Adaptive Space-Frequency Basis Selection Methods for Signal Frequency Band Sensing and Fast Wideband DOA Estimation With Passive Sonar Data Verification

Abstract

In the wideband direction of arrival (DOA) estimation problem, signal measurement equipment typically receives wideband signals with different frequency bands in practice. Thus, the commonly adopted assumption in most relevant methods that the incident signals share the entire frequency band under monitoring is inappropriate, leading to degradation in DOA estimation accuracy of the measurement equipment. To overcome this challenge, we propose the adaptive space-frequency basis selection (ASFS) based on maximum likelihood (ML), namely ASFS-ML, to adaptively sense the occupied frequency band of each signal. Then, the ML-based DOA refinement procedure is further constructed to realize more accurate DOA estimation by only utilizing the identified frequency band of each signal. Besides, we also propose a denoised ASFS-ML (DASFS-ML) method to eliminate the influence of the inappropriate noise initialization. Moreover, considering the limited computational resources of small measurement devices, our methods adopt the dot-estimation strategy instead of a general scheme that estimates the whole space-frequency spectrum. This strategy only activates the space-frequency bases occupied by the incident signals, greatly improving the computational efficiency. Numerical simulations in this article are set in underwater passive sonar detection scenarios, and the experimental results over 100 Monte-Carlo trials demonstrate that our methods achieve lower DOA estimation error and computational cost. On the SWellEx-96 underwater acoustic data measured by a passive sonar system, our methods (ASFS-ML and DASFS-ML) exhibit average absolute DOA estimation errors of 1.789° and 1.671°, with their average running times of 0.315 and 1.691 s, respectively. These results indicate the superiority of our methods over the existing sparse representation (SR)-based approaches.