How do I get the sound device
more bits?
Suppose we do not have DMA US, GS or
Such devices, which give the hardware volume. Otherwise there
would the problem was solved too simple (one subchannel in, for
example, left ear give a volume of 64, the other in the same ear
- With volume 1). We have a standard Spectrum with no more than
Covox'om.
So, for him there were at
least three fundamentally different methods.
1. PWM (pulse width modulation).
Signal with a frequency divided by
intervals shorter than 1 / 20000 seconds in each of which
contains two adjacent voltage levels. Depending on the ratio of
durations of these levels within the time intervals at the exit
from filter produced by these or other intermediate levels of
stress. The condition of particle quency it is necessary that
the signal is not whistling.
2. PWM. The same thing, but not alternate adjacent levels,
and 0 and level. In addition to high ripple and low volume,
this method contains the problem of implementation - unknown
how to arrange for him a table to
encode 16-bit real-time.
3. Method without a title. I thought to
it is the name of the PCM (Pulse Code Modulation), but Michael
Klinkovsky flatly refused to accept it, but without showing
their istochnikov.Zdes the DAC is given signal level, rounded
to the nearest snizu.Oshibka a presentation stored in the
register and when it exceeds unit, this unit is added to the
next in this TsAP.Iz register with the She, of course, this
method vychitaetsya.Esli implement in hardware, then under the
same switching frequency, it provides a higher frequency
whistles than the PWM (which is advantage: less whistle heard.)
disadvantage it on the ZX that cycle when programming in Z80 is
obtained is longer than the peak rate of change data for the
PWM. In PWM for the peak velocity in the port just spit out a
row 2 bytes and more are waiting until they pass a standard
time interval. If you want a level closer to the first spit
bytes, PWM, simply divide these vyplevyvaniya pause more. AND
so on. Need only a small block of routines, they are even
smaller than the variations of the fractional part of the
voltage level.
In the case of Beeper'a methods (1) and (2) are equivalent
and are being implemented particularly simply for 64 variants
of fractional level (at a frequency of 22.4kHz, close to the
standard 22100, it provides a margin of 96 strokes = 160-64
for thinking and crawling under subprogrammes).
In this case there are only four semiconductor output with
different mikropauzami modulo 4 and four semiconductors with
different pause mikropauzami on modulo 4. They contain a lot
before NOP and invoked with different addresses to adjust the
pauses. It is only 500 bytes, including transition table. Can
be implemented and more fractional levels, but then the number
of sub-bounce sharply (thinking will have to make all the
routines). Beeper a good substitute Covox and perfectly (!)
replaces the AY on MCC method of blessed memory Monster / Sage.
No sounds only on emulators (Unreal Speccy, ZXMAK, ZX Emul DOS,
ZX Emul WIN, Z80 Stealth and others probably too).
In the case of Covox the same 64-level fractional
impossible to implement. Beeper for extreme
levels were far from pure 0, and the net
1. They are difficult to use because distance
to the adjacent levels due to the finite speed Z80 would be
greater than the distance between all other adjacent levels.
For Covox net (integer) levels yet
must be used as a bridge between
the highest possible level for the fractional previous pair
(whole interleaved) and the lowest possible - for the next. A
makeshift for These higher and lower differ from the nearest
whole number of 11 processor cycles (themselves understand
why). Consequently, between adjacent fractional levels should
also be about 11 cycles of makeshift - for linearity. This is
the ideal, in practice, have settle for 5 cycles and 32 levels
(Instead of 16) when the sampling rate of 22.4
kHz, ie, the length of time interval
PWM 5 * 32 = 160 cycles. It sounds a little worse
13-bit (instead of 12), t.k.dva fractional level
fall. But on hearing 13-bit and 8 bit sound
For as the heavens and the earth. Try it. We John'om
tried.
Where in the case Covox take time out
the same thinking with the program? Done as follows:
with fractional level of <0.5 levels of spit at the port in
order junior-senior, at> 0.5 - in order of senior-junior. Then,
in any case remains poltsikla free time.
For Covox on KR572PA1 method (1) does not work - the noise
level generated more than the standard 8 bits. When I was at
the first experiment, I heard this noise, then
Think for a glitch in the program, because 8 bits PA1
You can check directly, right in
BASIC. Here's how:
FOR n = 0 TO 255: OUT 251, n: OUT 251,0: NEXT n
However, Vladimir Makarov, the question of
his interpretation of the issue said: "hang up the capacitor
between the output and the land. "Like, will help. But there
are very good reason to assume that the Makarov is wrong. The
point of current switches. They, unlike from the R-2R matrix
trigger slowly. This means that the rapid alternation of bytes
sent to the DAC 127-128-127 ... cause flipping keys in
different phases, that at certain moments may sound, even as 0
or 255. If it switches instantly and issued only two adjacent
levels, then the capacitor would not be needed. Capacitor
voltage will only lead to the middle level, and what he is? He
would have been in Between 127 and 128 (a necessary condition
for the method sounds better than 8 bits) only for a very
linear characteristic of the DAC. I do not believe so accurate
it linearity. PA1 could probably be
more effective in method (2) with alternating 0 and level, but
this method needs a large table, it is not suitable for
practical use.
Naturally, I hang capacitor vsetaki tried it, but it does
not help.
Under emulators Unreal Speccy, ZX Emul
0.34 (DOS) and Z80 Stealth PWM for Covox
improves the sound. Probably in the emulators used filters,
smoothing out Covox, however, it is possible that it does sound
karta.Drugie emulators have not been verified.
Source SHIM.H - in the appendix. To him
need 16-bit sempl.Pri hold buttons
"P" and "Enter" audio device varies
with Covox'a on Beeper or Covox-13, respectively. The default
Beeper also emulated by Covox (to make it easier to compare the
quality of sound), it is configured Tagged beeper. Port Covox
configured Tagged covox. Before playing a sample deteriorates
by 2 bits (divided by 4) to better hear the difference. But it
is also possible to disable.
A. Coder
