Amiga Sound Quality

[Mrs Beanbag]

What does the 6581 need the CPU for?

Even Paula doesn't need the CPU. It only needs samples to be in Chip RAM.

[pandy71]

Originally Posted by Mrs Beanbag

What does the 6581 need the CPU for?

Even Paula doesn't need the CPU. It only needs samples to be in Chip RAM.

To start output sound, particular register need to have particular content - same principle for all - so even if you provide such data from some ROM and counter (no CPU) then anyway you can provide (as principle) similar data for Paula.

[Mrs Beanbag]

Well exactly. Paula can be controlled by the Copper anyway so in principle a copperlist could play a tracker mod. But this isn't very profound. To say that SID is not a synth because it needs something external to tell it to play a note, is like saying my music keyboard downstairs is not a synth because it doesn't play any sound unless I press a key.

The point is, SID can produce sound on its own without any further input. You could rig up a simple manual control system to it, with a few switches and dials and some basic logic. It has its own internal ability to generate waveforms.

Paula needs to get its waveform data from memory via DMA, in the form of samples, either recorded or synthesised by an external CPU.

[pandy71]

Originally Posted by Mrs Beanbag

Well exactly. Paula can be controlled by the Copper anyway so in principle a copperlist could play a tracker mod. But this isn't very profound. To say that SID is not a synth because it needs something external to tell it to play a note, is like saying my music keyboard downstairs is not a synth because it doesn't play any sound unless I press a key.

The point is, SID can produce sound on its own without any further input. You could rig up a simple manual control system to it, with a few switches and dials and some basic logic. It has its own internal ability to generate waveforms.

Paula needs to get its waveform data from memory via DMA, in the form of samples, either recorded or synthesized by an external CPU.

I disagree - both need to be programmed in this or different way - without some data they are silent, Paula can play square wave (1 word - 2 sample buffer) almost without acess to RAM.
So going strict (as i understand You) on requirement - SID/Paula are not synth - synth can be only that produce sound without programming - for example analog generator (VCO, XTAL, astable RC, Wien bridge, etc) or digital generator (ring oscillator etc).

Btw - how you locate product like Access Virus? Is this is a synthesizer or not?

[Mrs Beanbag]

My guitar is silent if I don't strum it, that doesn't mean it is not an acoustic instrument.

This has fully degenerated into a silly word game now.

[hooverphonique]

if it can generate a waveform from inputting a few parameters, it is a synth, otherwise not.. it doesn't even have to be able to play back the waveform.. so paula is not a synth..

[TCD]

Maybe this thread go back on topic? Thank you in advance.

[pandy71]

Originally Posted by TCD

Maybe this thread go back on topic? Thank you in advance.

Ouch, it hurts, but sorry, my apologies.

Going back to strict beginning - it is inefficient from software emulator point of view trying to emulate real Amiga sound - it require to solve problem related to PWM distortions and to simulate real analog post-processing in Amiga - i believe that current implementation is functionally more close to intentions of the HW Amiga designers.

[commodorejohn]

Originally Posted by pandy71

i believe that current implementation is functionally more close to intentions of the HW Amiga designers.

Whether or not this is true, it's irrelevant, since what people want out of an emulator is the behavior of the emulated system, and the Amiga sound device behaves in a characteristic way whether its designers meant it to or not. By this logic you could say that all the people who modify Hammond organs to add a pleasing amount of key-click are stupid because Laurens Hammond thought that was a bug instead of a feature.

[Thorham]

Originally Posted by commodorejohn

and the Amiga sound device behaves in a characteristic way whether its designers meant it to or not.

Not to mention that the designers probably just meant for the audio system to not be too expensive and still sound good. All this Amiga specific sound (whatever that means) probably comes from the price/performance requirement.

[pandy71]

Originally Posted by commodorejohn

Whether or not this is true, it's irrelevant, since what people want out of an emulator is the behavior of the emulated system,

Whether or not this is true, it's irrelevant, as people expect to have functionality from emulator - for real behaviour use original equpment which is possible.

Qeustion: do you know what is UPWM and NPWM?

http://books.google.nl/books?id=givs...%20PWM&f=false

Originally Posted by commodorejohn

and the Amiga sound device behaves in a characteristic way whether its designers meant it to or not. By this logic you could say that all the people who modify Hammond organs to add a pleasing amount of key-click are stupid because Laurens Hammond thought that was a bug instead of a feature.

And? Once again - use original Amiga for original sound as each Amiga have slightly different sound.

[Mrs Beanbag]

There is something to be said for purposefully not emulating a machine exactly. Some emulators use intelligent algorithms to increase the resolution of the graphics. Obviously that is not aiming for authenticity, but just to make it look good. Same could apply to sound, and we could add sample interpolation &c, and have all the Amiga games with all the original gameplay, but which look and sound BETTER than before.

And why not?

But different people like different things so we should have options. Not everybody still has a real Amiga anymore.

[mc6809e]

Originally Posted by Thorham

Not to mention that the designers probably just meant for the audio system to not be too expensive and still sound good. All this Amiga specific sound (whatever that means) probably comes from the price/performance requirement.

I think the Amiga specific sound comes from the harmonics created as the output jumps from wavetable value to value.

A two value wavetable of -128 and +127 when output at 1000 values per second, for example, will look like a square wave which means there will be odd harmonics of the 500 Hz fundamental at 1500 Hz, 2500 Hz, etc.

This seems impossible. How can a sample at 1000 Hz produce frequencies above the Nyquist limit? The answer is the output is not a sample. It's a waveform.

[chaos]

Originally Posted by mc6809e

A two value wavetable of -128 and +127 when output at 1000 values per second, for example, will look like a square wave which means there will be odd harmonics of the 500 Hz fundamental at 1500 Hz, 2500 Hz, etc.

Well, you *are* asking for a square wave, aren't you?

This seems impossible. How can a sample at 1000 Hz produce frequencies above the Nyquist limit? The answer is the output is not a sample. It's a waveform.

Why is this impossible? You can't produce a square signal without a bunch of higher frequency components. How do you define a Nyquist limit in this case? Plus, isn't a waveform just a collection of samples?

[mc6809e]

Originally Posted by chaos

Well, you *are* asking for a square wave, aren't you?


Why is this impossible? You can't produce a square signal without a bunch of higher frequency components. How do you define a Nyquist limit in this case? Plus, isn't a waveform just a collection of samples?

You can't define a Nyquist limit is this case. That's my point. Approaching the Amiga sound quality question from the perspective of sampling theory doesn't work.

Take this example:

Suppose a 500 Hz sine wave is sampled at 1000 samples per second and stored.

Is the waveform a sample of a sine wave or square wave?

It's obviously a sample of a sine wave but will be output as a square wave and new frequencies will appear in the output that weren't in the originally sampled signal.

Now sample the 500 Hz sine wave at 10000 samples per second.

When output this will be look more like a sine wave. On the other hand, outputting the first sample at 10000 values per second will still look somewhat like a square wave.

The point is, when considering Amiga sound quality, the system should be seen as more than a simple recorder. It outputs arbitrary waveforms at arbitrary rates and volumes with a modest amount of filtering.

Seeing Paula as a mere sample player is what was responsible for some of the poor PC-based mod players that came out very early on. They were developed with the (wrong) idea that the Amiga was a sample player and so they used simple interpolation and filtering to get a 44.1KHz PC output. The sound was very different and was obviously missing something.

Proper PC mod players got better when they started to up sample instead.

[chaos]

Originally Posted by mc6809e

You can't define a Nyquist limit is this case. That's my point. Approaching the Amiga sound quality question from the perspective of sampling theory doesn't work.

Yes, the point I was trying to make is that you can't apply Nyquist limit to producing sounds like Paula does. I'd think that any sound card will act the same if given samples like you described.

Suppose a 500 Hz sine wave is sampled at 1000 samples per second and stored.

Is the waveform a sample of a sine wave or square wave?

It's obviously a sample of a sine wave but will be output as a square wave and new frequencies will appear in the output that weren't in the originally sampled signal.

You will get the same sine wave if you low-pass filter it, which is what you are supposed to do. You can't just play back samples and expect to get the original signal back, you must filter it.

[pandy71]

Originally Posted by mc6809e

You can't define a Nyquist limit is this case. That's my point. Approaching the Amiga sound quality question from the perspective of sampling theory doesn't work.

Take this example:

Suppose a 500 Hz sine wave is sampled at 1000 samples per second and stored.

Is the waveform a sample of a sine wave or square wave?

Nope, you are confused - this is common - people don't understand that Nyquist conditions require presence of the reconstruction filter - ideal reconstruction filter will produce sine wave.
http://en.wikipedia.org/wiki/Reconstruction_filter

[mc6809e]

Originally Posted by chaos

You will get the same sine wave if you low-pass filter it, which is what you are supposed to do. You can't just play back samples and expect to get the original signal back, you must filter it.

Well suppose the original signal was instead a square wave. How much filtering would be needed then? None, of course.

[mc6809e]

Originally Posted by pandy71

Nope, you are confused - this is common - people don't understand that Nyquist conditions require presence of the reconstruction filter - ideal reconstruction filter will produce sine wave.
http://en.wikipedia.org/wiki/Reconstruction_filter

I know all about reconstruction filters. I asked the question rhetorically so as to illustrate to others that there was an ambiguity in interpreting what the stored waveform really is.

If the Amiga had been built with a programmable reconstruction filter that could be tuned to the waveform output frequency, then it would be a sample player.

That fact that it only has a couple of gently sloping fixed low pass filters (and even one of these can be turned off) tells us that simple signal reconstruction and playback was not entirely the goal. Synthesis was also the goal.

[Mrs Beanbag]

Originally Posted by mc6809e

That fact that it only has a couple of gently sloping fixed low pass filters (and even one of these can be turned off) tells us that simple signal reconstruction and playback was not entirely the goal. Synthesis was also the goal.

Separate filters for each channel with programmable cut-off frequency would have been awesome though. You could do so much synth with that as well as eliminate the aliasing from the samples!

Incidentally I've lately been experimenting with compressing my samples by upsampling, and restoring them with a Lanczos interpolation. Works quite well.