Which Amiga computer technically has the best audio quality?

[Photon]

Yes, I would say you can go up to just below 7.5 kHz without losing the softer (LED-off) sound. And then not a too sharp knee - even the Amiga 500 LED on and off you can hear the difference quite clearly in the right band (e.g. not already filtered electric bass up to speech no esses).

[pandy71]

Originally Posted by no9

This example represents the real-life experience of an Amiga user in a common situation, not an abstract edge case.

lol - sounds like perfect used cars seller speech... there is nothing "abstract edge case" in playing sound with periods 128..124

Originally Posted by no9

The way Amiga, as a whole system, operates on sound defines the quality that users face. There is nothing misleading in this, and it is far more representative of the Amiga in common applications such as games or demos.

? oh common... almost like some corpo BS

Use module without poor samples sampled at 8kHz without proper antialiasing... i recall some demo where 4 windows with waveform visualization was present and samples was based on probably Janet Jackson song - i recall it was very good quality...

On opposite - perhaps some future composer chip tunes - for sure they will be way more correct from aliasing perspective.

[no9]

Originally Posted by pandy71

lol - sounds like perfect used cars seller speech... there is nothing "abstract edge case" in playing sound with periods 128..124

Yeah, yeah... thousands of Amiga games and demos proves just that. 128 and not a milisecond longer - that's our everyday Amiga experience.

Use module without poor samples sampled at 8kHz without proper antialiasing...

Place your order where someone actually cares about your criteria.

[daxb]

The main reason why the sound/sample quality was so low/bad was the limited size of storage and memory. I'm sure if the limit wasn't so high/low they chose for the better. I.e. (in most cases) instead of ~8kHz (C-2) to ~16kHz (C-3) the target would be >16kHz or 22kHz to 27kHz. This makes a big different in sound quality. How clean the source sound is or how good you can record it is also important. A good or bad example is the Agony piano mod which is great but sample quality is bad (a lot of noise). Try with led filter on an off.

[Karlos]

I'll tell you what we really need: Software-controllable, low-pass resonant filters.

Ideally that would one be per channel but it's probably not possible (I don't think there are independent analogue outputs per channel at any tapping point), so at least independently on the left and right side since as a musician these are probably going straight into a mixer as separate inputs anyway.

Then all we need is the means to control them from our favourite trackers.

[no9]

I was looking for something else but stumbled upon some deliberations about Amiga filters in E&MM magazine from '86. And more.

https://www.worldradiohistory.com/UK...6-08-S-OCR.pdf

[pandy71]

Originally Posted by no9

Yeah, yeah... thousands of Amiga games and demos proves just that. 128 and not a milisecond longer - that's our everyday Amiga experience.


Place your order where someone actually cares about your criteria.

Sorry, i've just realized you are from Poland...

[minator]

Originally Posted by ShK

Thanks for the input! Here's the updated table:

I'm pretty sure the A1000 didn't have the LED filter, certainly when I tried to activate it I got a visit from a Guru!

[minator]

The Amiga's sound is great. It's not exactly CD quality but it was of it's time - many commercial samplers at that time had very similar specs. The Ensoniq Mirage and ESQ (designed by the guy who designed the SID chip) were both 8-bit as was the PPG Wave 2 / 2.2 and even the mighty Fairlight (1).

These all created grungy low sample rate 8-bit sounds but put them through analogue filters so they're not quite as grungy as Paula could be.
What's really interesting is how many of these synths (and the Amiga) worked compared to later on. They changed pitch by speeding up or slowing down samples, this means any aliasing moves up and down with the pitch. It's a form of distortion, but because it's related to the notes being played, it sounds good. Indeed there's a lot of effort that goes into emulating these sounds in modern recreations. In fact the Modal 002 a few years back uses a similar method to create sounds and it's awesome.

The Amiga was a bit different because you had the added limitation of people trying to put multiple mods on an 880K disc. Hence the 8K sampling rate.
This low sampling rate accentuates the aliasing noise but because of the variable sample rate playback and the very basic hardware filter, it blends in creating the unique Amiga sound.

I found the old ST-01 sounds disc on line and plan to play them through a modern synth, however I want to resample them first so they don't get smoothed out. If it works, this should create a very Amiga-ish sound.

[ShK]

Originally Posted by minator

I'm pretty sure the A1000 didn't have the LED filter, certainly when I tried to activate it I got a visit from a Guru!

Thanks, let's try to make chart better:

[Jope]

Originally Posted by minator

I'm pretty sure the A1000 didn't have the LED filter, certainly when I tried to activate it I got a visit from a Guru!

No guru should happen if you toggle the LED on an A1000, but the end result is just a LED that changes brightness, as the filter is permanently on.

[8bitbubsy]

Originally Posted by h0ffman

A600
Due to an error in the audio circuit design it sounds really muffled. This can be fixed by removing C321 and C331 from the underside of the board then it matches the A1200. The only drawback is it reduces the overal output level and thus brings the noise floor up a little.

That's NOT the way to do it! You are ending up with a cutoff so high that the PWM artifacts (from volume PWM) are not filtered away. The fact that A1200's lp cutoff is 34kHz (and not removed all together) is for a reason. The proper "fix" is to match the RC values with that of the A1200.

Also there is nothing wrong with the audio circuitry in an A600, it sounds about the same as an A500 (though different kind of gain setting, IIRC), and that was intentional. They wanted to reduce aliasing in the sample reconstruction. It made sense, but at the same time it made the listening experience a bit dull. My guess is that they increased the cutoff on the A1200 because PCs started to have (somewhat) crystal clear 16-bit audio at the time.

Another thing: I saw the database of cutoffs someone made in here. It has some errors... A1200's filter cutoffs are the same for all revisions, even though the schematics you find on the internet say other wise. Don't trust the schematics, it's that simple. How do I know this? Well, just look at mobo scans for the different revisions, and you'll find that the resistor (part of the RC) values are the same. This is already proof enough that they never implemented the rev2 filter change.

EDIT: Also the same for Escom mobos!

[alexh]

Originally Posted by redblade

How hard was it to program the CD32 and CDTV CD quality sound?

CD32 and CDTV can only play CD quality sound from the CD-drive as CDDA. The data doesn't go into the Amiga but directly to a DAC from where it goes through an analog mixer to be combined with Paula.

Originally Posted by redblade

do the samples have to be a track number on the CD?

Yes

[Mick]

Originally Posted by 8bitbubsy

Another thing: I saw the database of cutoffs someone made in here. It has some errors... A1200's filter cutoffs are the same for all revisions, even though the schematics you find on the internet say other wise. Don't trust the schematics, it's that simple. How do I know this? Well, just look at mobo scans for the different revisions, and you'll find that the resistor (part of the RC) values are the same. This is already proof enough that they never implemented the rev2 filter change.

The Amiga 1200 1A used the 1.5K config or at least some of them did, it could be that Commodore transitioned in the middle of the production. I think it's the A600's that never used 27.2khz cutoff but only 4.4khz despite what the schematics say.

[ Show youtube player ]

Also, the CD32 seems to have a 70khz cutoff for Paula output so I don't think noise was necessarily a problem. From what I've read the capacitor is also to stabilise the OP AMP as a current to voltage converter. I think Commodore just couldn't get the level of amplification right and the changes in capacitor accompanies the change in resistor value.

[8bitbubsy]

Oh wow, look at that!

I found another A1200 rev1A on google images with the 1.5k R component. Makes me wonder if maybe all of them are 1.5k, and that I just concluded that all Amigas had the same values since I checked about every revision (maybe I forgot 1A). Either way, sorry for being cocky about it, I stand corrected.

[jaycee1980]

The use of 1K5 for the first stage I/V conversion was a mistake. The result is too much gain, and the signal is distorted. Commodore later used 680 ohms in the 1200, and 750 ohms in the 4000. These are appropriate gain values.

[jaycee1980]

I was sufficiently intrigued by the variations across the various Amiga versions that I decided to simulate the audio stage in LTspice.

The audio circuit has not actually changed that much since the A500. The A1000 circuit had a much more complicated multistage filter, but it is the only machine that has it. Everything after that has pretty much followed the A500 design. Yes that includes the AGA machines and the CD32.

Here's my model of the circuit. It's simplified a tiny bit but the essential elements are there.

U14A forms what is known as an I/V converter. Paula's audio outputs are current based, and this stage transforms that into a variable voltage. The gain is set by R331. Vref is approximately 2.5V - this is because Paula is a 5V chip, so for representing zero, half of the supply voltage is used. C1 is used to limit the bandwidth of the opamp so that it does not become unstable - as an effect of this, it also forms a lowpass filer.

U14B forms the well known "Amiga filter". It is a Butterworth filter implemented using the Sallen-Key topology common for opamp based filters. X1 represents the filter switch which is normally implemented using the JFET at Q331. With X1 closed, the filter is bypassed and the second opamp stage simply becomes a buffer.

Finally we have C334, R334, R335 and E303. C334 is used to remove the 2.5V DC offset caused by the first stage. R334 and R335 set the final output amplitude, and there is some HF rolloff caused by the capacitance of E303 - but it is negligible in the audio band. THe value of C334 does also cause some LF rolloff.

[jaycee1980]

So lets start with the 500/500+/200. These use R331 = 360, C1 = 100nF, R334 = 1K, R335 = 390. So what do we get?

With the Amiga filter off, C1 = 100nF gives a HF rolloff of ~4.3KHz at -3dB. Rather excessive. I think this was a design mistake and the designer actually intended to use a much more reasonable 10nF. The audio bandwidth is quite limited, and thats before the Amiga's filter is even enabled.

With the filter on, the HF rolloff point becomes ~2.3KHz.

With the resistor values shown the audio output is a maximum of 200mV p-p

[jaycee1980]

So now let's do the A1200. It doesnt help here that the schematics are often wrong. Most machines were manufactured with R331 = 680, C1 = 6800p, R334 = 1K, R335 = 360.

With these values, we get a HF rolloff of ~29.3KHz with the filter off, ~2.6KHz with it on, and an overall signal amplitude of 350mV p-p. However some A1200's have R335 = 2K, which gives an overall amplitude of 887mV p-p.

With the A1200s that had the incorrect R331 = 1K5, and R335 = 2K, this gives a very loud signal of nearly 2V p-p ! The problem with this is it causes distortion, because it causes the bottom of the waveform at the I/V stage to be clipped as it attempts to go below 0v.

[jaycee1980]

So what about the other machines ?
The A3000 uses pretty much identical values to the A500/2000, with two difference - C1 is not fitted, and the equivalent of E303 is 100nF.

This gives us ~5.2KHz with the filter off, ~2.2KHz with it on, and an overall amplitude of 200mV p-p. There is also an additional issue - the I/V stage may become unstable, as there is now a peak at around 880KHz of 10dB. This could lead to instability in the opamp, but fortunately it doesnt seem to.