──────────────────── ────────────────────
DIGITAL SOUND,
or dot the i's
──────────────────── ────────────────────
(c) VTS'99
I think this question is poorly covered
in the ZX press, so there are some
misconceptions The material published here is
I think it will eliminate some of them.
Signed (bipolar) DAC.
Someone said that this feature has
better sound quality than unipolar
(regular) DAC. This is probably purely subjective
tive assessment, because this could happen in nature
cannot.
As far as I understand, the bipolar code is
provide digitizers that will be used to digitize
samples (although I'm unfamiliar with PC). Same code
you can then throw it into a bipolar DAC, and
for unipolar you need to recalculate
(add #80). Everything would be fine, but the scheme
technically a bipolar DAC is usually a little
more difficult, but to recalculate samples into
unipolar code - a matter of seconds, i.e. is-
It is more profitable to use a unipolar DAC.
And so, they tell us that, they say,
your unipolar DACs are a must.
In fact, the only its distinguishing feature is
which - inthe formation of a constant component -
at level #80, which is then filtered
is checked by transition capacitors in
ULF. Moreover, level #80 is constant only
because we took a digitized bipolar
nal signal and shifted it to this value.
If you try to generate some RND
signal, then this level will shift and,
Moreover, it may change. If anyone doesn't
knows, I’ll explain the concept of a constant component -
signal (Fig. 1).
Fig. 1
So, the constant component U0 is
average signal value over some time
name T. Moreover, if you calculate the area, the limit
lined by the lines Ot, t=0, t=T and the
signal, then it will be equal to the area of the straight
triangle U0*T. And if you add up the areas
signal above and below the line ox, then they
will be equal. It was before this tension
transition capacitors are charged, and
then they “subtract” it from the signal, leavingload only has a variable component.
Moreover, the time T is determined by the inertia
capacitor (i.e. time constant
("tau") t=Rh*C).
So, if the previously used sig-
cash with U0<>#80, convert and submit to
bipolar DAC, there also will appear permanent
this componentand it will also be filtered
capacitor. Separately, I want to warn
supporters of bipolar transformation from
attempts to get rid of the hated condensation
sator, using an amplifier with galvanic
what connections. Hitting the mentioned post-
y component is fraught with damage to your
favorite 100-watt speakers (if you don't
provided for protection).
“Why then do we even need bipolar
DACs, someone invented them, right?” - ask
You. Yes, they are needed, but not for getting di-
gital sounds, but let's say for control
engine (changing the direction of rotation
niya), and you never know what else...
Dynamic range.
A very controversial concept. Some people think
its ratio of max signal value to
noise level, and others - to the min level
signal values, and the first ones get
a much larger figure (which advertises-
xia), and as the latter think, I andnot at all
I understand (since the criterion is min signal level
standssubjective assessmentof its quality
- [1], which is unacceptable). The whole joke is in
that in analogue technology these two op-
The divisions are identical, because at low levels
nyah signal active amplifier elements
practically do not introduce distortions, because
how to work in a narrow area of characteristic
tics, where it is practically linear (for
switching on mode B, which gives significant
distortion even with a large signal). A min
the signal level is limited by its diversity
cleanliness against the background of noise, i.e. actually
noise level.
In the digital channel there is only one non-
removable noise, "by nature" - quanto noise
vania. At the same time, all other noises are com-
mutational, from power circuits, active noise
nal elements of the analog path, thermal
noises, etc., apparently, are so easy to fix
so vulnerable that you might not even remember them
nat, and enter the coveted figure in the advertisement
96 dB... Let's leave these noises on the conscience of the rec-
Lama, and let's deal with quantization noise.
It arises for a banal reason - co-
a limited number of quantization steps,determined by the bit depth of the digital channel -
la. Because the original signal is continuous, and
digitized - discrete, then always su-
there is a quantization error, which is all
time changes. After recovery to
analog signal this error turns
into quantization noise. The spectrum of this noise ranges
placed higher in frequency than the useful spectrum
signal, and the peak occurs at the frequency
sampling. If this noise is filtered
Well, then it will turn out completely identical
input signal. But more on that later. Let u
We don't have any filters. Then according to
[2] the quantization noise level is determined
like:
Ksh = -(6*N+1.8) [dB],
where N is the number of digits.
Then with N=16 we get Ksh=-97.8 dB, and
at N=8 - Ksh=-49.8 dB
Now let's deal with the second definition of di-
namic range, truly reflecting
state of affairs.
Let's put the min amplitude of the signal in 1 unit -
nitsa of the junior category, and we get the following
expression:
D = 6*N [dB]
With N=16 we get D=96 dB, and with N=8
we have D=48 dB.
Actually, these figures are usually
sit down. However, from the initial conditions it follows
It is clear that at the minimum level the signal is received by one
nobitov! with all the attendant qualities
vital characteristics...
That is. it is necessary to set a threshold of nonli-
linear distortion. Actually in theory
it's not that simple, but in practice it's simple
appropriate measurements are made [2].
It is much easier to switch from max valid
nonlinear distortions to min permissible co-
number of ranks. Usually allowed
Khu=10%, which approximately corresponds to 4 times
row coding [2]. Maybe this is not
correct from a theoretical standpoint, but it allows
It's easy to estimate the dynamic range. Although
concepts of nonlinear distortion and bit depth
quantizations are not synonymous, one
however, both clearly define the distortion
marriage. Consequently, in the theoretical op-
division operating with max admissible
Khu.max, you can use min allowed
Nmin. Now the dynamic range can be
evaluate directly the number of times
nearby (i.e. using the binary logarithm
instead of decimal):
D2 =N-Nmin = dN [bit]
Taking Nmin=4, we have dN = N-4 [bit].
Or you can go to the decimal base -
nia:
D = 6*dN [dB]
For N=16 we have D=72 dB at Nmin=4 and
D=48 dB at Nmin=8.
Finally, I want to say that the described
The above cases are idealized. Actually
in fact, there is still a lot of noise in the tract that
not so easy to eliminate, and a lot of sources
niks of nonlinear distortions, starting from
DAC/ADC and ending with analog amplifiers
lyami, and even microphones/speakers...
You can’t just hit them head-on... But
this no longer applies to digital theory
signal processing.
Oversampling.
What is it and what is it eaten with - you can
read in SE#2 ([1]), in the article about computer
youtube music. In short, oversampling
tization is the filling of pro-
time interval between two neighboring
four additional reports, values
which are calculated by interpolation in accordance with
in accordance with the impulse responseidea
Alpha rectangular low-pass filter (note -
this is precisely the mysterious filter Ko-
Telnikova). In the simplest cases, you can
use regular linear interpolation
tion.
Thus. resampling - it's easy
digital filtering,with increased frequency
sampling. It is necessary because
in conventional sound cards there is no output
what low-pass filter? If for fkb=HHkHz it is almost
normal, then for 8 kHz samples this is
results in HF noise in the band from hkHz to
22kHz. However, quite high quality
filtering will be provided by a regular analog low-pass filter
about eighth to tenth order. So,
for 8th order you need 4 operational forces
body, i.e. only 1 housing K1401UD2 (for
stereo - two bodies). A similar filter
They are planning to do it for GS as well. According to research
yam X-Trade, its cutoff frequency for pain
The majority of samples is 10-12 kHz. By the way, in
around 16-18 kHz for most normal
people are within earshot. On TVby-
VOS 1OOHz-1OkHz (15.62SkHz - frequency
lines), on FM and audio cassettes about
1bkHz... But, for example, musicians hear
up to 22kHz. In addition, there is a hypothesis that
HF noise (far beyond 2OkHz) was still perceived
tossed around like distortions.
Okay, time to wrap things up. Actually I
I wanted to tell you about a couple more mules. But re-
I decided to put it aside for now.
Literature:
1. Spectrum Expert #2.
2. Radio, 1991, N11-12. Sound engineering,
Analog signal conversion devices
nals.
Share your thoughts about the article