Evaluation of In-Car Communication Systems - Digital Signal Processing for In-Vehicle Systems and Safety - page 80

Loud-
speaker
signals

Enhanced
loud-
speaker
spectra

Low delay
synthesis
filter bank

Time-domain
equalizer
and delay

Non-linear
and time-variant
postprocessing

Deemphasis
filter

Uncor-
related
parts of
loudspeaker
signals

Fig. 5.8 Synthesis filter bank and time-domain postprocessing

components for improving the playback for the listening passengers. If, for example,
a setup with seven microphones, as depicted in Fig. 5.4 , is used and only front-to-rear
communication should be supported, then the four front microphones would be
grouped as microphones used for the talking passengers . The three rear microphones
could be used, e.g. for estimating the background noise level at each rear seat. These
estimated noise levels can be utilized for determining appropriate playback volumes.
Thus, the three rear microphones would be grouped as microphones used for the
listening passengers .
Figure 5.8 depicts the synthesis part of ICC systems. In a first step, the spectra of
the loudspeakers signals are transformed back into the time domain by an appro-
priate filter bank. As mentioned before, special emphasis should be placed on using
low-delay approaches, e.g. as described in [ 12 , 13 ].
If a preemphasis filter has been applied before the analysis filter bank, its inverse
has to be computed after the corresponding synthesis filter bank. Furthermore, each
loudspeaker might be equalized, either in terms of correcting undesired ripple in its
frequency response or in terms of attenuating those frequencies where the feedback
to the microphones is largest. Equalization can be applied either in the time domain,
e.g. by means of allpass-based structures [ 14 ], in the subband domain by means of
weighting factors, or as a combination of both. Usually, time-domain approaches
are used to realize sharp notch or peak filters, and frequency domain approaches are
used for smooth corrections. If spatial effects that require a certain delay for each
loudspeaker should be exploited, also the delays might be adjusted individually for
each loudspeaker.
Finally, signal processing that reduces the mutual correlation between loud-
speaker signals can be applied. This is of special importance if the feedback and
echo paths from the loudspeakers to the microphones are to be identified individu-
ally. The solution of this MIMO 11 system identification problem is not unique if the

11 MIMO is the abbreviation for multiple input multiple output.