Underwater applications such as tomography often require the estimation of acoustic mode time series from vertical array data. Mode filters are designed to pass a desired mode while suppressing noise and interference. For transient signals or non-stationary environments, the mode filter must adapt its weights over time. We developed the Performance-Weighted Blended (PWB) mode filter that combines pseudo-inverse (PI) mode filters of different ranks to achieve results better than or equal to a single PI filter IEEE Oceans 2022, Asilomar 2023. The PWB mode filter is universal, similar to Buck and Singer's universal beamformer IEEE SAM 2018. Performance depends on the user's choice of filters to include in the set. This talk explores a different approach to blended mode filter design. Instead of blending a discrete filter set, we use a modified eigenspace beamformer operating in the signal subspace defined using Gram–Schmidt orthonormalization of the sampled modeshapes. This eliminates the ad hoc filter design required in the previous approach. Blending different diagonal loading levels in the signal subspace with the PWB algorithm protects against modeshape mismatch. This talk compares the original PWB mode filter with the new eigenspace design. Performance is evaluated using simulated and experimental data. Work supported by ONR.
Chakrabarti et al. (Tue,) studied this question.