Spatial wavelet packet denoising for improved DOA estimation

Abstract

The performance of direction-of-arrival (DOA) estimation techniques such as MUSIC degrades progressively with decreasing signal-to-noise ratio (SNR). The DOA estimation performance may be improved by employing a pre-processor that enhances the SNR, before performing the DOA estimation. In this paper, a denoising technique based on the use of wavelet packet transform in the spatial domain is proposed for enhancing the output SNR of a uniform linear array of sensors receiving narrowband signals in the form of plane waves from different directions. The technique involves the use of a spatial wavelet packet transform (SWPT) followed by a block thresholding scheme based on the norm of SWPT subvectors in different spatial frequency subbands. This method has the advantage of not requiring the high sampling rates demanded by the temporal wavelet denoising techniques. It is shown through simulations that SWPT denoising (SWD) requires a sampling rate that is just 2-4 times the signal frequency, whereas temporal wavelet denoising (TWD) requires a much higher sampling rate for achieving a comparable SNR gain. Consequently, at lower sampling rates, the DOA estimation performance indices, such as bias, mean square error and resolution, achieved by SWD are much superior to those achieved by TWD or by undenoised data.