A gradient-based target tracking method using cumulants

We present a gradient-based cumulant method to track the signal subspace in an array signal processing scenario. This method is combined with a non-adaptive singular value decomposition (SVD) and a non-adaptive eigenvalue decomposition (EVD) to yield an adaptive virtual-ESPRIT algorithm (VESPA) for target tracking. The resulting least-mean-squared VESPA (LMS-VESPA) is of complexity O(M/sup 2/P). In addition to hardware saving, we demonstrate through simulations that, when the signals are closely spaced, block-adaptive ESPRIT suffers even from slight colored noise, and that when the SNR is poor whether the signals are close or not, LMS-VESPA is still robust to such noise.