Wireless communication physical layer multi-task processing method based on pre-training large language model

Abstract

The invention discloses a wireless communication physical layer multi-task processing method based on a pre-trained large language model, and belongs to the technical field of wireless communication, a wireless communication system physical layer multi-task large model is constructed based on the pre-trained large language model, and is deployed at a base station side, and training is performed through user historical data, so that the multi-task processing of the wireless communication system physical layer is realized. On-line reasoning is carried out after training is completed, the method can be used for completing multiple tasks such as channel estimation, channel prediction, distance estimation, path loss estimation and beam forming, the model deployment difficulty is low, the training overhead is small, and the model precision is kept.

Description

Technical Field

[0001] The present invention belongs to the field of wireless communication technology, and relates to a physical layer task processing technology of a wireless communication system, and specifically relates to a solution for multi-task processing of the physical layer of a wireless communication system based on a pre-trained large language model. Background Art

[0002] With the popularization of Internet of Vehicles scenarios, the number and density of communication needs are gradually increasing. This poses a huge challenge to the speed and stability of the fifth-generation communication system (5G). This challenge affects all levels of communication design. At the network level, more efficient congestion control and traffic management strategies are needed to cope with dynamic loads; at the physical transmission level, since accurate channel state information (CSI) is a prerequisite for the normal operation of subsequent precoding, power allocation, modulation selection, and transmit antenna selection, with the introduction of massive multiple-input multiple-output (mMIMO) technology, the number of antennas has increased significantly, resulting in an increase in the pilot resource overhead of CSI estimation. At the same time, the channel aging problem caused by high mobility also exacerbates this phenomenon, resulting in a decrease in the system's spectrum efficiency; these challenges all require the design of appropriate solutions to solve them.

[0003] Due to the different optimization goals of these problems, traditional solutions focus on single task optimization or single module optimization. For example, for pilot-based channel estimation tasks, many methods use generative artificial intelligence (GAI) as a tool to achieve higher-precision channel estimation, because GAI's powerful generation capability is consistent with the task requirements of channel estimation; in channel prediction tasks, most methods use recurrent neural networks (RNN) due to their excellent performance in processing timing tasks; however, designing optimization methods for each module in a system will make the design process very complicated, and the stability of the system will be reduced because the error of information transmission between modules is not considered during the individual design. At the same time, the relatively independent optimization between each module can only obtain the local optimal solution, and it is difficult to obtain the global optimal solution of all modules.

[0004] The multi-task learning method can jointly train multiple communication tasks and obtain common representations between tasks to obtain better model performance, while also reducing the complexity of the training process. Therefore, some works consider the two key tasks of wireless signal recognition, wireless signal classification and modulation mode recognition, and use the underlying shared multi-task learning network for joint training; there are also works that consider the joint training and reasoning of direct channel (DC) and cascaded channel (CC) estimation in the reconfigurable intelligent surface (RIS) system. By jointly learning the two tasks, DC and CC can be obtained simultaneously using one network, reducing the pilot overhead in the actual system; however, these methods also have obvious disadvantages. The multi-task learning method based on underlying sharing is prone to data imbalance and seesaw phenomenon. At the same time, due to the small scale of the model, it is difficult to increase the types and scope of learning tasks. That is, most of the existing methods only jointly train two strongly related communication tasks, and how to jointly train more communication tasks or communication tasks with unclear correlation remains to be solved.

[0005] Large Language Model (LLM) is an emerging multi-task learner. Different from traditional deep learning methods, large models have shown amazing reasoning ability and generalization performance in various fields such as natural language processing, medical, legal, and financial fields. As the earliest large language model, the GPT series model has shown good performance in various natural language processing tasks. In the field of computer vision, large models such as BeiT have also achieved the best performance in multiple tasks such as target detection, semantic segmentation, and visual question answering. Although there have been attempts to apply pre-trained LLMs to channel prediction tasks through fine-tuning, this method is only suitable for a single task and does not fully exploit the multi-task characteristics of LLM. Therefore, there are still some difficulties in applying pre-trained large language models to multiple tasks in the physical layer of wireless communication systems through fine-tuning. Due to the diverse mapping relationships of physical layer tasks, there are many weakly related tasks, such as channel prediction and positioning enhancement, which increases the difficulty of joint training. In addition, due to the huge domain generation gap between CSI and natural language, and the need to avoid the occurrence of mutual exclusion caused by weak correlation between tasks, these characteristics of the physical layer of wireless communication systems further increase the difficulty of network design. Summary of the invention

[0006] In order to overcome the shortcomings of the above-mentioned prior art, the present invention proposes a physical layer multi-task processing solution for a multiple-input multiple-output-orthogonal frequency division multiplexing (MIMO-OFDM) wireless communication system based on a pre-trained large language model. The designed physical layer multi-task network model of the wireless communication system is trained on a physical layer multi-task mixed data set, thereby achieving high-precision physical layer multi-task effects and efficient training process.

[0007] The present invention is a technology applied to a MIMO-OFDM wireless communication system to fine-tune a pre-trained large language model so that it has the ability to solve multiple tasks at the physical layer, wherein the physical layer tasks include but are not limited to tasks such as wireless communication system channel estimation, channel prediction, beam management, and positioning enhancement. The present invention considers a dual-mode communication system with both low-frequency and millimeter-wave wireless communication links. A large multi-task model of the physical layer of the wireless communication system is constructed based on a pre-trained large language model, and the large multi-task model of the physical layer is deployed on the base station side. It is trained through user historical data, and online reasoning is performed after the training is completed. It can be used to complete multiple tasks such as channel estimation, channel prediction, distance estimation, path loss estimation, and beamforming. The input of the large multi-task model of the physical layer is the user uplink low-frequency pilot signal at the first P moments, and the output corresponds to the target output of each task, that is, the user uplink low-frequency CSI at the first P moments, the user downlink low-frequency CSI at the last L moments, the distance and path loss of the user from the base station, and the millimeter-wave downlink beam vector. By integrating multiple tasks into one model, the reasoning efficiency of the model is improved while maintaining the reasoning accuracy unchanged, which is convenient for model maintenance and system update.

[0008] To achieve the above-mentioned purpose, the present invention mainly includes constructing a wireless communication system physical layer multi-task network model based on a pre-trained large language model (GPT-2), including a pre-processing module, a multi-task adapter module, a pre-trained large model backbone network module and a multi-task output module, which is deployed on the base station side, and multiple physical layer tasks are completed through online reasoning of the user uplink low-frequency pilot signal. The GPT-2 model adopts an open source large language model based on the Transformer architecture developed by OpenAI. In order to improve the efficiency of multi-task learning of the model, most of the parameters of the GPT-2 model are frozen, and only the linear layer of the feedforward network is fine-tuned by the MoE-Lora designed by the present invention, wherein MoE-Lora refers to a multi-task fine-tuning method combining a hybrid expert model and Lora fine-tuning. The present invention applies the multi-task fine-tuning method to the physical layer tasks of the wireless communication system for the first time, and combines the pre-trained large model with multi-task learning to achieve the effect of efficient multi-task fine-tuning of the large model. At the same time, in view of the characteristics of the physical layer tasks, the present invention designs a multi-task expert weight distribution network to better dynamically combine the decisions of the expert network.

[0009] The solution for multi-tasking in the physical layer of a wireless communication system based on a pre-trained large language model includes the following steps:

[0010] 1) The user obtains the MIMO-OFDM channel prediction training data set through field channel measurement or simulation. Each sample includes the pilot signal received by the uplink low-frequency link at the first P time sampling points in a continuous period of time (model input), the CSI data of the downlink low-frequency link at the last L time sampling points, the distance and path loss between the user and the base station, and the millimeter wave downlink optimal beam number (model output). The sampling interval of each time sampling point is the same.

[0011] 2) Construct a physical layer multi-task network model of a wireless communication system based on a pre-trained large language model (GPT-2), including a pre-processing module, a multi-task adapter module, a pre-trained large model backbone network module and a multi-task output module. The network input is the user's uplink low-frequency pilot signal at the first P moments (the pilot signal received by the uplink low-frequency link at the first P time sampling points in the training set), and outputs the target output corresponding to each task, i.e., the user's uplink low-frequency CSI at the first P moments, the user's downlink low-frequency CSI at the last L moments, the user's distance from the base station and the path loss, and the millimeter wave downlink beam vector. The constructed physical layer multi-task network is trained on the training data set in step 1).

[0012] 3) Deploy the trained wireless communication system physical layer multi-task network model to the base station side of the actual wireless communication system scenario, and complete multiple physical layer tasks based on the user's uplink low-frequency pilot signal through online reasoning.

[0013] Through the above steps, the physical layer multi-task processing on the base station side of the wireless communication system is realized.

[0014] Compared with the prior art, the present invention has the following beneficial effects:

[0015] The present invention provides a physical layer multi-task processing solution for a wireless communication system based on a pre-trained large language model. Through network design and fine-tuning training, the pre-trained large language model is cross-modally applied to multiple tasks at the physical layer, making full use of the powerful modeling and generalization capabilities of the large language model, and achieving good performance in various tasks. At the same time, compared with the previous method of designing the network for each module separately, the overall design complexity of the model is greatly reduced, and the process of model training is simplified. The technology proposed by the present invention has the following advantages:

[0016] (I) Task processing accuracy: The large-model multi-task network extracts appropriate features for each task through the preprocessing module, which makes it easier for the model to capture task characteristics, and aligns the task features of each task with the semantic feature space of the pre-trained large model through the multi-task adapter module, and the constructed multi-task output head module selects the most appropriate model architecture for each task. These specially designed network structures improve the accuracy of the model on each task.

[0017] (II) Model generalization performance: The backbone network of the large-model multi-task network is a pre-trained GPT-2 model, most of which are frozen. Only MoE-Lora is used to fine-tune the feedforward network. In the face of different task models, the threshold routing network adaptively combines the Lora weights to achieve dynamic adaptation of the network with the input, thereby obtaining its powerful generalization ability. The proposed large-model multi-task network can maintain good model accuracy when the channel data distribution changes.

[0018] (III) Model training overhead: In the face of the needs of multiple physical layer tasks, the large-model multi-task network only needs one training to be deployed to the base station for operation. Compared with the traditional method of selecting a suitable network for each module for training and deployment, the training process of this method is simpler and the deployment difficulty is greatly reduced. At the same time, since the large-model multi-task network is fine-tuned based on the pre-trained large language model, the training cost is comparable to that of ordinary small models (such as CNN, RNN and other networks), and will not incur huge training overhead. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a network architecture block diagram based on a pre-trained large model when the present invention is implemented.

[0020] Figure 2 It is a block diagram of a multi-task adapter module constructed during the specific implementation of the present invention. DETAILED DESCRIPTION

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] The present invention is applied to a typical base station-single-user communication scenario, in which the base station side is equipped with a linear array multi-antenna and the user side is equipped with a single antenna. The signal is in the form of an orthogonal frequency division multiplexing (OFDM) signal. The tasks processed by the model include: channel estimation, channel prediction, Sub-6G assisted millimeter wave beamforming, user distance estimation, and path loss estimation. During the operation of the model, the user side sends an uplink low-frequency pilot signal to the base station. The base station obtains the uplink low-frequency CSI at the current moment, the downlink low-frequency CSI at the future moment, the millimeter wave beamforming vector, the user distance information, and the path loss information through the proposed physical layer multi-task model, thereby efficiently completing the downlink data transmission task.

[0023] The present invention includes three steps: data set construction, network construction and training, and real-time deployment. The specific working steps are as follows:

[0024] S10: Construct MIMO-OFDM multi-task hybrid dataset;

[0025] The data sets of physical layer tasks of multiple wireless communication systems are merged so that the subsequent model can perform multi-task joint training on the obtained multi-task mixed data set. Each sample in the multi-task mixed data set includes the pilot signal received by the uplink low-frequency link at the first P time sampling points in a continuous period of time, the CSI data of the downlink low-frequency link at the last L time sampling points, the distance and path loss between the user and the base station, and the millimeter wave downlink optimal beam number, and the sampling interval of each time sampling point is the same.

[0026] S20: Construct a physical layer multi-task network model and train it on the data set in S10. The physical layer multi-task network includes a preprocessing module, a multi-task adapter module, a pre-trained large model backbone network module, and a multi-task output module. The input of the network is the user uplink low-frequency pilot signal at the first P moments, and the output corresponds to the target output of each task, that is, the user uplink low-frequency CSI at the first P moments, the user downlink low-frequency CSI at the last L moments, the millimeter wave downlink beam vector, the distance between the user and the base station, and the path loss.

[0027] S30: deploy the trained network to an actual scenario or apply it to a test set for testing, input the uplink low-frequency pilot signal data of the user at P historical moments, and output the output data corresponding to each task.

[0028] In step S10: according to actual needs, it is necessary to determine system-related parameters and data set construction. This step includes the following processes S11-S12:

[0029] S11: First, determine the data set parameters, namely, the number of antennas in the low-frequency band and the millimeter-wave band, the center frequency, the number and interval of subcarriers, the CSI time domain sampling interval and the number of sampling points, and the configuration of the millimeter-wave band communication codebook. Assume that the low-frequency band on the base station side is equipped with N t =N h ×1 planar array antenna, millimeter wave band equipped with The user side is equipped with a single antenna, the time sampling interval is ΔT, the number of uplink low-frequency pilot signal sampling points is P, and the number of downlink low-frequency CSI sampling points is L. The uplink and downlink frequencies of the low-frequency band are f u and f d The bandwidth is the same, both use OFDM symbols, both contain K subcarriers and the subcarrier spacing is Δf, and the uplink and downlink frequencies in the millimeter wave band are and And the bandwidth is the same, both use OFDM symbols, and both contain subcarriers and the subcarrier spacing is Low-frequency communication uses digital precoding, and millimeter-wave communication uses analog precoding. The uplink pilot uses a comb-shaped pilot distribution with a pilot interval of m subcarriers. The millimeter-wave communication codebook uses a super-resolution discrete Fourier codebook C = {c1, c2, ... c B},in represents the b-th steering vector, and B represents the number of codebooks.

[0030] S12: Use Γ={T1,T2,…T N} represents the set of tasks to be processed by the model and the corresponding sub-datasets of the tasks, where N represents the total number of tasks. In this example, N = 5. For each task T i , each corresponds to a sub-dataset D i ={TI i ,TO i}, where TI i Represents the input data of the i-th task, TO i Represents the output data of the i-th task. Denotes the CSI of the kth subcarrier and the i-th time point in the downlink, and is expressed as Indicates the pilot signal of the kth subcarrier and the i-th time point in the uplink, and is expressed as represents the CSI of the kth subcarrier and the i-th time point in the uplink, then the downlink CSI data at all frequencies at the i-th time point is expressed as

[0031]

[0032] The value range of i is an integer in [P+1,P+L]. Similarly, the uplink CSI data at all frequencies at the i-th time and the uplink pilot data at the pilot frequency at the i-th time can be expressed as

[0033]

[0034] The value range of i is an integer in [1,P]. represents the optimal millimeter wave beamforming vector of the user at the i-th moment. i ∈C 1 represents the distance between the user and the base station at the i-th moment. i ∈C 1 represents the path loss of the user at the i-th moment. Therefore, for the channel estimation task,

[0035]

[0036] For the channel prediction task,

[0037]

[0038] For Sub-6G assisted millimeter wave beamforming tasks,

[0039]

[0040] For the distance estimation task,

[0041]

[0042] TO4=l P ∈C 1 ,

[0043] For the path loss estimation task,

[0044]

[0045] TO5=pl P ∈C 1

[0046] Step S20: construct a physical layer multi-task network, input TI = {TI1, TI2, ... TI N}, output TO={TO1,TO2,…TO N This step mainly includes the following processes S21 to S26:

[0047] Step S21: Since the channel characteristics required by each task are different, unified preprocessing is not conducive to fully utilizing the characteristics of each task. Therefore, a corresponding preprocessing function is designed for each task to preprocess the input data. The data after preprocessing of the i-th task is recorded as Right now

[0048]

[0049] The prediction processing operation of the channel estimation and channel prediction tasks is to tokenize the CSI at each moment, that is, to flatten the spatial and frequency domain features of the CSI as follows:

[0050]

[0051] Where flatten(X,n) means flattening the nth dimension and all subsequent dimensions of the tensor X into one dimension. Since the beamforming task, distance estimation task, and path loss estimation task require channel angular domain features, the CSI data will be transformed to convert the spatial domain CSI into the angular domain CSI.

[0052]

[0053] in N t dimensional DFT matrix.

[0054] Step S22: This step performs feature alignment on the preprocessed data so that the feature space of each task is aligned with the semantic feature space of the pre-trained large model. In order to fully consider the unique characteristics of each task, a multi-task adapter module is defined for each task, denoted as Adapter. in,i , its specific network architecture is as follows Figure 2 As shown in Figure 2, the multi-task adapter module includes a linear mapping layer and a residual connection network. The input and output are described in detail below.

[0055] The input of the multi-task adapter module is the preprocessed data of the i-th task First, the dimensions are simply aligned through the linear mapping layer and mapped to Where L is the length of the sequence input to the pre-trained LLM, and 768 is the hidden layer dimension of the selected pre-trained GPT-2.

[0056]

[0057] Linear(·) contains at least two full connection operations, that is, linearly mapping the first dimension and the second dimension to the specified dimension respectively. Then, the residual connection network is used to align the feature semantics to obtain the task representation of the i-th task. Right now:

[0058]

[0059] Res(·) contains two 1D convolution kernels and an activation function Relu, the convolution kernel size is 3, and the step size is 1. The above steps can be simplified as

[0060]

[0061] Step S23: This step inputs the aligned features into the backbone network, i.e., the pre-trained large language model for processing, corresponding to Figure 1 The backbone network adopts the GPT-2 architecture, which is composed of multiple layers of transformer decoders. Each layer of the decoder includes a multi-head attention layer, an addition and normalization layer, and a feedforward layer. In actual deployment, the number of layers of GPT-2 is N. L It can be adjusted flexibly, and the MoE-Lora fine-tuning method is used to fine-tune the linear layer weights in the feedforward layer. The final output of the large language model is expressed as Right now

[0062]

[0063] Where LLM(·) represents the forward reasoning process of the pre-trained large language model.

[0064] Step S24: This step sends the features output by the large model to the multi-task adapter again to further achieve feature alignment and obtain feature The specific operation is the same as step S22, which can be expressed as

[0065]

[0066] Adapter out,i is the multi-task adapter on the output side corresponding to the i-th task.

[0067] Step S25: This step mainly processes the output of the adapter to obtain the output results TO' of each task. i The specific operation is to use the multi-task output header module to out,i Output features of the i-th task The two dimensions of are mapped to the dimensions of the output results of the i-th task respectively. Since the output dimensions of beamforming, distance estimation, and path loss estimation are low, the MLP architecture is used as the multi-task output head module, while the output dimensions of channel estimation and channel prediction are high, so the CNN architecture is used as the multi-task output head module. The specific operation can be expressed as

[0068]

[0069] Output i Represents the output header of the i-th task.

[0070] The output head network is the network after the adapter, which can map the features of each task to the target output corresponding to the task. For example, in the distance estimation task, the output head (MLP) maps the L*D-dimensional feature vector to a 1*1 distance estimate [L=K / m,D=768].

[0071] Step S26: Steps S21 to S25 describe in detail each module of the network, and the input is {TI i ,i=1,2,…,5}, the output is {TO′ i , i=1,2,…,5}. The weighted sum of the losses of each task is used as the overall loss Loss of network training, that is,

[0072]

[0073] Among them, f i l Represents the loss function of task i. Except for the beam management task which uses the cross entropy loss function, all other tasks use NMSE as the loss function. i Indicates the loss ratio of each task, the default value is

[0074] In step S30, the network model trained in step S20 is deployed to the actual scene for testing. This step includes the following processes S31-S32:

[0075] Step S31: First, convert the uplink low-frequency pilot signal data of the user at P historical moments into a tensor format for network input. Then input the converted data into the trained network model for testing to obtain network output data corresponding to each physical layer task.

[0076] Step S32: Convert the tensor data output by the network into a matrix format, and then perform subsequent precoding and other operations based on the various channel information obtained to achieve online reasoning.

[0077] The present invention can be used to complete multiple tasks such as channel estimation, channel prediction, distance estimation, path loss estimation, beamforming, etc. The model training cost is low, the processing accuracy of each task can be improved, and it has strong generalization ability.

[0078] It should be noted that the purpose of publishing the embodiments is to help further understand the present invention, but those skilled in the art can understand that various substitutions and modifications are possible without departing from the scope of the present invention and the appended claims. Therefore, the present invention should not be limited to the contents disclosed in the embodiments, and the scope of protection claimed by the present invention shall be subject to the scope defined in the claims.

Claims

1. A wireless communication physical layer multi-task processing method based on a pre-trained large language model, characterized in that: A multi-task large model of the physical layer of the wireless communication system is constructed based on the pre-trained large language model and deployed on the base station side. It is trained with user historical data and then online inference is performed after training to complete multiple tasks of the physical layer of the wireless communication system. The steps include: 1) Obtain a multi-task mixed data set of a wireless communication system as a model training data set; Each data set sample includes the pilot signal received by the uplink low-frequency link at the first P time sampling points in a continuous period of time, the channel state information (CSI) data of the downlink low-frequency link at the last L time sampling points, as well as the distance and path loss between the user and the base station, and the optimal beam number of the millimeter wave downlink; 2) Building a physical layer multi-task network model of a wireless communication system based on a pre-trained large language model, including a pre-processing module, a multi-task adapter module, a pre-trained large model backbone network module and a multi-task output module; The input of the network model is the user uplink low-frequency pilot signal at the first P moments, and the output corresponds to the target output of each physical layer task, that is, the user uplink low-frequency CSI at the first P moments, the user downlink low-frequency CSI at the last L moments, the distance between the user and the base station and the path loss, and the millimeter wave downlink beam vector; Using the training data set obtained in step 1), the constructed wireless communication system physical layer multi-task network model is trained to obtain a trained wireless communication system physical layer multi-task network model; 3) Deploy the trained wireless communication system physical layer multi-task network model to the base station side of the wireless communication system, and complete multiple physical layer tasks based on the user's uplink low-frequency pilot signal online reasoning; Through the above steps, the physical layer multi-task processing on the base station side of the wireless communication system can be realized.

2. The wireless communication physical layer multi-task processing method based on a pre-trained large language model as claimed in claim 1, characterized in that: The wireless communication system is a base station-single user communication scenario, where the base station side is equipped with linear array multiple antennas and the user side is equipped with a single antenna; The signal is in the form of orthogonal frequency division multiplexing signal; the wireless communication physical layer tasks processed by the physical layer multi-task network model of the wireless communication system include: channel estimation, channel prediction, Sub-6G assisted millimeter wave beamforming, user distance estimation, and path loss estimation.

3. The wireless communication physical layer multi-task processing method based on a pre-trained large language model as claimed in claim 1, characterized in that: In step 1), the sampling intervals of the time sampling points are the same; constructing a wireless communication system physical layer multi-task hybrid data set specifically comprises merging data sets of multiple wireless communication system physical layer tasks; including: S11: Determine the parameters of the training data set, including the number of antennas in the low frequency band and the millimeter wave band, the center frequency, the number and interval of subcarriers, the CSI time domain sampling interval and the number of sampling points, and the configuration of the millimeter wave band communication codebook; S12: Define the wireless communication system physical layer task set Γ = {T1, T2, ..., T i , ..., T N } and the sub-datasets of the corresponding tasks; Each task T in the physical layer task set Γ i , each corresponding to a sub-dataset D i , expressed as: D i ={TI i ,TO i }, where TI i Represents the input data of the i-th task, TO i Represents the output data of the i-th task.

4. The wireless communication physical layer multi-task processing method based on a pre-trained large language model as claimed in claim 3, characterized in that: Step 2) construct a physical layer multi-task network model of the wireless communication system based on the pre-trained large language model. The input of the model is TI = {TI1, TI2, ...TI N }, the output is TO = {TO1, TO2, ...TO N }; Model construction includes the following steps: S21: Design corresponding preprocessing functions to preprocess the input data of different tasks; S22: performing a feature alignment operation on the preprocessed data so that the feature space of each task is aligned with the semantic feature space of the physical layer multi-task network model constructed based on the pre-trained large model; A multi-task adapter module is defined for each task separately; the multi-task adapter includes a linear mapping layer and a residual connection network; the pre-processed data of the i-th task is input into the multi-task adapter, and the dimensions are simply aligned through the linear mapping layer; Then, the feature semantics is aligned through the residual connection network to obtain the task representation of the i-th task; S23: Input the aligned features into the backbone network, i.e., the pre-trained large language model for processing; The backbone network is composed of multiple layers of transformer decoders, each of which includes a multi-head attention layer, an addition and normalization layer, and a feedforward layer. Adjust the number of layers of the backbone network and use the fine-tuning method to fine-tune the linear layer weights in the feed-forward layer; After the forward reasoning process of the large language model, the final output representation of the large language model is obtained; S24: Send the features output by the large language model to the multi-task adapter again to further achieve feature alignment, and output the further aligned features; S25: using the multi-task output head module to further process the output features of the multi-task adapter, and respectively map the two dimensions of the adapter output features to the dimensions of the corresponding task output data to obtain the output results of each task; S26: Use the weighted sum of the losses of each task as the overall loss function for network model training to perform model training, and obtain a trained multi-task network model of the physical layer of the wireless communication system based on the pre-trained large language model.

5. The wireless communication physical layer multi-task processing method based on a pre-trained large language model as claimed in claim 4, characterized in that: The input to the multitasking adapter is First, the dimensions are simply aligned through the linear mapping layer and mapped to Where L is the length of the sequence input to the pre-trained LLM, and 768 is the hidden layer dimension of the selected pre-trained GPT-2. Then, the feature semantics are aligned through the residual connection network to obtain It is expressed as: Adapter out，i is the multi-task adapter on the output side corresponding to the i-th task; Mapping through the linear mapping layer includes at least two full connection operations, that is, linearly mapping the first dimension and the second dimension to the specified dimension respectively; the residual connection network contains two 1-dimensional convolution kernels and an activation function Relu.

6. The wireless communication physical layer multi-task processing method based on a pre-trained large language model as claimed in claim 4, characterized in that: In S23, the backbone network adopts the GPT-2 large language model; the fine-tuning method adopts the MoE-Lora fine-tuning method.

7. The wireless communication physical layer multi-task processing method based on a pre-trained large language model as claimed in claim 4, characterized in that: S24 sends the features output by the large model to the multi-task adapter again to further achieve feature alignment, which is expressed as: in, To further realize the final output features after feature alignment; Adapter out，i is the multi-task adapter on the output side corresponding to the i-th task.

8. The wireless communication physical layer multi-task processing method based on a pre-trained large language model as claimed in claim 1, characterized in that: When using the trained wireless communication physical layer multi-task processing model to predict offline, the uplink low-frequency pilot signal data of the user at P historical moments is first converted into the network input format through tensorization operation.

Images