Compressive spectrum sensing with spectral priors for cognitive radar

Abstract

This paper proposes a new sparse spectrum sensing framework for cognitive radars, by combining the ideas of coprime sampling and atomic norm line spectrum estimation. Cognitive radars need to scan a large frequency band to detect presence of other radio users and find available spectral holes for opportunistic transmission. This necessitates the use of expensive A/D converters operating at very high sampling rates. Coprime sampling can be highly effective in such a scenario, since it allows spectrum sensing at significantly lower sub-Nyquist rates. This paper demonstrates how coprime sampling can enable sparse spectrum sensing at sub-Nyquist rates by using an atomic-norm minimization based reconstruction framework. Traditional atomic norm based methods, when used with coprime sensing, can lead to false detection of spectral lines, especially in presence of noise and limited data. By exploiting spectral priors (such as partial knowledge of available holes) available to the cognitive radar, coprime sampling and atomic norm based spectrum sensing can successfully avoid such false detection and enable efficient spectrum sensing at sub-Nyquist rates.