Digital predistorter model design method based on neural network

Abstract

The invention belongs to the technical field of digital pre-distortion in satellite communication transmission, and provides a digital pre-distorter model design method based on a neural network for solving the technical problems that an existing digital pre-distortion algorithm lacks utilization of effective information and phase deviation, and the digital pre-distorter model design method based on the neural network is designed by combining the amplitude / phase and orthogonal / in-phase shunt isolation and interaction characteristics of communication data. A digital predistorter model based on set vector decomposition and multi-expert judgment is designed on the basis of an LSTM model, in-phase / orthogonal components and amplitude / phase information are comprehensively considered, cross-layer splicing is performed on original distortion signals and phase recovery signals through a vector splicing module, and isolation and interaction among different features are learned.

Description

Technical Field

[0001] The invention belongs to the technical field of digital predistortion in satellite communication transmission. Background Art

[0002] Satellite communication has many advantages, such as wide coverage, long transmission distance, large communication capacity, good transmission quality, flexible and rapid networking, and high confidentiality, and has become a highly competitive means of communication today. As a key component in satellite communication systems, the linearization technology of RF power amplifiers is crucial to improving the quality of satellite communications; and DPD (digital pre-distortion technology) is an important means of RF power amplifier linearization technology. It introduces opposite distortion at the signal input to compensate for the nonlinear distortion at the output of the RF power amplifier, avoiding problems such as reduced signal transmission quality, reduced spectrum efficiency, and interference between adjacent channels.

[0003] For example, although the pre-distortion algorithm based on the polynomial model can reduce nonlinear distortion to a certain extent, it has limitations in dealing with complex nonlinear characteristics, especially in multi-band and multi-channel communication environments, and it is difficult to meet the high-precision requirements of modern communication systems. However, neural networks, due to their flexibility and adaptability, can better capture and compensate for complex nonlinear distortions and have been widely used in the field of digital pre-distortion in recent years. Existing neural network digital pre-distortion methods include:

[0004] Combined with RNN (recurrent neural network), the memory effect of RNN is used for modeling, which improves the pre-distortion effect in dynamic environments. However, due to the gradient vanishing or gradient exploding problems in the process of calculating long sequence features, it is difficult to capture long-distance time dependencies, which is manifested as a weak memory ability for historical inputs. Therefore, a variant of RNN, the LSTM model, is proposed. By adding cell memory units and gating mechanisms to select memory or forget important feature information, it can capture the memory effect in a longer time domain in digital pre-distortion technology. However, the LSTM model lacks the use of signal amplitude and phase information.

[0005] Furthermore, by constructing VDLSTM (Vector Decomposition Long Short-Term Memory Network) to perform nonlinear calculations on the signal amplitude, and then recovering the phase by linear weighting, the technical problem of the above LSTM model's lack of use of signal amplitude and phase information is solved. However, VDLSTM focuses entirely on amplitude and phase information in calculations, lacks the use of isolation and interaction between orthogonal and in-phase branches of the signal, which affects the final result of nonlinear compensation; at the same time, the phase is recovered by linear weighted summation, which ignores the nonlinearity and memory of the phase, resulting in phase offset. Summary of the invention

[0006] In order to improve the limitations of existing digital pre-distortion algorithms, such as the lack of utilization of effective information and the technical problem of phase shift, the present invention proposes a "digital pre-distorter model design method based on neural network".

[0007] The design method of digital predistorter model based on neural network, such as Figure 1 As shown, the predistorter model includes an amplitude learning module, a phase calculation module, a phase recovery module, a vector splicing module, a multi-expert network learning module and a fusion decision module;

[0008] After the input data is acquired, the distorted signal is decomposed into amplitude and phase, input into the amplitude learning module for amplitude learning to output the amplitude learning signal, and input into the phase calculation module for processing by using sine and cosine functions to generate phase information signals of the in-phase path and the orthogonal path respectively; the amplitude learning signal and the phase information signal are input into the phase recovery module through linear transformation and multiplied with the in-phase path and the orthogonal path phase signals to output the phase recovery signal; the phase recovery signal and the distorted signal are input into the vector splicing module, spliced ​​in the feature dimension to output the spliced ​​signal; the spliced ​​signal is input into the multi-expert network learning module, and the multi-expert decision signal is output through calculation; the multi-expert decision signal is input into the fusion decision module to output the final decision signal;

[0009] The amplitude learning module includes a gating mechanism and a cell memory unit. The decomposed distorted signal is updated through the gating mechanism and cyclic calculation to update the hidden layer state of the cell memory unit and the decomposed signal itself. When the cyclic calculation is completed, the amplitude learning signal is output; the gating mechanism includes an input gate, a forget gate and an output gate. The specific calculation formula is as follows:

[0010] ,

[0011] : The activation value of the forget gate, 1 means completely retained, and 0 means completely discarded;

[0012] : Sigmoid activation function;

[0013] : Weight and bias of the forget gate;

[0014] : The hidden state of the previous moment;

[0015] : Input at the current moment;

[0016] ,

[0017] : Activation value of input gate;

[0018] : weight and bias of input gate;

[0019] ,

[0020] : The activation value of the output gate;

[0021] : Weight and bias of the output gate;

[0022] ,

[0023] : The current state of the cell memory unit;

[0024] : The cell memory unit state at the last moment;

[0025] : Candidate memory at the current moment;

[0026] ,

[0027] () is the hyperbolic tangent function;

[0028] , : Weights and biases of cell memory units;

[0029] ,

[0030] : The hidden state at the current moment;

[0031] The multi-expert network learning module receives the spliced ​​signal, and respectively focuses on learning the orthogonal branch, the feature interaction branch and the in-phase branch through each channel expert network, and correspondingly outputs the orthogonal branch signal, the feature interaction branch signal and the in-phase branch signal;

[0032] The fusion decision module receives the orthogonal branch signal, the in-phase branch signal and the feature interaction branch signal, performs feature fusion in a weighted fusion manner, and outputs a final decision signal; after building a pre-distorter model and performing iterative training, the final VDMOE (Vector Decomposed Mixture Of Experts) model is obtained.

[0033] Technical effects:

[0034] The present invention combines the isolation and interaction characteristics of the amplitude / phase and orthogonal / in-phase branches of communication data, and designs a digital predistorter model of set vector decomposition and multi-expert decision based on the LSTM model.

[0035] 1. Comprehensively consider the in-phase / quadrature components and amplitude / phase information to avoid the problem of loss of effective input feature information;

[0036] 2. The original distorted signal and the phase-recovered signal are cross-layered through the vector splicing module to avoid forgetting the initial features as the network learning progresses. This solves the technical problem of existing methods ignoring the nonlinearity and memory of the phase. At the same time, it also avoids problems such as excessive compression or stretching of features due to nonlinear operations at a certain layer, as well as gradient disappearance or gradient explosion, thereby accelerating the convergence speed.

[0037] 3. Subsequent nonlinear time series learning was carried out through the multi-expert network learning module to learn the isolation and interaction between different features, further enhancing the adaptability of the algorithm. Finally, the final decision signal was obtained by weighted summation of the discrimination results of each expert channel. The nonlinear learning of the phase also solved the technical problem of phase offset.

[0038] In order to illustrate the excellent performance of the method used in the present invention, the VDMOE model of the present invention is used to predict the test data, and its performance is evaluated by multiple indicators compared with the existing LSTM model and VDLSTM model. The specific evaluation indicators include power spectrum density diagram, mean square error (MSE) and AM-AM and AM-PM characteristic curves, such as Figure 3 As shown, the power spectrum density diagram shows the spectrum characteristics of the signal after pre-distortion processing. The distortion of the signal is improved after adding the pre-distorter. By comparing the differences between the corresponding curves of the LSTM model and the VDLSTM model, it can be found that the present invention (green line) is closer to the original signal.

[0039] The mean square error (MSE) is a commonly used error measurement indicator that can quantify the difference between the predistorter output signal and the ideal signal, reflecting the accuracy and reliability of the predistorter in signal recovery. The following table shows that the model constructed by the present invention has the smallest MSE value, and the predicted value is closest to the ideal signal.

[0040] DPD MSE (dB) No DPD added -19.52 LSTM -57.13 VDLSTM -54.66 VDMOE -59.22

[0041] AM-AM and AM-PM characteristic curves can intuitively show the amplitude and phase changes of the signal under different input powers, such as Figure 4As shown in the figure, by comparing the AM-AM graphs of the signal before and after adding the VDMOE model, it can be found that there is obvious nonlinear distortion when the VDMOE model is not used. In contrast, the VDMOE model uses the vector decomposition module to perform time series calculations on the signal amplitude, and uses the multi-expert network to learn the isolation and interaction between multi-dimensional features, so that the nonlinear distortion performance on AM-AM is better. Figure 5 As shown, by comparing the AM-PM diagrams of the signals before and after different DPD models, it can be found that when the DPD is the VDLSTM model, the phase distortion is more serious. In contrast, the present invention takes into account the phase timing and memory characteristics and performs subsequent nonlinear learning through a multi-expert network learning module, so the phase distortion is smaller. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is the overall structural block diagram of the predistorter model of the present invention.

[0043] Figure 2 The present invention is a flowchart illustrating the training mechanism process according to an embodiment of the present invention.

[0044] Figure 3 It is a comparison diagram of the power spectrum density of different DPD models in technical effects.

[0045] Figure 4 It is a comparison diagram of the AM-AM characteristic curves of the signal before and after the present invention.

[0046] Figure 5 The figure shows the comparison of the AM-PM characteristic curves of the signal before and after passing through different DPD models. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by ordinary technicians in the field without creative work by adopting the embodiments of the present invention are within the scope of protection of the present invention.

[0048] In this embodiment, data acquisition is first performed. The signal source is modulated by binary phase shift keying (BPSK) to obtain an initial signal. The initial signal is preprocessed to obtain a two-dimensional distorted signal in the time dimension and the feature dimension. The initial signal and the distorted signal are aligned and collected, and the distorted signal is input into the constructed predistorter model.

[0049] Furthermore, the initial signal preprocessing is specifically as follows: after the frequency up-conversion operation, the distorted signal is obtained after passing through a radio frequency power amplifier and attenuation, and the radio frequency power amplifier operates in a nonlinear operating range.

[0050] The gating mechanism described in the amplitude learning module includes input gate, forget gate and output gate. By introducing the gating mechanism to regulate the flow of information, the gradient disappearance and explosion problems are effectively alleviated. The specific calculation formula is as follows:

[0051] ,①

[0052] , ②

[0053] , ③

[0054] , ④

[0055] ,⑤

[0056] ,⑥

[0057] Among them, t represents the current moment, t-1 represents the previous moment, is the input at the current moment, f represents the forget gate, i represents the input gate, o represents the output gate, and C represents the memory unit state. is the candidate memory at the current moment, h represents the hidden state, and are weight and bias parameters, Represents the activation function Sigmoid(), () represents the hyperbolic tangent function.

[0058] Furthermore, the activation function Sigmoid() is expressed as:

[0059] ,

[0060] The hyperbolic tangent function tanh() is expressed as:

[0061] ,

[0062] in, is a constant, is the input variable.

[0063] The vector concatenation module concatenates features of different dimensions horizontally to expand them into high-dimensional features. The specific calculation formula is as follows:

[0064] ,

[0065] ,

[0066] ,

[0067] in, and represent the in-phase and quadrature components of the distorted signal, respectively. and Represent the in-phase and quadrature components of the in-phase branch of the phase recovery signal, respectively. , , The spliced ​​signal input into the multi-expert network learning module, specifically, The concatenated vector of the orthogonal branch channel expert network is input, and the orthogonal branch signal is output; The concatenated vector of the input feature interaction branch channel expert network is output as feature interaction branch signal; To input the concatenated vector of the in-phase branch channel expert network and output the in-phase branch signal, each channel expert network is calculated separately, and the structure is LSTM neural network and fully connected layer. The calculation formula is the same as formula ① to formula ⑥, which will not be repeated here.

[0068] This embodiment uses a pre-training plus fine-tuning approach to iteratively train the pre-distorter model.

[0069] like Figure 2 As shown in the figure, the left side is the pre-training link. First, a post-inverse model is trained, and its weight is used as the initial weight of the pre-distorter model. The input of the post-inverse model is the distorted signal after attenuation processing. The ideal target is the direct output signal of the signal source. The post-inverse model is equivalent to complementing the power amplifier to form distortion compensation. After that, the indirect learning structure is provided to the loop fine-tuning training on the right side to further fine-tune the pre-distorter model. First, the ideal gain signal is defined. Assume that the output signal of the signal source is , after linear amplification, the ideal signal without distortion is ,In the experiment, we refer to the amplification gain of the power amplifier and the attenuation, and set , which corresponds to the post-distorter model training environment.

[0070] Furthermore, the back propagation algorithm used in iterative training is AdamW. To further improve the computational efficiency and robustness of the model, the network is batch trained and the batch size is set to 256. The initial learning rate of the network is set to 0.0005. The total number of training iterations is 50, and the validation set score is calculated once for each iteration. Finally, the model with the highest validation set score is selected and migrated to the test set for calculation. The overall model training process uses a GPU (graphics processing unit) server, and the GPU chip is NVIDIA GeForce RTX 4090 with a memory capacity of 24GB.

[0071] The contents not described in detail in this specification belong to the prior art known to those skilled in the art. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. A method for designing a digital predistorter model based on a neural network, characterized in that: The predistorter model includes an amplitude learning module, a phase calculation module, a phase recovery module, a vector splicing module, a multi-expert network learning module and a fusion decision module; After the input data is acquired, the distorted signal is decomposed into amplitude and phase, input into the amplitude learning module for amplitude learning to output the amplitude learning signal, and input into the phase calculation module for processing by using sine and cosine functions to generate phase information signals of the in-phase path and the orthogonal path respectively; the amplitude learning signal and the phase information signal are input into the phase recovery module through linear transformation and multiplied with the in-phase path and the orthogonal path phase signals to output the phase recovery signal; the phase recovery signal and the distorted signal are input into the vector splicing module, spliced ​​in the feature dimension to output the spliced ​​signal; the spliced ​​signal is input into the multi-expert network learning module, and the multi-expert decision signal is output through calculation; the multi-expert decision signal is input into the fusion decision module to output the final decision signal; The amplitude learning module includes a gating mechanism and a cell memory unit. The decomposed distorted signal is updated through the gating mechanism and cyclic calculation to update the hidden layer state of the cell memory unit and the decomposed signal itself. When the cyclic calculation is completed, the amplitude learning signal is output; the gating mechanism includes a forget gate f, an input gate i and an output gate o. The specific calculation formula is as follows: ， : The activation value of the forget gate, 1 means completely retained, and 0 means completely discarded; : Sigmoid activation function; : Weight and bias of the forget gate; : The hidden state of the previous moment; : Input at the current moment; ， : Activation value of input gate; : weight and bias of input gate; ， : The activation value of the output gate; : Weight and bias of the output gate; ， : The current state of the cell memory unit; : The cell memory unit state at the last moment; : Candidate memory at the current moment; ， () is the hyperbolic tangent function; , : Weights and biases of cell memory units; ， : The hidden state at the current moment; The multi-expert network learning module receives the spliced ​​signal, and respectively focuses on learning the orthogonal branch, the feature interaction branch and the in-phase branch through each channel expert network, and correspondingly outputs the orthogonal branch signal, the feature interaction branch signal and the in-phase branch signal; The fusion decision module receives the orthogonal branch signal, the in-phase branch signal and the feature interaction branch signal, performs feature fusion in a weighted fusion manner, and outputs a final decision signal; after building a pre-distorter model and performing iterative training, the final VDMOE model is obtained.

2. The method for designing a digital predistorter model based on a neural network according to claim 1, characterized in that: The input data acquisition is specifically as follows: the signal source is modulated by two-phase shift keying to obtain an initial signal, the initial signal is preprocessed to obtain a two-dimensional distorted signal in the time dimension and the feature dimension, the initial signal and the distorted signal are aligned and collected, and the distorted signal is input into the constructed predistorter model.

3. The method for designing a digital predistorter model based on a neural network according to claim 2, characterized in that: The initial signal preprocessing is specifically as follows: after the frequency up-conversion operation, the distorted signal is obtained after passing through a radio frequency power amplifier and attenuation, and the radio frequency power amplifier operates in a nonlinear working range.

4. The method for designing a digital predistorter model based on a neural network according to claim 1, characterized in that: The activation function in the amplitude learning module is specifically: , the hyperbolic tangent function is specifically , in, is a constant, is the input variable.

5. The method for designing a digital predistorter model based on a neural network according to claim 1, characterized in that: The specific calculation formula of the vector splicing module is as follows: ， ， ， in, and represent the in-phase and quadrature components of the distorted signal, respectively. and Represent the in-phase and quadrature components of the phase recovery signal, respectively. is the concatenation vector of the input orthogonal branch channel expert network, is the concatenated vector of the input feature interactive branch channel expert network, is the concatenation vector of the input in-phase branch channel expert network.

6. The method for designing a digital predistorter model based on a neural network according to claim 1, characterized in that: After the predistorter model is built, iterative training is performed as follows: first, a post-inverse model is trained, and its weight is used as the initial weight of the predistorter model. The input of the post-inverse model is the distorted signal after attenuation processing. The ideal goal is that the signal source directly outputs the signal. The post-inverse model and the power amplifier complement each other to form distortion compensation. After that, the predistorter model is further fine-tuned through the indirect learning structure, and the signal source output signal is , after linear amplification and without distortion, the ideal gain signal is , update the error of the predistorter model and set .

7. The method for designing a digital predistorter model based on a neural network according to claim 6, characterized in that: The back propagation algorithm used in iterative training is AdamW, the batch size is set to 256, the initial learning rate of the network is set to 0.0005, and the total number of training iterations is 50.

Images