"Principal component inverse" algorithm for detection in the presence of reverberation

Detection in the presence of reverberation is often difficult in active sonar, due to the reflection/diffusion/diffraction of the transmitted signal by the ocean surface, ground, and volume. A modelization of reverberation is often used to improve detection because classical algorithms are inefficient. A commonly used reverberation model is colored and nonstationary noise. This model leads to elaborate detection algorithms which normalize and whiten reverberation. In this paper, we focus on a more deterministic model which considers reverberation as a sum of echoes issued from the transmitted signal. The Principal Component Inverse (PCI) algorithm is used with this model to estimate and delete the reverberation echoes. A rank analysis of the observation matrix shows that PCI is efficient in this configuration under some conditions, such as when the transmitted signal is Frequency Modulated. Both methods are validated with real sonar surface reverberation noise. We show that whitening has poor performance when reverberation and target echo have the same properties, while PCI maintains the same performance whatever the reverberation characteristics. Further, we extend the algorithms to spatio-temporal data. We propose a new algorithm for PCI which allows better echo separation. This new method is shown to be more efficient on real spatio-temporal data.