CN118120156A - Training method, device, equipment and storage medium of channel information feedback model - Google Patents
Training method, device, equipment and storage medium of channel information feedback model Download PDFInfo
- Publication number
- CN118120156A CN118120156A CN202180103608.5A CN202180103608A CN118120156A CN 118120156 A CN118120156 A CN 118120156A CN 202180103608 A CN202180103608 A CN 202180103608A CN 118120156 A CN118120156 A CN 118120156A
- Authority
- CN
- China
- Prior art keywords
- information
- channel information
- encoder
- transfer learning
- decoder
- Prior art date
- Legal status (The legal status 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 status listed.)
- Pending
Links
- 238000012549 training Methods 0.000 title claims abstract description 134
- 238000000034 method Methods 0.000 title claims abstract description 81
- 230000000873 masking effect Effects 0.000 claims abstract description 127
- 239000011159 matrix material Substances 0.000 claims description 166
- 238000013526 transfer learning Methods 0.000 claims description 154
- 238000005070 sampling Methods 0.000 claims description 37
- 230000006835 compression Effects 0.000 claims description 35
- 238000007906 compression Methods 0.000 claims description 35
- 238000012545 processing Methods 0.000 claims description 23
- 238000013508 migration Methods 0.000 claims description 16
- 230000005012 migration Effects 0.000 claims description 16
- 238000011084 recovery Methods 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 9
- 230000002776 aggregation Effects 0.000 claims description 8
- 238000004220 aggregation Methods 0.000 claims description 8
- 230000004931 aggregating effect Effects 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 6
- 230000006837 decompression Effects 0.000 claims description 4
- 238000004891 communication Methods 0.000 abstract description 39
- 238000010586 diagram Methods 0.000 description 18
- 238000013135 deep learning Methods 0.000 description 14
- 238000005516 engineering process Methods 0.000 description 8
- 238000013528 artificial neural network Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 6
- 230000008859 change Effects 0.000 description 4
- 238000013461 design Methods 0.000 description 4
- 238000009826 distribution Methods 0.000 description 4
- 230000007774 longterm Effects 0.000 description 4
- 238000010295 mobile communication Methods 0.000 description 4
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- 238000009827 uniform distribution Methods 0.000 description 2
- 208000033748 Device issues Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B7/00—Radio transmission systems, i.e. using radiation field
- H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
- H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
- H04B7/06—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Cable Transmission Systems, Equalization Of Radio And Reduction Of Echo (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
The application discloses a training method, device and equipment of a channel information feedback model and a storage medium, and relates to the technical field of communication. The method comprises the following steps: masking operation is carried out on the initial channel information to obtain masking channel information; inputting the mask channel information into the channel information feedback model, and outputting recovered channel information; and training the channel information feedback model based on the error between the recovered channel information and the initial channel information.
Description
The present application relates to the field of communications technologies, and in particular, to a training method, apparatus, device, and storage medium for a channel information feedback model.
The terminal device typically generates channel Information by measuring Channel State Information (CSI), and feeds back the channel Information to the network device.
In the related art, a channel information feedback scheme based on deep learning has been introduced: the channel information is regarded as an image to be compressed, the encoder is used for compressing and feeding back the channel information, and the decoder is used for reconstructing the compressed channel information at the transmitting end, so that the channel information can be reserved to a greater extent.
Disclosure of Invention
The embodiment of the application provides a training method, a training device, training equipment and a training storage medium for a channel information feedback model. The technical scheme is as follows:
According to one aspect of the present application, there is provided a training method of a channel information feedback model, applied to a source side terminal, the method comprising:
masking operation is carried out on the initial channel information to obtain masking channel information;
inputting the mask channel information into the channel information feedback model, and outputting recovered channel information;
And training the channel information feedback model based on the error between the recovered channel information and the initial channel information.
According to one aspect of the present application, there is provided a training method of a channel information feedback model applied to a target side terminal, the channel information feedback model including: a second encoder and a second decoder, the method comprising:
Generating the second decoder;
Receiving second transfer learning information sent by the network device, where the second transfer learning information is used to assist in transfer learning, and the second transfer learning information includes: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;
jointly training the second encoder and the second decoder based on the matrix size information;
and sending the trained second decoder to the network equipment.
According to one aspect of the present application, there is provided a training method of a channel information feedback model, applied to a network device, the method comprising:
Transmitting second transfer learning information to the target side terminal, wherein the second transfer learning information is used for assisting in transfer learning, and the second transfer learning information comprises: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;
And receiving a second decoder sent by the target side terminal, wherein the second decoder is obtained by training after the target side terminal performs transfer learning based on the second transfer learning information.
According to an aspect of the present application, there is provided a training apparatus of a channel information feedback model, the apparatus comprising: the system comprises a mask module, a model processing module and a training module;
the mask module is used for performing mask operation on the initial channel information to obtain mask channel information;
The model processing module is used for inputting the mask channel information into the channel information feedback model and outputting recovered channel information;
The training module is configured to train the channel information feedback model based on an error between the recovered channel information and the initial channel information.
According to an aspect of the present application, there is provided a training apparatus of a channel information feedback model including: a second encoder and a second decoder, the apparatus comprising: the system comprises a decoder generating module, an information receiving module, a training module and a decoder sending module;
the decoder generating module is used for generating the second decoder;
The information receiving module is configured to receive second transfer learning information sent by the network device, where the second transfer learning information is used to assist in transfer learning, and the second transfer learning information includes: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;
The training module is used for carrying out joint training on the second encoder and the second decoder based on the matrix size information;
the decoder sending module is configured to send the trained second decoder to the network device.
According to an aspect of the present application, there is provided a training apparatus of a channel information feedback model, the apparatus comprising: an information transmitting module and a decoder receiving module;
The information sending module is configured to send second transfer learning information to the target side terminal, where the second transfer learning information is used to assist in transfer learning, and the second transfer learning information includes: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;
The decoder receiving module is configured to receive a second decoder sent by the target side terminal, where the second decoder is obtained by training the target side terminal after performing transfer learning based on the second transfer learning information.
According to an aspect of the present application, there is provided a terminal device including: a processor; wherein,
The processor is used for carrying out masking operation on the initial channel information to obtain masking channel information;
the processor is used for inputting the mask channel information into a channel information feedback model and outputting recovery channel information;
the processor is configured to train the channel information feedback model based on an error between the recovered channel information and the initial channel information.
According to an aspect of the present application, there is provided a terminal device including: a processor and a transceiver coupled to the processor; wherein,
The processor is used for generating a second decoder;
The transceiver is configured to receive second transfer learning information sent by the network device, where the second transfer learning information is used to assist in transfer learning, and the second transfer learning information includes: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;
The processor is configured to jointly train the second encoder and the second decoder based on the matrix size information;
the transceiver is configured to send the trained second decoder to the network device.
According to an aspect of the present application, there is provided a network device comprising: a transceiver; wherein,
The transceiver is configured to send second transfer learning information to the target-side terminal, where the second transfer learning information is used to assist in transfer learning, and the second transfer learning information includes: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;
The transceiver is configured to receive a second decoder sent by the target side terminal, where the second decoder is obtained by training the target side terminal after performing transfer learning based on the second transfer learning information.
According to one aspect of the present application, there is provided a computer readable storage medium having stored therein executable instructions that are loaded and executed by a processor to implement a training method of a channel information feedback model as described in the above aspect.
According to an aspect of an embodiment of the present application, there is provided a chip including programmable logic circuits and/or program instructions for implementing the training method of the channel information feedback model described in the above aspect when the chip is run on a computer device.
According to an aspect of the present application, there is provided a computer program product which, when run on a processor of a computer device, causes the computer device to perform the method of training a channel information feedback model as described in the above aspects.
The technical scheme provided by the embodiment of the application at least comprises the following beneficial effects:
Under the condition of executing a channel information feedback scheme based on deep learning, when model training is carried out, the mask operation is utilized to shield the initial channel information of a part, so that the redundant information input during the channel information feedback model training is reduced, the resource expenditure of the model training is reduced, the training speed of the model is accelerated, and the generalization capability of the training model is improved.
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic diagram of a channel information feedback system provided by an exemplary embodiment of the present application;
FIG. 2 is a schematic diagram of a pre-training-fine mode based transfer learning provided in an exemplary embodiment of the present application;
FIG. 3 is a block diagram of a communication system provided by an exemplary embodiment of the present application;
FIG. 4 is a flowchart of a training method of a channel information feedback model provided by an exemplary embodiment of the present application;
FIG. 5 is a schematic diagram of a masking operation provided by an exemplary embodiment of the present application;
FIG. 6 is a schematic diagram of a channel information feedback model in the form of an encoder-decoder provided by an exemplary embodiment of the present application;
FIG. 7 is a flowchart of a training method of a channel information feedback model provided by an exemplary embodiment of the present application;
FIG. 8 is a schematic diagram of a masking operation provided by an exemplary embodiment of the present application;
Fig. 9 is a schematic diagram of a channel information feedback system provided by an exemplary embodiment of the present application;
FIG. 10 is a flowchart of a training method of a channel information feedback model provided by an exemplary embodiment of the present application;
FIG. 11 is a flowchart of a training method for a channel information feedback model provided by an exemplary embodiment of the present application;
FIG. 12 is a flowchart of a training method for a channel information feedback model provided by an exemplary embodiment of the present application;
FIG. 13 is a schematic diagram of a training process for a channel information feedback model provided by an exemplary embodiment of the present application;
FIG. 14 is a block diagram of a training apparatus for a channel information feedback model provided by an exemplary embodiment of the present application;
FIG. 15 is a block diagram of a training apparatus for a channel information feedback model according to an exemplary embodiment of the present application;
FIG. 16 is a block diagram of a training apparatus for a channel information feedback model provided by an exemplary embodiment of the present application;
fig. 17 is a schematic structural diagram of a communication device according to an exemplary embodiment of the present application.
For the purpose of making the objects, technical solutions and advantages of the present application more apparent, the embodiments of the present application will be described in further detail with reference to the accompanying drawings.
First, technical knowledge involved in the embodiments of the present application will be briefly described:
codebook-based eigenvector feedback scheme
In the current New Radio (NR) system, a codebook-based eigenvector feedback scheme is generally adopted to enable a base station to acquire downlink CSI. Specifically, the base station sends a downlink channel state Information reference signal (CHANNEL STATE Information-REFERENCE SIGNAL, CSI-RS) to the terminal, the terminal estimates the CSI of the downlink channel by using the CSI-RS, and performs eigenvalue decomposition on the estimated downlink channel to obtain an eigenvector corresponding to the downlink channel. And the terminal calculates codeword coefficients corresponding to the feature vector in a preset codebook according to a certain rule, performs quantization feedback, and recovers the feature vector according to quantized CSI fed back by a user.
Channel information feedback scheme based on deep learning
In view of the great success of artificial intelligence (ARTIFICIAL INTELLIGENCE, AI) technology, especially deep learning, in computer vision, natural language processing, etc., the communication field has begun to try to solve the technical problems that are difficult to solve by the conventional communication method, such as deep learning, by using deep learning. The neural network architecture commonly used in deep learning is nonlinear and data-driven, can perform feature extraction on actual channel matrix data, restore channel matrix information fed back by a terminal side compression as much as possible at a base station side, and provides possibility for reducing CSI feedback overhead at the terminal side while ensuring restoring the channel information. Channel information is regarded as an image to be compressed based on the CSI feedback of the deep learning, the channel information is compressed and fed back by a deep learning self-encoder, and the compressed channel image is reconstructed at a transmitting end, so that the channel information can be reserved to a greater extent.
A typical channel information feedback system is shown in fig. 1. The whole feedback system is divided into an encoder part and a decoder part which are respectively arranged at a transmitting end and a receiving end. After the transmitting end obtains the channel information through channel estimation, the channel information matrix is compressed and encoded through a neural network of the encoder, the compressed bit stream is fed back to the receiving end through an air interface feedback link, and the receiving end recovers the channel information through the decoder according to the feedback bit stream, so that complete feedback channel information is obtained.
The encoder shown in fig. 1 uses a superposition of multiple fully-connected layers and the decoder uses a design of convolutional layers and residual structures. Illustratively, at one side of the encoder, information is input into the encoder, the information is convolved through a convolution (conv) layer, the dimension of the information is changed through a remolding (Reshape) layer, and the information is processed through a full-connection (dense) layer, so that the encoding of the information is completed; at the decoder side, the input information is processed through a full-join (dense) layer, and then the information is input to the semantic segmentation network REFINENET for processing, where REFINENET includes: and (3) reshaping (Reshape) the layer, at least one convolution (conv) layer and designing a residual structure, and then carrying out convolution (conv) on the information to finish decoding the information. With the codec framework unchanged, the network model structure inside the encoder and decoder can be flexibly designed.
Pre-training-fine-tuning mode based transfer learning
Transfer learning can be understood as utilizing existing knowledge, models, structures to help achieve learning goals on the target data. The transition learning based on the pretraining-fine tuning mode refers to: a network is trained in the source domain, directly used for data of the target domain, and fine-tuned on the target domain data, as shown in fig. 2. Therefore, the transfer learning based on the pre-training-fine tuning mode can better utilize limited computing resources and can also cope with the problem of insufficient data volume of the new scene.
The channel information feedback in the related art is a codebook-based eigenvector feedback scheme, however, the scheme simply selects an optimal feedback matrix and a corresponding feedback coefficient from the codebook according to the estimated channel, but the codebook itself is a preset finite sequence, i.e. the mapping process from the estimated channel to the channel in the codebook is quantized and lossy. Meanwhile, the fixed codebook design cannot be dynamically adjusted according to the change of the channel, so that the accuracy of the fed-back channel information is reduced, and the precoding performance is further reduced.
Further, the existing channel information feedback scheme based on deep learning utilizes a deep neural network (Deep Neural Networks, DNN), a convolutional neural network (Convolution Neural Networks, CNN) and the like to directly encode and compress the channel information obtained after channel estimation for feedback, and compared with the traditional channel information feedback based on codebooks, the feedback precision is remarkably improved. However, the model performance of the channel information feedback scheme based on deep learning is strongly related to data diversity, a large amount of real channel data is needed to provide support, the real channel data acquisition cost is high, and meanwhile, a great amount of calculation overhead is also brought in the training process.
In addition, the data distribution may change over time due to an insufficient stability of the wireless environment. Under a limited data set, even if the model is fully trained, the model performance is difficult to guarantee after the data distribution changes with the lapse of time.
Therefore, how to cope with the data distribution change brought by time lapse under different channel scenes and ensure the accuracy of channel vector compression feedback and recovery is a model generalization problem to be solved urgently.
In view of the above problems, an embodiment of the present application provides a training method for a channel information feedback model, where when a channel information feedback scheme based on deep learning is executed, initial channel information of a mask operation shielding part is utilized during model training, so as to reduce redundant information input during training of the channel information feedback model, reduce resource overhead of model training, accelerate training speed of the model, and improve generalization capability of the training model.
Fig. 3 shows a block diagram of a communication system provided by an exemplary embodiment of the present application, which may include: access network 12 and terminal equipment 14.
Access network 12 includes a number of network devices 120 therein. The network device 120 may be a base station, which is a means deployed in an access network to provide wireless communication functionality for terminals. The base stations may include various forms of macro base stations, micro base stations, relay stations, access points, and the like. In systems employing different radio access technologies, the names of base station capable devices may vary, for example in LTE systems, called enodebs or enbs; in the 5G NR-U system, it is called gNodeB or gNB. As communication technology evolves, the description of "base station" may change. For convenience, the above-described devices for providing the terminal device 14 with the wireless communication function are collectively referred to as network devices.
The terminal device 14 may include various handheld devices, vehicle mounted devices, wearable devices, computing devices, or other processing devices connected to a wireless modem, as well as various forms of user equipment, mobile Stations (MSs), terminals (TERMINAL DEVICE), and the like, having wireless communication capabilities. For convenience of description, the above-mentioned devices are collectively referred to as a terminal. The network device 120 and the terminal device 14 communicate with each other via some air interface technology, e.g. Uu interface.
Optionally, the terminal device 14 includes: a source side terminal and a target side terminal. The source terminal is a device for performing a pre-training phase of a model in transfer learning, and the target terminal is a device for performing a fine-tuning phase of a model in transfer learning.
The technical scheme of the embodiment of the application can be applied to various communication systems, such as: global system for mobile communications (Global System of Mobile Communication, GSM), code division multiple access (Code Division Multiple Access, CDMA) system, wideband code division multiple access (Wideband Code Division Multiple Access, WCDMA) system, general Packet Radio Service (GPRS), long term evolution (Long Term Evolution, LTE) system, LTE frequency division duplex (Frequency Division Duplex, FDD) system, LTE time division duplex (Time Division Duplex, TDD) system, long term evolution advanced (Advanced Long Term Evolution, LTE-a) system, new Radio (NR) system, NR system evolution system, LTE on unlicensed band (LTE-based access to Unlicensed spectrum, LTE-U) system, NR-U system, universal mobile telecommunications system (Universal Mobile Telecommunication System, UMTS), global interconnect microwave access (Worldwide Interoperability for Microwave Access, wiMAX) communication system, wireless local area network (Wireless Local Area Networks, WLAN), wireless fidelity (WIRELESS FIDELITY, WIFI), 6 th Generation mobile communication technology (6-Generation, 6G) system, next Generation communication system or other communication system, etc.
Generally, the number of connections supported by the conventional Communication system is limited and easy to implement, however, with the development of Communication technology, the mobile Communication system will support not only conventional Communication but also, for example, device-to-Device (D2D) Communication, machine-to-machine (Machine to Machine, M2M) Communication, machine type Communication (MACHINE TYPE Communication, MTC), inter-vehicle (Vehicle to Vehicle, V2V) Communication, and internet of vehicles (Vehicle to Everything, V2X) systems. The embodiments of the present application may also be applied to these communication systems.
Fig. 4 is a flowchart illustrating a training method of a channel information feedback model according to an exemplary embodiment of the present application. The method may be applied in a terminal device in a communication system as shown in fig. 3, the method comprising:
step 410: and carrying out masking operation on the initial channel information to obtain masking channel information.
The initial channel information is determined channel information after the terminal equipment performs channel estimation.
Wherein, the masking operation refers to an operation of masking part of information to reduce redundant information.
It will be appreciated that the channel information is characterized by a high degree of redundancy, and that to address this, the redundancy may be reduced by masking portions of the channel information. In the field of image processing, visual images are also characterized by high redundancy, e.g., missing pixel information can be recovered from neighboring pixel blocks. In the embodiment of the application, the method for masking the initial channel information by using the masking operation is provided, so that the redundant information is reduced.
Exemplary masking operations are schematically illustrated in fig. 5. After the masking operation is performed on the initial channel information H, masking channel information H 'is obtained, and the initial channel information H has more redundant information than the masking channel information H'. It will be appreciated that fig. 5 is merely an exemplary illustration, and that in practice, channel information may not be as similar to the presentation of the image shown in fig. 5.
Step 420: and inputting the mask channel information into a channel information feedback model, and outputting recovered channel information.
The channel information feedback model is used for carrying out compression feedback on input channel information and carrying out reconstruction recovery on the compressed channel information.
In the embodiment of the application, after masking operation is performed on the initial channel information to obtain the masked channel information, the masked channel information is used as the input of a channel information feedback model, and the masked channel information is predicted by using the channel information feedback model, so that the recovered channel information is output.
Optionally, the channel information feedback model is in the form of an encoder-decoder.
By way of example, referring in conjunction to fig. 6, a schematic diagram of processing channel information using a channel information feedback model in the form of an encoder-decoder is shown. In the current feedback period, after the transmitting end carries out channel estimation, the estimated channel information H is compressed and encoded by an encoder, and is fed back to the receiving end through a feedback link of an air interface. In more detail, the feedback link of the air interface actually transmits a feedback vector, and the feedback vector is obtained by the output of the neural network of the encoder at the transmitting end and is used as part of the input of the neural network at the receiving end for the receiving end to recover the channel information.
Step 430: the channel information feedback model is trained based on errors between the recovered channel information and the initial channel information.
After the recovered channel information output by the channel information feedback model is obtained, the recovered channel information is compared with the corresponding initial channel information to judge the accuracy of the masked content in the initial channel information predicted by the channel information feedback model, and when the recovered channel information and the corresponding initial channel information have errors, the channel information feedback model is correspondingly corrected according to the errors, so that the generated channel information feedback model has the reconstruction recovery capability of the channel information.
In summary, in the case of executing the channel information feedback scheme based on deep learning, the technical scheme provided in this embodiment uses the mask operation to mask the initial channel information of the portion during model training, so as to reduce the redundant information input during channel information feedback model training, reduce the resource overhead of model training, accelerate the training speed of the model, and improve the generalization capability of the training model.
The masking operation is further described below.
Fig. 7 is a flowchart illustrating a training method of a channel information feedback model according to an exemplary embodiment of the present application. The method may be applied in a terminal device in a communication system as shown in fig. 3, the method comprising:
Step 710: the channel matrix used to represent the initial channel information is divided into a plurality of non-overlapping matrix blocks.
Wherein, the matrix size information corresponding to each divided matrix block is the same.
Illustratively, the channel matrix used to represent the initial channel information is a 25 x 25 matrix divided into 25 5*5 matrix blocks.
Step 720: a position index is generated for the matrix blocks, constituting a matrix block sequence.
Wherein the position index is an index for characterizing the position of each matrix block in the sequence of matrix blocks.
Illustratively, 25 matrix blocks correspond to position indices of 0,1, …,24, respectively, to form a matrix block sequence.
Step 730: and sampling the matrix block sequence, and shielding matrix blocks which are not sampled in the matrix block sequence to obtain mask channel information.
That is, the sampled matrix blocks in the matrix block sequence are reserved, and the matrix blocks which are not sampled are deleted, so that the mask channel information is obtained.
Optionally, the sampling modes corresponding to the sampling include: randomly sampling; or, grid sampling. That is, the selection scheme of the masking operation includes a random masking policy and a grid masking policy. It is understood that the choice of masking operation is not limited to the two masking strategies described above, e.g. setting the masking distribution with other a priori knowledge, and is within the scope of the present application. Wherein the grid samples may be equally spaced grid samples.
Illustratively, the samples corresponding to the masking operation shown in fig. 5 are random samples, and the samples corresponding to the masking operation shown in fig. 8 are raster samples.
Illustratively, the terminal device randomly samples the matrix block sequence in a uniform distribution with a sampling rate of 50%. Illustratively, the terminal device performs grid sampling on the matrix block sequence according to uniform distribution, and the sampling rate is 25%.
It will be appreciated that the above sample rates are merely exemplary illustrations, and that embodiments of the present application do not limit the values of the sample rates. Illustratively, in the case that the number of channel information on the present side is large, a smaller sampling rate is adopted; in the case where the amount of channel information on the own side is small, a larger sampling rate is adopted.
Step 740: and inputting the mask channel information into a channel information feedback model, and outputting recovered channel information.
The specific embodiment of this step is referred to above step 420, and will not be described herein.
Step 750: the channel information feedback model is trained based on errors between the recovered channel information and the initial channel information.
The specific embodiment of this step is referred to above in step 430, and will not be described herein.
In summary, the technical solution provided in this embodiment provides different masking policies, such as a random masking policy and a grid masking policy, to execute the masking operation, so as to ensure the rationality of the masking operation.
Illustratively, in combination with the masking operation and the model structure of the encoder-decoder described above, the overall architecture of the present embodiment may be as shown in fig. 9.
In fig. 9, the following operational flow is mainly shown: masking operations, encoders, codeword stuffing, and decoders.
Masking operation: after masking operation is performed on the channel matrix H corresponding to the initial channel information, masking channel information H' is obtained.
An encoder: the mask channel information H' is input to an encoder and compression-encoded to obtain compression-encoded information.
Codeword filling: and performing codeword filling on the compressed coding information to obtain filled compressed coding information, namely, obtaining complete compressed coding information.
A decoder: the filled compression coding information is input into a decoder for decompression, and restored channel information H is obtained.
Correspondingly, if the channel information feedback model of the local side includes: a first encoder, a first decoder; the step of inputting the mask channel information into the channel information feedback model to obtain recovered channel information includes:
(1) The mask channel information is input as a model of the first encoder, and compression processing is performed on the mask channel information through the first encoder to obtain compression coding information.
(2) And filling the code word of the compression coding information to obtain filled compression coding information.
Codeword filling refers to filling codewords at mask locations.
Illustratively, the compressed coding information obtained by the first encoder is the coding of the visible matrix blocks in the matrix block sequence corresponding to the channel matrix, and the code word is filled in the corresponding position of the mask based on the position index.
(3) And taking the filled compression coding information as a model input of a first decoder, and decompressing the filled compression coding information through the first decoder to obtain the recovery channel information.
In an exemplary embodiment, the training of the channel information feedback model by the terminal device corresponds to a pre-training stage in the transfer learning of the pre-training-fine tuning mode, the terminal device is a source side terminal, the source side terminal also needs to upload the encoder pre-trained on the source side, the target side terminal executes the fine tuning stage, trains the decoder and uploads the decoder to the network device.
Fig. 10 is a flowchart illustrating a training method of a channel information feedback model according to an exemplary embodiment of the present application. The method may be applied in a communication system as shown in fig. 3, the method comprising:
Step 1010: after the training of the channel information feedback model is completed, the source side terminal sends first transfer learning information of the channel information feedback model to the network equipment.
Correspondingly, the network equipment receives first transfer learning information of the channel information feedback model sent by the source side terminal. The first transfer learning information is used for performing transfer learning on the channel information feedback model.
The channel information feedback model at the source side terminal comprises the following steps: the first encoder, the first migration learning information includes:
a first encoder.
That is, the first migration learning information carries model parameters of the first encoder.
The mask operation corresponds to the matrix size information.
That is, the first transfer learning information carries matrix size information corresponding to the masking operation performed by the source terminal. Wherein the matrix size information is used to indicate the size of each matrix block in the matrix block sequence of the input channel information feedback model.
Step 1020: and the network equipment sends second migration learning information to the target side terminal.
Correspondingly, the target side terminal receives second transfer learning information sent by the network equipment. The second transfer learning information is used for assisting transfer learning.
Wherein the second transfer learning information includes:
A second encoder.
That is, the second migration learning information carries model parameters of the second encoder. Wherein the second encoder is trained based on the masking operation.
The mask operation corresponds to the matrix size information.
That is, the second transfer learning information carries matrix size information corresponding to the masking operation performed by the source terminal. Wherein the matrix size information is used to indicate the size of each matrix block in the matrix block sequence of the input channel information feedback model.
It can be understood that the matrix size information corresponding to the mask operation in the second transfer learning information is the matrix size information corresponding to the mask operation in the first transfer learning information, and the second encoder in the second transfer learning information is obtained based on the first encoder in the first transfer learning information.
Step 1030: the target-side terminal generates a second decoder.
That is, the target-side terminal generates a new decoder.
Step 1040: the target-side terminal performs joint training on the second encoder and the second decoder based on the matrix size information.
That is, the target terminal jointly trains the second encoder and the new second decoder together under the new data set using the pre-trained second encoder to complete the transfer learning.
It can be understood that the transition learning of the pretraining-fine tuning mode refers to that a network is pretrained, and is directly used for data of a target scene, fine tuning is performed on the data of the target scene, so that an existing model in a certain scene can be realized, and other scenes can be helped to realize the same function. In the embodiment of the application, the second encoder is pre-trained, and the pre-trained second encoder and the new second decoder are used for retraining together, so that the computing resources of the target side terminal are saved.
Step 1050: and the target side terminal sends the trained second decoder to the network equipment.
Correspondingly, the network equipment receives a second decoder sent by the target side terminal, and the second decoder is obtained by training after the target side terminal performs transfer learning based on the second transfer learning information.
Optionally, after step 1050, the target side terminal uses the trained second encoder, the network device side uses the received second decoder, the target side terminal is used as a transmitting end of the channel information, the network device is used as a receiving end of the channel information, and the second encoder of the target side terminal and the second decoder of the network device side are used to implement a channel information feedback scheme based on deep learning.
In summary, according to the technical scheme provided by the embodiment, the channel information feedback model in the form of the encoder-decoder is enhanced and designed in the migration scene of the pre-training-fine tuning mode, and the mask operation is utilized to reduce the redundant information input in the pre-training stage, so that the pre-training speed of the model is accelerated, the generalization capability of the pre-training model is improved, and the model performance is improved.
Next, a mode in which the target terminal performs joint training on the second encoder and the second decoder based on the matrix size information will be described.
(1) The channel matrix for representing the initial channel information is divided into a plurality of non-overlapping matrix blocks according to the matrix size information, and the plurality of matrix blocks constitute a matrix block sequence.
Wherein the matrix size information is used to indicate the size of each matrix block in the matrix block sequence of the input channel information feedback model.
For example, if the matrix size information indicates that the size of the matrix block is 5*5, the target-side terminal divides the channel matrix corresponding to the channel information of the target-side terminal into a plurality of matrix blocks 5*5.
(2) And taking the matrix block sequence as a model input of a second encoder, and compressing the matrix block sequence through the second encoder to obtain compression coding information.
(3) And taking the compressed and encoded information as a model input of a second decoder, and decompressing the compressed and encoded information through the second decoder to obtain the recovery channel information.
(4) The second encoder and the second decoder are jointly trained based on an error between the recovered channel information and the original channel information.
After the migration to the target domain, the second encoder of the target domain supports variable length sequence input, and the second encoder inputs a complete channel matrix block sequence without masking, and accordingly, the second encoder does not need to perform codeword filling after outputting compressed encoding information, so as to fully utilize limited data of the current scene.
In one possible implementation, the second encoder is an encoder indicated to the network device by a source-side terminal.
That is, after receiving the first transfer learning information sent by the source side terminal, the network device directly sends the first transfer learning information to the target side terminal device as second transfer learning information. The second encoder in the second transfer learning information in the above embodiment is identical to the first encoder in the first transfer learning information.
In summary, according to the technical scheme provided by the embodiment, the encoder is obtained by pre-training by the source side terminal, and is migrated to the target side terminal, and the redundant information input in the pre-training stage is reduced by using the mask operation, so that the pre-training speed of the model is accelerated.
For example, the implementation is exemplarily described with reference to fig. 11, and as shown in fig. 11, the following steps are performed:
Step 1101: the source side terminal acquires the channel data and executes the mask strategy.
Illustratively, the source-side terminal divides the channel data into a regular, non-overlapping N small block matrices (patches). One position index 0,1,2, N-1 is generated for each matrix block, constituting a matrix block sequence. The matrix block sequence is then sampled and the remaining matrix blocks are masked (i.e., deleted).
Step 1102: the source side terminal jointly trains the encoder and decoder.
Illustratively, the masked channel information is used as an input to the encoder, and a codeword filling operation is added after the encoder, where the input to the decoder is the filled codeword, including the visible matrix block codeword and the filled codeword at the corresponding position of the mask.
In this embodiment, the decoder and encoder may be of asymmetric design, and the decoder may suitably reduce the number of network layers and the amount of parameters compared to the parameter scale of the encoder, thereby reducing the pre-training time.
Step 1103: and the source side terminal sends matrix size information corresponding to the encoder and the mask operation to the network equipment.
Step 1104: the network device sends matrix size information corresponding to the encoder and the masking operation to the target side terminal.
Step 1105: the target side terminal generates a new decoder.
Step 1106: and the target side terminal processes the channel information into a matrix block sequence adapted by the encoder according to the matrix size information.
Step 1107: the target-side terminal uses the pre-trained encoder in conjunction with the new decoder to retrain under the new data set to complete the model migration.
Step 1108: the target side terminal synchronizes the decoder to the network device.
In another possible implementation manner, the second encoder is a global encoder obtained by aggregating and calculating model parameters of a plurality of encoders by the network device, where the plurality of encoders are respectively from a plurality of source side terminals.
Fig. 12 is a flowchart illustrating a training method of a channel information feedback model according to an exemplary embodiment of the present application. The method may be applied in a communication system as shown in fig. 3, the method comprising:
step 1210: after training of the channel information feedback model is completed, the plurality of source side terminals respectively send first migration learning information of the channel information feedback model to the network equipment.
Correspondingly, the network equipment receives first transfer learning information of the channel information feedback model, which is respectively sent by a plurality of source side terminals. The first transfer learning information is used for performing transfer learning on the channel information feedback model.
The channel information feedback model at the source side terminal comprises the following steps: the first encoder, the first migration learning information includes:
a first encoder.
That is, the first migration learning information carries model parameters of the first encoder.
The mask operation corresponds to the matrix size information.
That is, the first transfer learning information carries matrix size information corresponding to the masking operation performed by the source terminal. Wherein the matrix size information is used to indicate the size of each matrix block in the matrix block sequence of the input channel information feedback model.
Optionally, in order to unify the masking operation, the network device issues the same masking policy parameter, which is a parameter related to the masking operation, to the plurality of source side terminals before step 1210.
Optionally, the mask policy parameter includes at least one of:
The mask operation corresponds to the matrix size information.
Wherein the matrix size information is used to indicate the size of each matrix block in the matrix block sequence of the input channel information feedback model.
The masking operation corresponds to the sampling information.
Wherein the sampling information is used to indicate the execution mode of the sampling in the masking operation. Exemplary, the sampling information includes at least one of: sampling mode; sampling rate.
Step 1220: and the network equipment performs aggregation calculation on the model parameters of the plurality of trained first encoders to obtain a global encoder.
Where aggregate computation refers to a way of computing a set of values and returning a single value. In the embodiment of the application, the model parameters of a plurality of first encoders are aggregated and calculated, and the model parameters of a final global encoder are returned.
Step 1230: and the network equipment sends second migration learning information to the target side terminal.
Correspondingly, the target side terminal receives second transfer learning information sent by the network equipment. The second transfer learning information is used for assisting transfer learning.
Wherein the second transfer learning information includes: the global encoder and the matrix size information corresponding to the masking operation are obtained by training the global encoder based on the masking operation.
It can be understood that the matrix size information corresponding to the mask operation in the second transfer learning information is the matrix size information corresponding to the mask operation in the first transfer learning information, and the global encoder in the second transfer learning information is obtained by performing aggregate calculation based on a plurality of first encoders in a plurality of first transfer learning information.
Step 1240: the target-side terminal generates a second decoder.
That is, the target-side terminal generates a new decoder.
Step 1250: the target side terminal performs joint training on the global encoder and the second decoder.
That is, the target terminal jointly trains the global encoder and the new second decoder together under the new data set using the pre-trained global encoder to complete the migration learning.
Step 1260: and the target side terminal sends the trained second decoder to the network equipment.
In summary, according to the technical scheme provided by the embodiment, the plurality of source side terminals cooperatively train to obtain the shared global encoder, the data redundancy degree under the plurality of terminal devices is higher, the data redundancy can be reduced more remarkably by using the mask operation, the characterization capability of extracting potential features of the model is enhanced, and the pre-training speed of the model is accelerated.
Illustratively, the above implementation is exemplarily described with reference to fig. 13, and as shown in fig. 13, the following steps are performed:
Step 1301, unify masking policy: the network device uniformly configures mask policy parameters and then uniformly distributes the mask policy parameters to n candidate source side terminals: source side terminal 1, source side terminal 2,..source side terminal n.
Step 1302, pre-training the encoder: the source terminals each perform a masking operation, and train the encoder-decoder with the masked masking channel information as input.
The self-encoder network architecture based on the masking policy is consistent for a single terminal device. The assembly comprises: masking operation, encoder, codeword filling, decoder. Each terminal device needs the above four components, and the working architecture and flow of each terminal device can be referred to the embodiment shown in fig. 9, which is not described in detail herein.
In this embodiment, the decoder and encoder may be of asymmetric design, and the decoder may suitably reduce the number of network layers and the amount of parameters compared to the parameter scale of the encoder, thereby reducing the pre-training time.
Step 1303, upload encoder: each source terminal deletes the decoder part to save the memory resource of the device, only reserves the encoder part, and uploads the encoder synchronization to the network device.
Step 1304, aggregate computing: and the base station server or the aerial computing node performs aggregation computation on the encoder model parameters of each collaborative source side terminal to obtain a global encoder.
Step 1305, issuing global encoder and matrix size information: the network device, such as a base station server or an air computing node, issues the matrix size information corresponding to the global encoder and the masking operation to the target terminal.
It will be appreciated that there may be a plurality of destination side terminals, and not limited to source side terminals, all terminals under the network device may be candidate destination side terminals, in particular looking at the system policy.
Step 1306, fine tuning: the target side terminal processes the existing channel information data into a matrix block sequence by utilizing the matrix size information, and directly inputs the complete matrix block sequence to the encoder-decoder without masking.
It should be noted that the encoder here is a global encoder, but the target-side terminal needs to regenerate a decoder of an initialized state. The encoder model size here can be scaled up appropriately to achieve better decoding performance.
Step 1307, upload encoder: the encoder is a model which is finally required to be deployed at the receiving end, so the target side terminal also needs to send the trained decoder to the network equipment, and the network equipment can be ensured to correctly analyze the code words sent by the encoder of the target side terminal and restore the code words into complete channel information.
As shown in the above steps, in the process of implementing the transfer learning, each participant does not need to share the data in the local device, so that the data privacy and security of the participants are fully ensured.
The above-described method embodiments may be implemented individually or in combination, and the present application is not limited thereto.
In the above embodiments, the step performed by the source side terminal may separately implement the training method for the channel information feedback model on the source side terminal, the step performed by the target side terminal may separately implement the training method for the channel information feedback model on the target side terminal, and the step performed by the network device may separately implement the training method for the channel information feedback model on the network device side.
Fig. 14 is a block diagram of a training apparatus for a channel information feedback model according to an exemplary embodiment of the present application, which may be implemented as a source side terminal or as a part of the source side terminal, the apparatus including: a masking module 1402, a model processing module 1404, and a training module 1406;
The masking module 1402 is configured to perform masking operation on the initial channel information to obtain masking channel information;
The model processing module 1404 is configured to input the mask channel information into the channel information feedback model, and output recovered channel information;
The training module 1406 is configured to train the channel information feedback model based on an error between the recovered channel information and the initial channel information.
In an alternative embodiment, the masking module 1402 is configured to:
dividing a channel matrix representing the initial channel information into a plurality of non-overlapping matrix blocks;
Generating a position index for the matrix blocks to form a matrix block sequence;
And sampling the matrix block sequence, and shielding matrix blocks which are not sampled in the matrix block sequence to obtain the mask channel information.
In an alternative embodiment, the sampling mode corresponding to the sampling includes:
Randomly sampling;
Or alternatively, the first and second heat exchangers may be,
And (5) grid sampling.
In an alternative embodiment, the channel information feedback model includes: a first encoder and a first decoder;
the model processing module 1404 is configured to:
the mask channel information is used as a model input of the first encoder, and compression processing is carried out on the mask channel information through the first encoder to obtain compression coding information;
Performing codeword filling on the compressed coding information to obtain filled compressed coding information;
And taking the filled compression coding information as a model input of the first decoder, and decompressing the filled compression coding information through the first decoder to obtain the recovery channel information.
In an alternative embodiment, the apparatus further comprises: an information reporting module;
The information reporting module is configured to send first transfer learning information of the channel information feedback model to a network device after training of the channel information feedback model is completed, where the first transfer learning information is used for performing transfer learning on the channel information feedback model.
In an alternative embodiment, the channel information feedback model includes: a first encoder, the first transfer learning information comprising:
The first encoder;
And the masking operation corresponds to the matrix size information.
In an alternative embodiment, the apparatus further comprises: a parameter receiving module;
The parameter receiving module is configured to receive a mask policy parameter issued by a network device, where the mask policy parameter is a parameter related to the mask operation.
In an alternative embodiment, the masking policy parameters include at least one of:
the matrix size information corresponding to the mask operation;
and the masking operation is used for corresponding sampling information.
Fig. 15 is a block diagram showing a training apparatus for a channel information feedback model according to an exemplary embodiment of the present application, which may be implemented as a target-side terminal or as a part of the target-side terminal, the apparatus including: a decoder generation module 1502, an information reception module 1504, a training module 1506, and a decoder transmission module 1508;
The decoder generating module 1502 is configured to generate the second decoder;
the information receiving module 1504 is configured to receive second transfer learning information sent by the network device, where the second transfer learning information is used to assist in transfer learning, and the second transfer learning information includes: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;
The training module 1506 is configured to jointly train the second encoder and the second decoder based on the matrix size information;
The decoder sending module 1508 is configured to send the trained second decoder to the network device.
In an alternative embodiment, the training module 1506 is configured to:
dividing a channel matrix for representing initial channel information into a plurality of non-overlapping matrix blocks according to the matrix size information, wherein the matrix blocks form a matrix block sequence;
The matrix block sequence is used as the model input of the second encoder, and compression processing is carried out on the matrix block sequence through the second encoder to obtain compression coding information;
The compression coding information is used as the model input of the second decoder, and decompression processing is carried out on the compression coding information through the second decoder to obtain recovery channel information;
The second encoder and the second decoder are jointly trained based on an error between the recovered channel information and the initial channel information.
In an alternative embodiment, the second encoder is an encoder indicated to the network device by a source side terminal.
In an optional embodiment, the second encoder is a global encoder obtained by aggregating, by the network device, model parameters of a plurality of encoders, where the plurality of encoders are respectively from a plurality of source-side terminals.
Fig. 16 is a block diagram of a training apparatus for a channel information feedback model according to an exemplary embodiment of the present application, which may be implemented as a network device or as a part of a network device, the apparatus including: an information sending module 1602 and a decoder receiving module 1604;
The information sending module 1602 is configured to send second transfer learning information to the target-side terminal, where the second transfer learning information is used to assist in transfer learning, and the second transfer learning information includes: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;
The decoder receiving module 1604 is configured to receive a second decoder sent by the target side terminal, where the second decoder is obtained by training the target side terminal after performing transfer learning based on the second transfer learning information.
In an alternative embodiment, the second encoder is an encoder indicated to the network device by a source side terminal;
The apparatus further comprises: an information receiving module;
The information receiving module is configured to receive first transfer learning information sent by the source side terminal, where the first transfer learning information is used to assist in transfer learning, and the first transfer learning information includes: the first encoder and the masking operation correspond to matrix size information.
In an optional embodiment, the second encoder is a global encoder obtained by performing aggregate calculation on model parameters of a plurality of encoders by a network device;
the apparatus further comprises: an information receiving module and an aggregation calculating module;
the information receiving module is configured to receive a plurality of first transfer learning information sent by a plurality of source side terminals, where the first transfer learning information is used to assist in transfer learning, and the first transfer learning information includes: matrix size information corresponding to the first encoder and the masking operation;
and the aggregation calculation module is used for carrying out aggregation calculation on the model parameters of the plurality of trained first encoders to obtain the global encoder.
In an alternative embodiment, the apparatus further comprises: a parameter configuration module;
The parameter configuration module is configured to issue the same mask policy parameter to a plurality of source side terminals, where the mask policy parameter is a parameter related to the mask operation.
In an alternative embodiment, the masking policy parameters include at least one of:
the matrix size information corresponding to the mask operation;
and the masking operation is used for corresponding sampling information.
Fig. 17 shows a schematic structural diagram of a communication device (terminal device or network device) provided in an exemplary embodiment of the present application, the communication device 1700 includes: a processor 1701, a transceiver 1702, and a memory 1703.
The processor 1701 includes one or more processing cores, and the processor 1701 executes various functional applications by running software programs and modules.
The transceiver 1702 may be used to receive and transmit information, and the transceiver 1702 may be a communication chip.
The memory 1703 may be used for storing a computer program, and the processor 1701 is used for executing the computer program to implement the steps performed by the communication device in the above-described method embodiment.
Further, memory 1703 may be implemented by any type of volatile or nonvolatile memory device or combination thereof, including but not limited to: random-Access Memory (RAM) and Read-Only Memory (ROM), erasable programmable Read-Only Memory (EPROM), electrically erasable programmable Read-Only Memory (ELECTRICALLY ERASABLE PROGRAMMABLE READ-Only Memory, EEPROM), flash Memory or other solid state Memory technology, compact disc Read-Only (Compact Disc Read-Only Memory, CD-ROM), high density digital video disc (Digital Video Disc, DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices.
When the communication device is implemented as a source terminal, the processor 1701 and the transceiver 1702 in the embodiments of the present application may execute steps executed by the source terminal in any of the methods shown in the foregoing embodiments, which are not described herein.
In one possible implementation, when the communication device is implemented as a source-side terminal,
The processor 1701 is configured to perform a masking operation on the initial channel information to obtain masking channel information;
The processor 1701 is configured to input the mask channel information into a channel information feedback model and output recovery channel information;
The processor 1701 is configured to train the channel information feedback model based on an error between the recovered channel information and the initial channel information.
When the communication device is implemented as the target terminal, the processor 1701 and the transceiver 1702 in the embodiments of the present application may execute steps executed by the target terminal in any of the methods shown in the foregoing embodiments, which are not described herein.
In one possible implementation, when the communication device is implemented as a target-side terminal,
The processor 1701 is configured to generate a second decoder;
The transceiver 1702 is configured to receive second transfer learning information sent by a network device, where the second transfer learning information is used to assist in transfer learning, and the second transfer learning information includes: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;
The processor 1701 is configured to jointly train the second encoder and the second decoder based on the matrix size information;
The transceiver 1702 is configured to send the trained second decoder to the network device.
When the communication device is implemented as a network device, the processor 1701 and the transceiver 1702 in the embodiments of the present application may execute steps executed by the network device in any of the methods shown in the foregoing embodiments, which are not described herein again.
In one possible implementation, when the communication device is implemented as a network device,
The transceiver 1702 is configured to send second transfer learning information to the target-side terminal, where the second transfer learning information is used to assist in transfer learning, and the second transfer learning information includes: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;
The transceiver 1702 is configured to receive a second decoder sent by the target side terminal, where the second decoder is obtained by training the target side terminal after performing transfer learning based on the second transfer learning information.
In an exemplary embodiment, there is also provided a computer readable storage medium having stored therein at least one instruction, at least one program, a set of codes, or a set of instructions, which are loaded and executed by a processor to implement the training method of the channel information feedback model performed by a communication device provided by the above respective method embodiments.
In an exemplary embodiment, a chip is also provided, the chip including programmable logic circuits and/or program instructions for implementing the training method of the channel information feedback model described in the above aspect when the chip is run on a computer device.
In an exemplary embodiment, there is also provided a computer program product, which when run on a processor of a computer device, causes the computer device to perform the training method of the channel information feedback model of the above aspect.
It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program for instructing relevant hardware, where the program may be stored in a computer readable storage medium, and the storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.
The foregoing description of the preferred embodiments of the present application is not intended to limit the application, but rather, the application is to be construed as limited to the appended claims.
Claims (40)
- A method for training a channel information feedback model, the method comprising:masking operation is carried out on the initial channel information to obtain masking channel information;inputting the mask channel information into the channel information feedback model, and outputting recovered channel information;And training the channel information feedback model based on the error between the recovered channel information and the initial channel information.
- The method of claim 1, wherein masking the initial channel information to obtain masked channel information comprises:dividing a channel matrix representing the initial channel information into a plurality of non-overlapping matrix blocks;Generating a position index for the matrix blocks to form a matrix block sequence;And sampling the matrix block sequence, and shielding matrix blocks which are not sampled in the matrix block sequence to obtain the mask channel information.
- The method according to claim 2, wherein the sampling mode corresponding to the sampling includes:Randomly sampling;Or alternatively, the first and second heat exchangers may be,And (5) grid sampling.
- A method according to any one of claims 1 to 3, wherein the channel information feedback model comprises: a first encoder and a first decoder;The step of inputting the mask channel information into the channel information feedback model to obtain recovered channel information includes:the mask channel information is used as a model input of the first encoder, and compression processing is carried out on the mask channel information through the first encoder to obtain compression coding information;Performing codeword filling on the compressed coding information to obtain filled compressed coding information;And taking the filled compression coding information as a model input of the first decoder, and decompressing the filled compression coding information through the first decoder to obtain the recovery channel information.
- A method according to any one of claims 1 to 3, wherein the method further comprises:After training of the channel information feedback model is completed, first transfer learning information of the channel information feedback model is sent to network equipment, and the first transfer learning information is used for transfer learning of the channel information feedback model.
- The method of claim 5, wherein the channel information feedback model comprises: a first encoder, the first transfer learning information comprising:The first encoder;And the masking operation corresponds to the matrix size information.
- A method according to any one of claims 1 to 3, wherein the method further comprises:A masking policy parameter issued by a network device is received, the masking policy parameter being a parameter related to the masking operation.
- The method of claim 7, wherein the mask policy parameters include at least one of:the matrix size information corresponding to the mask operation;and the masking operation is used for corresponding sampling information.
- The training method of the channel information feedback model is characterized in that the channel information feedback model comprises the following steps: a second encoder and a second decoder, the method comprising:Generating the second decoder;Receiving second transfer learning information sent by the network device, where the second transfer learning information is used to assist in transfer learning, and the second transfer learning information includes: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;jointly training the second encoder and the second decoder based on the matrix size information;and sending the trained second decoder to the network equipment.
- The method of claim 9, wherein the jointly training the second encoder and the second decoder based on matrix size information comprises:dividing a channel matrix for representing initial channel information into a plurality of non-overlapping matrix blocks according to the matrix size information, wherein the matrix blocks form a matrix block sequence;The matrix block sequence is used as the model input of the second encoder, and compression processing is carried out on the matrix block sequence through the second encoder to obtain compression coding information;The compression coding information is used as the model input of the second decoder, and decompression processing is carried out on the compression coding information through the second decoder to obtain recovery channel information;The second encoder and the second decoder are jointly trained based on an error between the recovered channel information and the initial channel information.
- The method according to claim 9 or 10, wherein,The second encoder is an encoder indicated to the network device by a source side terminal.
- The method according to claim 9 or 10, wherein,The second encoder is a global encoder obtained by aggregating and calculating model parameters of a plurality of encoders by the network equipment, wherein the plurality of encoders are respectively from a plurality of source side terminals.
- A method for training a channel information feedback model, the method comprising:Transmitting second transfer learning information to the target side terminal, wherein the second transfer learning information is used for assisting in transfer learning, and the second transfer learning information comprises: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;And receiving a second decoder sent by the target side terminal, wherein the second decoder is obtained by training after the target side terminal performs transfer learning based on the second transfer learning information.
- The method of claim 13, wherein the second encoder is an encoder indicated to the network device by a source side terminal;The method further comprises the steps of:Receiving first transfer learning information sent by the source side terminal, wherein the first transfer learning information is used for assisting in transfer learning, and the first transfer learning information comprises: the first encoder and the masking operation correspond to matrix size information.
- The method of claim 13, wherein the second encoder is a global encoder obtained by aggregating model parameters of a plurality of encoders by a network device;The method further comprises the steps of:Receiving a plurality of first transfer learning information sent by a plurality of source side terminals respectively, wherein the first transfer learning information is used for assisting in transfer learning, and the first transfer learning information comprises: matrix size information corresponding to the first encoder and the masking operation;and performing aggregate calculation on the model parameters of the plurality of trained first encoders to obtain the global encoder.
- The method of claim 15, wherein the method further comprises:And transmitting the same masking strategy parameters to a plurality of source side terminals, wherein the masking strategy parameters are parameters related to the masking operation.
- The method of claim 16, wherein the mask policy parameters include at least one of:the matrix size information corresponding to the mask operation;and the masking operation is used for corresponding sampling information.
- A training apparatus for a channel information feedback model, the apparatus comprising: the system comprises a mask module, a model processing module and a training module;the mask module is used for performing mask operation on the initial channel information to obtain mask channel information;The model processing module is used for inputting the mask channel information into the channel information feedback model and outputting recovered channel information;The training module is configured to train the channel information feedback model based on an error between the recovered channel information and the initial channel information.
- The apparatus of claim 18, wherein the masking module is configured to:dividing a channel matrix representing the initial channel information into a plurality of non-overlapping matrix blocks;Generating a position index for the matrix blocks to form a matrix block sequence;And sampling the matrix block sequence, and shielding matrix blocks which are not sampled in the matrix block sequence to obtain the mask channel information.
- The apparatus of claim 19, wherein the sampling pattern corresponding to the sampling comprises:Randomly sampling;Or alternatively, the first and second heat exchangers may be,And (5) grid sampling.
- The apparatus according to any one of claims 18 to 20, wherein the channel information feedback model comprises: a first encoder and a first decoder;The model processing module is used for:the mask channel information is used as a model input of the first encoder, and compression processing is carried out on the mask channel information through the first encoder to obtain compression coding information;Performing codeword filling on the compressed coding information to obtain filled compressed coding information;And taking the filled compression coding information as a model input of the first decoder, and decompressing the filled compression coding information through the first decoder to obtain the recovery channel information.
- The apparatus according to any one of claims 18 to 20, further comprising: an information reporting module;The information reporting module is configured to send first transfer learning information of the channel information feedback model to a network device after training of the channel information feedback model is completed, where the first transfer learning information is used for performing transfer learning on the channel information feedback model.
- The apparatus of claim 22, wherein the channel information feedback model comprises: a first encoder, the first transfer learning information comprising:The first encoder;And the masking operation corresponds to the matrix size information.
- The apparatus according to any one of claims 18 to 20, further comprising: a parameter receiving module;The parameter receiving module is configured to receive a mask policy parameter issued by a network device, where the mask policy parameter is a parameter related to the mask operation.
- The apparatus of claim 24, wherein the mask policy parameter comprises at least one of:the matrix size information corresponding to the mask operation;and the masking operation is used for corresponding sampling information.
- A training device for a channel information feedback model, wherein the channel information feedback model comprises: a second encoder and a second decoder, the apparatus comprising: the system comprises a decoder generating module, an information receiving module, a training module and a decoder sending module;the decoder generating module is used for generating the second decoder;The information receiving module is configured to receive second transfer learning information sent by the network device, where the second transfer learning information is used to assist in transfer learning, and the second transfer learning information includes: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;The training module is used for carrying out joint training on the second encoder and the second decoder based on the matrix size information;the decoder sending module is configured to send the trained second decoder to the network device.
- The apparatus of claim 26, wherein the training module is configured to:dividing a channel matrix for representing initial channel information into a plurality of non-overlapping matrix blocks according to the matrix size information, wherein the matrix blocks form a matrix block sequence;The matrix block sequence is used as the model input of the second encoder, and compression processing is carried out on the matrix block sequence through the second encoder to obtain compression coding information;The compression coding information is used as the model input of the second decoder, and decompression processing is carried out on the compression coding information through the second decoder to obtain recovery channel information;The second encoder and the second decoder are jointly trained based on an error between the recovered channel information and the initial channel information.
- The apparatus of claim 26 or 27, wherein the device comprises a plurality of sensors,The second encoder is an encoder indicated to the network device by a source side terminal.
- The apparatus of claim 26 or 27, wherein the device comprises a plurality of sensors,The second encoder is a global encoder obtained by aggregating and calculating model parameters of a plurality of encoders by the network equipment, wherein the plurality of encoders are respectively from a plurality of source side terminals.
- A training apparatus for a channel information feedback model, the apparatus comprising: an information transmitting module and a decoder receiving module;The information sending module is configured to send second transfer learning information to the target side terminal, where the second transfer learning information is used to assist in transfer learning, and the second transfer learning information includes: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;The decoder receiving module is configured to receive a second decoder sent by the target side terminal, where the second decoder is obtained by training the target side terminal after performing transfer learning based on the second transfer learning information.
- The apparatus of claim 30, wherein the second encoder is an encoder indicated to the network device by a source side terminal;The apparatus further comprises: an information receiving module;The information receiving module is configured to receive first transfer learning information sent by the source side terminal, where the first transfer learning information is used to assist in transfer learning, and the first transfer learning information includes: the first encoder and the masking operation correspond to matrix size information.
- The apparatus of claim 30, wherein the second encoder is a global encoder obtained by aggregating model parameters of a plurality of encoders by a network device;the apparatus further comprises: an information receiving module and an aggregation calculating module;the information receiving module is configured to receive a plurality of first transfer learning information sent by a plurality of source side terminals, where the first transfer learning information is used to assist in transfer learning, and the first transfer learning information includes: matrix size information corresponding to the first encoder and the masking operation;and the aggregation calculation module is used for carrying out aggregation calculation on the model parameters of the plurality of trained first encoders to obtain the global encoder.
- The apparatus of claim 32, wherein the apparatus further comprises: a parameter configuration module;The parameter configuration module is configured to issue the same mask policy parameter to a plurality of source side terminals, where the mask policy parameter is a parameter related to the mask operation.
- The apparatus of claim 33, wherein the mask policy parameter comprises at least one of:the matrix size information corresponding to the mask operation;and the masking operation is used for corresponding sampling information.
- A terminal device, characterized in that the terminal device comprises: a processor; wherein,The processor is used for carrying out masking operation on the initial channel information to obtain masking channel information;the processor is used for inputting the mask channel information into a channel information feedback model and outputting recovery channel information;the processor is configured to train the channel information feedback model based on an error between the recovered channel information and the initial channel information.
- A terminal device, characterized in that the terminal device comprises: a processor and a transceiver coupled to the processor; wherein,The processor is used for generating a second decoder;The transceiver is configured to receive second transfer learning information sent by the network device, where the second transfer learning information is used to assist in transfer learning, and the second transfer learning information includes: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;The processor is configured to jointly train the second encoder and the second decoder based on the matrix size information;the transceiver is configured to send the trained second decoder to the network device.
- A network device, the network device comprising: a transceiver; wherein,The transceiver is configured to send second migration learning information to the target-side terminal, where the second migration learning information is used to assist in migration learning, and the second migration learning information includes: the second encoder and the matrix size information corresponding to the masking operation are obtained by training based on the masking operation;The transceiver is configured to receive a second decoder sent by the target side terminal, where the second decoder is obtained by training the target side terminal after performing transfer learning based on the second transfer learning information.
- A computer readable storage medium having stored therein executable instructions that are loaded and executed by a processor to implement a method of training a channel information feedback model according to any of claims 1 to 17.
- A chip comprising programmable logic circuits and/or program instructions for implementing a training method of a channel information feedback model according to any of claims 1 to 17 when the chip is running.
- A computer program product or computer program, characterized in that it comprises computer instructions stored in a computer readable storage medium, from which a processor reads and executes the computer instructions to implement a training method of a channel information feedback model according to any of claims 1 to 17.
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/CN2021/143874 WO2023123429A1 (en) | 2021-12-31 | 2021-12-31 | Method and device for training channel information feedback model, apparatus, and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118120156A true CN118120156A (en) | 2024-05-31 |
Family
ID=86997128
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202180103608.5A Pending CN118120156A (en) | 2021-12-31 | 2021-12-31 | Training method, device, equipment and storage medium of channel information feedback model |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN118120156A (en) |
| WO (1) | WO2023123429A1 (en) |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2018403734A1 (en) * | 2018-01-22 | 2020-09-03 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Information determining method, device, and computer storage medium |
| CN108847876B (en) * | 2018-07-26 | 2021-03-02 | 东南大学 | Large-scale MIMO time-varying channel state information compression feedback and reconstruction method |
| KR20210138947A (en) * | 2020-05-13 | 2021-11-22 | 한양대학교 에리카산학협력단 | Method And System For Allocating Multi-Channel Resource |
| CN111901024B (en) * | 2020-07-29 | 2021-11-05 | 燕山大学 | MIMO channel state information feedback method based on fitting depth learning resistance |
-
2021
- 2021-12-31 CN CN202180103608.5A patent/CN118120156A/en active Pending
- 2021-12-31 WO PCT/CN2021/143874 patent/WO2023123429A1/en unknown
Also Published As
| Publication number | Publication date |
|---|---|
| WO2023123429A1 (en) | 2023-07-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116034381A (en) | Communication method and communication device | |
| US20230354081A1 (en) | Information quantization method, terminal device, and network device | |
| WO2023011472A1 (en) | Method for feeding back channel state information, method for receiving channel state information, and terminal, base station, and computer-readable storage medium | |
| CN118120156A (en) | Training method, device, equipment and storage medium of channel information feedback model | |
| WO2022236785A1 (en) | Channel information feedback method, receiving end device, and transmitting end device | |
| CN116436551A (en) | Channel information transmission method and device | |
| WO2023115254A1 (en) | Data processing method and device | |
| CN114157722A (en) | Data transmission method and device | |
| US20240154670A1 (en) | Method and apparatus for feedback channel status information based on machine learning in wireless communication system | |
| EP4398527A1 (en) | Model processing method, electronic device, network device, and terminal device | |
| US20240259072A1 (en) | Model processing method, electronic device, network device, and terminal device | |
| WO2023015499A1 (en) | Wireless communication method and device | |
| WO2023060503A1 (en) | Information processing method and apparatus, device, medium, chip, product, and program | |
| WO2022217502A1 (en) | Information processing method and apparatus, communication device, and storage medium | |
| TWI830543B (en) | Information processing methods, devices, terminals and network equipment | |
| US20240048207A1 (en) | Method and apparatus for transmitting and receiving feedback information based on artificial neural network | |
| WO2023116155A1 (en) | Neural network operation method and apparatus, and storage medium | |
| US20230354096A1 (en) | Binary variational (biv) csi coding | |
| WO2022151064A1 (en) | Information sending method and apparatus, information receiving method and apparatus, device, and medium | |
| WO2024032701A1 (en) | Channel state information processing method and apparatus | |
| WO2024026792A1 (en) | Communication method and apparatus, device, storage medium, chip, and program product | |
| CN117044145A (en) | Method for acquiring reference signal and communication equipment | |
| KR102689928B1 (en) | Decoding method and apparatus based on polar code in communication system | |
| WO2024008004A1 (en) | Communication method and apparatus | |
| WO2024032775A1 (en) | Quantization method and apparatus |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |