Robust spectral estimation in the presence of multipath

Summary form only given. Signals arising from realistic collections environments may contain unwanted multipath components resulting in overlapping wavelets. Such signal components introduce extra sinusoidal transient signals into the data record, and robust spectral estimation algorithms are required. It is reported that an L₁ solution to a set of linear prediction equations provides a spectral estimate of sinusoid location with less bias and no line splitting as compared to a least-squares solution. Furthermore, computing the kurtosis of the residual vector results in an effective hypothesis test for determining whether a multipath component is present in the collected signal. Thus, the added computational expense of computing a non-least-squares solution is incurred only if required. It has been found that, in the presence of a multipath component, the L ₁ solution to a set of FBLP equations provides sinusoidal frequency location with less bias and no line splitting in the test cases tried so far