基于深度学习的卫星信号调制识别算法

Satellite Signal Modulation Recognition Algorithm Based on Deep Learning

为实现卫星信号调制方式的分类,提出的高阶累积量与K最近邻算法(KNN)调制样式识别算法选取对噪声不敏感的5种高阶累积量特征参数用于信号的识别,通过KNN作为分类器对信号分类。实验结果表明,当信噪比(SNR)高于12 dB时,信号的调制方式可以被高效地识别,并且识别率趋近100%,但需要人工设计和提取特征参数。因此,提出了循环神经网络(Recurrent Neural Network,RNN)的卫星调制信号识别算法,以信号的IQ数据作为模型的输入,通过LSTM进行分时特征提取,全连接层进行分类,最终完成识别。在采样长度等于512,SNR大于4 dB时,识别率趋近100%。与KNN相比,LSTM网络的识别性能更为优越,尤其在低SNR的情况下,可以高效识别6种调制方式。

In order to classify the modulation modes of satellite signals,the proposed high-order cumulant and K-Nearest Neighbor(KNN)modulation pattern recognition algorithm selects five high-order cumulant characteristic parameters,which are insensitive to noise,to identify the signals,and uses KNN as the classifier to classify the signals.The experimental results show that when the Signal-to-Noise Ratio(SNR)is higher than 12 dB,the signal modulation method can be effectively identified,and the recognition rate is close to 100%,but it requires manual design and extraction of characteristic parameters.Therefore,a satellite modulation signal recognition algorithm based on Recurrent Neural Network(RNN)is proposed.The IQ data of signal is used as the input of the model,time-sharing feature extraction is carried out by LSTM,full-connection layer is classified,and finally the recognition is completed.When the sampling length is equal to 512 and the SNR is greater than 4 dB,the recognition rate approaches 100%.As compared with KNN,the recognition performance of LSTM network is superior,especially in the case of low SNR,it can efficiently identify six modulation modes.