6.2. Estimating the TF Matrix
Based on the simulation results in Figure 4 and Figure 6, it can be observed that ROBNCA and NCA achieve the minimum NSME in both the noise case and noise + outliers case. Moreover, the existence of outliers does not have an obvious impact on the performance of ROBNCA and NCA. In contrast, the performance of FastNCA and NINCA for estimating the TF matrix S is not robust to outliers. Unlike the good simulation results in estimating A, the performance of PosNCA for estimating S is significantly inferior to all other algorithms. That is probably because PosNCA utilizes the least-squares solution to derive S once obtaining the estimate of A, which is numerically unstable.