Weak sinusoidal signal extraction from white noise using convolutional neural network

Abstract

A great number of analog and digital data communications schemes use the sinusoidal waveform as a basic elementary signal, including the spread spectrum data exchange techniques. Detection of the presence of the sinusoidal waveform in a mixture of signal and noise is a common task, regardless the specific modulation scheme. This paper presents the machine learning-based approach for detection of the sinusoidal wave. It presents the structure of the convolutional neural network, as well as the performance metrics for the sinusoidal signals detection. The paper provides an assessment of the overall accuracy for the binary signals. It reports the overall accuracy value of 0.93 for the sinusoidal signal detection in the presence of additive white Gaussian noise at the signal-to-noise ratio value of −20 dB for a balanced dataset.