A parameter setting method and device, electronic equipment and storage medium

Abstract

Embodiments of the present application provide a parameter setting method and device, electronic equipment and a storage medium, wherein the method comprises: determining current performance measurement information of a communication processing model; determining a performance change trend according to the performance measurement information and historical performance measurement information; adjusting a learning rate of the communication processing model based on the performance change trend, and retraining the communication processing model according to the learning rate. Embodiments of the present application realize dynamic updating of the learning rate, improve the accuracy of the communication processing model, can reduce the training time of the communication processing model, thereby reducing the waiting time of the communication system, can enhance the communication efficiency, and improve the communication quality.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a parameter setting method, apparatus, electronic device, and storage medium. Background Technology

[0002] With the development of wireless communication technology, network latency has gradually decreased, and communication technology has become increasingly intelligent and efficient. Traditional methods can no longer meet the development trend. With the increase in hardware computing power and the maturity of Artificial Intelligence (AI) technology, more and more manufacturers are choosing to use AI technology to solve problems in wireless communication, such as channel estimation and power amplifier behavior model analysis. However, in the field of AI, the training of neural networks often depends on the setting of the learning rate. For example, setting the learning rate too high causes the performance indicators of the neural network to oscillate, leading to a failure to converge, and the resulting neural network will cause a decrease in communication quality. Setting the learning rate too low results in slow convergence of the neural network's performance indicators, and a longer training process will lead to increased communication latency. Therefore, how to set the learning rate has become a crucial factor in improving communication efficiency and quality.

[0003] Currently, there are two traditional methods for solving the problem of learning rate setting:

[0004] 1) Setting a fixed learning rate is often used to solve problems in fixed scenarios. However, in wireless communication, due to the complexity of application scenarios, a fixed learning rate often cannot meet the needs of multiple scenarios, causing the trained neural network to be unable to adapt to its complex scenarios, resulting in a decline in communication quality.

[0005] 2) Set a learning rate decay: Initialize a large learning rate and decay it during training. Common decay methods include piecewise decay and exponential decay. However, learning rate decay introduces new parameters, increasing the complexity of the communication system and making communication quality uncontrollable. In addition, the learning rate decay process is a one-way decay, and when the actual demand of the communication scenario is greater than the initially set learning rate, the neural network that has been trained extensively cannot improve the performance of the communication system.

[0006] To address the aforementioned issues, there is an urgent need for an adaptive learning rate setting method to reduce the training time of neural networks in communication networks, lower communication latency, and improve communication quality. Summary of the Invention

[0007] This application provides a parameter setting method, apparatus, electronic device, and storage medium to enable rapid training of a communication processing model, reduce communication latency, and improve system communication quality.

[0008] This application provides a parameter setting method, which includes the following steps:

[0009] Determine the current performance metrics of the communication processing model;

[0010] The performance change trend is determined based on the performance metric information and historical performance metric information;

[0011] The learning rate of the communication processing model is adjusted based on the performance change trend, and the communication processing model is retrained according to the learning rate.

[0012] This application embodiment also provides a parameter setting device, the device comprising:

[0013] The current parameter module is used to determine the current performance metrics of the communication processing model;

[0014] The training trend module is used to determine the performance change trend based on the performance metric information and historical performance metric information.

[0015] The parameter adjustment module is used to adjust the learning rate of the communication processing model based on the performance change trend, and to retrain the communication processing model according to the learning rate.

[0016] This application also provides an electronic device, which includes:

[0017] One or more processors;

[0018] Memory, used to store one or more programs;

[0019] When the one or more programs are executed by the one or more processors, the one or more processors implement the parameter setting method as described in any of the embodiments of this application.

[0020] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the parameter setting method as described in any of the embodiments of this application.

[0021] In this embodiment, by acquiring performance measurement information of the communication processing model, comparing this performance measurement information with historical performance measurement information to determine the performance change trend, adjusting the learning rate of the communication processing model according to the performance change trend, and retraining the communication processing model according to the learning rate, dynamic training of the communication processing model is achieved, improving the accuracy of the learning rate of the communication processing model, thereby improving the communication quality of the communication system. Adjusting the learning rate using the performance change trend can reduce the model training time, thereby reducing the waiting latency of the communication system and enhancing communication efficiency. Attached Figure Description

[0022] Figure 1 This is a flowchart of a parameter setting method provided in an embodiment of this application;

[0023] Figure 2 This is a flowchart of another parameter setting method provided in the embodiments of this application;

[0024] Figure 3 This is a flowchart of another parameter setting method provided in the embodiments of this application;

[0025] Figure 4 This is an example diagram of a parameter setting method provided in an embodiment of this application;

[0026] Figure 5 This is a verification example diagram of a parameter setting method provided in an embodiment of this application;

[0027] Figure 6 This is an example graph illustrating a learning rate change trend provided in an embodiment of this application;

[0028] Figure 7 This is a comparison chart of neural network convergence speeds provided in an embodiment of this application;

[0029] Figure 8 This is a training example diagram of an MP model with a fixed learning rate provided in an embodiment of this application;

[0030] Figure 9 This is a training example diagram of an MP model under a parameter setting method provided in an embodiment of this application;

[0031] Figure 10 This is a schematic diagram of the structure of a communication processing model provided in an embodiment of this application;

[0032] Figure 11 This is an example diagram illustrating the training effect of a communication processing model provided in an embodiment of this application;

[0033] Figure 12 This is a comparison chart of the training effects of a communication processing model provided in an embodiment of this application;

[0034] Figure 13 This is a schematic diagram of another communication processing model provided in an embodiment of this application;

[0035] Figure 14 This is an example diagram illustrating the training effect of an LMS model provided in an embodiment of this application;

[0036] Figure 15 This is an example diagram illustrating the training effect of another LMS model provided in an embodiment of this application;

[0037] Figure 16This is a schematic diagram of the structure of a parameter setting device provided in an embodiment of this application;

[0038] Figure 17 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0039] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0040] In the following description, the use of suffixes such as "module," "part," or "unit" to denote elements is solely for the purpose of illustrative purposes and has no inherent meaning. Therefore, "module," "part," or "unit" may be used interchangeably.

[0041] Figure 1 This is a flowchart illustrating a parameter setting method provided in an embodiment of this application. This embodiment can be applied to deep learning model training in communication systems. The method can be executed by a parameter setting device, which can be implemented through software and / or hardware, and is generally integrated into a central terminal or communication terminal. See [link to relevant documentation]. Figure 1 The method provided in this application specifically includes the following steps:

[0042] Step 110: Determine the current performance metrics of the communication processing model.

[0043] The communication processing model can be a deep learning model used to process communication parameters in a wireless communication system. This model requires pre-training with massive amounts of data. Its accuracy directly impacts the communication quality of the wireless communication system. The communication processing model can include power amplifier behavior models for signal processing, digital predistortion models for improving power amplifier linearity, and adaptive equalization models for estimating transmitted signals. Performance metrics reflect the accuracy and processing efficiency of the communication processing model, specifically including the error between the model's output and the validation set data, as well as the model's processing speed.

[0044] In this embodiment of the application, during the training process of the communication processing model, the performance measurement information of the current state of the communication processing model can be obtained. The method of obtaining the information may include: inputting the verification data into the communication processing model to obtain the output result, and comparing the output result with the verification data. The comparison result can be used as the current performance measurement information. Alternatively, timing can be started when the verification data is input into the communication processing model and ended when the communication processing model generates the output result. The timing length can be used as the performance measurement information of the communication processing model.

[0045] Step 120: Determine the performance change trend based on performance measurement information and historical performance measurement information.

[0046] Historical performance metrics can be the performance metrics determined during each validation step in the training process of the communication processing model. These historical metrics can be all performance metrics determined historically, or the best performance value among all historically determined performance metrics. Performance change trends can be the result of comparing the current performance metric with historical performance metrics, and can include performance increase, performance decrease, or no change. It can be understood that performance increase means the current performance metric is better than historical performance metrics, performance decrease means the current performance metric is worse than historical performance metrics, and no change in performance means the current performance metric is equal to historical performance metrics.

[0047] Specifically, performance metrics can be compared with historical performance metrics, and the comparison result can be used as a performance change trend. For example, the difference between the current performance metric and historical performance metrics can be determined, and the performance change trend can be determined based on the magnitude of the difference.

[0048] Step 130: Adjust the learning rate of the communication processing model based on the performance change trend, and retrain the communication processing model according to the learning rate.

[0049] The learning rate can be the rate at which data is learned in deep learning. It can be a hyperparameter used to determine whether the objective function of the communication processing model can converge to a local minimum and the convergence speed of the objective function.

[0050] In the embodiments of the application, the learning rate of the communication processing model can be adjusted according to the performance change trend. For example, when performance increases, the learning rate can be increased to improve the accuracy of the communication processing model, and when performance decreases, the learning rate can be decreased to improve the training speed of the communication processing model. After adjusting the learning rate, the communication processing model can be retrained using training data to obtain new weights for the communication processing model.

[0051] In this embodiment, by determining the performance metric information of the communication processing model, determining the performance change trend based on the comparison results of the performance metric information and historical performance metric information, adjusting the learning rate of the communication processing model according to the performance change trend, and retraining the communication processing model according to the learning rate, the communication processing model is rapidly trained. By dynamically adjusting the learning rate through the performance change trend, the accuracy of the communication processing model can be improved, the training time of the communication processing model can be reduced, thereby reducing the waiting latency of the communication system and enhancing communication efficiency.

[0052] Furthermore, based on the above-described embodiments, the performance measurement information includes at least one of mean squared error and normalized mean squared error.

[0053] Specifically, Mean Squared Error (MSE) can be defined as the mean of the squared differences between the output of the communication processing model and the validation data. The formula for calculating MSE is as follows:

[0054]

[0055] Where n is the number of validation data used to verify the performance of the communication processing model, and x i For the output result, y i To verify the data.

[0056] In this embodiment, the Normalized Mean Squared Error (NMSE) can be calculated by transforming the expression for mean squared error into a dimensionless scalar expression. The formula for calculating the normalized mean squared error is as follows:

[0057]

[0058] Where n is the number of validation data used to verify the performance of the communication processing model, and x i For the output result, y i To verify the data.

[0059] Furthermore, based on the above-described embodiments, the historical performance measurement information includes historical best performance measurement information and previous training performance measurement information.

[0060] In this embodiment, the historical best performance metric information can be the best information among the performance metric information determined in each iteration. "Best" can refer to the highest accuracy or fastest processing speed of the communication processing model, etc. The previous training performance metric information can be the performance metric information determined in the previous one or several iterations. Furthermore, after each performance metric information is determined, it can be compared with the historical best performance metric information. If the currently determined performance metric information is better than the historical best performance metric information, then the historical best performance metric information is replaced with the currently determined performance metric information. Additionally, the previous training performance metric information is replaced with the currently determined performance metric information.

[0061] Figure 2 This is a flowchart of another parameter setting method provided in this application embodiment. This application embodiment is a specific embodiment based on the above application embodiment. See also... Figure 2The method provided in this application specifically includes the following steps:

[0062] Step 210: Determine the current performance metrics of the communication processing model.

[0063] Step 220: Determine the first comparison result and the second comparison result of the performance metric information with the historical best performance metric information and the previous training performance metric information respectively; take the first comparison result and the second comparison result as the performance change trend.

[0064] In this embodiment of the application, the performance measurement information can be compared with the historical best performance measurement information and the previous training performance measurement information respectively. The comparison result of the performance measurement information with the historical best performance measurement information is recorded as the first comparison result, and the comparison result of the performance measurement information with the previous training performance measurement information is recorded as the second comparison result. The obtained first comparison result and second comparison result can be used as a reflection of the performance change trend of the training state of the communication processing model.

[0065] Step 230: If the first comparison result shows that the performance metric information is worse than the historical best performance metric information, and the second comparison result shows that the performance metric information is worse than the previous training performance metric information, reduce the value of the learning rate.

[0066] Specifically, if the first comparison result shows that the performance metric is worse than the historical best performance metric, and the second comparison result shows that the performance metric is worse than the previous training performance metric, then it is determined that the performance of the current communication processing model has declined after training. In this case, the learning rate is reduced so that the performance of the communication processing model will increase in subsequent training processes.

[0067] Step 240: If the first comparison result shows that the performance metric information is better than the historical best performance metric information, increase the value of the learning rate.

[0068] In this embodiment of the application, if the first comparison result shows that the performance metric information is better than the historical best performance metric information, then the performance of the current communication processing model is the highest historical value, and the learning rate can be further increased to further improve the performance of the communication processing model.

[0069] Furthermore, if the first comparison result shows that the performance metric is worse than the historical performance metric, but the second comparison result shows that the performance metric is better than the previous training performance metric, it indicates that the current learning rate helps to increase the performance of the communication processing model. Therefore, the learning rate does not need to be adjusted.

[0070] Step 250: Retrain the communication processing model based on the learning rate.

[0071] In this embodiment of the application, after adjusting the value of the learning rate, the communication processing model can be retrained using training data to obtain the weights of the new communication processing model.

[0072] In this embodiment, by acquiring the performance metric information of the current communication processing model, the performance metric information is compared with the historical best performance metric information and the previous training performance metric information to obtain a first comparison result and a second comparison result as the performance change trend. If the first comparison result shows that the performance metric information is worse than the historical best performance metric information and the second comparison result shows that the performance metric information is worse than the previous training performance metric information, the learning rate is reduced. If the first comparison result shows that the performance metric information is better than the historical best performance metric information, the learning rate is increased. The communication processing model is re-trained based on the adjusted learning rate, thus realizing rapid training of the communication processing model. By dynamically adjusting the learning rate through the performance change trend, the accuracy of the communication processing model can be improved, the training time of the communication processing model can be reduced, thereby reducing the waiting latency of the communication system and enhancing communication efficiency.

[0073] Figure 3 This is a flowchart of another parameter setting method provided in this application embodiment. This application embodiment is a specific embodiment based on the above application embodiment. See also... Figure 3 The method provided in this application specifically includes the following steps:

[0074] Step 310: Initialize the historical performance metrics and learning rate of the communication processing model.

[0075] In this embodiment, historical performance metrics and learning rate can be initialized at the beginning of training the communication processing model. An initial value is set for each of the historical performance metrics and learning rate. The method of this application does not require high accuracy in setting the initial value, and the setting range is large, which can reduce the impact of the learning rate setting range on the communication processing model.

[0076] Step 320: Determine the current performance metrics of the communication processing model.

[0077] Step 330: Determine the performance change trend based on performance measurement information and historical performance measurement information.

[0078] Step 340: Determine the adjustment factor corresponding to the performance change trend, and use the adjustment factor to update the learning rate value.

[0079] The adjustment factor can be a weighted coefficient used to adjust the value of the learning rate.

[0080] In this embodiment, different adjustment factors can be pre-set for different performance change trends. When a performance change trend is obtained, different adjustment factors can be selected for different performance change trends, and the learning rate can be adjusted using these adjustment factors. For example, the product of the adjustment factor and the learning rate can be used as the learning rate for training the communication processing model.

[0081] Step 350: Retrain the communication processing model based on the learning rate.

[0082] Step 360: Determine the fitting relationship between the input voltage and output voltage of the communication device based on the trained communication processing model.

[0083] Specifically, the communication processing model can be a communication power amplifier behavior model that processes the input and output voltages of the device. This model can be used to fit the relationship between the input and output voltages of the communication device, thereby achieving accurate voltage control of the communication device and reducing device energy consumption.

[0084] In this embodiment, by initializing the historical performance metrics and learning rate of the communication processing model, the current performance metrics of the communication processing model are determined during the training process. Based on the performance metrics and historical performance metrics, the performance change trend is determined, and the adjustment factor corresponding to the performance change trend is obtained. The value of the learning rate is updated according to the adjustment factor, and the communication processing model is retrained using the learning rate. After the communication processing model is trained, the input voltage and output voltage of the communication device are fitted to achieve accurate control of the communication device voltage, reduce voltage fluctuations, improve communication stability, and enhance communication quality.

[0085] Furthermore, based on the above-mentioned application embodiments, the communication processing model includes at least one of the following: a communication power amplifier behavior model, a digital predistortion model, and an adaptive equalization model.

[0086] In one exemplary implementation, Figure 4 This is an example diagram of a parameter setting method provided in an embodiment of this application. See also... Figure 4 Assume the neural network has parameters c, performance is measured by M, and the metric function is M = f(c). The learning rate is μ, where μ is a real number greater than 0.

[0087] Let c be the parameter at the completion of the nth training iteration, and M be the performance metric. n The learning rate is denoted as μ. n .

[0088] M best This represents the historical best value for the neural network performance metric at the end of the (n-1)th training iteration.

[0089] The contents that need to be protected in this invention are listed below:

[0090] (1) Learning rate adaptive mechanism: Calculate the network performance metric M when the nth training is completed. n =f(c),

[0091] ①If M n Better than M best Then let M best =M n μ n+1 =α1μ n ;

[0092] ②If M n Inferior to M best And M n Better than M n-1 Then μ n+1 =α2μ n ;

[0093] ③If M n Inferior to M best And M n Inferior to M n-1 Then μ n+1 =α3μ n ;

[0094] (2) As described in (1), if the performance metric is MSE or NMSE, the better is equivalent to less than, and the worse is equivalent to greater than;

[0095] (3) As described in (1), α1, α2, and α3 are learning rate adjustment factors, where α1 ≥ 1 and α3 ≤ 1;

[0096] (4) As described in (1), α1, α2, and α3 can be fixed values ​​or values ​​that change with the training process;

[0097] (5) As described in (1), if the maximum learning rate μ is configured max Then μ n+1 =min(μ n+1 ,μ max );

[0098] (6) As described in (1), if the minimum learning rate μ is configured min Then μ n+1 =max(μ n+1 ,μ min );

[0099] The above description is based on neural networks, but it does not mean that the embodiments of this application are only applicable to neural networks.

[0100] This application's embodiments demonstrate the effectiveness of the implementation method by using the parameter setting method of this application in a neural network scenario to improve the accuracy of sine function fitting. See [link to relevant documentation]. Figure 5 Suppose the input of a sinusoidal signal is x, starting from 0, with 1024 numbers spaced 0.01 between adjacent numbers. The output of the sinusoidal signal is y, where y = sin(x).

[0101] Now, we will build a neural network model with one input, one output, and one hidden layer with four neurons. The hidden layer activation function is the tanh function, and the output layer uses a linear activation function. We will initialize the parameters of this neural network, denoted as c0. The initial parameters include two sets: w and b, where b... 11 =b 12 =b 13 =b 14 =b 21 =0, w 21 =-0.4228, w 22 =-0.2863, w 23 =-0.2793, w 24 =0.0892, w 11 = -0.6490, w 12 =1.1812, w 13 =-0.7585, w 14 = -1.1096.

[0102] Using NMSE as a measure of neural network fitting performance, i.e., M n =NMSE(c n NMSE represents the normalized mean square error; a smaller value is better. The expression for NMSE is:

[0103]

[0104] in It is a neural network for the i-th sample point x i The output.

[0105] The initial learning rate is set to 0.001, the maximum learning rate is set to 0.1, and the minimum learning rate is set to 0.0001.

[0106] Set fixed learning rate adjustment factors α1 = 1.02, α2 = 1, and α3 = 0.99.

[0107] Calculate the performance of the neural network fitting using the initialization parameter c0, and let M... best =M0=NMSE(c0)=1.0691.

[0108] After completing the above steps, training begins using the Adam optimization algorithm. A portion of the training records is shown in the table below:

[0109] Table 1 Training Records 1-10

[0110]

[0111]

[0112] Table 1 shows the training records from n=1 to n=10. It can be seen that the neural network performance metric shows a monotonically decreasing trend during this stage, so the learning rate continues to increase.

[0113] Table 2 Training Records 271-279

[0114] n＝272 Mn = 1.5344e-02 Mbest = 3.6499e-03 μ = 0.04678 (↓) n＝273 Mn = 1.4601e-02 Mbest = 3.6499e-03 μ = 0.04678 n＝274 Mn = 6.4012e-03 Mbest = 3.6499e-03 μ = 0.04678 n＝275 Mn = 3.3701e-03 Mbest = 3.6499e-03 μ = 0.04772 (↑) n＝276 Mn = 3.0402e-03 Mbest = 3.3701e-03 μ = 0.04867 (↑) n＝277 Mn = 3.0309e-03 Mbest = 3.0402e-03 μ = 0.04965 (↑) n＝278 Mn = 2.7975e-03 Mbest = 3.0309e-03 μ = 0.05064 (↑) n＝279 Mn = 2.9533e-03 Mbest = 2.7975e-03 μ = 0.05013 (↓)

[0115] Figure 6 This is an example graph illustrating a learning rate change trend provided in an embodiment of this application; see also Figure 6 It shows the trend of the learning rate throughout the training process, which initially increases rapidly and then slowly decreases until the set minimum learning rate value. Figure 7 This is a comparison chart of neural network convergence speeds provided in an embodiment of this application. See also... Figure 7 By comparing the learning rate adjustment method in this embodiment with a fixed learning rate using a neural network, the adaptive learning rate mechanism proposed in this embodiment is shown to have a faster convergence speed than the fixed learning rate, and its final convergence performance is comparable to the optimal fixed learning rate with less fluctuation.

[0116] For example, taking the behavioral modeling of a power amplifier in a wireless communication system as an example, the relationship between the power amplifier input x and output y is determined, where x and y are both complex numbers. In this embodiment, the communication processing model is a memory polynomial (MP) model, which is commonly used for power amplifier behavioral modeling, and its expression is as follows:

[0117]

[0118] Where z is the output of the model, w k,q These are the model coefficients, where Q represents the memory depth and K represents the order of nonlinearity. The process of modeling a power amplifier using the MP model is equivalent to solving for the model coefficients w. k,q The process.

[0119] The Least Mean Square (LMS) algorithm is used to solve for the model coefficients w. k,qA common method is the complex LMS algorithm, which is similar to the training algorithm for neural networks. Both are iterative solutions based on gradient descent. The general principle of the complex LMS algorithm is as follows:

[0120] (1) Initialize w k,q Set the learning rate μ;

[0121] (2) The number of iterations cnt = 1;

[0122] (3) n = 1;

[0123] (4) Calculate the objective function For w k,q partial derivatives in

[0124] (5) Update parameters,

[0125] (6) n = n + 1, if n ≤ N, then return (4);

[0126] (7) cnt = cnt + 1, return (3);

[0127] The parameter setting method in this application embodiment can be combined with the ordinary LMS algorithm, and the specific steps are as follows:

[0128] (1) Initialize w k,q Set the learning rate μ1, and the fixed learning rate adjustment factors α1 = 1.3, α2 = 1, α3 = 0.9;

[0129] (2) Calculate the performance metrics of the initial model.

[0130] (3) The number of iterations cnt = 1;

[0131] (4) n = 1;

[0132] (5) Calculate the objective function For w k,q partial derivatives in

[0133] (6) Update parameters,

[0134] (7) n = n + 1, if n ≤ N, then return (5);

[0135] (8) Calculate the performance metrics for cnt iterations.

[0136] ①If M cnt Less than M bestThen let M best =M cnt μ cnt+1 =α1μ cnt ;

[0137] ②If M cnt Greater than M best And M cnt Less than M cnt-1 Then μ cnt+1 =α2μ cnt ;

[0138] ③If M cnt Greater than M best And M cnt Greater than M cnt-1 Then μ cnt+1 =α3μ cnt ;

[0139] (9) cnt = cnt + 1, return (4).

[0140] Figure 8 This diagram illustrates a training example of an MP model with a fixed learning rate, provided in an embodiment of this application. It shows how the NMSE changes with increasing training iterations using a common LMS algorithm set at different fixed learning rates in power amplifier modeling. It demonstrates that different learning rates result in significantly different final performance. (See also...) Figure 9 In the same power amplifier modeling, the LMS algorithm with parameter settings combined with the embodiments of this application was used, and different initial learning rates were set. The changes in NMSE as the number of training iterations increased were observed. Figure 9 As can be seen, even with different initial learning rates, the training performance converges to the same level quickly, and the convergence speed is much slower than that of the MP model with a fixed learning rate. Furthermore, the initial learning rate setting has no impact on the convergence speed of the learning rate in this application, thus reducing the impact of inaccurate initial learning rate settings on the training of the communication processing model.

[0141] In another exemplary embodiment, when modeling the behavior of a radio frequency power amplifier in a wireless communication system, a dynamic adjustment factor can be set to adjust the learning rate. Specifically, this can include the following steps:

[0142] (1) Initialize w k,q Set the learning rate μ1, and the fixed learning rate adjustment factors α1 = 1.3, α2 = 1, α3 = 0.9;

[0143] (2) Calculate the performance metrics of the initial model.

[0144] (3) The number of iterations cnt = 1;

[0145] (4) n = 1;

[0146] (5) Calculate the objective function For w k,q partial derivatives in

[0147] (6) Update parameters,

[0148] (7) n = n + 1, if n ≤ N, then return (5);

[0149] (8) Calculate the performance metrics for cnt iterations.

[0150] ①If M cnt Less than M best Then let M best =M cnt μ cnt+1 =α1μ cnt , α1=1.1α1, α3=1.1;

[0151] ②If M cnt Greater than M best And M cnt Less than M cnt-1 Then μ cnt+1 =α2μ cnt ;

[0152] ③If M cnt Greater than M best And M cnt Greater than M cnt-1 Then μ cnt+1 =α3μ cnt , α3=0.9α3, α1=1.3;

[0153] (9) cnt = cnt + 1, return (4);

[0154] In the embodiments of this application, the adjustment factors α1 and α3 can change dynamically as the MP model is trained.

[0155] Based on the above embodiments, Figure 10 This is a schematic diagram of the structure of a communication processing model provided in an embodiment of this application. Figure 10The communication processing model shown can be a Real-Valued Time-Delay Neural Network (RVTDNN) model used for RF power amplifier modeling. RVTDNN is a real-valued neural network that divides complex signals into real and imaginary parts to address the problem that traditional real-valued neural networks cannot handle complex data. RVTDNN consists of two parts: the first part is a time-delay structure used to simulate the nonlinear characteristics of the power amplifier, and the second part is a conventional Multilayer Perceptron (MLP) network used to simulate the nonlinear characteristics of the power amplifier. In the figure, the input x is divided into three terms: Real, imaginary, and abs. Each term contains four time-delay units in addition to itself. These data are fed into the subsequent MLP network, which contains a hidden layer consisting of 16 nodes with the tanh activation function. The parameters of this neural network are initialized, denoted as c0. NMSE is used as a metric for the fitting performance of the neural network, i.e., M n =NMSE(c n NMSE represents the normalized mean square error; a smaller value is better. The formula for calculating NMSE is: in It is a neural network for x i The output.

[0156] The parameters of the neural network are initialized, denoted as c0. The initial best historical performance is then set to M. best =NMSE(c0).

[0157] The initial learning rate is set to 0.001, the maximum learning rate is set to 0.1, and the minimum learning rate is set to 0.0001.

[0158] Set fixed learning rate adjustment factors α1 = 1.03, α2 = 1, and α3 = 0.99.

[0159] After completing the above steps, training begins using the Adam optimization algorithm. A portion of the training records is shown in the table below:

[0160] Table 3 Training Records 1-9

[0161]

[0162]

[0163] Table 3 shows the training records from n=1 to n=9. It can be seen that the neural network performance metric shows a monotonically decreasing trend during this stage, so the learning rate continues to increase.

[0164] Table 4 Training Records 271-279

[0165] n＝58 Mn = 4.2117e-04 Mbest = 3.9255e-04 μ3.9255e-045 n＝59 Mn = 3.8949e-04 Mbest = 3.9255e-04 μ3.9255e-045e-0 n＝60 Mn = 5.3353e-04 Mbest = 3.8949e-04 μ3.8949e-049e-0 n＝61 Mn = 3.9073e-04 Mbest = 3.8949e-04 μ3.8949e-049 n＝62 Mn = 3.8795e-04 Mbest = 3.8949e-04 μ3.8949e-049e-0 n＝63 Mn = 3.5915e-04 Mbest = 3.8795e-04 μ3.8795e-045e-0 n＝64 Mn = 4.2375e-04 Mbest = 3.5915e-04 μ3.5915e-045e-0 n＝65 Mn = 4.2229e-04 Mbest = 3.5915e-04 μ3.5915e-045

[0166] Table 4 shows the training records of the RVTDNN model from n=57 to n=65. It can be seen that the neural network performance metric fluctuates during this stage, with the learning rate sometimes increasing, sometimes decreasing, and sometimes remaining constant. Specifically, the learning rate remains constant when n equals 58, 61, and 65; it increases when n equals 59, 62, and 63; and decreases at other times. Figure 11 This is an example diagram illustrating the training effect of a communication processing model provided in an embodiment of this application. See also... Figure 11 The learning rate changes rapidly at the beginning of the training process, and then slowly decreases until it reaches the set minimum value. Figure 12 This is a comparison chart of the training effects of a communication processing model provided in an embodiment of this application. Figure 12 The diagram shows a comparison between training the RVTDNN model using the adaptive learning rate set by the parameter setting method of this application embodiment and training using a fixed learning rate. Compared with the fixed learning rate, the adaptive learning rate set by the parameter setting method provided in this application embodiment has a faster convergence speed and better final convergence performance.

[0167] In one exemplary implementation, Figure 13 This is a schematic diagram of another communication processing model provided in this application embodiment. In wireless communication systems, the nonlinearity of the transmitter power amplifier can cause distortion of the transmitted signal, leading to a decrease in communication performance. To solve this problem, a communication processing model is needed to linearly amplify the original signal. This communication processing model can be a Digital Pre-Distortion (DPD) model. This model allows the transmitted signal to be processed by the DPD module before being fed into the power amplifier, so that the signal output by the power amplifier is a linear amplification of the original signal. The memory polynomial model is a commonly used model for DPD modules, and its expression is as follows:

[0168]

[0169] Where z is the output of the model, w k,q These are the coefficients of the model, where Q represents the memory depth and K represents the order of nonlinearity. The core of the DPD problem lies in solving for the coefficients w of the DPD model. k,q . Figure 13 An LMS algorithm for solving DPD coefficients using an indirect architecture is shown. The training process of this LMS algorithm is similar to that of neural network algorithms, and the parameter setting method in the embodiments of this application can be used. The processing procedure of a common LMS algorithm may include the following steps:

[0170] (1) Initialize w k,q Set the learning rate μ;

[0171] (2) The number of iterations cnt = 1;

[0172] (3) n = 1;

[0173] (4) Calculate the error, e(n)=y(n) / Gx(n), where G is the amplification factor of the power amplifier;

[0174] (5) Update parameters, w k,q =w k,q -μ(x[nq]x[nq] k-1 )*e;

[0175] (6) n = n + 1, if n ≤ N, then return (4);

[0176] (7) cnt = cnt + 1, return (3).

[0177] The processing steps of the LMS algorithm in conjunction with the parameter setting method in the embodiments of this application may include the following steps:

[0178] (1) Initialize w k , q Set an initial learning rate μ1 and a fixed learning rate adjustment factor. Acceleration factors v1 = 1.01, v3 = 0.999;

[0179] (2) Calculate the performance metrics of the initial model.

[0180] (3) The number of iterations cnt = 1;

[0181] (4) n = 1;

[0182] (5) Calculate the error, e(n) = y(n) / Gx(n);

[0183] (6) Update parameters, w k,q =w k,q -μ(x[nq]|x[nq]| k-1 ) * e;

[0184] (7) n = n + 1, if n ≤ N, then return (5);

[0185] (8) Calculate the performance metrics for cnt iterations.

[0186] a) If M cnt Less than M bestThen let M best =M cnt μ cnt+1 =α1μ cnt , α1=v1α1,

[0187] b) If M cnt Greater than M best And M cnt Less than M cnt-1 Then μ cnt+1 =μ cnt ;

[0188] c) If M cnt Greater than M best And M cnt Greater than M cnt-1 Then μ cnt+1 =α3μ cnt , α3=v3α3,

[0189] (9) cnt = cnt + 1, return (4).

[0190] Figure 14 This is an example diagram illustrating the training effect of an LMS model provided in an embodiment of this application. See also... Figure 14 This paper demonstrates how the NMSE changes with increasing training iterations when using the common LMS algorithm in a digital predistortion model with different fixed learning rates. It shows that different learning rates result in significantly different final effects. The learning rate trained by existing technology is extremely unstable and can easily lead to signal distortion in the communication system, thus reducing communication quality. Figure 15 This is an example diagram illustrating the training effect of another LMS model provided in this application embodiment. See also... Figure 15 In the same digital predistortion model, using the LMS algorithm incorporating the parameter setting method described in this application embodiment, different initial learning rates were set. The changes in NMSE with increasing training iterations show that even with different initial learning rates, the training performance quickly converges to the same level. In this application embodiment, the initial learning rate setting is insensitive, facilitating user training of the digital predistortion model. In this embodiment, adjustment factors α1 and α3 dynamically change with the training process.

[0191] In another exemplary embodiment, in a wireless communication system, the signal emitted by the transmitter is affected by multipath propagation when passing through the wireless channel. At the receiver side, an equalization module is needed to recover the original signal. Assuming the transmitted signal is x, the received signal is y, and the channel impulse response is h, the expression for this process is:

[0192]

[0193] Equilibrium is the process of recovering x from y. Equilibrium can employ a linear model, for example:

[0194]

[0195] The coefficients w of the model can be solved using the LMS algorithm. The equalization can also be achieved using the RVTDNN model described in the above embodiments, and the model coefficients can be solved using the back propagation (BP) algorithm. Regardless of the model used, the learning rate can be set using the parameter setting method proposed in the embodiments of this application.

[0196] Figure 16 This is a schematic diagram of a parameter setting device provided in an embodiment of this application. It can execute the parameter setting method provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects of the execution method. The device can be implemented by software and / or hardware, and specifically includes: a current parameter module 401, a training trend module 402, and a parameter adjustment module 403.

[0197] The current parameter module 401 is used to determine the current performance metrics of the communication processing model.

[0198] The training trend module 402 is used to determine the performance change trend based on the performance metric information and historical performance metric information.

[0199] The parameter adjustment module 403 is used to adjust the learning rate of the communication processing model based on the performance change trend, and to retrain the communication processing model according to the learning rate.

[0200] In this embodiment, the current parameter module determines the performance metric information of the communication processing model, the training trend module determines the performance change trend based on the comparison between the performance metric information and historical performance metric information, and the parameter adjustment module adjusts the learning rate of the communication processing model according to the performance change trend and retrains the communication processing model based on the learning rate. This achieves rapid training of the communication processing model. By dynamically adjusting the learning rate through the performance change trend, the accuracy of the communication processing model can be improved, the training time of the communication processing model can be reduced, thereby reducing the waiting latency of the communication system and enhancing communication efficiency.

[0201] Furthermore, based on the above application embodiments, the performance measurement information in the current parameter module 301 includes at least one of mean square error and normalized mean square error.

[0202] Furthermore, based on the above-mentioned application embodiments, the historical performance measurement information in the training trend module 302 includes historical best performance measurement information and previous training performance measurement information.

[0203] Furthermore, based on the above-described embodiments, the training trend module 302 includes:

[0204] The comparison execution unit is used to determine a first comparison result and a second comparison result between the performance metric information and the historical best performance metric information and the previous training performance metric information, respectively.

[0205] A trend determination unit is used to take the first comparison result and the second comparison result as the performance change trend.

[0206] Furthermore, based on the above-described embodiments, the parameter adjustment module 303 includes:

[0207] The first processing unit is configured to reduce the value of the learning rate when the first comparison result indicates that the performance metric information is inferior to the historical best performance metric information, and the second comparison result indicates that the performance metric information is inferior to the previous training performance metric information.

[0208] The second processing unit is configured to increase the value of the learning rate when the first comparison result indicates that the performance metric information is better than the historical best performance metric information.

[0209] Furthermore, based on the above-described embodiments, the parameter adjustment module 303 in the device further includes:

[0210] The factor adjustment unit is used to determine the adjustment factor corresponding to the performance change trend, and to update the value of the learning rate using the adjustment factor.

[0211] Furthermore, based on the above-described embodiments, the apparatus further includes:

[0212] The parameter initialization module is used to initialize the historical performance metric information and the learning rate of the communication processing model.

[0213] Furthermore, based on the above-described embodiments, the apparatus further includes:

[0214] The model uses a module to determine the fitting relationship between the input voltage and output voltage of the communication device based on the trained communication processing model.

[0215] Furthermore, based on the above-mentioned application embodiments, the communication processing model in the device includes at least one of the following: a communication power amplifier behavior model, a predistortion model, and an adaptive equalization model.

[0216] Figure 17This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device includes a processor 50, a memory 51, an input device 52, and an output device 53; the number of processors 50 in the electronic device can be one or more. Figure 17 Taking a processor 50 as an example; in an electronic device, the processor 50, memory 51, input device 52, and output device 53 can be connected via a bus or other means. Figure 17 Taking the example of a connection between China and Israel via a bus.

[0217] The memory 51, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the modules corresponding to the parameter setting device in this embodiment (current parameter module 401, training trend module 402, and parameter adjustment module 403). The processor 50 executes various functional applications and data processing of the electronic device by running the software programs, instructions, and modules stored in the memory 51, thereby implementing the above-described parameter setting method.

[0218] The memory 51 may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a given function; the data storage area may store data created based on the use of the electronic device. Furthermore, the memory 51 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 51 may further include memory remotely located relative to the processor 50, which can be connected to the electronic device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0219] Input device 52 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the electronic device. Output device 53 may include display devices such as a display screen.

[0220] This application also provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform a parameter setting method, the method comprising:

[0221] Determine the current performance metrics of the communication processing model;

[0222] The performance change trend is determined based on the performance metric information and historical performance metric information;

[0223] The learning rate of the communication processing model is adjusted based on the performance change trend, and the communication processing model is retrained according to the learning rate.

[0224] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0225] It is worth noting that in the embodiments of the above-mentioned device, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the scope of protection of this application.

[0226] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0227] In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0228] The preferred embodiments of the present invention have been described above with reference to the accompanying drawings, but this does not limit the scope of the invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and spirit of the present invention should be considered within the scope of the present invention.

Claims

1. A parameter setting method, characterized in that, The method includes: Determine the current performance metrics of the communication processing model; The performance change trend is determined based on the performance metric information and historical performance metric information; wherein, the historical performance metric information includes historical best performance metric information and previous training performance metric information; the step of determining the performance change trend based on the performance metric information and historical performance metric information includes: determining a first comparison result and a second comparison result between the performance metric information and the historical best performance metric information and the previous training performance metric information, respectively; and using the first comparison result and the second comparison result as the performance change trend; The learning rate of the communication processing model is adjusted based on the performance change trend, and the communication processing model is retrained according to the learning rate.

2. The method according to claim 1, characterized in that, The performance metric information includes at least one of mean squared error and normalized mean squared error.

3. The method according to claim 1, characterized in that, The adjustment of the learning rate of the communication processing model based on the performance change trend includes at least one of the following: If the first comparison result indicates that the performance metric is inferior to the historical best performance metric, and the second comparison result indicates that the performance metric is inferior to the previous training performance metric, then the value of the learning rate is reduced. If the first comparison result indicates that the performance metric is better than the historical best performance metric, the value of the learning rate is increased.

4. The method according to claim 1, characterized in that, Adjusting the learning rate of the communication processing model based on the performance change trend includes: Determine the adjustment factor corresponding to the performance change trend, and update the value of the learning rate using the adjustment factor.

5. The method according to claim 4, wherein the adjustment factor is different for different performance change trends.

6. The method according to claim 1, characterized in that, Also includes: Initialize the historical performance metric information and the learning rate of the communication processing model.

7. The method according to claim 1, characterized in that, Also includes: The fitting relationship between the input voltage and output voltage of the communication device is determined based on the communication processing model after training.

8. The method according to claim 1, characterized in that, The communication processing model includes at least one of the following: a communication power amplifier behavior model, a digital predistortion model, and an adaptive equalization model.

9. A parameter setting device, characterized in that, The device includes: The current parameter module is used to determine the current performance metrics of the communication processing model; The training trend module is used to determine the performance change trend based on the performance metric information and historical performance metric information; wherein, the historical performance metric information includes historical best performance metric information and previous training performance metric information; the training trend module is specifically used to: determine a first comparison result and a second comparison result between the performance metric information and the historical best performance metric information and the previous training performance metric information respectively; and use the first comparison result and the second comparison result as the performance change trend; The parameter adjustment module is used to adjust the learning rate of the communication processing model based on the performance change trend, and to retrain the communication processing model according to the learning rate.

10. An electronic device, characterized in that, The electronic device includes: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the parameter setting method as described in any one of claims 1-8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the parameter setting method as described in any one of claims 1-8.

Images