块,还用于:接收第二参考信号;发送第三信息,该第三信息与该第一信息相同,或者,该第三信息指示该发送端间隔第一时间后再发送参考信号,该第三信息是根据该第二参考信号得到的;接收该发送端在该第一时间内发送的数据。
第十方面,本申请提供了一种通信装置,包括:
通信模块,用于接收第一参考信号;
处理模块,用于将该第一参考信号输入至第一目标模型中进行处理,得到该第一目标模型中的K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重,该第一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数;
该通信模块,还用于发送第四信息,该第四信息指示K个第二目标分支自适应层和该K个第二目标分支自适应层的权重,该K个第二目标分支自适应层和该K个第二目标分支自适应层的权重用于发送端对待发送的编码数据进行处理,该K个第二目标分支自适应层和该K个第二目标分支自适应层的权重是根据该K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重确定的。
在一种可能的实现方式中,该通信模块,还用于:
接收第一数据,该第一数据是该发送端对待发送的编码数据进行处理得到的;
该处理模块,还用于将该第一数据输入至该K个第一目标分支自适应层中进行处理,得到处理后的数据。
在一种可能的实现方式中,该通信模块,还用于:
发送第一信道字典集,该第一信道字典集用于该发送端确定该N个第一目标分支自适应层和该发送端的M个第二目标分支自适应层之间的对应关系,M为正整数;
接收第一指示信息,该第一指示信息用于指示该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,该通信模块,还用于:
接收第二信道字典集;
该处理模块,还用于将第一信道字典集和该第二信道字典集进行字典对齐,得到该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间 的对应关系。
在一种可能的实现方式中,该第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该通信模块,还用于:
接收第二参考信号;
发送第五信息,该第五信息与该第四信息相同,或者,该第五信息指示该发送端间隔第一时间后再发送参考信号,该第五信息是根据该第二参考信号得到的;
接收该发送端在该第一时间内发送的数据。
第十一方面,本申请提供了一种通信装置,包括:通信模块,用于发送第一参考信号;该通信模块,还用于接收第一信息,该第一信息指示该第一参考信号接收端的第一目标模型中的K个第一目标分支自适应层和该K个第一目标分支自适应层的权重,该第一信息是根据该第一参考信号得到的,K为正整数。
在一种可能的实现方式中,该装置还包括处理模块,用于:根据该第一信息,确定与该K个第一目标分支自适应层对应的K个第二目标分支自适应层和该K个第二目标分支自适应层的权重;该处理模块,还用于将待发送的编码数据输入至第二目标模型的该K个第二目标分支自适应层中进行处理,得到第一数据;该通信模块,还用于发送该第一数据。
在一种可能的实现方式中,当M小于K时,该处理模块,还用于更新第三目标模型,得到该第二目标模型,使得该第二目标模型包括该K个第二目标分支自适应层,其中该第三目标模型包括M个第二目标分支自适应层,M为正整数。
在一种可能的实现方式中,该通信模块,还用于:接收第一信道字典集;该处理模块,还用于将该第一信道字典集和第二信道字典集进行字典对齐,得到该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
在另一种可能的实现方式中,该通信模块,还用于:发送第二信道字典集;接收第一指示信息,该第一指示信息用于指示该第二信道字典集中的第二目标分支自适应层与第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,该第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该通信模块,还用于:发送第二参考信号;接收第五信息,该第五信息与该第四信息相同,或者,该第五信息指示间隔第一时间后再发送参考信号,该第五信息是根据该第二参考信号得到的;在该第一时间内发送数据。
第十二方面,本申请提供了一种通信装置,包括:
通信模块,用于发送第一参考信号;
该通信模块,还用于接收第四信息,该第四信息指示K个第二目标分支自适应层和该K个第二目标分支自适应层的权重,该第四信息是根据该第一参考信号得到的,K为正整数。
在一种可能的实现方式中,该装置还包括处理模块,用于:
根据该第四信息,将待发送的编码数据输入至第二目标模型的该K个第二目标分支自适应层中进行处理,得到第一数据;
该通信模块,还用于发送该第一数据。
在一种可能的实现方式中,该通信模块,还用于:
接收第一信道字典集;
该处理模块,还用于将该第一信道字典集和第二信道字典集进行字典对齐,得到该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系;
该通信模块,还用于发送第一指示信息,该第一指示信息用于指示该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,该通信模块,还用于:
发送第二信道字典集,该第二信道字典集用于参考信号接收端确定该参考信号接收端的N个第一目标分支自适应层和发送端的M个第二目标分支自适应层之间的对应关系,M、N均为正整数。
在一种可能的实现方式中,该第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该通信模块,还用于:
发送第二参考信号;
接收第五信息,该第五信息与该第四信息相同,或者,该第五信息指示间隔第一时间后再发送参考信号,该第五信息是根据该第二参考信号得到的;
在该第一时间内发送数据。
第十三方面,本申请提供了一种通信装置,包括:通信模块,用于接收第一参考信号;处理模块,用于将该第一参考信号输入至第一目标模型中进行处理,得到该第一目标模型中的K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重,该第一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数;该通信模块,还用于接收第一数据;该处理模块,还用于将该第一数据输入至该K个第一目标分支自适应层中进行处理,得到处理后的数据,其中,该K个第一目标分支自适应层的权重中最大权重不小于预设值。
在一种可能的实现方式中,该第一目标模型还包括第一目标信道特征提取网络和目标稀疏门控模块,该第一目标信道特征提取网络用于对该第一参考信号进行处理,得到信道分布信息和该信道分布信息对应的信道类别信息;该目标稀疏门控模块用于根据该信道分布信息和该信道类别信息计算得到该K个第一目标分支自适应层和该K个第一目标分支自适应层的权重。
在一种可能的实现方式中,该K个第一目标分支自适应层用于分别对该第一数据进行处理,得到该K个第一目标分支自适应层分别处理后的数据;该K个第一目标分支自适应层的权重用于对该K个第一目标分支自适应层分别处理后的数据进行加权求和处理,得到该处理后的数据。
在一种可能的实现方式中,该处理模块,还用于:当该K个第一目标分支自适应层的权重中最大权重小于该预设值时,对该第一数据进行解码处理,得到处理后的数据。
在一种可能的实现方式中,该通信模块,还用于:发送第一信息,该第一信息指示该K个第一目标分支自适应层和该K个第一目标分支自适应层的权重,该K个第一目标分支自适应层和该K个第一目标分支自适应层的权重用于发送端对待发送的编码数据进行处理得到该第一数据。
在一种可能的实现方式中,该通信模块,还用于:当该K个第一目标分支自适应层的权 重中最大权重小于该预设值,发送第二信息,该第二信息指示发送端直接发送该第一数据。
在一种可能的实现方式中,该通信模块,还用于:发送第一信道字典集,该第一信道字典集用于发送端确定该N个第一目标分支自适应层和该发送端的M个第二目标分支自适应层之间的对应关系,M为正整数。
在另一种可能的实现方式中,该通信模块,还用于:接收第二信道字典集;该处理模块,还用于将第一信道字典集和该第二信道字典集进行字典对齐,得到该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系;该通信模块,还用于发送第一指示信息,该第一指示信息用于指示该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,该通信模块,还用于:
发送第四信息,该第四信息指示K个第二目标分支自适应层和该K个第二目标分支自适应层的权重,该K个第二目标分支自适应层和该K个第二目标分支自适应层的权重用于发送端对待发送的编码数据进行处理得到该第一数据,该K个第二目标分支自适应层和该K个第二目标分支自适应层的权重是根据该K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重确定的。
在一种可能的实现方式中,该通信模块,还用于:
发送第一信道字典集;
接收第一指示信息,该第一指示信息用于指示该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
在另一种可能的实现方式中,该通信模块,还用于:
接收第二信道字典集;
该处理模块,还用于将第一信道字典集和该第二信道字典集进行字典对齐,得到该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,该第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该通信模块,还用于:接收第二参考信号;发送第三信息,该第三信息与该第一信息相同,或者,该第三信息指示发送端间隔第一时间后再发送参考信号,该第三信息是根据该第二参考信号得到的;接收该发送端在该第一时间内发送的数据。
在另一种可能的实现方式中,该通信模块,还用于:接收第二参考信号;发送第五信息,该第五信息与该第四信息相同,或者,该第五信息指示发送端间隔第一时间后再发送参考信号,该第五信息是根据该第二参考信号得到的;接收该发送端在该第一时间内发送的数据。
第十四方面,本申请提供一种通信装置,包括:通信模块,用于发送第一参考信号;处理模块,用于当接收到第一信息时,该第一信息指示该第一参考信号接收端的第一目标模型中的K个第一目标分支自适应层和该K个第一目标分支自适应层的权重,根据该第一信息,确定与该K个第一目标分支自适应层对应的K个第二目标分支自适应层和该K个第二目标分支自适应层的权重,该第一信息是根据该第一参考信号得到的,K为正整数;该处理模块,还用于将待发送的编码数据输入至第二目标模型的该K个第二目标分支自适应层中进行处理,得到第一数据;该通信模块,还用于发送该第一数据。
在一种可能的实现方式中,当M小于K时,该处理模块,还用于更新第三目标模型,得 到该第二目标模型,使得该第二目标模型包括该K个第二目标分支自适应层,其中该第三目标模型包括M个第二目标分支自适应层,M为正整数。
在一种可能的实现方式中,该K个第二目标分支自适应层用于分别对该待发送的编码数据进行处理,得到该K个第二目标分支自适应层分别处理后的数据;该K个第二目标分支自适应层的权重用于对该K个第二目标分支自适应层分别处理后的数据进行加权求和处理,得到该第一数据。
在一种可能的实现方式中,该通信模块,还用于:当接收到第二信息时,该第二信息指示直接发送该待发送的编码数据,则发送该待发送的编码数据,该第二信息是根据该第一参考信号得到的。
在一种可能的实现方式中,该通信模块,还用于:接收第一信道字典集;该处理模块,还用于将该第一信道字典集和第二信道字典集进行字典对齐,得到该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系。
在另一种可能的实现方式中,该通信模块,还用于:发送第二信道字典集;接收第一指示信息,该第一指示信息用于指示该第二信道字典集中的第二目标分支自适应层与第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,该第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该通信模块,还用于:发送第二参考信号;接收第三信息,该第三信息与该第一信息相同;或者,该第三信息用于指示间隔第一时间后再发送参考信号,该第三信息是根据该第二参考信号得到的;在该第一时间内发送数据。
第十五方面,本申请提供一种通信装置,包括:
通信模块,用于发送第一参考信号;
处理模块,用于当接收到第四信息时,该第四信息指示K个第二目标分支自适应层和该K个第二目标分支自适应层的权重,根据该第四信息,将待发送的编码数据输入至第二目标模型的该K个第二目标分支自适应层中进行处理,得到第一数据,该第四信息是根据该第一参考信号得到的,K为正整数;
该通信模块,还用于发送该第一数据。
在一种可能的实现方式中,该K个第二目标分支自适应层用于分别对该待发送的编码数据进行处理,得到该K个第二目标分支自适应层分别处理后的数据;该K个第二目标分支自适应层的权重用于对该K个第二目标分支自适应层分别处理后的数据进行加权求和处理,得到该第一数据。
在一种可能的实现方式中,该通信模块,还用于:
当接收到第二信息时,该第二信息指示直接发送该待发送的编码数据,则发送该待发送的编码数据,该第二信息是根据该第一参考信号得到的。
在一种可能的实现方式中,该通信模块,还用于:
接收第一信道字典集;
该处理模块,还用于将该第一信道字典集和第二信道字典集进行字典对齐,得到该第二信道字典集中的第二目标分支自适应层与该第一信道字典集中的第一目标分支自适应层之间的对应关系;
该通信模块,还用于发送第一指示信息,该第一指示信息用于指示该第二信道字典集中 的第二目标分支自适应层与第一信道字典集中的第一目标分支自适应层之间的对应关系。
在另一种可能的实现方式中,该通信模块,还用于:
发送第二信道字典集,该第二信道字典集用于参考信号接收端确定该参考信号接收端的N个第一目标分支自适应层和发送端的M个第二目标分支自适应层之间的对应关系,M、N均为正整数。
在一种可能的实现方式中,该第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该处理模块,还用于:
发送第二参考信号;
接收第五信息,该第五信息与该第四信息相同;或者,该第五信息用于指示间隔第一时间后再发送参考信号,该第五信息是根据该第二参考信号得到的;
在该第一时间内发送数据。
第十六方面,本申请实施例提供一种目标模型训练装置。该目标模型包括目标编码网络、目标解码网络和第一目标模型,该第一目标模型包括第一目标信道特征提取网络、N个第一目标分支自适应层和目标稀疏门控模块。该装置包括:第一训练模块,用于对初始信道特征提取网络进行训练,得到该第一目标信道特征提取网络以及第一信道字典集;第二训练模块,用于根据该目标编码网络和该目标解码网络,对N个初始分支自适应层进行训练,得到该N个第一目标分支自适应层,该N为该第一信道字典集中的第一目标分支自适应层的个数;第三训练模块,用于根据该目标编码网络、该目标解码网络、该第一目标信道特征提取网络和该N个第一目标分支自适应层,对初始稀疏门控模块进行训练,得到该目标稀疏门控模块。
可选的,该第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该第一训练模块,用于:对该初始信道特征提取网络进行多次训练,得到该第一目标信道特征提取网络以及第一信道字典集。其中,该初始信道特征提取网络是根据带标签的历史信道信息或者预设信道模型进行预训练得到的。
其中,当对信道特征提取网络U t-1进行第t次训练时,将无标签的信道信息与该带标签的历史信道信息均输入至信道特征提取网络U t-1中进行处理,得到该无标签的信道信息与该带标签的历史信道信息分别对应的信道分布信息。根据该信道分布信息得到N t-1个聚类中心。根据该N t-1个聚类中心和该信道分布信息,得到预测的信道字典集。然后,根据该预测的信道字典集与标注的信道字典集计算得到第一损失值,并根据该第一损失值调整该信道特征提取网络U t-1的参数。令t=t+1,并重复执行上述步骤,直到迭代次数达到预设次数,将该信道特征提取网络U t-1作为该第一目标信道特征提取网络,将该预测的信道字典集作为该第一信道字典集。
在一种可能的实现方式中,第二训练模块,用于:对于第m个初始分支自适应层进行第t次训练时,将第一样本数据输入至该目标编码网络和预设信道中进行处理,得到第二样本数据。将该第二样本数据输入至该第m个初始分支自适应层中进行处理,得到该第m个初始分支自适应层处理得到的数据,m为正整数,m不大于N。将该第m个初始分支自适应层处理得到的数据输入至该目标解码网络中进行处理,得到第三样本数据。然后,根据该第一样本数据和该第三样本数据计算得到第二损失值,并根据该第二损失值调整该第m个初始分支自适应层参数。令t=t+1,并重复上述步骤,直到达到停止条件,将该第m个初始分支自适应层作为第m个第一目标分支自适应层。
令m=m+1,并重复执行上述步骤,直到m=N,得到该N个第一目标分支自适应层。
在一种可能的实现方式中,该装置还包括通信模块,用于发送第二指示信息,该第二指示信息指示该第m个初始分支自适应层进行训练。
在一种可能的实现方式中,第三训练模块,用于:对该初始稀疏门控模块进行多次训练,得到该目标稀疏门控模块。
其中,当对稀疏门控模块X t-1进行第t次训练时,将第四样本数据输入至该目标编码网络和预设信道中进行处理,得到第五样本数据。将该第五样本数据输入至该第一目标信道特征提取网络中进行处理,得到信道分布信息训练数据和信道类别信息训练数据。将该信道分布信息训练数据和该信道类别信息训练数据均输入至该稀疏门控模块X t-1中进行处理,得到与该信道分布信息训练数据对应的R个第一目标分支自适应层样本和该R个第一目标分支自适应层的权重样本,R为正整数,R不大于N。根据该R个第一目标分支自适应层样本和该R个第一目标分支自适应层的权重样本对该第五样本数据进行处理,得到处理后的训练数据。将该处理后的训练数据输入至该目标解码网络中进行处理,得到第六样本数据。然后,根据该第四样本数据、该第六样本数据和该R个第一目标分支自适应层的权重样本计算得到第三损失值,并根据该第三损失值调整该稀疏门控模块X t-1的参数。令t=t+1,并重复执行上述步骤,直到达到停止条件,将该稀疏门控模块X t-1作为该目标稀疏门控模块。
针对上述第九方面至第十五方面中的任一方面,在一种可能的实现方式中,所述处理模块可以为处理器,通信模块可以为收发模块、收发器或通信接口。可以理解的,所述通信模块可以是所述装置中的收发器,例如通过所述装置中的天线、馈线和编解码器等实现,或者,如果通信装置为设置在设备中的芯片,则通信模块可以是该芯片的输入/输出接口,例如输入/输出电路、管脚等。
针对上述第十六方面中的任一实现方式,在一种可能的实现方式中,所述第一训练模块、所述第二训练模块和所述第三训练模块可以均为处理器。
在一种可能的实现方式中,通信模块可以为收发模块、收发器或通信接口。可以理解的,所述通信模块可以是所述装置中的收发器,例如通过所述装置中的天线、馈线和编解码器等实现,或者,如果通信装置为设置在设备中的芯片,则通信模块可以是该芯片的输入/输出接口,例如输入/输出电路、管脚等。
第十七方面,本申请实施例提供一种通信装置,所述通信装置包括一个或多个处理器;其中,所述一个或多个处理器用于执行一个或多个存储器存储的计算机程序,使得所述通信装置实现如第一方面任一项所述的方法,或,实现如第二方面任一项所述的方法,或,实现如第三方面任一项所述的方法,或,实现如第四方面任一项所述的方法,或,实现如第五方面任一项所述的方法,或,实现如第六方面任一项所述的方法,或,实现如第七方面任一项所述的方法。
在一种可能的实现方式中,所述通信装置还包括所述一个或多个存储器。
在一种可能的实现方式中,所述通信装置为芯片或芯片系统。
第十八方面,本申请实施例提供一种计算机可读存储介质,所述计算机可读存储介质存储有指令,当所述指令被处理器执行时,实现如第一方面任一项所述的方法,或,实现如第二方面任一项所述的方法,或,实现如第三方面任一项所述的方法,或,实现如第四方面任一项所述的方法,或,实现如第五方面任一项所述的方法,或,实现如第六方面任一项所述 的方法,或,实现如第七方面任一项所述的方法。
第十九方面,本申请实施例提供一种计算机程序产品,包括计算机程序,当所述计算机程序被执行时,实现如第一方面任一项所述的方法,或,实现如第二方面任一项所述的方法,或,实现如第三方面任一项所述的方法,或,实现如第四方面任一项所述的方法,或,实现如第五方面任一项所述的方法,或,实现如第六方面任一项所述的方法,或,实现如第七方面任一项所述的方法。
第二十方面,本申请实施例提供一种通信系统,包括如第九方面任一项所述的装置或包括如第十方面任一项所述的装置,还包括如第十二方面任一项所述的装置;或者,该通信系统包括如第十一方面任一项所述的装置,以及如第十三方面任一项所述的装置;或者,该通信系统包括如第十四方面任一项所述的装置,以及如第十五方面任一项所述的装置或如第十六方面任一项所述的装置。
可以理解地,上述提供的第九方面所述的装置、第十方面所述的装置、第十一方面所述的装置、第十二方面所述的装置、第十三方面所述的装置、第十四方面所述的装置、第十五方面所述的装置、第十六方面所述的装置、第十七方面所述的装置、第十八方面所述的计算机存储介质或者第十九方面所述的计算机程序产品、第二十方面所述的系统,均用于执行第一方面中任一所提供的方法、第二方面中任一所提供的方法、第三方面中任一所提供的方法、第四方面中任一所提供的方法、第五方面中任一所提供的方法、第六方面中任一所提供的方法、第七方面中任一所提供的方法或第八方面中任一所提供的方法。因此,其所能达到的有益效果可参考对应方法中的有益效果,此处不再赘述。
附图说明
下面对本申请实施例用到的附图进行介绍。
图1a是本申请实施例提供的一种通信系统的架构示意图;
图1b是本申请实施例提供的一种通信系统的架构示意图;
图1c是本申请实施例提供的一种通信系统的架构示意图;
图2a是本申请实施例提供的一种通信方法的流程示意图;
图2b是本申请实施例提供的一种目标稀疏门控模块示意图;
图3是本申请实施例提供的一种通信系统示意图;
图4是本申请实施例提供的另一种通信方法的流程示意图;
图5a是本申请实施例提供的另一种通信系统示意图;
图5b是本申请实施例提供的一种帧结构示意图;
图6a是本申请实施例提供的一种目标模型训练方法的流程示意图;
图6b是本申请实施例提供的另一种帧结构示意图;
图7a是本申请实施例提供的一种通信装置的结构示意图;
图7b是本申请实施例提供的另一种通信装置的结构示意图;
图8是本申请实施例提供的又一种通信装置的结构示意图。
具体实施方式
下面结合本申请实施例中的附图对本申请实施例进行描述。本申请实施例的实施方式部 分使用的术语仅用于对本申请的具体实施例进行解释,而非旨在限定本申请。
由于现有神经网络算法并不能解决实际环境中的信道动态变化的问题,有鉴于此,本申请提供一种通信方法、装置及系统,能够应对信道动态变化,实现对未知信道的迁移适配。
以下将结合附图,来详细介绍本申请实施例的系统架构。本申请的通信方法可以应用于蜂窝通信系统。请参见图1a,图1a是本申请实施例适用的一种通信系统的示意图,该系统包括网络设备101和终端102。
网络设备101可以是基站(base station)、演进型基站(evolved NodeB,eNodeB)、发送接收点(transmission reception point,TRP)、第五代(5th generation,5G)移动通信系统中的下一代基站(next generation NodeB,gNB)、第六代(6th generation,6G)移动通信系统中的下一代基站、未来移动通信系统中的基站或WiFi系统中的接入节点等;也可以是完成基站部分功能的模块或单元,例如,可以是集中式单元(central unit,CU),也可以是分布式单元(distributed unit,DU)。网络设备101可以是宏基站,也可以是微基站或室内站,还可以是中继节点或施主节点等。本申请的实施例对网络设备101所采用的具体技术和具体设备形态不做限定。为了便于描述,下文以基站作为网络设备的例子进行描述。
该终端102也可以称为终端设备、用户设备(user equipment,UE)、移动台、移动终端等。终端可以广泛应用于各种场景,例如,设备到设备(device-to-device,D2D)、车物(vehicle to everything,V2X)通信、机器类通信(machine-type communication,MTC)、物联网(internet of things,IOT)、虚拟现实、增强现实、工业控制、自动驾驶、远程医疗、智能电网、智能家具、智能办公、智能穿戴、智能交通、智慧城市等。终端可以是手机、平板电脑、带无线收发功能的电脑、可穿戴设备、车辆、无人机、直升机、飞机、轮船、机器人、机械臂、智能家居设备等。本申请的实施例对终端所采用的具体技术和具体设备形态不做限定。
基站和终端可以是固定位置的,也可以是可移动的。基站和终端可以部署在陆地上,包括室内或室外、手持或车载;也可以部署在水面上;还可以部署在空中的飞机、气球和人造卫星上。本申请的实施例对基站和终端的应用场景不做限定。
基站和终端之间、基站和基站之间、终端和终端之间可以通过授权频谱进行通信,也可以通过免授权频谱进行通信,也可以同时通过授权频谱和免授权频谱进行通信;可以通过6千兆赫(gigahertz,GHz)以下的频谱进行通信,也可以通过6GHz以上的频谱进行通信,还可以同时使用6GHz以下的频谱和6GHz以上的频谱进行通信。本申请的实施例对无线通信所使用的频谱资源不做限定。
在本申请的实施例中,基站的功能也可以由基站中的模块(如芯片)来执行,也可以由包含有基站功能的控制子系统来执行。这里的包含有基站功能的控制子系统可以是智能电网、工业控制、智能交通等上述应用场景中的控制中心。终端的功能也可以由终端中的模块(如芯片或调制解调器)来执行,也可以由包含有终端功能的装置来执行。
示例地,本方案可以应用于5G、卫星通信等无线通信系统中。无线通信系统通常由小区组成,每个小区包含一个基站(Base Station,BS),基站向多个移动台(Mobile Station,MS)提供通信服务。其中基站包含基带单元(Baseband Unit,BBU)和远端射频单元(Remote Radio Unit,RRU)。BBU和RRU可以放置在不同的地方,例如:RRU拉远,放置于高话务量的区域,BBU放置于中心机房。BBU和RRU也可以放置在同一机房。BBU和RRU也可以为一个机架下的不同部件。
其中,上述无线通信系统包括但不限于:窄带物联网系统(Narrow Band-Internet of Things, NB-IoT)、全球移动通信系统(Global System for Mobile Communications,GSM)、增强型数据速率GSM演进系统(Enhanced Data rate for GSM Evolution,EDGE)、宽带码分多址系统(Wideband Code Division Multiple Access,WCDMA)、码分多址2000系统(Code Division Multiple Access,CDMA2000)、时分同步码分多址系统(Time Division-Synchronization Code Division Multiple Access,TD-SCDMA),长期演进系统(Long Term Evolution,LTE)以及下一代5G移动通信系统的三大应用场景:增强移动宽带(Enhanced Mobile Broadband,eMBB),超高可靠与低时延通信(Ultra-reliable and Low Latency Communications,URLLC)和基于LTE演进的物联网技术(LTE enhanced MTO,eMTC)。
请参见图1b,图1b是本申请实施例适用的另一种通信系统的示意图,该系统包括卫星基站103和终端104。
其中,卫星基站103可以是无人机、热气球、低轨卫星、中轨卫星或高轨卫星等。或者,卫星基站103也可以是指非地面的基站或非地面的设备等。卫星基站103既可以作为网络设备,也可以作为终端设备。卫星基站103可以不具备基站的功能,也可以具备部分或者全部基站的功能,本申请对此不做限定。
针对终端104的介绍可参阅图1a中终端102的记载,在此不再赘述。
卫星基站103可以为终端104提供通信服务。其中,卫星基站103向终端104传输下行数据。该数据采用信道编码进行编码。该信道编码后的数据经过星座调制后传输给终端104。终端104向卫星基站103传输上行数据。该上行数据也可以采用信道编码进行编码。编码后的数据经过星座调制后传输给卫星基站103。
本申请的通信方法还可以应用于星间通信系统。请参见图1c,图1c是本申请实施例适用的又一种通信系统的示意图,该系统包括卫星105和卫星106。
传统的卫星星间链路通信系统可以分为:空间光束捕获跟踪对准技术(acquisition,pointing and tracking,APT)子系统和通信子系统两大部分。通信子系统负责星间信息的传输,是星间通信系统的主体;APT子系统负责卫星之间的捕获、对准和跟踪。其中,捕获是指确定入射信号的来波方向;对准是指调整发射波瞄准接收方向;跟踪是指在整个通信过程中,不断调整对准和捕获。为了尽量减少信道中的衰减和干扰影响,同时要求具有较高的保密性和传输率,必须实时的调整APT子系统来不断适应变化。现有的APT系统均为光学系统,缺点在于光学对准难度大,需要机械调整指向。现有的通信子系统,多数为光通信系统,也有部分微波波段的系统,多采用单个高增益天线。现有的APT系统和通信子系统为独立的系统。缺点在于光通信容易受震动等影响,速率不稳定;毫米波频率低,通信容量低,天线需要机械调整指向。
如图1c所示,卫星105和卫星106均包括通信模块、收发天线、APT模块、APT发射/接收。本申请的通信方法可应用于该通信模块。
综上,本方案的通信方法可以部署在网络设备101端、或者卫星基站103端、或者卫星105端,还可以部署在终端102、终端104或者卫星106等,本方案对此不作限制。
例如,当部署在卫星基站103端时,通过多分支自适应层来对应不同信道分布下的终端网络,从而提高性能。
应理解,上述图中所示的各设备的数量为示意性的,根据实际场景的需求,可以具有任意数量的设计,本方案对此不作限制。
上面说明了本申请实施例的架构,下面对本申请实施例的方法进行详细介绍。
参照图2a所示,是本申请实施例提供的一种通信方法的流程示意图。可选的,该方法可以应用于前述的通信系统,例如图1a所示的通信系统。如图2a所示的通信方法可以包括步骤201-206。应理解,本申请为了方便描述,故通过201-206这一顺序进行描述,并不旨在限定一定通过上述顺序进行执行。本申请实施例对于上述一个或多个步骤的执行的先后顺序、执行的时间、执行的次数等不做限定。下文以通信方法的步骤201和204的执行主体为基站、202、203、205和206的执行主体为终端为例进行描述,对于其他执行主体本申请同样也适用。步骤201-206具体如下:
201、基站发送第一参考信号;
该第一参考信号可以是导频信号。导频信号是基站连续发射未经调制的直接序列扩频信号。该导频信号使得终端能够获得前向码分多址信道时限,提供相关解调相位参考等。
202、终端接收该第一参考信号;
终端接收来自基站的该第一参考信号。
203、终端将所述第一参考信号输入至第一目标模型中进行处理,得到所述第一目标模型中的K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重,所述第一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数;
其中,本申请实施例中的每个第一目标分支自适应层,可以理解为,其分别对应不同种类的信道分布。也即,N个第一目标分支自适应层对应N种信道分布的类型。
当然,还可以理解为,N个第一目标分支自适应层对应N’种聚类的信道分布的类型,其中,N大于N’。也即,可以是至少一个第一目标分支自适应层对应一种聚类的信道分布的类型。本方案对此不作严格限制。该聚类,可以理解为,将信道分布相似或相同的信道分布合并为一个信道分布的类型。
也就是说,通过将第一参考信号输入至第一目标模型中进行处理,可以确定该第一目标模型的N个第一目标分支自适应层中的K个第一目标分支自适应层,以及该K个第一目标分支自适应层的权重。也即,当前的信道分布可以采用该确定的K个第一目标分支自适应层对应的信道分布的类型及其对应的权重来综合表征。
在一种可能的实现方式中,所述第一目标模型还包括第一目标信道特征提取网络和目标稀疏门控模块。
其中,所述第一目标信道特征提取网络用于对所述第一参考信号进行处理,得到信道分布信息和所述信道分布信息对应的信道类别信息。
该信道分布信息,可以是用来表征当前信道分布在特征空间中的特征向量。
该信道分布信息对应的信道类别信息,可以是对当前的信道分布所属的信道类型进行预测的概率。
所述目标稀疏门控模块用于根据所述信道分布信息和所述信道类别信息计算得到所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重。
目标稀疏门控模块基于上述信道分布信息和信道类别信息可以得出当前信道分布对应的类型。其通过K个第一目标分支自适应层及其对应的权重来综合表征。
可选的,目标稀疏门控模块基于如下方式可计算得到各个第一目标分支自适应层的权重:
α=softmax(TopK(W SG*h+p cluster));
其中softmax为归一化指数函数;W SG为目标稀疏门控模块的参数。h为信道分布信息。p cluster为信道分布信息对应的信道类别信息。α为分支自适应层的权重。TopK表示最终选择的 分支个数。例如K=3,则表示α 12,…,α N中只有3个对应有数值,其余为0。这样可以使得网络的推演时间不随分支个数增加而增大。
当然还可以采用其他方式计算得到各个第一目标分支自适应层的权重,本方案对此不作限制。
在一种可能的实现方式中,所述K个第一目标分支自适应层用于分别对所述第一数据进行处理,得到所述K个第一目标分支自适应层分别处理后的数据。
例如,该每个第一目标分支自适应层所做的处理是对第一数据做仿射变换。例如,第一数据是f(x)∈C 72×14×1×1,该维度表示72个子载波,14个符号,1个接收端的天线,1个发送端的天线的复数域。该每个第一目标分支自适应层分别对应一组参数(W i,b i),W∈C 72×14×1×1,b∈C 72×14×1×1。各分支自适应层所对应的仿射变换运算可以表示为:t i(f(x))=W i·f(x)+b i。i表示该K个第一目标分支自适应层的标号,i为正整数。
当然,该每个第一目标分支自适应层所做的处理还可以是其他处理,本方案对此不作严格限制。
所述K个第一目标分支自适应层的权重用于对所述K个第一目标分支自适应层分别处理后的数据进行加权求和处理,得到所述处理后的数据。
例如,该加权求和处理可表示为
Figure PCTCN2022132654-appb-000001
α i表示该K个第一目标分支自适应层的权重。
204、基站发送第一数据;
205、终端接收第一数据;
终端接收来自基站的第一数据。
206、终端将所述第一数据输入至所述K个第一目标分支自适应层中进行处理,得到处理后的数据,其中,所述K个第一目标分支自适应层的权重中最大权重不小于预设值。
例如,通过将K个第一目标分支自适应层的权重进行排序,找到其中最大的权重值。当该最大权重值不小于预设值,则表明该得到的K个第一目标分支自适应层的可靠性较高。因此,将接收到的第一数据输入到该K个第一目标分支自适应层中进行处理。
该预设值可以是任意不大于1的正数,当然,其还可以为0,本方案对此不作限制。可选的,该预设值取值为0.8。
可选的,该处理可以是对输入的第一数据做仿射变换。例如,基站侧对应正交频分复用技术(Orthogonal Frequency Division Multiplexing,OFDM)波形输出的第一数据f(x)∈C 72×14×1×1,各分支自适应层所对应的仿射变换运算可以表示为:t i(f(x))=W i·f(x)+b i
基于各个第一目标分支自适应层所做的上述处理,以及各个第一目标分支自适应层对应的权重,通过进行加权求和处理,可计算得到:
Figure PCTCN2022132654-appb-000002
针对该部分介绍可参阅步骤203的介绍,在此不再赘述。
在一种可能的实现方式中,该目标稀疏门控模块还可以包含多个稀疏门控子模块。如图2b所示,通过对时延扩展与速度分级,分支级别A通过训练不同时延扩展的情况(例如分支A1:30ns;分支A2:100ns),分支级别B训练不同速度的情况(例如分支B1:3km/s;分支B2:30km/s)。最后加权求和处理可表示为:
Figure PCTCN2022132654-appb-000003
t Bi、t Ai分别表示各分支自适应层所对应的仿射变换运算。i的取值范围为1到N的整数,α Ai为分支级别A的各分支自适应层的权重,β Bi为分支级别B的各分支自适应层的权重。
也就是说,通过多个稀疏门控子模块来组合实现对信道分布的多级动态刻画。这样通过 多个参数空间内的组合,得到非线性的信道分布刻画,提高了泛化迁移能力。
上述处理后的数据,可以是对该数据T进行解码处理后的数据,例如,通过将数据T输入到目标解码网络中进行解码处理,可得到上述解码后的数据。
如图3所示,为本申请实施例提供的一种通信系统的示意图。该第一目标模型包括第一目标信道特征提取网络、目标稀疏门控模块和N个第一目标分支自适应层。终端基于接收到的第一参考信号输入到第一目标信道特征提取网络中进行处理,得到信道分布信息和信道类别信息。然后,将该信道分布信息和信道类别信息输入到目标稀疏门控模块进行处理,得到N个第一目标分支自适应层中的K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重。当该K个第一目标分支自适应层的权重中最大权重不小于预设值,终端则将接收到的第一数据输入到该K个第一目标分支自适应层中进行处理,并进行加权求和计算,然后将得到的数据T输入到目标解码网络中进行解码处理,即得到上述处理后的数据。
在一种可能的实现方式中,当该K个第一目标分支自适应层的权重中最大权重小于预设值,则终端直接对该第一数据进行解码处理,得到处理后的数据。
当该最大权重值小于预设值,则表明确定的该K个第一目标分支自适应层的可靠性较低。因此,如图3所示的回退操作,终端将接收到的第一数据直接进行解码,而不经过该K个第一目标分支自适应层进行处理。
可选的,还可以保存当前信道数据,并将该当前信道数据作为训练数据,来重新训练第一目标模型,以得到性能更好的模型。针对第一目标模型的训练的介绍可参阅后续模型训练部分的记载,在此不再赘述。
需要说明的是,该实施例以基站作为发端、终端作为收端为例进行介绍。其还可以是终端作为发端、基站作为收端,其处理过程与上述一致,在此不再赘述。
本申请实施例,终端基于接收到的参考信号确定第一目标模型的N个第一目标分支自适应层中的K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重。当该K个第一目标分支自适应层的权重中最大权重不小于预设值,通过将接收到的第一数据输入至该K个第一目标分支自适应层中进行处理,得到处理后的数据。采用该手段,基于确定的K个第一目标分支自适应层和K个权重来表征实际的信道分布,以应对信道的动态变化,这样可以提高系统性能。
参照图4所示,是本申请实施例提供的另一种通信方法的流程示意图。可选的,该方法可以应用于前述的通信系统,例如图1a所示的通信系统。如图4所示的通信方法可以包括步骤401-409。应理解,本申请为了方便描述,故通过401-409这一顺序进行描述,并不旨在限定一定通过上述顺序进行执行。本申请实施例对于上述一个或多个步骤的执行的先后顺序、执行的时间、执行的次数等不做限定。下文以通信方法的步骤401、405、406和407的执行主体为基站、步骤402-404、408和409的执行主体为终端为例进行描述,对于其他执行主体本申请同样也适用。步骤401-409具体如下:
401、基站发送第一参考信号;
针对该步骤的介绍可参阅图2a所示实施例中步骤201的记载,在此不再赘述。
402、终端接收第一参考信号;
针对该步骤的介绍可参阅图2a所示实施例中步骤202的记载,在此不再赘述。
403、终端将所述第一参考信号输入至第一目标模型中进行处理,得到所述第一目标模型中的K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重,所述第一目标 模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数;
针对该步骤的介绍可参阅图2a所示实施例中步骤203的记载,在此不再赘述。
404、当所述K个第一目标分支自适应层的权重中最大权重不小于预设值,终端发送第一信息,所述第一信息指示所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重,所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重用于基站对待发送的编码数据进行处理得到第一数据;
也就是说,当K个第一目标分支自适应层的权重中最大权重不小于预设值时,终端向基站发送信息,以便基站侧对待发送的编码数据进行处理后再发送。这样做,基站侧按照信息指示,选择对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
需要说明的是,本申请实施例仅以当所述K个第一目标分支自适应层的权重中最大权重不小于预设值,终端发送第一信息为例进行介绍。其还可以是在任何条件下,终端均发送第一信息。例如,该预设值取值为0,则终端均发送第一信息。本方案对此不作严格限制。
405、基站接收第一信息,并根据所述第一信息,确定与所述K个第一目标分支自适应层对应的K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重;
可选的,基站侧部署有第二目标模型。如图5a所示,为本申请实施例提供的另一种通信系统的示意图。其中,第二目标模型包括M个第二目标分支自适应层。第一目标模型包括上述第一目标信道特征提取网络、目标稀疏门控模块和N个第一目标分支自适应层。
该第二目标模型中包含该K个第二目标分支自适应层。该K个第二目标分支自适应层与上述K个第一目标分支自适应层对应。
也就是说,终端侧基于前述K个第一目标分支自适应层表征当前信道分布。终端侧将该结果反馈给基站侧,以便基站侧确定出表征当前信道分布的K个第二目标分支自适应层,以便对待发送的编码数据进行处理。
下面对基站侧确定与所述K个第一目标分支自适应层对应的K个第二目标分支自适应层的方式进行介绍。
示例1:终端向基站发送第一信道字典集。
基站接收该第一信道字典集,并将所述第一信道字典集和第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
这样,基站可以基于该对应关系确定与所述K个第一目标分支自适应层对应的K个第二目标分支自适应层。进而基于该K个第一目标分支自适应层的权重即可确定该K个第二目标分支自适应层的权重。
示例2:基站向终端发送第二信道字典集。
终端接收该第二信道字典集,并将第一信道字典集和该第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
然后,终端向基站发送第一指示信息。该第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
基站基于接收到的第一指示信息,即可确定与所述K个第一目标分支自适应层对应的K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重。
在示例1和示例2的基础上,可选的,该第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。该第二信道字典集包含第二目标分支自适应层与信道标签之间的 对应关系。
也即通过基于相同的信道标签可确定分支自适应层之间的对应关系。
当然,还可以采用其他方式来确定与所述K个第一目标分支自适应层对应的K个第二目标分支自适应层,本方案对此不作严格限制。
可选的,终端基于第一目标模型中的第一目标信道特征提取网络可得到第一信道字典集。其中,第二目标模型还包括第二目标信道特征提取网络。相应地,基站基于第二目标模型中的第二目标信道特征提取网络可得到第二信道字典集。
其中,进行上述字典对齐的操作,是由于信道特征提取网络的训练仅与信道训练集相关,发端和收端两端可以分别独立训练该模块。由于两侧的信道训练集可能不完全相同,得到的信道字典集可能不同,因此通过字典对齐的方式来实现各端自适应分支层的匹配。
例如,用带标签的历史信道信息或者第三代合作伙伴计划(The 3rd Generation Partnership Project,3GPP)协议中的抽头延时线(Tapped Delay Line,TDL)或者簇延时线(Clustered Delay Line,CDL)信道模型的采样数据{H,Y label}作为网络训练的信道数据集。H表示所述信道模型的采样,Y label表示所述信道模型的标签。通过输入信道数据H,输出分类结果:信道数据属于不同信道标签结果的概率
Figure PCTCN2022132654-appb-000004
根据
Figure PCTCN2022132654-appb-000005
与Y label计算交叉熵作为训练的损失函数L classify,进而更新网络。其中,损失函数L classify可表示为:
Figure PCTCN2022132654-appb-000006
N为参与训练的不同的信道标签的种类总数。
两端分别基于该损失函数对信道特征提取网络进行训练,得到不同的信道字典集。例如对于终端有D UE={(1:TDL-A30v3,TDL-C30v3),(2:TDL-A300v100)}。对于基站有D BS={(a:CDL-A30v100),(b:TDL-A30v3,TDL-C30v3),(c:TDL-A300v100,TDL-C300v100)}。其中,TDL-A30v3、TDL-C30v3、CDL-A30v100等均为信道标签。(1:TDL-A30v3,TDL-C30v3)表示信道标签为TDL-A30v3,TDL-C30v3的信道对应的信道类别(也即分支自适应层)的标号为1。(a:CDL-A30v100)表示信道标签为CDL-A30v100的信道对应的信道类别(也即分支自适应层)的标号为a。a、b、c以及前述1、2分别表示不同的信道类别的标号。例如TDL-A30v3表示时延扩展为30ns,速度为3km/s下的TDL-A信道模型。
例如,终端发送自身的信道字典集D UE。其中,若信道字典集中存在的信道标签不属于协议中的TDL或者CDL信道模型(例如自定义的历史信道信息建模出的模型),则终端需要额外发送最接近聚类中心点的信道数据h p,以便基站重定义信道分布信息。
基站接收该信道字典集D UE,并与自身得到的信道字典集D BS匹配,得到信道匹配集D m。例如,对于上述示例中,信道匹配集D m={(b:1),(c:2)}。也就是说,终端侧的信道自适应层1与基站侧的信道自适应层b对应。终端侧的信道自适应层2与基站侧的信道自适应层c对应。
对于额外信道数据h p,基站可通过输入自己的信道特征提取网络,获得信道分布信息,从而确定所属的信道标签。若存在某些无法匹配的情况,则说明在该信道下两端的多分支自适应层并不能完全匹配,则需要重新进行模型训练,以得到可以匹配的信道字典集。
基站还可以将信道匹配集D m反馈给终端供推演时使用。
其中,对于上述无法匹配的情况,例如当M小于K时,则基站更新第三目标模型,得到 第二目标模型,使得所述第二目标模型包括所述K个第二目标分支自适应层,其中所述第三目标模型包括M个第二目标分支自适应层。
也就是说,当基站侧仅部署了第三目标模型中的M个第二目标分支自适应层,其并不能完全包含与上述K个第一目标分支自适应层对应的分支自适应层。因此,通过对第三目标模型进行重新训练,更新该第三目标模型,得到第二目标模型。该第二目标模型可以满足包括该K个第二目标分支自适应层的条件。
这样,基站即可确定与所述K个第一目标分支自适应层对应的K个第二目标分支自适应层。
在一种可能的实现方式中,所述K个第二目标分支自适应层用于分别对所述待发送的编码数据进行处理,得到所述K个第二目标分支自适应层分别处理后的数据。
例如,该每个第二目标分支自适应层所做的处理是对待发送的编码数据做仿射变换。例如,待发送的编码数据是f(x)∈C 72×14×1×1,该维度表示72个子载波,14个符号,1个接收端的天线,1个发送端的天线的复数域。该每个第二目标分支自适应层分别对应一组参数(W j,b j),W∈C 72×14×1×1,b∈C 72×14×1×1。各分支自适应层所对应的仿射变换运算可以表示为:t j(f(x))=W j·f(x)+b j。j表示该K个第二目标分支自适应层的标号,j为正整数。
当然,该每个第二目标分支自适应层所做的处理还可以是其他处理,本方案对此不作严格限制。
所述K个第二目标分支自适应层的权重用于对所述K个第二目标分支自适应层分别处理后的数据进行加权求和处理,得到第一数据。
例如,该加权求和处理可表示为
Figure PCTCN2022132654-appb-000007
α j表示该K个第二目标分支自适应层的权重。
406、基站将待发送的编码数据输入至第二目标模型的所述K个第二目标分支自适应层中进行处理,得到第一数据;
针对该步骤的介绍可参阅步骤405的记载,在此不再赘述。
407、基站发送第一数据;
基站将经过K个第二目标分支自适应层处理后的数据发送给终端。
408、终端接收第一数据;
409、终端将所述第一数据输入至所述K个第一目标分支自适应层中进行处理,得到处理后的数据。
针对该步骤的介绍可参阅图2a所示实施例中步骤206的记载,在此不再赘述。
在一种可能的实现方式中,当所述K个第一目标分支自适应层的权重中最大权重小于所述预设值,终端发送第二信息,所述第二信息指示基站直接发送所述第一数据。
也就是说,基站不需要对待发送的编码数据(第一数据)基于第二目标分支自适应层进行处理。相应地,终端基于接收到的第一数据,直接解码处理,也不需要经过第一目标分支自适应层进行处理。
在一种可能的实现方式中,基站还发送第二参考信号。
终端接收该第二参考信号。然后,终端基于该第二参考信号发送第三信息。其中,该第三信息与第一信息相同。或者,该第三信息指示基站间隔第一时间后再发送参考信号。
也就是说,终端基于第二参考信号确定出当前信道的分布与前述第一参考信号对应的信 道的分布是一致的,也即该信道分布在对应第一参考信号到第二参考信号的该段时间内相对没有发生变化。终端则指示基站间隔一段时间后再发送参考信号。
可选的,当前子帧的信道分布信息未发生改变,如图5b所示,则令下一时隙(时隙#1)减少参考信号的发送。例如时隙#1中没有参考信号。
基站接收到该第三信息后,则在第一时间内发送数据。
该在第一时间内发送数据,可以理解为,基站在第一时间内不再发送参考信号。
相应地,终端接收在所述第一时间内发送的数据。也即,终端在所述第一时间内不再接收到参考信号。
这样,可以减少导频传输开销,增加数据传输的频谱效率。
在另一种可能的实现方式中,基站发送第二参考信号。终端接收该第二参考信号。然后,终端基于该第二参考信号发送第三信息。其中,该第三信息与第一信息不同。也就是说,当前信道分布发生改变。因此终端需要将新确定的各第一目标分支自适应层以及其权重反馈给基站。
可选的,当信道分布信息发生改变,例如权重改变了,如图5b所示,则以时隙#0的帧结构进行发送。
其中,当触发回退操作(最大权重小于预设值)时,则回退到原有3GPP标准中的数据传输帧结构。
在一种可能的实现方式中,本申请实施例中的信息的反馈维度与第一目标分支自适应层中分支的个数N、预设值Γ和目标稀疏门控模块的TopK值相关。
例如:
(1)第一目标分支自适应层的分支标号的比特bit数
Figure PCTCN2022132654-appb-000008
其中,
Figure PCTCN2022132654-appb-000009
表示对log 2N向上取整,N为第一目标分支自适应层的总数。
(2)量化bit数
Figure PCTCN2022132654-appb-000010
其中设定的量化损失为Δ。默认最大权重值放在第一个发送。
(3)将目标稀疏门控模块的TopK设置为K,表示选取权重值最大的K个权重。对于权重值计算:
设定最大权重计算公式为
Figure PCTCN2022132654-appb-000011
B为将上述量化的bit数转化为十进制的转化因子;
其余权重计算公式为
Figure PCTCN2022132654-appb-000012
其中,由于各权重的和为1,因此由前K-1个分支的权重计算得到最后一个分支的权重为
Figure PCTCN2022132654-appb-000013
B 1为最大权重值的十进制表示。B i为第i个权重值的十进制表示。
其中,当最大权重小于预设值时,设定用1bit指示回退操作。
基于上述设定,可计算反馈表所需反馈bit数为:K*N index+(K-1)*N quan。这样可以降低反馈信息开销。
上述参数可以在信道字典集中给定配置,以便可以实现收发端同步。
例如,N=6,Γ=0.8,Δ=0.02,TopK=3,则可设计如下控制信道反馈表,共需3*3+3*2=15bits表示。其中,各分支采用二进制进行表示。由于分支4的权重可基于前述分支1和分支2计 算得到,因此可以不进行反馈。
表一
分支标号(3bit) 权重(3bit)
分支1(001) 0.9(100)
分支2(010) 0.05(010)
分支4(100) 0.05(/)
其中,在时分双工(Time Division Duplexing,TDD)条件下,由于权重与信道分布信息相关,发送端可通过信道互易性,获得信道估计信息,并类似的可通过信道特征提取网络与稀疏门控模块直接估计分支权重。发送端可以通过上面的设计反馈给接收端同步,以保证收发两端分支一致。
其中,通过对采用本方案的通信方法的系统进行仿真,得到如下结果:通过增加多分支自适应层,可以提高高信噪比下混合信道训练的目标编码网络和目标解码网络在特定信道的性能。并且对于未知的信道,通过目标稀疏门控模块对聚类信道的组合,可以得到性能上的提升。
且,本方案关注信道分布类别的表达而非单帧信道反馈的表达。通过实验发现,本方案对信道的自适应调整对反馈时延的变化鲁棒,且对信道分布信息的输出维度鲁棒。
需要说明的是,该实施例以基站作为发端、终端作为收端为例进行介绍。其还可以是终端作为发端、基站作为收端,其处理过程与上述一致,在此不再赘述。
该实施例以参考信号接收端发送第一信息,该第一信息指示K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重为例进行介绍。
其中,需要说明的是,还可以是参考信号接收端发送第四信息。其中,所述第四信息指示K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重。所述K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重用于发送端对待发送的编码数据进行处理。
这样,发送端可直接根据该信息对待发送的编码数据进行处理即可。
其中,所述K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重是根据所述K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重确定的。
可选的,参考信号接收端发送第一信道字典集,所述第一信道字典集用于发送端确定N个第一目标分支自适应层和所述发送端的M个第二目标分支自适应层之间的对应关系,M为正整数。
然后,发送端接收第一信道字典集。并将所述第一信道字典集和第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
发送端还发送第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
参考信号接收端基于接收到的第一指示信息,进而确定与所述K个第一目标分支自适应层对应的K个第二目标分支自适应层,以及该K个第二目标分支自适应层的权重。
或者,发送端发送第二信道字典集。
参考信号接收端接收该第二信道字典集。然后将第一信道字典集和所述第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集 中的第一目标分支自适应层之间的对应关系。
进而参考信号接收端确定与所述K个第一目标分支自适应层对应的K个第二目标分支自适应层,以及该K个第二目标分支自适应层的权重。
相应地,在一种可能的实现方式中,基站还发送第二参考信号。
终端接收该第二参考信号。然后,终端基于该第二参考信号发送第五信息。其中,该第五信息与第四信息相同。或者,该第五信息指示基站间隔第一时间后再发送参考信号。
也就是说,终端基于第二参考信号确定出当前信道的分布与前述第一参考信号对应的信道的分布是一致的,也即该信道分布在对应第一参考信号到第二参考信号的该段时间内相对没有发生变化。终端则指示基站间隔一段时间后再发送参考信号。
本申请实施例,终端基于接收到的参考信号确定第一目标模型的N个第一目标分支自适应层中的K个第一目标分支自适应层以及该K个第一目标分支自适应层的权重。当该K个第一目标分支自适应层的权重中最大权重不小于预设值,终端向基站发送信息,以便基站基于信息将待发送的编码数据输入至K个第二目标分支自适应层中进行处理,并发送处理后的第一数据。然后,终端通过将接收到的第一数据输入至该K个第一目标分支自适应层中进行处理,得到处理后的数据。采用该手段,终端基于确定的K个第一目标分支自适应层和K个权重来表征实际的信道分布,以应对信道的动态变化,这样可以提高系统性能。且,终端将该实际的信道分布对应的K个第一目标分支自适应层和K个权重反馈给基站,以便基站按信息指示,选择对应信道分布下的最优处理方式对编码数据进行处理,可以提高通信性能。
以上对本申请实施例中的通信方法进行了介绍。下面对本申请实施例中的目标模型的训练方法进行介绍。参照图6a所示,是本申请实施例提供的一种目标模型的训练方法的流程示意图。如图6a所示的目标模型的训练方法可以包括步骤601-603。应理解,本申请为了方便描述,故通过601-603这一顺序进行描述,并不旨在限定一定通过上述顺序进行执行。本申请实施例对于上述一个或多个步骤的执行的先后顺序、执行的时间、执行的次数等不做限定。步骤601-603具体如下:
601、对初始信道特征提取网络进行训练,得到第一目标信道特征提取网络以及第一信道字典集。
在一种可能的实现方式中,步骤601可包括:
对所述初始信道特征提取网络进行多次训练,得到所述第一目标信道特征提取网络以及第一信道字典集。所述初始信道特征提取网络是根据带标签的历史信道信息或者预设信道模型进行预训练得到的。
该预设模型可以是3GPP协议中的TDL模型或者CDL信道模型。例如,从训练集(带标签的历史信道信息)中采样,输入信道特征提取网络,输出样本属于各信道类别的概率,然后通过计算与标签的交叉熵损失函数,进而训练网络参数,可得到初始信道特征提取网络。
其中,当对信道特征提取网络U t-1进行第t次训练时,将无标签的信道信息与所述带标签的历史信道信息均输入至信道特征提取网络U t-1中进行处理,得到所述无标签的信道信息与所述带标签的历史信道信息分别对应的信道分布信息。
根据所述信道分布信息得到N t-1个聚类中心。
例如,通过K均值聚类算法(k-means clustering algorithm,K-means)或者子空间K-means方法进行聚类处理得到上述聚类中心。
根据所述N t-1个聚类中心和所述信道分布信息,得到预测的信道字典集。
例如,根据各聚类中心和各信道分布信息计算任意两者之间的欧氏距离,基于该欧氏距离,利用已有信道标签推断其最可能的信道类型标签。通过合并各聚类中心中相同的信道类型标签,即可得到预测的信道字典集。
根据所述预测的信道字典集与标注的信道字典集计算得到第一损失值,并根据所述第一损失值调整所述信道特征提取网络U t-1的参数;
令t=t+1,并重复执行上述步骤,直到迭代次数达到预设次数,将所述信道特征提取网络U t-1作为所述第一目标信道特征提取网络,将所述预测的信道字典集作为所述第一信道字典集。
可选的,信道特征提取网络可包括提取器网络和分类器网络。其中提取器网络的输入为参考信息,当然还可以是信道估计结果H est。H est可以是基于参考信息得到的最小二乘法(Least Squares,LS)信道估计或最小均方误差(Minimum Mean Squared Error,MMSE)信道估计,也可以是由收端神经网络得到的信道估计结果。提取器网络的输出为信道分布信息h,用来表征当前信道分布在特征空间中的特征向量。分类器网络的输入为信道分布信息h,输出为信道分布信息对应的信道类别信息p cluster。该信道类别信息是基于信道字典集表示的。
其中,该损失函数L1可表示为:
11=λL classify+(1-λ)L cluster
其中,λ∈[0,1],λ为超参数,用来调节L classify和L cluster两个损失函数的影响占比;L classify为交叉熵损失函数;L cluster为聚类损失函数;
Figure PCTCN2022132654-appb-000014
其中,KL(P||Q)表示辅助目标分布P与样本的软聚类分布Q之间的(Kullback-Leibler,KL)散度;q ij表示样本i属于聚类j的概率,
Figure PCTCN2022132654-appb-000015
h i为样本i经过提取器网络输出的信道分布信息,μ j为第j个聚类的质心,j′为1,2…N,N为聚类中心的个数;p ij表示辅助目标分布;
Figure PCTCN2022132654-appb-000016
为1,2…M,M为样本的个数。
基于上述损失函数进行训练,即可得到训练好的第一目标信道特征提取网络。
602、根据目标编码网络和目标解码网络,对N个初始分支自适应层进行训练,得到N个第一目标分支自适应层,所述N为所述第一信道字典集中的第一目标分支自适应层的个数;
在一种可能的实现方式中,步骤602可包括:
对于第m个初始分支自适应层进行第t次训练时,将第一样本数据输入至所述目标编码网络和预设信道中进行处理,得到第二样本数据。
将所述第二样本数据输入至所述第m个初始分支自适应层中进行处理,得到所述第m个初始分支自适应层处理得到的数据,m为正整数,m不大于N。
将所述第m个初始分支自适应层处理得到的数据输入至所述目标解码网络中进行处理,得到第三样本数据。
根据所述第一样本数据和所述第三样本数据计算得到第二损失值,并根据所述第二损失值调整所述第m个初始分支自适应层参数。
令t=t+1,并重复上述步骤,直到达到停止条件,将所述第m个初始分支自适应层作为第m个第一目标分支自适应层。
令m=m+1,并重复执行上述步骤,直到m=N,得到所述N个第一目标分支自适应层。
在一种可能的实现方式中,所述方法还包括:
发送第二指示信息,所述第二指示信息指示所述第m个初始分支自适应层进行训练。
例如,发端向收端发送该指示信息,以指示其训练第m个分支。然后,发端向收端发送训练数据。
或者,收端向发端发送该指示信息,以告知收端,发端要训练第m个分支。这样,以便发端向收端发送训练数据。
可选的,如图6b所示,对于多分支自适应层的训练需要大量数据集用于刻画信道分布的特征。在1个子帧上所有时隙上发送导频(参考信号),然后接收训练的梯度信息。对于各信道类型对应的分支,可在帧头额外指示信道字典集标号。这样做,以指示该帧所训练的自适应层。
603、根据所述目标编码网络、所述目标解码网络、所述第一目标信道特征提取网络和所述N个第一目标分支自适应层,对初始稀疏门控模块进行训练,得到目标稀疏门控模块。
也就是说,基于训练好的目标编码网络、目标解码网络、第一目标信道特征提取网络和N个第一目标分支自适应层,来对初始稀疏门控模块进行训练。
基于上述训练,即可得到如图3所示的目标模型。其中,该目标模型包括所述目标编码网络、所述目标解码网络和第一目标模型,所述第一目标模型包括第一目标信道特征提取网络、N个第一目标分支自适应层和目标稀疏门控模块。
在一种可能的实现方式中,步骤603可包括:
对所述初始稀疏门控模块进行多次训练,得到所述目标稀疏门控模块,
其中,当对稀疏门控模块X t-1进行第t次训练时,将第四样本数据输入至所述目标编码网络和预设信道中进行处理,得到第五样本数据;
将所述第五样本数据输入至所述第一目标信道特征提取网络中进行处理,得到信道分布信息训练数据和信道类别信息训练数据;
将所述信道分布信息训练数据和所述信道类别信息训练数据均输入至所述稀疏门控模块X t-1中进行处理,得到与所述信道分布信息训练数据对应的R个第一目标分支自适应层样本和所述R个第一目标分支自适应层的权重样本,R为正整数,R不大于N;
根据所述R个第一目标分支自适应层样本和所述R个第一目标分支自适应层的权重样本对所述第五样本数据进行处理,得到处理后的训练数据;
将所述处理后的训练数据输入至所述目标解码网络中进行处理,得到第六样本数据;
根据所述第四样本数据、所述第六样本数据和所述R个第一目标分支自适应层的权重样本计算得到第三损失值,并根据所述第三损失值调整所述稀疏门控模块X t-1的参数;
令t=t+1,并重复执行上述步骤,直到达到停止条件,将所述稀疏门控模块X t-1作为所述目标稀疏门控模块。
可选的,该训练过程增加熵正则约束,使信道分布的判决更加明确。其中该损失函数L2可表示为:
L2=L llr+βEntropy(α);
其中,L llr为目标编码网络和目标解码网络联合训练时的损失函数。α为分支自适应层的权重,Entropy(α)表示计算分支自适应层权重的熵,β为不小于0的超参数。
基于上述损失函数进行训练,即可得到训练好的目标稀疏门控模块。
在一种可能的实现方式中,所述目标模型还包括第二目标模型,所述第二目标模型包括第二目标信道特征提取网络和M个第二目标分支自适应层。
其中,针对该第二目标信道特征提取网络的训练过程可参阅步骤601的记载,在此不再赘述。
针对M个第二目标分支自适应层的训练,其可以是对M个初始分支自适应层和前述N个初始分支自适应层联合进行多次训练得到,其中:
对于第m’个初始分支自适应层进行第t次训练时,将第七样本数据输入至所述目标编码网络中进行处理,得到第八样本数据。
将所述第八样本数据输入至所述第m’个初始分支自适应层中进行处理,得到所述第m’个初始分支自适应层处理得到的数据,m’为正整数,m’不大于M。
将所述第m’个初始分支自适应层处理得到的数据输入至预设信道、第m个初始分支自适应层和所述目标解码网络中进行处理,得到第九样本数据。其中,该第二目标模型中的第m’个初始分支自适应层与第一目标模型中的第m个初始分支自适应层是对应关系。例如,通过基于前述信道特征提取网络训练可分别得到第一信道字典集和第二信道字典集。第一信道字典集与第一目标模型对应,第二信道字典集与第二目标模型对应。这样,在训练时,两个模型可以进行自适应层的对齐,进而一起训练。
根据所述第七样本数据和所述第九样本数据计算得到第四损失值,并根据所述第四损失值调整所述第m’个初始分支自适应层参数和第m个初始分支自适应层。
令t=t+1,并重复上述步骤,直到达到停止条件,将所述第m个初始分支自适应层作为第m个第一目标分支自适应层,将所述第m’个初始分支自适应层作为第m’个第二目标分支自适应层。
令m’=m’+1,并重复执行上述步骤,直到m’=M,得到所述N个第一目标分支自适应层和所述M个第二目标分支自适应层。
可选的,M=N。当然M、N还可以是其他关系,本方案对此不作限制。
本申请实施例,基于对初始信道特征提取网络进行训练,得到该第一目标信道特征提取网络。然后,根据对N个初始分支自适应层进行训练,得到N个第一目标分支自适应层。最后,根据训练好的目标编码网络、目标解码网络、第一目标信道特征提取网络和N个第一目标分支自适应层对初始稀疏门控模块进行训练,得到目标稀疏门控模块。采用该手段得到的目标模型,可以解决由于环境中的信道动态变化而导致的神经网络训练与推演的问题。
另一方面,本方案通过训练分支自适应层权重的方式来调整分支自适应层的输出,而非对目标编码网络、目标解码网络参数进行训练,这样能够减少网络训练的开销。
本申请实施例的上述目标模型以包括多个网络模块为例进行介绍。需要说明的是,该多个网络模块可以是独立的网络模块,其还可以是集成一体的,或者部分网络模块集成一体等,本方案对此不作严格限制。
需要说明的是,本申请实施例以第一目标模型(包括第一目标信道特征提取网络、N个第一目标分支自适应层、以及目标稀疏门控模块)部署在收端、第二目标模型(包括第二目标信道特征提取网络和M个第二目标分支自适应层)部署在发端为例进行介绍。其还可以是第一目标模型部署在发端、第二目标模型部署在收端等。本申请实施例还可以仅部署第一目标模型等。本方案对此不作严格限制。
需要说明的是,在本申请的各个实施例中,如果没有特殊说明以及逻辑冲突,各个实施例之间的术语和/或描述具有一致性、且可以相互引用,不同的实施例中的技术特征根据其内在的逻辑关系可以组合形成新的实施例。
上述详细阐述了本申请实施例的方法,下面提供了本申请实施例的装置。可以理解的,本申请各个装置实施例中,对多个单元或者模块的划分仅是一种根据功能进行的逻辑划分,不作为对装置具体的结构的限定。在具体实现中,其中部分功能模块可能被细分为更多细小的功能模块,部分功能模块也可能组合成一个功能模块,但无论这些功能模块是进行了细分还是组合,装置所执行的大致流程是相同的。例如,一些装置中包含接收单元和发送单元。一些设计中,发送单元和接收单元也可以集成为通信单元,该通信单元可以实现接收单元和发送单元所实现的功能。通常,每个单元都对应有各自的程序代码(或者说程序指令),这些单元各自对应的程序代码在处理器上运行时,使得该单元受处理单元的控制而执行相应的流程从而实现相应功能。
本申请实施例还提供用于实现以上任一种方法的装置,例如,提供一种通信装置包括用以实现以上任一种方法中终端所执行的各步骤的模块(或手段)。再如,还提供另一种通信装置,包括用以实现以上任一种方法中基站所执行的各步骤的模块(或手段)。
例如,参照图7a所示,是本申请实施例提供的一种通信装置的结构示意图。该通信装置用于实现前述的通信方法,例如图2a、图4所示的通信方法。
如图7a所示,该装置可包括通信模块701和处理模块702,具体如下:
通信模块701,用于接收第一参考信号;
处理模块702,用于将所述第一参考信号输入至第一目标模型中进行处理,得到所述第一目标模型中的K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重,所述第一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数;
所述通信模块701,还用于接收第一数据;
所述处理模块702,还用于将所述第一数据输入至所述K个第一目标分支自适应层中进行处理,得到处理后的数据,其中,所述K个第一目标分支自适应层的权重中最大权重不小于预设值。
在一种可能的实现方式中,所述第一目标模型还包括第一目标信道特征提取网络和目标稀疏门控模块,所述第一目标信道特征提取网络用于对所述第一参考信号进行处理,得到信道分布信息和所述信道分布信息对应的信道类别信息;所述目标稀疏门控模块用于根据所述信道分布信息和所述信道类别信息计算得到所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重。
在一种可能的实现方式中,所述K个第一目标分支自适应层用于分别对所述第一数据进行处理,得到所述K个第一目标分支自适应层分别处理后的数据;所述K个第一目标分支自适应层的权重用于对所述K个第一目标分支自适应层分别处理后的数据进行加权求和处理,得到所述处理后的数据。
在一种可能的实现方式中,所述处理模块702,还用于:当所述K个第一目标分支自适应层的权重中最大权重小于所述预设值时,对所述第一数据进行解码处理,得到处理后的数据。
在一种可能的实现方式中,所述通信模块701,还用于:发送第一信息,所述第一信息 指示所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重,所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重用于发送端对待发送的编码数据进行处理得到所述第一数据。
在一种可能的实现方式中,所述通信模块701,还用于:当所述K个第一目标分支自适应层的权重中最大权重小于所述预设值,发送第二信息,所述第二信息指示发送端直接发送所述第一数据。
在一种可能的实现方式中,所述通信模块701,还用于:发送第一信道字典集,所述第一信道字典集用于发送端确定所述N个第一目标分支自适应层和所述发送端的M个第二目标分支自适应层之间的对应关系,M为正整数。
在另一种可能的实现方式中,所述通信模块701,还用于:
接收第二信道字典集;
所述处理模块702,还用于将第一信道字典集和所述第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系;
所述通信模块701,还用于发送第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,所述第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,所述通信模块701,还用于:
发送第四信息,所述第四信息指示K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重,所述K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重用于发送端对待发送的编码数据进行处理得到所述第一数据,所述K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重是根据所述K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重确定的。
在一种可能的实现方式中,所述通信模块701,还用于:
发送第一信道字典集;
接收第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,所述通信模块701,还用于:
接收第二信道字典集;
所述处理模块702,还用于将第一信道字典集和所述第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,所述通信模块701,还用于:接收第二参考信号;
发送第三信息,所述第三信息与所述第一信息相同,或者,所述第三信息指示发送端间隔第一时间后再发送参考信号,所述第三信息是根据所述第二参考信号得到的;
接收所述发送端在所述第一时间内发送的数据。
针对上述各模块的介绍,可参阅前述实施例的记载,在此不再赘述。
再如,参照图7b所示,是本申请实施例提供的另一种通信装置的结构示意图。该通信装 置用于实现前述的通信方法,例如图2a、图4所示的通信方法。
该通信装置包括通信模块703和处理模块704,具体如下:
通信模块703,用于发送第一参考信号;
处理模块704,用于当接收到第一信息时,所述第一信息指示所述第一参考信号接收端的第一目标模型中的K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重,根据所述第一信息,确定与所述K个第一目标分支自适应层对应的K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重,所述第一信息是根据所述第一参考信号得到的,K为正整数;
所述处理模块704,还用于将待发送的编码数据输入至第二目标模型的所述K个第二目标分支自适应层中进行处理,得到第一数据;
所述通信模块703,还用于发送所述第一数据。
在一种可能的实现方式中,当M小于K时,所述处理模块704,还用于更新第三目标模型,得到所述第二目标模型,使得所述第二目标模型包括所述K个第二目标分支自适应层,其中所述第三目标模型包括M个第二目标分支自适应层,M为正整数。
在一种可能的实现方式中,所述K个第二目标分支自适应层用于分别对所述待发送的编码数据进行处理,得到所述K个第二目标分支自适应层分别处理后的数据;所述K个第二目标分支自适应层的权重用于对所述K个第二目标分支自适应层分别处理后的数据进行加权求和处理,得到所述第一数据。
在一种可能的实现方式中,所述通信模块703,还用于:当接收到第二信息时,所述第二信息指示直接发送所述待发送的编码数据,则发送所述待发送的编码数据,所述第二信息是根据所述第一参考信号得到的。
在一种可能的实现方式中,所述通信模块703,还用于:接收第一信道字典集;
所述处理模块704,还用于将所述第一信道字典集和第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
在另一种可能的实现方式中,所述通信模块703,还用于:
发送第二信道字典集;
接收第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,所述第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,所述通信模块703,还用于:发送第二参考信号;
接收第三信息,所述第三信息与所述第一信息相同;或者,所述第三信息用于指示间隔第一时间后再发送参考信号,所述第三信息是根据所述第二参考信号得到的;
在所述第一时间内发送数据。
针对上述各模块的介绍,可参阅前述实施例的记载,在此不再赘述。
本申请提供还一种通信装置,包括:
通信模块,用于接收第一参考信号;
处理模块,用于将所述第一参考信号输入至第一目标模型中进行处理,得到所述第一目标模型中的K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重,所述第 一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数;所述通信模块,还用于发送第一信息,所述第一信息指示所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重,所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重用于发送端对待发送的编码数据进行处理。
在一种可能的实现方式中,所述通信模块,还用于:接收第一数据,所述第一数据是所述发送端对待发送的编码数据进行处理得到的;所述处理模块,还用于将所述第一数据输入至所述K个第一目标分支自适应层中进行处理,得到处理后的数据。
在一种可能的实现方式中,所述通信模块,还用于:发送第一信道字典集,所述第一信道字典集用于所述发送端确定所述N个第一目标分支自适应层和所述发送端的M个第二目标分支自适应层之间的对应关系,M为正整数。
在另一种可能的实现方式中,所述通信模块,还用于:接收第二信道字典集;
所述处理模块,还用于将第一信道字典集和所述第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系;
所述通信模块,还用于发送第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,所述第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,所述通信模块,还用于:接收第二参考信号;发送第三信息,所述第三信息与所述第一信息相同,或者,所述第三信息指示所述发送端间隔第一时间后再发送参考信号,所述第三信息是根据所述第二参考信号得到的;接收所述发送端在所述第一时间内发送的数据。
针对上述各模块的介绍,可参阅前述实施例的记载,在此不再赘述。
另一方面,本申请还提供一种通信装置,包括:
通信模块,用于发送第一参考信号;
所述通信模块,还用于接收第一信息,所述第一信息指示所述第一参考信号接收端的第一目标模型中的K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重,所述第一信息是根据所述第一参考信号得到的,K为正整数。
在一种可能的实现方式中,所述装置还包括处理模块,用于:根据所述第一信息,确定与所述K个第一目标分支自适应层对应的K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重;
所述处理模块,还用于将待发送的编码数据输入至第二目标模型的所述K个第二目标分支自适应层中进行处理,得到第一数据;
所述通信模块,还用于发送所述第一数据。
在一种可能的实现方式中,当M小于K时,所述处理模块,还用于更新第三目标模型,得到所述第二目标模型,使得所述第二目标模型包括所述K个第二目标分支自适应层,其中所述第三目标模型包括M个第二目标分支自适应层,M为正整数。
在一种可能的实现方式中,所述通信模块,还用于:接收第一信道字典集;
所述处理模块,还用于将所述第一信道字典集和第二信道字典集进行字典对齐,得到所 述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
在另一种可能的实现方式中,所述通信模块,还用于:发送第二信道字典集;接收第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,所述第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,所述通信模块,还用于:发送第二参考信号;接收第三信息,所述第三信息与所述第一信息相同,或者,所述第三信息指示间隔第一时间后再发送参考信号,所述第三信息是根据所述第二参考信号得到的;在所述第一时间内发送数据。
针对上述各模块的介绍,可参阅前述实施例的记载,在此不再赘述。
本申请实施例还提供一种通信装置,包括:
通信模块,用于接收第一参考信号;
处理模块,用于将所述第一参考信号输入至第一目标模型中进行处理,得到所述第一目标模型中的K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重,所述第一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数;
所述通信模块,还用于发送第四信息,所述第四信息指示K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重,所述K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重用于发送端对待发送的编码数据进行处理,所述K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重是根据所述K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重确定的。
在一种可能的实现方式中,所述通信模块,还用于:
接收第一数据,所述第一数据是所述发送端对待发送的编码数据进行处理得到的;
所述处理模块,还用于将所述第一数据输入至所述K个第一目标分支自适应层中进行处理,得到处理后的数据。
在一种可能的实现方式中,所述通信模块,还用于:
发送第一信道字典集,所述第一信道字典集用于所述发送端确定所述N个第一目标分支自适应层和所述发送端的M个第二目标分支自适应层之间的对应关系,M为正整数;
接收第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,所述通信模块,还用于:
接收第二信道字典集;
所述处理模块,还用于将第一信道字典集和所述第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,所述第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,所述通信模块,还用于:
接收第二参考信号;
发送第三信息,所述第三信息与所述第一信息相同,或者,所述第三信息指示所述发送 端间隔第一时间后再发送参考信号,所述第三信息是根据所述第二参考信号得到的;
接收所述发送端在所述第一时间内发送的数据。
本申请实施例还提供一种通信装置,包括:
通信模块,用于发送第一参考信号;
所述通信模块,还用于接收第四信息,所述第四信息指示K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重,所述第四信息是根据所述第一参考信号得到的,K为正整数。
在一种可能的实现方式中,所述装置还包括处理模块,用于:
根据所述第四信息,将待发送的编码数据输入至第二目标模型的所述K个第二目标分支自适应层中进行处理,得到第一数据;
所述通信模块,还用于发送所述第一数据。
在一种可能的实现方式中,所述通信模块,还用于:
接收第一信道字典集;
所述处理模块,还用于将所述第一信道字典集和第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系;
所述通信模块,还用于发送第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,所述通信模块,还用于:
发送第二信道字典集,所述第二信道字典集用于参考信号接收端确定所述参考信号接收端的N个第一目标分支自适应层和发送端的M个第二目标分支自适应层之间的对应关系,M、N均为正整数。
在一种可能的实现方式中,所述第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,所述通信模块,还用于:
发送第二参考信号;
接收第三信息,所述第三信息与所述第一信息相同,或者,所述第三信息指示间隔第一时间后再发送参考信号,所述第三信息是根据所述第二参考信号得到的;
在所述第一时间内发送数据。
本申请实施例还提供一种通信装置,包括:
通信模块,用于发送第一参考信号;
处理模块,用于当接收到第四信息时,所述第四信息指示K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重,根据所述第四信息,将待发送的编码数据输入至第二目标模型的所述K个第二目标分支自适应层中进行处理,得到第一数据,所述第四信息是根据所述第一参考信号得到的,K为正整数;
所述通信模块,还用于发送所述第一数据。
在一种可能的实现方式中,所述K个第二目标分支自适应层用于分别对所述待发送的编码数据进行处理,得到所述K个第二目标分支自适应层分别处理后的数据;所述K个第二目 标分支自适应层的权重用于对所述K个第二目标分支自适应层分别处理后的数据进行加权求和处理,得到所述第一数据。
在一种可能的实现方式中,所述通信模块,还用于:
当接收到第二信息时,所述第二信息指示直接发送所述待发送的编码数据,则发送所述待发送的编码数据,所述第二信息是根据所述第一参考信号得到的。
在一种可能的实现方式中,所述通信模块,还用于:
接收第一信道字典集;
所述处理模块,还用于将所述第一信道字典集和第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系;
所述通信模块,还用于发送第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与第一信道字典集中的第一目标分支自适应层之间的对应关系。
在一种可能的实现方式中,所述通信模块,还用于:
发送第二信道字典集,所述第二信道字典集用于参考信号接收端确定所述参考信号接收端的N个第一目标分支自适应层和发送端的M个第二目标分支自适应层之间的对应关系,M、N均为正整数。
在一种可能的实现方式中,所述第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,所述处理模块,还用于:
发送第二参考信号;
接收第三信息,所述第三信息与所述第一信息相同;或者,所述第三信息用于指示间隔第一时间后再发送参考信号,所述第三信息是根据所述第二参考信号得到的;
在所述第一时间内发送数据。
本申请实施例还提供一种目标模型训练装置。该目标模型包括目标编码网络、目标解码网络和第一目标模型,该第一目标模型包括第一目标信道特征提取网络、N个第一目标分支自适应层和目标稀疏门控模块。该装置包括:
第一训练模块,用于对初始信道特征提取网络进行训练,得到该第一目标信道特征提取网络以及第一信道字典集;
第二训练模块,用于根据该目标编码网络和该目标解码网络,对N个初始分支自适应层进行训练,得到该N个第一目标分支自适应层,该N为该第一信道字典集中的第一目标分支自适应层的个数;
第三训练模块,用于根据该目标编码网络、该目标解码网络、该第一目标信道特征提取网络和该N个第一目标分支自适应层,对初始稀疏门控模块进行训练,得到该目标稀疏门控模块。
可选的,该第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
在一种可能的实现方式中,该第一训练模块,用于:对该初始信道特征提取网络进行多次训练,得到该第一目标信道特征提取网络以及第一信道字典集。其中,该初始信道特征提取网络是根据带标签的历史信道信息或者预设信道模型进行预训练得到的。
其中,当对信道特征提取网络U t-1进行第t次训练时,将无标签的信道信息与该带标签的 历史信道信息均输入至信道特征提取网络U t-1中进行处理,得到该无标签的信道信息与该带标签的历史信道信息分别对应的信道分布信息。根据该信道分布信息得到N t-1个聚类中心。根据该N t-1个聚类中心和该信道分布信息,得到预测的信道字典集。然后,根据该预测的信道字典集与标注的信道字典集计算得到第一损失值,并根据该第一损失值调整该信道特征提取网络U t-1的参数。令t=t+1,并重复执行上述步骤,直到迭代次数达到预设次数,将该信道特征提取网络U t-1作为该第一目标信道特征提取网络,将该预测的信道字典集作为该第一信道字典集。
在一种可能的实现方式中,第二训练模块,用于:对于第m个初始分支自适应层进行第t次训练时,将第一样本数据输入至该目标编码网络和预设信道中进行处理,得到第二样本数据。将该第二样本数据输入至该第m个初始分支自适应层中进行处理,得到该第m个初始分支自适应层处理得到的数据,m为正整数,m不大于N。将该第m个初始分支自适应层处理得到的数据输入至该目标解码网络中进行处理,得到第三样本数据。然后,根据该第一样本数据和该第三样本数据计算得到第二损失值,并根据该第二损失值调整该第m个初始分支自适应层参数。令t=t+1,并重复上述步骤,直到达到停止条件,将该第m个初始分支自适应层作为第m个第一目标分支自适应层。
令m=m+1,并重复执行上述步骤,直到m=N,得到该N个第一目标分支自适应层。
在一种可能的实现方式中,该装置还包括通信模块,用于:发送第二指示信息,该第二指示信息指示该第m个初始分支自适应层进行训练。
在一种可能的实现方式中,第三训练模块,用于:对该初始稀疏门控模块进行多次训练,得到该目标稀疏门控模块。
其中,当对稀疏门控模块X t-1进行第t次训练时,将第四样本数据输入至该目标编码网络和预设信道中进行处理,得到第五样本数据。将该第五样本数据输入至该第一目标信道特征提取网络中进行处理,得到信道分布信息训练数据和信道类别信息训练数据。将该信道分布信息训练数据和该信道类别信息训练数据均输入至该稀疏门控模块X t-1中进行处理,得到与该信道分布信息训练数据对应的R个第一目标分支自适应层样本和该R个第一目标分支自适应层的权重样本,R为正整数,R不大于N。根据该R个第一目标分支自适应层样本和该R个第一目标分支自适应层的权重样本对该第五样本数据进行处理,得到处理后的训练数据。将该处理后的训练数据输入至该目标解码网络中进行处理,得到第六样本数据。然后,根据该第四样本数据、该第六样本数据和该R个第一目标分支自适应层的权重样本计算得到第三损失值,并根据该第三损失值调整该稀疏门控模块X t-1的参数。令t=t+1,并重复执行上述步骤,直到达到停止条件,将该稀疏门控模块X t-1作为该目标稀疏门控模块。
针对上述各模块的介绍,可参阅前述实施例的记载,在此不再赘述。
应理解以上各个装置中各模块的划分仅是一种逻辑功能的划分,实际实现时可以全部或部分集成到一个物理实体上,也可以物理上分开。此外,通信装置或目标模型的训练装置中的模块可以以处理器调用软件的形式实现;例如通信装置包括处理器,处理器与存储器连接,存储器中存储有指令,处理器调用存储器中存储的指令,以实现以上任一种方法或实现该装置各模块的功能,其中处理器例如为通用处理器,比如中央处理单元(central processing unit,CPU)或微处理器,存储器为装置内的存储器或装置外的存储器。或者,装置中的模块可以以硬件电路的形式实现,可以通过对硬件电路的设计实现部分或全部单元的功能,该硬件电路可以理解为一个或多个处理器;例如,在一种实现中,该硬件电路为专用集成电路 (application-specific integrated circuit,ASIC),通过对电路内元件逻辑关系的设计,实现以上部分或全部单元的功能;再如,在另一种实现中,该硬件电路为可以通过可编程逻辑器件(programmable logic device,PLD)实现,以现场可编程门阵列(field programmable gate array,FPGA)为例,其可以包括大量逻辑门电路,通过配置文件来配置逻辑门电路之间的连接关系,从而实现以上部分或全部单元的功能。以上装置的所有模块可以全部通过处理器调用软件的形式实现,或全部通过硬件电路的形式实现,或部分通过处理器调用软件的形式实现,剩余部分通过硬件电路的形式实现。
参照图8所示,是本申请实施例提供的又一种通信装置的硬件结构示意图。如图8所示的通信装置800(该装置800具体可以是一种计算机设备)包括存储器801、处理器802、通信接口803以及总线804。其中,存储器801、处理器802、通信接口803通过总线804实现彼此之间的通信连接。
存储器801可以是只读存储器(read only memory,ROM),静态存储设备,动态存储设备或者随机存取存储器(random access memory,RAM)。
存储器801可以存储程序,当存储器801中存储的程序被处理器802执行时,处理器802和通信接口803用于执行本申请实施例的通信方法或目标模型的训练方法的各个步骤。
处理器802是一种具有信号的处理能力的电路,在一种实现中,处理器802可以是具有指令读取与运行能力的电路,例如中央处理单元CPU、微处理器、图形处理器(graphics processing unit,GPU)(可以理解为一种微处理器)、或数字信号处理器(digital singnal processor,DSP)等;在另一种实现中,处理器802可以通过硬件电路的逻辑关系实现一定功能,该硬件电路的逻辑关系是固定的或可以重构的,例如处理器802为ASIC或可编程逻辑器件PLD实现的硬件电路,比如FPGA。在可重构的硬件电路中,处理器加载配置文档,实现硬件电路配置的过程,可以理解为处理器加载指令,以实现以上部分或全部模块的功能的过程。此外,还可以是针对人工智能设计的硬件电路,其可以理解为一种ASIC,例如神经网络处理单元(neural network processing unit,NPU)、张量处理单元(tensor processing unit,TPU)、深度学习处理单元(deep learning processing unit,DPU)等。处理器802用于执行相关程序,以实现本申请实施例的通信装置或目标模型的训练装置中的单元所需执行的功能,或者执行本申请方法实施例的通信方法或目标模型的训练方法。
可见,以上装置中的各模块可以是被配置成实施以上方法的一个或多个处理器(或处理电路),例如:CPU、GPU、NPU、TPU、DPU、微处理器、DSP、ASIC、FPGA,或这些处理器形式中至少两种的组合。
此外,以上装置中的各模块可以全部或部分可以集成在一起,或者可以独立实现。在一种实现中,这些模块集成在一起,以片上系统(system-on-a-chip,SOC)的形式实现。该SOC中可以包括至少一个处理器,用于实现以上任一种方法或实现该装置各模块的功能,该至少一个处理器的种类可以不同,例如包括CPU和FPGA,CPU和人工智能处理器,CPU和GPU等。
通信接口803使用例如但不限于收发器一类的收发装置,来实现装置800与其他设备或通信网络之间的通信。例如,可以通过通信接口803获取数据。
总线804可包括在装置800各个部件(例如,存储器801、处理器802、通信接口803)之间传送信息的通路。
应注意,尽管图8所示的装置800仅仅示出了存储器、处理器、通信接口,但是在具体实现过程中,本领域的技术人员应当理解,装置800还包括实现正常运行所必须的其他器件。 同时,根据具体需要,本领域的技术人员应当理解,装置800还可包括实现其他附加功能的硬件器件。此外,本领域的技术人员应当理解,装置800也可仅仅包括实现本申请实施例所必须的器件,而不必包括图8中所示的全部器件。
本申请实施例还提供一种通信系统,例如包括上述如图7a所示的装置,以及如图7b所示的装置。
本申请实施例还提供了一种计算机可读存储介质,该计算机可读存储介质中存储有指令,当其在计算机或处理器上运行时,使得计算机或处理器执行上述任一个方法中的一个或多个步骤。
本申请实施例还提供了一种包含指令的计算机程序产品。当该计算机程序产品在计算机或处理器上运行时,使得计算机或处理器执行上述任一个方法中的一个或多个步骤。
应理解,在本申请的描述中,除非另有说明,“/”表示前后关联的对象是一种“或”的关系,例如,A/B可以表示A或B;其中A,B可以是单数或者复数。并且,在本申请的描述中,除非另有说明,“多个”是指两个或多于两个。“以下至少一项(个)”或其类似表达,是指的这些项中的任意组合,包括单项(个)或复数项(个)的任意组合。例如,a,b,或c中的至少一项(个),可以表示:a,b,c,a-b,a-c,b-c,或a-b-c,其中a,b,c可以是单个,也可以是多个。另外,为了便于清楚描述本申请实施例的技术方案,在本申请的实施例中,采用了“第一”、“第二”等字样对功能和作用基本相同的相同项或相似项进行区分。本领域技术人员可以理解“第一”、“第二”等字样并不对数量和执行次序进行限定,并且“第一”、“第二”等字样也并不限定一定不同。同时,在本申请实施例中,“示例性的”或者“例如”等词用于表示作例子、例证或说明。本申请实施例中被描述为“示例性的”或者“例如”的任何实施例或设计方案不应被解释为比其它实施例或设计方案更优选或更具优势。确切而言,使用“示例性的”或者“例如”等词旨在以具体方式呈现相关概念,便于理解。
在本申请所提供的几个实施例中,应该理解到,所揭露的系统、装置和方法,可以通过其它的方式实现。例如,该单元的划分,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式,例如,多个单元或组件可以结合或者可以集成到另一个系统,或一些特征可以忽略,或不执行。所显示或讨论的相互之间的耦合、或直接耦合、或通信连接可以是通过一些接口,装置或单元的间接耦合或通信连接,可以是电性,机械或其它的形式。
作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
在上述实施例中,可以全部或部分地通过软件、硬件、固件或者其任意组合来实现。当使用软件实现时,可以全部或部分地以计算机程序产品的形式实现。该计算机程序产品包括一个或多个计算机指令。在计算机上加载和执行该计算机程序指令时,全部或部分地产生按照本申请实施例的流程或功能。该计算机可以是通用计算机、专用计算机、计算机网络、或者其他可编程装置。该计算机指令可以存储在计算机可读存储介质中,或者通过该计算机可读存储介质进行传输。该计算机指令可以从一个网站站点、计算机、服务器或数据中心通过有线(例如同轴电缆、光纤、数字用户线(digital subscriber line,DSL))或无线(例如红外、无线、微波等)方式向另一个网站站点、计算机、服务器或数据中心进行传输。该计算机可读存储介质可以是计算机能够存取的任何可用介质或者是包含一个或多个可用介质集成的服务器、数据中心等数据存储设备。该可用介质可以是只读存储器(read-only memory,ROM),或随机存取存储器(random access memory,RAM),或磁性介质,例如,软盘、硬盘、磁带、 磁碟、或光介质,例如,数字通用光盘(digital versatile disc,DVD)、或者半导体介质,例如,固态硬盘(solid state disk,SSD)等。
以上所述,仅为本申请实施例的具体实施方式,但本申请实施例的保护范围并不局限于此,任何在本申请实施例揭露的技术范围内的变化或替换,都应涵盖在本申请实施例的保护范围之内。因此,本申请实施例的保护范围应以所述权利要求的保护范围为准。

Claims (112)

  1. 一种通信方法,其特征在于,包括:
    接收第一参考信号;
    将所述第一参考信号输入至第一目标模型中进行处理,得到所述第一目标模型中的K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重,所述第一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数;
    发送第一信息,所述第一信息指示所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重,所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重用于发送端对待发送的编码数据进行处理。
  2. 根据权利要求1所述的方法,其特征在于,所述方法还包括:
    接收第一数据,所述第一数据是所述发送端对待发送的编码数据进行处理得到的;
    将所述第一数据输入至所述K个第一目标分支自适应层中进行处理,得到处理后的数据。
  3. 根据权利要求1或2所述的方法,其特征在于,所述方法还包括:
    发送第一信道字典集,所述第一信道字典集用于所述发送端确定所述N个第一目标分支自适应层和所述发送端的M个第二目标分支自适应层之间的对应关系,M为正整数。
  4. 根据权利要求1或2所述的方法,其特征在于,所述方法还包括:
    接收第二信道字典集;
    将第一信道字典集和所述第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系;
    发送第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  5. 根据权利要求3或4所述的方法,其特征在于,所述第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
  6. 根据权利要求1至5任一项所述的方法,其特征在于,所述方法还包括:
    接收第二参考信号;
    发送第三信息,所述第三信息与所述第一信息相同,或者,所述第三信息指示所述发送端间隔第一时间后再发送参考信号,所述第三信息是根据所述第二参考信号得到的;
    接收所述发送端在所述第一时间内发送的数据。
  7. 一种通信方法,其特征在于,包括:
    接收第一参考信号;
    将所述第一参考信号输入至第一目标模型中进行处理,得到所述第一目标模型中的K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重,所述第一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数;
    发送第四信息,所述第四信息指示K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重,所述K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重用于发送端对待发送的编码数据进行处理,所述K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重是根据所述K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重确定的。
  8. 根据权利要求7所述的方法,其特征在于,所述方法还包括:
    接收第一数据,所述第一数据是所述发送端对待发送的编码数据进行处理得到的;
    将所述第一数据输入至所述K个第一目标分支自适应层中进行处理,得到处理后的数据。
  9. 根据权利要求7或8所述的方法,其特征在于,所述方法还包括:
    发送第一信道字典集,所述第一信道字典集用于所述发送端确定所述N个第一目标分支自适应层和所述发送端的M个第二目标分支自适应层之间的对应关系,M为正整数;
    接收第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  10. 根据权利要求7或8所述的方法,其特征在于,所述方法还包括:
    接收第二信道字典集;
    将第一信道字典集和所述第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  11. 根据权利要求9或10所述的方法,其特征在于,所述第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
  12. 根据权利要求7至11任一项所述的方法,其特征在于,所述方法还包括:
    接收第二参考信号;
    发送第五信息,所述第五信息与所述第四信息相同,或者,所述第五信息指示所述发送端间隔第一时间后再发送参考信号,所述第五信息是根据所述第二参考信号得到的;
    接收所述发送端在所述第一时间内发送的数据。
  13. 一种通信方法,其特征在于,包括:
    发送第一参考信号;
    接收第一信息,所述第一信息指示所述第一参考信号接收端的第一目标模型中的K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重,所述第一信息是根据所述第一参考信号得到的,K为正整数。
  14. 根据权利要求13所述的方法,其特征在于,所述方法还包括:
    根据所述第一信息,确定与所述K个第一目标分支自适应层对应的K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重;
    将待发送的编码数据输入至第二目标模型的所述K个第二目标分支自适应层中进行处理,得到第一数据;
    发送所述第一数据。
  15. 根据权利要求14所述的方法,其特征在于,
    当M小于K时,更新第三目标模型,得到所述第二目标模型,使得所述第二目标模型包括所述K个第二目标分支自适应层,其中所述第三目标模型包括M个第二目标分支自适应层,M为正整数。
  16. 根据权利要求13至15任一项所述的方法,其特征在于,所述方法还包括:
    接收第一信道字典集;
    将所述第一信道字典集和第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  17. 根据权利要求13至15任一项所述的方法,其特征在于,所述方法还包括:
    发送第二信道字典集;
    接收第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支 自适应层与第一信道字典集中的第一目标分支自适应层之间的对应关系。
  18. 根据权利要求16或17所述的方法,其特征在于,所述第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
  19. 根据权利要求13至18任一项所述的方法,其特征在于,所述方法还包括:
    发送第二参考信号;
    接收第三信息,所述第三信息与所述第一信息相同,或者,所述第三信息指示间隔第一时间后再发送参考信号,所述第三信息是根据所述第二参考信号得到的;
    在所述第一时间内发送数据。
  20. 一种通信方法,其特征在于,包括:
    发送第一参考信号;
    接收第四信息,所述第四信息指示K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重,所述第四信息是根据所述第一参考信号得到的,K为正整数。
  21. 根据权利要求20所述的方法,其特征在于,所述方法还包括:
    根据所述第四信息,将待发送的编码数据输入至第二目标模型的所述K个第二目标分支自适应层中进行处理,得到第一数据;
    发送所述第一数据。
  22. 根据权利要求20或21所述的方法,其特征在于,所述方法还包括:
    接收第一信道字典集;
    将所述第一信道字典集和第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系;
    发送第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  23. 根据权利要求20或21所述的方法,其特征在于,所述方法还包括:
    发送第二信道字典集,所述第二信道字典集用于参考信号接收端确定所述参考信号接收端的N个第一目标分支自适应层和发送端的M个第二目标分支自适应层之间的对应关系,M、N均为正整数。
  24. 根据权利要求22或23所述的方法,其特征在于,所述第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
  25. 根据权利要求20至24任一项所述的方法,其特征在于,所述方法还包括:
    发送第二参考信号;
    接收第五信息,所述第五信息与所述第四信息相同,或者,所述第五信息指示间隔第一时间后再发送参考信号,所述第五信息是根据所述第二参考信号得到的;
    在所述第一时间内发送数据。
  26. 一种通信方法,其特征在于,包括:
    接收第一参考信号;
    将所述第一参考信号输入至第一目标模型中进行处理,得到所述第一目标模型中的K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重,所述第一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数;
    接收第一数据;
    将所述第一数据输入至所述K个第一目标分支自适应层中进行处理,得到处理后的数据,其中,所述K个第一目标分支自适应层的权重中最大权重不小于预设值。
  27. 根据权利要求26所述的方法,其特征在于,所述第一目标模型还包括第一目标信道特征提取网络和目标稀疏门控模块,
    所述第一目标信道特征提取网络用于对所述第一参考信号进行处理,得到信道分布信息和所述信道分布信息对应的信道类别信息;
    所述目标稀疏门控模块用于根据所述信道分布信息和所述信道类别信息计算得到所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重。
  28. 根据权利要求26或27所述的方法,其特征在于,所述K个第一目标分支自适应层用于分别对所述第一数据进行处理,得到所述K个第一目标分支自适应层分别处理后的数据;所述K个第一目标分支自适应层的权重用于对所述K个第一目标分支自适应层分别处理后的数据进行加权求和处理,得到所述处理后的数据。
  29. 根据权利要求26至28任一项所述的方法,其特征在于,所述方法还包括:
    当所述K个第一目标分支自适应层的权重中最大权重小于所述预设值时,对所述第一数据进行解码处理,得到处理后的数据。
  30. 根据权利要求26至29任一项所述的方法,其特征在于,所述方法还包括:
    发送第一信息,所述第一信息指示所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重,所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重用于发送端对待发送的编码数据进行处理得到所述第一数据。
  31. 根据权利要求26至30任一项所述的方法,其特征在于,所述方法还包括:
    发送第一信道字典集,所述第一信道字典集用于发送端确定所述N个第一目标分支自适应层和所述发送端的M个第二目标分支自适应层之间的对应关系,M为正整数。
  32. 根据权利要求26至30任一项所述的方法,其特征在于,所述方法还包括:
    接收第二信道字典集;
    将第一信道字典集和所述第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系;
    发送第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  33. 根据权利要求26至29任一项所述的方法,其特征在于,所述方法还包括:
    发送第四信息,所述第四信息指示K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重,所述K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重用于发送端对待发送的编码数据进行处理得到所述第一数据,所述K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重是根据所述K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重确定的。
  34. 根据权利要求26至29、33任一项所述的方法,其特征在于,所述方法还包括:
    发送第一信道字典集;
    接收第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  35. 根据权利要求26至29、33任一项所述的方法,其特征在于,所述方法还包括:
    接收第二信道字典集;
    将第一信道字典集和所述第二信道字典集进行字典对齐,得到所述第二信道字典集中的 第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  36. 根据权利要求31、32、34或35所述的方法,其特征在于,所述第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
  37. 根据权利要求26至36任一项所述的方法,其特征在于,所述方法还包括:
    当所述K个第一目标分支自适应层的权重中最大权重小于所述预设值,发送第二信息,所述第二信息指示发送端直接发送所述第一数据。
  38. 根据权利要求30所述的方法,其特征在于,所述方法还包括:
    接收第二参考信号;
    发送第三信息,所述第三信息与所述第一信息相同,或者,所述第三信息指示发送端间隔第一时间后再发送参考信号,所述第三信息是根据所述第二参考信号得到的;
    接收所述发送端在所述第一时间内发送的数据。
  39. 一种通信方法,其特征在于,包括:
    发送第一参考信号;
    当接收到第一信息时,所述第一信息指示所述第一参考信号接收端的第一目标模型中的K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重,根据所述第一信息,确定与所述K个第一目标分支自适应层对应的K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重,所述第一信息是根据所述第一参考信号得到的,K为正整数;
    将待发送的编码数据输入至第二目标模型的所述K个第二目标分支自适应层中进行处理,得到第一数据;
    发送所述第一数据。
  40. 根据权利要求39所述的方法,其特征在于,
    当M小于K时,更新第三目标模型,得到所述第二目标模型,使得所述第二目标模型包括所述K个第二目标分支自适应层,其中所述第三目标模型包括M个第二目标分支自适应层,M为正整数。
  41. 根据权利要求39或40所述的方法,其特征在于,
    所述K个第二目标分支自适应层用于分别对所述待发送的编码数据进行处理,得到所述K个第二目标分支自适应层分别处理后的数据;所述K个第二目标分支自适应层的权重用于对所述K个第二目标分支自适应层分别处理后的数据进行加权求和处理,得到所述第一数据。
  42. 根据权利要求39至41任一项所述的方法,其特征在于,所述方法还包括:
    当接收到第二信息时,所述第二信息指示直接发送所述待发送的编码数据,则发送所述待发送的编码数据,所述第二信息是根据所述第一参考信号得到的。
  43. 根据权利要求39至42任一项所述的方法,其特征在于,所述方法还包括:
    接收第一信道字典集;
    将所述第一信道字典集和第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  44. 根据权利要求39至42任一项所述的方法,其特征在于,所述方法还包括:
    发送第二信道字典集;
    接收第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与第一信道字典集中的第一目标分支自适应层之间的对应关系。
  45. 根据权利要求43或44所述的方法,其特征在于,所述第二信道字典集包含第二目标 分支自适应层与信道标签之间的对应关系。
  46. 根据权利要求39至45任一项所述的方法,其特征在于,所述方法还包括:
    发送第二参考信号;
    接收第三信息,所述第三信息与所述第一信息相同;或者,所述第三信息用于指示间隔第一时间后再发送参考信号,所述第三信息是根据所述第二参考信号得到的;
    在所述第一时间内发送数据。
  47. 一种通信方法,其特征在于,包括:
    发送第一参考信号;
    当接收到第四信息时,所述第四信息指示K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重,根据所述第四信息,将待发送的编码数据输入至第二目标模型的所述K个第二目标分支自适应层中进行处理,得到第一数据,所述第四信息是根据所述第一参考信号得到的,K为正整数;
    发送所述第一数据。
  48. 根据权利要求47所述的方法,其特征在于,
    所述K个第二目标分支自适应层用于分别对所述待发送的编码数据进行处理,得到所述K个第二目标分支自适应层分别处理后的数据;所述K个第二目标分支自适应层的权重用于对所述K个第二目标分支自适应层分别处理后的数据进行加权求和处理,得到所述第一数据。
  49. 根据权利要求47或48所述的方法,其特征在于,所述方法还包括:
    当接收到第二信息时,所述第二信息指示直接发送所述待发送的编码数据,则发送所述待发送的编码数据,所述第二信息是根据所述第一参考信号得到的。
  50. 根据权利要求47至49任一项所述的方法,其特征在于,所述方法还包括:
    接收第一信道字典集;
    将所述第一信道字典集和第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系;
    发送第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与第一信道字典集中的第一目标分支自适应层之间的对应关系。
  51. 根据权利要求47至49任一项所述的方法,其特征在于,所述方法还包括:
    发送第二信道字典集,所述第二信道字典集用于参考信号接收端确定所述参考信号接收端的N个第一目标分支自适应层和发送端的M个第二目标分支自适应层之间的对应关系,M、N均为正整数。
  52. 根据权利要求50或51所述的方法,其特征在于,所述第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
  53. 根据权利要求47至52任一项所述的方法,其特征在于,所述方法还包括:
    发送第二参考信号;
    接收第五信息,所述第五信息与所述第四信息相同;或者,所述第五信息用于指示间隔第一时间后再发送参考信号,所述第五信息是根据所述第二参考信号得到的;
    在所述第一时间内发送数据。
  54. 一种通信装置,其特征在于,包括:
    通信模块,用于接收第一参考信号;
    处理模块,用于将所述第一参考信号输入至第一目标模型中进行处理,得到所述第一目标模型中的K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重,所述第一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数;
    所述通信模块,还用于发送第一信息,所述第一信息指示所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重,所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重用于发送端对待发送的编码数据进行处理。
  55. 根据权利要求54所述的装置,其特征在于,所述通信模块,还用于:
    接收第一数据,所述第一数据是所述发送端对待发送的编码数据进行处理得到的;
    所述处理模块,还用于将所述第一数据输入至所述K个第一目标分支自适应层中进行处理,得到处理后的数据。
  56. 根据权利要求54或55所述的装置,其特征在于,所述通信模块,还用于:
    发送第一信道字典集,所述第一信道字典集用于所述发送端确定所述N个第一目标分支自适应层和所述发送端的M个第二目标分支自适应层之间的对应关系,M为正整数。
  57. 根据权利要求54或55所述的装置,其特征在于,所述通信模块,还用于:
    接收第二信道字典集;
    所述处理模块,还用于将第一信道字典集和所述第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系;
    所述通信模块,还用于发送第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  58. 根据权利要求56或57所述的装置,其特征在于,所述第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
  59. 根据权利要求54至58任一项所述的装置,其特征在于,所述通信模块,还用于:
    接收第二参考信号;
    发送第三信息,所述第三信息与所述第一信息相同,或者,所述第三信息指示所述发送端间隔第一时间后再发送参考信号,所述第三信息是根据所述第二参考信号得到的;
    接收所述发送端在所述第一时间内发送的数据。
  60. 一种通信装置,其特征在于,包括:
    通信模块,用于接收第一参考信号;
    处理模块,用于将所述第一参考信号输入至第一目标模型中进行处理,得到所述第一目标模型中的K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重,所述第一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数;
    所述通信模块,还用于发送第四信息,所述第四信息指示K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重,所述K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重用于发送端对待发送的编码数据进行处理,所述K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重是根据所述K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重确定的。
  61. 根据权利要求60所述的装置,其特征在于,所述通信模块,还用于:
    接收第一数据,所述第一数据是所述发送端对待发送的编码数据进行处理得到的;
    所述处理模块,还用于将所述第一数据输入至所述K个第一目标分支自适应层中进行处理,得到处理后的数据。
  62. 根据权利要求60或61所述的装置,其特征在于,所述通信模块,还用于:
    发送第一信道字典集,所述第一信道字典集用于所述发送端确定所述N个第一目标分支自适应层和所述发送端的M个第二目标分支自适应层之间的对应关系,M为正整数;
    接收第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  63. 根据权利要求60或61所述的装置,其特征在于,所述通信模块,还用于:
    接收第二信道字典集;
    所述处理模块,还用于将第一信道字典集和所述第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  64. 根据权利要求62或63所述的装置,其特征在于,所述第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
  65. 根据权利要求60至64任一项所述的装置,其特征在于,所述通信模块,还用于:
    接收第二参考信号;
    发送第五信息,所述第五信息与所述第四信息相同,或者,所述第五信息指示所述发送端间隔第一时间后再发送参考信号,所述第五信息是根据所述第二参考信号得到的;
    接收所述发送端在所述第一时间内发送的数据。
  66. 一种通信装置,其特征在于,包括:
    通信模块,用于发送第一参考信号;
    所述通信模块,还用于接收第一信息,所述第一信息指示所述第一参考信号接收端的第一目标模型中的K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重,所述第一信息是根据所述第一参考信号得到的,K为正整数。
  67. 根据权利要求66所述的装置,其特征在于,所述装置还包括处理模块,用于:
    根据所述第一信息,确定与所述K个第一目标分支自适应层对应的K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重;
    所述处理模块,还用于将待发送的编码数据输入至第二目标模型的所述K个第二目标分支自适应层中进行处理,得到第一数据;
    所述通信模块,还用于发送所述第一数据。
  68. 根据权利要求67所述的装置,其特征在于,
    当M小于K时,所述处理模块,还用于更新第三目标模型,得到所述第二目标模型,使得所述第二目标模型包括所述K个第二目标分支自适应层,其中所述第三目标模型包括M个第二目标分支自适应层,M为正整数。
  69. 根据权利要求66至68任一项所述的装置,其特征在于,所述通信模块,还用于:
    接收第一信道字典集;
    所述处理模块,还用于将所述第一信道字典集和第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  70. 根据权利要求66至68任一项所述的装置,其特征在于,所述通信模块,还用于:
    发送第二信道字典集;
    接收第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与第一信道字典集中的第一目标分支自适应层之间的对应关系。
  71. 根据权利要求69或70所述的装置,其特征在于,所述第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
  72. 根据权利要求66至71任一项所述的装置,其特征在于,所述通信模块,还用于:
    发送第二参考信号;
    接收第三信息,所述第三信息与所述第一信息相同,或者,所述第三信息指示间隔第一时间后再发送参考信号,所述第三信息是根据所述第二参考信号得到的;
    在所述第一时间内发送数据。
  73. 一种通信装置,其特征在于,包括:
    通信模块,用于发送第一参考信号;
    所述通信模块,还用于接收第四信息,所述第四信息指示K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重,所述第四信息是根据所述第一参考信号得到的,K为正整数。
  74. 根据权利要求73所述的装置,其特征在于,所述装置还包括处理模块,用于:
    根据所述第四信息,将待发送的编码数据输入至第二目标模型的所述K个第二目标分支自适应层中进行处理,得到第一数据;
    所述通信模块,还用于发送所述第一数据。
  75. 根据权利要求73或74所述的装置,其特征在于,所述通信模块,还用于:
    接收第一信道字典集;
    所述处理模块,还用于将所述第一信道字典集和第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系;
    所述通信模块,还用于发送第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  76. 根据权利要求73或74所述的装置,其特征在于,所述通信模块,还用于:
    发送第二信道字典集,所述第二信道字典集用于参考信号接收端确定所述参考信号接收端的N个第一目标分支自适应层和发送端的M个第二目标分支自适应层之间的对应关系,M、N均为正整数。
  77. 根据权利要求75或76所述的装置,其特征在于,所述第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
  78. 根据权利要求73至77任一项所述的装置,其特征在于,所述通信模块,还用于:
    发送第二参考信号;
    接收第五信息,所述第五信息与所述第四信息相同,或者,所述第五信息指示间隔第一时间后再发送参考信号,所述第五信息是根据所述第二参考信号得到的;
    在所述第一时间内发送数据。
  79. 一种通信装置,其特征在于,包括:
    通信模块,用于接收第一参考信号;
    处理模块,用于将所述第一参考信号输入至第一目标模型中进行处理,得到所述第一目标模型中的K个第一目标分支自适应层以及所述K个第一目标分支自适应层的权重,所述第一目标模型包括N个第一目标分支自适应层,K不大于N,K、N均为正整数;
    所述通信模块,还用于接收第一数据;
    所述处理模块,还用于将所述第一数据输入至所述K个第一目标分支自适应层中进行处理,得到处理后的数据,其中,所述K个第一目标分支自适应层的权重中最大权重不小于预设值。
  80. 根据权利要求79所述的装置,其特征在于,所述第一目标模型还包括第一目标信道特征提取网络和目标稀疏门控模块,
    所述第一目标信道特征提取网络用于对所述第一参考信号进行处理,得到信道分布信息和所述信道分布信息对应的信道类别信息;
    所述目标稀疏门控模块用于根据所述信道分布信息和所述信道类别信息计算得到所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重。
  81. 根据权利要求79或80所述的装置,其特征在于,所述K个第一目标分支自适应层用于分别对所述第一数据进行处理,得到所述K个第一目标分支自适应层分别处理后的数据;所述K个第一目标分支自适应层的权重用于对所述K个第一目标分支自适应层分别处理后的数据进行加权求和处理,得到所述处理后的数据。
  82. 根据权利要求79至81任一项所述的装置,其特征在于,所述处理模块,还用于:
    当所述K个第一目标分支自适应层的权重中最大权重小于所述预设值时,对所述第一数据进行解码处理,得到处理后的数据。
  83. 根据权利要求79至82任一项所述的装置,其特征在于,所述通信模块,还用于:
    发送第一信息,所述第一信息指示所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重,所述K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重用于发送端对待发送的编码数据进行处理得到所述第一数据。
  84. 根据权利要求79至83任一项所述的装置,其特征在于,所述通信模块,还用于:
    发送第一信道字典集,所述第一信道字典集用于发送端确定所述N个第一目标分支自适应层和所述发送端的M个第二目标分支自适应层之间的对应关系,M为正整数。
  85. 根据权利要求79至83任一项所述的装置,其特征在于,所述通信模块,还用于:
    接收第二信道字典集;
    所述处理模块,还用于将第一信道字典集和所述第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系;
    所述通信模块,还用于发送第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  86. 根据权利要求79至82任一项所述的装置,其特征在于,所述通信模块,还用于:
    发送第四信息,所述第四信息指示K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重,所述K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重用于发送端对待发送的编码数据进行处理得到所述第一数据,所述K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重是根据所述K个第一目标分支自适应层以及所述 K个第一目标分支自适应层的权重确定的。
  87. 根据权利要求79至82、86任一项所述的装置,其特征在于,所述通信模块,还用于:
    发送第一信道字典集;
    接收第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  88. 根据权利要求79至82、86任一项所述的装置,其特征在于,所述通信模块,还用于:
    接收第二信道字典集;
    所述处理模块,还用于将第一信道字典集和所述第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  89. 根据权利要求84、85、87或88所述的装置,其特征在于,所述第一信道字典集包含第一目标分支自适应层与信道标签之间的对应关系。
  90. 根据权利要求79至89任一项所述的装置,其特征在于,所述通信模块,还用于:
    当所述K个第一目标分支自适应层的权重中最大权重小于所述预设值,发送第二信息,所述第二信息指示发送端直接发送所述第一数据。
  91. 根据权利要求83所述的装置,其特征在于,所述通信模块,还用于:
    接收第二参考信号;
    发送第三信息,所述第三信息与所述第一信息相同,或者,所述第三信息指示发送端间隔第一时间后再发送参考信号,所述第三信息是根据所述第二参考信号得到的;
    接收所述发送端在所述第一时间内发送的数据。
  92. 一种通信装置,其特征在于,包括:
    通信模块,用于发送第一参考信号;
    处理模块,用于当接收到第一信息时,所述第一信息指示所述第一参考信号接收端的第一目标模型中的K个第一目标分支自适应层和所述K个第一目标分支自适应层的权重,根据所述第一信息,确定与所述K个第一目标分支自适应层对应的K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重,所述第一信息是根据所述第一参考信号得到的,K为正整数;
    所述处理模块,还用于将待发送的编码数据输入至第二目标模型的所述K个第二目标分支自适应层中进行处理,得到第一数据;
    所述通信模块,还用于发送所述第一数据。
  93. 根据权利要求92所述的装置,其特征在于,
    当M小于K时,所述处理模块,还用于更新第三目标模型,得到所述第二目标模型,使得所述第二目标模型包括所述K个第二目标分支自适应层,其中所述第三目标模型包括M个第二目标分支自适应层,M为正整数。
  94. 根据权利要求92或93所述的装置,其特征在于,
    所述K个第二目标分支自适应层用于分别对所述待发送的编码数据进行处理,得到所述K个第二目标分支自适应层分别处理后的数据;所述K个第二目标分支自适应层的权重用于对所述K个第二目标分支自适应层分别处理后的数据进行加权求和处理,得到所述第一数据。
  95. 根据权利要求92至94任一项所述的装置,其特征在于,所述通信模块,还用于:
    当接收到第二信息时,所述第二信息指示直接发送所述待发送的编码数据,则发送所述待发送的编码数据,所述第二信息是根据所述第一参考信号得到的。
  96. 根据权利要求92至95任一项所述的装置,其特征在于,所述通信模块,还用于:
    接收第一信道字典集;
    所述处理模块,还用于将所述第一信道字典集和第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系。
  97. 根据权利要求92至95任一项所述的装置,其特征在于,所述通信模块,还用于:
    发送第二信道字典集;
    接收第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与第一信道字典集中的第一目标分支自适应层之间的对应关系。
  98. 根据权利要求96或97所述的装置,其特征在于,所述第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
  99. 根据权利要求92至98任一项所述的装置,其特征在于,所述通信模块,还用于:
    发送第二参考信号;
    接收第三信息,所述第三信息与所述第一信息相同;或者,所述第三信息用于指示间隔第一时间后再发送参考信号,所述第三信息是根据所述第二参考信号得到的;
    在所述第一时间内发送数据。
  100. 一种通信装置,其特征在于,包括:
    通信模块,用于发送第一参考信号;
    处理模块,用于当接收到第四信息时,所述第四信息指示K个第二目标分支自适应层和所述K个第二目标分支自适应层的权重,根据所述第四信息,将待发送的编码数据输入至第二目标模型的所述K个第二目标分支自适应层中进行处理,得到第一数据,所述第四信息是根据所述第一参考信号得到的,K为正整数;
    所述通信模块,还用于发送所述第一数据。
  101. 根据权利要求100所述的装置,其特征在于,
    所述K个第二目标分支自适应层用于分别对所述待发送的编码数据进行处理,得到所述K个第二目标分支自适应层分别处理后的数据;所述K个第二目标分支自适应层的权重用于对所述K个第二目标分支自适应层分别处理后的数据进行加权求和处理,得到所述第一数据。
  102. 根据权利要求100或101所述的装置,其特征在于,所述通信模块,还用于:
    当接收到第二信息时,所述第二信息指示直接发送所述待发送的编码数据,则发送所述待发送的编码数据,所述第二信息是根据所述第一参考信号得到的。
  103. 根据权利要求100至102任一项所述的装置,其特征在于,所述通信模块,还用于:
    接收第一信道字典集;
    所述处理模块,还用于将所述第一信道字典集和第二信道字典集进行字典对齐,得到所述第二信道字典集中的第二目标分支自适应层与所述第一信道字典集中的第一目标分支自适应层之间的对应关系;
    所述通信模块,还用于发送第一指示信息,所述第一指示信息用于指示所述第二信道字典集中的第二目标分支自适应层与第一信道字典集中的第一目标分支自适应层之间的对应关 系。
  104. 根据权利要求100至102任一项所述的装置,其特征在于,所述通信模块,还用于:
    发送第二信道字典集,所述第二信道字典集用于参考信号接收端确定所述参考信号接收端的N个第一目标分支自适应层和发送端的M个第二目标分支自适应层之间的对应关系,M、N均为正整数。
  105. 根据权利要求103或104所述的装置,其特征在于,所述第二信道字典集包含第二目标分支自适应层与信道标签之间的对应关系。
  106. 根据权利要求100至105任一项所述的方法,其特征在于,所述处理模块,还用于:
    发送第二参考信号;
    接收第五信息,所述第五信息与所述第四信息相同;或者,所述第五信息用于指示间隔第一时间后再发送参考信号,所述第五信息是根据所述第二参考信号得到的;
    在所述第一时间内发送数据。
  107. 一种通信装置,其特征在于,所述通信装置包括一个或多个处理器;其中,所述一个或多个处理器用于执行一个或多个存储器存储的计算机程序,使得所述通信装置实现如权利要求1至6任一项所述的方法,或,实现如权利要求7至12任一项所述的方法,或,实现如权利要求13至19任一项所述的方法,或,实现如权利要求20至25任一项所述的方法,或,实现如权利要求26至38任一项所述的方法,或,实现如权利要求39至46任一项所述的方法,或,实现如权利要求47至53任一项所述的方法。
  108. 根据权利要求107所述的通信装置,其特征在于,所述通信装置还包括所述一个或多个存储器。
  109. 根据权利要求107或108所述的通信装置,其特征在于,所述通信装置为芯片或芯片系统。
  110. 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质存储有指令,当所述指令被处理器执行时,实现如权利要求1至6任一项所述的方法,或,实现如权利要求7至12任一项所述的方法,或,实现如权利要求13至19任一项所述的方法,或,实现如权利要求20至25任一项所述的方法,或,实现如权利要求26至38任一项所述的方法,或,实现如权利要求39至46任一项所述的方法,或,实现如权利要求47至53任一项所述的方法。
  111. 一种计算机程序产品,其特征在于,包括计算机程序,当所述计算机程序被执行时,实现如权利要求1至6任一项所述的方法,或,实现如权利要求7至12任一项所述的方法,或,实现如权利要求13至19任一项所述的方法,或,实现如权利要求20至25任一项所述的方法,或,实现如权利要求26至38任一项所述的方法,或,实现如权利要求39至46任一项所述的方法,或,实现如权利要求47至53任一项所述的方法。
  112. 一种通信系统,其特征在于,包括如权利要求54-59任一项所述的装置,以及如权利要求66-72任一项所述的装置,或,所述通信系统包括如权利要求60-65任一项所述的装 置,以及如权利要求73-78任一项所述的装置,或,所述通信系统包括如权利要求79-91任一项所述的装置,以及如权利要求92-99任一项所述的装置或如权利要求100-106任一项所述的装置。
PCT/CN2022/132654 2022-11-17 2022-11-17 一种通信方法、装置及系统 WO2024103352A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/132654 WO2024103352A1 (zh) 2022-11-17 2022-11-17 一种通信方法、装置及系统

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/132654 WO2024103352A1 (zh) 2022-11-17 2022-11-17 一种通信方法、装置及系统

Publications (1)

Publication Number Publication Date
WO2024103352A1 true WO2024103352A1 (zh) 2024-05-23

Family

ID=91083485

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/132654 WO2024103352A1 (zh) 2022-11-17 2022-11-17 一种通信方法、装置及系统

Country Status (1)

Country Link
WO (1) WO2024103352A1 (zh)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190268185A1 (en) * 2017-09-30 2019-08-29 Zte Corporation Information transmission method and apparatus
CN111819892A (zh) * 2018-03-06 2020-10-23 三星电子株式会社 使用上行链路srs测量进行基于ai的ue速度估计的方法和装置
WO2021051987A1 (zh) * 2019-09-18 2021-03-25 华为技术有限公司 神经网络模型训练的方法和装置
WO2022040160A2 (en) * 2020-08-18 2022-02-24 Qualcomm Incorporated Neural network or layer configuration indicator for a channel state information scheme
CN115136505A (zh) * 2020-02-28 2022-09-30 高通股份有限公司 基于神经网络的信道状态信息反馈
EP4068169A1 (en) * 2019-12-31 2022-10-05 Huawei Technologies Co., Ltd. Search method for machine learning model and related apparatus and device

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190268185A1 (en) * 2017-09-30 2019-08-29 Zte Corporation Information transmission method and apparatus
CN111819892A (zh) * 2018-03-06 2020-10-23 三星电子株式会社 使用上行链路srs测量进行基于ai的ue速度估计的方法和装置
WO2021051987A1 (zh) * 2019-09-18 2021-03-25 华为技术有限公司 神经网络模型训练的方法和装置
EP4068169A1 (en) * 2019-12-31 2022-10-05 Huawei Technologies Co., Ltd. Search method for machine learning model and related apparatus and device
CN115136505A (zh) * 2020-02-28 2022-09-30 高通股份有限公司 基于神经网络的信道状态信息反馈
WO2022040160A2 (en) * 2020-08-18 2022-02-24 Qualcomm Incorporated Neural network or layer configuration indicator for a channel state information scheme

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
ERICSSON: "Addition of indoor industrial channel model", 3GPP TSG-WG1 MEETING #97 R1-1907919, 20 May 2019 (2019-05-20), XP051740177 *

Similar Documents

Publication Publication Date Title
US11528048B2 (en) Mixing coefficient data for processing mode selection
US11082149B2 (en) Communication system having a configurable modulation order and an associated method and apparatus
US11070410B1 (en) Transmitter
US20240211770A1 (en) Communication method and apparatus
US11139866B2 (en) Machine-learning-based detection and reconstruction from low-resolution samples
WO2020228374A1 (zh) 通信方法和装置
CN111542111B (zh) 用于提供定时同步的方法、装置和系统
CN115118566B (zh) 改进无线通信中信息的发送
US9184895B2 (en) Method and apparatus of interference alignment in cellular network
Garcia Marti et al. A mixture density channel model for deep learning-based wireless physical layer design
WO2024103352A1 (zh) 一种通信方法、装置及系统
US10872045B1 (en) Method and apparatus for symbol detection
CN116938658A (zh) 数据处理方法、装置及系统
KR20190093019A (ko) 클라우드 무선 접속 네트워크의 수신 간섭 상관도에 따른 신호 송수신 방법 및 장치
US10797930B1 (en) Apparatus and method for detection of received signals
WO2021151980A1 (en) Finite resolution decomposition of a matrix and matrix-vector multiplication
US20230259742A1 (en) Communication method, apparatus, and system
US11617183B2 (en) Demapping received data
WO2024208316A1 (zh) 一种压缩信息的反馈方法及装置
WO2024027424A1 (zh) 一种数据质量度量方法及装置
WO2023125996A1 (zh) 一种上行预编码方法及装置
WO2024131900A1 (zh) 一种通信的方法和通信装置
WO2024008004A1 (zh) 一种通信方法及装置
US20240305402A1 (en) Method and apparatus for performing signal compensation based on frequency resource allocation
WO2022262687A1 (zh) 一种数据处理方法及装置

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: 22965534

Country of ref document: EP

Kind code of ref document: A1