Amiga (Paula) vs Atari ST (Yamaha AY)

[pandy71]

Originally Posted by AmigaHope

What he's saying is that it's not nonlinear from the center of the waveform (i.e. signed, like a-law, mu-law, etc.), but rather that it's nonlinear from the bottom of the unsigned waveform, so that the most resolution is at one half of the extreme side of the waveform and the least is at the other extreme side of the waveform. It means you'd never be able to, for instance, make a good sine wave, instead it would have a nice sine shape on the bottom and a square shape on the top of the waveform.

Now I don't know if he's *correct* or not, but that's what I got from what he said.

Don't get me wrong but i see no problem to convert signed to unsigned and vice versa - important is quantization steps and seem curve is correct - higher level less accurate (bigger) quantization step. But i may wrongly interpret datasheet - poor quality especially for YM.
Once again - from my perspective volume control act as non linear DAC and if this logarithmic characteristic is not addressed correctly then you have severe distortion. Key is to map uniformly quantized PCM to quasi-logarithmic characteristic of this "DAC" .

[Bruce Abbott]

The AY DAC is a 'single-ended' programmable current generator. When this current is passed through a resistor connected to 0V, it produces a voltage proportional to the current. With standard load resistance the output voltage varies from 0V (amplitude 0) to 1V (amplitude 15), with 0.5V achieved at amplitude 13.

The datasheet is a little misleading because in one place it shows a linear ramp, while the actual output is logarithmic. Some versions of the AY datasheet also have an oscilloscope trace which shows the actual output waveform. Here you can see the 0V level (horizontal line at bottom) and voltage steps as the amplitude is ramped down and up. The finest steps are at the bottom, which equates to the negative peaks of an 'AC' sound waveform.

Tones are square waves whose voltage oscillates between 0V and the voltage set by the amplitude. Envelope shaping can be used to produce a tone with decaying volume, the logarithmic output reduction matching the decay characteristics of a typical physical instrument. However since the output is referenced to 0V, the 'mid-point' voltage of the waveform also reduces logarithmically, producing a lower frequency which is audible at fast envelope rates.

When I first tried out an AY-3-8912 in the early 1980's, I was quite disappointed by the audible steps and 'interference' from the envelope shaping. However at that time I was just trying to generate the cleanest possible notes, not realizing that the AY's 'flaws' could be used to make much richer and more interesting sounds!

[alexh]

Almost completely off topic (different sound chip) but if you've got an STe and never tried this sound demo before you should

http://www.pouet.net/prod.php?which=52336

[Hewitson]

What's so impressive about an STe playing modules? People have been doing that for decades.

[alexh]

Originally Posted by Hewitson

What's so impressive about an STe playing modules? People have been doing that for decades.

You must never have played modules on a real STe using the players commonly available? If you had, and you then used this demo you would know what is so great about it. Or there's something wrong with your ears :-)

[pandy71]

Originally Posted by alexh

You must never have played modules on a real STe using the players commonly available? If you had, and you then used this demo you would know what is so great about it. Or there's something wrong with your ears :-)

After brief source check, Paula emulation is done with fixed frequency (obvious as STE do not support variable sample rate as Paula) - and this means that anything sampled faster than 25033Hz will be not played correctly. Only NTSC Paula is emulated (PAL Amiga i.e. whole European demoscene is played inaccurately).
Don't get me wrong, I agree - this can be spectacular for Atari STE and Falcon users but from Amiga perspective this still not Paula due radical difference in both architectures (i assume Falcon DSP could be capable to perform proper signal processing to deal with above problems).

I have strong impression that good coders on Atari ST added trough pure software many unique Amiga HW features (of course at a cost of CPU utilization) and at some point Atari ST coders pushed limitations beyond Amiga. Why this not happen on Amiga?
This is obvious example of Atari ST coders skills http://www.leonik.net/dml/sec_pcs.py and http://www.leonik.net/dml/sec_crypto.py

[frank_b]

Originally Posted by pandy71

After brief source check, Paula emulation is done with fixed frequency (obvious as STE do not support variable sample rate as Paula) - and this means that anything sampled faster than 25033Hz will be not played correctly. Only NTSC Paula is emulated (PAL Amiga i.e. whole European demoscene is played inaccurately).
Don't get me wrong, I agree - this can be spectacular for Atari STE and Falcon users but from Amiga perspective this still not Paula due radical difference in both architectures (i assume Falcon DSP could be capable to perform proper signal processing to deal with above problems).

I have strong impression that good coders on Atari ST added trough pure software many unique Amiga HW features (of course at a cost of CPU utilization) and at some point Atari ST coders pushed limitations beyond Amiga. Why this not happen on Amiga?
This is obvious example of Atari ST coders skills http://www.leonik.net/dml/sec_pcs.py and http://www.leonik.net/dml/sec_crypto.py

DML is in a league of his own as a coder. He's an absolute genius. There are probably very few coders as good as he is in the world.

[pandy71]

Originally Posted by frank_b

DML is in a league of his own as a coder. He's an absolute genius. There are probably very few coders as good as he is in the world.

Well... from my perspective he is passionate, determined and wiling to do research - classical example is color reduction to efficiently use CLUT machines limitation (valid for Amiga and Atari ST) - can be done with objective approach but also can be done on subjective fashion.
Last years bringing more and more machine learning - image decomposition, human visual system - those things happen of course not on ordinary Amiga (as even with fastest CPU it will be slow).

For example from my personal naive research is quite obvious that using https://people.eecs.berkeley.edu/~dc...loads/scolorq/ as pre-processor to perform CLUT selection and classical error distribution (even Floyd Steinberg) produce very decent results usually way better than any other "objective" color requantization algorithm. Same for HAM - with proper CLUT and wise error distribution (edge detection based) probably most of HAM limitations (ramp) can be removed.
Using YCoCg instead YCbCr may improve speed of video decoding. All those techniques exist and can be beneficial for limited HW.

Similar for audio - strong ATH noiseshaping may deliver near 16 bit perceived quality on Paula without "14 bit" trick.