Previous | Next
| From: TLC (all posts)
Date: 2000-09-07
Subject: TED sound experiments
|
Hi all!
Some of you (emulator writers at the first place) will probably be interested of this thing. Past few days, I made some experiments with the Plus/4 and here are the results.
I experimented with the DC offsets one gets when setting various values to the $ff11 (volume + control) register. I also wanted to know the reason of the effect some of you know: by setting very high values to the frequency registers, some more DC levels can be generated that one can use for improving digi playback quality (as most of my music converter routines do anyway - those days, I obtained the idea from Doky, but until now I had just suspects how and why it works).
First, I wrote a routine for the Plus/4 that plays an about 2khz squarewave signal using a digi routine. It first sets osc1 frequency =
$03fd, osc2 = $03fe. After a delay, the digi routine starts and alternates $ff11 between $00 and a given value. This value increases one by every 4000 samples. I digitized the resulting wave with my PC/ GUS PnP @ 16bit 48khz.
Then, wrote a small routine in Pascal that picks out a piece from the big Wav file and measures its average AC power (with the assumption that it must be proportional to the TED's actual DC offsets). In the measuring process, each first 3/4 part of a same-level-wave section is skipped, as well as the last 1/10 part, avoiding transient problems.
Here are the results (in descending order of the measured power).
$ff11 Power Power in %
FB 15681 100
FF 15681 100
FA 15680 99
FE 15680 99
F9 15679 99
FC 15677 99
FD 15677 99
F8 15676 99
BB 15675 99
B8 15674 99
BC 15673 99
BF 15673 99
BA 15671 99
BE 15670 99
B9 15668 99
BD 15668 99
F7 13621 86
B7 13614 86
F6 11341 72
B6 11333 72
7B 11317 72
7F 11316 72
7C 11315 72
7E 11315 72
78 11313 72
7A 11313 72
7D 11313 72
79 11312 72
3E 11309 72
3A 11307 72
3B 11307 72
3F 11307 72
3D 11306 72
38 11305 72
39 11305 72
3C 11303 72
77 9912 63
37 9905 63
F5 9176 58
B5 9170 58
76 8309 52
36 8306 52
D8 7339 46
DD 7339 46
D9 7338 46
DC 7338 46
DE 7338 46
9B 7337 46
DA 7337 46
DB 7337 46
9A 7336 46
9C 7336 46
9E 7336 46
9F 7336 46
DF 7336 46
99 7335 46
98 7334 46
9D 7334 46
E9 7319 46
ED 7319 46
EC 7318 46
EE 7318 46
EF 7318 46
E8 7317 46
EA 7317 46
EB 7316 46
AB 7315 46
AE 7315 46
AF 7315 46
A8 7314 46
AC 7314 46
68 7313 46
6C 7313 46
6F 7313 46
AA 7313 46
AD 7313 46
6B 7312 46
A9 7312 46
6E 7311 46
69 7310 46
6A 7310 46
6D 7310 46
28 7309 46
2F 7308 46
2B 7308 46
2A 7307 46
2C 7307 46
29 7307 46
2D 7306 46
2E 7306 46
F4 7076 45
B4 7075 45
75 6757 43
35 6753 43
D7 6467 41
97 6464 41
E7 6444 41
A7 6443 41
67 6436 41
27 6430 41
D6 5451 34
96 5450 34
E6 5432 34
A6 5429 34
66 5422 34
26 5418 34
74 5229 33
34 5226 33
F3 5025 32
B3 5023 32
D5 4457 28
95 4454 28
E5 4438 28
A5 4434 28
65 4428 28
25 4425 28
73 3732 23
33 3729 23
5B 3598 22
58 3598 22
5F 3597 22
5C 3597 22
5D 3597 22
5E 3597 22
59 3597 22
1A 3596 22
18 3596 22
1D 3596 22
5A 3596 22
1B 3595 22
19 3595 22
1F 3595 22
1E 3595 22
1C 3595 22
D4 3465 22
94 3462 22
E4 3445 21
A4 3443 21
64 3435 21
24 3433 21
57 3179 20
17 3176 20
F2 3061 19
B2 3058 19
56 2689 17
16 2687 17
93 2485 15
D3 2485 15
A3 2465 15
E3 2465 15
63 2463 15
23 2461 15
72 2278 14
32 2276 14
55 2205 14
15 2204 14
54 1719 10
14 1717 10
D2 1528 9
92 1528 9
A2 1508 9
E2 1508 9
62 1506 9
22 1505 9
53 1237 7
13 1236 7
F1 1103 7
B1 1102 7
71 821 5
31 820 5
52 762 4
12 761 4
D1 562 3
91 562 3
61 542 3
E1 542 3
A1 542 3
21 541 3
51 281 1
11 280 1
4A 31 0
45 31 0
09 31 0
08 31 0
0A 31 0
0B 31 0
C1 31 0
C2 31 0
C3 31 0
C4 31 0
C5 31 0
C6 31 0
C7 31 0
C8 31 0
C9 31 0
CD 31 0
CE 31 0
CF 31 0
D0 31 0
46 31 0
4C 31 0
82 31 0
86 31 0
90 31 0
30 31 0
81 31 0
4B 31 0
E0 31 0
B0 31 0
49 31 0
60 31 0
40 31 0
41 31 0
20 31 0
A0 31 0
48 31 0
80 31 0
10 31 0
07 31 0
04 31 0
05 31 0
06 31 0
0F 30 0
8C 30 0
0D 30 0
44 30 0
0E 30 0
C0 30 0
88 30 0
70 30 0
47 30 0
0C 30 0
83 30 0
89 30 0
8F 30 0
CA 30 0
8D 30 0
84 30 0
4E 30 0
85 30 0
F0 30 0
8A 30 0
8E 30 0
87 30 0
50 30 0
4F 30 0
42 30 0
CB 30 0
CC 30 0
4D 30 0
02 30 0
00 30 0
8B 30 0
43 30 0
03 30 0
01 30 0
Some thoughts that I can derive, after checking this (and after seeing the wave in Sound Forge)
- The output is asymmetric (in DC). I'd say it's positive. Well, not a surprise after all...
- The TED's sound output is quite noisy. The noise seems to be more noticeable if higher levels are generated (this is really strange: I only noticed it if the level was changing dynamically (for example,
alternating between $00 and $b8); a constant high level ($b8), when measured, did not show more noise than a constant 0). The 50Hz component is clearly noticeable - would suspect that it comes from the video part of the TED, since there are plenty of other possibly video-originate noise in the signal too.
- Setting 0 volume ($x0) or turning no oscillator on ($0x) all generates the same silence (no noticeable difference).
- Volume values above 8 give the same DC offset as volume 8 (the power values of the higher $ff11 settings tend to have higher power in the list, but there is very small difference (not even consequent) and I'd rather consider it a side-effect of my measuring method (AC analysis, on a wave that is clearly asymmetric and one should wait 'enough' until all transients go down before measuring) or the noise that I didn't want to filter out.
- This is probably more interesting: writing $03fe to the frequency register locks up the oscillator completely. The oscillator then gives the same DC offsets as you get in constant DC level mode (bit7 of $ff11 =
1), e.g. if frq(osc2) = $03fe, writing $24 or $94 or $a4 into $ff11 all give the same DC offset.
- Setting $03fe as frequency in osc2 also locks up the noise generator.
Depending on something that I can't derive, setting $4x values to $ff11
either gives the same constant DC offset as above ($9x or $Ax) or constant zero. Should depend on the minute the oscillator value is written, as I suspect.
- Frequency of $03fd and the like work the following way: try it,
$03fb - $03fd give the same DC offset if the oscillator is on. This is because these don't stop the oscillator, they're just very high frequencies that one doesn't hear. On a PAL machine, $03fb lets the oscillator generate an about 22khz signal, that anyway my GUS PnP just filtered out completely. $03fd is about 55khz - I don't know if it can even reach the audio output. Since this is a 50-50% pulse wave, its average power should be half of the same volume constant DC level. (As one could have suspected that without this explanation).
The levels, if going through the volume settings as 0,1,2...8 seem to be slightly non-linear. The first step (0-1) is about half the second step
(1-2), whichever oscillators are active. Check the power difference through the steps of $b0...$b8. The steps slightly rise, then the last step is again smaller. ...I don't know the reason.
Then, what to use all this for?
If I were an emulator programmer, I could improve sound emulation.
- Checking one of the highest series of osc/volume setting ($b0...$b8,
for example) could give a fair approximation of the TED volume steps,
related to the TED volume setting.
A proper emulation of the TED behaviour, including digi tricks could be then the following:
- Each oscillators could be emulated as simple (incrementing counter type) square generators. The output level is either 0 or 100%. If the oscillator is set off, assume 0. If on, then... The correct formula for the output is 'f=Phi(single clock)/8/(1024-((frqvalue+1)&1023))'. If
'frqvalue' is between $03fb and $03fd, assume constant 50% output. If
'frqvalue'=$03fe, or 'constant level mode' is selected (bit 8 of $ff11
is set), assume 100% output.
- (Something should be done in order to emulate noise correctly; if someone wants, I sample the lowest frq TED noise wave with my PC - I think it could be easy to play this back at the speed derived from the above oscillator formula.)
- Multiple each current output level by the current volume (in %).
This current volume is given by the volume step table, indexed with the TED volume.
- Add the two resulting values together and store to the next position of the output wave (to the sound card). See next one...
(Expect some aliasing problems at least at the higher frequencies:
generating square waveforms of some 10-20khz at the sampling rate of
44khz or even 48khz implies lots of undersampling problems. The
'theoretical' squarewave one tries to play contains plenty of strong overtones that appear as aliasing (if one knows how a SID music converter routine on the Plus/4 sounds, he for sure knows the typical sound that I mean). SID emulators on PC used to solve this by oversampling in software, applying a low-pass filter algorithm to the high-sampled stream and playing this through the sound output (this definitely improves sound quality). But this is a different problem).
If I were experimenting with digi playback on a real machine...
Well I'd probably try to get a good sounding digi $ff11 table, by optimizing a series that give the closest output levels to the linear characteristic (split the '100%' range into the needed number of pieces,
and select the nearest DC offset for each needed value - better do that using the absolute results, not percents since it's more accurate, but the idea is the same). (I sure did it ;-) Well not much improvements to the previous digi tables - if anyone wants, I can give you my results).
BTW, Richard Atkinson also promised to make some measurements - expect that sometime on the list. I'd say these should be more accurate.
With regards,
L.
Ps. Some of you could also be curious to know how the TED actually generates the output. I thought I better include Richard's original post here, that he sent to the cbm-hackers mailing list about one month ago.
...As a comment: it's very interesting how Commodore avoided completely any analog design in this sound generator (unlike the SID design is).
The TED is fully digital. ...Well, it makes sense, after all, not putting some analog circuits to the closest neigborhood of a video generator circuit ;-)
>
The waveform coming out of the SND pin is digital, pulse-width modulated. I think the width is expressed in master clock cycles (14.3 or 17.7 MHz) and the period of the cycle depends on whether both channels are on or just one. The volume bits set the length of the pulse and if only one channel is on the period is doubled so that the DC voltage level is halved. I have yet to make a quantitative table of the mark-space ratios with different register settings but that's definitely the next thing to do.
>
|
| |
Copyright © Plus/4 World Team, 2001-2024 |
