Combining Capon and Bartlett spectral estimators for detection of multiple sinusoids in colored noise environments

Abstract

Detection of multiple sinusoids in colored noise environments has many potential applications, such as sonar, radar, and communication. Most of conventional algorithms often use the local signal-to-noise-ratio (LSNR) as a test statistic to detect the sinusoids, where the LSNR is estimated in the frequency domain by using the Bartlett spectral estimator (BSE). Unfortunately, the BSE has a relatively low frequency resolution, which may degrade the detection performance significantly. To solve the frequency resolution problem of the BSE, this paper proposes a two-stage hybrid algorithm to estimate the LSNR. In the first stage, the BSE is used to estimate the noise power spectral density over frequency. After obtaining the noise power spectral density, the second stage employs the Capon spectral estimator (CSE) to estimate the LSNR. The proposed hybrid algorithm significantly improves the detection performance, especially when the sinusoids are closely spaced. Simulation results show that the proposed algorit...