Paula's audio range is 9 octaves

[demolition]

If you listen carefully to MIDI patches for example, you can also hear that they use different samples for different ranges. One sample may only cover one octave, and then you notice that the timbre changes a bit from one key to the next as it moves to another sample.
For a synthetic sample like a square wave or triangle, one sample can be used for all octaves, but for a recorded acoustic instrument, it will not sound right if you move too far from the recorded sampling frequency as daxb writes since the harmonics will not be right.

[Amiga1992]

Originally Posted by Foebane

Marketing

I assumed as much, but wanted to be sure I wasn't missing anything. It did sound like a load of old shit.
Here lies an advantage of a machine that has a tone generator, you can get more octaves sounding out of it. Sure its sounds are simpler, but the range makes it very accessible right off the bat (this is why people shouldn't compare DACs to tone generators).

Anyway yeah, multisampling, which is what demolition describes, is always the way to go with a sample-based machine, it sucks that tools like Protracker don't support something like that directly (creating an instrument where you specify different samples for different ranges). I believe there were other trackers that did, like Musicline. but it's not a feature I really ever used back then.

Also about this, even if you use a one cycle wave sample, on the Amiga, you still can only map it to these pitiful 3 octaves, and resample it much, and it will start sounding wrong. But they are tiny, so generating by other means 4 or 5 samples at different octaves should take no space and be very possible (until you hit Protracker's ridiculously low sample slots amount )

[pandy71]

Originally Posted by demolition

All DACs use a low pass filter, also known as a reconstruction filter to remove overtones/distortion.
In this case, I was referring specifically to the Amiga implementation with Paula + the external analog low pass filter which is located well down in the audible range. So you can either choose to leave it on and make everything sound dull or leave it off and have distortion artifacts.

Properly sampled signal doesn't have overtones, overtones can be created by DAC imperfections however some DAC's (like Sigma Delta) can be inherently linear so they don't create too much distortions.
Paula distortions are usually more complex natures as in Paula volume regulation is implemented PWM like way.
Problem is that in Amiga times all samplers was quite poor quality and also people had limited knowledge - nowadays as samples can be processed on PC and converted properly to Amiga this should be not a problem. Most imperfections you referring to are outcome of 80's/90's.
I bet that DAC in Paula should give noise floor around -45 .. -43dBFS at least.

And i agree - Paula lacks for variable low pass filter (but i understand that in time when Amiga was designed there was no sane way to implement such filter as it will require some DSP and modern DSP was born only few years before Paula was designed - uPD7220 1980 and TMS32010 1983)

[demolition]

Originally Posted by pandy71

Properly sampled signal doesn't have overtones, overtones can be created by DAC imperfections however some DAC's (like Sigma Delta) can be inherently linear so they don't create too much distortions.

Sigma Delta converters are definitely not linear, however they use oversampling and noise shaping techniques to shift those overtones to much higher frequencies. This makes it so much easier to remove them as you can use a simple low-order low pass filter. Yes, the overtones appear due to DAC imperfections, but unfortunately there is no such thing as a perfect DAC.

Originally Posted by pandy71

And i agree - Paula lacks for variable low pass filter (but i understand that in time when Amiga was designed there was no sane way to implement such filter as it will require some DSP and modern DSP was born only few years before Paula was designed - uPD7220 1980 and TMS32010 1983)

Well, the SID chip in C64 has a variable low pass filter and that predates Paula. Or at least it intended to be a low pass filter in the specs. But it can be done without DSP using well-established analog techniques as used in amplifiers, mixers, guitar pedals etc. The hard part is mixing digital logic with analog components in a single IC since the requirements for the silicon processes are very different for the two.

[Amiga1992]

Originally Posted by pandy71

And i agree - Paula lacks for variable low pass filter (but i understand that in time when Amiga was designed there was no sane way to implement such filter as it will require some DSP

Nah that's total BS man, there were plenty of analog filters available, like in the C64.
They should have created 4 discrete voices and add an analog filter to each.
Which I am sure is what Yannes proposed before leaving C=, forming Ensoniq and taking that idea with him to make the Mirage.

[pandy71]

Originally Posted by demolition

Sigma Delta converters are definitely not linear, however they use oversampling and noise shaping techniques to shift those overtones to much higher frequencies. This makes it so much easier to remove them as you can use a simple low-order low pass filter. Yes, the overtones appear due to DAC imperfections, but unfortunately there is no such thing as a perfect DAC.

DS DAC are as linear as signal can have one of two values. Oversampling is something else than linearity. From practical perspective DAC with SFDR like 110dB are feasible and practically used widely nowadays - 110dB is way more than 8 bit non oversampled system can deliver.

Originally Posted by demolition

Well, the SID chip in C64 has a variable low pass filter and that predates Paula. Or at least it intended to be a low pass filter in the specs. But it can be done without DSP using well-established analog techniques as used in amplifiers, mixers, guitar pedals etc. The hard part is mixing digital logic with analog components in a single IC since the requirements for the silicon processes are very different for the two.

SID use single linear filter - such filter can be programmable by changing values of components or reconfigured to create other than low pass filter.
Paula in theory could use switched capacitor filter but still such techniques was too costly to be implemented in Paula (and practical implementation may be not possible due a way how signal level is handled by Paula).

Originally Posted by Akira

Nah that's total BS man, there were plenty of analog filters available, like in the C64.
They should have created 4 discrete voices and add an analog filter to each.
Which I am sure is what Yannes proposed before leaving C=, forming Ensoniq and taking that idea with him to make the Mirage.

Nope... way how Paula produce audio (variable samplerate) push requirements for analog filter very high - simple case will be settling time for such analog filter - filter may be practically not implementable (to cover sample rates from 6 - 7 kHz up to 56kHz and beyond to 1.79MHz). Yannes (@Ensoniq) and SID use NCO (DDS) approach (where sample rate is constant), Amiga use Wavetable with variable sample rate. Those techniques are different fundamentally.

[Amiga1992]

Originally Posted by pandy71

Yannes (@Ensoniq) and SID use NCO (DDS) approach (where sample rate is constant)

No man.
The Ensoniq Mirage is a variable rate 8-bit sampler with discrete, low pass, variable analog filters.
The ESQ-1/SQ-80 are wavetable synths with, again, same filters.

There's always a limit to a reconstructing filter if you wanna keep costs down, so no, you can't go to 56Khz (the Amiga does not even go ther eso I don't understand why this mention), the MIrage only lets you go up to 32k or something, but you could buy a replacement filter package that let you go to 48K.

[NorthWay]

Originally Posted by Akira

you can't go to 56Khz (the Amiga does not even go ther eso I don't understand why this mention)

?
DBLPAL/NTSC / Multiscan screenmodes are 56KHz. If you fiddle the display enough (preferably with the video signal off) you should be able to get some silly high frequencies from poor old Paula.
(AFAIK some gfx cards have a 56KHz setting that makes sure this is active when viewing gfx-card modes.)

[Amiga1992]

Originally Posted by NorthWay

?
DBLPAL/NTSC / Multiscan screenmodes are 56KHz. If you fiddle the display enough (preferably with the video signal off)

Another theoretical height that has no practical purpose.
You can't actually, like, track music, if you have no screen.
This claim approaches the "9 octaves" marketing blurb.

[ross]

Originally Posted by Akira

Another theoretical height that has no practical purpose.
You can't actually, like, track music, if you have no screen.

With some magic it's possible in all resolutions
But in fact it is rather complicated and heavy.

[buggs]

Originally Posted by Akira

Another theoretical height that has no practical purpose.
You can't actually, like, track music, if you have no screen.
This claim approaches the "9 octaves" marketing blurb.

I disagree. All you need to do on ECS/AGA Amigas is to switch to a double scan screenmode where twice the number of DMA slots for Paula are available. Picasso96 (as mentioned above) can do that on it's own for RTG screenmodes. This way you can easily play with up to 56 kHz DMA rate.

Whether the popular trackers support these capabilities is another matter, of course.

[pandy71]

Originally Posted by Akira

No man.
The Ensoniq Mirage is a variable rate 8-bit sampler with discrete, low pass, variable analog filters.
The ESQ-1/SQ-80 are wavetable synths with, again, same filters.

There's always a limit to a reconstructing filter if you wanna keep costs down, so no, you can't go to 56Khz (the Amiga does not even go ther eso I don't understand why this mention), the MIrage only lets you go up to 32k or something, but you could buy a replacement filter package that let you go to 48K.

Yes man, Ensoniq use DCO (NCO) and use AWG as foundation for waveform generation (Paula is unable to skip samples in wavetables - it can only play faster or slower complete wavetable where Ensoniq can use single sample rate waveform and play it with different speed by skipping samples accordingly to programmed frequency (phase accuracy is variable with fixed granularity)).

And reconstruction filter limitation are always trade-off between implementation cost and efficiency - nowadays such filters are implemented digitally and with help of oversampling they require very cheap and simple analog reconstruction part (btw Amiga has reconstruction filter always active - filter frequency is controlled and with clever programming some intermediate frequencies can be achieved so you can simulate different cut-off frequencies).

I've mentioned 56kHz as it is frequently used "feature" (for ECS/AGA by HW, for ICS/OCS it should be possible to be simulated by CPU with Copper). Technically you can go way higher than 56kHz and with help of aggressive noise shaping push Paula above 16 bit quality (anyway reconstruction filer cut-off should be around 20 -21kHz maximum).
Any sane RTG compatible tracker can benefit from increased horizontal frequency.