A communication data processing method, apparatus, device, and medium

Abstract

The application discloses a communication data processing method and device, equipment and medium, the method comprises the following steps: acquiring the actual communication performance parameter and channel environment parameter of the base station communication model in real time; determining the model weight update parameter of the base station communication model based on the preset model weight update algorithm and the channel environment parameter in the case that the actual communication performance parameter is less than the preset communication performance threshold; performing incremental update and first communication data processing on the base station communication model based on the model weight update parameter; determining the model update configuration parameter matched with the model weight update parameter, encapsulating the model weight update parameter and the model update configuration parameter, and generating a weight update instruction; and sending the weight update instruction to the corresponding terminal, so that the terminal can perform terminal communication model synchronous update and second communication data processing at the data transmission layer, thereby reducing the time delay and computing power cost of model update, improving the flexibility of the update mode and the accuracy of data processing.

Description

Technical Field

[0001] This application belongs to the field of data processing technology, specifically relating to a communication data processing method, apparatus, device, and medium. Background Technology

[0002] With the deep application of artificial intelligence technology in the physical layer of wireless communication, base stations and terminals need to maintain synchronized communication models. The data processing performance of the communication model directly determines the effectiveness of channel awareness and data transmission optimization during wireless communication, and is a core foundation for ensuring communication quality. To ensure the accuracy of communication data processing, timely updates to the communication model to address actual communication performance parameter degradation have become an urgent problem to be solved in the field of communication technology.

[0003] In existing technologies, communication model updates between base stations and terminals primarily rely on RRC (Radio Resource Control) signaling to complete the entire process of parameter transmission and control. After detecting a degradation in model performance, the base station calculates all the parameters required for model updates and sends all parameters and configuration information to the terminal via RRC reconfiguration signaling. The terminal receives this information, parses it at the RRC layer, updates its local model, and processes communication data based on the updated model. However, existing technologies rely on RRC signaling for model updates, which involves complex processes such as RRC reconfiguration, security activation, and radio bearer establishment. Furthermore, transmitting all model weights results in a large amount of data being updated, leading to high update latency, high computational overhead, and inflexible update methods. This, in turn, results in inaccurate communication data processing results. Summary of the Invention

[0004] This application provides a communication data processing method, apparatus, device, and medium, which solves the problems of high model update latency, large computing power overhead, and inflexible update methods in the prior art, leading to inaccurate communication data processing results. By determining the model weight update parameters of the base station communication model based on a preset model weight update algorithm and channel environment parameters, incremental updates and communication data processing are performed on the base station communication model. Weight update instructions are generated based on the model weight update parameters and model update configuration parameters, and weight update instructions are issued based on the physical downlink shared channel. This achieves the purpose of incremental updates to the communication model, reduces the model update latency and large computing power overhead, and improves the flexibility of the update method and the accuracy of data processing.

[0005] In a first aspect, embodiments of this application provide a communication data processing method, the method comprising: The actual communication performance parameters and channel environment parameters of the base station communication model are obtained in real time. When the actual communication performance parameters are less than the preset communication performance threshold, the model weight update parameters of the base station communication model are determined based on the preset model weight update algorithm and the channel environment parameters. The model weight update parameters include the first update low-rank matrix, the second update low-rank matrix, and the update matrix scaling factor. The base station communication model is incrementally updated based on the model weight update parameters, and the first communication data processing is performed based on the incrementally updated base station communication model. Determine the model update configuration parameters that match the model weight update parameters, and encapsulate the first update low-rank matrix, the second update low-rank matrix, the update matrix scaling factor, and the model update configuration parameters according to the preset data transmission frame structure to generate a weight update instruction. Weight update instructions are sent to the corresponding terminals based on the physical downlink shared channel, so that the terminals can perform terminal communication model synchronization updates and second communication data processing at the data transmission layer.

[0006] Furthermore, determine the model update configuration parameters that match the model weight update parameters, including: Obtain the current air interface transmission bandwidth and the target update accuracy of the base station communication model, and determine the quantization format parameters of the first and second update low-rank matrices based on the current air interface transmission bandwidth and the target update accuracy. Read the model index parameters of the base station communication model and the network layer index parameters corresponding to the model index parameters. Aggregate the quantization format parameters, model index parameters, and layer index parameters to obtain the model update configuration parameters that match the model weight update parameters.

[0007] Furthermore, based on the current air interface transmission bandwidth and target update accuracy, the quantization format parameters of the first and second updated low-rank matrices are determined, including: Determine the available quantization accuracy range corresponding to the current air interface transmission bandwidth, and match the target update accuracy with the available quantization accuracy range; When the target update precision is within the available quantization precision range, the target update precision is used as the quantization format parameter for the first and second updated low-rank matrices. When the target update precision is outside the range of available quantization precision, the multiple available quantization precisions within the range of available quantization precision are sorted by precision, and the maximum available quantization precision is used as the quantization format parameter for the first and second updated low-rank matrices.

[0008] Furthermore, the preset data transmission frame structure includes multiple data payload fields and one control field; The first updated low-rank matrix, the second updated low-rank matrix, the update matrix scaling factor, and the model update configuration parameters are encapsulated according to the preset data transmission frame structure to generate a weight update instruction, including: Based on the quantization format parameters in the model update configuration parameters, the first updated low-rank matrix and the second updated low-rank matrix are subjected to low-precision quantization processing respectively, and the relationship between the total amount of data of the first low-precision quantization processing result and the second low-precision quantization processing result and the maximum data carrying capacity of the preset data transmission frame structure is determined. When the total amount of data is less than or equal to the maximum data carrying capacity of the preset data transmission frame structure, the first low-precision quantization result and the second low-precision quantization result are respectively encapsulated into the first data payload field and the second data payload field in the preset data transmission frame structure. Based on the preset scaling factor index value list, the target scaling factor index value corresponding to the scaling factor of the update matrix is ​​determined, and the target scaling factor index value, the quantization format parameter, the model index parameter and the layer index parameter in the model update configuration parameters are encapsulated into the control field in the preset data transmission frame structure to obtain the weight update instruction.

[0009] Furthermore, the number of preset data transmission frame structures is multiple; After determining the relationship between the total amount of data from the first low-precision quantization result and the second low-precision quantization result and the maximum data carrying capacity of the preset data transmission frame structure, the method further includes: When the total amount of data exceeds the maximum data carrying capacity of the preset data transmission frame structure, the first low-precision quantization processing result and the second low-precision quantization processing result are split according to the maximum data carrying capacity to obtain multiple first quantization processing data segments and multiple second quantization processing data segments. According to the splitting order, each first quantization processing data segment is marked with a first segmentation identifier, and each second quantization processing data segment is marked with a second segmentation identifier. The results of each first segmentation identifier are encapsulated into the first data payload field of the corresponding preset data transmission frame structure, and the results of each second segmentation identifier are encapsulated into the second data payload field of the corresponding preset data transmission frame structure.

[0010] Furthermore, the model weight update parameters for the base station communication model are determined based on a preset model weight update algorithm and channel environment parameters, including: Determine the first preset low-rank matrix, the second preset low-rank matrix, and the initial update matrix scaling factor corresponding to the preset model weight update algorithm; predict the channel theoretical characteristic parameters based on the channel environment parameters, the first preset low-rank matrix, the second preset low-rank matrix, and the initial update matrix scaling factor. Obtain the actual channel characteristic parameters corresponding to the channel environment parameters, and calculate the channel characteristic deviation between the theoretical channel characteristic parameters and the actual channel characteristic parameters; Based on the channel feature deviation, the first and second preset low-rank matrices are iteratively optimized in multiple rounds, and the scaling factor of the initial update matrix is ​​corrected based on the optimization results of each round to obtain the model weight update parameters.

[0011] Furthermore, the base station communication model is incrementally updated based on the model weight update parameters, including: The model weight increment matrix is ​​obtained by fusing the first updated low-rank matrix, the second updated low-rank matrix, and the update matrix scaling factor in the model weight update parameters. The base weight matrix of the base station communication model is updated based on the model weight increment matrix to obtain the incremental update result of the base station communication model.

[0012] Secondly, embodiments of this application provide a communication data processing apparatus, the apparatus comprising: The parameter update determination module is used to obtain the actual communication performance parameters and channel environment parameters of the base station communication model in real time. When the actual communication performance parameters are less than the preset communication performance threshold, the module determines the model weight update parameters of the base station communication model based on the preset model weight update algorithm and the channel environment parameters. The model weight update parameters include the first update low-rank matrix, the second update low-rank matrix, and the update matrix scaling factor. The data processing module is used to incrementally update the base station communication model based on the model weight update parameters, and to perform the first communication data processing based on the incrementally updated base station communication model. The update instruction generation module is used to determine the model update configuration parameters that match the model weight update parameters. It encapsulates the first update low-rank matrix, the second update low-rank matrix, the update matrix scaling factor, and the model update configuration parameters according to the preset data transmission frame structure to generate weight update instructions. The update synchronization module is used to send weight update instructions to the corresponding terminal based on the physical downlink shared channel, so that the terminal can perform terminal communication model synchronization update and second communication data processing at the data transmission layer.

[0013] Thirdly, embodiments of this application provide an electronic device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method described in the first aspect.

[0014] Fourthly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0015] Fifthly, embodiments of this application also provide a computer program product comprising a computer program stored in a computer-readable storage medium, wherein at least one processor of the device reads from the computer-readable storage medium and executes the computer program, causing the device to perform the method described in the first aspect.

[0016] In this embodiment, the actual communication performance parameters and channel environment parameters of the base station communication model are acquired in real time. When the actual communication performance parameters are less than a preset communication performance threshold, the model weight update parameters of the base station communication model are determined based on a preset model weight update algorithm and channel environment parameters. The model weight update parameters include a first updated low-rank matrix, a second updated low-rank matrix, and an update matrix scaling factor. The base station communication model is incrementally updated based on the model weight update parameters, and the first communication data processing is performed based on the incrementally updated base station communication model. A model update configuration parameter matching the model weight update parameters is determined, and the first updated low-rank matrix, the second updated low-rank matrix, the update matrix scaling factor, and the model update configuration parameters are encapsulated according to a preset data transmission frame structure to generate a weight update instruction. The weight update instruction is sent to the corresponding terminal based on the physical downlink shared channel for the terminal to perform terminal communication model synchronization update and second communication data processing at the data transmission layer. The aforementioned communication data processing method solves the problems of high model update latency, large computational overhead, and inflexible update methods in existing technologies, which lead to inaccurate communication data processing results. By determining the model weight update parameters of the base station communication model based on a preset model weight update algorithm and channel environment parameters, incremental updates and communication data processing are performed on the base station communication model. Weight update instructions are generated based on the model weight update parameters and model update configuration parameters, and the weight update instructions are issued based on the physical downlink shared channel. This achieves the goal of incremental updates to the communication model, reduces model update latency and large computational overhead, and improves the flexibility of the update method and the accuracy of data processing. Attached Figure Description

[0017] Figure 1 This is a flowchart of a communication data processing method provided in an embodiment of this application; Figure 2 This is a flowchart of determining model update configuration parameters provided in an embodiment of this application; Figure 3 This is a flowchart of the generation of weight update instructions provided in an embodiment of this application; Figure 4This is a structural block diagram of a communication data processing device provided in an embodiment of this application; Figure 5 This is a structural block diagram of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, specific embodiments of this application are described in detail below with reference to the accompanying drawings. It is understood that the specific embodiments described herein are merely for explaining this application and not for limiting it. Furthermore, it should be noted that, for ease of description, only the parts relevant to this application are shown in the drawings, not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0019] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0020] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0021] Firstly, this solution can be used in scenarios involving data processing of communication data between base stations and terminals, particularly in scenarios where communication models are used to process communication data to ensure communication performance between base stations and terminals. By determining the model weight update parameters of the base station communication model based on a preset model weight update algorithm and channel environment parameters, incremental updates and communication data processing are performed on the base station communication model. Weight update instructions are generated based on the model weight update parameters and model update configuration parameters, and these instructions are issued based on the physical downlink shared channel. This achieves the goal of incrementally updating the communication model, reducing model update latency and computational overhead, and improving the flexibility of the update method and the accuracy of data processing.

[0022] Based on the above usage scenarios, it is understood that the execution subject of each step in this solution can be a computer device. The computer device refers to any electronic device with data computing, processing and storage capabilities, such as a PC (Personal Computer) or other terminal devices, or a server or other devices. This application embodiment does not limit this.

[0023] The following description, in conjunction with the accompanying drawings, details a communication data processing method, apparatus, device, and medium provided in this application through specific embodiments and application scenarios.

[0024] Figure 1 This is a flowchart of a communication data processing method provided in an embodiment of this application. Figure 1 As shown, the method is applied to a base station and specifically includes the following steps: S101: Real-time acquisition of actual communication performance parameters and channel environment parameters of the base station communication model. When the actual communication performance parameters are less than the preset communication performance threshold, the model weight update parameters of the base station communication model are determined based on the preset model weight update algorithm and the channel environment parameters. The model weight update parameters include the first updated low-rank matrix, the second updated low-rank matrix, and the update matrix scaling factor.

[0025] The base station communication model can be an artificial intelligence or machine learning model deployed on the base station side to implement functions related to the physical layer of wireless communication. Examples include CSI (Channel State Information) compression models, beam prediction models, and positioning enhancement models. In this solution, the base station and terminal sides synchronously store communication models with identical configuration parameters. If the base station detects that the actual communication performance parameters of the base station communication model are lower than a preset communication performance threshold, both the base station and terminal sides need to synchronously update the corresponding stored communication models to ensure the synergy, accuracy, and overall communication performance of AI applications in the 5G-Advanced (5G Evolution System) / 6G wireless communication physical layer, adapt to the time-varying and fast-fading characteristics of wireless channels, and solve problems such as communication link failures, data processing errors, and wasted air interface resources caused by model inconsistencies. Actual communication performance parameters can be quantitative indicators characterizing the inference effect and operating status of the base station communication model. Channel environment parameters can be characteristic data reflecting the real-time state of the wireless communication air interface channel. Examples include channel state information, signal fading coefficient, UE (User Equipment) movement trajectory, and propagation environment characteristics. The preset communication performance threshold can be a pre-set minimum communication performance value under the condition that the base station communication model does not experience performance drift. The preset model weight update algorithm can be a pre-set lightweight algorithm for calculating the weight correction amount of the base station communication model. The preset model weight update algorithm in this scheme is mainly the LoRA (Low-Rank Adaptation) algorithm. The model weight update parameters can be a set of parameters used to incrementally update the basic weights of the base station communication model. The first updated low-rank matrix can be one of the low-rank matrices obtained after performing low-rank decomposition on the model weight increment in the preset model weight update algorithm. The second updated low-rank matrix can be another low-rank matrix in the preset model weight update algorithm that works with the first updated low-rank matrix to complete the low-rank decomposition of the model weight increment. The update matrix scaling factor can be a parameter used to adjust the magnitude of the product of the first updated low-rank matrix and the second updated low-rank matrix.

[0026] In one embodiment, the operating status of the communication model can be continuously monitored, and the actual inference performance index of the base station communication model can be collected in real time as the actual communication performance parameter. Time-varying characteristic data of the wireless channel can also be collected in real time as the channel environment parameter. The collected actual communication performance parameter is compared with a preset communication performance threshold. If the actual communication performance parameter is less than the preset threshold, the model weight update process is triggered. An algorithm based on low-rank adaptation (LoRA) is called as the preset model weight update algorithm. The real-time channel environment parameter is substituted into the algorithm to calculate the model weight update parameters used to update the model's basic weights. This parameter set includes a first updated low-rank matrix after low-rank decomposition, a second updated low-rank matrix, and an update matrix scaling factor used to adjust the magnitude of the matrix product.

[0027] In one embodiment, determining the model weight update parameters of the base station communication model based on a preset model weight update algorithm and channel environment parameters includes: determining a first preset low-rank matrix, a second preset low-rank matrix, and an initial update matrix scaling factor corresponding to the preset model weight update algorithm; predicting theoretical channel characteristic parameters based on the channel environment parameters, the first preset low-rank matrix, the second preset low-rank matrix, and the initial update matrix scaling factor; obtaining actual channel characteristic parameters corresponding to the channel environment parameters; calculating the channel characteristic deviation between the theoretical channel characteristic parameters and the actual channel characteristic parameters; performing multiple rounds of iterative optimization on the first preset low-rank matrix and the second preset low-rank matrix based on the channel characteristic deviation; and adjusting the scaling factor of the initial update matrix based on the optimization results of each round to obtain the model weight update parameters.

[0028] The first preset low-rank matrix can be the initial low-rank matrix pre-set for low-rank decomposition in the preset model weight update algorithm. In this scheme, the first preset low-rank matrix is ​​a matrix A initialized using a Gaussian distribution, with dimension 1. ( For rank, (Input dimension for the layer to be updated in the model). The second preset low-rank matrix can be another initial low-rank matrix pre-set for low-rank decomposition in the preset model weight update algorithm, in conjunction with the first preset low-rank matrix. In this scheme, the second preset low-rank matrix is ​​matrix B initialized to a 0 matrix, with dimension [missing information]. ( Output the dimension of the layer to be updated in the model. (Rank). The initial update matrix scaling factor can be an initial scalar parameter preset in the pre-defined model weight update algorithm to adjust the magnitude of the low-rank matrix product. In this scheme, the initial update matrix scaling factor is the initial scalar parameter in the LoRA algorithm. The theoretical channel characteristic parameters can be theoretical values ​​of channel characteristics predicted by a base station communication model based on a preset low-rank matrix, an initial scaling factor, and real-time channel environment parameters. Examples include predicted channel state information, beam prediction angle, and signal transmission gain. The actual channel characteristic parameters can be actual measured values ​​of channel characteristics corresponding to the theoretical parameters, acquired through the wireless communication air interface. Examples include actually detected channel state information, actual beam propagation angle, and actual signal transmission gain. Channel characteristic deviation can be the quantized difference between the theoretical and actual channel characteristic parameters. Examples include parameter difference, mean square error, and absolute error. The optimization results of each round can be the updated matrix parameter values ​​obtained after each round of iterative optimization of the first and second preset low-rank matrices. The optimization results of each round are intermediate results of gradually correcting the low-rank matrix and reducing the channel characteristic deviation.

[0029] In one embodiment, based on a preset model weight update algorithm, the corresponding initial iteration parameters can be determined, namely, a first preset low-rank matrix initialized with a Gaussian distribution, a second preset low-rank matrix initialized to zero, and an initial update matrix scaling factor. Real-time collected channel environment parameters are substituted into the base station communication model, and the theoretical channel characteristic parameters are predicted through inference calculations using the base station communication model and the aforementioned initial parameters. The actual channel characteristic parameters corresponding to the theoretical channel characteristic parameters are collected by the wireless communication air interface detection module, and the difference between the theoretical and actual channel characteristic parameters is calculated to obtain the channel characteristic deviation. If the channel characteristic deviation does not reach a preset deviation threshold, a loss function is constructed based on this deviation, and multiple rounds of iterative optimization are performed on the first and second preset low-rank matrices using optimization algorithms such as gradient descent. Based on the adjustment magnitude and direction of the matrix parameters in each round of optimization, the initial update matrix scaling factor is adjusted accordingly to reduce the magnitude deviation of the matrix product. The above iterative optimization and scaling steps are repeated until the channel characteristic deviation reaches the preset deviation threshold. At this point, the optimized low-rank matrix and the corrected scaling factor from the final round are used as the model weight update parameters.

[0030] This scheme predicts and calculates the channel characteristic deviation between the theoretical and actual channel characteristic parameters. Based on the channel characteristic deviation, it performs multiple rounds of iterative optimization on the first and second preset low-rank matrices and corrects the scaling factor of the initial update matrix to obtain the model weight update parameters. This enables the accurate and personalized generation of model weight update parameters, improving the adaptability of the base station communication model to time-varying channels.

[0031] S102, the base station communication model is incrementally updated based on the model weight update parameters, and the first communication data processing is performed based on the incrementally updated base station communication model.

[0032] Incremental updates can be performed without replacing the base station communication model's base weights. Instead, the weight increments calculated from the model's weight update parameters are added to the base weights, achieving a lightweight update of the base station communication model. In this scheme, the incremental update is based on the LoRA algorithm, which adds the weight increment matrix to the original base weights of the model, enabling rapid hot updates. The first communication data processing step involves using the incrementally updated base station communication model to process communication data related to the wireless communication physical layer. Examples include CSI compression, beam prediction, and positioning enhancement data processing.

[0033] In one embodiment, the first updated low-rank matrix, the second updated low-rank matrix, and the update matrix scaling factor in the model weight update parameters can be first superimposed onto the local base weights of the base station communication model based on a preset model weight increment matrix calculation formula. This completes the incremental update of the base station communication model, and the incrementally updated base station communication model is then used to perform the first communication data processing on the communication data. For example, if the base station communication model is a CSI compression model, the collected channel state information is compressed and encoded; if the base station communication model is a beam prediction model, inference is performed on the channel environment data to complete the beam direction prediction processing; if the base station communication model is a positioning enhancement model, the location-related data of the terminal is analyzed to enhance the terminal positioning accuracy, and the updated model can adapt to the real-time channel environment, improving the accuracy and effectiveness of data processing.

[0034] In one embodiment, incrementally updating the base station communication model based on model weight update parameters includes: fusing the first updated low-rank matrix, the second updated low-rank matrix, and the update matrix scaling factor in the model weight update parameters to obtain the model weight increment matrix; updating the base weight matrix of the base station communication model based on the model weight increment matrix to obtain the incremental update result of the base station communication model.

[0035] The model weight increment matrix can be used to incrementally correct the basic weights of the base station communication model. In this scheme, the model weight increment matrix is ​​mainly calculated based on the LoRA algorithm. The basic weight matrix can be the original weight matrix of the base station communication model, pre-synchronized and stored by the base station and the terminal. The basic weight matrix in this scheme is... , dimension , is the initial weight matrix when the model has not undergone any incremental updates.

[0036] In one embodiment, the first low-rank update matrix (matrix A, dimension 1) can be extracted from the model weight update parameters. The second updated low-rank matrix (matrix B, dimension 1) ) and update matrix scaling factor (scalar) According to the fusion calculation formula of the LoRA algorithm By performing a fusion calculation of matrix multiplication and scalar scaling, the corresponding model weight increment matrix is ​​obtained. This matrix has the same dimension as the basic weight matrix. Matrix. The calculated model weight increment matrix. Superimposed on the pre-stored basic weight matrix of the base station communication model Above, through the formula Update the base weight matrix to obtain the updated model weight matrix. This refers to the incremental update result of the base station communication model.

[0037] This scheme calculates the model weight increment matrix by fusing the first updated low-rank matrix, the second updated low-rank matrix, and the update matrix scaling factor in the model weight update parameters. Based on the model weight increment matrix, the basic weight matrix of the base station communication model is updated, which can realize lightweight and fast hot update of the base station communication model and avoid the large computing power overhead of full weight transmission and model update.

[0038] S103, determine the model update configuration parameters that match the model weight update parameters, and encapsulate the first update low-rank matrix, the second update low-rank matrix, the update matrix scaling factor and the model update configuration parameters according to the preset data transmission frame structure to generate a weight update instruction.

[0039] The model update configuration parameters can be matched with the model weight update parameters, guiding the terminal side to complete various control and configuration information for weight parameter parsing and incremental model updates. In this scheme, the model update configuration parameters include parameters such as model index, rank indicator, quantization format, layer index, and segmentation indicator, which correspond one-to-one with weight parameters such as the first updated low-rank matrix and the second updated low-rank matrix. The preset data transmission frame structure can be a predefined, dedicated MAC CE (Media Access Control Control Element) frame structure used to encapsulate the model weight update parameters and model update configuration parameters. It is a standardized data format adapted to wireless air interface MAC layer transmission. The preset data transmission frame structure in this scheme is an AI-Update MAC CE (AI Update MAC Control Element) bit-level structure, containing fixed and variable-length combinations of MAC header, length field, control field, structure field, parameter field, and weight payload field.

[0040] In one embodiment, the model update configuration parameters can be determined based on the determined model weight update parameters. For example: the Model_ID (model index) is determined based on the base station communication model to be updated; the Rank_Idx (rank indicator) is determined based on the rank of the first and second updated low-rank matrices; the Q_Fmt (quantization format) is determined based on the quantization processing method of the low-rank matrix; the Layer_ID (layer index) is determined based on the neural network layer to be updated; and the Seg_Ind (segmentation indicator) is determined based on the data volume of the weight parameters. Following the preset AI-Update MAC CE bit-level frame structure, various parameters are encapsulated in the corresponding fields sequentially. For example: Write a dedicated LCID (Logical Channel Identifier) ​​(e.g., 45) in the MAC subheader, and set R (Reserved bit) in bit 7 of the MAC subheader for the scalability of the frame structure. Set F (Format bit) in bit 6 of the MAC subheader to control the format of the length field L. When F=0, the length field is L (8), i.e., 8 bits wide. When F=1, the length field is L (16), i.e., 16 bits wide. Then encapsulate Seg_Ind (2) and Model_ID (6) in the control field, encapsulate Rank_Idx (3) and Layer_ID (5) in the structure field, encapsulate Q_Fmt (3) and the quantization index Scale_Idx (5) of the update matrix scaling factor in the parameter field. Scale_Idx is the scaling factor index. Finally, encapsulate the bitstream data of the first updated low-rank matrix and the second updated low-rank matrix after quantization in the weight payload field in sequence. After all parameters are standardized and encapsulated, a weight update instruction that can be transmitted at the MAC layer is generated. This instruction is a dedicated MAC CE containing all update-related parameters. The weight update instruction can be sent to the terminal side via the PDSCH (Physical Downlink Shared Channel) over the air interface.

[0041] S104, based on the physical downlink shared channel, the weight update instruction is sent to the corresponding terminal so that the terminal can perform terminal communication model synchronization update and second communication data processing at the data transmission layer.

[0042] The physical downlink shared channel (PDSCH) in this scheme can be a common physical channel used by the base station to transmit downlink data and control signaling to the terminal in a wireless communication system. In this scheme, the PDSCH is a dedicated transmission channel carrying MAC CEs encapsulated with weight update instructions, adapting to the high-speed transmission requirements of MAC layer signaling. Terminal communication model synchronization update refers to the process where, after receiving the weight update instruction from the base station, the terminal performs incremental weight updates to its local communication model at the data transmission layer, ensuring weight synchronization between the terminal-side model and the updated model at the base station. In this scheme, the synchronization update is performed by the terminal at the MAC layer, completing parameter parsing and weight superposition without the need for the RRC layer, achieving rapid hot synchronization of the model. Secondary communication data processing refers to the wireless communication physical layer data processing performed by the terminal using the updated local model after completing the communication model synchronization update, matching the base station's approach. This is a terminal-side collaborative processing step within the base station's primary communication data processing. Examples include CSI decompression processing, beam reception adaptation processing, and positioning data feedback processing.

[0043] In one embodiment, a dedicated MAC CE encapsulating weight update instructions can be embedded into the base station's MACPDU (Media Access Control Protocol Data Unit). This MAC PDU is then transmitted to the corresponding target terminal via the Physical Downlink Shared Channel (PHS) as a radio frequency signal, completing the air interface instruction transmission. The terminal's PHY (Physical Layer) receiver performs physical layer decoding on the received PDSCH signal, obtaining a transport block and sending it to the MAC layer. The terminal's MAC layer identifies the dedicated LCID (101101) corresponding to the weight update instruction and directly extracts the model update configuration parameters and weight update parameters from the instruction at the data transmission layer, without forwarding to the RRC layer. Subsequently, it performs parameter parsing and fusion calculation using the LoRA algorithm consistent with the base station side, superimposing the obtained model weight increment matrix onto the basic weight matrix of the local terminal communication model, thus completing the synchronous incremental update of the terminal communication model and the base station-side model. After the terminal completes the model synchronization update, it immediately uses the updated local model to carry out the second communication data processing. If the base station side uses CSI compression processing, the terminal side performs the corresponding CSI decompression processing; if the base station side uses beam prediction processing, the terminal side performs beam receiving direction adaptation and verification processing; if the base station side uses positioning enhancement processing, the terminal side performs positioning data acquisition and feedback processing. This enables the base station and the terminal to complete collaborative physical layer communication data processing based on the synchronized updated communication model, adapting to the real-time changing channel environment and improving overall communication performance.

[0044] The technical solution provided in this application embodiment acquires the actual communication performance parameters and channel environment parameters of the base station communication model in real time. When the actual communication performance parameters are less than a preset communication performance threshold, the model weight update parameters of the base station communication model are determined based on a preset model weight update algorithm and channel environment parameters. The model weight update parameters include a first updated low-rank matrix, a second updated low-rank matrix, and an update matrix scaling factor. The base station communication model is incrementally updated based on the model weight update parameters, and the first communication data processing is performed based on the incrementally updated base station communication model. A model update configuration parameter matching the model weight update parameters is determined, and the first updated low-rank matrix, the second updated low-rank matrix, the update matrix scaling factor, and the model update configuration parameters are encapsulated according to a preset data transmission frame structure to generate a weight update instruction. The weight update instruction is sent to the corresponding terminal based on the physical downlink shared channel for the terminal to perform terminal communication model synchronization update and second communication data processing at the data transmission layer. The aforementioned communication data processing method solves the problems of high model update latency, large computational overhead, and inflexible update methods in existing technologies, which lead to inaccurate communication data processing results. By determining the model weight update parameters of the base station communication model based on a preset model weight update algorithm and channel environment parameters, incremental updates and communication data processing are performed on the base station communication model. Weight update instructions are generated based on the model weight update parameters and model update configuration parameters, and the weight update instructions are issued based on the physical downlink shared channel. This achieves the goal of incremental updates to the communication model, reduces model update latency and large computational overhead, and improves the flexibility of the update method and the accuracy of data processing.

[0045] Figure 2 This is a flowchart illustrating the determination of model update configuration parameters provided in an embodiment of this application. For example... Figure 2 As shown, the specific steps include the following: S201, obtain the current air interface transmission bandwidth and the target update accuracy of the base station communication model, and determine the quantization format parameters of the first and second update low-rank matrices based on the current air interface transmission bandwidth and the target update accuracy.

[0046] The current air interface transmission bandwidth can be the actual downlink data transmission bandwidth available between the base station and the terminal under the current channel environment. The current air interface transmission bandwidth is a core physical layer parameter that determines the air interface data transmission rate and the amount of data transmitted in a single transmission. In this scheme, the current air interface transmission bandwidth can constrain the amount of data transmitted for model update parameters. The target update accuracy can be a pre-set performance indicator such as the inference accuracy or fitting effect that the base station communication model needs to achieve after incremental updates. The quantization format parameter can be the format configuration information used to quantize and encode the values ​​of the first and second updated low-rank matrices. The quantization format parameter is a core configuration used to balance the transmission overhead of update parameters with the model update accuracy. For example, types such as INT8, INT4, NF4, and FP16 correspond to different quantization accuracies and data storage bytes.

[0047] In one embodiment, the available downlink transmission bandwidth of the current wireless air interface can be collected and calculated in real time through the bandwidth detection module of the base station physical layer to obtain the current air interface transmission bandwidth. The bandwidth detection module of the base station physical layer is a dedicated signal detection and processing unit deployed in the base station physical layer (PHY). It is the core functional module for the base station to realize wireless air interface bandwidth awareness, and is primarily responsible for real-time collection, detection, and calculation of the actual available downlink transmission bandwidth of the current wireless air interface. Based on the service application requirements of the base station communication model, the target update accuracy to be achieved in this incremental model update can be determined, and the current air interface transmission bandwidth and the target update accuracy can be used as constraints to determine the quantization format parameters. If the current air interface transmission bandwidth is sufficient and the target update accuracy requirement is high (such as beam prediction models in high-reliability communication scenarios), then high-precision quantization format parameters (such as FP16, INT8) should be selected to ensure the numerical accuracy of the first and second updated low-rank matrices and meet the performance requirements of model updates. If the current air interface transmission bandwidth is limited (such as severe channel fading or air interface resources being occupied by other services), and the target update accuracy requirement is normal (such as CSI compression models in ordinary scenarios), then lightweight quantization format parameters (such as INT4, NF4) should be selected to significantly compress the amount of transmitted data of the low-rank matrix, adapt to the limited air interface transmission bandwidth, and ensure that the update parameters can be transmitted quickly and completely.

[0048] S202, read the model index parameters of the base station communication model and the network layer index parameters corresponding to the model index parameters, aggregate the quantization format parameters, model index parameters and layer index parameters to obtain the model update configuration parameters that match the model weight update parameters.

[0049] The model index parameter can be a unique identifier assigned to the base station communication model. In this scheme, the model index parameter is a 6-bit Model_ID, with a value range of 0~63, which can correspond to different base station communication models such as CSI compression, beam prediction, and positioning enhancement. The network layer index parameter can be a unique identifier assigned to different neural network layers within a single base station communication model, used to accurately specify the specific network layer in the model that needs to undergo incremental weight updates. In this scheme, the network layer index parameter is a 5-bit Layer_ID, with a value range of 0~31, which can correspond to different neural network layers in the model such as fully connected layers and convolutional layers.

[0050] In one embodiment, the unique model index parameter corresponding to the base station communication model that needs to be updated can be read from the base station's model management module. Based on the impact of channel environment changes on the model, the network layer index parameter corresponding to the specific neural network layer in the base station communication model pointed to by this model index parameter is then read. The determined quantization format parameters, the read model index parameters, and the network layer index parameters are aggregated and integrated, along with the rank indicator, scaling factor index, segmentation indicator, and other supporting configuration information for this model update, to form a complete set of model update configuration parameters that matches the calculated model weight update parameters one-to-one. The model update configuration parameters are used to guide the terminal side in identifying the target update model, locating the model layer to be updated, and restoring the low-rank matrix according to the corresponding quantization format.

[0051] In one embodiment, determining the quantization format parameters of the first and second updated low-rank matrices based on the current air interface transmission bandwidth and the target update precision includes: determining the available quantization precision range corresponding to the current air interface transmission bandwidth and matching the target update precision with the available quantization precision range; if the target update precision is within the available quantization precision range, using the target update precision as the quantization format parameters of the first and second updated low-rank matrices; if the target update precision is outside the available quantization precision range, sorting the multiple available quantization precisions within the available quantization precision range and using the maximum available quantization precision as the quantization format parameters of the first and second updated low-rank matrices.

[0052] The available quantization precision range is a set of quantization precisions selected based on the current air interface transmission bandwidth capacity, combined with constraints such as air interface transmission delay and resource consumption. This set enables fast and complete transmission of low-rank matrix parameters and represents the selectable range of quantization precisions suitable for air interface transmission under the current channel environment. In this scheme, the available quantization precision range can be a subset selected from preset quantization precision types (INT8, INT4, NF4, FP16). Different air interface transmission bandwidths correspond to different ranges; the more abundant the bandwidth, the wider the available quantization precision range and the more high-precision types it includes.

[0053] In one embodiment, when determining the quantization format parameters based on the current air interface transmission bandwidth and the target update precision, the amount of data to be transmitted and the required transmission time slots for low-rank matrix parameters under different quantization precisions are calculated according to the specific value of the current air interface transmission bandwidth and the preset bandwidth-to-quantization precision mapping rules. The quantization precision types that meet the air interface transmission resource and real-time requirements are then selected to form an available quantization precision range. For example, when the current air interface transmission bandwidth is sufficient, the available quantization precision range includes FP16, INT8, NF4, and INT4; when the current air interface transmission bandwidth is limited and can only support small data transmissions, the available quantization precision range only includes NF4 and INT4. The target update precision of this model update is matched with this available quantization precision range. If the target update precision falls within the available quantization precision range, it means that the current air interface bandwidth can support the transmission of parameters with that precision. The target update precision is then directly used as the quantization format parameters for the first and second updated low-rank matrices, taking into account both model update precision and air interface transmission requirements. If the target update accuracy is outside the range of available quantization accuracy, it means that the current air interface bandwidth cannot support the transmission of such high-precision parameters. In order to ensure the model update effect as much as possible under the bandwidth constraint, all available quantization accuracies within the range of available quantization accuracy are sorted from high to low accuracy. The maximum value of the sorted available quantization accuracy is selected as the quantization format parameter, so as to achieve the optimal model update accuracy under the current channel conditions while meeting the bandwidth constraints of the current air interface transmission.

[0054] This scheme maximizes model update accuracy under transmission bandwidth constraints by determining the available quantization accuracy range corresponding to the current air interface transmission bandwidth, and by determining the quantization format parameters of the first and second updated low-rank matrices based on the target update accuracy and the available quantization accuracy range. This improves the transmission efficiency of model update weights and ensures the performance of the updated model.

[0055] The technical solution provided in this application determines the quantization format parameters of the first and second updated low-rank matrices based on the current air interface transmission bandwidth and target update accuracy. It aggregates the quantization format parameters, model index parameters, and layer index parameters to obtain model update configuration parameters that match the model weight update parameters. This achieves the goal of accurately adapting the model weight update parameters to the current transmission environment and model update requirements, thereby improving the rationality and success rate of model updates.

[0056] Figure 3 This is a flowchart illustrating the generation of weight update instructions provided in an embodiment of this application. For example... Figure 3 As shown, the preset data transmission frame structure includes multiple data payload fields and one control field, specifically including the following steps: S301, perform low-precision quantization processing on the first updated low-rank matrix and the second updated low-rank matrix according to the quantization format parameters in the model update configuration parameters, and determine the relationship between the total amount of data of the first low-precision quantization processing result and the second low-precision quantization processing result and the maximum data carrying capacity of the preset data transmission frame structure.

[0057] The data payload field can be a region in the preset data transmission frame structure specifically used to carry bitstream data. In this scheme, the data payload field mainly consists of Octet 4~K in the AI-Update MAC CE frame structure, i.e., Payload A (data payload field A) starting from Octet 4 (byte 4), and Octet K+1~N, i.e., Payload B (data payload field B) starting from Octet K+1 (byte K+1). The control field can be a region in the preset data transmission frame structure used to carry model update control configuration parameters. In this scheme, the control field mainly consists of Octet 1 (byte 1) in the AI-Update MAC CE frame structure, containing core control information such as segmentation indicators and model index parameters. Low-precision quantization processing can be an encoding process that converts the original high-precision values ​​(such as FP32) of the first and second updated low-rank matrices into low-precision numerical formats (such as INT8, INT4, NF4) according to the quantization format parameters in the model update configuration parameters. Low-precision quantization processing adapts to the air interface transmission bandwidth and frame structure carrying capacity by compressing the matrix data volume. The maximum data carrying capacity of the preset data transmission frame structure can be the maximum number of data bytes or bits that the data payload field can carry after reserving the number of bytes occupied by configuration parameters such as the control field, structure field, and parameter field. In this scheme, the maximum data carrying capacity of the preset data transmission frame structure is the upper limit of the quantized low-rank matrix data volume that can be transmitted by a single MAC CE frame.

[0058] In one embodiment, a predetermined quantization format parameter can be extracted from the model update configuration parameters. Low-precision quantization processing is then performed on the original matrix values ​​of the first and second updated low-rank matrices, respectively. The parameter configuration of a preset data transmission frame structure is retrieved, and the maximum data carrying capacity of the frame structure is calculated. The total data volume of the first and second low-precision quantization results is compared with the calculated maximum data carrying capacity of the preset data transmission frame structure. The result of this comparison determines whether the quantized data needs to be segmented and encapsulated for transmission. If the total data volume is less than or equal to the maximum carrying capacity, the first and second low-precision quantization results can be encapsulated and transmitted as a single frame. If the total data volume is greater than the maximum carrying capacity, the first and second low-precision quantization results need to be segmented and encapsulated into multiple frames according to the segmentation rules of the frame structure.

[0059] In one embodiment, the number of preset data transmission frame structures is multiple. After determining the relationship between the total amount of data of the first low-precision quantization processing result and the second low-precision quantization processing result and the maximum data carrying capacity of the preset data transmission frame structure, the method further includes: when the total amount of data is greater than the maximum data carrying capacity of the preset data transmission frame structure, splitting the first low-precision quantization processing result and the second low-precision quantization processing result according to the maximum data carrying capacity to obtain multiple first quantization processing data segments and multiple second quantization processing data segments; performing first segmentation identification on each first quantization processing data segment and second segmentation identification on each second quantization processing data segment according to the splitting order, and encapsulating each first segmentation identification result into the first data payload field of the corresponding preset data transmission frame structure, and encapsulating each second segmentation identification result into the second data payload field of the corresponding preset data transmission frame structure.

[0060] The first quantized data segment can be multiple data fragments obtained by splitting the low-precision quantization result of the first updated low-rank matrix according to the maximum data carrying capacity of the preset data transmission frame structure. The second quantized data segment can be multiple data fragments obtained by splitting the low-precision quantization result of the second updated low-rank matrix according to the same splitting rules. The first segment identifier can be an identifier added to each first quantized data segment to indicate its splitting order and segment type. The second segment identifier can be an identifier added to each second quantized data segment, consistent with the first segment identifier rules, used for the orderly reassembly of the second low-rank matrix quantized data on the terminal side. The first data payload field can be a dedicated area in the data payload field of the preset data transmission frame structure, specifically used to carry the first updated low-rank matrix quantized data and the corresponding first segment identifier. In this scheme, it is the Payload A (Matrix A Quantized Bits) area of ​​AI-Update MAC CE. The second data payload field can be a dedicated area in the data payload field of a preset data transmission frame structure, specifically used to carry the second updated low-rank matrix quantized data and the corresponding second segment identifier. In this scheme, it is the Payload B (Matrix B Quantized Bits) area of ​​AI-Update MAC CE.

[0061] In one embodiment, there are multiple preset AI-Update MAC CE data transmission frame structures. When it is determined that the total amount of data from the first low-precision quantization processing result and the second low-precision quantization processing result exceeds the maximum data carrying capacity of a single frame, the continuous bit streams of the first low-precision quantization processing result and the second low-precision quantization processing result are firstly split into multiple first quantization processing data segments and multiple second quantization processing data segments, using the maximum data carrying capacity of a single frame as the splitting benchmark. According to the splitting order, a first segment identifier is added to each first quantization processing data segment, and a second segment identifier is added to each second quantization processing data segment. Finally, multiple preset AI-Update MAC CE frame structures are invoked. The first segment identifier result and the corresponding first quantization processing data segment of each group are encapsulated into the first data payload field of the corresponding frame structure according to the splitting order. The second segment identifier result and the corresponding second quantization processing data segment are encapsulated into the second data payload field of the frame structure. Simultaneously, a corresponding segmentation indicator (Seg_Ind) field is configured for each frame, completing the standardized encapsulation of multiple frames. This ensures that the terminal side can sequentially reassemble the complete first and second low-rank matrix quantization data according to the identifier information and segmentation indicator.

[0062] This solution splits the quantization results and adds segment identifiers when the total data volume exceeds the maximum carrying capacity of a single frame. Each data segment is then encapsulated into the corresponding data payload domain. This enables the orderly transmission of model update parameters across multiple frames, ensuring that the parameters are transmitted completely and without error. This improves the reliability and parsing accuracy of large-scale data transmission over the air interface, thereby enhancing the accuracy of terminal model updates.

[0063] S302, when the total amount of data is less than or equal to the maximum data carrying capacity of the preset data transmission frame structure, the first low-precision quantization processing result and the second low-precision quantization processing result are respectively encapsulated into the first data payload field and the second data payload field in the preset data transmission frame structure.

[0064] In one embodiment, when the total amount of data is less than or equal to the maximum data carrying capacity of the preset data transmission frame structure, it is explained that all the data of the first updated low-rank matrix and the second updated low-rank matrix after low-precision quantization can be completely carried by a single preset data transmission frame structure without the need for splitting and segmenting. At this time, the first low-precision quantization result can be completely encapsulated into the first data payload field of the preset data transmission frame structure that is specifically for carrying this type of data, according to the preset frame structure data mapping rules. At the same time, the second low-precision quantization result can be completely encapsulated into the second data payload field of the frame structure that is immediately following the first data payload field and is specifically for carrying this type of data.

[0065] S303, based on the preset scaling factor index value list, determine the target scaling factor index value corresponding to the scaling factor of the update matrix, and encapsulate the target scaling factor index value, the quantization format parameter, the model index parameter and the layer index parameter in the model update configuration parameters into the control field in the preset data transmission frame structure to obtain the weight update instruction.

[0066] The preset scaling factor index value list can be pre-stored synchronously on the base station and terminal sides, containing a mapping table between the actual scaling factor values ​​and their corresponding index values. In this scheme, the preset scaling factor index value list is a 5-bit Scale_Idx mapping table, with index values ​​ranging from 0 to 31. Each index value uniquely corresponds to a specific scalar value of the scaling factor in the update matrix. The base station and terminal sides share the same list to ensure consistent parsing. The target scaling factor index value can be the index number matched from the preset scaling factor index value list, corresponding to the actual value of the scaling factor in the update matrix calculated in this model update. It is the specific carrier for transmitting the scaling factor in the frame structure, replacing the original scalar value transmission to reduce air interface data overhead.

[0067] In one embodiment, the pre-stored list of preset scaling factor index values ​​on the base station side is first retrieved. The actual scalar value of the scaling factor of the update matrix obtained through iterative optimization is precisely matched with the scaling factor value in the list to find the unique corresponding index number, which is determined as the target scaling factor index value. The target scaling factor index value, along with the quantization format parameter, model index parameter, and layer index parameter already determined in the model update configuration parameters, is encapsulated according to the control domain field definition rules of the preset data transmission frame structure. The model index parameter and segmentation indicator are encapsulated into frame structure Octet1 (control domain), the layer index parameter and rank indicator are encapsulated into Octet2 (structure domain), and the target scaling factor index value and quantization format parameter are encapsulated into Octet3 (parameter domain). At the same time, the low-rank matrix quantization data already encapsulated into the first and second data payload domains is combined with the above configuration parameters to complete the full field encapsulation of the entire preset data transmission frame structure, ultimately forming a complete weight update instruction that can be transmitted over the air interface. The list of AI-Update MAC CE bit-level structures obtained after full field encapsulation of the entire preset data transmission frame structure is shown in Table 1 below.

[0068] Table 1: List of AI-Update MAC CE bit-level structures

[0069] The technical solution provided in this application embodiment performs low-precision quantization processing on the first and second updated low-rank matrices, encapsulates the low-precision quantization processing results into the data payload field of a preset data transmission frame structure, and encapsulates the target scaling factor index value, quantization format parameters, model index parameters, and layer index parameters in the model update configuration parameters into the control field of the preset data transmission frame structure to obtain weight update instructions. This can significantly reduce the amount of data for model update parameters, improve the encapsulation efficiency and air interface transmission efficiency of weight update instructions, thereby improving model update efficiency and enhancing the accuracy of communication data processing.

[0070] Figure 4 This is a structural block diagram of a communication data processing device provided in an embodiment of this application. Figure 4 As shown, it specifically includes: The update parameter determination module 401 is used to obtain the actual communication performance parameters and channel environment parameters of the base station communication model in real time. When the actual communication performance parameters are less than the preset communication performance threshold, the model weight update parameters of the base station communication model are determined based on the preset model weight update algorithm and the channel environment parameters. The model weight update parameters include the first update low-rank matrix, the second update low-rank matrix and the update matrix scaling factor. The data processing module 402 is used to incrementally update the base station communication model based on the model weight update parameters, and to perform the first communication data processing based on the incrementally updated base station communication model. The update instruction generation module 403 is used to determine the model update configuration parameters that match the model weight update parameters, and encapsulate the first update low-rank matrix, the second update low-rank matrix, the update matrix scaling factor and the model update configuration parameters according to the preset data transmission frame structure to generate weight update instructions. The update synchronization module 404 is used to send weight update instructions to the corresponding terminal based on the physical downlink shared channel, so that the terminal can perform terminal communication model synchronization update and second communication data processing at the data transmission layer.

[0071] Furthermore, the update instruction generation module 403 is specifically used for: Obtain the current air interface transmission bandwidth and the target update accuracy of the base station communication model, and determine the quantization format parameters of the first and second update low-rank matrices based on the current air interface transmission bandwidth and the target update accuracy. Read the model index parameters of the base station communication model and the network layer index parameters corresponding to the model index parameters. Aggregate the quantization format parameters, model index parameters, and layer index parameters to obtain the model update configuration parameters that match the model weight update parameters.

[0072] Furthermore, the update instruction generation module 403 is specifically used for: Determine the available quantization accuracy range corresponding to the current air interface transmission bandwidth, and match the target update accuracy with the available quantization accuracy range; When the target update precision is within the available quantization precision range, the target update precision is used as the quantization format parameter for the first and second updated low-rank matrices. When the target update precision is outside the range of available quantization precision, the multiple available quantization precisions within the range of available quantization precision are sorted by precision, and the maximum available quantization precision is used as the quantization format parameter for the first and second updated low-rank matrices.

[0073] Furthermore, the preset data transmission frame structure includes multiple data payload fields and one control field; Update instruction generation module 403, specifically used for: Based on the quantization format parameters in the model update configuration parameters, the first updated low-rank matrix and the second updated low-rank matrix are subjected to low-precision quantization processing respectively, and the relationship between the total amount of data of the first low-precision quantization processing result and the second low-precision quantization processing result and the maximum data carrying capacity of the preset data transmission frame structure is determined. When the total amount of data is less than or equal to the maximum data carrying capacity of the preset data transmission frame structure, the first low-precision quantization result and the second low-precision quantization result are respectively encapsulated into the first data payload field and the second data payload field in the preset data transmission frame structure. Based on the preset scaling factor index value list, the target scaling factor index value corresponding to the scaling factor of the update matrix is ​​determined, and the target scaling factor index value, the quantization format parameter, the model index parameter and the layer index parameter in the model update configuration parameters are encapsulated into the control field in the preset data transmission frame structure to obtain the weight update instruction.

[0074] Furthermore, the number of preset data transmission frame structures is multiple; The update instruction generation module 403 is also used for: When the total amount of data exceeds the maximum data carrying capacity of the preset data transmission frame structure, the first low-precision quantization processing result and the second low-precision quantization processing result are split according to the maximum data carrying capacity to obtain multiple first quantization processing data segments and multiple second quantization processing data segments. According to the splitting order, each first quantization processing data segment is marked with a first segmentation identifier, and each second quantization processing data segment is marked with a second segmentation identifier. The results of each first segmentation identifier are encapsulated into the first data payload field of the corresponding preset data transmission frame structure, and the results of each second segmentation identifier are encapsulated into the second data payload field of the corresponding preset data transmission frame structure.

[0075] Furthermore, the parameter determination module 401 is specifically used for: Determine the first preset low-rank matrix, the second preset low-rank matrix, and the initial update matrix scaling factor corresponding to the preset model weight update algorithm; predict the channel theoretical characteristic parameters based on the channel environment parameters, the first preset low-rank matrix, the second preset low-rank matrix, and the initial update matrix scaling factor. Obtain the actual channel characteristic parameters corresponding to the channel environment parameters, and calculate the channel characteristic deviation between the theoretical channel characteristic parameters and the actual channel characteristic parameters; Based on the channel feature deviation, the first and second preset low-rank matrices are iteratively optimized in multiple rounds, and the scaling factor of the initial update matrix is ​​corrected based on the optimization results of each round to obtain the model weight update parameters.

[0076] Furthermore, the data processing module 402 is specifically used for: The model weight increment matrix is ​​obtained by fusing the first updated low-rank matrix, the second updated low-rank matrix, and the update matrix scaling factor in the model weight update parameters. The base weight matrix of the base station communication model is updated based on the model weight increment matrix to obtain the incremental update result of the base station communication model.

[0077] The technical solution provided in this application includes an update parameter determination module, which is used to obtain the actual communication performance parameters and channel environment parameters of the base station communication model in real time. When the actual communication performance parameters are less than a preset communication performance threshold, the module determines the model weight update parameters of the base station communication model based on a preset model weight update algorithm and the channel environment parameters. The model weight update parameters include a first update low-rank matrix, a second update low-rank matrix, and an update matrix scaling factor. A data processing module is used to incrementally update the base station communication model based on the model weight update parameters and perform first communication data processing based on the incrementally updated base station communication model. An update instruction generation module is used to determine the model update configuration parameters that match the model weight update parameters, encapsulate the first update low-rank matrix, the second update low-rank matrix, the update matrix scaling factor, and the model update configuration parameters according to a preset data transmission frame structure, and generate a weight update instruction. An update synchronization module is used to send the weight update instruction to the corresponding terminal based on the physical downlink shared channel, so that the terminal can perform terminal communication model synchronization update and second communication data processing at the data transmission layer. The aforementioned communication data processing device solves the problems of high model update latency, large computational overhead, and inflexible update methods in existing technologies, which lead to inaccurate communication data processing results. By determining the model weight update parameters of the base station communication model based on a preset model weight update algorithm and channel environment parameters, the device performs incremental updates and communication data processing on the base station communication model. It also generates weight update instructions based on the model weight update parameters and model update configuration parameters, and issues the weight update instructions based on the physical downlink shared channel. This achieves the goal of incrementally updating the communication model, reducing model update latency and large computational overhead, and improving the flexibility of the update method and the accuracy of data processing.

[0078] The communication data processing device in this application embodiment can be configured in a device, or it can be configured in a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, mobile electronic devices can be mobile phones, tablets, laptops, PDAs, in-vehicle electronic devices, wearable devices, ultra-mobile personal computers (UMPCs), netbooks, or personal digital assistants (PDAs), etc., while non-mobile electronic devices can be servers, network attached storage (NAS), personal computers (PCs), televisions (TVs), ATMs, or self-service machines, etc. This application embodiment does not specifically limit the scope.

[0079] One communication data processing device in this application embodiment can be an operating system. The operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit the specific operating system.

[0080] The communication data processing apparatus provided in this application embodiment can implement the various processes implemented in the above method embodiments. To avoid repetition, it will not be described again here.

[0081] like Figure 5 As shown, this application embodiment also provides an electronic device 500, including a processor 501, a memory 502, and a program or instructions stored in the memory 502 and executable on the processor 501. When the program or instructions are executed by the processor 501, they implement the various processes of the above-described communication data processing method embodiment and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0082] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0083] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described communication data processing method embodiment and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0084] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0085] This application also provides a program product including program code. When the program product is run on a computer device, the program code causes the computer device to perform the steps of the methods described above according to various exemplary embodiments of this application. For example, the computer device can execute a communication data processing method described in the embodiments of this application. The program product can be implemented using any combination of one or more readable media.

[0086] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0087] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0088] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0089] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of this application, the scope of which is determined by the scope of the claims.

Claims

1. A communication data processing method, characterized in that, The method is applied to a base station and includes: The actual communication performance parameters and channel environment parameters of the base station communication model are acquired in real time. When the actual communication performance parameters are less than a preset communication performance threshold, the model weight update parameters of the base station communication model are determined based on a preset model weight update algorithm and the channel environment parameters. The model weight update parameters include a first update low-rank matrix, a second update low-rank matrix, and an update matrix scaling factor. The base station communication model is incrementally updated based on the model weight update parameters, and the first communication data processing is performed based on the incrementally updated base station communication model. Determine the model update configuration parameters that match the model weight update parameters, and encapsulate the first update low-rank matrix, the second update low-rank matrix, the update matrix scaling factor, and the model update configuration parameters according to a preset data transmission frame structure to generate a weight update instruction; The weight update instruction is sent to the corresponding terminal based on the physical downlink shared channel, so that the terminal can perform terminal communication model synchronization update and second communication data processing at the data transmission layer.

2. The communication data processing method according to claim 1, characterized in that, The step of determining the model update configuration parameters that match the model weight update parameters includes: Obtain the current air interface transmission bandwidth and the target update accuracy of the base station communication model, and determine the quantization format parameters of the first update low-rank matrix and the second update low-rank matrix based on the current air interface transmission bandwidth and the target update accuracy; Read the model index parameters of the base station communication model and the network layer index parameters corresponding to the model index parameters, and aggregate the quantization format parameters, the model index parameters, and the layer index parameters to obtain model update configuration parameters that match the model weight update parameters.

3. The communication data processing method according to claim 2, characterized in that, The step of determining the quantization format parameters of the first updated low-rank matrix and the second updated low-rank matrix based on the current air interface transmission bandwidth and the target update precision includes: Determine the available quantization precision range corresponding to the current air interface transmission bandwidth, and match the target update precision with the available quantization precision range; If the target update precision is within the range of available quantization precision, the target update precision is used as the quantization format parameter for the first updated low-rank matrix and the second updated low-rank matrix. If the target update precision is outside the range of available quantization precision, the multiple available quantization precisions within the range of available quantization precision are sorted by precision, and the maximum available quantization precision is used as the quantization format parameter of the first updated low-rank matrix and the second updated low-rank matrix.

4. The communication data processing method according to claim 2, characterized in that, The preset data transmission frame structure includes multiple data payload fields and one control field; The step of encapsulating the first updated low-rank matrix, the second updated low-rank matrix, the update matrix scaling factor, and the model update configuration parameters according to a preset data transmission frame structure to generate a weight update instruction includes: According to the quantization format parameter in the model update configuration parameters, the first updated low-rank matrix and the second updated low-rank matrix are subjected to low-precision quantization processing respectively, and the relationship between the total amount of data of the first low-precision quantization processing result and the second low-precision quantization processing result and the maximum data carrying capacity of the preset data transmission frame structure is determined. When the total amount of data is less than or equal to the maximum data carrying capacity of the preset data transmission frame structure, the first low-precision quantization result and the second low-precision quantization result are respectively encapsulated into the first data payload field and the second data payload field in the preset data transmission frame structure. Based on the preset scaling factor index value list, the target scaling factor index value corresponding to the scaling factor of the update matrix is ​​determined, and the target scaling factor index value, the quantization format parameter in the model update configuration parameters, the model index parameter and the layer index parameter are encapsulated into the control field in the preset data transmission frame structure to obtain the weight update instruction.

5. The communication data processing method according to claim 4, characterized in that, The number of the preset data transmission frame structures is multiple; After determining the relationship between the total amount of data from the first low-precision quantization result and the second low-precision quantization result and the maximum data carrying capacity of the preset data transmission frame structure, the method further includes: When the total amount of data exceeds the maximum data carrying capacity of the preset data transmission frame structure, the first low-precision quantization processing result and the second low-precision quantization processing result are split according to the maximum data carrying capacity to obtain multiple first quantization processing data segments and multiple second quantization processing data segments. According to the splitting order, each of the first quantized data segments is marked with a first segmentation identifier, and each of the second quantized data segments is marked with a second segmentation identifier. The results of each first segmentation identifier are encapsulated into the first data payload field of the corresponding preset data transmission frame structure, and the results of each second segmentation identifier are encapsulated into the second data payload field of the corresponding preset data transmission frame structure.

6. The communication data processing method according to claim 1, characterized in that, The step of determining the model weight update parameters of the base station communication model based on the preset model weight update algorithm and the channel environment parameters includes: Determine the first preset low-rank matrix, the second preset low-rank matrix, and the initial update matrix scaling factor corresponding to the preset model weight update algorithm; predict the channel theoretical characteristic parameters based on the channel environment parameters, the first preset low-rank matrix, the second preset low-rank matrix, and the initial update matrix scaling factor. Obtain the actual channel characteristic parameters corresponding to the channel environment parameters, and calculate the channel characteristic deviation between the theoretical channel characteristic parameters and the actual channel characteristic parameters; Based on the channel feature deviation, the first preset low-rank matrix and the second preset low-rank matrix are optimized in multiple rounds of iterations, and the scaling factor of the initial update matrix is ​​corrected based on the optimization results of each round to obtain the model weight update parameters.

7. The communication data processing method according to claim 1, characterized in that, The incremental update of the base station communication model based on the model weight update parameters includes: The first low-rank update matrix, the second low-rank update matrix, and the update matrix scaling factor in the model weight update parameters are fused together to obtain the model weight increment matrix. The base weight matrix of the base station communication model is updated based on the model weight increment matrix to obtain the incremental update result of the base station communication model.

8. A communication data processing device, characterized in that, The device includes: The update parameter determination module is used to obtain the actual communication performance parameters and channel environment parameters of the base station communication model in real time. When the actual communication performance parameters are less than the preset communication performance threshold, the module determines the model weight update parameters of the base station communication model based on the preset model weight update algorithm and the channel environment parameters. The model weight update parameters include a first update low-rank matrix, a second update low-rank matrix, and an update matrix scaling factor. The data processing module is used to incrementally update the base station communication model based on the model weight update parameters, and to perform first communication data processing based on the incrementally updated base station communication model. The update instruction generation module is used to determine the model update configuration parameters that match the model weight update parameters, and encapsulate the first update low-rank matrix, the second update low-rank matrix, the update matrix scaling factor and the model update configuration parameters according to a preset data transmission frame structure to generate a weight update instruction. The update synchronization module is used to send the weight update instruction to the corresponding terminal based on the physical downlink shared channel, so that the terminal can perform terminal communication model synchronization update and second communication data processing at the data transmission layer.

9. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of a communication data processing method as described in any one of claims 1-7.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of a communication data processing method as described in any one of claims 1-7.

Images