A model generation method, device, equipment and readable storage medium

Abstract

The application discloses a model generation method and device, equipment and a readable storage medium, and relates to the technical field of communication, to reduce air interface overhead. The method comprises the following steps: sending first information to a terminal, wherein the first information comprises information of a generator model and information of a discriminator model; receiving information of a first generator model sent by the terminal, wherein the first generator model is generated by the terminal according to the first information and an initial model; and the initial model is obtained by the terminal from the network device or pre-configured in the terminal. The embodiment of the application can reduce air interface overhead.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a model generation method, apparatus, device and readable storage medium. Background Technology

[0002] There are usually two methods for acquiring existing downlink channel data: (1) generated by a link-level simulation system; (2) measured and estimated by the terminal, and then periodically fed back the estimated results after statistical quantization.

[0003] Existing add-on AI (Artificial Intelligence) application models decouple AI model training and inference, allowing AI model validation to be performed only after the fact, lacking effective verification and assurance mechanisms. Currently, intelligent model training and iterative optimization are completed offline, resulting in poor real-time performance and a lack of direct correlation. When network performance metrics fall short of expectations after the model is deployed, only a "rollback" mechanism can mitigate the negative impact of the AI ​​model, which is inherently lagging.

[0004] Future networks based on intrinsic AI will need to incorporate digital twin simulation networks to enable pre-validation of models. However, digital twin simulation networks incur significant overhead when collecting data from the physical network, especially data related to the physical layer channels. Due to the small time granularity, this places higher demands on air interface transmission.

[0005] If terminal feedback is used to generate the data required for digital twin simulation networks, the channel data fed back by existing terminals (such as CQI (Channel Quality Indicator) / PMI (Precoding Matrix Indicator) / RI (Rank Indicator)) is statistically quantized data, and the reporting period is relatively long, which cannot meet the requirements of channel modeling. If the terminal is required to directly report complete channel data, such as the channel H matrix, the data volume is very large due to the small time granularity (milliseconds), resulting in high overhead on the air interface. Summary of the Invention

[0006] This application provides a model generation method, apparatus, device, and readable storage medium to reduce air interface overhead.

[0007] In a first aspect, embodiments of this application provide a model generation method applied to a network device, comprising:

[0008] Send first information to the terminal, the first information including information about the generator model and information about the discriminator model;

[0009] The terminal receives information about a first generator model, wherein the first generator model is generated by the terminal based on the first information and an initial model; the initial model is obtained by the terminal from the network device or pre-configured on the terminal.

[0010] Prior to receiving the information of the first generator model sent by the terminal, the method further includes:

[0011] The initial model is sent to the terminal, and the initial model includes: a generator initial model and a discriminator initial model.

[0012] The method further includes, after receiving the information of the first generator model sent by the terminal:

[0013] By aggregating multiple first generator models, information about the second generator model is obtained.

[0014] The method further includes, after aggregating the first generator model to obtain the information of the second generator model, the method further comprising:

[0015] The information of the second generator model is sent to the simulated wireless access network.

[0016] The step of sending the first information to the terminal includes:

[0017] The first information is sent to the terminal via control plane signaling;

[0018] The generator model information includes the generator's input dimension; the discriminator model information includes the discriminator's input dimension.

[0019] The generator model information also includes one or more of the following:

[0020] The generator's input, and the format of the generator's model file;

[0021] The information of the discriminator model also includes one or more of the following:

[0022] The input to the discriminator, and the format of the discriminator's model file.

[0023] Sending the initial model to the terminal includes any one of the following:

[0024] In the control plane signaling, the initial model is sent to the terminal via a container.

[0025] The initial model is sent to the terminal via user plane signaling.

[0026] When the initial model includes multiple fragments, the control plane signaling includes a first indication information and a second indication information, wherein the first indication information is used to indicate whether the fragment is the last fragment, and the second indication information is used to indicate the fragment number.

[0027] The user plane signaling packet header includes third indication information and / or fourth indication information; the third indication information is used to indicate that the user plane signaling is used to transmit an initial model; the fourth indication information is used to indicate the identifier of the transmission session.

[0028] When the initial model includes multiple fragments, the user plane signaling includes a fifth indication information and a sixth indication information, wherein the fifth indication information is used to indicate whether the fragment is the last fragment, and the sixth indication information is used to indicate the fragment number.

[0029] The step of aggregating multiple first generator models to obtain information about the second generator model includes:

[0030] Multiple intermediate data are obtained through multiple first generator models, wherein the input of each first generator model is a first random sequence corresponding to each first generator model;

[0031] Multiple intermediate data points are used as input to the discriminator to be trained, and a second random sequence is used as input to the generator to be trained. The generator and the discriminator to be trained are trained iteratively to obtain information about the second generator model.

[0032] The method further includes, after obtaining multiple intermediate data through multiple first generator models:

[0033] Preprocess multiple intermediate data;

[0034] The step of using multiple intermediate data points as input to the discriminator to be trained includes:

[0035] Multiple preprocessed intermediate data are used as input to the discriminator to be trained.

[0036] The step of sending the information of the second generator model to the simulated wireless access network includes:

[0037] Send the metafile information of the second generator model and / or the model file of the second generator model to the simulated wireless access network;

[0038] The metafile information of the second generator model includes one or more of the following: the input dimension of the second generator model, and the output dimension of the second generator model.

[0039] The metafile information of the second generator model also includes one or more of the following:

[0040] The network architecture description of the second generator model; the format of the model file of the second generator model; weights; gradients.

[0041] Secondly, embodiments of this application provide a model generation method applied to a terminal, including:

[0042] Receive first information sent by the network device, the first information including information of the generator model and information of the discriminator model;

[0043] Generate a first generator model based on the first information and the initial model;

[0044] The information of the first generator model is sent to the network device.

[0045] The initial model is obtained by the terminal from the network device or pre-configured on the terminal; the initial model includes a generator initial model and a discriminator initial model.

[0046] The first information sent by the receiving network device includes:

[0047] Receive the first information sent by the network device via control plane signaling;

[0048] The generator model information includes the generator's input dimension; the discriminator model information includes the discriminator's input dimension.

[0049] The generator model information also includes one or more of the following:

[0050] The generator's input, and the format of the generator's model file;

[0051] The information of the discriminator model also includes one or more of the following:

[0052] The input to the discriminator, and the format of the discriminator's model file.

[0053] The initial model sent by the network device is received through one or more of the following methods:

[0054] Receive the initial model sent by the network device via control plane signaling;

[0055] Receive the initial model sent by the network device via user plane signaling.

[0056] The step of generating the first generator model based on the first information and the initial model includes:

[0057] A first vector is generated based on the input dimension of the generator, wherein the dimension of the first vector is the same as the input dimension of the generator;

[0058] A second vector is generated based on the input dimension of the discriminator, wherein the dimension of the second vector is the same as the input dimension of the discriminator;

[0059] The first vector and the second vector are used as inputs to the generator to be trained and the discriminator to be trained, respectively, to obtain the first generator model.

[0060] The step of sending the information of the first generator model to the network device includes any one of the following:

[0061] In the control plane signaling, the information of the first generator model is sent to the network device via a container;

[0062] The information of the first generator model is sent to the network device via user plane signaling.

[0063] When the first generator model includes multiple fragments, the control plane signaling includes a seventh indication information and an eighth indication information, wherein the seventh indication information is used to indicate whether the fragment is the last fragment, and the eighth indication information is used to indicate the fragment number.

[0064] The user plane signaling packet header includes a ninth indication information and / or a tenth indication information; the ninth indication information is used to indicate that the user plane signaling is used to transmit an initial model; the tenth indication information is used to indicate the identifier of the transmission session.

[0065] When the first generator model includes multiple fragments, the user plane signaling includes an eleventh indication information and a twelfth indication information, wherein the eleventh indication information is used to indicate whether the fragment is the last fragment, and the twelfth indication information is used to indicate the fragment number.

[0066] Thirdly, embodiments of this application provide a data generation apparatus, applied to a network device, comprising:

[0067] The first sending module is used to send first information to the terminal, the first information including information of the generator model and information of the discriminator model;

[0068] The first receiving module is configured to receive information about a first generator model sent by the terminal, wherein the generator model is generated by the terminal based on the first information and an initial model; the initial model is obtained by the terminal from the network device or pre-configured on the terminal.

[0069] The device further includes:

[0070] The second sending module is used to send the initial model to the terminal. The initial model includes a generator initial model and a discriminator initial model.

[0071] The device further includes:

[0072] The first processing module is used to aggregate multiple first generator models to obtain information about the second generator model.

[0073] The device further includes:

[0074] The third sending module is used to send information about the second generator model to the simulated wireless access network.

[0075] The first sending module is used to send the first information to the terminal via control plane signaling;

[0076] The generator model information includes the generator's input dimension; the discriminator model information includes the discriminator's input dimension.

[0077] The generator model information also includes one or more of the following:

[0078] The generator's input, and the format of the generator's model file;

[0079] The information of the discriminator model also includes one or more of the following:

[0080] The input to the discriminator, and the format of the discriminator's model file.

[0081] The second sending module is used to send the initial model to the terminal via a container in the control plane signaling.

[0082] The initial model is sent to the terminal via user plane signaling.

[0083] When the initial model includes multiple fragments, the control plane signaling includes a first indication information and a second indication information, wherein the first indication information is used to indicate whether the fragment is the last fragment, and the second indication information is used to indicate the fragment number.

[0084] The user plane signaling packet header includes third indication information and / or fourth indication information; the third indication information is used to indicate that the user plane signaling is used to transmit an initial model; the fourth indication information is used to indicate the identifier of the transmission session.

[0085] When the initial model includes multiple fragments, the user plane signaling includes a fifth indication information and a sixth indication information, wherein the fifth indication information is used to indicate whether the fragment is the last fragment, and the sixth indication information is used to indicate the fragment number.

[0086] The first processing module includes:

[0087] The first acquisition submodule is used to obtain multiple intermediate data through multiple first generator models, wherein the input of each first generator model is a first random sequence corresponding to each first generator model;

[0088] The first processing submodule is used to take multiple intermediate data as input to the discriminator to be trained, take the second random sequence as input to the generator to be trained, and iteratively train the generator and the discriminator to be trained to obtain information of the second generator model.

[0089] The first processing module further includes:

[0090] The preprocessing submodule is used to preprocess multiple intermediate data.

[0091] The first processing submodule is also used to take the preprocessed intermediate data as input to the discriminator to be trained.

[0092] The third sending module is used to send the meta-file information of the second generator model and / or the model file of the second generator model to the simulated wireless access network.

[0093] The metafile information of the second generator model includes one or more of the following: the input dimension of the second generator model, and the output dimension of the second generator model.

[0094] The metafile information of the second generator model also includes one or more of the following:

[0095] The network architecture description of the second generator model; the format of the model file of the second generator model; weights; gradients.

[0096] Fourthly, embodiments of this application provide a data generation apparatus applied to a terminal, comprising:

[0097] The first receiving module is used to receive first information sent by the network device, the first information including information of the generator model and information of the discriminator model;

[0098] The first generation module is used to generate a first generator model based on the first information and the initial model;

[0099] The first sending module is used to send information about the first generator model to the network device.

[0100] The initial model is obtained by the terminal from the network device or pre-configured on the terminal; the initial model includes a generator initial model and a discriminator initial model.

[0101] The first receiving module is used to receive the first information sent by the network device through control plane signaling;

[0102] The generator model information includes the generator's input dimension; the discriminator model information includes the discriminator's input dimension.

[0103] The generator model information also includes one or more of the following:

[0104] The generator's input, and the format of the generator's model file;

[0105] The information of the discriminator model also includes one or more of the following:

[0106] The input to the discriminator, and the format of the discriminator's model file.

[0107] The initial model sent by the network device is received through one or more of the following methods:

[0108] Receive the initial model sent by the network device via control plane signaling;

[0109] Receive the initial model sent by the network device via user plane signaling.

[0110] The first generation module includes:

[0111] The first generation submodule is used to generate a first vector based on the input dimension of the generator, wherein the dimension of the first vector is the same as the input dimension of the generator;

[0112] The second generation submodule is used to generate a second vector based on the input dimension of the discriminator, wherein the dimension of the second vector is the same as the input dimension of the discriminator;

[0113] The third generation submodule is used to take the first vector and the second vector as inputs to the generator to be trained and the discriminator to be trained, respectively, to obtain the first generator model.

[0114] The first sending module is configured to perform any one of the following:

[0115] In the control plane signaling, the information of the first generator model is sent to the network device via a container;

[0116] The information of the first generator model is sent to the network device via user plane signaling.

[0117] When the first generator model includes multiple fragments, the control plane signaling includes a seventh indication information and an eighth indication information, wherein the seventh indication information is used to indicate whether the fragment is the last fragment, and the eighth indication information is used to indicate the fragment number.

[0118] The user plane signaling packet header includes a ninth indication information and / or a tenth indication information; the ninth indication information is used to indicate that the user plane signaling is used to transmit an initial model; the tenth indication information is used to indicate the identifier of the transmission session.

[0119] When the first generator model includes multiple fragments, the user plane signaling includes an eleventh indication information and a twelfth indication information, wherein the eleventh indication information is used to indicate whether the fragment is the last fragment, and the twelfth indication information is used to indicate the fragment number.

[0120] Fifthly, embodiments of this application provide a data generation apparatus applied to a network device, including: a processor and a transceiver;

[0121] The transceiver is configured to send first information to the terminal, the first information including generator model information and discriminator model information; and to receive first generator model information sent by the terminal, wherein the first generator model is generated by the terminal based on the first information and an initial model; the initial model is obtained by the terminal from the network device or pre-configured on the terminal.

[0122] The transceiver is further configured to send the initial model to the terminal, the initial model including: a generator initial model and a discriminator initial model.

[0123] The processor is further configured to aggregate multiple first generator models to obtain information about a second generator model.

[0124] The transceiver is also used to send information about the second generator model to a simulated wireless access network.

[0125] The transceiver is also used for:

[0126] The first information is sent to the terminal via control plane signaling;

[0127] The generator model information includes the generator's input dimension; the discriminator model information includes the discriminator's input dimension.

[0128] The generator model information also includes one or more of the following:

[0129] The generator's input, and the format of the generator's model file;

[0130] The information of the discriminator model also includes one or more of the following:

[0131] The input to the discriminator, and the format of the discriminator's model file.

[0132] The transceiver is also used to perform any of the following:

[0133] In the control plane signaling, the initial model is sent to the terminal via a container.

[0134] The initial model is sent to the terminal via user plane signaling.

[0135] When the initial model includes multiple fragments, the control plane signaling includes a first indication information and a second indication information, wherein the first indication information is used to indicate whether the fragment is the last fragment, and the second indication information is used to indicate the fragment number.

[0136] The user plane signaling packet header includes third indication information and / or fourth indication information; the third indication information is used to indicate that the user plane signaling is used to transmit an initial model; the fourth indication information is used to indicate the identifier of the transmission session.

[0137] When the initial model includes multiple fragments, the user plane signaling includes a fifth indication information and a sixth indication information, wherein the fifth indication information is used to indicate whether the fragment is the last fragment, and the sixth indication information is used to indicate the fragment number.

[0138] The processor is further configured to:

[0139] Multiple intermediate data are obtained through multiple first generator models, wherein the input of each first generator model is a first random sequence corresponding to each first generator model;

[0140] Multiple intermediate data points are used as input to the discriminator to be trained, and a second random sequence is used as input to the generator to be trained. The generator and the discriminator to be trained are trained iteratively to obtain information about the second generator model.

[0141] The processor is further configured to:

[0142] Preprocess multiple intermediate data;

[0143] Multiple preprocessed intermediate data are used as input to the discriminator to be trained.

[0144] The transceiver is also used for:

[0145] Send the metafile information of the second generator model and / or the model file of the second generator model to the simulated wireless access network;

[0146] The metafile information of the second generator model includes one or more of the following: the input dimension of the second generator model, and the output dimension of the second generator model.

[0147] The metafile information of the second generator model also includes one or more of the following:

[0148] The network architecture description of the second generator model; the format of the model file of the second generator model; weights; gradients.

[0149] Sixthly, embodiments of this application provide a data generation apparatus applied to a terminal, comprising: a processor and a transceiver;

[0150] The transceiver is used to receive first information sent by the network device, the first information including information of the generator model and information of the discriminator model;

[0151] The processor is configured to generate a first generator model based on the first information and the initial model;

[0152] The transceiver is also used to send information about the first generator model to the network device.

[0153] The initial model is obtained by the terminal from the network device or pre-configured on the terminal; the initial model includes a generator initial model and a discriminator initial model.

[0154] The transceiver is further configured to: receive the first information sent by the network device via control plane signaling;

[0155] The generator model information includes the generator's input dimension; the discriminator model information includes the discriminator's input dimension.

[0156] The generator model information also includes one or more of the following:

[0157] The generator's input, and the format of the generator's model file;

[0158] The information of the discriminator model also includes one or more of the following:

[0159] The input to the discriminator, and the format of the discriminator's model file.

[0160] The transceiver is further configured to receive the initial model sent by the network device in one or more of the following ways:

[0161] Receive the initial model sent by the network device via control plane signaling;

[0162] Receive the initial model sent by the network device via user plane signaling.

[0163] The processor is further configured to:

[0164] A first vector is generated based on the input dimension of the generator, wherein the dimension of the first vector is the same as the input dimension of the generator;

[0165] A second vector is generated based on the input dimension of the discriminator, wherein the dimension of the second vector is the same as the input dimension of the discriminator;

[0166] The first vector and the second vector are used as inputs to the generator to be trained and the discriminator to be trained, respectively, to obtain the first generator model.

[0167] The transceiver is also used to perform any of the following:

[0168] In the control plane signaling, the information of the first generator model is sent to the network device via a container;

[0169] The information of the first generator model is sent to the network device via user plane signaling.

[0170] When the first generator model includes multiple fragments, the control plane signaling includes a seventh indication information and an eighth indication information, wherein the seventh indication information is used to indicate whether the fragment is the last fragment, and the eighth indication information is used to indicate the fragment number.

[0171] The user plane signaling packet header includes a ninth indication information and / or a tenth indication information; the ninth indication information is used to indicate that the user plane signaling is used to transmit an initial model; the tenth indication information is used to indicate the identifier of the transmission session.

[0172] When the first generator model includes multiple fragments, the user plane signaling includes an eleventh indication information and a twelfth indication information, wherein the eleventh indication information is used to indicate whether the fragment is the last fragment, and the twelfth indication information is used to indicate the fragment number.

[0173] In a seventh aspect, embodiments of this application provide a communication device, including: a transceiver, a memory, a processor, and a program stored in the memory and executable on the processor; the processor is configured to read the program in the memory to implement the steps in the model generation method described above.

[0174] Eighthly, embodiments of this application provide a readable storage medium for storing a program that, when executed by a processor, implements the steps in the model generation method described above.

[0175] In this embodiment, the network device sends first information to the terminal, and the terminal generates a first generator model based on the first information and an initial model, and then sends it to the network device. As can be seen from the above analysis, in this embodiment, the terminal does not need to upload a large amount of data; it only needs to upload the generator model. Therefore, the solution using this embodiment can reduce air interface overhead. Attached Figure Description

[0176] Figure 1 This is one of the flowcharts of the model generation method provided in the embodiments of this application;

[0177] Figure 2 This is a schematic diagram of the first information in the embodiments of this application;

[0178] Figure 3 This is a schematic diagram of the RRC signaling for transmitting the initial model in an embodiment of this application;

[0179] Figure 4 This is a schematic diagram illustrating the extension of SDAP in the embodiments of this application;

[0180] Figure 5 This is a schematic diagram illustrating the process by which a network device aggregates generator models uploaded from multiple terminals in an embodiment of this application.

[0181] Figure 6 This is the second flowchart of the model generation method provided in the embodiments of this application;

[0182] Figure 7 This is the third flowchart of the model generation method provided in the embodiments of this application;

[0183] Figure 8 This is a schematic diagram of the generator processing provided in the embodiments of this application;

[0184] Figure 9 This is a schematic diagram of the processing of the discriminator provided in the embodiments of this application;

[0185] Figure 10 This is a schematic diagram illustrating the effect of an embodiment of this application;

[0186] Figure 11This is one of the structural diagrams of the data generation apparatus provided in the embodiments of this application;

[0187] Figure 12 This is a second structural diagram of the data generation apparatus provided in the embodiments of this application;

[0188] Figure 13 This is the third structural diagram of the data generation apparatus provided in the embodiments of this application;

[0189] Figure 14 This is the fourth structural diagram of the data generation device provided in the embodiments of this application. Detailed Implementation

[0190] In the embodiments of this application, the term "and / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0191] In the embodiments of this application, the term "multiple" refers to two or more, and other quantifiers are similar.

[0192] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0193] See Figure 1 , Figure 1 This is a flowchart of the model generation method provided in the embodiments of this application, applied to network devices, such as... Figure 1 As shown, it includes the following steps:

[0194] Step 101: Send the first message to the terminal.

[0195] The first information is used to enable the terminal to generate a generator model and a discriminator model. This information may include: information about the generator model and information about the discriminator model. Specifically, the generator model information includes the generator's input dimensions; the discriminator model information includes the discriminator's input dimensions. Further, to improve the efficiency of the terminal's model generation, the generator model information may also include one or more of the following: the generator's input and the format of the generator's model file. The discriminator model information may also include one or more of the following: the discriminator's input and the format of the discriminator's model file.

[0196] The input dimensions of the generator include:

[0197] Algorithms for generating random input variables: such as the normal distribution N(0,1), Bernoulli distribution, etc.

[0198] LatentDimension of random input variables: for example, 100.

[0199] After the terminal obtains the generator's input dimension configuration, it generates a vector with dimension 100 as the generator's input according to the configured random input algorithm.

[0200] The input dimensions of the discriminator include: the number of antenna ports M and the number of taps based on multipath delay N. After obtaining the input configuration of the discriminator, the terminal obtains the M*N channel impulse response matrix as the input of the discriminator.

[0201] The mode of the generative model is used to indicate the mode of the model. For example, Generative Adversarial Network (GAN) can be used as the default mode 1, but other modes can also be used.

[0202] The generator's model file or the discriminator's model file can be in formats such as h5 or pkl.

[0203] like Figure 2 The diagram shown is a schematic representation of the first information. For example, the first information, RadioChannelModelGenerationConfig, may include: ModelMode (the mode for generating the model), GeneratorInputShape (the input dimension of the generator), DiscriminatorInputShape (the input dimension of the discriminator), and ModelFormat (the format of the model file).

[0204] In this embodiment, the first information can be used for generating channel models, generating image processing models, etc. The network device can send the first information to the terminal via control plane signaling. The control plane signaling can be, for example, an RRC (Radio Resource Control) reconfiguration message.

[0205] Step 102: Receive information about a first generator model sent by the terminal, wherein the first generator model is generated by the terminal based on the first information and the initial model; the initial model is obtained by the terminal from the network device or pre-configured on the terminal.

[0206] The information of the first generator model includes one or more of the following: the structure of the first generator model, the model file, the input dimension, the output dimension, the weights, the gradient, etc.

[0207] After receiving the initial information from the network device, the terminal starts training the GAN model, including the generator model G and the discriminator model D, by combining the initial generator model and the initial discriminator model, until Nash equilibrium is reached. From this point, the obtained generator model G can be used as the first generator model.

[0208] In this step, the network device may also receive a first generator model sent by the terminal via control plane signaling (such as RRC signaling) or user plane signaling. This first generator model is used for generating channel models, image processing models, etc.

[0209] In this embodiment, the network device sends first information to the terminal, and the terminal generates a first generator model based on the first information and an initial model, and then sends it to the network device. As can be seen from the above analysis, in this embodiment, the terminal does not need to upload a large amount of data; it only needs to upload the generator model. Therefore, the solution using this embodiment can reduce air interface overhead.

[0210] Based on the above embodiments, in order to improve the efficiency of obtaining the generator model, before step 102, the network device may also send an initial model to the terminal. The initial model includes: a generator initial model and a discriminator initial model.

[0211] In this step, the network device can send the initial model to the terminal in at least two of the following ways.

[0212] Method 1: In the control plane signaling, the initial model is sent to the terminal via a container.

[0213] When the initial model includes multiple fragments, the control plane signaling includes a first indication information and a second indication information, wherein the first indication information is used to indicate whether the fragment is the last fragment, and the second indication information is used to indicate the fragment number.

[0214] For example, network devices can transmit the initial model via a container in RRC signaling. If the model file is large, it can be divided into different fragments for transmission. The SegmentType indicates whether a fragment is the last fragment, the SegmentNumber indicates the fragment number, and the container is used to pass the generator and discriminator models.

[0215] like Figure 3 The diagram shows a schematic of the RRC signaling for transmitting the initial model. The RadioChannelModel includes: SegmentType, SegmentNumber, and container.

[0216] Method 2: Send the initial model to the terminal via user plane signaling.

[0217] In this embodiment, the transmission of model data packets is indicated by extending the user plane header. The user plane signaling header includes third and / or fourth indication information; the third indication information indicates that the user plane signaling is used to transmit an initial model; the fourth indication information indicates the identifier of the transmission session. When the initial model includes multiple fragments, the user plane signaling includes fifth and sixth indication information, wherein the fifth indication information indicates whether the fragment is the last fragment, and the sixth indication information indicates the fragment number.

[0218] In practical applications, new fields can be added to the header of the user plane signaling packet to carry third and fourth indication information, respectively. For example, Figure 4 As shown, network devices can indicate model transmission by extending the SDAP (Service Data Adaptation Protocol) packet header and adding a TransactionID to specify the identifier of this transmission session. If the model data is large, the SegmentNum indicates the fragment number, and the EndMarker indicates whether the current fragment is the last packet or fragment of the model transmission.

[0219] Based on the above embodiments, after step 103, the method may further include: aggregating multiple first generator models to obtain information about a second generator model.

[0220] The information of the second generator model includes one or more of the following: metafile information, model file of the second generator model; the metafile information of the second generator model includes one or more of the following: network architecture description of the second generator model, input dimension, output dimension, weights, gradients of the second generator model.

[0221] After receiving the first generator model uploaded by multiple terminals, the network device inputs a random sequence into the first generator model to generate corresponding intermediate data. It then uses a two-layer GAN method to aggregate the models and retrain new generator and discriminator models (G, D). By aggregating multiple channel models using a two-layer GAN method, the network device reduces the transmission overhead of the data models.

[0222] Specifically, the network device obtains multiple intermediate data points through multiple first generator models, where the input to each first generator model is a first random sequence corresponding to that model. The multiple intermediate data points are used as input to a discriminator to be trained, and the second random sequence is used as input to a generator to be trained. The generator and discriminator are iteratively trained to obtain information about the second generator model. The second random sequence can be any random sequence arbitrarily selected by the network device.

[0223] In the above process, to improve data processing efficiency, multiple intermediate data points can be preprocessed. Then, these preprocessed intermediate data points are used as input to the discriminator to be trained.

[0224] The preprocessing may include data optimization processes such as data filtering and deletion. Taking channel model generation as an example, network devices can obtain terminal location and speed information, as well as channel quality measurement information. During preprocessing, the network device can cluster users with similar locations or channels, and select channel data representing typical users within the same cluster as input to the discriminator. Alternatively, it can calculate the sample distribution of multi-user channel generation data, filtering channel data samples with similar characteristics to reduce the processing overhead of the channel discriminator. Mainstream algorithms for calculating sample distribution include, but are not limited to:

[0225] (1) MMD (maximize mean discrepancy): The distributions P1 and P2 are mapped to a high-dimensional Hilbert space through a kernel function to obtain high-dimensional features, and then the distance between them is calculated.

[0226] (2) Bulldozer distance: The cost required to move a soil pile distribution P1 to a soil pile distribution P2 under path planning. The bulldozer distance is the minimum cost under optimal path planning.

[0227] Considering that the computing power of the wireless access network is stronger than that of the terminal, the design of the channel generator model and the channel discriminator model takes into account the use of a deeper network and a larger space of random input variables.

[0228] like Figure 5 As shown, taking the channel model as an example, multiple channel generator models (such as G1, G2, G3, etc.) are received from multiple terminals. The network device inputs random sequences (random sequence 1, random sequence 2, ...) into the channel generator model to generate corresponding channel data (channel data 1, channel data 2, channel data 3). The input of the generator is a random sequence N. A two-layer GAN method is used to aggregate the model and retrain a new generator and discriminator model (G, D).

[0229] Based on the above, the network device can also send information about the second generator model to the simulated wireless access network.

[0230] Specifically, the network device can send the metafile information of the second generator model and / or the model file of the second generator model to the simulated wireless access network; wherein, the metafile information of the second generator model includes one or more of the following: the input dimension of the second generator model, and the output dimension of the second generator model. Further, to improve the processing efficiency of the simulated network, the metafile information of the second generator model also includes one or more of the following: a network architecture description of the second generator model; the format of the model file of the second generator model; weights; and gradients.

[0231] The input dimensions of the second generator model include:

[0232] Algorithms for generating random input variables: such as the normal distribution N(0,1), Bernoulli distribution, etc.

[0233] LatentDimension of random input variables: for example, 100.

[0234] The output dimensions of the second generator model include: the number of antenna ports M and the number of taps based on multipath delay N. The simulated wireless access network generates an M*N channel impulse response matrix through the channel generator model as input to the simulated channel data.

[0235] The second generator model includes a description of its network architecture, such as descriptions of convolutional, fully connected, and pooling network structures. The model file format for the second generator model can be, for example, h5 or pkl.

[0236] The model file of the second generator model can be transmitted via a binary bitstream file.

[0237] See Figure 6 , Figure 6 This is a flowchart of the model generation method provided in the embodiments of this application, applied to a terminal, such as... Figure 6 As shown, it includes the following steps:

[0238] Step 601: Receive first information sent by the network device, the first information including information of the generator model and information of the discriminator model.

[0239] The terminal receives the first information sent by the network device via control plane signaling (such as RRC signaling). The meaning of the first information can be found in the description of the foregoing embodiments.

[0240] Step 602: Generate the first generator model based on the first information and the initial model.

[0241] The initial model is obtained by the terminal from the network device or pre-configured on the terminal; the initial model includes a generator initial model and a discriminator initial model.

[0242] In this step, the terminal can receive the initial model sent by the network device in at least two of the following ways:

[0243] (1) Receive the initial model sent by the network device through control plane signaling, the initial model including generator initial model and discriminator initial model.

[0244] (2) Receive the initial model sent by the network device through user plane signaling, the initial model including generator initial model and discriminator initial model.

[0245] The content and meaning of the user plane signaling and control plane signaling can be referred to the description in the foregoing embodiments.

[0246] Specifically, in this step, the terminal generates a first vector based on the input dimension of the generator, the first vector having the same dimension as the input dimension of the generator; it generates a second vector based on the input dimension of the discriminator, the second vector having the same dimension as the input dimension of the discriminator; the first vector and the second vector are used as the inputs to the generator to be trained and the discriminator to be trained, respectively, until a Nash equilibrium is reached, thus obtaining the first generator model.

[0247] Step 603: Send the information of the first generator model to the network device.

[0248] The information of the first generator model includes one or more of the following: the structure of the first generator model, the model file, the input dimension, the output dimension, the weights, the gradient, etc.

[0249] In this step, the terminal can send the information of the first generator model to the network device in at least two of the following ways.

[0250] (1) In the control plane signaling, the information of the first generator model is sent to the network device in the form of a container. When the first generator model includes multiple fragments, the control plane signaling includes a seventh indication information and an eighth indication information, wherein the seventh indication information is used to indicate whether the fragment is the last fragment, and the eighth indication information is used to indicate the fragment number.

[0251] For example, the terminal can send information about the first generator model to the network device via RRC signaling.

[0252] (2) Send the information of the first generator model to the network device via user plane signaling.

[0253] For example, the terminal can send information about the first generator model by extending the SDAP packet header. The user plane signaling packet header includes a ninth indication and / or a tenth indication; the ninth indication indicates that the user plane signaling is used to transmit the initial model; the tenth indication indicates the identifier of the transmission session. When the first generator model includes multiple fragments, the user plane signaling includes an eleventh indication and a twelfth indication, wherein the eleventh indication indicates whether the fragment is the last fragment, and the twelfth indication indicates the fragment number.

[0254] In this embodiment, the network device sends first information to the terminal, and the terminal generates a first generator model based on the first information and an initial model, and then sends it to the network device. As can be seen from the above analysis, in this embodiment, the terminal does not need to upload a large amount of data; it only needs to upload the generator model. Therefore, the solution using this embodiment can reduce air interface overhead.

[0255] It should be noted that, in the embodiments of this application, the information of the generator model is not limited to the content listed in the examples above, and can be extended as needed.

[0256] See Figure 7 , Figure 7 This is a flowchart of the model generation method provided in the embodiments of this application. Figure 7 The following steps can be included in the description of generating a channel model:

[0257] Step 701: The physical wireless access network (e.g., a base station) transmits channel first information to the terminal via RRC signaling. This information includes generator model information and discriminator model information. In this embodiment, the following is taken as an example:

[0258] (1) The input dimension of the generator model G, for example, the random sequence of input to the generator G (e.g., conforming to a normal distribution N(0,1)).

[0259] (2) Channel data input dimensions and format of discriminator model D: for example, based on the frequency impulse response H matrix of the antenna port and time-frequency domain.

[0260] (3) Model format of generator G and discriminator D.

[0261] Step 702: The physical wireless access network transmits the initial model of the channel (G generator initial model and D discriminator initial model) through RRC or user plane.

[0262] Step 703: After receiving the first information from the physical wireless access network, the terminal collects channel data and starts training the GAN model, including training the generator G model and the discriminator D model, until Nash equilibrium is reached, resulting in the trained generator G model and discriminator D model. The generator G model is used for channel data generation.

[0263] Step 704: The terminal uploads the information of the generated generator G model via RRC signaling or user plane data, including one or more of the following: generator model structure, model file, input dimension, output dimension, weights, and gradients.

[0264] Step 705: After the physical wireless access network (e.g., a base station) receives generator G models from multiple terminals, it aggregates the models using the following method:

[0265] Each user's corresponding random sequence is input into the corresponding generator G model to generate a certain amount of channel data. This channel data is then used as real sample data and input into a new GAN network. Additionally, random sequences are input into the generator G network to be trained to generate fake samples, which are then distinguished from real samples by a discriminator D. The generator G network and discriminator D network are iterated until the network reaches Nash equilibrium. This yields the aggregated generator G model and discriminator D model.

[0266] Step 706: The physical wireless access network (e.g., a base station) uploads the information of the aggregated generator G model to the simulation network, including one or more of the following: the structure of the generator model, the model file, the input dimension, the output dimension, the weights, and the gradient.

[0267] Step 707: Simulate the wireless access network by generating data and a simulation environment based on the generator G model.

[0268] In the above embodiments, considering the potential computing power limitations of ordinary terminals, it is advisable to use drive test terminals for model data collection and training.

[0269] The following is an example of using a GAN model to generate channel data:

[0270] (1) The generator model uses fully connected layers + 2 upsampling layers + 3 convolutional layers to generate data (e.g. Figure 8 (As shown).

[0271] (2) The discriminator uses a three-layer convolutional + fully connected approach to identify real and fake data (e.g., Figure 9 (As shown).

[0272] like Figure 10 As shown, after 500 training iterations, the distribution deviation (Wasserstein distance) between the real channel data and the data generated by the generator model gradually decreased. Therefore, the GAN method can effectively generate data with a distribution similar to the actual channel data.

[0273] In the above embodiments, since the terminal only needs to transmit the channel data generation model, the overhead of the terminal uploading channel data can be reduced. Furthermore, the network aggregates multiple channel models using a two-layer GAN method, further reducing the transmission overhead of the data model.

[0274] It should be noted that, in the embodiments of this application, the information of the generator model is not limited to the content listed in the examples above, and can be extended as needed.

[0275] This application also provides a data generation apparatus, applied to a network device. See also... Figure 11 , Figure 11 This is a structural diagram of the data generation apparatus provided in the embodiments of this application. Since the principle of the data generation apparatus in solving the problem is similar to that of the model generation method in the embodiments of this application, the implementation of this data generation apparatus can refer to the implementation of the method, and the repeated parts will not be described again.

[0276] like Figure 11 As shown, the data generation device 1100 includes:

[0277] The first sending module 1101 is used to send first information to the terminal, the first information including generator model information and discriminator model information; the first receiving module 1102 is used to receive the first generator model information sent by the terminal, wherein the generator model is generated by the terminal based on the first information and an initial model; the initial model is obtained by the terminal from the network device or pre-configured on the terminal.

[0278] The device further includes:

[0279] The second sending module is used to send the initial model to the terminal. The initial model includes a generator initial model and a discriminator initial model.

[0280] The device further includes:

[0281] The first processing module is used to aggregate multiple first generator models to obtain information about the second generator model.

[0282] The device further includes:

[0283] The third sending module is used to send information about the second generator model to the simulated wireless access network.

[0284] The first sending module is used to send the first information to the terminal via control plane signaling;

[0285] The generator model information includes the generator's input dimension; the discriminator model information includes the discriminator's input dimension.

[0286] The generator model information also includes one or more of the following:

[0287] The generator's input, and the format of the generator's model file;

[0288] The information of the discriminator model also includes one or more of the following:

[0289] The input to the discriminator, and the format of the discriminator's model file.

[0290] The second sending module is used to send the initial model to the terminal via a container in the control plane signaling.

[0291] The initial model is sent to the terminal via user plane signaling.

[0292] When the initial model includes multiple fragments, the control plane signaling includes a first indication information and a second indication information, wherein the first indication information is used to indicate whether the fragment is the last fragment, and the second indication information is used to indicate the fragment number.

[0293] The user plane signaling packet header includes third indication information and / or fourth indication information; the third indication information is used to indicate that the user plane signaling is used to transmit an initial model; the fourth indication information is used to indicate the identifier of the transmission session.

[0294] When the initial model includes multiple fragments, the user plane signaling includes a fifth indication information and a sixth indication information, wherein the fifth indication information is used to indicate whether the fragment is the last fragment, and the sixth indication information is used to indicate the fragment number.

[0295] The first processing module includes:

[0296] The first acquisition submodule is used to obtain multiple intermediate data through multiple first generator models, wherein the input of each first generator model is a first random sequence corresponding to each first generator model;

[0297] The first processing submodule is used to take multiple intermediate data as input to the discriminator to be trained, take the second random sequence as input to the generator to be trained, and iteratively train the generator and the discriminator to be trained to obtain information of the second generator model.

[0298] The first processing module further includes:

[0299] The preprocessing submodule is used to preprocess multiple intermediate data.

[0300] The first processing submodule is also used to take the preprocessed intermediate data as input to the discriminator to be trained.

[0301] The third sending module is used to send the meta-file information of the second generator model and / or the model file of the second generator model to the simulated wireless access network.

[0302] The metafile information of the second generator model includes one or more of the following: the input dimension of the second generator model, and the output dimension of the second generator model.

[0303] The metafile information of the second generator model also includes one or more of the following:

[0304] The network architecture description of the second generator model; the format of the model file of the second generator model; weights; gradients.

[0305] The apparatus provided in this application embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0306] This application also provides a data generation apparatus for use in a terminal. See also... Figure 12 , Figure 12 This is a structural diagram of the data generation apparatus provided in the embodiments of this application. Since the principle of the data generation apparatus in solving the problem is similar to that of the model generation method in the embodiments of this application, the implementation of this data generation apparatus can refer to the implementation of the method, and the repeated parts will not be described again.

[0307] like Figure 12 As shown, the data generation device 1200 includes:

[0308] The first receiving module 1201 is used to receive first information sent by the network device, the first information including information of the generator model and information of the discriminator model; the first generating module 1202 is used to generate a first generator model according to the first information and the initial model; the first sending module 1203 is used to send the information of the first generator model to the network device.

[0309] The initial model is obtained by the terminal from the network device or pre-configured on the terminal; the initial model includes a generator initial model and a discriminator initial model.

[0310] The first receiving module is used to receive the first information sent by the network device through control plane signaling;

[0311] The generator model information includes the generator's input dimension; the discriminator model information includes the discriminator's input dimension.

[0312] The generator model information also includes one or more of the following:

[0313] The generator's input, and the format of the generator's model file;

[0314] The information of the discriminator model also includes one or more of the following:

[0315] The input to the discriminator, and the format of the discriminator's model file.

[0316] The initial model sent by the network device is received through one or more of the following methods:

[0317] Receive the initial model sent by the network device via control plane signaling;

[0318] Receive the initial model sent by the network device via user plane signaling.

[0319] The first generation module includes:

[0320] The first generation submodule is used to generate a first vector based on the input dimension of the generator, wherein the dimension of the first vector is the same as the input dimension of the generator;

[0321] The second generation submodule is used to generate a second vector based on the input dimension of the discriminator, wherein the dimension of the second vector is the same as the input dimension of the discriminator;

[0322] The third generation submodule is used to take the first vector and the second vector as inputs to the generator to be trained and the discriminator to be trained, respectively, to obtain the first generator model.

[0323] The first sending module is configured to perform any one of the following:

[0324] In the control plane signaling, the information of the first generator model is sent to the network device via a container;

[0325] The information of the first generator model is sent to the network device via user plane signaling.

[0326] When the first generator model includes multiple fragments, the control plane signaling includes a seventh indication information and an eighth indication information, wherein the seventh indication information is used to indicate whether the fragment is the last fragment, and the eighth indication information is used to indicate the fragment number.

[0327] The user plane signaling packet header includes a ninth indication information and / or a tenth indication information; the ninth indication information is used to indicate that the user plane signaling is used to transmit an initial model; the tenth indication information is used to indicate the identifier of the transmission session.

[0328] When the first generator model includes multiple fragments, the user plane signaling includes an eleventh indication information and a twelfth indication information, wherein the eleventh indication information is used to indicate whether the fragment is the last fragment, and the twelfth indication information is used to indicate the fragment number.

[0329] The apparatus provided in this application embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0330] It should be noted that the division of units in the embodiments of this application is illustrative and only represents one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units.

[0331] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a processor-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0332] like Figure 13 As shown, this application provides a data generation apparatus for use in network devices, including a processor 1301 and a transceiver 1302.

[0333] The transceiver 1302 is configured to send first information to the terminal, the first information including generator model information and discriminator model information; and receive the first generator model information sent by the terminal, wherein the first generator model is generated by the terminal based on the first information and an initial model; the initial model is obtained by the terminal from the network device or pre-configured on the terminal.

[0334] The transceiver 1302 is further configured to send the initial model to the terminal, the initial model including: a generator initial model and a discriminator initial model.

[0335] The processor 1301 is further configured to aggregate multiple first generator models to obtain information about a second generator model.

[0336] The transceiver 1302 is further configured to send information about the second generator model to a simulated wireless access network.

[0337] The transceiver 1302 is further used for:

[0338] The first information is sent to the terminal via control plane signaling;

[0339] The generator model information includes the generator's input dimension; the discriminator model information includes the discriminator's input dimension.

[0340] The generator model information also includes one or more of the following:

[0341] The generator's input, and the format of the generator's model file;

[0342] The information of the discriminator model also includes one or more of the following:

[0343] The input to the discriminator, and the format of the discriminator's model file.

[0344] The transceiver 1302 is also used to perform any of the following:

[0345] In the control plane signaling, the initial model is sent to the terminal via a container.

[0346] The initial model is sent to the terminal via user plane signaling.

[0347] When the initial model includes multiple fragments, the control plane signaling includes a first indication information and a second indication information, wherein the first indication information is used to indicate whether the fragment is the last fragment, and the second indication information is used to indicate the fragment number.

[0348] The user plane signaling packet header includes third indication information and / or fourth indication information; the third indication information is used to indicate that the user plane signaling is used to transmit an initial model; the fourth indication information is used to indicate the identifier of the transmission session.

[0349] When the initial model includes multiple fragments, the user plane signaling includes a fifth indication information and a sixth indication information, wherein the fifth indication information is used to indicate whether the fragment is the last fragment, and the sixth indication information is used to indicate the fragment number.

[0350] The processor 1301 is further configured to:

[0351] Multiple intermediate data are obtained through multiple first generator models, wherein the input of each first generator model is a first random sequence corresponding to each first generator model;

[0352] Multiple intermediate data points are used as input to the discriminator to be trained, and a second random sequence is used as input to the generator to be trained. The generator and the discriminator to be trained are trained iteratively to obtain information about the second generator model.

[0353] The processor 1301 is further configured to:

[0354] Preprocess multiple intermediate data;

[0355] Multiple preprocessed intermediate data are used as input to the discriminator to be trained.

[0356] The transceiver 1302 is further used for:

[0357] Send the metafile information of the second generator model and / or the model file of the second generator model to the simulated wireless access network;

[0358] The metafile information of the second generator model includes one or more of the following: the input dimension of the second generator model, and the output dimension of the second generator model.

[0359] The metafile information of the second generator model also includes one or more of the following:

[0360] The network architecture description of the second generator model; the format of the model file of the second generator model; weights; gradients.

[0361] The apparatus provided in this application embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0362] like Figure 14 As shown, this application embodiment provides a data generation device applied to a terminal, including: a processor 1401 and a transceiver 1402;

[0363] The transceiver 1402 is used to receive first information sent by the network device, the first information including information of the generator model and information of the discriminator model;

[0364] The processor 1401 is configured to generate a first generator model based on the first information and the initial model;

[0365] The transceiver 1402 is also used to send information about the first generator model to the network device.

[0366] The initial model is obtained by the terminal from the network device or pre-configured on the terminal; the initial model includes a generator initial model and a discriminator initial model.

[0367] The transceiver 1402 is further configured to: receive the first information sent by the network device via control plane signaling;

[0368] The generator model information includes the generator's input dimension; the discriminator model information includes the discriminator's input dimension.

[0369] The generator model information also includes one or more of the following:

[0370] The generator's input, and the format of the generator's model file;

[0371] The information of the discriminator model also includes one or more of the following:

[0372] The input to the discriminator, and the format of the discriminator's model file.

[0373] The transceiver 1402 is further configured to receive the initial model sent by the network device through one or more of the following methods:

[0374] Receive the initial model sent by the network device via control plane signaling;

[0375] Receive the initial model sent by the network device via user plane signaling.

[0376] The processor 1401 is further configured to:

[0377] A first vector is generated based on the input dimension of the generator, wherein the dimension of the first vector is the same as the input dimension of the generator;

[0378] A second vector is generated based on the input dimension of the discriminator, wherein the dimension of the second vector is the same as the input dimension of the discriminator;

[0379] The first vector and the second vector are used as inputs to the generator to be trained and the discriminator to be trained, respectively, to obtain the first generator model.

[0380] The transceiver 1402 is also used to perform any of the following:

[0381] In the control plane signaling, the information of the first generator model is sent to the network device via a container;

[0382] The information of the first generator model is sent to the network device via user plane signaling.

[0383] When the first generator model includes multiple fragments, the control plane signaling includes a seventh indication information and an eighth indication information, wherein the seventh indication information is used to indicate whether the fragment is the last fragment, and the eighth indication information is used to indicate the fragment number.

[0384] The user plane signaling packet header includes a ninth indication information and / or a tenth indication information; the ninth indication information is used to indicate that the user plane signaling is used to transmit an initial model; the tenth indication information is used to indicate the identifier of the transmission session.

[0385] When the first generator model includes multiple fragments, the user plane signaling includes an eleventh indication information and a twelfth indication information, wherein the eleventh indication information is used to indicate whether the fragment is the last fragment, and the twelfth indication information is used to indicate the fragment number.

[0386] The apparatus provided in this application embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0387] This application provides a communication device, including: a transceiver, a memory, a processor, and a program stored in the memory and executable on the processor; the processor is used to read the program in the memory to implement the steps in the model generation method described above.

[0388] This application also provides a readable storage medium storing a program. When executed by a processor, this program implements the various processes of the above-described model generation method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The readable storage medium can be any available medium or data storage device that a processor can access, including but not limited to magnetic storage (e.g., floppy disks, hard disks, magnetic tapes, magneto-optical disks (MO), etc.), optical storage (e.g., CDs, DVDs, BDs, HVDs, etc.), and semiconductor storage (e.g., ROMs, EPROMs, EEPROMs, non-volatile memory (NAND flash), solid-state drives (SSDs)).

[0389] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0390] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0391] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A model generation method applied to a network device, characterized in that, include: Send first information to the terminal, the first information including information about the generator model and information about the discriminator model; The terminal receives information about a first generator model sent by the terminal, wherein the first generator model is generated by the terminal based on the first information and an initial model; the initial model is obtained by the terminal from the network device or pre-configured on the terminal. The method further includes, after receiving the information of the first generator model sent by the terminal: Aggregate multiple first generator models to obtain information about the second generator model; The information of the second generator model is sent to the simulated wireless access network.

2. The method according to claim 1, characterized in that, Before receiving the information of the first generator model sent by the terminal, the method further includes: The initial model is sent to the terminal, and the initial model includes: a generator initial model and a discriminator initial model.

3. The method according to claim 1, characterized in that, Sending the first information to the terminal includes: The first information is sent to the terminal via control plane signaling; The generator model information includes the generator's input dimension; the discriminator model information includes the discriminator's input dimension.

4. The method according to claim 3, characterized in that, The generator model information also includes one or more of the following: The generator's input, and the format of the generator's model file; The information of the discriminator model also includes one or more of the following: The input to the discriminator, and the format of the discriminator's model file.

5. The method according to claim 2, characterized in that, Sending the initial model to the terminal includes any one of the following: In the control plane signaling, the initial model is sent to the terminal via a container. The initial model is sent to the terminal via user plane signaling.

6. The method according to claim 5, characterized in that, When the initial model includes multiple fragments, the control plane signaling includes a first indication information and a second indication information, wherein the first indication information is used to indicate whether the fragment is the last fragment, and the second indication information is used to indicate the fragment number.

7. The method according to claim 5, characterized in that, The header of the user plane signaling includes third indication information and / or fourth indication information; the third indication information is used to indicate that the user plane signaling is used to transmit an initial model; the fourth indication information is used to indicate the identifier of the transmission session.

8. The method according to claim 7, characterized in that, When the initial model includes multiple fragments, the user plane signaling includes a fifth indication information and a sixth indication information, wherein the fifth indication information is used to indicate whether the fragment is the last fragment, and the sixth indication information is used to indicate the fragment number.

9. The method according to claim 1, characterized in that, The aggregation of multiple first generator models to obtain information about the second generator model includes: Multiple intermediate data are obtained through multiple first generator models, wherein the input of each first generator model is a first random sequence corresponding to each first generator model; Multiple intermediate data points are used as input to the discriminator to be trained, and a second random sequence is used as input to the generator to be trained. The generator and the discriminator to be trained are trained iteratively to obtain information about the second generator model.

10. The method according to claim 9, characterized in that, After obtaining multiple intermediate data through multiple first generator models, the method further includes: Preprocess multiple intermediate data; The step of using multiple intermediate data points as input to the discriminator to be trained includes: Multiple preprocessed intermediate data are used as input to the discriminator to be trained.

11. The method according to claim 1, characterized in that, Sending the information of the second generator model to the simulated wireless access network includes: Send the metafile information of the second generator model and / or the model file of the second generator model to the simulated wireless access network; The metafile information of the second generator model includes one or more of the following: the input dimension of the second generator model, and the output dimension of the second generator model.

12. The method according to claim 11, characterized in that, The metafile information of the second generator model also includes one or more of the following: The network architecture description of the second generator model; the format of the model file of the second generator model; weights; gradients.

13. A model generation method, applied to a terminal, characterized in that, include: Receive first information sent by the network device, the first information including information of the generator model and information of the discriminator model; Generate a first generator model based on the first information and the initial model; The information of the first generator model is sent only to the network device. The information of the first generator model is used by the network device to aggregate multiple first generator models to obtain the information of the second generator model, and then the information of the second generator model is sent to the simulated wireless access network.

14. The method according to claim 13, characterized in that, The initial model is obtained by the terminal from the network device or pre-configured on the terminal; the initial model includes a generator initial model and a discriminator initial model.

15. The method according to claim 13, characterized in that, The first information sent by the receiving network device includes: Receive the first information sent by the network device via control plane signaling; The generator model information includes the generator's input dimension; the discriminator model information includes the discriminator's input dimension.

16. The method according to claim 15, characterized in that, The generator model information also includes one or more of the following: The generator's input, and the format of the generator's model file; The information of the discriminator model also includes one or more of the following: The input to the discriminator, and the format of the discriminator's model file.

17. The method according to claim 14, characterized in that, The initial model sent by the network device may be received in one or more of the following ways: Receive the initial model sent by the network device via control plane signaling; Receive the initial model sent by the network device via user plane signaling.

18. The method according to claim 13, characterized in that, The step of generating the first generator model based on the first information and the initial model includes: A first vector is generated based on the input dimension of the generator, wherein the dimension of the first vector is the same as the input dimension of the generator; A second vector is generated based on the input dimension of the discriminator, wherein the dimension of the second vector is the same as the input dimension of the discriminator; The first vector and the second vector are used as inputs to the generator to be trained and the discriminator to be trained, respectively, to obtain the first generator model.

19. The method according to claim 13, characterized in that, Sending the information of the first generator model only to the network device includes any one of the following: In control plane signaling, information about the first generator model is sent to the network device only via a container; The information of the first generator model is sent to the network device only via user plane signaling.

20. The method according to claim 19, characterized in that, When the first generator model includes multiple fragments, the control plane signaling includes a seventh indication information and an eighth indication information, wherein the seventh indication information is used to indicate whether the fragment is the last fragment, and the eighth indication information is used to indicate the fragment number.

21. The method according to claim 19, characterized in that, The header of the user plane signaling includes a ninth indication information and / or a tenth indication information; the ninth indication information is used to indicate that the user plane signaling is used to transmit an initial model; the tenth indication information is used to indicate the identifier of the transmission session.

22. The method according to claim 19, characterized in that, When the first generator model includes multiple fragments, the user plane signaling includes an eleventh indication information and a twelfth indication information, wherein the eleventh indication information is used to indicate whether the fragment is the last fragment, and the twelfth indication information is used to indicate the fragment number.

23. A data generation device, applied to network equipment, characterized in that, include: The first sending module is used to send first information to the terminal, the first information including information of the generator model and information of the discriminator model; The first receiving module is configured to receive information about a first generator model sent only by the terminal, wherein the generator model is generated by the terminal based on the first information and an initial model; the initial model is obtained by the terminal from the network device or pre-configured on the terminal. The device further includes: The first processing module is used to aggregate multiple first generator models to obtain information about the second generator model; The third sending module is used to send information about the second generator model to the simulated wireless access network.

24. A data generation device, applied to a terminal, characterized in that, include: The first receiving module is used to receive first information sent by the network device, the first information including information of the generator model and information of the discriminator model; The first generation module is used to generate a first generator model based on the first information and the initial model; The first sending module is used to send information about the first generator model only to the network device. The information about the first generator model is used by the network device to aggregate multiple first generator models to obtain information about the second generator model, and then send the information about the second generator model to the simulated wireless access network.

25. A data generation apparatus, applied to network equipment, characterized in that, include: Processor and transceiver; The transceiver is used to send first information to the terminal, the first information including information of the generator model and information of the discriminator model; The terminal receives information about a first generator model sent by the terminal, wherein the first generator model is generated by the terminal based on the first information and an initial model; the initial model is obtained by the terminal from the network device or pre-configured on the terminal. The processor is further configured to aggregate multiple first generator models to obtain information about a second generator model; The transceiver is also used to send information about the second generator model to a simulated wireless access network.

26. A data generation device, applied to a terminal, characterized in that, include: Processor and transceiver; The transceiver is used to receive first information sent by the network device, the first information including information of the generator model and information of the discriminator model; The processor is configured to generate a first generator model based on the first information and the initial model; The transceiver is further configured to send information about the first generator model only to the network device. The information about the first generator model is used by the network device to aggregate multiple first generator models to obtain information about the second generator model, and then send the information about the second generator model to the simulated wireless access network.

27. A communication device, comprising: A transceiver, a memory, a processor, and a program stored in the memory and executable on the processor; characterized in that, The processor is configured to read a program from the memory to implement the steps in the model generation method as described in any one of claims 1 to 12; or to implement the steps in the model generation method as described in any one of claims 13 to 22.

28. A readable storage medium for storing a program, characterized in that, When the program is executed by a processor, it implements the steps of the model generation method as described in any one of claims 1 to 12; or implements the steps of the model generation method as described in any one of claims 13 to 22.

Images