A Weak Signal Detection Method Based on Spatial Spectrum-LSTM Neural Network

In this paper, we propose a weak signal detection method based on spatial spectrum-long short-term memory (LSTM) neural network to address the problem that the traditional blind detection method of weak signals is not effective in the condition of low signal-to-noise ratios. We firstly exploit the difference between the spatial spectrum transformed signal and noise to determine whether there is a weak signal. Then, the LSTM neural network is used for feature learning to classify different samples. It can avoid the influence of the detection threshold on the detection performance of the system. Numerical results show that the detection performance of our method outperforms LSTM neural network, radial basis function neural network, traditional maximum-minimum eigenvalue, and energy detection methods.

I. Introduction

The iterative update of communication technology has made the electromagnetic environment more complex, and many other electromagnetic signals will seriously interfere with the transmission and reception of target signals. Furthermore, the effects of noise and fading of the wireless channel cause the signal to be easily annihilated and difficult to detect. This will significantly degrade the quality of the received signal, and it will become very difficult for the receiver to process the signal. The weak signal detection technology is a promising method to resolve the problem. Various kinds of weak signal detection methods have been widely proposed, and the linear and nonlinear theories in different fields have been applied. In the weak signal detection field, the signal-to-noise ratio (SNR) of the receiving end is improved by weakening the background noise or enhancing the strength of the useful signal, so as to achieve the purpose of effectively detecting and extracting the useful signal. However, certain prior information is required. The problem of signal existence detection simplifies the detection purpose. Blind detection methods in the field of spectrum sensing can be introduced for detection, mainly including energy detection (ED) and covariance-based detection methods.