CN116073916A

CN116073916A - Model generation method, device, equipment and readable storage medium

Info

Publication number: CN116073916A
Application number: CN202111282277.4A
Authority: CN
Inventors: 李刚
Original assignee: China Mobile Communications Group Co Ltd; China Mobile Communications Ltd Research Institute
Current assignee: China Mobile Communications Group Co Ltd; China Mobile Communications Ltd Research Institute
Priority date: 2021-11-01
Filing date: 2021-11-01
Publication date: 2023-05-05

Abstract

The application discloses a model generation method, device and equipment and a readable storage medium, and relates to the technical field of communication so as to reduce air interface overhead. The method comprises the following steps: transmitting 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; the initial model is obtained from the network device by the terminal or is preconfigured at the terminal. The embodiment of the application can reduce the overhead of the air interface.

Description

Model generation method, device, equipment and readable storage medium

Technical Field

The present disclosure relates to the field of communications technologies, and in particular, to a method, an apparatus, a device, and a readable storage medium for generating a model.

Background

There are two general methods for acquiring existing downlink channel data: (1) generated by a link-level simulation system; (2) And after the terminal measures and carries out channel estimation, the estimation result is subjected to statistical quantization and then is fed back periodically.

The existing external AI (Artificial Intelligence ) application mode enables training and reasoning of an AI model to be decoupled, verification based on the AI model can only be performed afterwards, and effective verification and guarantee means are lacked. The training and iterative optimization of the current intelligent model are completed on line, the real-time performance is poor, and the direct relevance is lacking. When the network performance index is lower than expected after the model is online, the negative influence of the AI model can be avoided only through a back-off mechanism, and hysteresis exists.

In the future, a network based on an endogenous AI needs to introduce a digital twin simulation network to realize the prior verification of a model. However, the digital twin simulation network has high data overhead in acquiring data from the physical network, especially data related to the physical layer channel, and has high transmission requirement on an air interface because of small time granularity.

If the feedback of the terminal is adopted to generate the data required by the digital twin simulation network, but the channel data (such as CQI (Channel quality indicator, channel quality indicator)/PMI (Precoding matrix indicator )/RI (Rank indicator)) fed back by the existing terminal are all data after statistical quantization processing, and the reporting period is relatively large, so that the requirement of channel modeling cannot be met. If the terminal is required to report complete channel data directly, such as channel H matrix, the data size is very large due to small time granularity (millisecond level), and the overhead to the air interface is large.

Disclosure of Invention

The embodiment of the application provides a method, a device, equipment and a readable storage medium for generating a model so as to reduce air interface expenditure.

In a first aspect, an embodiment of the present application provides a method for generating a model, which is applied to a network device, including:

transmitting 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; the initial model is obtained from the network device by the terminal or is preconfigured at the terminal.

Wherein, before the receiving the information of the first generator model sent by the terminal, the method further comprises:

transmitting the initial model to the terminal, wherein the initial model comprises: a generator initial model and a discriminator initial model.

Wherein after receiving the information of the first generator model sent by the terminal, the method further comprises:

and aggregating a plurality of the first generator models to obtain information of a second generator model.

Wherein after said aggregating said first generator model to obtain information of a second generator model, said method further comprises:

And sending the information of the second generator model to the simulated wireless access network.

The sending the first information to the terminal includes:

transmitting the first information to the terminal through control plane signaling;

wherein the information of the generator model includes an input dimension of the generator; the information of the discriminator model includes: the input dimension of the discriminator.

Wherein the information of the generator model further comprises one or more of the following:

input to the generator, format of the model file of the generator;

the information of the discriminator model further includes one or more of:

input to the discriminator, format of the model file of the discriminator.

The sending the initial model to the terminal comprises any one of the following steps:

in control plane signaling, the initial model is sent to the terminal in a container mode;

and sending the initial model to the terminal through user plane signaling.

When the initial model comprises a plurality of fragments, the control plane signaling comprises first indication information and second indication information, wherein the first indication information is used for indicating whether the fragments are last fragments or not, and the second indication information is used for indicating fragment numbers.

Wherein, the packet header of the user plane signaling comprises third indication information and/or fourth indication information; the third indication information is used for indicating the user plane signaling to be used for transmitting an initial model; the fourth indication information is used to represent an identification of the transmission session.

When the initial model includes a plurality of slices, fifth indicating information and sixth indicating information are included in the user plane signaling, wherein the fifth indicating information is used for indicating whether the slices are last slices, and the sixth indicating information is used for indicating the slice numbers.

Wherein the aggregating the plurality of first generator models to obtain information of a second generator model includes:

obtaining a plurality of intermediate data through a plurality of first generator models, wherein the input of each first generator model is a first random sequence corresponding to each first generator model;

and taking the plurality of intermediate data as the input of the discriminator to be trained, taking the second random sequence as the input of the generator to be trained, and iteratively training the generator to be trained and the discriminator to be trained to obtain the information of a second generator model.

Wherein after the obtaining the plurality of intermediate data by the plurality of first generator models, the method further comprises:

Preprocessing a plurality of intermediate data;

the taking the plurality of intermediate data as input of the discriminator to be trained comprises:

and taking the preprocessed plurality of intermediate data as the input of the discriminator to be trained.

Wherein the sending the information of the second generator model to the simulated wireless access network comprises:

transmitting metafile information of the second generator model and/or a 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: an input dimension of the second generator model, an output dimension of the second generator model.

Wherein the metafile information of the second generator model further comprises one or more of:

a network architecture description of the second generator model; the format of the model file of the second generator model; weighting; gradient.

In a second aspect, an embodiment of the present application provides a method for generating a model, which is applied to a terminal, including:

receiving first information sent by network equipment, wherein the first information comprises information of a generator model and information of a discriminator model;

Generating a first generator model according to the first information and the initial model;

and sending information of the first generator model to the network equipment.

Wherein the initial model is acquired from the network device by the terminal or is preconfigured in the terminal; the initial model includes a generator initial model and a discriminator initial model.

The receiving, by the network device, the first information sent by the network device includes:

receiving the first information sent by the network equipment through control plane signaling;

input to the generator, format of the model file of the generator;

the information of the discriminator model further includes one or more of:

input to the discriminator, format of the model file of the discriminator.

Wherein the initial model sent by the network device is received by one or more of:

receiving the initial model sent by the network equipment through control plane signaling;

And receiving the initial model sent by the network equipment through user plane signaling.

Wherein the generating a first generator model according to the first information and the initial model includes:

generating a first vector according to the input dimension of the generator, wherein the dimension of the first vector is the same as the input dimension of the generator;

generating a second vector according to the input dimension of the discriminator, wherein the dimension of the second vector is the same as the input dimension of the discriminator;

and respectively taking the first vector and the second vector as the input of a generator to be trained and the input of a discriminator to be trained to obtain the first generator model.

Wherein the sending the information of the first generator model to the network device includes any one of the following:

in control plane signaling, sending information of the first generator model to the network equipment in a container mode;

and sending the information of the first generator model to the network equipment through user plane signaling.

When the first generator model comprises a plurality of fragments, seventh indication information and eighth indication information are included in the control plane signaling, wherein the seventh indication information is used for indicating whether the fragments are last fragments, and the eighth indication information is used for indicating fragment numbers.

The packet header of the user plane signaling comprises ninth indication information and/or tenth indication information; the ninth indication information is used for indicating that the user plane signaling is used for transmitting an initial model; the tenth indication information is used to represent an identification of the transmission session.

When the first generator model includes a plurality of slices, the user plane signaling includes eleventh indication information and twelfth indication information, wherein the eleventh indication information is used for indicating whether a slice is a last slice, and the twelfth indication information is used for indicating a slice number.

In a third aspect, an embodiment of the present application provides a data generating apparatus, applied to a network device, including:

the first sending module is used for sending first information to the terminal, wherein the first information comprises information of a generator model and information of a discriminator model;

the first receiving module is used for receiving information of a first generator model sent by the terminal, wherein the generator model is generated by the terminal according to the first information and an initial model; the initial model is obtained from the network device by the terminal or is preconfigured at the terminal.

Wherein the apparatus further comprises:

the second sending module is configured to send the initial model to the terminal, where the initial model includes: a generator initial model and a discriminator initial model.

Wherein the apparatus further comprises:

and the first processing module is used for aggregating a plurality of the first generator models to obtain information of a second generator model.

Wherein the apparatus further comprises:

and the third sending module is used for sending the information of the second generator model to the simulated wireless access network.

The first sending module is configured to send the first information to the terminal through control plane signaling;

input to the generator, format of the model file of the generator;

the information of the discriminator model further includes one or more of:

input to the discriminator, format of the model file of the discriminator.

The second sending module is configured to send the initial model to the terminal in a container manner in control plane signaling;

And sending the initial model to the terminal through user plane signaling.

Wherein the first processing module comprises:

the first acquisition submodule is used for obtaining a plurality of intermediate data through a plurality of first generator models, wherein the input of each first generator model is a first random sequence corresponding to each first generator model;

The first processing sub-module is used for taking a plurality of intermediate data as input of a discriminator to be trained, taking a second random sequence as input of a generator to be trained, and iteratively training the generator to be trained and the discriminator to be trained to obtain information of a second generator model.

Wherein the first processing module further comprises:

the preprocessing sub-module is used for preprocessing a plurality of intermediate data;

the first processing sub-module is further configured to take the preprocessed plurality of intermediate data as input of a discriminator to be trained.

The third sending module is configured to send metafile information of the second generator model and/or a model file of the second generator model to the simulated wireless access network;

In a fourth aspect, an embodiment of the present application provides a data generating device, which is applied to a terminal, including:

the first receiving module is used for receiving first information sent by the network equipment, wherein the first information comprises information of a generator model and information of a discriminator model;

the first generation module is used for generating a first generator model according to the first information and the initial model;

and the first sending module is used for sending the information of the first generator model to the network equipment.

The first receiving module is used for receiving the first information sent by the network equipment through control plane signaling;

input to the generator, format of the model file of the generator;

the information of the discriminator model further includes one or more of:

Input to the discriminator, format of the model file of the discriminator.

Wherein the first generation module comprises:

a first generation sub-module, configured to generate a first vector according to an input dimension of the generator, where a dimension of the first vector is the same as the input dimension of the generator;

a second generation sub-module, configured to generate a second vector according to an input dimension of the discriminator, where a dimension of the second vector is the same as the input dimension of the discriminator;

and the third generation submodule is used for respectively taking the first vector and the second vector as the input of a generator to be trained and the input of a discriminator to be trained to obtain the first generator model.

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

In a fifth aspect, an embodiment of the present application provides a data generating apparatus, applied to a network device, including: a processor and a transceiver;

the transceiver is used for sending first information to the 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; the initial model is obtained from the network device by the terminal or is preconfigured at the terminal.

Wherein 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.

The processor is further configured to aggregate a plurality of the first generator models to obtain information of a second generator model.

Wherein the transceiver is further configured to send information of the second generator model to an emulated wireless access network.

Wherein the transceiver is further configured to:

input to the generator, format of the model file of the generator;

the information of the discriminator model further includes one or more of:

input to the discriminator, format of the model file of the discriminator.

Wherein the transceiver is further configured to perform any one of:

And sending the initial model to the terminal through user plane signaling.

Wherein the processor is further configured to:

preprocessing a plurality of intermediate data;

Wherein the transceiver is further configured to:

In a sixth aspect, an embodiment of the present application provides a data generating apparatus, which is applied to a terminal, including: comprising the following steps: a processor and a transceiver;

The transceiver is used for receiving first information sent by the network equipment, wherein the first information comprises information of a generator model and information of a discriminator model;

the processor is used for generating a first generator model according to the first information and the initial model;

the transceiver is further configured to send information of the first generator model to the network device.

Wherein the transceiver is further configured to: receiving the first information sent by the network equipment through control plane signaling;

input to the generator, format of the model file of the generator;

the information of the discriminator model further includes one or more of:

input to the discriminator, format of the model file of the discriminator.

Wherein the transceiver is further configured to: receiving an initial model sent by the network device by one or more of:

Wherein the processor is further configured to:

Wherein the transceiver is further configured to perform any one of:

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

In an eighth aspect, embodiments of the present application provide a readable storage medium storing a program, which when executed by a processor, implements steps in a model generation method as described above.

In the embodiment of the application, the network device sends first information to the terminal, and the terminal generates a first generator model according to the first information and the initial model and sends the first generator model to the network device. From the above analysis, it can be seen that in the embodiment of the present application, the terminal does not need to upload a large amount of data, but only needs to upload the generator model. Therefore, by utilizing the scheme of the embodiment of the application, the overhead of an air interface can be reduced.

Drawings

FIG. 1 is one of the flowcharts of the model generation method provided in the embodiments of the present application;

FIG. 2 is a schematic diagram of first information in an embodiment of the present application;

fig. 3 is a schematic diagram of RRC signaling transmitting an initial model in an embodiment of the present application;

FIG. 4 is an expanded schematic diagram of SDAP in an embodiment of the present application;

FIG. 5 is a schematic diagram of a process of a generator model for aggregating multiple terminal uploads by a network device in an embodiment of the present application;

FIG. 6 is a second flowchart of a model generation method according to an embodiment of the present disclosure;

FIG. 7 is a third flowchart of a model generation method provided in an embodiment of the present application;

FIG. 8 is a process schematic of a generator provided by an embodiment of the present application;

FIG. 9 is a process schematic of a discriminator provided by an embodiment of the present application;

FIG. 10 is an effect schematic of an embodiment of the present application;

FIG. 11 is one of the block diagrams of the data generating apparatus provided in the embodiment of the present application;

FIG. 12 is a second block diagram of a data generating apparatus according to an embodiment of the present application;

FIG. 13 is a third block diagram of the data generating apparatus provided in the embodiment of the present application;

fig. 14 is a diagram showing a structure of a data generating apparatus according to an embodiment of the present invention.

Detailed Description

In the embodiment of the application, the term "and/or" describes the association relationship of the association objects, which means that three relationships may exist, for example, a and/or B may be represented: a exists alone, A and B exist together, and B exists alone. The character "/" generally indicates that the context-dependent object is an "or" relationship.

The term "plurality" in the embodiments of the present application means two or more, and other adjectives are similar thereto.

The following description of the technical solutions in the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings in the embodiments of the present application, and it is apparent that the described embodiments are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.

Referring to fig. 1, fig. 1 is a flowchart of a model generation method provided in an embodiment of the present application, which is applied to a network device, as shown in fig. 1, and includes the following steps:

step 101, first information is sent to a terminal.

Wherein the first information is for causing the terminal to generate a generator model and a discriminator model. The information may include: information of the generator model and information of the discriminator model. Wherein the information of the generator model includes an input dimension of the generator; the information of the discriminator model includes: the input dimension of the discriminator. Further, in order to improve the efficiency of terminal generation model, the information of the generator model further includes one or more of the following: the input to the generator, the format of the model file of the generator. The information of the discriminator model further includes one or more of: input to the discriminator, format of the model file of the discriminator.

Wherein the input dimensions of the generator include:

algorithm for generating random input variables (random algorithm): such as normal distribution N (0, 1), bernoulli distribution, etc.;

dimension of random input variable (latex dimension): such as 100.

After the terminal obtains the input dimension configuration of the generator, a vector with dimension 100 is generated as the input of the generator according to the configured random input algorithm.

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

The pattern of the generative model is used to indicate the pattern of the model, e.g. the countermeasure generation network (Generative Adversarial Network, GAN) may be used by default as pattern 1, although other patterns may be used.

The format of the model file of the generator or the model file of the discriminator may be in the format of h5, pkl, etc.

Fig. 2 is a schematic diagram of the first information. The first information radiochannelModelModelgenerationConfig may include, among others, by way of example: modelMode (model of generative model), generator Inputshape (input dimension of generator), identifier Inputshape (input dimension of discriminator), modelFormat (format of model file).

In the embodiment of the present application, the first information may be used for generation of a channel model, generation of an image processing model, and the like. The network device may send the first information to the terminal through control plane signaling. The control plane signaling may be, for example, an RRC (Radio Resource Control ) reconfiguration message or the like.

102, 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; the initial model is obtained from the network device by the terminal or is preconfigured at the terminal.

Wherein the information of the first generator model includes one or more of: the structure of the first generator model, the model file, the input dimension, the output dimension, the weight, the gradient and the like of the first generator model.

After receiving the first information issued by the network equipment, the terminal combines the generator initial model and the discriminator initial model, starts the model training of the GAN, including the generator G model and the discriminator D model, until Nash equilibrium is reached. From here on, the resulting generator G model can be taken as the first generator model here.

In this step, the network device may also receive a first generator model sent by the terminal through control plane signaling (e.g. RRC signaling) or user plane signaling. Wherein the first generator model is used for generating a channel model, generating an image processing model and the like.

On the basis of the above embodiment, the network device may also send an initial model to the terminal before step 102 in order to improve the efficiency of obtaining the generator model. The initial model includes: a generator initial model and a discriminator initial model.

In this step, the network device may send the initial model to the terminal in at least two ways.

Mode one: in control plane signalling, the initial model is sent to the terminal by means of a container,

For example, the network device may transmit the initial model by means of a container in RRC signaling. If the model file is large, it can be cut into different slices for transmission. Whether a slice is the last slice is indicated by a SegmentType, the slice number is indicated by a SegmentNumber, and the producer and discriminator models are passed by a container.

As shown in fig. 3, a schematic diagram of RRC signaling of the initial model is transmitted. Wherein the RadioChannelModel comprises: segmentType, segmentNumber, container.

Mode two: and sending the initial model to the terminal through user plane signaling.

In the embodiment of the application, the transmission of the model data packet is indicated by expanding the packet header of the user plane. The packet header of the user plane signaling comprises third indication information and/or fourth indication information; the third indication information is used for indicating the user plane signaling to be used for transmitting an initial model; the fourth indication information is used to represent an identification of the transmission session. When the initial model includes a plurality of slices, in the user plane signaling, fifth indicating information and sixth indicating information are included, wherein the fifth indicating information is used for indicating whether the slices are last slices, and the sixth indicating information is used for indicating the slice numbers.

In practical application, a new field may be added to the packet header of the user plane signaling, which is used to carry the third indication information and the fourth indication information respectively. For example, as shown in fig. 4, the network device may instruct the model transfer by extending the header of the SDAP (Service Data Adaptation Protocol ), and add the TransactionID to specify the identity of the current transfer session. If the model data is large, the number of the slice is indicated by SegmentNum, and whether the current slice is the last packet or slice transmitted by the model is indicated by EndMarker.

On the basis of the above embodiment, after step 103, the method may further include: and aggregating a plurality of the first generator models to obtain information of a second generator model.

Wherein the information of the second generator model includes one or more of: metafile information, model files of the second generator model; the metafile information of the second generator model includes one or more of: the network architecture description of the second generator model, the input dimension, the output dimension, the weight, the gradient of the second generator model.

After receiving the first generator model uploaded by the plurality of terminals, the network device inputs a random sequence into the first generator model to generate corresponding intermediate data, adopts a double-layer GAN method to aggregate the models, and retrains new generator and discriminator models (G, D). The network equipment aggregates a plurality of channel models by a double-layer GAN method, so that the transmission cost of the data model is reduced.

Specifically, the network device obtains a plurality of intermediate data through a plurality of first generator models, wherein the input of each first generator model is a first random sequence corresponding to each first generator model. And taking the plurality of intermediate data as the input of the discriminator to be trained, taking the second random sequence as the input of the generator to be trained, and iteratively training the generator to be trained and the discriminator to be trained to obtain the information of a second generator model. The second random sequence may be a random sequence optionally selected by the network device.

In the above process, in order to improve the data processing efficiency, a plurality of intermediate data may be preprocessed. Then, correspondingly, the preprocessed plurality of intermediate data are used as inputs of the discriminator to be trained.

The preprocessing may include, for example, a data optimization process such as data filtering, deletion, etc. Taking the generation of a channel model as an example, the network device can acquire the position and moving speed information of the terminal and the channel quality measurement information, and then in the preprocessing process, the network device can cluster users with similar positions or similar channels, and select channel data with typical users in the same cluster as the input of the discriminator. Or, the sample distribution calculation is carried out on the channel generation data of multiple users, the channel data with similar channel data samples is filtered, and the processing cost of the channel discriminator is reduced. Mainstream algorithms for calculating the sample distribution include, but are not limited to:

(1) MMD (maximize mean discrepancy, maximum mean difference): the distribution P1 and the distribution P2 are mapped to a high-dimensional Hilbert space through a kernel function, high-dimensional characteristics are obtained, and then the distance between the high-dimensional characteristics is calculated.

(2) Bulldozer distance: the consumption required to move the pile distribution P1 to the pile distribution P2 under path planning. While bulldozer distance is the minimum consumption under optimal path planning.

Since the calculation of the radio access network is considered to be stronger than that of the terminal, the use of deeper networks and larger random input variable spaces are considered in the design of the channel generator model and the channel discriminator model.

As shown in fig. 5, taking a channel model as an example, a channel generator model (such as G1, G2, G3 and …) uploaded by a plurality of terminals is received, the network device inputs a random sequence (random sequence 1, random sequence 2 and … …) into the channel generator model to generate corresponding channel data (channel data 1, channel data 2 and channel data 3), the input of the generator is a random sequence N, the aggregation model is performed by adopting a double-layer GAN method, and new generator and discriminator models (G and D) are retrained.

On the above basis, the network device may further send information of the second generator model to the simulated wireless access network.

Specifically, the network device may send metafile information of the second generator model and/or a 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: an input dimension of the second generator model, an output dimension of the second generator model. Further, in order to improve the processing efficiency of the simulation network, the metafile information of the second generator model further 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; weighting; gradient.

Wherein the input dimensions of the second generator model include:

dimension of random input variable (latex dimension): such as 100.

The output dimension of the second generator model comprises: the number of antenna ports M, and the number of taps N based on multipath delay. The simulated wireless access network generates a channel impulse response matrix of M x N through a channel generator model as the input of simulated channel data.

The network architecture description of the second generator model includes, for example, a description of a network structure including convolution, full connection, pooling, and the like. The format of the model file of the second generator model may be, for example, h5, pkl, or the like.

Wherein the model file of the second generator model may be transmitted via a binary bitstream file.

Referring to fig. 6, fig. 6 is a flowchart of a model generation method provided in an embodiment of the present application, applied to a terminal, as shown in fig. 6, including the following steps:

step 601, receiving first information sent by a network device, wherein the first information comprises information of a generator model and information of a discriminator model.

Wherein the terminal receives the first information sent by the network device through control plane signaling (such as RRC signaling). The meaning of the first information may refer to the description of the foregoing embodiment.

Step 602, generating a first generator model according to the first information and the initial model.

The initial model is acquired from the network equipment by the terminal or is preconfigured in the terminal; the initial model includes a generator initial model and a discriminator initial model.

In this step, the terminal may receive the initial model sent by the network device in at least two ways:

(1) The initial model sent by the network equipment through control plane signaling is received, and the initial model comprises a generator initial model and a discriminator initial model.

(2) And receiving the initial model sent by the network equipment through user plane signaling, wherein the initial model comprises a generator initial model and a discriminator initial model.

Wherein the content and meaning of the user plane signaling and the control plane signaling can be referred to the description of the foregoing embodiments.

Specifically, in this step, the terminal generates a first vector according to the input dimension of the generator, where the dimension of the first vector is the same as the input dimension of the generator; generating a second vector according to the input dimension of the discriminator, wherein the dimension of the second vector is the same as the input dimension of the discriminator; and respectively taking the first vector and the second vector as the input of a generator to be trained and the input of a discriminator to be trained until Nash equilibrium is achieved, so as to obtain the first generator model.

Step 603, sending information of the first generator model to the network device.

In this step, the terminal may transmit information of the first generator model to a network device in at least two ways.

(1) In control plane signalling, information of the first generator model is sent to the network device by means of a container. When the first generator model includes a plurality of slices, in the control plane signaling, seventh indication information and eighth indication information are included, wherein the seventh indication information is used for indicating whether a slice is a last slice, and the eighth indication information is used for indicating a slice number.

For example, the terminal may send information of the first generator model to a network device through RRC signaling.

(2) And sending the information of the first generator model to the network equipment through user plane signaling.

For example, the terminal may send the information of the first generator model by expanding an SDAP header. The packet header of the user plane signaling comprises ninth indication information and/or tenth indication information; the ninth indication information is used for indicating that the user plane signaling is used for transmitting an initial model; the tenth indication information is used to represent an identification of the transmission session. When the first generator model includes a plurality of slices, eleventh indication information and twelfth indication information are included in the user plane signaling, wherein the eleventh indication information is used for indicating whether the slices are last slices, and the twelfth indication information is used for indicating the slice numbers.

It should be noted that, in the embodiment of the present application, the information of the model of the generator may not be limited to include those listed by the above examples, and may be extended as needed.

Referring to fig. 7, fig. 7 is a flowchart of a model generating method provided in an embodiment of the present application. In fig. 7, the generation of the channel model is described as an example, and may include the following steps:

step 701, a physical radio access network (e.g. a base station) transmits channel first information to a terminal through RRC signaling, the information including information of a generator model and information of a discriminator model. In this embodiment, the following are taken as examples:

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

(2) Channel data input dimension and format of the discriminant model D: such as a frequency impulse response H matrix based on antenna ports and the time-frequency domain.

(3) The model format of generator G model and discriminator D.

Step 702, the physical radio access network transmits a channel initial model ((G generator initial model and D discriminator initial model) through RRC or user plane.

Step 703, after receiving the first information issued by the physical wireless access network, the terminal collects channel data and starts training the GAN model, including training the generator G model and training the arbiter D model, until nash equalization is achieved, to obtain a trained generator G model and a trained arbiter D model. Wherein the generator G model is used for channel data generation.

Step 704, the information of the generator G model generated by the terminal through RRC signaling or user plane data uploading includes one or more of the following: the structure of the generator model, the model file, the input dimension, the output dimension, the weight and the gradient.

Step 705, after the physical wireless access network (such as a base station) receives the generator G models of the plurality of terminals, the following method is adopted to aggregate the models:

inputting a random sequence corresponding to each user into a corresponding generator G model to generate a certain amount of channel data; inputting the channel data as real sample data into a new GAN network; in addition, a random sequence is input into a generator G network to be trained, a fake sample is generated, a discriminator D is utilized to distinguish true from false, and the generator G network and the discriminator D network are iterated until the network reaches Nash balance. Thus, the aggregated generator G model and the arbiter D model are obtained.

Step 706, the physical wireless access network (such as a base station) uploads the aggregated information of the 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 weight and the gradient.

Step 707, the emulated wireless access network generates data and emulated environment according to the generator G model.

In the above embodiment, taking into account the problem that the ordinary terminal may have computing power, the drive test terminal may be considered to be used for collecting and training the model data.

The following is one embodiment of channel data generation using the GAN model:

(1) The generator model uses full concatenation +2 upsampling + three layer convolution for data generation (as shown in figure 8).

(2) The discriminator adopts a three-layer convolution and full connection mode to identify true and false data (as shown in fig. 9).

As shown in fig. 10, after 500 training iterations, it was found that the distribution deviation (Wasserstein distance, bulldozer distance) between the real channel data and the data generated by the generator model was gradually decreasing. Therefore, the method using GAN can well generate data similar to the actual channel data distribution.

In the above embodiment, since the terminal only needs to transfer the channel data generation model, the overhead of uploading the channel data by the terminal can be reduced. In addition, the network aggregates a plurality of channel models by a double-layer GAN method, so that the transmission cost of the data model is reduced.

The embodiment of the application also provides a data generation device which is applied to the network equipment. Referring to fig. 11, fig. 11 is a block diagram of a data generating apparatus provided in an embodiment of the present application. Since the principle of the data generating device for solving the problem is similar to that of the model generating method in the embodiment of the present application, the implementation of the data generating device may refer to the implementation of the method, and the repetition is omitted.

As shown in fig. 11, the data generating apparatus 1100 includes:

a first transmitting module 1101 for transmitting first information to a terminal, the first information including information of a generator model and information of a discriminator model; a first receiving module 1102, configured to receive information of a first generator model sent by the terminal, where the generator model is generated by the terminal according to the first information and an initial model; the initial model is obtained from the network device by the terminal or is preconfigured at the terminal.

Wherein the apparatus further comprises:

input to the generator, format of the model file of the generator;

the information of the discriminator model further includes one or more of:

input to the discriminator, format of the model file of the discriminator.

And sending the initial model to the terminal through user plane signaling.

Wherein the first processing module comprises:

Wherein the first processing module further comprises:

The device provided in the embodiment of the present application may execute the above method embodiment, and its implementation principle and technical effects are similar, and this embodiment will not be described herein again.

The embodiment of the application also provides a data generation device which is applied to the terminal. Referring to fig. 12, fig. 12 is a block diagram of a data generating apparatus provided in an embodiment of the present application. Since the principle of the data generating device for solving the problem is similar to that of the model generating method in the embodiment of the present application, the implementation of the data generating device may refer to the implementation of the method, and the repetition is omitted.

As shown in fig. 12, the data generating apparatus 1200 includes:

a first receiving module 1201, configured to receive first information sent by a network device, where the first information includes information of a generator model and information of a discriminator model; a first generation module 1202 for generating a first generator model according to the first information and the initial model; a first sending module 1203 is configured to send information of the first generator model to the network device.

input to the generator, format of the model file of the generator;

the information of the discriminator model further includes one or more of:

input to the discriminator, format of the model file of the discriminator.

Wherein the first generation module comprises:

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

It should be noted that, in the embodiment of the present application, the division of the units is schematic, which is merely a logic function division, and other division manners may be implemented in actual practice. In addition, each functional unit in each embodiment of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a processor-readable storage medium. Based on such understanding, the technical solution of the present application may be embodied in essence or a part contributing to the prior art or all or part of the technical solution, in the form of a software product stored in a storage medium, including several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor (processor) to perform all or part of the steps of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

As shown in fig. 13, an embodiment of the present application provides a data generating apparatus, which is applied to a network device, including: a processor 1301 and a transceiver 1302;

the transceiver 1302 is configured to send first information to a terminal, where the first information includes 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; the initial model is obtained from the network device by the terminal or is preconfigured at the terminal.

Wherein the transceiver 1302 is further configured to send the initial model to the terminal, where the initial model includes: a generator initial model and a discriminator initial model.

The processor 1301 is further configured to aggregate a plurality of the first generator models to obtain information of a second generator model.

Wherein the transceiver 1302 is further configured to send information of the second generator model to an emulated wireless access network.

Wherein the transceiver 1302 is further configured to:

input to the generator, format of the model file of the generator;

the information of the discriminator model further includes one or more of:

input to the discriminator, format of the model file of the discriminator.

Wherein the transceiver 1302 is further configured to perform any of:

and sending the initial model to the terminal through user plane signaling.

Wherein the processor 1301 is further configured to:

preprocessing a plurality of intermediate data;

Wherein the transceiver 1302 is further configured to:

As shown in fig. 14, an embodiment of the present application provides a data generating apparatus, applied to a terminal, including: comprising the following steps: a processor 1401 and a transceiver 1402;

the transceiver 1402 is configured to receive first information sent by a network device, where the first information includes information of a generator model and information of a discriminator model;

the processor 1401 is configured to generate a first generator model according to the first information and an initial model;

the transceiver 1402 is further configured to send information of the first generator model to the network device.

Wherein the transceiver 1402 is further configured to: receiving the first information sent by the network equipment through control plane signaling;

input to the generator, format of the model file of the generator;

the information of the discriminator model further includes one or more of:

input to the discriminator, format of the model file of the discriminator.

Wherein the transceiver 1402 is further configured to: receiving an initial model sent by the network device by one or more of:

Wherein the processor 1401 is further configured to:

Wherein the transceiver 1402 is further configured to perform any one of:

The embodiment of the application provides a communication device, which comprises: a transceiver, a memory, a processor, and a program stored on the memory and executable on the processor; the processor is configured to read a program in the memory to implement the steps in the model generating method as described above.

The embodiment of the present application further provides a readable storage medium, where a program is stored, and when the program is executed by a processor, the processes of the embodiment of the model generating method are implemented, and the same technical effects can be achieved, so that repetition is avoided, and no further description is given here. The readable storage medium may be any available medium or data storage device that can be accessed by a processor, including, but not limited to, magnetic memories (e.g., floppy disks, hard disks, magnetic tapes, magneto-optical disks (MO), etc.), optical memories (e.g., CD, DVD, BD, HVD, etc.), semiconductor memories (e.g., ROM, EPROM, EEPROM, nonvolatile memories (NAND FLASH), solid State Disks (SSD)), etc.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. In light of such understanding, the technical solutions of the present application may be embodied essentially or in part in the form of a software product stored in a storage medium (e.g., ROM/RAM, magnetic disk, optical disk) and including instructions for causing a terminal (which may be a cell phone, computer, server, air conditioner, or network device, etc.) to perform the methods described in the embodiments of the present application.

The embodiments of the present application have been described above with reference to the accompanying drawings, but the present application is not limited to the above-described embodiments, which are merely illustrative and not restrictive, and many forms may be made by those of ordinary skill in the art without departing from the spirit of the present application and the scope of the claims, which are also within the protection of the present application.

Claims

1. A model generation method applied to a network device, comprising:

2. The method of claim 1, wherein prior to said receiving information of the first generator model transmitted by the terminal, the method further comprises:

3. The method of claim 1, wherein after said receiving information of the first generator model transmitted by the terminal, the method further comprises:

4. A method according to claim 3, wherein after said aggregating said first generator model to obtain information of a second generator model, the method further comprises:

5. The method of claim 1, wherein the sending the first information to the terminal comprises:

6. The method of claim 5, wherein the step of determining the position of the probe is performed,

the information of the generator model further includes one or more of the following:

input to the generator, format of the model file of the generator;

the information of the discriminator model further includes one or more of:

Input to the discriminator, format of the model file of the discriminator.

7. The method according to claim 2, wherein said sending the initial model to the terminal comprises any one of:

and sending the initial model to the terminal through user plane signaling.

8. The method of claim 7, wherein when the initial model includes a plurality of slices, including first indication information and second indication information in the control plane signaling, wherein the first indication information is used to indicate whether a slice is a last slice, and the second indication information is used to indicate a slice number.

9. The method according to claim 7, wherein a packet header of the user plane signaling includes third indication information and/or fourth indication information; the third indication information is used for indicating the user plane signaling to be used for transmitting an initial model; the fourth indication information is used to represent an identification of the transmission session.

10. The method of claim 9, wherein the step of determining the position of the substrate comprises,

when the initial model includes a plurality of slices, in the user plane signaling, fifth indicating information and sixth indicating information are included, wherein the fifth indicating information is used for indicating whether the slices are last slices, and the sixth indicating information is used for indicating the slice numbers.

11. A method according to claim 3, wherein said aggregating a plurality of said first generator models to obtain information of a second generator model comprises:

12. The method of claim 11, wherein after the obtaining a plurality of intermediate data by a plurality of the first generator models, the method further comprises:

preprocessing a plurality of intermediate data;

13. The method of claim 4, wherein the sending information of the second generator model to the simulated wireless access network comprises:

14. The method of claim 13, wherein the metafile information of the second generator model further comprises one or more of:

15. The model generation method is applied to a terminal and is characterized by comprising the following steps:

and sending information of the first generator model to the network equipment.

16. The method according to claim 15, wherein the initial model is obtained by the terminal from the network device or pre-configured at the terminal; the initial model includes a generator initial model and a discriminator initial model.

17. The method of claim 15, wherein the receiving the first information sent by the network device comprises:

18. The method of claim 17, wherein the step of determining the position of the probe is performed,

input to the generator, format of the model file of the generator;

the information of the discriminator model further includes one or more of:

input to the discriminator, format of the model file of the discriminator.

19. The method of claim 16, wherein the initial model sent by the network device is received by one or more of:

20. The method of claim 15, wherein generating a first generator model from the first information and an initial model comprises:

21. The method of claim 15, wherein said sending information of said first generator model to said network device comprises any one of:

22. The method of claim 21, wherein when the first generator model includes a plurality of slices, including seventh indication information and eighth indication information in the control plane signaling, wherein the seventh indication information is used to indicate whether a slice is a last slice, and the eighth indication information is used to indicate a slice number.

23. The method according to claim 21, wherein a packet header of the user plane signaling includes ninth indication information and/or tenth indication information; the ninth indication information is used for indicating that the user plane signaling is used for transmitting an initial model; the tenth indication information is used to represent an identification of the transmission session.

24. The method of claim 21, wherein when the first generator model includes a plurality of slices, including eleventh indication information and twelfth indication information in the user plane signaling, wherein the eleventh indication information is used to indicate whether a slice is a last slice, and the twelfth indication information is used to indicate a slice number.

25. A data generating apparatus applied to a network device, comprising:

26. A data generating apparatus applied to a terminal, comprising:

27. A data generating apparatus applied to a network device, comprising: a processor and a transceiver;

28. A data generating apparatus applied to a terminal, comprising: comprising the following steps: a processor and a transceiver;

29. A communication device, comprising: a transceiver, a memory, a processor, and a program stored on the memory and executable on the processor; it is characterized in that the method comprises the steps of,

the processor for reading a program in a memory to implement the steps in the model generation method according to any one of claims 1 to 14; or to implement the steps in a model generation method as claimed in any one of claims 15 to 24.

30. A readable storage medium storing a program, wherein the program when executed by a processor implements the steps in the model generation method according to any one of claims 1 to 14; or to implement the steps in a model generation method as claimed in any one of claims 15 to 24.