Signal detection by detecting departure from noise

Abstract

The short term sample spectrum is examined to determine if the sample spectrum is noise-like. The parameters measured are nonparametric and include the sample coefficient of variation, the center frequency, the bandwidth, and higher order measures of the spectrum. For a typical application in a voice grade channel, the performance of the detector is orders of magnitude better than the usual adaptive short term energy threshold detectors. The time between false alarms goes from minutes to days for a given detection probability. The detection is independent of the signal amplitude. There are no adaptive thresholds requiring signal history for operation. Signals may be detected in a very short time after the analysis is begun, the time required to accumulate enough samples to form one sample spectrum.<>