A Novel Noise-Enhanced Back-Propagation Technique for Weak Signal Detection in Neyman–Pearson Framework

Abstract

In this paper, we propose a noise enhanced neural network based detector. The proposed method can detect the known weak signal in additive non-Gaussian noise. Carefully injected noise in a neural network enhances the weak signal detection performance. During training, the back-propagation algorithm achieves less error and it converges faster with the addition of the external noise. The optimum value of external noise is calculated theoretically and justified by simulation. This method excels over the traditional neural network based detectors in terms of its performance characteristics i.e., the probability of detection ($$P_D$$) at some specified probability of false alarm ($$P_{FA}$$). Performance of the noise enhanced neural network based detector under several signal-to-noise ratio environments are also compared with state-of-the-art detectors.