Periodic Signal Recovery and Detection from 1-Bit Measurements Using Convolutional Neural Network

Abstract

In order to lower the hardware requirements of digital systems and to reduce the amount of collected data, 1-Bit sampling technique has received much attention. Following the trend, in this paper we try to address a classic problem based on the 1-Bit sampling data—the problem of detection and estimation of periodic signals in White Gaussian Noise. To achieve the tasks, the 1-D convolutional neural networks (CNN) are used to recover the waveforms of the periodic signals from the 1-Bit measurements. Subsequently, the method of generalized likelihood ratio test (GLRT) is applied on the recovered waveforms to detect the periodic signals and to estimate their unknown parameters. The simulation results show that CNN can recover the waveforms of periodic signals with a reasonable accuracy, and the parameters of frequency, time delay, initial phase, and relative amplitude can be obtained.