Amiga Sound Quality

[Gorf]

Originally Posted by pandy71

Nope, sound and disk are linked to horizontal frequency and as fundamental principle for Amiga...

yes I know that very well!
I meant horizontal of course
(don't know why I mix those two up all the time ...)

DMAC eat CPU cycles - all bus activity performed by DMAC use CPU time and DMAC can't use chipset timing slots.

Sure - but serial and parallel do so now anyways as the CPU has to manage it - as does any hard disk no matter via DMA or PIO - audio would not be that much obviously.
On the other hand you would free Agnus-DMA-stots for 8 more sprites... and you could use FastRAM for audio-data

Yes and no... twice faster bus should solve issues with audio but this is not possible due RAM access time when Amiga architecture was designed

I did not say anything about a faster bus.
I am very well aware of the RAM roundtrip times back then

Faster and wider RAM would have been helpful in later Amiga models and would have made transition to HD-floppy-drives and more audio-channels easier, that would have been a nice side effect from a DMAC-Paula-combination.

But I am talking here about the original design from 85 - even there it would have made more sense to separate the video-dma from audio and IO

[Thomas Richter]

Originally Posted by Gorf

But I am talking here about the original design from 85 - even there it would have made more sense to separate the video-dma from audio and IO

It would, but it would have required another DMA controller (despite Agnus, which is the DMA controller now), or a more flexible slot assignment for Angus. I afraid the bus timing with Agnus was already at the edge of possibilities back then.



The goal with the fixed slots was probably to keep the design as simple as possible and to avoid any complex arbitration logic besides that for the blitter.

Paula should have really received an upgade for AGA. Double bandwidth was available, so HD floppy support would have been possible, and 16-bit sound or 44kHz sound (CD-quality) would have been possible.



The logic is pretty similar to that of the ANTIC, the look-alike chip of Agnus in the 8-bit Ataris. It also had a fixed DMA raster for multiple DMA sources. "Bitplane" aka "playfield DMA", "sprite" aka "Player/Missle DMA", "Display list DMA" (related to the copper), and memory refresh, all at fixed raster cycles.



Thus, in a sense, Agnus is an upgraded "super-ANTIC", which now also did floppy and sound DMA.

[Gorf]

Originally Posted by Thomas Richter

It would, but it would have required another DMA controller (despite Agnus, which is the DMA controller now)

Well that is what I said here all along and why I mentioned the CBM 8727 DMA or the DAMC in the A590 - that should have been combined with Paula and CIAs

[Bruce Abbott]

Originally Posted by pandy71

there is no such thing as free lunch - few small improvements will enrich Paula without altering anything on system bus - anything above - better use dedicated audio card with local RAM (a least FIFO) and perhaps neat DSP to offload CPU.

In reality yes, but we are not talking reality here.

Originally Posted by Thomas Richter

Paula should have really received an upgade for AGA. Double bandwidth was available, so HD floppy support would have been possible, and 16-bit sound or 44kHz sound (CD-quality) would have been possible.

Originally Posted by Gorf

Well that is what I said here all along and why I mentioned the CBM 8727 DMA or the DAMC in the A590 - that should have been combined with Paula and CIAs

Ah yes, so many things Commodore 'should' have done. Well I for one am glad they didn't.

Commodore tried to make a chipset with all the things being talked about, but failed. It was just too much change at once with too many things to go wrong. So they reset and went for a safer incremental change that was more reliable. If they had done this in the first place instead of setting their sights too high, we would have had AGA sooner and Commodore could have made a lot more money. Then they might have had enough behind them to produce a real next generation machine.

Paula sound is great if you work within its limitations and don't try to make it something it isn't. Sometimes that means accepting that you can't have everything you want. But if Paula had 8 channels of 44kHz stereo sound, would it be that much better? I listen to PC music and most of it sounds like it was composed on an Adlib card - no artistic expression because the media doesn't encourage it.

There has always been a tradition in the Amiga community of saying 'look how wonderful our machine is', then in the next breath 'but it should have been so much more' - followed by wish lists not grounded in reality. Familiarity breeds contempt?

[no9]

Originally Posted by Bruce Abbott

But if Paula had 8 channels of 44kHz stereo sound, would it be that much better?

Yes, it would. You could finally stop nagging players with this one 8-bitish question MUSIC OR SFX? and cutting sound voices so drastically like most Amiga shooters did.

[daxb]

If you would upgrade Paula audio and hear the demands then you would go 16 bit, 44kHz and at least 16 channels with free panning. That is what musicians needed as minimum in the '90 and game developers would also like. Later they would demand VST. With powerful enough hardware this would fixed the audio part at that time. Instead there was no upgrade.

[no9]

While we can't please demanding professional audio consumers we still can address certain problems known for years from 8-bit computers, and later continued with A500. Even 6 channels Paula still would be improvement for games. With 3 channels you can create reasonable game soundtrack without resorting to some dubious tricks and you still have the other 3 left to address events in a game. I would even prefer more 4-bit sound channels for sound effects instead less but 8-bit ones. And yet we were left with games like Lotus where is no music during game at all or you can have music from the radio but without sound effects like in Lotus 3.

I'm not an expert but Paula's 4 hard-panned voices suggest that games wasn't taken into account during design. It was rather some video purposes where you can stream 2 stereo tracks and x-fade between them.

[roondar]

Well, I suppose the Amiga designers could have dedicated more slots to audio than they did. That would have allowed more channels or more bits per channel. With the standard Amiga setup, these extra slots are actually available in the horizontal blanking interval (though not at the start but rather at the end of the rasterline).

Of course, nothing is free so assigning those slots would've meant less maximum overscan and slightly more DMA contention.

[Thomas Richter]

Originally Posted by Bruce Abbott

Commodore tried to make a chipset with all the things being talked about, but failed.

But not because the ideas were bad, but the timing was. Too little, too late.

Originally Posted by Bruce Abbott

It was just too much change at once with too many things to go wrong.

That's why you start such projects early. Unlike CBM. At the time the chipset was designed, it was ahead of its time. CBM waited until it was behind its time. At this time, it was too late.

Originally Posted by Bruce Abbott

So they reset and went for a safer incremental change that was more reliable.

Incremental changes are fine, really, but you have to make them often, and soon enough. There wasn't really much of an incremental change until AGA. There was only the chip ram increase from 512K to 2MB, and some incremental change to Denise, but Chip RAM bandwidth remained so limited that the new possibilities like SHIRES were practically not usable.

Originally Posted by Bruce Abbott

But if Paula had 8 channels of 44kHz stereo sound, would it be that much better?

Would it be a game changer? No. But was AGA a game changer? Neither. The problem was not that the change wasn't right, it was just too late. If, instead of ECS, it would have been AGA at the time ECS became available - that would have been right.

[Thomas Richter]

Originally Posted by no9

I'm not an expert but Paula's 4 hard-panned voices suggest that games wasn't taken into account during design. It was rather some video purposes where you can stream 2 stereo tracks and x-fade between them.

My guess is that there was no particular reason for four voices except that the predicessor of PAULA, POKEY namely, had four, and it worked well enough for its users, so there was no need to change. Atari had 4 players, 4 missles, 4 voices, and 4 joystick ports (at least the original machine), so you see where this was targetted at: at most 4 people playing together.


Now, 4 sprites would have been too small (and turned out to be too small on the 8-bits), so that was changed. But the 4 voices worked well enough.

[no9]

Yeah, this might be the case. And also in the 1984-85 they could not consider music as immanent part of games so there was no pressure on that. C64 games of 1984 had very limited music
[ Show youtube player ]


But in the 1990 it was hard to consider a game release as complete without soundtrack so there was no excuse for not extending Paula voices, even if they had to stay 8-bit.

Amiga games were hitting hard 4 channels ceiling while being far below it's sample frequency limit. The latter was limited by disk space and chip memory in the first place, but not by the Paula yet.

[pandy71]

In theory there is simple workaround for mentioned bus timing limitation - using invisible video lines time to transfer (in bursts / bulk) Audio data (and store it internally in Paula in some kind of FIFO) - it was doable even in Amiga beginning - of course at a cost of CPU time but still HW wise (for example this feature can be available side to regular Audio cycles as for example additional Audio voices independent from already present 4 channels) - 8 - 16 lines would give 48000Hz sample rate. This could be still added to Amiga by using video lines (albeit less efficient than using data bus) - in theory with 16:9 video Amiga have plenty of bandwidth to create 48kHz 16 bit stereo audio even today. Another lost opportunity that could be nicely integrated without significant change to primary architecture.

[commodorejohn]

I mean, you can say anything was possible in retrospect if you don't mind completely reinventing the broader cultural/technological context of the time as compared to, y'know, actual history. 16:9 video? Sure, why not. Definitely suited to the 12"-15" 4:3 CRTs people were actually using in 1990. That's why pan-'n-scan never caught on and letterboxing was never necessary for home video releases targeted at cinephiles, right?

[chb]

Originally Posted by pandy71

In theory there is simple workaround for mentioned bus timing limitation - using invisible video lines time to transfer (in bursts / bulk) Audio data (and store it internally in Paula in some kind of FIFO) - it was doable even in Amiga beginning - of course at a cost of CPU time but still HW wise (for example this feature can be available side to regular Audio cycles as for example additional Audio voices independent from already present 4 channels) - 8 - 16 lines would give 48000Hz sample rate. This could be still added to Amiga by using video lines (albeit less efficient than using data bus) - in theory with 16:9 video Amiga have plenty of bandwidth to create 48kHz 16 bit stereo audio even today. Another lost opportunity that could be nicely integrated without significant change to primary architecture.

I think having 8-16 lines or about 4kB of buffers in Paula would be absolutely outside the possibilities of the time. That's about 200k transistors just for the buffers. An additional bus for Paula and some (S)RAM, similar to the VIC II's color ram, would have been doable probably, but still expensive.
The idea with using the video lines is interesting for some add-on hardware with today's technology. If I understand correctly, a kind of Graffiti expansion that gets its data from the digital video pins during pseudo-vblank (display data is still being sent but the add-on blanks the screen and fills the audio buffer)? Could be nice, especially if combined with some kind of Graffiti-like c2p hardware and maybe some DSP for sample compression + playback engine. Latency could be a bit high, bbut sufficient for music replay and game sound effects.

But ofc probably almost no one would use it, as no one used the Graffiti.

[Bruce Abbott]

Originally Posted by Thomas Richter

But not because the ideas were bad, but the timing was. Too little, too late.

In practice the ideas were bad, because they couldn't implement them in a timely manner. That's why successful designers don't try to do too much at once. You see this in most electronic devices, in automobiles etc., incremental changes that can be backed out of if it doesn't work or the market doesn't respond well.

Mind you, Commodore weren't the only ones who wasted a lot of time and money on failed projects. It's just that Intel and IBM (to name two) had enough resources to burn that they could get away with it.

That's why you start such projects early. Unlike CBM. At the time the chipset was designed, it was ahead of its time. CBM waited until it was behind its time. At this time, it was too late.

Commodore had enough trouble just trying to get the original Amiga out, let alone making another quantum leap. But I'm pretty sure they could done AGA 'before it was too late' (since it didn't take them long after failing to get AAA working) if they had started with that goal.

Incremental changes are fine, really, but you have to make them often, and soon enough. There wasn't really much of an incremental change until AGA. There was only the chip ram increase from 512K to 2MB, and some incremental change to Denise, but Chip RAM bandwidth remained so limited that the new possibilities like SHIRES were practically not usable.

We could say the same in the PC world too. After IBM released VGA, many manufacturers brought out their own chipsets and continued using them with virtually no changes for many years. Take Trident for example:-

About Those Trident VGAs…

With these improved numbers, the TVGA8900B would still do very poorly compared to the other cards, and especially the DOOM figures are abysmal. Can anyone do worse?

Why yes. This card:

it’s a newer card (very early 1993) with a newer chip (TVGA9000B) but only half the video memory (512 KB). The performance is more or less identical to the 8900B in its initial configuration, only DOOM is slightly slower.

Seeing how awful the performance of these cards is compared to the good ones (1.3 vs. 5.5 MB/s bandwidth), one might suppose that the Tridents were by far the slowest ISA VGA cards on the market in the 1990s. Surprisingly enough, that’s not the case. More about the real slowpokes next time.

Yet they sold bucket-loads.

Crappy cards, sold dirt cheap - and customers were none the wiser. In the 10 years that I sold PCs in my shop (1991 to 2000), very few customers cared about the actual performance of the VGA card, or the sound card, or even the CPU (386SX? That must be better than a 386DX, right?).

Would it be a game changer? No. But was AGA a game changer? Neither. The problem was not that the change wasn't right, it was just too late. If, instead of ECS, it would have been AGA at the time ECS became available - that would have been right.

I'm not sure they could have managed AGA that soon, but certainly any time earlier would have helped. AGA in the A3000 would have made me very happy!

But all this is just idle speculation. The fact is that in reality none of it was possible. Commodore was already having trouble just getting the A1000 finished, and during all that time crappy PCs were swamping the market. It's a miracle that they managed to produce what they did.

[Bruce Abbott]

Originally Posted by no9

I'm not an expert but Paula's 4 hard-panned voices suggest that games wasn't taken into account during design. It was rather some video purposes where you can stream 2 stereo tracks and x-fade between them.

Or perhaps they didn't think panning was worth it when most people would be using a mono TV? I used to think that not having panning was a mistake, but now I don't mind. Fixed outputs make the individual sounds more distinctive, which I find more appealing than having them all mixed together. On a typical monitor the separation isn't that great anyway since the speakers are close together, and when you are concentrating on playing a fast action game exactly where the sound is coming from doesn't matter.

Even 6 channels Paula still would be improvement for games. With 3 channels you can create reasonable game soundtrack without resorting to some dubious tricks and you still have the other 3 left to address events in a game.

More is always 'better', but is 4 channels really that bad? Most earlier home computers had no digital sound and 3 or fewer synth channels, yet they still managed pretty well. Even the ZX Spectrum with its bit-banged beeper could produce enough noise to liven up a game.

I recently made an expansion for the Mattel Aquarius that included an AY-3-8910, and I have collected a large number of 'PT3' music tracks for it. I first tried out this chip in the early 1980's and was not impressed with its performance - but I am no musician. Some of the music that has been produced for it is amazing! Only 3 channels and no panning though...

I would even prefer more 4-bit sound channels for sound effects instead less but 8-bit ones. And yet we were left with games like Lotus where is no music during game at all or you can have music from the radio but without sound effects like in Lotus 3.

I'm not sure why that is, but there no theoretical reason they couldn't have had in-game music and sound effects at the same time. Perhaps it was lack of memory, or they couldn't be bothered producing music that worked with the sound effects, or they figured that the sound of the engine was the only music a real petrol-head needs!

[chb]

Originally Posted by Bruce Abbott

More is always 'better', but is 4 channels really that bad? Most earlier home computers had no digital sound and 3 or fewer synth channels, yet they still managed pretty well. Even the ZX Spectrum with its bit-banged beeper could produce enough noise to liven up a game.

Four channels is slightly inconvenient for music + SFX, because for a line-up very common in pop/rock music you'd need at least drums, bass, chords and melody. On the other hand, four voices were pretty standard for 1984 (the C64 had 3, the NES 5 voices), and a sample can contain a full chord or complex drum pattern. And there's quite a lot of music that was great on the C64 and used only two voices (R-Type for example) to allow SFX + music. So it was probably more composers/programmers not bothering to much.

Additionally, the drum track very often uses samples at constant pitch, and that's rather easy and efficient to mix with SFX. Roondar showed some mixer code that takes just about 4% CPU load to mix two channels on an A500, at a sample rate (11 kHz) that's sufficient for drums and SFX. So that could have been done, too.

[no9]

Originally Posted by Bruce Abbott

Or perhaps they

Or perhaps you can't accept the answers that are already there.

I'm not interested in such futile conversation.

[TEG]

Originally Posted by Bruce Abbott

More is always 'better', but is 4 channels really that bad? Most earlier home computers had no digital sound and 3 or fewer synth channels, yet they still managed pretty well. Even the ZX Spectrum with its bit-banged beeper could produce enough noise to liven up a game.

Yeah, the sound of the Amiga would be "better", we had lost what make this sound so unique and magic.

And I agree, the ZX Spectrum sound is another interesting particularity. So much personality. This is what made those years so great for those who had the opportunity to live in this context.

The limitations made the Amiga sound beautiful. It's enough good to be listened for a long time, which can't be say for the ST sound, and enough limited to create a particular experience. It's brilliant.

[pandy71]

Originally Posted by chb

I think having 8-16 lines or about 4kB of buffers in Paula would be absolutely outside the possibilities of the time. That's about 200k transistors just for the buffers. An additional bus for Paula and some (S)RAM, similar to the VIC II's color ram, would have been doable probably, but still expensive.

Well - it could be made similarly as simple BBD only digital - around 70 - 80k of transistors - also CBM could use process with higher density similarly to Lisa (HP source) - in fact integrate Agnus, Denise and Paula within single IC - this could be also very important first step toward Amiga with PCI + external graphic card... (such single IC could act as HW legacy emulator where graphics and sound could be seamlessly integrated with graphics and video from modern solutions).

Originally Posted by chb

The idea with using the video lines is interesting for some add-on hardware with today's technology. If I understand correctly, a kind of Graffiti expansion that gets its data from the digital video pins during pseudo-vblank (display data is still being sent but the add-on blanks the screen and fills the audio buffer)? Could be nice, especially if combined with some kind of Graffiti-like c2p hardware and maybe some DSP for sample compression + playback engine. Latency could be a bit high, bbut sufficient for music replay and game sound effects.

But ofc probably almost no one would use it, as no one used the Graffiti.

Yes, this is idea - of course VBLANK time can be used only by integrated approach, using regular video lines can be added even today (using digital video lines at the video port as 4 bit serial data port) - data lines can be masked in logic so from user perspective it will be like letter-boxed video - black bar (bars on top and bottom).

TI has neat PRU co-processors that can be used for such HW/SW activity (BeagleBone) - it can even emulate for example 68k so act as alternative to FPGA Amiga accelerators.
https://www.embeddedrelated.com/showarticle/586.php