More than 4 sound channels - how it works?

[demolition]

Read here: https://en.wikipedia.org/wiki/Original_Chip_Set#Audio.

Paula has 4 independent 8 bit DACs where each DAC has a 6 bit analog volume control which can also be modulated. When the 8 bit in the DAC is combined with 6 bit from the volume control, you get something like 14 bit depending on how quickly the volume control can be modulated.

I would also reckon that you could combine the two channels on each side so instead of 4x14 bit, you could get something close to 2x16 bit.

[Samurai_Crow]

For playing OctaMed Soundstudio sound you can see the source from Aminet though the rights to use it directly rather than in the original .library form may be dubious now that the rights to OMed SS have been bought out.

Also, the way that 14-bit sounds work is by playing the low 6 bits through a second voice at a volume of 1. Stereo is accomplished by doing this through the other two voices. This means that you can play one voice of 14-bit stereo sound without CPU mixing at all. Anything beyond this is either CPU or added hardware.

[Thorham]

Originally Posted by Samurai_Crow

the low 6 bits through a second voice at a volume of 1.

No, it's the six upper bits of the low byte. Bits 15-8 are played at volume 63, bits 7-2 are played at volume 1, and bits 1-0 are discarded.

[Samurai_Crow]

Originally Posted by Thorham

No, it's the six upper bits of the low byte. Bits 15-8 are played at volume 63, bits 7-2 are played at volume 1, and bits 1-0 are discarded.

I was talking about a 14-bit sample but if I had been talking about a 16-bit sample you'd have been right with your correction.

[meynaf]

Originally Posted by kovacm

I must ask something that is not related strictly to topic but 14bit sound on Amiga is achieved by combining two 8 bit channels but Amiga still have 8bit DAC at output? right? So internally (in memory and CPU) you work with 14bit but at output you have 8bit OR Amiga have 4 separate 8bit DAC connected to stereo output chinches?

Amiga really has two 8bit DAC and not more, yet it can still output 14bit quality. Sounds strange, huh ?

14bit is done by using two channels, one at maximum volume (64) and the other at minimum nonzero volume (1).
Internally (in memory and CPU) it's actually 16bit that's used.

The Amiga DAC do the volume by being muted (turned off) during (64-n) clock ticks out of 64. Then the filtering hardware removes the high frequencies and you get proper volume (i hope my explanation was clear enough ?).

[meynaf]

Originally Posted by demolition

Paula has 4 independent 8 bit DACs where each DAC has a 6 bit analog volume control which can also be modulated.

Paula has only 2 8bit DACs, not 4. Mixing is done by alternating data of each channel at high frequency.

Originally Posted by demolition

I would also reckon that you could combine the two channels on each side so instead of 4x14 bit, you could get something close to 2x16 bit.

Nope. Doesn't work this way ; 14bit already takes 2 channels to be done.

Originally Posted by Thorham

No, it's the six upper bits of the low byte. Bits 15-8 are played at volume 63, bits 7-2 are played at volume 1, and bits 1-0 are discarded.

Volume 64, not 63

[Thorham]

Originally Posted by Samurai_Crow

I was talking about a 14-bit sample but if I had been talking about a 16-bit sample you'd have been right with your correction.

Yeah, I somehow didn't get that

Originally Posted by meynaf

Volume 64, not 63

Indeed

                          Agnus/
                  Read/   Denise/
Register Address  Write   Paula         Function
-------- -------  -----   -------       --------
AUDxVOL    0A8      W       P      Audio channel x volume

                 This register contains the volume setting for
                 audio channel x.  Bits 6,5,4,3,2,1,0 specify 65
                 linear volume levels as shown below.

                 Bit#    Use
                 ----    --------------------------
                  15-07  Not used
                  06     Forces volume to max (64 ones, no zeros)
                  05-00  Sets one of 64 levels (000000=no output
                         (111111=63 1s, one 0)

Shows what I know

[Trachu]

Originally Posted by Don_Adan

First Amiga 7V soundformat was Jochen Hippel 7V, created in 1990 year.
Second Amiga 7V soundformat was TFMX 7V.
Third Amiga 7V soundformat was Mugician II (aka Mugician 7V).
All Amiga 7V soundformats has 3 normal channels and 4 mixed channels.
Many years ago I asked Chris Huelsbeck's about TFMX 7V details, and he wrote
that Jochen Hippel was author (idea and mixing code) of 7V routine (exactly
mixing 4 channels in 1 channel).
Many years ago I asked also Reinier van Vliet (Mugician author), he wrote
that he created Mugician II replay, after Turrican 2 was published.
Turrican 2 TFMX 7V music was inspiration for him.
All Amiga 7V replays with sources code are available on the Wanted Team page.

Well thanx for this info and i have found more information regarding this type of mixing.
Jochen Hippel original routine uses 268 cycles per mixed byte. Optimised code uses 254 (or 246 in non osfriendly way)

Digital Mugician 2 original playing routine uses 272cycles per mixed byte, but optimised code only 211 (or 203 in non os friendly way)!!!
You can find their assembly code here
http://wt.exotica.org.uk/players.html

Overally we have to have more mixed bytes per frame to obtain decent quality, so this limits the possible usefulness of this technology.

On C64 we could see a progres in music quality all the time, but Amiga go stuck in Protracker era. Protracker is great, but maybe someone could find a way to enhance number of channels with very limited cost of CPU.

[Don_Adan]

Originally Posted by Trachu

Well thanx for this info and i have found more information regarding this type of mixing.
Jochen Hippel original routine uses 268 cycles per mixed byte. Optimised code uses 254 (or 246 in non osfriendly way)

Digital Mugician 2 original playing routine uses 272cycles per mixed byte, but optimised code only 211 (or 203 in non os friendly way)!!!
You can find their assembly code here
http://wt.exotica.org.uk/players.html

Overally we have to have more mixed bytes per frame to obtain decent quality, so this limits the possible usefulness of this technology.

On C64 we could see a progres in music quality all the time, but Amiga go stuck in Protracker era. Protracker is great, but maybe someone could find a way to enhance number of channels with very limited cost of CPU.

Thanks, but this is my group page.

[ReadOnlyCat]

Originally Posted by Trachu

Well thanx for this info and i have found more information regarding this type of mixing.
Jochen Hippel original routine uses 268 cycles per mixed byte. Optimised code uses 254 (or 246 in non osfriendly way)

Digital Mugician 2 original playing routine uses 272cycles per mixed byte, but optimised code only 211 (or 203 in non os friendly way)!!!
You can find their assembly code here
http://wt.exotica.org.uk/players.html

Overally we have to have more mixed bytes per frame to obtain decent quality, so this limits the possible usefulness of this technology.

I am a bit surprised by these high numbers. I guess they must mean that these routines are mixing voices with different volumes and frequencies.

This seems like a surprising choice given that the instruments usually selected for mixing are usually percussions and the likes which have mostly constant volume and frequencies. It seems sensible to constrain the trackers to use constant frequencies for mixed voices which should greatly reduce the complexity and time needed for each byte: for n voices, only n adds + one divide (simple table).

Is there such a thing on the market?

[4mat_]

That's a terrible idea, even with percussion you want a bit of pitch control. Apart from the obvious things like toms and cymbals getting a bit of variance in your drums needs pitch effects.

Even using this to play long sample loops wouldn't work because they all need individual tuning to the master bpm.

[pandy71]

Originally Posted by meynaf

Paula has only 2 8bit DACs, not 4. Mixing is done by alternating data of each channel at high frequency.

Prove it... Based on all Paula descriptions it is quite clear that Volume regulation in Amiga is performed by PWM - as such independent DAC for each channel is almost mandatory... (to simplify design but it is possible with additional muxer switched at 2*CCK but this mean that 8 bit DAC must be capable to convert over 7 Mega samples per second)

If Paula perform sample interleaving then it is possible to increase sample rate by using 2 DAC's and without reprogramming horizontal scan rate, only by proper sample interleaving in memory (accordingly to Your claim - odd samples in first channel, even on second).

I can imagine that cost of silicone may justify to have 4 DAC's (multiplier 8*6 even single, working in sample interleaving mode, probably will occupy maybe even more area on die than whole CPU).

[Mrs Beanbag]

Originally Posted by 4mat_

That's a terrible idea, even with percussion you want a bit of pitch control. Apart from the obvious things like toms and cymbals getting a bit of variance in your drums needs pitch effects.

Even using this to play long sample loops wouldn't work because they all need individual tuning to the master bpm.

hm i dunno, most of my music, the only percussion instrument i use at different pitches is the tom, and only then in 3 different pitches, which i could resample beforehand if i wanted.

But i'd have thought one of the most popular uses of more channels would be chords.

[pandy71]

Originally Posted by Mrs Beanbag

hm i dunno, most of my music, the only percussion instrument i use at different pitches is the tom, and only then in 3 different pitches, which i could resample beforehand if i wanted.

But i'd have thought one of the most popular uses of more channels would be chords.

PC audio world is sampling fixed so one sampling rate is not a problem - similar approach can be applied on Amiga - some PC audio hw support only one native sampling frequency... Probably only GUS was somehow comparable to Amiga audio HW (similar principle - variable sampling rate).

[meynaf]

Originally Posted by pandy71

Prove it... Based on all Paula descriptions it is quite clear that Volume regulation in Amiga is performed by PWM - as such independent DAC for each channel is almost mandatory... (to simplify design but it is possible with additional muxer switched at 2*CCK but this mean that 8 bit DAC must be capable to convert over 7 Mega samples per second)

The PWM frequency is about 56Khz, not 3.5Mhz.

Originally Posted by pandy71

If Paula perform sample interleaving then it is possible to increase sample rate by using 2 DAC's and without reprogramming horizontal scan rate, only by proper sample interleaving in memory (accordingly to Your claim - odd samples in first channel, even on second).

This is not possible, because you'd have to make the two channels in perfect phase opposition (half-a-sample sync) - something the DMA simply can't do.

Originally Posted by Mrs Beanbag

But i'd have thought one of the most popular uses of more channels would be chords.

I'm not a musician but from what i've heard, drums don't use many channels in comparison to tones. So the fixed pitch mixing wouldn't be useful. A fixed per-channel volume can bring a few cpu cycles but i don't think it's worth.

[pandy71]

Originally Posted by meynaf

The PWM frequency is about 56Khz, not 3.5Mhz.

56kHz * 64=3.58MHz - DAC must be updated with this frequency (unless DAC_Vref is controlled by PWM but then DAC must be multiplying type which itself affecting significantly DAC bandwith).

My point is - until someone not map Paula silicone or Paula schematics will be available we can only speculate how many DAC's is on Paula.
Block schematics which are available suggest 4 independent DAC's and all channels are combined in analog domain.

Originally Posted by meynaf

This is not possible, because you'd have to make the two channels in perfect phase opposition (half-a-sample sync) - something the DMA simply can't do.

Well you suggested sample interleaving (don't get me wrong - this is OK from design perspective and probably i would design Paula in such way to reduce silicone size) - this half sample time offset is not related to DMA but to internal DAC buffer + muxer + ZOH + PWM - relatively easy test.
Stacking to 8 bit DAC's in parallel you should be able to receive 9 bit resolution (if there is no sample interleaving), if there is sample interleaving you should be able double sample rate by using 2 channels where one channel play odd samples and second even samples (maybe with 1 byte offset).

If some one is able to accept some limitations (2 virtual channels must have same sampling rate + sample rate reduced by half i.e. relatively lofi sound) then 8 channels can be created by interleaving odd samples and interleaving even samples in one channel buffer.

[meynaf]

Originally Posted by pandy71

56kHz * 64=3.58MHz - DAC must be updated with this frequency (unless DAC_Vref is controlled by PWM but then DAC must be multiplying type which itself affecting significantly DAC bandwith).

My point is - until someone not map Paula silicone or Paula schematics will be available we can only speculate how many DAC's is on Paula.
Block schematics which are available suggest 4 independent DAC's and all channels are combined in analog domain.

Maybe it has 4x8bit DAC after all. But if you can go with just 2x8bit, why not just doing it this way ?
If DACs occupy a large die size, using 2 instead of 4 makes sense.

If i recall correctly - which is dubious - i've read somewhere that the mixing was done this way - which might have been wrong as well.

Originally Posted by pandy71

Well you suggested sample interleaving (don't get me wrong - this is OK from design perspective and probably i would design Paula in such way to reduce silicone size) - this half sample time offset is not related to DMA but to internal DAC buffer + muxer + ZOH + PWM - relatively easy test.
Stacking to 8 bit DAC's in parallel you should be able to receive 9 bit resolution (if there is no sample interleaving), if there is sample interleaving you should be able double sample rate by using 2 channels where one channel play odd samples and second even samples (maybe with 1 byte offset).

If you can have two channels 180 degrees of each other, then you can have double sample rate regardless of the mixing method (because your output level will get updated every half a sample).

Originally Posted by pandy71

If some one is able to accept some limitations (2 virtual channels must have same sampling rate + sample rate reduced by half i.e. relatively lofi sound) then 8 channels can be created by interleaving odd samples and interleaving even samples in one channel buffer.

For 2 channels out of one, with the same replay freq, you'll be better off by simply averaging the two samples with the cpu.

Anyway, for 8 channels or more, i'll always prefer using the cpu. And even an unexpanded A1200 can play many channels with acceptable quality ; for example the menu music in superstardust is a 12-channel S3M.

[pandy71]

Originally Posted by meynaf

Maybe it has 4x8bit DAC after all. But if you can go with just 2x8bit, why not just doing it this way ?
If DACs occupy a large die size, using 2 instead of 4 makes sense.

If i recall correctly - which is dubious - i've read somewhere that the mixing was done this way - which might have been wrong as well.

I agree that 2 seem to be cheaper but... it imply to have fast and accurate DAC, for single channel DAC is updated with 3.58MHz speed so for 2 channels it must be updated twice faster. If volume regulation is performed not by selecting sample value vs zero but as a DAC Vref change this means that DAC must be multiplying type and such DAC's are very limited in terms of bandwidth - perhaps few hundreds kHz max.
So from one side silicone designer is focus on reducing die size but on second he is limited by technology available and expected parameters.

Seem that on picture: two large areas are very well visible - one area consist 4 blocks (audio digital part?, audio digital + analog?) second are consist 2 large block utilising roughly 40% of die (flopy + other logic?, audio DAC's?)

Based on this picture still nothing is clear...
And you can imagine even different approach - analog switches controlled by PWM where they switch between DAC output and analog ground - as such PWM is applied in analog domain...
Nothing is clear and after longer picture observation i almost convinced that audio part occupy approx 60% and that there are 4 DAC's and that second large block is probably chipset bus with address decoder + other bus logic.

Picture from http://obligement.free.fr/articles/p...microscope.php but real source is here http://www.tinytransistors.net/index...d=31&Itemid=32

Originally Posted by meynaf

If you can have two channels 180 degrees of each other, then you can have double sample rate regardless of the mixing method (because your output level will get updated every half a sample).

Not necessary... imagine that samples from 2 channels are interleaved 1.2.1.2...1.2.1.2 (case for no volume applied i.e. level 65) - if this is performed with 2*3.58 then single DAC can be shared, when level is applied samples from each channel are no longer interleaved in continuous way but they can be replaced by 0 - after averaging analog signal will be restored with level proportional to ratio between time of sample/time of zeros (proportional to level) - this can be easily made with some muxer that switch between sample buffer and 0, another muxer selecting sample from 2 channels (muxing twice faster) and it should work.

Originally Posted by meynaf

For 2 channels out of one, with the same replay freq, you'll be better off by simply averaging the two samples with the cpu.

Anyway, for 8 channels or more, i'll always prefer using the cpu. And even an unexpanded A1200 can play many channels with acceptable quality ; for example the menu music in superstardust is a 12-channel S3M.

This means every sample must be multiplied by volume so 8x6 gives us 14 bit, then two 14 bit samples must be added (15 bit product) and shifted down by 7 bits (or only 8 MSB selected). Plain 68k is relatively slow in DSP...

[Toni Wilen]

US Patent 4,829,473 says ""Each of the data registers 29,31 transfers its data into a plurality of audio control counters and registers 33 and 35, which in turn each drive an individual one of the digital to analog (D/A) converters 37,39 of which there are four in all." and FIG 4 shows 4 identical channel blocks and each has single analog out line.

[meynaf]

Originally Posted by pandy71

This means every sample must be multiplied by volume so 8x6 gives us 14 bit, then two 14 bit samples must be added (15 bit product) and shifted down by 7 bits (or only 8 MSB selected). Plain 68k is relatively slow in DSP...

Multiplication by volume is done with a LUT so no slow multiply is involved.
A good routine can do 4ch on bare 68000 and 32ch with interpolation on 68030 - not what i'd call slow.