Model training method, communication system, and storage medium

Abstract

The present disclosure provides a model training method, a communication device, a communication system, and a storage medium. The method is executed by a first device and comprises: acquiring at least one of a first model, a second model, and first information, the first information being used for training the first model and / or the second model; and on the basis of a data set measured by a first device and at least one of the first information, the first model, and the second model, determining a third model and / or a fourth model, the third model being the same as the first model or the third model being obtained by retraining the first model, the third model being used for the first device to perform channel state information (CSI) compression, and the fourth model being used for a second device to perform CSI recovery. A third model and / or a fourth model can be determined on the basis of a data set measured by a first device, so that a model can be trained by using actually measured data, thereby avoiding the problem of performance degradation of a trained model caused by inconsistent data distributions of a first device side and a second device side when the trained model is used for reasoning.

Description

Model training method, communication system and storage medium TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of communication, and particularly relates to a model training method, a communication system and a storage medium. BACKGROUND

[0002] In the field of communication, a bilateral artificial intelligence model can be used to feed back channel state information, reduce feedback overhead of a terminal or improve feedback accuracy of the channel state information. At present, the terminal side and the network side can train the bilateral artificial intelligence model independently or cooperatively, but the training method may cause inconsistency between data distribution of the terminal side and data distribution of the network side, resulting in performance degradation when the trained model is directly used for inference.

[0003] SUMMARY

[0004] The present disclosure provides a model training method, a communication device, a communication system and a storage medium.

[0005] According to a first aspect of an embodiment of the present disclosure, a model training method is provided, which is performed by a first device, and the method comprises: obtaining at least one of a first model, a second model and first information, the first information being used for training the first model and / or the second model; determining a third model and / or a fourth model based on a data set measured by the first device and at least one of the first information, the first model and the second model, the third model being the same as the first model or the third model being obtained by retraining the first model, the third model being used for channel state information (CSI) compression by the first device, and the fourth model being used for CSI recovery by a second device.

[0006] In the above method, the third model and / or the fourth model can be determined according to the data set measured by the first device, so that the model is trained by using the actually measured data, and the inference performance of the model is improved.

[0007] According to a second aspect of an embodiment of the present disclosure, a model training method is provided, which is performed by a second device, and the method comprises: sending at least one of a first model, a second model and first information to a first device, the first information being used for training the first model and / or the second model; determining a third model and / or a fourth model based on a data set measured by the first device and at least one of the first information, the first model and the second model, the third model being the same as the first model or the third model being obtained by retraining the first model, the fourth model being the same as the second model or the fourth model being obtained by retraining the second model, the third model being used for channel state information (CSI) compression by the first device, and the fourth model being used for CSI recovery by the second device.

[0008] In the method, the third model and / or the fourth model can be determined according to the data set measured by the first device, the model is trained by using the actually measured data, and the inference performance of the model is improved.

[0009] According to a third aspect of the embodiments of the present disclosure, a first device is provided, including a transceiver module, configured to obtain at least one of a first model, a second model, and first information, the first information being used for training the first model and / or the second model; and a processing module, configured to determine a third model and / or a fourth model based on a data set measured by the first device and at least one of the first information, the first model, and the second model, the third model being the same as the first model or the third model being obtained by retraining the first model, the third model being used for channel state information (CSI) compression by the first device, and the fourth model being used for CSI recovery by a second device.

[0010] According to a fourth aspect of the embodiments of the present disclosure, a second device is provided, including a transceiver module, configured to send at least one of a first model, a second model, and first information to a first device, the first information being used for training the first model and / or the second model; and a processing module, configured to determine a third model and / or a fourth model based on a data set measured by the first device and at least one of the first information, the first model, and the second model, the third model being the same as the first model or the third model being obtained by retraining the first model, the third model being used for channel state information (CSI) compression by the first device, and the fourth model being used for CSI recovery by the second device.

[0011] According to a fifth aspect of the embodiments of the present disclosure, a communication device is provided, including one or more processors; and wherein the one or more processors are configured to invoke instructions to cause the communication device to perform the method described in any one of the first aspect of the present disclosure, or to perform the method described in any one of the second aspect of the present disclosure.

[0012] According to a sixth aspect of the embodiments of the present disclosure, a communication system is provided, including a first device and a second device, wherein the first device is configured to implement the method of the first aspect, and the second device is configured to implement the method of the second aspect.

[0013] According to a seventh aspect of the embodiments of the present disclosure, a storage medium is provided, the storage medium storing instructions, when the instructions are executed on a communication device, causing the communication device to perform the method of any one of the first aspect or the second aspect.

[0014] According to an eighth aspect of the embodiments of the present disclosure, a computer program product is provided, including a computer program, when the computer program is executed by a processor, implementing the method of any one of the first aspect or the second aspect. BRIEF DESCRIPTION OF DRAWINGS

[0015] The above mentioned and / or additional aspects and advantages of the present disclosure will become apparent and more readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:

[0016] FIG. 1 is a schematic diagram of an architecture of some communication systems according to embodiments of the present disclosure;

[0017] FIG. 2 is a schematic diagram of an interaction of a model training method according to embodiments of the present disclosure;

[0018] FIG. 3 is a schematic diagram of a flow of a model training method according to embodiments of the present disclosure;

[0019] FIG. 4 is a schematic diagram of a flow of another model training method according to embodiments of the present disclosure;

[0020] FIG. 5 is a schematic diagram of a flow of CSI compression feedback of a bilateral AI model based on CSI generation sub-model and CSI recovery sub-model according to embodiments of the present disclosure;

[0021] FIG. 6a is a schematic diagram of a structure of a first device according to an embodiment of the present disclosure;

[0022] FIG. 6b is a schematic diagram of a structure of a second device according to an embodiment of the present disclosure;

[0023] FIG. 7a is a schematic diagram of a structure of a communication device according to an embodiment of the present disclosure;

[0024] FIG. 7b is a schematic diagram of a structure of a chip according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0025] The present disclosure provides a model training method and a communication device, a communication system and a storage medium.

[0026] In a first aspect, the present disclosure provides a model training method, executed by a first device, the method comprising: obtaining at least one of a first model, a second model and first information, the first information being used for training the first model and / or the second model; determining a third model and / or a fourth model based on at least one of a data set measured by the first device and the first information, the first model and the second model, the third model being the same as the first model or the third model being obtained by retraining the first model, the fourth model being the same as the second model or the fourth model being obtained by retraining the second model, the third model being used for CSI compression by the first device, the fourth model being used for CSI recovery by a second device.

[0027] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, the model is trained by using the actually measured data, and the inference performance of the model is improved.

[0028] In some embodiments of the first aspect, in some embodiments, obtaining at least one of the first model, the second model, and the first information includes: obtaining at least one of the first model, the second model, and the first information from the second device, the first information including model parameters of the first model and / or the second model and a training data set, and the first model and / or the second model being trained by the second device based on the training data set.

[0029] In the above embodiments, at least one of the first model, the second model, and the first information can be obtained from the second device, so as to determine the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model.

[0030] In some embodiments of the first aspect, in some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model includes: training a fifth model based on the first model and model parameters of the first model by using the training data set; and training the third model based on the data set measured by the first device and the fifth model, wherein the second model is the fourth model.

[0031] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, the model is trained by using the actually measured data, and the inference performance of the model is improved.

[0032] In some embodiments of the first aspect, in some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model includes: training the third model by using a second training type Type2 based on the data set measured by the first device, the training data set, the first model, and model parameters of the first model, wherein the second model is the fourth model.

[0033] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, the model is trained by using the actually measured data, and the inference performance of the model is improved.

[0034] In some embodiments of the first aspect, determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model, and the second model comprises: sending the first device measured data set to the second device, the first device measured data set being used by the second device to train the third model; and receiving the third model and / or model parameters of the third model sent by the second device, wherein the second model is used as the fourth model.

[0035] In the above embodiments, the third model and / or the fourth model can be determined based on the first device measured data set, so that the model is trained using actually measured data, and the inference performance of the model is improved.

[0036] In some embodiments of the first aspect, determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model, and the second model comprises: training the third model using the training data set and / or the first device measured data set, the second model, and model parameters of the second model, wherein the second model is used as the fourth model.

[0037] In the above embodiments, the third model and / or the fourth model can be determined based on the first device measured data set, so that the model is trained using actually measured data, and the inference performance of the model is improved.

[0038] In some embodiments of the first aspect, determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model, and the second model comprises: sending the first device measured data set to the second device, the first device measured data set being used by the second device to train the second model; receiving the second model and / or model parameters of the second model sent by the second device; and training the third model based on the second model and / or the model parameters of the second model, wherein the second model is used as the fourth model.

[0039] In the above embodiments, the third model and / or the fourth model can be determined based on the first device measured data set, so that the model is trained using actually measured data, and the inference performance of the model is improved.

[0040] In some embodiments of the first aspect, obtaining at least one of the first model, the second model, and the first information comprises: obtaining at least one of the first model, the second model, and the first information trained by the second device or the first device, wherein the first information comprises at least one of target CSI, feedback CSI, and recovered CSI, the target CSI is an input of training the first model, the feedback CSI is an output of training the first model and an input of training the second model, and the recovered CSI is an output of training the second model.

[0041] In the above embodiments, at least one of the first model, the second model, and the first information can be acquired to determine the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model.

[0042] With reference to some embodiments of the first aspect, in some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model comprises: training a sixth model based on the feedback CSI and the target CSI; training the sixth model based on the data set measured by the first device and the second model by using a second training type Type2 to obtain the third model, wherein the second model is the fourth model.

[0043] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so that the model is trained by using the actually measured data, and the inference performance of the model is improved.

[0044] With reference to some embodiments of the first aspect, in some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model comprises: sending the data set measured by the first device to the second device, the data set measured by the first device being used by the second device to train a seventh model; receiving the feedback CSI and / or the target CSI sent by the second device for training the seventh model; and training the third model based on the feedback CSI and / or the target CSI for training the seventh model, wherein the second model is the fourth model.

[0045] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so that the model is trained by using the actually measured data, and the inference performance of the model is improved.

[0046] With reference to some embodiments of the first aspect, in some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model comprises: training an eighth model based on the feedback CSI and the recovered CSI; training a ninth model based on the data set measured by the first device and the eighth model; and training the third model based on the data set measured by the first device, the ninth model, and the second model by using a second training type Type2, wherein the second model is the fourth model.

[0047] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so that the model is trained by using the actually measured data, and the inference performance of the model is improved.

[0048] In some embodiments of the first aspect, based on the first device measured data set and at least one of the first information, the first model, and the second model, determining the third model and / or the fourth model comprises: training an eighth model based on the feedback CSI and the recovered CSI; and training the third model based on the first device measured data set and the eighth model using a first training type Type 1, wherein the second model is the fourth model.

[0049] In the above embodiments, the third model and / or the fourth model can be determined based on the first device measured data set, so that the model is trained using actually measured data, and the inference performance of the model is improved.

[0050] In some embodiments of the first aspect, based on the first device measured data set and at least one of the first information, the first model, and the second model, determining the third model and / or the fourth model comprises: sending the first device measured data set to the second device, the first device measured data set being used to train the first model and / or the second model; receiving the second model and / or the model parameters of the second model sent by the second device; and training the third model based on the first device measured data set, the second model, and / or the model parameters of the second model, wherein the second model is the fourth model.

[0051] In the above embodiments, the third model and / or the fourth model can be determined based on the first device measured data set, so that the model is trained using actually measured data, and the inference performance of the model is improved.

[0052] In some embodiments of the first aspect, obtaining at least one of the first model, the second model, and the first information comprises: obtaining at least one of a standardized first model, a standardized second model, and the first information, the first information comprising at least one of the model parameters of the first model and / or the second model, and the standardized training data set.

[0053] In the above embodiments, the standardized first model, the second model, and the corresponding parameters can be obtained, so that the third model and / or the fourth model can be determined based on the first device measured data set and at least one of the first information, the first model, and the second model.

[0054] In some embodiments of the first aspect, based on the first device measured data set and at least one of the first information, the first model, and the second model, determining the third model and / or the fourth model comprises: determining the normalized first model as the third model; and the method further comprises: sending the first device measured data set to the second device, the first device measured data set being used by the second device to obtain the fourth model through the first training type Type1.

[0055] In the above embodiments, the third model and / or the fourth model can be determined according to the first device measured data set, so that the model is trained by using the actually measured data, and the inference performance of the model is improved.

[0056] In some embodiments of the first aspect, based on the first device measured data set and at least one of the first information, the first model, and the second model, determining the third model and / or the fourth model comprises: determining the normalized first model as the third model; and the method further comprises: sending the first device measured data set to the second device, the first device measured data set being used by the second device to obtain the fourth model through the second training type Type2.

[0057] In the above embodiments, the third model and / or the fourth model can be determined according to the first device measured data set, so that the model is trained by using the actually measured data, and the inference performance of the model is improved.

[0058] In some embodiments of the first aspect, based on the first device measured data set and at least one of the first information, the first model, and the second model, determining the third model and / or the fourth model comprises: receiving the fourth model sent by the second device, the fourth model being obtained by the second device through the first training type Type1 based on a data set collected by the second device and the first model; and obtaining the third model through the second training type Type2 based on the first device measured data set, the first model, and the fourth model.

[0059] In the above embodiments, the third model and / or the fourth model can be determined according to the first device measured data set, so that the model is trained by using the actually measured data, and the inference performance of the model is improved.

[0060] In some embodiments of the first aspect, determining the third model and / or the fourth model based on the first information, the first model, and the second model and the data set measured by the first device comprises: determining the normalized first model as the third model; and the method further comprises: receiving a tenth model sent by the second device, the tenth model being obtained by the second device based on a data set collected by the second device and the first model through a first training type Type 1; and determining the fourth model through a second training type Type 2 based on the data set measured by the first device, the first model, and the tenth model.

[0061] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so that the model is trained using actually measured data, and the inference performance of the model is improved.

[0062] In some embodiments of the first aspect, determining the third model and / or the fourth model based on the first information, the first model, and the second model and the data set measured by the first device comprises: sending the data set measured by the first device to the second device, the data set measured by the first device being used by the second device to train the third model based on the first model and the fourth model, the fourth model being obtained by the second device based on a data set collected by the second device and the first model through a first training type Type 1; and receiving the third model sent by the second device.

[0063] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so that the model is trained using actually measured data, and the inference performance of the model is improved.

[0064] In some embodiments of the first aspect, determining the third model and / or the fourth model based on the first information, the first model, and the second model and the data set measured by the first device comprises: training the third model through a first training type Type 1 based on the data set measured by the first device and the second model, wherein the second model is used as the fourth model.

[0065] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so that the model is trained using actually measured data, and the inference performance of the model is improved.

[0066] In some embodiments of the first aspect, based on the first device measured data set and at least one of the first information, the first model, and the second model, determining the third model and / or the fourth model comprises: sending the first device measured data set to the second device, the first device measured data set being used by the second device to train the third model based on the second model through a first training type Type 1; and receiving the third model sent by the second device, wherein the second model is the fourth model.

[0067] In the above embodiments, the third model and / or the fourth model can be determined according to the first device measured data set, so that the model is trained using actually measured data, and the inference performance of the model is improved.

[0068] In some embodiments of the first aspect, based on the first device measured data set and at least one of the first information, the first model, and the second model, determining the third model and / or the fourth model comprises: training the third model based on the first device measured data set and the second model through a second training type Type 2, wherein the second model is the fourth model.

[0069] In the above embodiments, the third model and / or the fourth model can be determined according to the first device measured data set, so that the model is trained using actually measured data, and the inference performance of the model is improved.

[0070] In the second aspect, the embodiments of the present disclosure provide a model training method, the method being performed by a second device, and the method comprising: sending, to a first device, at least one of a first model, a second model, and first information, the first information being used to train the first model and / or the second model; determining a third model and / or a fourth model based on a first device measured data set and at least one of the first information, the first model, and the second model, the third model being the same as the first model or the third model being obtained by retraining the first model, and the fourth model being the same as the second model or the fourth model being obtained by retraining the second model, the third model being used by the first device to perform channel state information (CSI) compression, and the fourth model being used by the second device to perform CSI recovery.

[0071] In the above embodiments, the third model and / or the fourth model can be determined according to the first device measured data set, so that the model is trained using actually measured data, and the inference performance of the model is improved; and the problem of the inference performance of the model being reduced due to the inconsistency of the data distribution on the first device and the second device can be avoided.

[0072] In some embodiments of the second aspect, the method further comprises: training the first model and / or the second model based on the training data set, wherein the first information comprises model parameters of the first model and / or the second model and the training data set.

[0073] In the above embodiments, the second device can train the first model and the second model, so as to determine the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model and the second model, so as to train the model by using the actually measured data and improve the inference performance of the model.

[0074] In some embodiments of the second aspect, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model and the second model comprises: taking the second model as the fourth model; and wherein the third model is a fifth model trained by the first device based on the first model and model parameters of the first model using the training data set, and the third model is obtained based on the data set measured by the first device and the fifth model.

[0075] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so as to train the model by using the actually measured data and improve the inference performance of the model.

[0076] In some embodiments of the second aspect, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model and the second model comprises: taking the second model as the fourth model; and wherein the third model is obtained by the first device based on the data set measured by the first device, the training data set, the first model and model parameters of the first model through a second training type Type2.

[0077] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so as to train the model by using the actually measured data and improve the inference performance of the model.

[0078] In some embodiments of the second aspect, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model and the second model comprises: taking the second model as the fourth model; and wherein the method further comprises: receiving the data set measured by the first device sent by the first device; training the third model based on the data set measured by the first device; and sending the third model and / or model parameters of the third model to the first device.

[0079] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so that the model is trained by using actually measured data, and the inference performance of the model is improved.

[0080] In combination with some embodiments of the second aspect, in some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model comprises: determining the second model as the fourth model; and wherein the third model is trained by the first device using the training data set and / or the data set measured by the first device, the second model, and model parameters of the second model.

[0081] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so that the model is trained by using actually measured data, and the inference performance of the model is improved.

[0082] In combination with some embodiments of the second aspect, in some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model comprises: determining the second model as the fourth model; receiving the data set measured by the first device sent by the first device; training the model based on the data set measured by the first device to obtain the second model; and sending the second model and / or model parameters of the second model to the first device, wherein the third model is trained by the first device based on the second model and / or the model parameters of the second model.

[0083] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so that the model is trained by using actually measured data, and the inference performance of the model is improved.

[0084] In combination with some embodiments of the second aspect, in some embodiments, the method further comprises: training the first model and / or the second model based on the training data set; or receiving at least one of the first model, the second model trained by the first device, and the first information, wherein the first information comprises at least one of target CSI, feedback CSI, and recovered CSI, the target CSI is an input of training the first model, the feedback CSI is an output of training the first model and an input of training the second model, and the recovered CSI is an output of training the second model.

[0085] In the above embodiments, the second device can obtain the first model and the second model, so as to determine the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model, train the model by using actually measured data, and improve the inference performance of the model.

[0086] In some embodiments of the second aspect, determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model, and the second model comprises: taking the second model as the fourth model; wherein the third model is a sixth model trained by the first device based on the target CSI and the feedback CSI, and the sixth model is trained by the first device based on the first device measured data set and the second model by using a second training type Type2.

[0087] In the above embodiments, the third model and / or the fourth model can be determined according to the first device measured data set, so that the model is trained by using the actually measured data, and the inference performance of the model is improved.

[0088] In some embodiments of the second aspect, determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model, and the second model comprises: taking the second model as the fourth model; receiving the first device measured data set sent by the first device; training a seventh model based on the first device measured data set; and sending the feedback CSI and / or the target CSI used for training the seventh model to the first device, wherein the third model is trained by the first device based on the feedback CSI and / or the target CSI used for training the seventh model.

[0089] In the above embodiments, the third model and / or the fourth model can be determined according to the first device measured data set, so that the model is trained by using the actually measured data, and the inference performance of the model is improved.

[0090] In some embodiments of the second aspect, determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model, and the second model comprises: taking the second model as the fourth model; wherein the third model is an eighth model trained by the first device based on the feedback CSI and the recovered CSI, a ninth model trained by the first device based on the first device measured data set and the eighth model, and a model trained by the first device based on the first device measured data set, the ninth model, and the second model by using a second training type Type2.

[0091] In the above embodiments, the third model and / or the fourth model can be determined according to the first device measured data set, so that the model is trained by using the actually measured data, and the inference performance of the model is improved.

[0092] In some embodiments of the second aspect, determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model, and the second model comprises: taking the second model as the fourth model; and receiving the first device measured data set sent by the first device; and training the first model and / or the second model based on the first device measured data set; and sending the second model and / or the model parameters of the second model to the first device, wherein the third model is trained by the first device based on the first device measured data set, the second model, and / or the model parameters of the second model.

[0093] In the above embodiments, the third model and / or the fourth model can be determined based on the first device measured data set, so that the model is trained using actually measured data, and the inference performance of the model is improved.

[0094] In some embodiments of the second aspect, determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model, and the second model comprises: taking the second model as the fourth model; and receiving the first device measured data set sent by the first device; and training the first model and / or the second model based on the first device measured data set; and sending the second model and / or the model parameters of the second model to the first device, wherein the third model is trained by the first device based on the first device measured data set, the second model, and / or the model parameters of the second model.

[0095] In the above embodiments, the third model and / or the fourth model can be determined based on the first device measured data set, so that the model is trained using actually measured data, and the inference performance of the model is improved.

[0096] In some embodiments of the second aspect, the method further comprises: obtaining at least one of the standardized first model, the standardized second model, and the first information, wherein the first information comprises the model parameters of the first model and / or the second model, and the standardized training data set.

[0097] In the above embodiments, the standardized first model and the second model and the corresponding parameters can be obtained, so that the third model and / or the fourth model can be determined based on the first device measured data set and at least one of the first information, the first model, and the second model.

[0098] In some embodiments of the second aspect, determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model, and the second model comprises: receiving the first device measured data set sent by the first device; and training the fourth model based on the first device measured data set by the first training type Type 1, wherein the standardized first model is taken as the third model.

[0099] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so that the model is trained by using actually measured data, and the inference performance of the model is improved.

[0100] In combination with some embodiments of the second aspect, in some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model comprises: receiving the data set measured by the first device sent by the first device; obtaining the fourth model by Type2 training based on the data set measured by the first device, wherein the standardized first model is the third model.

[0101] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so that the model is trained by using actually measured data, and the inference performance of the model is improved.

[0102] In combination with some embodiments of the second aspect, in some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model comprises: obtaining the fourth model by Type1 training based on the data set collected by the second device and the first model; and sending the fourth model to the first device, wherein the third model is obtained by Type2 training based on the data set measured by the first device, the first model, and the fourth model.

[0103] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so that the model is trained by using actually measured data, and the inference performance of the model is improved.

[0104] In combination with some embodiments of the second aspect, in some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model comprises: obtaining the tenth model by Type1 training based on the data set collected by the second device and the first model; sending the tenth model to the first device, wherein the fourth model is obtained by Type2 training based on the data set measured by the first device, the first model, and the tenth model, and the standardized first model is the third model; and receiving the fourth model sent by the first device.

[0105] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so that the model is trained by using actually measured data, and the inference performance of the model is improved.

[0106] In some embodiments of the second aspect, determining the third model and / or the fourth model based on the first device measured dataset and at least one of the first information, the first model, and the second model comprises: receiving the first device measured dataset sent by the first device; training the third model based on the first device measured dataset and the second model by the first training type Type1; and sending the third model to the first device.

[0107] In the above embodiments, the third model and / or the fourth model can be determined based on the first device measured dataset, so that the model is trained by using actually measured data, and the inference performance of the model is improved.

[0108] In some embodiments of the second aspect, determining the third model and / or the fourth model based on the first device measured dataset and at least one of the first information, the first model, and the second model comprises: taking the second model as the fourth model; and the third model is trained by the first training type Type1 based on the first device measured dataset and the second model.

[0109] In the above embodiments, the third model and / or the fourth model can be determined based on the first device measured dataset, so that the model is trained by using actually measured data, and the inference performance of the model is improved.

[0110] In some embodiments of the second aspect, determining the third model and / or the fourth model based on the first device measured dataset and at least one of the first information, the first model, and the second model comprises: receiving the first device measured dataset sent by the first device; training the third model based on the first device measured dataset and the second model by the first training type Type1; and sending the third model to the first device.

[0111] In the above embodiments, the third model and / or the fourth model can be determined based on the first device measured dataset, so that the model is trained by using actually measured data, and the inference performance of the model is improved.

[0112] In some embodiments of the second aspect, determining the third model and / or the fourth model based on the first device measured dataset and at least one of the first information, the first model, and the second model comprises: taking the second model as the fourth model; and the third model is trained by the first training type Type1 based on the first device measured dataset and the second model.

[0113] In the above embodiments, the third model and / or the fourth model can be determined according to the data set measured by the first device, so that the model is trained by using actually measured data, and the inference performance of the model is improved.

[0114] In a third aspect, the embodiments of the present disclosure provide a first device, including: a transceiver module, configured to obtain at least one of a first model, a second model, and first information, the first information being used for training the first model and / or the second model; and a processing module, configured to determine a third model and / or a fourth model based on a data set measured by the first device and at least one of the first information, the first model, and the second model, the third model being the same as the first model or being obtained by retraining the first model, the third model being used for channel state information (CSI) compression by the first device, and the fourth model being used for CSI recovery by a second device.

[0115] In a fourth aspect, the embodiments of the present disclosure provide a second device, including: a transceiver module, configured to send at least one of a first model, a second model, and first information to a first device, the first information being used for training the first model and / or the second model; and a processing module, configured to determine a third model and / or a fourth model based on a data set measured by the first device and at least one of the first information, the first model, and the second model, the third model being the same as the first model or being obtained by retraining the first model, the third model being used for channel state information (CSI) compression by the first device, and the fourth model being used for CSI recovery by the second device.

[0116] In a fifth aspect, the embodiments of the present disclosure provide a communication device, including: one or more processors; and wherein the one or more processors are configured to invoke instructions to cause the communication device to perform the method of any one of the first aspect or the method of any one of the second aspect.

[0117] In a sixth aspect, the embodiments of the present disclosure provide a communication system, including: a first device and a second device; wherein the first device is configured to perform the method described in the first aspect and the optional implementation of the first aspect, and the second device is configured to perform the method described in the second aspect and the optional implementation of the second aspect.

[0118] In a seventh aspect, the embodiments of the present disclosure provide a storage medium, and the computer storage medium stores computer executable instructions; and the computer executable instructions are executed by a processor to perform the method described in the first aspect, the optional implementation of the first aspect, the second aspect, and the optional implementation of the second aspect.

[0119] Eighthly, embodiments of this disclosure provide a computer program product, characterized in that it includes a computer program that, when executed by a processor, implements the method as described in either the first or second aspect.

[0120] It is understood that the first device, the second device, the communication device, the communication system, and the storage medium described above are all used to execute the methods proposed in the embodiments of this disclosure. Therefore, the beneficial effects that can be achieved can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0121] This disclosure provides a method for determining time units, as well as communication devices, communication systems, and storage media. In some embodiments, the terms "time unit determination method" and "information processing method," "communication method," etc., can be used interchangeably; the terms "terminal," "network device," and "communication apparatus," etc., can be used interchangeably; and the terms "information processing system" and "communication system," etc., can be used interchangeably.

[0122] This disclosure is not exhaustive, but merely illustrative of some embodiments, and is not intended to limit the scope of protection of this disclosure. Unless otherwise specified, each step in a particular embodiment can be implemented as an independent embodiment, and the steps can be arbitrarily combined. For example, a solution after removing some steps in a particular embodiment can also be implemented as an independent embodiment, and the order of the steps in a particular embodiment can be arbitrarily interchanged. Furthermore, the optional implementation methods in a particular embodiment can be arbitrarily combined; moreover, the embodiments can be arbitrarily combined, for example, some or all steps of different embodiments can be arbitrarily combined, and a particular embodiment can be arbitrarily combined with the optional implementation methods of other embodiments.

[0123] In each of the disclosed embodiments, unless otherwise specified or in case of logical conflict, the terminology and / or descriptions of the embodiments are consistent and can be referenced by each other. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0124] The terminology used in the embodiments of this disclosure is for the purpose of describing particular embodiments only and is not intended to limit the scope of this disclosure.

[0125] In this embodiment of the disclosure, unless otherwise stated, elements expressed in the singular form, such as "a," "an," "the," "the," "the," "the," "the," "the," "this," etc., can mean "one and only one," or "one or more," "at least one," etc. For example, when using articles such as "a," "an," "the," etc. in translation, the noun following the article can be understood as either a singular expression or a plural expression.

[0126] In the embodiments of the present disclosure, "multiple" refers to two or more.

[0127] In some embodiments, the terms "at least one of", "at least one of", "at least one of", "one or more", "a plurality of", "multiple", and the like can be replaced with each other.

[0128] In the embodiments of the present disclosure, the description modes such as "at least one of A, B, C, and the like", "A and / or B and / or C, and the like" include any one of A, B, C, and the like existing alone, and also include any combination of any multiple of A, B, C, and the like, each of which can exist alone; for example, "at least one of A, B, C" includes the cases of A alone, B alone, C alone, A and B in combination, A and C in combination, B and C in combination, A and B and C in combination; for example, A and / or B includes the cases of A alone, B alone, and the combination of A and B.

[0129] In some embodiments, the description modes such as "A in one case and B in another case", "in response to one case A and in response to another case B", and the like can include the following technical solutions according to the cases: A is executed regardless of B, that is, A in some embodiments; B is executed regardless of A, that is, B in some embodiments; A and B are selectively executed, that is, A and B are selected from A and B to be executed in some embodiments; A and B are both executed, that is, A and B in some embodiments. When there are more branches of A, B, C, and the like, it is similar to the above.

[0130] The prefix words of "first", "second" and the like in the embodiments of the present disclosure are merely used to distinguish different description objects, and do not constitute limitation on the position, order, priority, quantity or content of the description objects. The description objects are described in the claims or embodiments, and should not be construed as redundant limitation because of the use of the prefix words. For example, the description object is "field", and the ordinal words before "field" in "first field" and "second field" do not limit the position or order between "fields", and "first" and "second" do not limit whether the "fields" modified thereby are in the same message or not, nor limit the order of "first field" and "second field". For another example, the description object is "level", and the ordinal words before "level" in "first level" and "second level" do not limit the priority between "levels". For another example, the quantity of the description object is not limited by the ordinal words, and can be one or more. For example, "first device", wherein the quantity of "device" can be one or more. In addition, the objects modified by different prefix words can be the same or different, for example, the description object is "device", and "first device" and "second device" can be the same device or different devices, and the types thereof can be the same or different. For another example, the description object is "information", and "first information" and "second information" can be the same information or different information, and the contents thereof can be the same or different.

[0131] In some embodiments, "including A", "containing A", "for indicating A", "carrying A" can be interpreted as directly carrying A, or indirectly indicating A.

[0132] In some embodiments, the terms of "in response to", "in response to determining", "in the case of", "when", "when", "if", "if" and the like can be replaced with each other.

[0133] In some embodiments, the terms of "greater than", "greater than or equal to", "not less than", "more than", "more than or equal to", "not less than", "higher than", "higher than or equal to", "not lower than", "above" and the like can be replaced with each other, and the terms of "less than", "less than or equal to", "not greater than", "less than", "less than or equal to", "not more than", "lower than", "lower than or equal to", "not higher than", "below" and the like can be replaced with each other.

[0134] In some embodiments, an apparatus or the like can be interpreted as an entity, and can also be interpreted as virtual, and the name thereof is not limited to the name described in the embodiments, and the terms "apparatus", "equipment", "device", "circuit", "network element", "node", "function", "unit", "section", "system", "network", "chip", "chip system", "entity", "subject" and the like can be replaced with each other.

[0135] In some embodiments, the terms "access network device (AN device)", "radio access network device (RAN device)", "base station (BS)", "radio base station", "fixed station", "node", "access point", "transmission point (TP)", "reception point (RP)", "transmission / reception point (TRP)", "panel", "antenna panel", "antenna array", "cell", "macro cell", "small cell", "femto cell", "pico cell", "sector", "cell group", "carrier", "component carrier", "bandwidth part (BWP)" and the like can be replaced with each other.

[0136] In some embodiments, the terms "terminal," "terminal device," "user equipment (UE)," "user terminal," "mobile station (MS)," "mobile terminal (MT)," "subscriber station," "mobile unit," "subscriber unit," "wireless unit," "remote unit," "mobile device," "wireless device," "wireless communication device," "remote device," "mobile subscriber station," "access terminal," "mobile terminal," "wireless terminal," "remote terminal," "handset," "user agent," "mobile client," "client," and so on can be replaced with each other.

[0137] In some embodiments, an access network device, a core network device, or a network device can be replaced with a terminal. For example, for a structure in which communication between an access network device, a core network device, or a network device and a terminal is replaced with communication between a plurality of terminals (for example, also referred to as device-to-device (D2D), vehicle-to-everything (V2X), and so on), embodiments of the present disclosure can also be applied. In this case, a structure in which a terminal has all or part of the functions of an access network device can also be provided. Furthermore, the language of "uplink," "downlink," and so on can also be replaced with language corresponding to communication between terminals (for example, "side"). For example, an uplink channel, a downlink channel, and so on can be replaced with a side channel, and an uplink, a downlink, and so on can be replaced with a side link.

[0138] In some embodiments, a terminal can be replaced with an access network device, a core network device, or a network device. In this case, a structure in which an access network device, a core network device, or a network device has all or part of the functions of a terminal can also be provided.

[0139] In some embodiments, the names of information and the like are not limited to the names described in the embodiments, and terms such as "information", "message", "signal", "signaling", "report", "configuration", "indication", "instruction", "command", "channel", "parameter", "field", "symbol", "symbol", "codebook", "codeword", "codepoint", "bit", "data", "program", "chip", and the like can be replaced with each other.

[0140] In some embodiments, terms such as "uplink", "uplink", "physical uplink", and the like can be replaced with each other, terms such as "downlink", "downlink", "physical downlink", and the like can be replaced with each other, terms such as "side", "sidelink", "sidelink communication", "sidelink communication", "direct connection", "direct connection link", "direct connection", "direct connection link communication", and the like can be replaced with each other.

[0141] In some embodiments, terms such as "downlink control information (DCI)", "downlink (DL) assignment", "DL DCI", "uplink (UL) grant", "UL DCI", and the like can be replaced with each other.

[0142] In some embodiments, terms such as "physical downlink shared channel (PDSCH)", "DL data", and the like can be replaced with each other, and terms such as "physical uplink shared channel (PUSCH)", "UL data", and the like can be replaced with each other.

[0143] In some embodiments, terms such as "radio", "wireless", "radio access network (RAN)", "access network (AN)", "RAN-based", and the like can be replaced with each other.

[0144] In some embodiments, the terms “synchronization signal (SS)”, “synchronization signal block (SSB)”, “reference signal (RS)”, “pilot”, “pilot signal” and the like can be replaced with each other.

[0145] In some embodiments, the terms “moment”, “time point”, “time”, “time position” and the like can be replaced with each other, and the terms “time length”, “time period”, “time window”, “window”, “time” and the like can be replaced with each other.

[0146] In some embodiments, “acquire”, “obtain”, “get”, “receive”, “transmit”, “bidirectional transmission”, “send and / or receive” and the like can be replaced with each other, which can be interpreted as receiving from other subjects, obtaining from protocols, obtaining by oneself, autonomously implementing and the like.

[0147] In some embodiments, the terms “send”, “transmit”, “report”, “issue”, “transmit”, “bidirectional transmission”, “send and / or receive” and the like can be replaced with each other.

[0148] In some embodiments, “predetermined” and “preset” can be interpreted as being previously specified in protocols and the like, or can be interpreted as being previously set by devices and the like.

[0149] In some embodiments, determining can be interpreted as judging, deciding, judging, calculating, computing, processing, deriving, investigating, searching, looking up, retrieving, inquiring, ascertaining, receiving, transmitting, inputting, outputting, accessing, resolving, selecting, choosing, establishing, comparing, "assuming", "expecting", "considering", broadcasting, notifying, communicating, forwarding, configuring, reconfiguring, allocating, mapping, assigning, etc., but is not limited thereto.

[0150] In some embodiments, deciding or judging can be performed by a value represented by 1 bit (0 or 1), or by a true or false value (Boolean value) represented by true or false, or by comparison of numerical values (for example, comparison with a predetermined value), but is not limited thereto.

[0151] In some embodiments, "network" can be interpreted as devices (for example, access network devices, core network devices, etc.) contained in the network.

[0152] In some embodiments, "not expecting to receive" can be interpreted as not receiving on time domain resources and / or frequency domain resources, or can be interpreted as, after receiving data, etc., not performing subsequent processing on the data, etc.; "not expecting to send" can be interpreted as not sending, or can be interpreted as sending but not expecting the receiving party to respond to the content of the sending.

[0153] In some embodiments, obtaining data, information, etc. can comply with the laws and regulations of the country where the location is located.

[0154] In some embodiments, data, information, etc. can be acquired after obtaining user consent. To solve the above problems, the present disclosure proposes an information indication method and a communication device, a communication system, and a storage medium.

[0155] FIG. 1 is a schematic diagram of an architecture of a communication system according to an embodiment of the present disclosure. As shown in FIG. 1, the communication system 100 can include a first device 101 and a second device 102.

[0156] In some embodiments, the first device can be a terminal or a network device, for example.

[0157] In some embodiments, the second device can be a terminal or a network device, for example.

[0158] In some embodiments, the terminal includes at least one of a mobile phone, a wearable device, an Internet of Things device, a car with communication function, a smart car, a Pad, a computer with wireless transceiver function, a virtual reality (VR) terminal device, an augmented reality (AR) terminal device, a wireless terminal device in industrial control, a wireless terminal device in self-driving, a wireless terminal device in remote medical surgery, a wireless terminal device in smart grid, a wireless terminal device in transportation safety, a wireless terminal device in smart city, a wireless terminal device in smart home, etc., but is not limited thereto.

[0159] In some embodiments, the access network device is, for example, a node or device that accesses a terminal to a wireless network, and the access network device can include at least one of an evolved NodeB (eNB) in a 5G communication system, a next generation eNB (ng-eNB), a next generation NodeB (gNB), a node B (NB), a home node B (HNB), a home evolved node B (HeNB), a wireless backhaul device, a radio network controller (RNC), a base station controller (BSC), a base transceiver station (BTS), a base band unit (BBU), a mobile switching center, a base station in a 6th generation mobile communication system (6th generation mobile networks or 6th generation wireless systems, 6G), an open base station (Open RAN), a cloud base station (Cloud RAN), a base station in other communication systems, an access node in a wireless fidelity (WiFi) system, but is not limited thereto.

[0160] In some embodiments, the technical solutions of the present disclosure can be applied to an Open RAN architecture, at this time, the interfaces between or within the access network devices involved in the embodiments of the present disclosure can become internal interfaces of the Open RAN, and the processes and information interactions between these internal interfaces can be realized through software or programs.

[0161] In some embodiments, the access network device can be composed of a central unit (CU) and a distributed unit (DU), wherein the CU can also be referred to as a control unit (control unit). The CU-DU structure can split the protocol layers of the access network device, and the functions of part of the protocol layers are controlled by the CU, and the functions of the remaining part or all of the protocol layers are distributed in the DU and controlled by the CU, but are not limited thereto.

[0162] In some embodiments, the core network device can be one device including one or more network elements, or can be multiple devices or device groups each including all or part of the one or more network elements. The network element can be virtual or physical. The core network includes, for example, at least one of an evolved packet core (EPC), a 5G core network (5GCN), and a next generation core (NGC).

[0163] In some embodiments, the one or more network elements described above can include, for example, an AMF, a UPF, an MME, and the like, and can also include other network elements such as a policy control function (PCF), an application function (AF), a network application function (NAF), an authentication and key management for applications anchor function (AAnF), a bootstrapping server function (BSF), a session management function (SMF), and the like.

[0164] It can be understood that the communication system described in the embodiments of the present disclosure is for more clearly illustrating the technical solutions of the embodiments of the present disclosure, and does not constitute a limitation on the technical solutions proposed in the embodiments of the present disclosure. It can be known by those skilled in the art that, as the system architecture evolves and new business scenarios appear, the technical solutions proposed in the embodiments of the present disclosure are also applicable to similar technical problems.

[0165] The following embodiments of the present disclosure can be applied to the communication system 100 shown in FIG. 1 or part of the subject, but are not limited thereto. The subjects shown in FIG. 1 are exemplary, and the communication system can include all or part of the subjects in FIG. 1, or other subjects other than those in FIG. 1. The number and form of each subject is arbitrary, and the connection relationship between the subjects is exemplary. The subjects can not be connected or can be connected, and the connection can be in any manner, can be direct connection or indirect connection, and can be wired connection or wireless connection.

[0166] Embodiments of the present disclosure can be applied to Long Term Evolution (LTE), LTE-Advanced (LTE-A), LTE-Beyond (LTE-B), SUPER 3G, IMT-Advanced, 4th generation mobile communication system (4G), 5th generation mobile communication system (5G), 5G new radio (NR), Future Radio Access (FRA), New-Radio Access Technology (RAT), New Radio (NR), New radio access (NX), Future generation radio access (FX), Global System for Mobile communications (GSM (registered trademark)), CDMA2000, Ultra Mobile Broadband (UMB), IEEE 802.11 (Wi-Fi (registered trademark)), IEEE 802.16 (WiMAX (registered trademark)), IEEE 802.20, Ultra-WideBand (UWB), Bluetooth (Bluetooth (registered trademark)), Public Land Mobile Network (PLMN) network, Device-to-Device (D2D) system, Machine to Machine (M2M) system, Internet of Things (IoT) system, Vehicle-to-Everything (V2X), system using other communication methods, next-generation system expanded based thereon, and the like. Further, a plurality of systems can be applied in combination (for example, combination of LTE or LTE-A and 5G, and the like).

[0167] The current research and simulation results show that by using artificial intelligence (AI) technology, the feedback overhead of the terminal can be reduced or the channel status information (CSI) feedback accuracy can be improved, and in the 3GPP standardization research, a bilateral AI / Machine Learning (ML) model based on the terminal-side CSI generation part model and the network-side CSI recovery part model is used to realize the compression feedback and recovery of CSI respectively. In one implementation method of the bilateral AI / ML model for realizing CSI compression feedback and recovery, the UE side compresses the downlink channel information H through the CSI generation part model (defined as an encoder) and sends it to the gNB through quantization as a binary bit stream, and the gNB side recovers H' similar to the original downlink information through the CSI recovery part model (defined as a decoder).

[0168] Among them, the CSI generation part model and the CSI recovery part model need to be trained through the collected data set. The types of training the CSI generation part model and the CSI recovery part model include the following three types:

[0169] Training type 1: training of the model is completed on one side (such as the terminal side or the network side), and then the trained part of the model is sent to the other side;

[0170] Training type 2: the CSI generation part model and the CSI recovery part model are respectively trained by joint training on the terminal side and the network side. Alternatively, after one end of the terminal side or the network side trains the part of the model, the other part of the bilateral model is trained on the other end, and the model parameters of the first trained part are not updated;

[0171] Training type 3: first train the model on one side, and then send the training data or other auxiliary information to the other side to complete the training of the other part of the model. Specifically, the training Type 3 can be divided into:

[0172] Network-side first training (NW-first training): the network side first trains the CSI generation part model and the CSI recovery part model, and then sends the data set and / or other auxiliary information of the trained CSI generation part model to the terminal side.

[0173] Terminal-side first training (UE-first training): the terminal side first trains the CSI generation part model and the CSI recovery part model, and then sends the data set and / or other auxiliary information of the trained CSI recovery part model to the network side

[0174] To reduce or mitigate the complexity of the training of the bilateral model, the following options for collaborative training between device vendors have been proposed:

[0175] Option 1: Standardize the model structure and parameters; Option 2: Standardize the dataset; Option 3: Standardize the model structure, while the parameters of the model are transferred between the network side and the terminal side; Option 4: Standardize the format of the data, while the data is transferred between the network side and the terminal side; Option 5: Standardize the model format, while the reference model is transferred between the network side and the terminal side.

[0176] Among them, Option 1 can be divided into: Option 1-1: Standardize the encoder; Option 1-2: Standardize the decoder; Option 1-3: Standardize the encoder and decoder.

[0177] According to the model parameters or model post-terminal execution behavior transmitted to the terminal side, Option 3 and Option 5 can be further divided into:

[0178] For Option 3:

[0179] Option 3a: Receive model parameters and retrain the model, redevelop a different model or test.

[0180] Option 3a-1: Transfer the parameters of the encoder; Option 3a-2: Transfer the parameters of the decoder; Option 3a-3: Transfer the parameters of the encoder and decoder.

[0181] Option 3b: The received model parameters are directly used for model inference.

[0182] For Option 5:

[0183] Option 5a: Train the received model, redevelop a different model or test.

[0184] Option 5a-1: Transfer the encoder model; Option 5a-2: Transfer the decoder model; Option 5a-3: Transfer the encoder and decoder model.

[0185] According to the different contents of the transferred data, Option 4 can be further divided into:

[0186] For Option 4, the dataset is transferred from the network side to the terminal side.

[0187] Option 4-1: The dataset is the target CSI (target CSI) and the feedback CSI (CSI feedback); Option 4-2: The dataset is the CSI feedback and the reconstructed target CSI (reconstructed target CSI); Option 4-3: The dataset is the target CSI, the CSI feedback and the reconstructed target CSI.

[0188] Based on the above trained bilateral model, it is possible that the data distribution of the terminal side and the data distribution of the network side are inconsistent, resulting in performance degradation when the trained model is directly used for inference. How to solve the performance degradation caused by data mismatch is a problem to be solved.

[0189] To solve the above problem, the present disclosure proposes a model training method, which can propose a bilateral model training method to solve the performance degradation problem caused by the mismatch of terminal side data and network side data for different device cooperation training options. The specific content of the method is as shown below.

[0190] FIG. 2 is an interaction schematic diagram of a time unit determination method according to an embodiment of the present disclosure. As shown in FIG. 2, the present disclosure relates to a model training method for a communication system 100, which can include a first device 101 and a second device 102. The method includes the following steps.

[0191] In step 2101, the second device sends at least one of the first model, the second model, and the first information to the first device.

[0192] In some embodiments, the first device can obtain at least one of the first model, the second model, and the first information, wherein the first information is used to train the first model and / or the second model. For example, the first device can be a terminal, and the second device can be a network device. In this case, the first model can be a reference model used by the terminal for encoding, and the second model can be a reference model used by the network device for decoding.

[0193] In some embodiments, the first device obtaining at least one of the first model, the second model, and the first information includes: obtaining at least one of the first model, the second model, and the first information from the second device, the first information including model parameters of the first model and / or the second model and a training data set, and the first model and / or the second model being trained by the second device based on the training data set.

[0194] That is, the second device can train the first model and / or the second model using the training data set and send at least one of the first model, the second model, and the first information to the first device. In this case, the first information can include the model parameters of the first model and / or the second model and the training data set. For example, when the first device is a terminal and the second device is a network device, the network device can train the first model and / or the second model based on the training data set, and send the first model and / or the second model, the model parameters corresponding to the model, and the training data set used by the network device to train the first model and the second model to the first device, so that the first device can perform model training according to the above model, parameters, or training data set.

[0195] In some embodiments, the first device obtaining the at least one of the first model, the second model, and the first information comprises: obtaining the at least one of the first model, the second model, and the first information trained by the second device or the first device, the first information comprising at least one of target channel status information (CSI), feedback CSI, and recovered CSI, the target CSI being an input of training the first model, the feedback CSI being an output of training the first model and an input of training the second model, and the recovered CSI being an output of training the second model.

[0196] That is, the first model and the second model can be trained at the first device or the second device, when the second device trains the first model and the second model, the first device can receive the trained first model and the trained second model from the second device; when the first device trains the first model and the second model, the first device can directly use the trained first model and the trained second model. Optionally, the first information can comprise at least one of the target CSI, the feedback CSI, and the recovered CSI, wherein the target CSI is an input of training the first model, the feedback CSI is an output of training the first model and an input of training the second model, and the recovered CSI is an output of training the second model, that is, the first information can be training data used for training the first model and the second model, and the first device can use the training data for training the first model and the second model to perform model training.

[0197] In some embodiments, the first device obtaining the at least one of the first model, the second model, and the first information comprises: obtaining the at least one of the first model, the second model, and the first information, the first information comprising at least one of model parameters of the first model and / or the second model, and a standardized training data set.

[0198] In other words, the first model and the second model can be standardized models, and at this time, the first information can be at least one of parameters of the standardized models and a training data set of the standardized models, that is, the first device can use the standardized models, the parameters of the standardized models, and the training data set to perform model training.

[0199] Step 2102: The first device determines a third model and / or a fourth model based on a data set measured by the first device and at least one of the first information, the first model, and the second model.

[0200] In some embodiments, the third model is the same as the first model or the third model is obtained by retraining the first model, the fourth model is the same as the second model or the fourth model is obtained by retraining the second model, the third model is used for channel status information (CSI) compression by the first device, and the fourth model is used for CSI recovery by the second device.

[0201] In other words, the first model can be retrained to obtain a third model, and the second model can be retrained to obtain a fourth model, or the second model can not be retrained, and the second model is directly determined as the fourth model, that is, the second model can be used for CSI compression. For example, the first device can be a terminal, and the second device can be a network device, that is, the terminal can use the third model for channel state information (CSI) compression and feed back the compressed CSI to the network device, and the network device can use the fourth model to recover the compressed CSI.

[0202] Example 1: The second device trains to obtain the first model and / or the second model, and the first device and the second device use the cooperative training option 3a or the cooperative training option 5a, that is, use the standardized model structure or the standardized model format for cooperative training. At this time, the second device sends at least one of the first model, the second model, the parameters of the first model, the parameters of the second model, and the training data set for training the first model and the second model to the first device, and the first device can determine the third model and / or the fourth model according to the above information sent by the second device.

[0203] When the second device can send the first model and the parameters of the first model to the first device, the optional example is as follows.

[0204] Example 1-1:

[0205] In some embodiments, determining the third model and / or the fourth model based on at least one of the first information, the first model, the second model, and the data set measured by the first device includes: training a fifth model based on the first model and the model parameters of the first model using the training data set; and training the third model based on the data set measured by the first device and the fifth model, wherein the second model is used as the fourth model.

[0206] In the above embodiments, the second device can train the first model and the second model, and send the first model, the model parameters of the first model, and the training data set for training the first model and the second model to the first device, so that the first device determines the third device. The first device can train the fifth model according to the first model and the parameters of the first model, where the first model can be represented as encoder#1, the second model can be represented as decoder#1, the third model can be represented as encoder#2, and the fifth model can be represented as decoder#2. The first device can perform model training according to the data set measured by the first device and the fifth model, to obtain the third model. At this time, the first device can use the third model to compress the CSI. At this time, the second model can be determined as the fourth model, and the second device can use the second model to restore the CSI. At this time, the fourth model can be represented as decoder#1.

[0207] The data set measured by the first device can be channel state information actually measured by the first device, that is, the first device can use the actually measured data to train the third model, so that the third model can be applicable to the first device, and the third model can obtain better performance when used by the first device for CSI compression.

[0208] In other words, when the first device is a terminal, the terminal side trains decoder#2 based on the received encoder#1 model and parameters. The network device sends the data set for training encoder#1 and decoder#1 to the terminal, so that the terminal can train decoder#2 based on the training data set. The terminal side trains encoder#2 based on the data set measured by the terminal side and decoder#2. For example, the model structures of encoder#1 and encoder#2 are the same, and at this time, encoder#2 and decoder#1 are used for inference in the inference stage.

[0209] Example 1-2:

[0210] In some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model includes: performing model training by a second training type Type2 based on the data set measured by the first device, the training data set, the first model, and the model parameters of the first model, to obtain the third model, where the second model is the fourth model.

[0211] In some embodiments, the first device can be a terminal, and the model training can be performed by the terminal using training type 2 to obtain a third model, wherein the training type 2 is to train a CSI generation part model and a CSI recovery part model respectively on the terminal side and the network side by joint training. Alternatively, after training a part model on one of the terminal side or the network side, the other part model of the bilateral model is trained on the other side, and the model parameters of the first training part are not updated.

[0212] That is, the second device can train the first model and the second model, and send the first model, the parameters of the first model, and the data set for training the first model and the second model to the first device, so as to facilitate the first device to determine the third model based on the above information. At this time, since the second model and the parameters of the second model are on the second device side, the first device can perform model training by the second training type Type 2 based on the data set measured by the first device, the training data set, the first model, and the model parameters of the first model, to obtain the third model. At this time, the first device can use the third model to compress the CSI, and it can be determined that the second model is the fourth model, that is, the second device can use the second model to recover the CSI.

[0213] For example, when the first device is a terminal and the second device is a network device, the terminal side trains an encoder#2 based on the data set measured by the terminal side and / or the training data set sent by the network device side, and the received encoder#1 model and parameters. The training data set sent is the data set for training the encoder#1 and the decoder#1 on the network device side, and at this time, the encoder#2 and the decoder#1 are used for inference in the inference stage.

[0214] Example 1-3:

[0215] In some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device, and at least one of the first information, the first model, and the second model includes: sending the data set measured by the first device to the second device, and the data set measured by the first device is used by the second device to train the third model; receiving the third model and / or the model parameters of the third model sent by the second device, wherein the second model is the fourth model.

[0216] In some embodiments, the first device can send the measured data set to the second device, and the second device can retrain the first model based on the data set measured by the first device to obtain the third model, and send the third model and / or the model parameters of the third model to the first device after the training is completed, so that the first device uses the third device to compress the CSI. At this time, it can be determined that the second model is the fourth model, that is, the second device uses the second model to recover the CSI.

[0217] For example, when the first device is a terminal and the second device is a network device, the terminal sends the measured data set to the network device, the network device re-trains the encoder#2 based on the data set measured by the terminal, and then sends the model / parameters of the encoder#2 to the terminal.

[0218] When the second device can send the second model and the parameters of the second model to the first device, an optional example is as follows.

[0219] Example 2-1:

[0220] In some embodiments, determining the third model and / or the fourth model based on at least one of the first information, the first model, the second model, and the data set measured by the first device comprises: training the third model using the training data set and / or the data set measured by the first device, the second model, and the model parameters of the second model, wherein the second model is the fourth model.

[0221] In some embodiments, the first device can receive the second model, the model parameters of the second model, and the training data set used by the second device to train the first model and the second model, and at this time, the first device can train the third model using the training data set and / or the data set measured by the first device, the second model, and the model parameters of the second model, and at this time, the second model can be determined as the fourth model, that is, the second device can use the second model to recover the CSI.

[0222] For example, when the first device is a terminal and the second device is a network device, the terminal side trains the encoder#2 based on the received model and parameters of the decoder#1 and the data set measured by the terminal side and / or the data set sent by the network device, and at this time, the model structures of the encoder#2 and the encoder#1 can be the same or different.

[0223] Example 2-2:

[0224] In some embodiments, determining the third model and / or the fourth model based on at least one of the first information, the first model, the second model, and the data set measured by the first device comprises: sending the data set measured by the first device to the second device, the data set measured by the first device being used by the second device to train the second model; receiving the second model and / or the model parameters of the second model sent by the second device; and training the third model based on the second model and / or the model parameters of the second model, wherein the second model is the fourth model.

[0225] In some embodiments, the first device can send the measured data set to the second device, the second device can train a second model according to the data set measured by the first device, and send the second model and parameters of the second model to the first device, and the first device can train a third model according to the received second model and / or model parameters of the second model. At this time, the second model can be determined as the fourth model, that is, the second device can use the second model to recover the CSI.

[0226] For example, when the first device is a terminal and the second device is a network device, the terminal can send the measured data set to the network device, the network device can retrain a decoder#1 based on the data set measured by the terminal, and then send the model and parameters of the decoder#1 to the terminal. The terminal can train a model according to the received model and parameters of the decoder#1 to obtain an encoder#2.

[0227] When the second device can send the first model, the second model, and the parameters of the first model and the second model to the first device, optional examples are as follows.

[0228] Example 3-1:

[0229] In some embodiments, the second device can send the first model, the second model, the parameters of the first model and the second model, and the training data set for training the first model and the second model to the first device. At this time, the first device can use at least one of the information sent by the second device to train a model to determine the third model and / or the fourth model.

[0230] For example, the first device can determine the third model and / or the fourth model using any of the methods in Examples 1-1 to 1-3 above, or can determine the third model and / or the fourth model using any of the methods in Examples 2-1 to 2-2 above.

[0231] Example 2: The first device and the second device use cooperative training option 4, that is, cooperative training is performed using a standardized data format. At this time, the second device sends a training data set to the first device, and the first device can determine the third model and / or the fourth model according to the data set sent by the second device.

[0232] When the data set is a target CSI (target CSI) and a feedback CSI (CSI feedback), optional examples are as follows.

[0233] Example 4-1:

[0234] In some embodiments, determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model, and the second model comprises: training a sixth model based on the target CSI and the feedback CSI; and training the sixth model based on the first device measured data set and the second model by employing a second training type Type2 to obtain the third model, wherein the second model is the fourth model.

[0235] In some embodiments, the first device can be a terminal, i.e., the terminal can receive the target CSI and the feedback CSI sent by the second device, train a sixth model based on the target CSI and the feedback CSI, and train the sixth model based on the first device measured data set and the second model by employing a second training type Type2 to obtain the third model, and it can be determined that the second model is the fourth model, i.e., the second device can use the second model to recover the CSI.

[0236] Example 4-2:

[0237] In some embodiments, determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model, and the second model comprises: sending the first device measured data set to the second device, the first device measured data set being used by the second device to train a seventh model; receiving the feedback CSI and / or the target CSI sent by the second device for training the seventh model; and training the third model based on the feedback CSI and / or the target CSI for training the seventh model, wherein the second model is the fourth model.

[0238] In some embodiments, the first device can be a terminal, and the second device can be a network device, i.e., the terminal can send the measured data set to the network device, the network device can use the measured data set to train a seventh model, the network device can send the feedback CSI and / or the target CSI for training the seventh model to the terminal, and the terminal can train a model based on the feedback CSI and / or the target CSI for training the seventh model to obtain the third model, at this time, it can be determined that the second model is the fourth model, i.e., the second device can use the second model to recover the CSI.

[0239] When the data set is the feedback CSI and the reconstructed target CSI, optional examples are as follows.

[0240] Example 5-1:

[0241] In some embodiments, determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the data set measured by the first device comprises: training an eighth model based on the feedback CSI and the reconstructed CSI; training a ninth model based on the data set measured by the first device and the eighth model; and training the third model based on the data set measured by the first device, the ninth model and the second model using a second training type Type2, wherein the second model is the fourth model.

[0242] In some embodiments, the eighth model can be retrained from the second model based on the feedback CSI and the reconstructed CSI, denoted as decoder#2, the ninth model can be retrained from the first model based on the data set measured by the first device and the eighth model, denoted as encoder#2, and the first device can train the third model based on the data set measured by the first device, the ninth model and the second model using the second training type Type2, wherein the third model can be denoted as encoder#3, and the second model can be determined as the fourth model, i.e., the second device can use the second model to reconstruct the CSI.

[0243] In some embodiments, the first device can be a terminal and the second device can be a network device, i.e., the terminal side trains the decoder#2 based on the received CSI feedback and the reconstructed target CSI. Then the terminal side trains the encoder#2 based on the data set measured by the UE and the decoder#2, and trains the encoder#3 based on the measured data set and the decoder#1 using the training type Type2.

[0244] Example 5-2:

[0245] In some embodiments, determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the data set measured by the first device comprises: training an eighth model based on the feedback CSI and the reconstructed CSI; training a ninth model based on the data set measured by the first device and the eighth model; and training the third model based on the data set measured by the first device, the ninth model and the second model using a second training type Type2, wherein the second model is the fourth model.

[0246] In some embodiments, the eighth model can be retrained based on the feedback CSI and the recovered CSI from the second model, denoted as decoder#2, the first device can use the first training type to train the third model based on the data set measured by the first device and the eighth model, where the third model can be denoted as encoder#3, and the first training type can be to complete the training of the model at one side (such as the terminal side or the network side), and then send the trained part of the model to the other side. At this time, the second model can be determined as the fourth model, that is, the second device can use the second model to recover the CSI.

[0247] In some embodiments, the first device can be a terminal, and the second device can be a network device, that is, the terminal side trains decoder#2 based on the received CSI feedback and the reconstructed target CSI. Then the terminal side trains encoder#3 based on the data set measured by the terminal and decoder#2 through the training type Type 1.

[0248] Example 5-3:

[0249] In some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model includes: sending the data set measured by the first device to the second device, the data set measured by the first device being used to train the first model and / or the second model; receiving the second model and / or the model parameters of the second model sent by the second device; and training the third model based on the data set measured by the first device, the second model, and / or the model parameters of the second model, where the second model is the fourth model.

[0250] In some embodiments, the first device can send the measured data set to the second device, and the second device can train the model according to the measured data set to obtain the first model encoder#1 and the second model decoder#1. The second device can send the second model and / or the model parameters of the second model to the first device, and the first device can train the third model based on the data set measured by the first device, the second model, and / or the model parameters of the second model. At this time, the third model can be denoted as encoder#2, and the second model can be determined as the fourth model, that is, the second device can use the second model to recover the CSI.

[0251] In some embodiments, the first device can be a terminal, and the second device can be a network device, i.e., the terminal side reports the measured data set to the network device, the network device trains the encoder #1 and the decoder #1 based on the received data set, and then sends the model or parameters of the decoder #1 to the terminal, and the terminal side trains the encoder #2 based on the decoder #1 and the measured data set.

[0252] When the data set is the target CSI, the feedback CSI, and the recovered target CSI, the optional examples are as follows.

[0253] Example 6-1

[0254] In some embodiments, when the data set is the target CSI, the feedback CSI, and the recovered target CSI, the first device can train the model based on the measured data set according to at least one of the above, to determine the third model and / or the fourth model. For example, the first device can use any one of the above examples 4-1 to 4-2, or use the method of any one of the above examples 5-1 to 5-3 to determine the third model and / or the fourth model.

[0255] In some embodiments, the model structure of the first model and the third model in the above example 2 can be the same or different, and when performing inference, the first device uses the third model to compress the CSI, and the second device uses the fourth model to recover the CSI.

[0256] Example 3: The first model and the second model are standardized models, for example, the first model and the second model can be standardized AI / ML models.

[0257] When the model structure and parameters of the standardized first model are standardized, the optional examples are as follows.

[0258] Example 7-1:

[0259] In some embodiments, based on the measured data set of the first device, and at least one of the first information, the first model, and the second model, the third model and / or the fourth model are determined, including: determining the standardized first model as the third model; the method further includes: sending the measured data set of the first device to the second device, and the measured data set of the first device is used by the second device to train the fourth model through the first training type Type1.

[0260] In some embodiments, the first device can send the measured data set of the first device to the second device, and the second device can train the fourth model through the first training type Type1 based on the measured data set of the first device. When performing inference, the standardized first model can be used as the third model to compress the CSI, and the fourth model can be used to recover the CSI.

[0261] For example, when the first device is a terminal and the second device is a network device, the terminal side reports the measured data set to the network device, and the network device side trains the decoder #2 by using the standardized encoder #1 and the received measured data set by using the training type Type 1. In the inference stage, the encoder #1 and the decoder #2 are used for inference to realize the compression and recovery of the CSI.

[0262] Example 7-2:

[0263] In some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model includes determining the standardized first model as the third model, and the method further includes sending the data set measured by the first device to the second device, and the data set measured by the first device is used by the second device to train the fourth model by using the second training type Type 2.

[0264] In some embodiments, the first device can send the data set measured by the first device to the second device, and the second device can train the fourth model based on the data set measured by the first device by using the second training type Type 2, and in the inference stage, the standardized first model can be used as the third model to compress the CSI, and the fourth model can be used to recover the CSI.

[0265] For example, when the first device is a terminal and the second device is a network device, N (N≥1) terminals input the measured data to the encoder #1, and then the network device side trains the decoder #2 by using the training type Type 2 and the N terminals. In the inference stage, the encoder #1 and the decoder #2 are used for inference to realize the compression and recovery of the CSI.

[0266] For example, when the network device has obtained the data set for training the fourth model, that is, the second device can collect the data set for model training to obtain the fourth model, the optional example is as follows.

[0267] Example 7-3:

[0268] In some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model includes receiving the fourth model sent by the second device, the fourth model being trained by the second device based on the data set collected by the second device and the first model by using the first training type Type 1, and training the third model by using the second training type Type 2 based on the data set measured by the first device, the first model, and the fourth model.

[0269] In some embodiments, the second device can train a fourth model decoder2 based on the collected dataset and the first model by the first training type Type1, and send the fourth model to the first device, and the first device can train a third model encoder2 based on the dataset measured by the first device, the first model and the fourth model by the second training type Type2, at this time the first device can use the third model for CSI compression, and the second device can use the fourth model for CSI recovery.

[0270] For example, when the first device is a terminal and the second device is a network device, the network device side trains a decoder#2 based on the dataset collected by the network device and the encoder#1 by the training Type 1. Then, the terminal trains an encoder#2 based on the measured dataset, the encoder#1 and the decoder#2 by the training type Type 2. In the inference stage, the encoder#2 and the decoder#2 are used for inference to realize the compression and recovery of CSI.

[0271] Example 7-4:

[0272] In some embodiments, determining the third model and / or the fourth model based on the dataset measured by the first device, and at least one of the first information, the first model and the second model comprises: determining the standardized first model as the third model; the method further comprises: receiving a tenth model sent by the second device, the tenth model being trained by the second device based on a dataset collected by the second device and the first model by the first training type Type1; training a fourth model based on the dataset measured by the first device, the first model and the tenth model by the second training type Type2.

[0273] In some embodiments, the second device can train a fourth model decoder2 based on the collected dataset and the first model by the first training type Type1, and send the fourth model to the first device, and the first device can train a third model encoder2 based on the dataset measured by the first device, the first model and the fourth model by the second training type Type2, at this time the first device can use the third model for CSI compression, and the second device can use the fourth model for CSI recovery.

[0274] For example, when the first device is a terminal and the second device is a network device, the network device side trains decoder#2 based on a data set collected by the network device and encoder#1 through training Type 1. Then, the terminal trains decoder#3 based on a measured data set, encoder#1 and decoder#2 through training Type 2, and performs inference in the inference stage to realize compression and recovery of CSI by using encoder#1 and decoder#3.

[0275] Example 7-5:

[0276] In some embodiments, determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model and the second model comprises: sending the data set measured by the first device to the second device, the data set measured by the first device being used by the second device to train the third model based on the first model and the fourth model, the fourth model being obtained by the second device based on a data set collected by the second device and the first model through a first training type Type 1; and receiving the third model sent by the second device.

[0277] In some embodiments, the second device can train the fourth model decoder2 based on a collected data set and the first model through a first training type Type 1, the first device can send the data set measured by the first device to the second device, the second device can train the third model encoder#2 based on the measured data set, the first model and the fourth model, and at this time the first device can use the third model for CSI compression and the second device can use the fourth model for CSI recovery.

[0278] For example, when the first device is a terminal and the second device is a network device, the network device side trains decoder#2 based on a data set collected by the network device and encoder#1 through training Type 1. The terminal reports the measured data set to the network device, and the network device trains encoder#2 based on the measured data set, encoder#1 and decoder#2 through training Type 1. Finally, the network device sends the trained encoder#2 to the terminal. In the inference stage, encoder#2 and decoder#2 are used to perform inference to realize compression and recovery of CSI.

[0279] When the model structure and parameters of the standardized second model are standardized, optional examples are as follows.

[0280] Example 8-1:

[0281] In some embodiments, determining the third model and / or the fourth model based on the first device measured dataset and at least one of the first information, the first model, and the second model comprises: training the third model based on the first device measured dataset and the second model by the first training type Type 1, wherein the second model is used as the fourth model.

[0282] In some embodiments, the first device can train the third model encoder#2 based on the first device measured dataset and the second model by the first training type Type 1, and use the standardized second model as the fourth model, so that the first device can use the third model to compress CSI, and the second device can use the standardized fourth model to recover CSI.

[0283] For example, when the first device is a terminal and the second device is a network device, the terminal trains the encoder#2 based on the measured dataset and the decoder#1 by the training type Type 1, and uses the encoder#2 and the decoder#1 to perform inference to realize CSI compression and recovery in the inference stage.

[0284] Example 8-2:

[0285] In some embodiments, determining the third model and / or the fourth model based on the first device measured dataset and at least one of the first information, the first model, and the second model comprises: sending the first device measured dataset to the second device, and the first device measured dataset is used by the second device to train the third model based on the second model by the first training type Type 1; and receiving the third model sent by the second device, wherein the second model is used as the fourth model.

[0286] In some embodiments, optionally, the first device can send the first device measured dataset to the second device, the second device can train the third model encoder#2 based on the first device measured dataset and the second model by the first training type Type 1, and send the third model to the first device, and use the standardized second model as the fourth model, so that the first device can use the third model to compress CSI, and the second device can use the standardized fourth model to recover CSI.

[0287] For example, when the first device is a terminal and the second device is a network device, the terminal reports the measured dataset to the network device, the network trains the encoder#2 based on the measured dataset and the decoder#1 by the training type Type 1, and then sends the encoder#2 to the terminal, and the encoder#2 and the decoder#1 are used to perform inference to realize CSI compression and recovery in the inference stage.

[0288] Example 8-3:

[0289] In some embodiments, determining the third model and / or the fourth model based on the first device measured data set, and at least one of the first information, the first model, the second model comprises: training the third model based on the first device measured data set, the second model by the second training type Type2, wherein the second model is the fourth model.

[0290] In some embodiments, the first device can train the third model based on the first device measured data set, the second model by the second training type Type2, and use the standardized second model as the fourth model. In this case, the first device can use the third model for CSI compression, and the second device can use the standardized fourth model for CSI recovery.

[0291] For example, when the first device is a terminal and the second device is a network device, N (N > 1) terminals train the encoder #2 of each terminal based on the measured data set and the decoder #1 by the training type Type2. In the inference stage, the encoder #2 and the decoder #1 are used for inference to realize the compression and recovery of CSI.

[0292] When the model structure and parameters of the first model are standardized, and the model structure and parameters of the second model are standardized, the optional examples are as follows.

[0293] Example 9-1:

[0294] In some embodiments, when the model structure and parameters of the first model and the second model are standardized, the first device can determine the third model and / or the fourth model by using the method of any one of the above examples 7-1 to 7-5, or the above examples 8-1 to 8-3.

[0295] In some embodiments, in addition to transmitting the parameters, model or training model data set of the model, the network device can also send auxiliary information of the training model to the terminal, and the auxiliary information includes one or more of the performance target, the target value of the training loss function, and the performance criterion. Alternatively, the performance target, the mapping relationship between the feedback overhead and the performance target, the target value of the training loss function, and the performance criterion are determined by negotiation between the network device and the terminal.

[0296] In some embodiments, the method in example 1, example 2, and example 3 above can use the data set measured by the first device for model training, can make the trained model applicable to the first device, can avoid the problem of performance reduction caused by mismatching of training data, and improve the performance of the model. For example, the above methods can independently solve the problem of model performance reduction, but are not limited thereto. The above methods of the examples can be combined in a non-contradictory manner to achieve model training using the data set measured by the first device, make the trained model applicable to the first device, and improve the performance of the model.

[0297] For example, using the method of example 3 above, the first model and the second model can be standardized models. The first device can perform model training based on the standardized first model and / or second model and the measured data set to obtain a third model and / or a fourth model, and transmit the parameters of the third model and / or the fourth model obtained by training to the second device based on the method of example 1, and the like.

[0298] In step 2103, the second device determines a third model and / or a fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model.

[0299] In some embodiments, the second device can determine a third model and / or a fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model. The method used by the second device to determine the third model and / or the fourth model is consistent with the method used by the first device. That is, when the first device is a terminal and the second device is a network device, the network device can determine the third model and / or the fourth model. For details, refer to the method shown in step 2102, which will not be described here.

[0300] In an embodiment of the present disclosure, the method further includes training a first model and / or a second model based on a training data set, wherein the first information includes model parameters of the first model and / or the second model and the training data set.

[0301] That is, the second device can train a first model and a second model based on a training data set, so that the first device or the second device determines a third model and / or a fourth model based on the first model and the second model.

[0302] FIG. 3 is a flow diagram of a model training method according to an embodiment of the present disclosure. As shown in FIG. 3, the present disclosure relates to a model training method for a first device. The above method includes:

[0303] In step 3101, at least one of a first model, a second model, and first information is obtained.

[0304] The optional implementation of step 3101 can refer to the optional implementation of step 2101 in FIG. 2 and other associated parts in the embodiments involved in FIG. 2, which will not be repeated here.

[0305] In some embodiments, the first device receives at least one of the first model, the second model, and the first information sent by the second device, but is not limited thereto, and can also receive at least one of the first model, the second model, and the first information sent by other subjects.

[0306] In some embodiments, the first device obtains the second indication information specified by a protocol.

[0307] In some embodiments, the first device obtains the second indication information from an upper layer.

[0308] In some embodiments, the first device processes to obtain the second indication information.

[0309] In step 3102, the third model and / or the fourth model are determined based on the data set measured by the first device and at least one of the first information, the first model, and the second model.

[0310] The optional implementation of step 3102 can refer to the optional implementation of step 2102 in FIG. 2 and other associated parts in the embodiments involved in FIG. 2, which will not be repeated here.

[0311] FIG. 4 is a flow diagram of a model training method according to an embodiment of the present disclosure. As shown in FIG. 4a, the present embodiment relates to a model training method for a second device, and the above method comprises:

[0312] In step 4101, at least one of the first model, the second model, and the first information is sent.

[0313] The optional implementation of step 4101 can refer to the optional implementation of step 2101 in FIG. 2, the optional implementation of step 3101 in FIG. 3, and other associated parts in the embodiments involved in FIG. 2 and FIG. 3a, which will not be repeated here.

[0314] In some embodiments, the first device can receive at least one of the first model, the second model, and the first information.

[0315] In some embodiments, the second device can send at least one of the first model, the second model, and the first information to the first device, but is not limited thereto, and the second device can also send at least one of the first model, the second model, and the first information to other subjects.

[0316] At step 4102, determine the third model and / or the fourth model based on the first device measured data set, and at least one of the first information, the first model, and the second model.

[0317] The optional implementation of step 4102 can refer to the optional implementation of step 2103 of FIG. 2 and other associated parts in the embodiments involved in FIG. 2, which will not be described here.

[0318] The following is an exemplary introduction to the above method.

[0319] The method shown in the embodiments of the present disclosure relates to a kind of bilateral AI model training method, as shown in Figure 5, it is the flow diagram of the CSI compression feedback of bilateral AI model based on CSI generation partial model and CSI recovery partial model, the method of this example can be applied to the bilateral AI model as shown in Figure 5, as shown in Figure 5, UE side is compressed by CSI generation partial model (defined as encoder) and is sent to gNB by quantization as binary bit stream by downlink channel information H, gNB side is restored by CSI recovery partial model (defined as decoder) H' similar to original downlink information.

[0320] But because the training method of current bilateral AI model can appear UE side data distribution and NW side data distribution are inconsistent, leading to the problem of performance decline when the model trained is directly used for inference, therefore how to solve the performance reduction caused by data mismatch is a problem to be solved.

[0321] The method of this example proposes a bilateral model training method to solve the performance decline problem caused by terminal side data and network side data mismatch for the cooperation training option between different devices, in order to facilitate description, the following encoder represents terminal side CSI generation partial model, and decoder represents network side (NetWork, NW) CSI recovery partial model. The complete content of the method is as follows.

[0322] Scenario 1: for bilateral cooperation training option 3a / 5a scenario. Assume that the NW side has trained encoder#1 and decoder#1 model.

[0323] For option 3a / 5a-1: pass the model / parameter of encoder.

[0324] Scheme 1-1: UE trains decoder#2 based on the received encoder#1 model / parameters, wherein the dataset for training decoder#2 is the dataset of encoder#1 and decoder#1 trained at the NW side and sent to the UE by the NW. UE trains encoder#2 based on the dataset measured at the UE side and decoder#2, wherein the model structure of encoder#1 and encoder#2 is the same, and the scheme adopts encoder#2 and decoder#1 for inference in the inference stage.

[0325] Scheme 1-2: UE trains encoder#2 based on the dataset measured at the UE side and / or the dataset sent by the NW side and the received encoder#1 model / parameters by using training type 2, wherein the sent dataset is the dataset of encoder#1 and decoder#1 trained at the NW side, and the scheme adopts encoder#2 and decoder#1 for inference in the inference stage.

[0326] Scheme 1-3: UE sends the measured dataset to the NW, and the NW re-trains encoder#1 based on the dataset measured at the UE, and then sends the model / parameters of encoder#1 to the UE.

[0327] For option 3a / 5a-2: pass the model / parameters of decoder scenario.

[0328] Scheme 2-1: UE trains encoder#2 based on the received model / parameters of decoder#1 and the dataset measured at the UE side and / or the dataset sent by the NW, wherein the model structure of encoder#2 and encoder#1 can be the same or different.

[0329] Scheme 2-2: UE sends the measured dataset to the NW, and the NW re-trains decoder#1 based on the dataset measured at the UE, and then sends the model / parameters of decoder#1 to the UE.

[0330] For option 3a / 5a-3: pass the model / parameters of encoder and decoder scenario.

[0331] Scheme 3-1: encoder#2 can be trained by using the methods of schemes 1-1-1-3 or schemes 2-1-2-2.

[0332] Wherein, the model structure of the trained encoder#2 and encoder#1 can be the same or different, and encoder#2 and decoder#1 are adopted for inference in the inference stage to realize the compression and recovery of CSI.

[0333] Scenario 2: For bilateral coordination training Option 4 scenario.

[0334] For Option 4-1: Dataset is target CSI and CSI feedback scenario.

[0335] Solution 4-1: UE side trains encoder#2 based on received target CSI and CSI feedback, then trains encoder#3 based on UE measured dataset and NW side decoder#1 by using training type 2.

[0336] Solution 4-2: UE side first sends measured dataset to NW, NW trains encoder#1 and decoder#1 based on UE measured dataset, then sends CSI feedback of trained encoder#2 to UE. Optionally, target CSI of trained encoder#2 is also sent to UE, UE trains encoder#3 based on received CSI feedback and / or target CSI.

[0337] For Option 4-2: Dataset is CSI feedback and reconstructed target CSI scenario.

[0338] Solution 5-1: UE side trains decoder#2 based on received CSI feedback and reconstructed target CSI. Then trains encoder#2 based on UE measured dataset and decoder#2, UE side trains encoder#3 by using training type 2 with measured dataset and decoder#1.

[0339] Solution 5-2: UE side trains decoder#2 based on received CSI feedback and reconstructed target CSI. Then UE side trains encoder#2 based on UE measured dataset and decoder#2 by using training type 1.

[0340] Solution 5-3: UE side reports measured dataset to NW, NW trains encoder#1 and decoder#1 based on received dataset, then sends model or parameters of decoder#1 to UE, UE side trains encoder#2 based on that decoder#1 and measured dataset.

[0341] For Option 4-3: The dataset is target CSI, CSI feedback and reconstructed target CSI scenario.

[0342] Scheme 6-1: The encoder#2 or encoder#3 can be trained by using the method of Alt4-1~4-2 or Alt5-1~5-3.

[0343] Wherein, the model structure of the above trained encoder#1, encoder#2, encoder#3 can be the same or different, and the encoder#3 and decoder#1 are used for inference in the inference stage to realize the compression and recovery of CSI; the above options 3 / 4 / 5 are assumed that the training of encoder and decoder has been completed on the NW side; the above method can also be applied to the training of encoder and decoder on the UE side, and the model, parameters or dataset are sent from the UE to the NW.

[0344] Scenario 3: For bilateral cooperative training Option 1 scenario. Assume that the standardized AI / ML model is represented as encoder#1 and decoder#1.

[0345] Scenario 3-1: Standardized encoder model structure and parameter scenario.

[0346] Scheme 7-1: The UE side reports the measured dataset to the NW, and the NW side trains the decoder#2 by using the standardized encoder#1 and the received measured dataset through training type 1, and the encoder#1 and decoder#2 are used for inference in the inference stage to realize the compression and recovery of CSI.

[0347] Scheme 7-2: N (N≥1) UEs input the measured data to the encoder#1, and then the NW side trains the decoder#2 by using the training type Type 2 and N UEs for joint training, and the encoder#1 and decoder#2 are used for inference in the inference stage to realize the compression and recovery of CSI.

[0348] The following schemes 7-3~7-4 assume that the NW has obtained the dataset for training the decoder:

[0349] Scheme 7-3: NW side trains decoder#2 based on NW collected dataset and encoder#1 by training Type 1, then UE trains encoder#2 based on measured dataset, encoder#1 and decoder#2 by training Type 2, and finally, encoder#2 and decoder#2 are used for inference to realize the compressed recovery of CSI in the inference stage.

[0350] Scheme 7-4: NW side trains decoder#2 based on NW collected dataset and encoder#1 by training Type 1, UE trains decoder#3 based on measured dataset, encoder#1 and decoder#2 by training Type 2, and finally, encoder#1 and decoder#3 are used for inference to realize the compressed recovery of CSI in the inference stage.

[0351] Scheme 7-5: NW side trains decoder#2 based on NW collected dataset and encoder#1 by training Type 1. UE reports measured dataset to NW, NW trains encoder#2 based on measured dataset, encoder#1 and decoder#2 by training Type 1, and finally, NW sends trained encoder#2 to UE, and encoder#2 and decoder#2 are used for inference to realize the compressed recovery of CSI in the inference stage.

[0352] Scenario 3-2: Standardized model structure and parameter scenario of decoder.

[0353] Scheme 8-1: UE trains encoder#2 based on measured dataset and decoder#1 by training Type 1, and encoder#2 and decoder#1 are used for inference to realize the compressed recovery of CSI in the inference stage. Optionally, UE reports measured dataset to NW, NW trains encoder#2 based on measured dataset and decoder#1 by training Type 1, and then sends encoder#2 to UE.

[0354] Scheme 8-2: N (N>1) UEs train their encoder#2 based on measured dataset and decoder#1 by training Type 2, and encoder#2 and decoder#1 are used for inference to realize the compressed recovery of CSI in the inference stage.

[0355] Scenario 3-3: Standardized model structure and parameter scenario of encoder and decoder.

[0356] Scheme 9-1: One of Alt7-1~7-4 and Alt8-1~8-2 described above can be used.

[0357] In the above, the NW can send the UE, in addition to the parameters of the model, the model or the dataset for training the model, auxiliary information for training the model, including one or more of the performance target, the target value of the training loss function, and the performance criterion. Alternatively, the performance target, the mapping relationship between the feedback overhead and the performance target, the target value of the training loss function, and the performance criterion are determined by predefinition through negotiation between the NW and the UE.

[0358] The above schemes in the above scenarios are explained and described below through specific embodiments.

[0359] Embodiment 1 (Scenario 1):

[0360] Suppose UE1 accesses the cell NW1, and the NW1 side trains an encoder#1 and a decoder#2 based on the collected training dataset, and the collected dataset does not contain the data of UE1. Suppose UE1 trains an encoder through the option 3a. The NW sends the dataset of encoder#1 to UE1, and UE1 determines that there is no encoder#1 according to the standardized model structure and the received model parameters. Limited by the terminal capability, UE1 can not have the capability to train the encoder and the decoder, in which case the training of the encoder or the decoder will be completed by the server on the UE side (the UE side training in this proposal refers to the training of the AI / ML model by the UE itself or the server on the UE side). In order to train an encoder that matches the decoder#1 on the NW side and obtain better inference performance, the following methods can be used to complete the training of the encoder:

[0361] (Scheme 1-1): In addition to sending the parameters of encoder#1, the NW side also sends the UE side all or part of the training dataset and the auxiliary information of the trained model of encoder#1 and decoder#1, such as the performance criterion of the trained model SGCS, and the feedback overhead performance target cosine similarity square (Square Generalized Cosine Similarity, SGCS) is 0.75. The UE side trains decoder#2 based on the received dataset and the auxiliary information of the trained model of encoder#1. Then, the UE side measures the downlink channel information according to the downlink pilot signal such as the channel status information-reference signal (Channel Status Information-Reference Signal, CSI-RS), and trains encoder#2 based on the measured downlink dataset and the trained decoder#2. During the training, the model parameters of decoder#2 are fixed. Finally, UE1 and NW1 use encoder#2 and decoder#1 for inference respectively.

[0362] (Scheme 1-2): In the above example (Scheme 1-1), the trained encoder# and decoder#2 are both completed on the UE side, which can be regarded as being completed based on training type Type 1. This example is completed using training type Type 2. Specifically, the UE still obtains encoder#1 first, and measures the channel information of UE1 based on the downlink pilot as the training dataset. Then, the UE side and the NW respectively use encoder#1 and decoder#1 to train encoder#2 through training type Type 2, and it is noted that the model parameters of decoder#1 are fixed during the training process. Finally, UE1 and NW1 still use encoder#2 and decoder#1 for inference respectively.

[0363] (Scheme 1-3): UE1 sends the measured downlink channel information to the NW, and the NW takes it as the dataset of the trained model. The NW uses encoder#1 and decoder#1 to train the model by using training type Type 1, and still keeps the model parameters of decoder#1 unchanged during the training process, thereby training encoder#2. Finally, UE1 and NW1 still use encoder#2 and decoder#1 for inference respectively.

[0364] The above is an example of passing the model parameters of encoder#1. If the model parameters of decoder#1 are passed, the training method is similar, and the training type Type 1 and the measured channel information can also be used as the training data set to complete the model training to obtain encoder#2. The detailed process is not repeated. Finally, UE1 and NW1 still use encoder#2 and decoder#1 for inference, respectively. If the NW side passes the model parameters of encoder#1 and decoder#1 to the UE, the UE side can also complete the model training to obtain the encoder#2 matched with the decoder#1 by the above method to obtain better model inference.

[0365] Embodiment 2 (Scenario 2):

[0366] It is still assumed that UE1 accesses the cell NW1, and the NW1 side trains encoder#1 and decoder#2 based on the collected training data set, and the data set does not contain the data of UE1. In the process of training encoder#1 and decoder#1, the NW can obtain the input data set target CSI and the output data set CSI feedback of encoder#1, and the input data set CSI feedback and the output data set reconstructed CSI of decoder#1.

[0367] (Scheme 4-1) In addition to sending the training data set target CSI and CSI feedback parameters, the NW side also sends auxiliary information of the training model to the UE side, such as the performance criterion of the training model SGCS, and the feedback overhead is 60 bits. The performance target SGCS is 0.75. First, the UE side can train encoder#2 based on the received training data set and auxiliary information. Then, the UE side and the NW use encoder#2 and decoder#1, respectively, to train encoder#3 through training type Type 2. In the training process, the model parameters of decoder#1 do not change, and only the model parameters of encoder#2 are updated. Finally, UE1 and NW1 use encoder#3 and decoder#1 for inference, respectively.

[0368] (Scheme 4-2) UE side sends the measured channel information as the training dataset to NW, NW trains the model of encoder#2 based on the dataset measured by UE and encoder#1 / decoder#1, and the model parameters of decoder#2 are not changed during the training process. Then, NW sends the CSI feedback generated during the training of encoder#2 to UE, and UE side trains encoder#3 based on the received CSI feedback and auxiliary information. Finally, UE1 and NW1 perform inference using encoder#3 and decoder#1, respectively.

[0369] Similarly, if NW sends CSI feedback and reconstructed CSI to UE side, UE side completes the training of encoder based on the above schemes 5-1~5-3, and UE and NW perform inference using the trained encoder and decoder#1, respectively. The detailed steps are not repeated. If NW sends the dataset as target CSI, CSI feedback and reconstructed target CSI to UE, UE side can also complete the training of the model parameters of encoder using the above example method.

[0370] Embodiment 3 (Scenario 3):

[0371] It is assumed that the model structure and parameters of encoder#1 are standardized.

[0372] (Scheme 7-1) The model structure and parameters of the standardized encoder#1 are kept unchanged, UE1 sends the measured data to NW as the training dataset, and NW side trains decoder#2 based on encoder#1 and the training dataset through training type Type 1. In the inference stage, encoder#1 and decoder#2 are used for inference to realize the compression and recovery of CSI.

[0373] (Scheme 7-2) In order to reduce the overhead of uplink transmission data, decoder#2 can be trained by N(N≥1) UEs and NW through training type type 2. Specifically, it is assumed that each UE has deployed an encoder, N UEs send the measured data as the input data of encoder#1, and N UEs and NW perform joint training in multiple-to-one mode, and the model parameters of the encoder on the UE side are not updated during the training process, so that decoder#2 is trained. UE and NW perform inference using encoder#1 and decoder#2, respectively, to realize the compression and recovery of CSI.

[0374] Assuming that UE1 accesses to cell NW1, NW1 side trains encoder#1 and decoder#2 based on the collected training data set, and the collected data set does not contain the data of UE1. Among them, encoder#1 is the standardized model. As described in the example of (solution 7-2), NW can update the model of encoder to obtain encoder#2 (solution 7-3) based on decoder#2 and the data set measured by UE, or update the parameters of the decoder model to obtain decoder#3 (solution 7-4). Alternatively, (solution 7-5) UE sends the measured data set to NW, and NW trains and updates encoder#1 to obtain encoder#2 through training type 1, and finally NW sends encoder#2 to UE, and encoder#2 and decoder#2 are used for inference in the inference stage to realize the compression and recovery of CSI.

[0375] Assuming that the model structure and parameters of decoder#1 are standardized. Encoder#2 can also be trained through similar steps in solutions 8-1 to 8-2 described above. If the model structure and parameters of encoder#1 and decoder#1 are standardized, encoder#2 and / or decoder#2 can also be obtained through similar steps, so that the trained model matches part of the model of the opposite end, and the trained model is suitable for the measurement data of UE1.

[0376] In summary, the above embodiments of the present solution can solve the problem of mismatch between terminal data and network data by training the model using the data set measured by the terminal, so as to improve the inference performance of the AI / ML model.

[0377] The method is as follows: FIG. 6a is a structural schematic diagram of the first device 101 according to an embodiment of the present disclosure. As shown in FIG. 6a, the first device 101 comprises a transceiver module 6101 configured to obtain at least one of a first model, a second model, and first information, the first information being used for training the first model and / or the second model; optionally, the transceiver module is configured to perform at least one of the steps related to the transceiving performed by the first device 101 in any of the above methods (for example, step 2101, but not limited thereto), which will not be described herein again.

[0378] In some embodiments, the first device 101 further includes a processing module 6102 configured to determine a third model and / or a fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model, the third model being the same as the first model or being obtained by retraining the first model, the third model being used for channel state information (CSI) compression by the first device, and the fourth model being used for CSI recovery by the second device. Optionally, the processing module is configured to perform at least one of the steps related to the processing performed by the first device 101 in any of the methods described above (for example, step 2102, and the like, but not limited thereto), details of which are not described herein again.

[0379] Fig. 6b is a structural schematic diagram of the second device 102 according to an embodiment of the present disclosure. As shown in Fig. 6b, the second device 102 includes a transceiver module 6201 configured to send at least one of the first model, the second model, and the first information to the first device, the first information being used for training the first model and / or the second model. Optionally, the transceiver module is configured to perform at least one of the steps of transceiving and the like performed by the second device 102 in any of the methods described above (for example, step 2101, and the like, but not limited thereto), details of which are not described herein again.

[0380] In some embodiments, the second device 102 further includes a processing module 6202 configured to determine a third model and / or a fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model, the third model being the same as the first model or being obtained by retraining the first model, the third model being used for channel state information (CSI) compression by the first device, and the fourth model being used for CSI recovery by the second device. Optionally, the processing module is configured to perform at least one of the steps of processing and the like performed by the second device 102 in any of the methods described above (for example, step 2103, and the like, but not limited thereto), details of which are not described herein again.

[0381] As shown in Fig. 7a, the communication device 7100 includes one or more processors 7101. The processor 7101 can be a general processor or a special-purpose processor, for example, a baseband processor or a central processor. The baseband processor can be configured to process communication protocols and communication data, and the central processor can be configured to control the communication device (for example, a base station, a baseband chip, a terminal device, a terminal device chip, a DU, or a CU), execute programs, and process data of the programs. The processor 7101 is configured to invoke instructions to enable the communication device 7100 to perform any of the methods described above.

[0382] In some embodiments, the communication device 7100 further includes one or more memories 7102 configured to store instructions. Optionally, all or part of the memory 7102 can also be located outside the communication device 7100.

[0383] In some embodiments, the communication device 7100 further includes one or more transceivers 7103. When the communication device 7100 includes one or more transceivers 7103, the communication steps in the above methods, such as sending and receiving, are performed by the transceiver 7103, and other steps are performed by the processor 7101.

[0384] In some embodiments, the transceiver can include a receiver and a transmitter, which can be separate or integrated together. Optionally, the terms transceiver, transceiving unit, transceiver, transceiving circuit, etc. can be replaced by each other, the terms transmitter, transmitting unit, transmitter, transmitting circuit, etc. can be replaced by each other, and the terms receiver, receiving unit, receiver, receiving circuit, etc. can be replaced by each other.

[0385] Optionally, the communication device 7100 further includes one or more interface circuits 7104, which are connected to the memory 7102, and can be used to receive signals from the memory 7102 or other devices, and can be used to send signals to the memory 7102 or other devices. For example, the interface circuit 7104 can read the instructions stored in the memory 7102 and send them to the processor 7101.

[0386] The communication device 7100 described in the above embodiments can be a network device or a terminal, but the scope of the communication device 7100 described in the present disclosure is not limited to this, and the structure of the communication device 7100 can not be limited by Figure 7a. The communication device can be a standalone device or can be part of a larger device. For example, the communication device can be: 1) a standalone integrated circuit (IC), or a chip, or a chip system or subsystem; (2) a set of one or more ICs, which can optionally also include storage components for storing data, programs; (3) an ASIC, such as a Modem; (4) a module that can be embedded in other devices; (5) a receiver, terminal device, smart terminal device, cellular phone, wireless device, handset, mobile unit, vehicle-mounted device, network device, cloud device, artificial intelligence device, etc.; (6) other, etc.

[0387] Figure 7b is a structural schematic diagram of a chip 7200 according to an embodiment of the present disclosure. For the case where the communication device 7100 is a chip or a chip system, the structural schematic diagram of the chip 7200 shown in Figure 7b can be referred to, but is not limited thereto.

[0388] The chip 7200 includes one or more processors 7201, which are used to invoke instructions to cause the chip 7200 to perform any of the above methods.

[0389] In some embodiments, chip 7200 further includes one or more interface circuits 7202 connected to memory 7203, which can be used to receive signals from or send signals to memory 7203 or other devices. For example, interface circuit 7202 can read instructions stored in memory 7203 and send the instructions to processor 7201. Optionally, the terms interface circuit, interface, transceiver pin, transceiver, and the like can be replaced with each other.

[0390] In some embodiments, chip 7200 further includes one or more memories 7203 for storing instructions. Optionally, all or part of memory 7203 can be outside chip 7200.

[0391] The present disclosure further proposes a storage medium, which has instructions stored thereon, and when the instructions are run on communication device 7100, communication device 7100 performs any of the above methods. Optionally, the storage medium is an electronic storage medium. Optionally, the storage medium is a computer readable storage medium, but is not limited to this, and it can also be a storage medium readable by other devices. Optionally, the storage medium can be a non-transitory storage medium, but is not limited to this, and it can also be a transitory storage medium.

[0392] The present disclosure further proposes a program product, which is executed by communication device 7100, and causes communication device 7100 to perform any of the above methods. Optionally, the program product is a computer program product.

[0393] The present disclosure further proposes a computer program, which, when run on a computer, causes the computer to perform any of the above methods.

[0394] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer programs. When the computer programs are loaded on a computer and executed, all or part of the processes or functions described in the embodiments of the present disclosure are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer programs can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another computer-readable storage medium, for example, the computer programs can be transferred from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that includes one or more available media sets. The available media can be magnetic media (such as floppy disk, hard disk, magnetic tape), optical media (such as high-density digital video disc (digital video disc, DVD)), or semiconductor media (such as solid state disk (solid state disk, SSD)) and the like.

[0395] The correspondence relationship shown in each table in the present disclosure can be configured or predefined. The values of the information in each table are only examples, and other values can be configured, and the present disclosure does not limit. When configuring the correspondence relationship between the information and each parameter, it is not necessarily required to configure all the correspondence relationships shown in each table. For example, the correspondence relationship shown in some rows in the table in the present disclosure can also not be configured. For another example, the above table can be appropriately deformed and adjusted, for example, split, merged, etc. The parameter name shown in the title of each table above can also use other names understandable by the communication device, and the parameter value or representation method can also use other values or representation methods understandable by the communication device. Each table above can also use other data structures when implemented, for example, arrays, queues, containers, stacks, linear tables, pointers, linked lists, trees, graphs, structures, classes, heaps, hash tables or hash tables, etc.

[0396] The predefinition in the present disclosure can be understood as definition, predefinition, storage, prestorage, prenegotiation, preconfiguration, solidification or pre-burning.

[0397] Those skilled in the art can clearly understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software manner depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the disclosure.

[0398] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0399] The above is only a specific implementation of the present disclosure, but the protection scope of the present disclosure is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present disclosure, which should be covered within the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be subject to the protection scope of the claims.

Claims

1. A model training method, characterized in that, The method is performed by a first device, comprising: obtaining at least one of a first model, a second model, and first information used for training the first model and / or the second model; determining a third model and / or a fourth model based on at least one of a data set measured by the first device, the first information, the first model, and the second model, the third model being the same as the first model or being obtained by retraining the first model, the fourth model being the same as the second model or being obtained by retraining the second model, the third model being used for channel state information (CSI) compression by the first device, and the fourth model being used for CSI recovery by a second device.

2. The method of claim 1, wherein, The obtaining at least one of the first model, the second model, and the first information comprises: obtaining at least one of the first model, the second model, and the first information from the second device, the first information comprising model parameters of the first model and / or the second model and a training data set, the first model and / or the second model being obtained by the second device based on the training data set.

3. The method of claim 2, wherein, The determining the third model and / or the fourth model based on at least one of the data set measured by the first device, the first information, the first model, and the second model comprises: training a fifth model based on the first model and model parameters of the first model using the training data set; performing model training based on the data set measured by the first device and the fifth model to obtain the third model, wherein the second model is the fourth model.

4. The method of claim 2, wherein, The determining the third model and / or the fourth model based on at least one of the data set measured by the first device, the first information, the first model, and the second model comprises: performing model training by a second training type (Type 2) based on the data set measured by the first device, the training data set, the first model, and model parameters of the first model to obtain the third model, wherein the second model is the fourth model.

5. The method of claim 2, wherein, The determining the third model and / or the fourth model based on at least one of the data set measured by the first device, the first information, the first model, and the second model comprises: sending the data set measured by the first device to the second device, the data set measured by the first device being used by the second device to perform model training to obtain the third model; receiving the third model and / or model parameters of the third model sent by the second device, wherein the second model is the fourth model.

6. The method of claim 2, wherein, The determining the third model and / or the fourth model based on at least one of the data set measured by the first device, the first information, the first model, and the second model comprises: training the third model using the training data set and / or the data set measured by the first device, the second model, and model parameters of the second model, wherein the second model is the fourth model.

7. The method of claim 2, wherein, The determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the first device measured data set comprises: sending the first device measured data set to the second device, the first device measured data set being used by the second device to train the second model; receiving the second model and / or the model parameters of the second model sent by the second device; training the third model based on the second model and / or the model parameters of the second model, wherein the second model is used as the fourth model.

8. The method of claim 1, wherein, The obtaining the at least one of the first information, the first model and the second model comprises: obtaining the at least one of the first model, the second model and the first information trained by the second device or the first device, the first information comprising at least one of target CSI, feedback CSI and recovered CSI, the target CSI being an input of training the first model, the feedback CSI being an output of training the first model and an input of training the second model, and the recovered CSI being an output of training the second model.

9. The method of claim 8, wherein, The determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the first device measured data set comprises: training a sixth model based on the target CSI and the feedback CSI; training the sixth model based on the first device measured data set and the second model by using a second training type Type2 to obtain the third model, wherein the second model is used as the fourth model.

10. The method of claim 8, wherein, The determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the first device measured data set comprises: sending the first device measured data set to the second device, the first device measured data set being used by the second device to train a seventh model; receiving the feedback CSI and / or the target CSI sent by the second device for training the seventh model; training the third model based on the feedback CSI and / or the target CSI for training the seventh model, wherein the second model is used as the fourth model.

11. The method of claim 8, wherein, The determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the first device measured data set comprises: training an eighth model based on the feedback CSI and the recovered CSI; training a ninth model based on the first device measured data set and the eighth model; training the third model based on the first device measured data set, the ninth model and the second model by using a second training type Type2, wherein the second model is used as the fourth model.

12. The method of claim 8, wherein, The determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the data set measured by the first device comprises: training an eighth model based on the feedback CSI and the recovered CSI; training the third model based on the data set measured by the first device and the eighth model by using a first training type Type 1, wherein the second model is used as the fourth model.

13. The method of claim 8, wherein, The determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the data set measured by the first device comprises: sending the data set measured by the first device to the second device, wherein the data set measured by the first device is used to train the first model and / or the second model; receiving the second model and / or the model parameters of the second model sent by the second device; training the third model based on the data set measured by the first device, the second model and / or the model parameters of the second model, wherein the second model is used as the fourth model.

14. The method of claim 1, wherein, The obtaining the at least one of the first model, the second model and the first information comprises: obtaining the at least one of the standardized first model, the standardized second model and the first information, wherein the first information comprises at least one of the model parameters of the first model and / or the second model and the standardized training data set.

15. The method of claim 14, wherein, The determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the data set measured by the first device comprises: determining the standardized first model as the third model; The method further comprises: sending the data set measured by the first device to the second device, wherein the data set measured by the first device is used for the second device to train the fourth model by using a first training type Type 1.

16. The method of claim 14, wherein, The determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the data set measured by the first device comprises: determining the standardized first model as the third model; The method further comprises: sending the data set measured by the first device to the second device, wherein the data set measured by the first device is used for the second device to train the fourth model by using a second training type Type 2.

17. The method of claim 14, wherein, The determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the data set measured by the first device comprises: receiving the fourth model sent by the second device, wherein the fourth model is trained by the second device based on a data set collected by the second device and the first model is trained by using a first training type Type 1; training the third model by using a second training type Type 2 based on the data set measured by the first device, the first model and the fourth model.

18. The method of claim 14, wherein, The determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the data set measured by the first device comprises: determining the normalized first model as the third model; The method further comprises: receiving a tenth model sent by a second device, the tenth model being obtained by the second device based on a data set collected by the second device and the first model through a first training type Type 1; obtaining the fourth model through a second training type Type 2 based on the data set measured by the first device, the first model and the tenth model.

19. The method of claim 14, wherein, The determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the data set measured by the first device comprises: sending the data set measured by the first device to a second device, the data set measured by the first device being used by the second device to obtain the third model based on the first model and the fourth model, the fourth model being obtained by the second device based on a data set collected by the second device and the first model through a first training type Type 1; receiving the third model sent by the second device.

20. The method of claim 14, wherein, The determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the data set measured by the first device comprises: obtaining the third model through a first training type Type 1 based on the data set measured by the first device and the second model, wherein the second model is the fourth model.

21. The method of claim 14, wherein, The determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the data set measured by the first device comprises: sending the data set measured by the first device to a second device, the data set measured by the first device being used by the second device to obtain the third model based on the second model through a first training type Type 1; receiving the third model sent by the second device, wherein the second model is the fourth model.

22. The method of claim 14, wherein, The determining the third model and / or the fourth model based on the at least one of the first information, the first model, the second model and the data set measured by the first device comprises: obtaining the third model through a second training type Type 2 based on the data set measured by the first device and the second model, wherein the second model is the fourth model.

23. A model training method, comprising: The method is performed by a second device and comprises: sending at least one of a first model, a second model and first information to a first device, the first information being used to train the first model and / or the second model; determine a third model and / or a fourth model based on at least one of the first information, the first model, the second model and the data set measured by the first device, the third model being the same as the first model or being obtained by retraining the first model, and the fourth model being the same as the second model or being obtained by retraining the second model, the third model being used for channel state information (CSI) compression by the first device, and the fourth model being used for CSI recovery by a second device.

24. The method of claim 23, wherein, The method further comprises: training the first model and / or the second model based on a training data set, wherein the first information comprises model parameters of the first model and / or the second model and the training data set.

25. The method of claim 24, wherein, The determining the third model and / or the fourth model based on at least one of the first information, the first model, the second model and the data set measured by the first device comprises: taking the second model as the fourth model. wherein the third model is obtained by the first device using the training data set, training a fifth model based on the first model and model parameters of the first model, and performing model training based on the data set measured by the first device and the fifth model.

26. The method of claim 24, wherein, The determining the third model and / or the fourth model based on at least one of the first information, the first model, the second model and the data set measured by the first device comprises: taking the second model as the fourth model. wherein the third model is obtained by the first device performing model training based on the data set measured by the first device, the training data set, the first model and model parameters of the first model by a second training type (Type 2).

27. The method of claim 24, wherein, The determining the third model and / or the fourth model based on at least one of the first information, the first model, the second model and the data set measured by the first device comprises: taking the second model as the fourth model. The method further comprises: receiving the data set measured by the first device sent by the first device; performing model training based on the data set measured by the first device to obtain the third model; sending the third model and / or model parameters of the third model to the first device.

28. The method of claim 24, wherein, The determining the third model and / or the fourth model based on at least one of the first information, the first model, the second model and the data set measured by the first device comprises: determining the second model as the fourth model. wherein the third model is obtained by the first device using the training data set and / or the data set measured by the first device, the second model, and model parameters of the second model.

29. The method of claim 24, wherein, The determining the third model and / or the fourth model based on at least one of the first information, the first model, the second model and the data set measured by the first device comprises: taking the second model as the fourth model. receiving the first device measured data set sent by the first device; training the second model based on the first device measured data set; sending the second model and / or model parameters of the second model to the first device, wherein the third model is trained by the first device based on the second model and / or the model parameters of the second model.

30. The method of claim 23, wherein, The method further comprises: training the first model and / or the second model based on a training data set; Or, receiving at least one of the first model, the second model and the first information trained by the first device; The first information includes at least one of target CSI, feedback CSI and recovered CSI, the target CSI is the input of training the first model, the feedback CSI is the output of training the first model and the input of training the second model, and the recovered CSI is the output of training the second model.

31. The method of claim 30, wherein, The third model and / or the fourth model are determined based on at least one of the first device measured data set, the first information, the first model and the second model, comprising: The second model is taken as the fourth model; The third model is a sixth model trained by the first device based on the target CSI and the feedback CSI, and is trained by the first device based on the first device measured data set, the second model and the second training type Type2.

32. The method of claim 30, wherein, The third model and / or the fourth model are determined based on at least one of the first device measured data set, the first information, the first model and the second model, comprising: The second model is taken as the fourth model; receiving the first device measured data set sent by the first device; training a seventh model based on the first device measured data set; sending the feedback CSI and / or the target CSI used for training the seventh model to the first device, wherein the third model is trained by the first device based on the feedback CSI and / or the target CSI used for training the seventh model.

33. The method of claim 30, wherein, The third model and / or the fourth model are determined based on at least one of the first device measured data set, the first information, the first model and the second model, comprising: The second model is taken as the fourth model; The third model is an eighth model trained by the first device based on the feedback CSI and the recovered CSI, a ninth model trained by the first device based on the first device measured data set, the eighth model and the second training type Type2, and a model trained by the first device based on the first device measured data set, the ninth model and the second model. The third model and / or the fourth model are determined based on at least one of the first device measured data set, the first information, the first model and the second model, comprising:

34. The method of claim 30, wherein, The second model is taken as the fourth model; ​ The third model is an eighth model trained by the first device based on the feedback CSI and the recovered CSI, and is trained based on a data set measured by the first device and the eighth model by using a first training type Type 1.

35. The method of claim 30, wherein, The determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model comprises: The second model is used as the fourth model. The data set measured by the first device sent by the first device is received. The first model and / or the second model are trained based on the data set measured by the first device. The second model and / or the model parameters of the second model are sent to the first device, and the third model is an eighth model trained by the first device based on the data set measured by the first device, the second model, and / or the model parameters of the second model.

36. The method of claim 23, wherein, The method further comprises: At least one of the standardized first model, the standardized second model, and the first information is obtained, and the first information comprises model parameters of the first model and / or the second model and a standardized training data set.

37. The method of claim 36, wherein, The determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model comprises: The data set measured by the first device sent by the first device is received. The fourth model is trained by the first training type Type 1 based on the data set measured by the first device, and the standardized first model is used as the third model.

38. The method of claim 36, wherein, The determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model comprises: The data set measured by the first device sent by the first device is received. The fourth model is trained by the second training type Type 2 based on the data set measured by the first device, and the standardized first model is used as the third model.

39. The method of claim 36, wherein, The determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model comprises: The fourth model is trained by the first training type Type 1 based on the data set collected by the second device and the first model; The fourth model is sent to the first device, and the third model is trained by the second training type Type 2 based on the data set measured by the first device, the first model, and the fourth model.

40. The method of claim 36, wherein, The determining the third model and / or the fourth model based on the data set measured by the first device and at least one of the first information, the first model, and the second model comprises: The tenth model is trained by the first training type Type 1 based on the data set collected by the second device and the first model. sending the tenth model to the first device, wherein the fourth model is obtained by a second training Type 2 based on the first device measured data set, the first model and the tenth model, and the normalized first model is used as the third model; receiving the fourth model sent by the first device.

41. The method of claim 36, wherein, The determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model and the second model comprises: obtaining the fourth model by a first training Type 1 based on the second device collected data set and the first model; receiving the first device measured data set sent by the first device; obtaining the third model by training based on the first device measured data set, the first model and the fourth model; sending the third model to the first device.

42. The method of claim 36, wherein, The determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model and the second model comprises: using the second model as the fourth model; wherein the third model is obtained by a first training Type 1 based on the first device measured data set and the second model.

43. The method of claim 36, wherein, The determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model and the second model comprises: receiving the first device measured data set sent by the first device; obtaining the third model by a first training Type 1 based on the first device measured data set and the second model; sending the third model to the first device.

44. The method of claim 36, wherein, The determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model and the second model comprises: using the second model as the fourth model; wherein the third model is obtained by a second training Type 2 based on the first device measured data set and the second model.

45. A first device, comprising: comprises: receiving the first device measured data set sent by the first device; obtaining the third model by a first training Type 1 based on the first device measured data set and the second model; 46. A second device, comprising: sending the third model to the first device. The determining the third model and / or the fourth model based on the first device measured data set and at least one of the first information, the first model and the second model comprises: using the second model as the fourth model; wherein the third model is obtained by a second training Type 2 based on the first device measured data set and the second model. comprises: a transceiver, configured to acquire at least one of a first model, a second model and first information, the first information being used for training the first model and / or the second model; a processing module, configured to determine a third model and / or a fourth model based on a first device measured data set and at least one of the first information, the first model and the second model, the third model being the same as the first model or being obtained by retraining the first model, the third model being used for channel state information (CSI) compression of the first device, and the fourth model being used for CSI recovery of a second device. comprises: transmitting, to a first device, at least one of a first model, a second model, and first information used for training the first model and / or the second model; determining, based on a data set measured by the first device and at least one of the first information, the first model, and the second model, a third model and / or a fourth model, the third model being the same as the first model or being obtained by retraining the first model, the third model being used for channel state information, CSI, compression by the first device, and the fourth model being used for CSI recovery by a second device.

47. A communication system, characterized by comprising: a first device, a transmitting entity of a first protocol layer of the first device being configured to perform the method according to any one of claims 1 to 22; a second device, a receiving entity of a first protocol layer of the second device being configured to perform the method according to any one of claims 23 to 44.

48. A communications device, comprising: comprising: a transceiver; a memory; a processor connected with the transceiver and the memory respectively, configured to control wireless signal transceiving of the transceiver by executing computer executable instructions on the memory, and capable of implementing the method according to any one of claims 1 to 44.

49. A computer storage medium, wherein, The computer storage medium stores computer executable instructions; the computer executable instructions are executed by the processor, and capable of implementing the method according to any one of claims 1 to 44. The computer storage medium stores computer executable instructions; the computer executable instructions are executed by the processor, and capable of implementing the method according to any one of claims 1 to 44.

Images