AD1893_15 Datasheet, PDF (12/20 Pages) Analog Devices – Low Cost SamplePort 16-Bit Stereo Asynchronous Sample Rate Converter

AD1893
Cutoff Frequency Modification
The final important operating concept of the ASRC is the modi-
fication of the filter cutoff frequency when the output sample
rate (F_SOUT) drops below the input sample rate (F_SIN), i.e.,
during downsampling operation. The AD1893 automatically
reduces the polyphase filter cutoff frequency under this condi-
tion. This lowering of the cutoff frequency (i.e., the reduction of
the input signal bandwidth) is required to avoid alias distortion.
The AD1893 SamplePort takes advantage of the scaling prop-
erty of the Fourier transform which can be stated as follows: if
the Fourier transform of f(t) is F(w), then the Fourier transform
of f(k × t) is F(w/k). This property can be used to linearly com-
press the frequency response of the filter, simply by multiplying
the coefficient ROM addresses (shown in Figure 6) by the ratio
of F_SOUT to F_SIN whenever F_SOUT is less than F_SIN. This scaling
property works without spectral distortion because the time scale
of the interpolated signal is so dense (300 ps resolution) with
respect to the cutoff frequency that the discrete-time representa-
tion is a close approximation to the continuous time function.
The cutoff frequency (–3 dB down) of the FIR filter during
downsampling is given by the following relation:
Downsampling Cutoff Frequency = (F_SOUT/44.1 kHz) × 20 kHz
The AD1893 frequency response compression circuit includes a
first order low-pass filter to smooth the filter cutoff frequency
selection during dynamic sample rate conditions. This allows
the ASRC to avoid objectionable clicking sounds that would
otherwise be imposed on the output while the loop settles to a
new sample rate ratio. Hysteresis is also applied to the filter
selection with approximately 300 Hz of cutoff frequency “noise
margin,” which limits the available selection of cutoff frequen-
cies to those falling on an approximately 300 Hz frequency grid.
Thus if a particular sample frequency ratio was reached by slid-
ing the output sample frequency up, it is possible that a filter
will be chosen with a cutoff frequency that could differ by as
much as 300 Hz from the filter chosen when the same sample
frequency ratio was reached by sliding the output sample fre-
quency down. This is necessary to ensure that the filter selection
is stable even with severely jittered input sample clocks.
Note that when the filter cutoff frequency is reduced, the transi-
tion band of the filter becomes narrower since the scaling prop-
erty affects all filter characteristics. The number of FIR filter
taps necessarily increases because there are now a smaller num-
ber of longer length polyphase filters. Nominally, when F_SOUT is
greater than F_SIN, the number of taps is 64. When F_SOUT is less
than F_SIN, the number of taps linearly increase to a maximum
of 128 when the ratio of F_SOUT to F_SIN equals 1:2. The number
of filter taps as a function of sample clock ratio is illustrated in
Figure 8. The natural consequence of this increase in filter taps
is an increase in group delay.
DOWN-
SAMPLING
128
UPSAMPLING
64
0.5
1.0
1.5
2.0 F_SOUT/F_SIN
Figure 8. Number of Filter Taps as a Function of F_SOUT/F_SlN
When the AD1893 output sample frequency is higher than the
input sample frequency (i.e., upsampling operation), the cutoff
frequency of the FIR polyphase filter can be greater than 20 kHz.
The cutoff frequency of the FIR filter during upsampling is
given by the following relation:
Upsampling Cutoff Frequency = (F_SIN/44.1 kHz) × 20 kHz
Noise and Distortion Phenomena
There are three noise/distortion phenomena that limit the per-
formance of the AD1893 ASRC. First, there is broadband,
Gaussian noise that results from polyphase filter selection
quantization. Even though the AD1893 has a large number of
polyphase filters (the equivalent of 65,536) from which to
choose, the selection is not infinite. Second, there is narrowband
noise that results from the nonideal synchronization of the
sample clocks to the 16 MHz system clock, which leads to a
nonideal computation of the sample clock ratio, which leads
to a nonideal polyphase filter selection. This noise source is
narrowband because the digital servo control loop averages the
polyphase filter selection, leading to a strong correlation be-
tween selections from output to output. In slow mode, the selec-
tion of polyphase filters is completely unaffected by the clock
synchronization. In fast mode, some narrowband noise modula-
tion may be observed with very long FFT measurements. This
situation is analogous to the behavior of a phase locked loop
when presented with a noisy or jittered input. Third, there are
distortion components that are due to the noninfinite stopband
rejection of the low-pass filter response. Noninfinite stopband
rejection means that some amount of out-of-band spectral en-
ergy will alias into the baseband. The AD1893 performance
specifications include the effects of these phenomena.
Note that Figures 16 through 18 are shown with full-scale input
signals. The distortion and noise components will scale with the
input signal amplitude. In other words, if the input signal is
attenuated by –20 dB, the distortion and noise components will
also be attenuated by –20 dB. This dependency holds until the
effects of the 16-bit input quantization are reached.
–12–
REV. A