Signal detection in non-stationary broadband noise

An algorithm for signal detection across a wide band of frequencies in which the noise statistics may be considered stationary only over very short intervals is proposed. The algorithm estimates the noise level in each channel of a channelized receiver, based on the levels in the quietest channels, after compensating for time-varying passband ripple. Methods for estimating the actual processing gain achieved, adjusting for misestimates in the noise levels, subtracting coarse ripple using calibrations, and tracking a channel´s noise-level drift are developed. Theoretical results are compared with those obtained from digital simulations and in laboratory tests. It is shown that the actual loss in detection sensitivity, because of the margin required to compensate for drift, coarseness of the calibrations, and loss of actual processing gain versus theoretical gain, is dependent on environmental and system details and must be determined from extensive live testing. Such tests enable the algorithm parameters to be set and the level of detection sensitivity to be determined