Abstract
Chi and Wu proposed a unified class of inverse filter criteria J <sub>r,m</sub> using an rth-order and an mth-order cumulants (where r is even and m>r≥2) which includes Wiggins' criterion, Shalvi and Weinstein's criterion and Tugnait's criteria as special cases, for blind deconvolution and equalization with only non-Gaussian output measurements of a nonminimum phase linear time-invariant (LTI) system (channel) h(n). In this paper, we theoretically prove that for finite SNR, as Mendel's (nonblind) minimum-variance deconvolution (MVD) filter, the optimum inverse filter v(n) associated with the criteria J <sub>2,m</sub> is a perfect phase equalizer but not a perfect amplitude equalizer, and the latter approaches the former as either m or SNR is increased or as the system h(n) has wider bandwith. For the other J <sub>r,m</sub> (r≥4), perfect equalization can be attained at the expense of SNR degradation. Finally, some simulation results are provided to support the proposed analytic results.