The biquad filter before the detector can be used to implement a
frequency-dependent compression threshold. For example, assume
that the overload point of the woofer is very frequency depen-
dent. In this case, one would have to set the compressor threshold
to a value that corresponded to the most sensitive overload fre-
quency of the woofer. If the input signal happened to be mostly
in a frequency range where the woofer was not so sensitive to
overload, then the compressor would be too pessimistic and the
volume of the woofer would be reduced. If, on the other hand,
the biquad filter were designed to follow the woofer excursion
curve of the speaker, then the volume of the woofer could be
maximized under all conditions.This is illustrated in Figure 15.
incoming sampling rate. However, when the de-emphasis filter is
implemented digitally, the response will scale with the sampling
rate unless the filter coefficients are altered to suit each possible
input sampling rate. For this reason, the AD1954 includes three
separate de-emphasis curves: one each for sampling rates of
32 kHz, 44.1 kHz, and 48 kHz.These curves are selected by
writing to Bits 5 and 4 of Control Register 1 over the SPI port.
Alternatively, the 44.1 kHz curve can be called upon using the
DEEMP/SDATA_AUX pin.This pin is included for compatibility
with CD decoder chips that have a de-emphasis output pin.
Using the Sub Reinjection Paths for Systems with No Subwoofer
Many systems will not use a subwoofer but would still benefit
from two-band compression/limiting.This can be accommodated
by using sub reinjection paths in the program flow.These param-
eters are programmed by entering two numbers (in 2.20 format)
into the parameter RAM. Note that if the biquad filters are not
properly designed, the frequency response at the crossover point
may not be flat. Many crossover filters are designed to be flat in
the sense of adding the powers together, but nonflat if the sum is
done in voltage mode.The user must take care to design an appro-
priate set of crossover filters.
20Hz
200Hz
20Hz
200Hz
FREQUENCY
FREQUENCY
Figure 15. Optimizing Woofer Loudness Using the
Subwoofer rms Biquad Filter
Interpolation Filters
The left and right channels have a 128:1 interpolation filter with
70 dB stop-band attenuation that precedes the digital -
modu-
lator.This filter has a group delay of approximately 24.1875/fS
taps, where fS is the sampling rate.The sub channel does not use
an interpolation filter.The reason for this (besides saving valuable
MIPS) is that it is expected that the bandwidth of the sub output
will be limited to less than 1 kHz.With no interpolation filter, the
first image will therefore be at 43.1 kHz (which is fS – 1 kHz for
CD audio).The standard external filter used for both the main
and sub channels is a third order, single op amp filter. If the cut-
off frequency of the external subwoofer filter is 2 kHz, then there
are more than four octaves between 2 kHz and the first image
at 43.1 kHz. A third order filter will roll off by approximately
18 dB/oct
4 octaves = 72 dB attenuation.This is approximately the same as the digital attenuation used in the main channel
filters, so no internal interpolation filter is required to remove the
out-of-band images.
When using a filter in front of the detector, a confusing side effect
occurs. If one measures the frequency response by using a swept
sine wave with an amplitude large enough to be above the com-
pressor threshold, the resulting frequency response will not look
flat. However, this is not real in the sense that, as the sine wave is
swept through the system, the gain is being slowly modulated up
and down according to the response of the biquad filter in front of
the detector. If one measures the response using a pink noise gen-
erator, the result will look much better, since the detector will settle
on only one gain value.The perceptual effect of the swept sine wave
test is not at all what would be predicted by simply looking at the
frequency response curve; it is only the signal path filters that will
affect the perception of the frequency response, not the detector
path filters.
De-emphasis Filtering
The standard for encoding CDs allows the use of a pre-emphasis
curve during encoding, which must be compensated for by a
de-emphasis curve during playback. The de-emphasis curve
is defined as a first order shelving filter with a single pole at
(1/(2
50 µs)) followed by a single zero at (1/(2
15 µs)).
This curve may be accurately modeled using a first order digital
filter. This filter is included in the AD1954; it is not part of the
bank of biquad filters and so does not take away from the num-
Note that by having interpolation filters in the main channels
but not the subwoofer channel, there is a potential time-delay
mismatch between the main and sub channels. The group delay
of the digital interpolation filters used in the main left/right
channels is about 0.5 ms. This must be compared to the group
delay of the external analog filter used in the subwoofer path. If
the group-delay mismatch causes a frequency response error
(when the two signals are acoustically added), then the pro-
grammable delay feature can be used to put extra delay in either
the subwoofer path or the main left/right path.
Since the specification of the de-emphasis filter is based on an
analog filter, the response of the filter should not depend on the
REV. A
ADI [ ADI ]
