Star flash communication signal enhancement method based on deep neural network

Abstract

The invention belongs to the field of industrial data processing, and particularly relates to a satellite flash communication signal enhancement method based on a deep neural network. The method comprises the following steps: firstly, acquiring a signal sample and collecting related indexes by using distributed star flash nodes; secondly, constructing an improved residual recurrent neural network model, and enhancing signals through preprocessing, introduction of a gating residual unit, a multi-layer stacked network and multi-scale feature fusion; then quantitatively evaluating the enhancement effect based on a peak signal-to-noise ratio, a mean square error and a structural similarity index; and constructing reinforcement learning environment optimization for substandard signals, taking a signal quality index as a state, taking network parameter adjustment as an action, and training and updating a Q value function according to a specific strategy. According to the method, the defects of feature redundancy, information loss, gradient problem, poor adaptability and the like of a traditional method are overcome, signal features can be effectively reserved, noise is suppressed, the reconstruction precision is improved, adaptive enhancement is realized, and the robustness and generalization ability of the enhancement effect are improved.

Description

Technical Field

[0001] The present invention belongs to the field of industrial data processing, and particularly relates to a method for enhancing SparkLink communication signals based on a deep neural network. Background Art

[0002] Although SparkLink technology performs excellently in signal acquisition and transmission, there are still deficiencies in the field of signal enhancement. In an actual communication environment, the quality of signals is often affected by various factors, such as environmental noise, interference signals, and multipath effects. Especially in a complex interference environment, relying solely on the hardware synchronization and distributed acquisition of SparkLink technology is difficult to fundamentally solve the problem of signal quality degradation. Random noise in the environment can mask important features of the target signal, and the multipath propagation of signals may lead to severe attenuation and distortion. These problems will not only reduce the stability of signal transmission but also may affect the overall performance of the communication system. The existing signal enhancement means are mainly implemented through simple neural networks, without considering that the simple neural network architecture is relatively single and lacks the ability to deeply analyze the complex characteristics of signals. Especially when facing the dynamic changes of signals under multi-factor interference, it seems powerless. In addition, when removing noise, some useful signals may be mistakenly eliminated as noise, or when retaining signal features, the interference of noise cannot be effectively suppressed, resulting in a still relatively high noise level in the processed signal and poor signal enhancement effect. Summary of the Invention

[0003] In view of the technical problems existing in the above background art, the present invention proposes a method for enhancing SparkLink communication signals based on a deep neural network, which is simple in method and strong in theory.

[0004] To achieve the above object, the technical solution adopted by the present invention is as follows, including the following steps:

[0005] S1. First, use distributed SparkLink nodes to collect signal samples, and collect the signal-to-noise ratio, time delay, and bit error rate of the signals;

[0006] S2. Construct a signal enhancement model of an improved residual recursive neural network;

[0007] S3. Compare and analyze the enhanced signal with the expected signal standard, and quantitatively evaluate the effect of signal enhancement;

[0008] S4. Construct a reinforcement learning environment to enhance and optimize signals that do not meet the quantitative evaluation. The specific implementation is as follows:

[0009] S41. First, abstract the signal enhancement process into a reinforcement learning framework. The state is the current peak signal-to-noise ratio, mean square error, and structural similarity index of the signal, the action A is the adjustment of the enhancement network parameters; the reward function is ;

[0010] S42. The input layer is the current state , using a two - layer fully connected network, with 128 neurons in each layer, the activation function is ReLU, each action corresponds to an action value, and the optimal action is selected;

[0011] S43. Start training, initialize the parameters of the reinforcement learning model, use - greedy policy to select actions, and enhance the signal according to the selected actions, calculate the new state , use the reward function to calculate the reward value;

[0012] S44. Update the Q - value function, and the optimized policy is: , where is the maximum Q - value of all possible actions in the next state ;

[0013] S45. If is greater than 40, stop enhancement, otherwise continue to enhance;

[0014] S5. Finally, obtain the enhanced signal data.

[0015] Preferably, the construction of the signal enhancement model of the improved residual recurrent neural network in step S2 is as follows:

[0016] S21. First, denoise the input signal through pre - processing, and standardize the signal data to meet the input requirements of the model;

[0017] S22. Introduce a gated residual unit to control the information flow of the signal, and the state update of the recursive step is: , where is the hidden state at time step t, is the residual mapping function, capturing the dependence between the current - moment signal and the historical signal, is the gating function, is the denoised input signal, represents the element - wise product;

[0018] S23. Adopt a multi - layer stacked residual recurrent neural network, with each layer using the output of the previous layer as the input and processing it through residual units;

[0019] S24. Finally, optimize the signal reconstruction performance through multi - scale feature fusion.

[0020] Preferably, after introducing a gated residual unit to control the information flow of the signal, a complexity index of the signal is calculated according to the key features of the signal: , where is a preset maximum signal-to-noise ratio threshold, is the signal-to-noise ratio of the current signal, represents a preset delay tolerance threshold, is the delay of the current signal, is the current bit error rate, is a preset bit error rate threshold; by calculating the complexity index of the signal, if triggers model adjustment, that is, increasing the number of gated residual units in the residual recurrent neural network and expanding the network depth.

[0021] Preferably, the gating function in step S22 is learned through an independent network module, which dynamically determines the signal transmission method according to the characteristics of the current signal and is calculated by the following formula: , where is a trainable parameter, is the Sigmoid activation function.

[0022] Preferably, it is characterized in that the implementation steps of optimizing the signal reconstruction performance by multi-scale feature fusion in step S24 are:

[0023] S241. First, the hidden state outputs of each layer of the residual recurrent neural network are

[0024] combined to form a multi-scale feature set; , where represents the importance score of a specific scale feature and is automatically learned by network training;

[0025] S243. The fused feature is processed through a linear transformation and a non-linear activation function to generate a reconstructed signal: , where are the weight matrix and bias respectively.

[0026] Preferably, the combined evaluation score after fusing to quantitatively evaluate the effect of signal enhancement in step S3 is: , where are the peak signal-to-noise ratio, mean square error, and structural similarity index of the signal respectively. If is greater than 40, it indicates that the enhancement is effective.

[0027] Compared with the prior art, the advantages and positive effects of the present invention are as follows. In terms of model construction, gated residual units are introduced to dynamically control the signal feature propagation path, effectively retaining useful features and suppressing interference from irrelevant information, and solving the problems of feature redundancy and information loss in traditional models. Through the multi-scale feature fusion technology, the attention mechanism is used to weight and fuse features of different scales, comprehensively capturing multi-level features of the signal and improving the signal reconstruction accuracy and robustness. During training, the residual structure avoids gradient vanishing or explosion, ensuring stable convergence of training. In terms of the optimization strategy, a reinforcement learning environment is constructed, abstracting signal enhancement as a reinforcement learning problem, using the signal quality index as the state, model parameter adjustment as the action, and enhancement effect as the reward to achieve adaptive enhancement and improve the robustness and generalization ability of the enhancement effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0029] Figure 1 It is a schematic structural flow diagram implemented by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] In order to more clearly understand the above objects, features and advantages of the present invention, the following will further illustrate the present invention with reference to the drawings and embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0031] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the present invention is not limited by the specific embodiments disclosed in the following specification.

[0032] Embodiment. In modern communication systems, due to noise interference, multipath effects and signal attenuation in complex communication environments, the signal quality often fails to meet the requirements of high-performance communication. Especially in high-speed, long-distance or multi-interference-source scenarios, the signal distortion and error code phenomena increase significantly, greatly limiting the practical application ability of the system. To overcome the above problems, the present invention proposes a signal enhancement method for SparkLink communication based on a deep neural network. The implementation process is as Figure 1 shown.

[0033] To improve the accuracy and spatio-temporal consistency of signal acquisition, a distributed SparkLink technology collaborative acquisition scheme is adopted. Considering arranging multiple SparkLink acquisition nodes within the communication area to collect signals, in order to ensure the synchronization accuracy. The nodes perform clock calibration through the ultra-low latency synchronization link of the SparkLink technology, adopting the master-slave synchronization method. One high-precision node is used as the master node, and the remaining nodes are used as slave nodes. The slave nodes adjust their own acquisition moments according to the master node's clock information, making the overall acquisition network like a closely collaborating whole, obtaining signal samples with extremely high spatio-temporal resolution, and collecting the signal-to-noise ratio, latency, and bit error rate of the signals.

[0034] Considering the traditional recurrent neural network, due to the lack of a dynamic control mechanism, the traditional model is prone to introducing irrelevant features or losing key signal information, resulting in an unsatisfactory signal enhancement effect. Its ability to capture multi-scale features is weak, and it cannot effectively restore complex signals. In addition, the traditional model is prone to being restricted by gradient vanishing or explosion during deep training, affecting the training effect and performance of the model. The present invention introduces a gated residual unit, which can dynamically control the propagation path of signal features, effectively retain useful features, and suppress the interference of irrelevant information, solving the problems of signal feature redundancy and information loss in the traditional model. In addition, through the multi-scale feature fusion technology, the model uses the attention mechanism to weight and uniformly fuse features of different scales, enabling it to more comprehensively capture the multi-level features of signals, significantly improving the accuracy and robustness of signal reconstruction. In the training of the deep network, the improved model adopts a residual structure, avoiding the problem of gradient vanishing or explosion that occurs when the traditional recurrent neural network is stacked in multiple layers, ensuring the stability and convergence of training. The construction of the signal enhancement model of the improved residual recurrent neural network is as follows: First, the input signal is denoised through preprocessing, and the signal data is normalized to meet the input requirements of the model. A gated residual unit is introduced to control the information flow of the signal, and the state update of the recursive step is: , where is the hidden state at time step t, is the residual mapping function, capturing the dependence between the current signal and the historical signal, is the gating function, dynamically determining the signal transmission method according to the characteristics of the current signal, and calculated by the following formula: , where are trainable parameters, is the Sigmoid activation function, is the denoised input signal, represents element-wise multiplication. After introducing the gated residual unit to control the information flow of the signal, the complexity index of the signal is calculated according to the key features of the signal: , where is the preset maximum signal-to-noise ratio threshold, is the current signal-to-noise ratio, represents the preset delay tolerance threshold, is the current signal delay, is the current bit error rate, is the preset bit error rate threshold; by calculating the complexity index of the signal, if triggers model adjustment, that is, increasing the number of gated residual units in the residual recurrent neural network and expanding the network depth. Then, a multi-layer stacked residual recurrent neural network is adopted, where each layer uses the output of the previous layer as input and is processed through residual units. Finally, the signal reconstruction performance is optimized through multi-scale feature fusion. The implementation steps are as follows: First, the hidden state outputs of each layer of the residual recurrent neural network are combined to form a multi-scale feature set; weights are assigned to different features through an attention mechanism and fused into a unified representation: where represents the importance score of a specific scale feature, which is automatically learned through network training; the fused feature is processed through a linear transformation and a non-linear activation function to generate the reconstructed signal: where

[0035] are the weight matrix and bias respectively. where are the peak signal-to-noise ratio, mean square error, and structural similarity index of the signal respectively. If is greater than 40, it indicates that the enhancement is effective. This evaluation method intuitively reflects the actual effect of signal enhancement and can accurately guide the subsequent optimization process. Then, for less than 40, a reinforcement learning-based optimization model is adopted.

[0036] The present invention constructs a reinforcement learning environment, abstracts the signal enhancement process as a reinforcement learning problem, uses signal quality metrics (such as peak signal-to-noise ratio, mean square error, and structural similarity index) as states, regards the adjustment of model parameters as actions, and uses the enhancement effect as the reward. The model can dynamically optimize the parameter settings to achieve adaptive enhancement. First, the signal enhancement process is abstracted as a reinforcement learning framework. The state is the current peak signal-to-noise ratio, mean square error, and structural similarity index of the signal, the action A is the adjustment of the enhancement network parameters; the reward function is ; the input layer is the current state , a two - layer fully - connected network is used, with 128 neurons in each layer, the activation function is ReLU, and each action corresponds to an action value to select the optimal action; start training, initialize the parameters of the reinforcement learning model, and use -greedy policy to select actions, and enhance the signal according to the selected actions, and calculate the new state , use the reward function to calculate the reward value; update the Q - value function, and the optimization strategy is: , where is the maximum Q - value of all possible actions in the next state ; if is greater than 40, stop enhancement, otherwise continue enhancement. This design enables the model to flexibly adjust the strategy in different complex environments, improving the robustness and generalization ability of the enhancement effect. At the same time, the reinforcement learning model adopts -greedy policy to explore the optimal action combination, making the signal enhancement process not only have the ability of local optimization but also achieve the maximization of long - term effects. In addition, this optimization framework combines a deep neural network as the policy model, which can efficiently process multi - dimensional complex signals, ensuring the stability and accuracy of the optimization process.

[0037] The above - mentioned are only the preferred embodiments of the present invention, and are not limitations to the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for enhancing star flash communication signals based on deep neural network, characterized in that: The following steps are involved: S1. First, distributed star flash nodes are used to collect signal samples and collect signal noise ratio, delay and bit error rate; S2, construct a signal enhancement model of improved residual recurrent neural network; S3, comparing and analyzing the enhanced signal with the expected signal standard to quantitatively evaluate the effect of signal enhancement; S4. Construct a reinforcement learning environment to enhance and optimize the signals that do not meet the quantitative evaluation requirements. The specific implementation is as follows: S41. First, the signal enhancement process is abstracted into a reinforcement learning framework. is the current peak signal-to-noise ratio, mean square error and structural similarity index of the signal, action A is the adjustment of the enhanced network parameters; the reward function is ; S42, input layer is the current state , using a two-layer fully connected network, each layer has 128 neurons, the activation function is ReLU, each action corresponds to an action value, and the optimal action is selected; S43, start training, initialize reinforcement learning model parameters, use - Greedy strategy selects actions, enhances the signal based on the selected actions, and calculates the new state , using the reward function Calculate reward value; S44, update the Q value function, the optimization strategy is: ,in For the next state All possible actions The maximum Q value of S45, if If it is greater than 40, the enhancement stops, otherwise it continues; S5. Finally, the enhanced signal data is obtained.

2. According to the method for enhancing star flash communication signals based on deep neural network in claim 1, it is characterized in that: The construction of the signal enhancement model of the improved residual recurrent neural network in step S2 is: S21, first input signal After preprocessing, the signal data is denoised and standardized to meet the input requirements of the model; S22, introduce the gated residual unit to control the information flow of the signal, and the state of the recursive step is updated as follows: ,in is the hidden state at time step t, is the residual mapping function, which captures the dependency between the current signal and the historical signal. is the gating function, is the denoised input signal, represents element-wise product; S23, using a multi-layer stacked residual recurrent neural network, each layer uses the output of the previous layer as input and processes it through a residual unit; S24. Finally, the signal reconstruction performance is optimized through multi-scale feature fusion.

3. According to the method for enhancing star flash communication signals based on deep neural network in claim 2, it is characterized in that: After introducing the gated residual unit to control the information flow of the signal, the complexity index of the signal is calculated according to the key characteristics of the signal: ,in To preset the maximum signal-to-noise ratio threshold, is the current signal-to-noise ratio, Represents the preset delay tolerance threshold, is the current signal delay, is the current bit error rate, is the preset bit error rate threshold; by calculating the complexity index of the signal, if Trigger model adjustment, that is, increase the number of gated residual units in the residual recurrent neural network and expand the network depth.

4. According to the method for enhancing star flash communication signals based on deep neural network in claim 2, it is characterized in that: The gate function in step S22 It is learned through an independent network module, which dynamically determines the signal transmission method according to the characteristics of the current signal, and is calculated by the following formula: ,in is a trainable parameter, is the Sigmoid activation function.

5. According to a method for enhancing star flash communication signals based on deep neural network according to claim 2, it is characterized in that: The steps for implementing the optimization of signal reconstruction performance by multi-scale feature fusion in step S24 are as follows: S241, first output the hidden state of each layer of residual recurrent neural network Composing a multi-scale feature set; S242. Empower different features through the attention mechanism and integrate them into a unified representation: ,in The importance score of the feature at a specific scale is automatically learned by network training; S243, the fused features The reconstructed signal is generated by processing through linear transformation and nonlinear activation function: ,in are the weight matrix and bias respectively.

6. The method for enhancing star flash communication signals based on deep neural network according to claim 1, characterized in that: The integrated evaluation score after fusion of the quantitative evaluation signal enhancement effect in step S3 is: ,in are the peak signal-to-noise ratio, mean square error, and structural similarity index of the signal respectively. If A value greater than 40 indicates that the enhancement is effective.

Images