Wireless optical communication equalization method and system based on base large model

Abstract

The invention discloses a wireless optical communication equalization method and system based on a base large model, and relates to the technical field of wireless optical communication, and the method comprises the steps: when a first user side transmits a preset known signal to a second user side through a wireless optical communication system, according to the preset known signal and a signal received by the second user side, carrying out the equalization of the wireless optical communication according to the preset known signal; determining a current characteristic of a channel of the wireless optical communication system; inputting the current characteristics of the channel of the wireless optical communication system into the trained base large model to obtain a target equalization parameter; adjusting a current communication index of a channel of the wireless optical communication system according to the target equalization parameter; and when the first user side continues to send the to-be-sent signal to the second user side, sending the to-be-sent signal through the channel after state adjustment. The method can adapt to continuously changing channel environments in real time, and the communication stability of a wireless optical communication system is improved to the maximum extent.

Description

Background Art

[0002] Visible light wireless communication, also known as "light fidelity technology", is called Light Fidelity (Li-Fi for short) in English. It is a new wireless transmission technology that uses visible light spectrum (such as the light emitted by a light bulb) for data transmission. The Li-Fi communication system consists of a light source and a receiving end. The light source generally uses a light-emitting diode (LED for short) to convert electrical signals into high-frequency flickering light signals that are invisible to the naked eye; the receiving end uses a photodiode to convert the received light signal emitted by the light source LED into an electrical signal, thus realizing the transmission of wireless signals. The system composition is as follows Figure 1 shown.

[0003] Wireless optical communication systems use light waves to transmit data. With the increasing shortage of wireless spectrum resources, wireless optical communication systems have attracted great attention due to their potential for high-speed data transmission, which is several times higher than that of radio frequency communication, as well as their advantages of no electromagnetic interference and high security. However, due to channel distortion caused by ambient light interference, multipath propagation, optical signal fading, and especially the nonlinear characteristics of LEDs, the performance of wireless optical communication systems is significantly affected. Visible light communication system equalization includes pre-equalization and post-equalization. Pre-equalization is to add an equalization module to the front stage of the LED, that is, the transmitter, to improve its frequency response and expand the bandwidth. The purpose is to correct the LED signal at the transmitter so that it can better adapt to the channel characteristics during transmission, thereby improving the performance of the communication system. The main function of post-equalization is to compensate for the channel at the receiving end to compensate for the distortion generated by the signal when passing through the channel to ensure that the signal quality meets the requirements. Traditional equalization methods, such as linear equalization and decision feedback equalization, are designed based on certain assumptions about channel conditions and are suitable for environments with relatively stable channel conditions. However, in the face of the highly dynamic characteristics of the wireless optical communication system channel, it may not be possible to achieve the best equalization performance. Therefore, an upgraded adaptive equalization method that can adapt to high dynamics more quickly is needed. The equalization parameters can be adjusted dynamically and in real time according to the current channel conditions to more effectively cope with the dynamic characteristics of the wireless optical communication system channel, thereby improving system performance and reliability. The existing technologies are as follows:

[0004] The invention patent with the publication number of "CN116455706A" and the subject name of "A design method and system for channel estimation and equalization" discloses: the received data is time synchronized, cyclic prefix removed and OFDM demodulated at the receiving end, and then processed by the channel estimation and equalization module, and then through channel decoding, and finally the received bits are output; the channel estimation and equalization module includes three parts: data merging and real and imaginary part splitting, channel estimation and equalization deep convolutional neural network and masker; when each channel of data is input into the module, the data is first merged, and then the real and imaginary parts of the merged data are split; then it is input into the channel estimation and equalization deep convolutional neural network to obtain the data after channel estimation and channel equalization; then it is input into the masker, and the corresponding mask length is configured according to the order of QAM, and finally the processed data is output from the module, so as to use fewer resources to overcome the rapidly changing channel and improve the accuracy of channel estimation and equalization, but the technical solution in the patent has multiple shortcomings, which makes the technical solution in the patent insufficient to cope with complex channel environments. The specific shortcomings are as follows:

[0005] 1) Complex system design:

[0006] The architecture design of CNN needs to be optimized for specific tasks, including multi-layer convolution, pooling, activation function selection, etc. This increases the complexity of system design, especially when facing a changing channel environment, where the model structure may need to be constantly adjusted and optimized. In addition, for adaptive equalization tasks, it is particularly important to design an effective feature extractor, and CNN requires manual design and adjustment of feature extractors to capture channel state information and received signal characteristics, which further increases the design complexity.

[0007] 2) Poor versatility:

[0008] CNNs usually need to be trained on specific tasks and datasets and lack a broad pre-training foundation. In different environments and tasks, CNNs have weak versatility and migration capabilities.

[0009] 3) Limited model size:

[0010] Due to computing resource and storage constraints, the number of parameters and layers of CNN models are usually limited. This causes the model to perform poorly in characterizing complex channel characteristics and capturing nonlinear relationships, and it is difficult to cope with complex and dynamically changing channel environments. CNN is mainly good at capturing local features and expanding the receptive field by stacking multiple convolutional layers, but it has limitations in capturing global information and long-distance dependencies, which is not conducive to dealing with complex channel environments. Summary of the invention

[0011] The technical problem to be solved by the present invention is to address the deficiencies of the prior art and specifically provide a wireless optical communication equalization method and system based on a large base model, as follows:

[0012] 1) In the first aspect, the present invention provides a wireless optical communication equalization method based on a large base model, and the specific technical solution is as follows:

[0013] When the first user terminal sends a preset known signal to the second user terminal through the wireless optical communication system, the current characteristics of the channel of the wireless optical communication system are determined according to the preset known signal and the signal received by the second user terminal;

[0014] Input the current characteristics of the channel of the wireless optical communication system into the trained base large model to obtain the target equalization parameters;

[0015] According to the target equalization parameter, the current communication index of the channel of the wireless optical communication system is adjusted by a programmable driving circuit of the wireless optical communication system;

[0016] When the first user end continues to send the signal to be sent to the second user end, the signal to be sent is sent through the channel with the state adjusted.

[0017] The beneficial effects of a wireless optical communication equalization method based on a base large model provided by the present invention are as follows:

[0018] The large base model has better versatility and migration capabilities. During the training process, it can effectively capture the characteristics of the channel, and can well capture global information and long-distance dependencies, improve the accuracy of the output equalization parameters (i.e., target equalization parameters), and does not require manual design and adjustment, reducing complexity. Moreover, when facing a changing channel environment, it reduces the frequency of adjusting and optimizing the large base model, and can adapt to the changing channel environment in real time, thereby maximizing the communication stability of the wireless optical communication system.

[0019] On the basis of the above scheme, the wireless optical communication equalization method based on the base large model of the present invention can also be improved as follows.

[0020] Furthermore, the process of obtaining the trained large base model includes:

[0021] Establishing a sample set including a plurality of samples, wherein a set of characteristics of a channel of a wireless optical communication system and an equalization parameter corresponding to the set of characteristics constitute one sample;

[0022] Based on the sample set, the base large model is pre-trained to obtain a pre-trained base large model;

[0023] Fine-tune the pre-trained base large model to obtain a trained base large model.

[0024] Furthermore, it also includes:

[0025] Multiple sets of features of the channels in the sample set are obtained from historical transmitted signals, historical received signals, historical channel model data, and historical equalized signals.

[0026] Furthermore, the base large model is BERT, GPT, RoBERTa, T5 or Sora.

[0027] 2) In the second aspect, the present invention also provides a wireless optical communication equalization system based on a large base model, and the specific technical solution is as follows:

[0028] It includes a channel characteristic determination module, a target equalization parameter acquisition module, a channel index adjustment module and a communication module;

[0029] The channel characteristic determination module is used to: when the first user terminal sends a preset known signal to the second user terminal through the wireless optical communication system, determine the current characteristics of the channel of the wireless optical communication system according to the preset known signal and the signal received by the second user terminal;

[0030] The target equalization parameter acquisition module is used to: input the current characteristics of the channel of the wireless optical communication system into the trained base large model to obtain the target equalization parameters;

[0031] The channel index adjustment module is used to: adjust the current communication index of the channel of the wireless optical communication system through the programmable driving circuit of the wireless optical communication system according to the target equalization parameter;

[0032] The communication module is used for: when the first user terminal continues to send the signal to be sent to the second user terminal, sending the signal to be sent through the channel after the state is adjusted.

[0033] On the basis of the above scheme, the wireless optical communication equalization system based on the base large model of the present invention can also be improved as follows.

[0034] Furthermore, a model determination module is included, and the model determination module is used to:

[0035] Establishing a sample set including a plurality of samples, wherein a set of characteristics of a channel of a wireless optical communication system and an equalization parameter corresponding to the set of characteristics constitute one sample;

[0036] Based on the sample set, the base large model is pre-trained to obtain a pre-trained base large model;

[0037] Fine-tune the pre-trained base large model to obtain a trained base large model.

[0038] Furthermore, it also includes a data acquisition module, which is used to obtain multiple sets of characteristics of the channels in the sample set from historical transmitted signals, historical received signals, historical channel model data and historical equalized signals.

[0039] Furthermore, the base large model is BERT, GPT, RoBERTa, T5 or Sora.

[0040] 3) In a third aspect, the present invention also provides an adaptive equalization wireless optical communication system, including a wireless optical communication system, wherein an adaptive equalizer of the wireless optical communication system is used to implement any of the above-mentioned wireless optical communication equalization methods based on a large base model.

[0041] 4) In a fourth aspect, the present invention further provides an electronic device, comprising a processor, the processor being coupled to a memory, the memory storing at least one computer program, and the at least one computer program being loaded and executed by the processor so that the electronic device implements any of the above-mentioned wireless optical communication equalization methods based on the base large model.

[0042] 5) In a fifth aspect, the present invention also provides a computer-readable storage medium, in which at least one computer program is stored, and at least one computer program is loaded and executed by a processor so that the computer implements any of the above-mentioned wireless optical communication equalization methods based on the base large model.

[0043] It should be noted that the beneficial effects achieved by the technical solutions of the second to fifth aspects of the present invention and the corresponding possible implementation methods can be found in the above-mentioned technical effects of the first aspect and its corresponding possible implementation methods, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0045] Figure 1 It is a structural schematic diagram of a visible light communication system;

[0046] Figure 2 A schematic diagram of a flow chart of a wireless optical communication equalization method based on a large base model according to an embodiment of the present invention;

[0047] Figure 3 This is one of the application schematic diagrams of a wireless optical communication equalization method based on a large base model of the present invention;

[0048] Figure 4 This is a second application schematic diagram of a wireless optical communication equalization method based on a large base model of the present invention;

[0049] Figure 5A flowchart of the process of obtaining a trained large base model;

[0050] Figure 6 It is a structural schematic diagram of a wireless optical communication equalization system based on a large base model according to an embodiment of the present invention;

[0051] Figure 7 The figure is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0053] like Figure 2 As shown, a wireless optical communication equalization method based on a base large model according to an embodiment of the present invention includes the following steps:

[0054] S1. When a first user terminal sends a preset known signal to a second user terminal through a wireless optical communication system, a current characteristic of a channel of the wireless optical communication system is determined according to the preset known signal and a signal received by the second user terminal;

[0055] S2, inputting the current characteristics of the channel of the wireless optical communication system into the trained base large model to obtain the target equalization parameters;

[0056] Among them, the base model generally refers to the infrastructure or framework used to build larger and more complex models in the field of machine learning and artificial intelligence. The model is based on deep learning technology and is trained using large amounts of data to perform various tasks. The base model is usually used in fields such as natural language processing, computer vision, and speech recognition. Its original design was to solve language and vision-related problems. A typical example of a base model is the GPT (Generative Pre-trained Transformer) model series, which was developed by OpenAI and uses the Transformer architecture to obtain general language understanding capabilities through large-scale pre-training. These models can be applied to various natural language processing tasks, such as text generation, language understanding, translation, etc. The base model can also be BERT, GPT, RoBERTa, T5 or Sora, and can also be set according to actual conditions.

[0057] S3, adjusting the current communication index of the channel of the wireless optical communication system through the programmable driving circuit of the wireless optical communication system according to the target equalization parameter;

[0058] S4. When the first user end continues to send the signal to be sent to the second user end, the signal to be sent is sent through the channel after the state is adjusted.

[0059] Optionally, sending a preset known signal and sending a signal to be sent may be performed alternately, or after sending a preset known signal once (i.e. obtaining a target equalization parameter and adjusting a communication index once), the signal to be sent may be sent multiple times, which may be set according to actual conditions.

[0060] Among them, the wireless optical communication system specifically includes a transmitting system, a receiving system and an adaptive equalizer. The transmitter uses an LED lamp to send an optical signal, and the receiver uses a photodetector PD to receive the optical signal after passing through the channel. The adaptive equalizer outputs the equalization parameter, that is, the target equalization parameter, through the trained base large model. The specific explanation is as follows:

[0061] 1) Launch system:

[0062] The transmitting system includes a light source, a programmable driving circuit, a modulation circuit and a signal source. The modulation circuit is used to modulate the input data (i.e., the preset known signal and the signal to be sent) onto the optical signal. The programmable driving circuit is used to dynamically control the amplitude of the optical signal and communication parameters and other communication indicators.

[0063] 2) Receiving system:

[0064] The receiving circuit includes a photodetector and a signal processing circuit (such as a filter, an attenuator, a transimpedance amplifier, etc.). The photodetector converts the received optical signal into an electrical signal, and the signal processed by the signal processing circuit is used as a received signal (referring to the signal received by the second user end), and is fed back to the adaptive equalizer at the same time. The adaptive equalizer determines the current characteristics of the channel of the wireless optical communication system based on the preset known signal and the signal received by the second user end; the current characteristics of the channel of the wireless optical communication system are input into the trained base large model to obtain the target equalization parameters, that is, to adjust the equalization parameters, and then adjust the current communication indicators of the channel of the wireless optical communication system according to the target equalization parameters.

[0065] 3) Adaptive Equalizer:

[0066] The input trained base large model is set in the adaptive equalizer. When the first user terminal sends a preset known signal to the second user terminal through the wireless optical communication system, the current characteristics of the channel of the wireless optical communication system are determined according to the preset known signal and the signal received by the second user terminal; the current characteristics of the channel of the wireless optical communication system are input into the trained base large model to obtain the target equalization parameters; based on the target equalization parameters, the current communication indicators of the channel of the wireless optical communication system are adjusted through the programmable driving circuit, specifically the current communication parameters and modulation signal amplitude of the channel of the wireless optical communication system are adjusted to realize the intelligent optimization of the LED modulation bandwidth, so as to optimize the communication rate of the wireless optical communication system. The trained base large model effectively improves the performance of the unstable system through efficient calculation.

[0067] like Figure 3 As shown, the present invention is described:

[0068] 1) User terminal A is used as the first user terminal, user terminal B is used as the second user terminal, and the wireless optical communication system is specifically a wireless optical communication system. The trained base large model is deployed at the transmitting end (user terminal A and user terminal B are both transmitting ends), then:

[0069] When user terminal A sends a preset known signal to user terminal B through the first VLC transceiver-in-one communication device, user terminal B receives the preset known signal through the second VLC transceiver-in-one communication device, and determines the current characteristics of the channel according to the preset known signal and the signal received by user terminal B; inputs the current characteristics of the channel of the wireless optical communication system into the trained base large model to obtain the target equalization parameters; adjusts the current communication indicators of the channel of the wireless optical communication system according to the target equalization parameters; when user terminal A continues to send the signal to be sent to user terminal B, the signal to be sent is sent through the channel after the state is adjusted.

[0070] 2) With the user terminal B as the first user terminal and the user terminal A as the second user terminal, the wireless optical communication system is specifically a wireless optical communication system, then:

[0071] When user terminal B sends a preset known signal to user terminal A through the second VLC transceiver-in-one communication device, user terminal A receives the preset known signal through the first VLC transceiver-in-one communication device, and determines the current characteristics of the channel according to the preset known signal and the signal received by user terminal A; inputs the current characteristics of the channel of the wireless optical communication system into the trained base large model to obtain the target equalization parameters; adjusts the current communication indicators of the channel of the wireless optical communication system according to the target equalization parameters; when user terminal B continues to send the signal to be sent to user terminal A, the signal to be sent is sent through the channel after the state is adjusted.

[0072] In this way, two-way communication between user terminal A and user terminal B can be achieved. It should be noted that the trained base large model is deployed at the transmitting end, and the transmitting end can be used as the executor of the wireless optical communication equalization method based on the base large model of the present invention.

[0073] like Figure 4 As shown, the trained large base model can also be deployed at the receiving end; the receiving end can be used as the execution subject of the wireless optical communication equalization method based on the large base model of the present invention.

[0074] Optionally, in the above technical solution, if Figure 5 As shown in the figure, the process of obtaining the trained large base model includes:

[0075] S020, establishing a sample set including a plurality of samples, wherein a set of characteristics of a channel of a wireless optical communication system and equalization parameters corresponding to the set of characteristics constitute one sample;

[0076] S021. Pre-train the base large model based on the sample set to obtain a pre-trained base large model;

[0077] S022. Fine-tune the pre-trained large base model to obtain a trained large base model.

[0078] Optionally, in the above technical solution, it also includes:

[0079] Multiple sets of features of the channels in the sample set are obtained from historical transmitted signals, historical received signals, historical channel model data, and historical equalized signals.

[0080] The process of obtaining the trained base model is described as follows:

[0081] S101, establishing a data set: establishing a data set for describing the input-output relationship of the wireless optical communication system.

[0082] The data set includes input data and output data. The input data includes multiple groups of transmitted signals, received signals, channel model data and equalized signals. The output data includes multiple groups of equalization parameters. Each group of transmitted signals, received signals, channel model data and equalized signals corresponds one by one to each group of equalization parameters. Among them, the equalization parameters determine the modulation bandwidth and frequency response of the LED, the signal-to-noise ratio of the wireless optical communication system and other communication performance.

[0083] The transmitted signal, received signal, channel model data and equalized signal are described as follows:

[0084] 1) Transmitted signal: a signal sent from a transmitting end (the first user end and the second user end can both serve as transmitting ends), that is, an original signal that has not been affected by a channel.

[0085] 2) Received signal: An electrical signal converted from an optical signal received by a receiver (both the first user end and the second user end can serve as a receiver) under the influence of the channel model, which contains signal distortion caused by channel conditions.

[0086] 3) Channel model data: refers to the data generated by the mathematical expression used to characterize the channel model. The channel model parameters correspond one-to-one to the received signal after passing through the channel.

[0087] 4) Equalized signal: The equalized signal is the output signal after adaptive equalization processing, and it should be as close to the original transmitted signal as possible.

[0088] The process of obtaining the channel model data is as follows:

[0089] Through experiments, a large amount of data related to the free space channel of visible light communication (channel model data) is collected, including channel characteristics and signal propagation paths under different environments, different angles, and different light intensities. At the same time, simulation software (such as MATLAB tools) is used to generate a large amount of simulated channel data as channel model data, and data enhancement techniques (such as rotation, scaling, adding noise, etc.) are used to expand the data set to supplement the deficiencies of experimental data. Or visible light communication channel data can be obtained through open source communication knowledge as channel model data.

[0090] The above channel data includes deterministic and random components. The deterministic component represents the known characteristics of the transmission medium (free space), such as the attenuation and dispersion of the optical signal. The random component represents the random fluctuations in the channel response, such as environmental noise and equipment errors. The channel model is specifically:

[0091] ① Fluctuation of light source intensity: The light intensity fluctuates due to factors such as the working state of the light source, power supply voltage, and different divergence angles.

[0092] ② Receiver sensitivity fluctuations: The receiver sensitivity may fluctuate due to changes in environmental factors such as temperature and humidity.

[0093] ③ Changes in the optical path: Changes in the communication distance and the light propagation path may be affected by factors such as air flow and object movement, causing random fluctuations in the channel response.

[0094] ④Multipath effect: In an indoor environment, light may reach the receiver after multiple reflections and refractions. This multipath effect will also cause random fluctuations in the channel response.

[0095] ⑤ Changes in ambient light: Changes in ambient light will also affect channel response. When the ambient light intensity increases, the receiver may be affected by stronger background light noise.

[0096] ⑥Shadow effect: The movement of an object may block the propagation path of light, causing changes in the light intensity received by the receiver.

[0097] S102: Establish a sample set, specifically:

[0098] S1020, preprocessing the data set to convert it into a data format suitable for the base large model, specifically:

[0099] Read, parse and process the input data in the data set, including cleaning, conversion and formatting, to obtain the data after the first processing. The formatted data types include: data types include digital data, text data, model data and other formats.

[0100] Extract key parameters and features from the data after the first processing, including parameter values, experimental results, trend analysis, etc., to obtain the data after the second processing;

[0101] The data format of the second processed data is converted as needed, such as for the ChatGPT data format conversion module, and the extracted data can be converted into natural language text through a natural language generation algorithm to interact with ChatGPT. In this embodiment, the second processed data is converted into data in a format used by the large model of the adaptation base.

[0102] The converted data in the format used by the large model of the adapted base can also be verified and adjusted: verify whether the data generated by the conversion is consistent with expectations, and if not, make necessary adjustments.

[0103] S1021, normalizing the data in the format used by the large model of the adaptation base, and extracting channel features;

[0104] Normalization is performed to ensure consistency in the scale of the data input into the large base model, and representative channel features are extracted, including using existing non-blind channel estimation methods or blind channel estimation methods to solve the amplitude, phase, and time-varying characteristics of the channel transfer function, such as path loss, reflection, scattering, etc., thereby establishing multiple samples, each of which includes a channel feature and an equalization parameter corresponding to the channel feature.

[0105] The sample set is divided into training set, validation set and test set. The training set is used to train the base model, the validation set is used to adjust the parameters and hyperparameters of the base model, and the test set is used to evaluate the final performance of the base model.

[0106] S103, select the large base model:

[0107] Select the appropriate base model architecture according to the task requirements and data characteristics, and select a suitable model from BERT, GPT, RoBERTa, T5 and Sora as the base model for subsequent training. These models (BERT, GPT, RoBERTa, T5 and Sora) have large-scale parameters and extensive pre-training data, which can provide powerful and flexible tools for adaptive equalization of wireless optical communication systems.

[0108] It should be noted that other models can be selected as the large base model according to actual conditions.

[0109] S104, pre-training the large base model selected in S103:

[0110] The pre-training process processes a large amount of unlabeled data, with the goal of capturing the underlying patterns, structures, and semantic knowledge in the sample set. Pre-training uses the Transformer architecture and is usually an unsupervised learning process. The fine-tuning process is performed after pre-training. By further training the pre-trained model on a small amount of labeled data for the new task, the model can learn specific features and rules related to the target task, thereby better adapting to the new task.

[0111] The training process is based on the training data set mentioned above. During the training process, a large number of prompts and output examples are used to "guide" the model to predict the optimal equalization parameters. These parameters are automatically adjusted during the training process, and the adjustment goal is to make the signal received by the receiver as close as possible to the original transmitted signal.

[0112] S105, fine-tuning: fine-tune the pre-trained base large model to obtain a trained base large model.

[0113] The specific fine-tuning process includes forward propagation, back propagation, parameter update, etc. In the forward propagation stage, the model receives input data, that is, the original transmitted signal after format conversion, the signal received by the photodetector after the channel influence, and the channel model parameters, etc., and calculates through all layers of the large model until the predicted output, that is, the predicted equalization parameters, is obtained. This process starts from the input layer and proceeds forward layer by layer. The output of each layer is the input of the next layer. When the data is propagated through all layers, the model will get the predicted output. Then, the model will calculate the loss function, which is a measure of the gap between the predicted output and the ideal output.

[0114] In this embodiment, the SFT (Soft Prompt Tuning) method is used to fine-tune the pre-trained base large model. The SFT method does not directly fine-tune the parameters of the model, but adds a learnable "soft prompt" at the input end of the model. The purpose of fine-tuning is achieved by optimizing this "soft prompt". The specific implementation process is as follows:

[0115] S1050, initialization: Use the pre-trained base model as the basis and do not directly update the equalization parameter θ.

[0116] S1051. Define soft hints. A learnable soft hint vector p is introduced, and the length of p is usually smaller than the length of the input sequence.

[0117] S1052, calculate the loss function. The input sequence and the soft prompt are concatenated to obtain the vector y = [p; θ], which is sent to the base large model to obtain the output, and the loss function L(θ, p) is defined by the mean square error (MSE):

[0118]

[0119] Where N is the number of training data, y i is the i-th item in the vector y, representing the true output of the training data, is the predicted value of the training data.

[0120] S1053, update soft prompts. Use the gradient descent algorithm to update the soft prompt parameters. Since the gradient of the loss function with respect to the model output is the partial derivative of the loss function with respect to each predicted value, the calculation formula is:

[0121]

[0122] Arranged into vector form:

[0123]

[0124] in, and j (j=1,2,…,N) are the predicted values ​​and true outputs of the training data respectively.

[0125] The update process is carried out in the opposite direction of the gradient, with the goal of minimizing the loss function. The parameter update formula is:

[0126]

[0127] Among them, η is the learning rate, which determines the step size of updating parameters in the gradient direction; is the gradient of the loss function L(θ,p) with respect to the equilibrium parameter θ, that is, the partial derivative of the above loss function with respect to each predicted value

[0128] S1054, iterative optimization. Repeat and return to execute S1052 until the loss function L(θ, p) converges, and a trained base large model is obtained.

[0129] S106. Re-evaluate the trained large base model.

[0130] After training, the performance of the trained large base model is evaluated on the previously segmented test set, and evaluation indicators such as accuracy, recall, F1 score and AUC are calculated to ensure that it can accurately predict the optimal equalization parameters. If the evaluation indicators of the trained large base model fail to pass the evaluation, S104 and S105 can be repeatedly executed by expanding the sample set, etc., until the evaluation indicators of the trained large base model pass the evaluation.

[0131] S107. Deploy the trained large base model in the wireless optical communication system.

[0132] The trained base model can be deployed in the adaptive equalizer in the wireless optical communication system to dynamically adjust the output signal of the LED. The specific process includes:

[0133] The adaptive equalizer infers the current real-time channel conditions through the distorted signal received by the wireless optical communication transceiver under the current channel conditions, continuously monitors and outputs the equalization parameters through the results generated by the base large model calculation, so that the signal emitted by the LED is adjusted according to the real-time channel status, improving the signal quality and thus reducing the impact of channel characteristics and noise on the system.

[0134] The present invention demonstrates the design and implementation method of pre-equalization at the transmitting end. This idea and method can also be used for post-equalization at the receiving end, that is, to compensate the channel at the receiving end of the system, thereby compensating for the distortion generated when the signal passes through the channel. By training the base large model, it can identify the channel conditions through the received signal, dynamically adjust the equalization parameters based on reasonable assumptions about the channel conditions, effectively cope with the dynamic characteristics of the wireless optical communication channel, and thus improve the system performance.

[0135] The present invention is based on an open source or closed source base model (such as GPT, RoBERTa or Sora, etc.), and provides an efficient and accurate equalization solution for wireless optical communication, which can bring the following beneficial effects:

[0136] 1) Effectively increase system bandwidth, optimize signal-to-noise ratio, and thus improve communication rate. Using traditional equalization, experimental tests show that a single optical path can achieve a communication speed of 700Mbit / s at a communication distance of 1m. In contrast, using an adaptive equalization solution based on a large base model, the communication speed can exceed 2Gbit / s after simulation;

[0137] 2) The proposed equalization method is a digital equalization method, which can be combined with analog equalization to further improve the system communication rate;

[0138] 3) It is able to dynamically and adaptively adjust the equalization parameters according to the changing channels, introduce large models to improve the pre-training efficiency and the migration and generalization capabilities of multi-task learning, and make the design of the equalization algorithm more flexible.

[0139] 4) An adaptive equalization scheme is designed using a large base model. Compared with traditional schemes, it does not need to consider LED performance and channel status, and can realize intelligent optimization of LED modulation bandwidth, thereby achieving the optimal system communication rate.

[0140] In the above embodiments, although the steps are numbered S1, S2, etc., these are only specific embodiments given by the present invention. Those skilled in the art may adjust the execution order of S1, S2, etc. according to actual conditions, which is also within the protection scope of the present invention. It can be understood that in some embodiments, some or all of the above embodiments may be included.

[0141] like Figure 6 As shown, a wireless optical communication equalization system 200 based on a base large model according to an embodiment of the present invention includes a channel characteristic determination module 201, a target equalization parameter acquisition module 202, a channel index adjustment module 203 and a communication module 204;

[0142] The channel characteristic determination module 201 is used to: when the first user terminal sends a preset known signal to the second user terminal through the wireless optical communication system, determine the current characteristics of the channel of the wireless optical communication system according to the preset known signal and the signal received by the second user terminal;

[0143] The target equalization parameter acquisition module 202 is used to: input the current characteristics of the channel of the wireless optical communication system into the trained base large model to obtain the target equalization parameters;

[0144] The channel index adjustment module 203 is used to adjust the current communication index of the channel of the wireless optical communication system according to the target equalization parameter;

[0145] The communication module 204 is used to: when the first user terminal continues to send the signal to be sent to the second user terminal, send the signal to be sent through the channel after the state is adjusted.

[0146] Optionally, in the above technical solution, a model determination module is further included, and the model determination module is used to:

[0147] Establishing a sample set including a plurality of samples, wherein a set of characteristics of a channel of a wireless optical communication system and an equalization parameter corresponding to the set of characteristics constitute one sample;

[0148] Based on the sample set, the base large model is pre-trained to obtain a pre-trained base large model;

[0149] Fine-tune the pre-trained base large model to obtain a trained base large model.

[0150] Optionally, the above technical solution further includes a data acquisition module, which is used to obtain multiple sets of features of the channels in the sample set from historical transmitted signals, historical received signals, historical channel model data and historical equalized signals.

[0151] Optionally, in the above technical solution, the base large model is BERT, GPT, RoBERTa, T5 or Sora.

[0152] It should be noted that the beneficial effects of the wireless optical communication equalization system 200 based on the large base model provided in the above embodiment are the same as the beneficial effects of the wireless optical communication equalization method based on the large base model, which will not be repeated here. In addition, when the system provided in the above embodiment realizes its functions, it only takes the division of the above functional modules as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the system can be divided into different functional modules according to actual conditions to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0153] like Figure 7 As shown, an electronic device 300 according to an embodiment of the present invention includes a processor 320, the processor 320 is coupled to a memory 310, and the memory 310 stores at least one computer program 330, and the at least one computer program 330 is loaded and executed by the processor 320, so that the electronic device 300 implements any of the above-mentioned wireless optical communication equalization methods based on a large base model, specifically:

[0154] The electronic device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors 320 (Central Processing Units, CPU) and one or more memories 310, wherein the one or more memories 310 store at least one computer program 330, and the at least one computer program 330 is loaded and executed by the one or more processors 320, so that the electronic device 300 implements any one of the wireless optical communication equalization methods based on the base large model provided in the above embodiments. Of course, the electronic device 300 may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output. The electronic device 300 may also include other components for realizing the functions of the device, which will not be repeated here. The electronic device may specifically be a computer, etc.

[0155] An adaptive equalization wireless optical communication system according to an embodiment of the present invention includes a wireless optical communication system, wherein an adaptive equalizer of the wireless optical communication system is used to implement any of the above-mentioned wireless optical communication equalization methods based on a large base model.

[0156] A computer-readable storage medium according to an embodiment of the present invention stores at least one computer program, and the at least one computer program is loaded and executed by a processor so that the computer implements any of the above-mentioned wireless optical communication equalization methods based on the base large model.

[0157] Optionally, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc (CD-ROM), a magnetic tape, a floppy disk, an optical data storage device, and the like.

[0158] In an exemplary embodiment, a computer program product or a computer program is also provided, the computer program product or the computer program comprising computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device performs any of the above-mentioned wireless optical communication equalization methods based on a large base model.

[0159] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects and represent the definition of a specific order or sequence. The order of use of similar objects can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than the order shown or described.

[0160] Those skilled in the art know that the present invention can be implemented as a system, method or computer program product. Therefore, the present invention can be specifically implemented in the following forms, namely: it can be complete hardware, it can be complete software (including firmware, resident software, microcode, etc.), or it can be a combination of hardware and software, which is generally referred to as "circuit", "module" or "system" herein. In addition, in some embodiments, the present invention can also be implemented in the form of a computer program product in one or more computer-readable media, and the computer-readable medium contains computer-readable program code.

[0161] Any combination of one or more computer-readable media can be used. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0162] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A wireless optical communication equalization method based on a base large model, characterized in that: include: When the first user terminal sends a preset known signal to the second user terminal through the wireless optical communication system, determining the current characteristics of the channel of the wireless optical communication system according to the preset known signal and the signal received by the second user terminal; Inputting the current characteristics of the channel of the wireless optical communication system into the trained base large model to obtain target equalization parameters; According to the target equalization parameter, adjusting the current communication index of the channel of the wireless optical communication system through the programmable driving circuit of the wireless optical communication system; When the first user terminal continues to send the signal to be sent to the second user terminal, the signal to be sent is sent through the channel with the state adjusted.

2. According to the wireless optical communication equalization method based on the base large model according to claim 1, it is characterized in that: The process of acquiring the trained large base model includes: Establishing a sample set including a plurality of samples, wherein a set of characteristics of a channel of the wireless optical communication system and equalization parameters corresponding to the set of characteristics constitute one sample; Based on the sample set, pre-training the base large model to obtain a pre-trained base large model; The pre-trained large base model is fine-tuned to obtain the trained large base model.

3. A wireless optical communication equalization method based on a base large model according to claim 2, characterized in that: Also includes: Multiple sets of features of the channels in the sample set are obtained from historical transmitted signals, historical received signals, historical channel model data, and historical equalized signals.

4. A wireless optical communication equalization method based on a base large model according to any one of claims 1 to 3, characterized in that: The base model is BERT, GPT, RoBERTa, T5, or Sora.

5. A wireless optical communication equalization system based on a base large model, characterized in that: It includes a channel characteristic determination module, a target equalization parameter acquisition module, a channel index adjustment module and a communication module; The channel characteristic determination module is used to: when the first user terminal sends a preset known signal to the second user terminal through the wireless optical communication system, determine the current characteristics of the channel of the wireless optical communication system according to the preset known signal and the signal received by the second user terminal; The target equalization parameter acquisition module is used to: input the current characteristics of the channel of the wireless optical communication system into the trained base large model to obtain the target equalization parameter; The channel indicator adjustment module is used to: adjust the current communication indicator of the channel of the wireless optical communication system through the programmable driving circuit of the wireless optical communication system according to the target equalization parameter; The communication module is used for: when the first user terminal continues to send the signal to be sent to the second user terminal, sending the signal to be sent through the channel with the state adjusted.

6. A wireless optical communication equalization system based on a base large model according to claim 5, characterized in that: It also includes a model determination module, wherein the model determination module is used to: Establishing a sample set including a plurality of samples, wherein a set of characteristics of a channel of the wireless optical communication system and equalization parameters corresponding to the set of characteristics constitute one sample; Based on the sample set, pre-training the base large model to obtain a pre-trained base large model; The pre-trained large base model is fine-tuned to obtain the trained large base model.

7. A wireless optical communication equalization system based on a base large model according to claim 6, characterized in that: It also includes a data acquisition module, which is used to obtain multiple sets of characteristics of the channels in the sample set from historical transmitted signals, historical received signals, historical channel model data and historical equalized signals.

8. A wireless optical communication equalization system based on a base large model according to any one of claims 5 to 7, characterized in that: The base model is BERT, GPT, RoBERTa, T5, or Sora.

9. An adaptive equalization wireless optical communication system, characterized in that: It comprises a wireless optical communication system, wherein the adaptive equalizer of the wireless optical communication system is used to implement a wireless optical communication equalization method based on a base large model as described in any one of claims 1 to 4.

10. An electronic device, characterized in that: The electronic device includes a processor, which is coupled to a memory. The memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor so that the electronic device implements a wireless optical communication equalization method based on a base large model as described in any one of claims 1 to 4.

Images