Digital Signal Processing Reference
In-Depth Information
Figure 10-7.
LAR-distances between the clean input and various output signals: unprocessed
output (Noisy), outputs of the SAF, STDWF and SAFWF techniques in four different noisy
environments.
The results of objective evaluations are shown in Figure 10-7. For
reference, the LAR distance between the clean input (signal
s
(
n
) in Figure
10-1) and the unprocessed noisy output (signal
z
(
n
) in Figure 10-1 with all
SPBs inhibited) is also presented. The low LAR-distance improvement
obtained by the SAF indicates that this method has difficulty rejecting noise
in diffuse noise environments. Evidently, SAF has better performance in the
third and forth environmental conditions (CarWOpen and CarWClose). This
can be justified by considering Figures 10-5 and 10-6 as follows: 1) In the
CarWOpen and CarWClose cases, the noise spectrum is dominated by
lowpass and coherent components. This results in improved noise reduction
of the important lowpass components by adaptive filtering. 2) In particular,
there are highly coherent engine-related noises in these two cases that are
efficiently cancelled by the subband adaptive filters.
As is evident from Figure 10-7, the hybrid system (SAFWF) outperforms
both the SAF and the STDWF systems for all four test sets. Especially in
CarWOpen and CarWClose cases, the SAFWF method produces better LAR
distance improvements. This demonstrates that an improvement in the
adaptive filter performance leads to better performance of the Wiener filter
