Detection of unknown signals in unknown, non-stationary noise

This work considers the problem of low-latency detection of unknown signals that appear intermittently in noise having unknown and perhaps non-stationary statistics. We show how to realize a detector having a guaranteed probability of false alarm with best in-class probability of detection and lowest complexity. The detector adaptation is based on a state-machine driven feedback path that selectively extracts the noise statistics required for setting a test threshold. Performance and complexity comparisons are made between the new detector, existing detectors in its class, and the ideal (clairvoyant) detector.