(A) nonparametric signal detector based on zero-crossings and its performance improvement using a neural network

Abstract

In this thesis we consider the problem of nonparametric detection of signals in noisy observations, where zero-crossings are used to get the test statistic. We apply zero-crossings to known and random signal detection, and investigate the performance of detectors via computer simulation under several noise circumstance. In the known signal detection, the zero-crossing detector shows quite good performance when the signal to noise ratio is moderate-to-high. In the correlated random signal case, the zero-crossing detector also reveals quite good performance. To show this more explicitly, we compare the performance characteristics of the zero-crossing detector with those of the linear correlator and sign correlator detectors in the known signal case and with those of the correlation and sign-correlation detectors in the random signal case. Because of its simplicity and easy hardware implementation, the zero-crossing detector can be used in many real applications at low cost.