Evaluation of In-Car Communication Systems - Digital Signal Processing for In-Vehicle Systems and Safety - page 79

Microphone
spectra used
for listening
passengers

Signal
distribution

Micro-
phone
signals

Highpass
filter

Preemphasis
filter

Low delay
analysis filter bank

Signal analysis
(clipping and
failure detection)

Microphone
spectra used
for talking
passengers

Fig. 5.7 Preprocessing and analysis filter bank

5.3.1 Analysis and Synthesis Filter Banks

Since typical background noise in automotive environments is dominating over
speech components in the low-frequency range, high-pass filtering is applied as a
first processing stage applied to all microphones - see Fig. 5.7 . The cutoff fre-
quency of the high-pass filters should be chosen between 80 and 300 Hz depending
on the preferences of the user and on the type of vehicle (higher frequencies for
sports cars, lower frequencies for sedans and vans).
A few signal processing components such as feedback and echo cancellation or
adaptive beamforming (these units will be described in the following sections)
assume a linear transmission between the loudspeakers and the microphones
installed in the vehicle. If a microphone clips, this assumption is definitely violated.
In order to exclude these periods, e.g. during the adaptation phase of cancellation
filters, each microphone is analyzed not only in terms of a clipping detection but
also in terms of a complete failure as caused, for example, by a breakdown of the
power supply for the microphone amplifier. If the failure of a microphone is
detected, the sensor is excluded from the succeeding processing.
A preemphasis filter and a low-delay analysis filter bank follow the high-pass
filter. The preemphasis stage can be realized, e.g. as a fixed prediction-error filter
that operates in the time domain. Its coefficients are adjusted to whiten the signal,
leading to a smooth high-pass characteristic. The inverse filter will be applied
behind the corresponding synthesis filter bank. Preemphasis and corresponding
de-emphasis filters help to utilize the limited spectral resolution of filter banks
(due to aliasing effects) in the best way.
Finally, the spectra of the microphone signals are grouped into two classes: Spectra
of the first class are used by those processing units that enhance the signals of the
talking passengers. Spectra of the second class are utilized for adjusting algorithmic