Research On Satellite Signal Modulation Pattern Recognition Algorithm Based On Deep Learning

In recent years,the application scenarios of satellite communication have been broadened,and usually different signal modulation patterns are used because of different communication environments or communication needs.In order to effectively identify the modulation pattern of satellite signals,and considering the limitation of computing power and storage space of on-board computer,this paper researches an embeddable deep learning based satellite signal modulation pattern identification algorithm,and the research process is as follows.(1)This paper first summarizes and analyzes the current research status of signal modulation pattern recognition through data research,and points out the problems of decision theory-based methods,feature extraction-based methods and deep learningbased methods at the present stage.The recognition method based on decision theory requires a large amount of a priori knowledge,while the complexity of operation is high and the adaptability is poor;the recognition method based on feature extraction carries out modulation recognition by extracting expert features,however,one expert feature is not applicable to all modulation patterns,so the generality of the method is poor;the deep learning algorithm is driven by data and has excellent feature extraction capability,so the deep learning-based recognition method has enough robustness and versatility to cope with the task of identifying multiple modulation styles in various communication states.However,the deep learning-based signal modulation pattern recognition algorithm at the present stage is usually of high complexity and requires a large amount of hardware and software resources for data processing equipment,which makes it difficult to deploy the application on embedded devices.(2)To address the problem of high complexity of the current deep learning-based signal modulation pattern recognition algorithm,this paper proposes a signal modulation pattern recognition algorithm based on multi-scale timing features,which significantly reduces the complexity of the algorithm without degrading the recognition performance.The algorithm constructs the algorithm model by cascaded convolutional network and long short-term memory network,takes the original I/Q signal as the input of timing data,firstly extracts the timing data of different time scales by using onedimensional convolutional layers,and then achieves the fusion of features by connecting the outputs of different convolutional layers across layers,so that the LSTM network layers can extract timing features better and more efficiently,which finally makes the algorithm for signal modulation pattern The final result is that the algorithm achieves excellent results in the recognition of signal modulation patterns.In addition,this paper designs and optimizes the network structure by reducing the network width,using small-sized convolutional kernels and using reduced-dimensional convolutional units,which reduces the complexity of the proposed algorithm.The experimental results show that the recognition accuracy of the proposed algorithm for 11 modulated signals reaches more than 90% when the signal-to-noise ratio is greater than 4 d B;compared with the deep learning algorithms with the same recognition accuracy,the proposed algorithm is more efficient,the number of model parameters is reduced by more than 89%,and the training time is shortened by more than 17%.The algorithm is also verified on embedded devices Jetson Nano and Raspberry Pi 4B,and the inference time of the proposed algorithm is shortened by more than 67% compared with the algorithm with the same recognition accuracy,so the proposed algorithm is more in line with the requirements of embedded applications.(3)For satellite communication scenarios,this paper uses the GNU Radio software radio development framework to build a satellite digital video signal processing system and simulate the generation of satellite digital video signals.After slicing,normalization and category labeling of the simulated satellite signal data,a satellite communication signal dataset is obtained.This dataset is then used to verify the ability of the proposed algorithm in this paper to identify the modulation pattern of satellite signals,while research experiments are conducted for different input data lengths and amplitude phase data inputs.The data length of the acquired signal samples may be different due to the variation of transmission rate and receiver sampling rate.The experimental results show that the proposed algorithm can effectively identify the signal samples with different data lengths under high SNR conditions;under low SNR conditions,the identification accuracy of the proposed algorithm is higher as the data length of the signal samples grows,because the sample data contains more information.Amplitude phase data is a common form of data input in satellite signal modulation style recognition,which can be obtained by I/Q data conversion.The experimental results show that the recognition accuracy of amplitude phase data input is lower in the low signal-to-noise interval compared with I/Q data input,mainly because the I/Q data contains more original information.