Roland Alpha Juno-2 pwm sawtooth

[aciddose]

Here's a .flac attached of me doing a quick doodle-version of the test. It should be quite easy to hear the harmonics of the same voice, as well as the chorus of the other five voices slightly lower in volume. I'd say maybe the other voices could be ~-20 dB or so but that's a guess. In any case they don't show up well because I didn't use the oscillator pitch envelope but only swept the pitch using the pitchbend.

juno_1_ramp_minimum_frequency_pitchbend_sweeps.zip

Looks like KVR doesn't support flac/mp3 so here's:
https://soundcloud.com/aciddose-1/juno- ... end-sweeps
You can use the download link on soundcloud and it should be able to grab the flac. I don't have a great buffer connected this is not an ideal recording, but, good enough to "hear".

[TechnoManiac]

Here's a .flac attached of me doing a quick doodle-version of the test. It should be quite easy to hear the harmonics of the same voice, as well as the chorus of the other five voices slightly lower in volume.

Thanks for the detailed answers

Did I understand correctly that this is the reason for these harmonics:

So we're hearing not just interference from other voices, but also interference from another division of the master clock before the final division and wave-shaping produces the output waveform.

[TechnoManiac]

I'd describe the effect of timbre change as "interference", and it's likely the result of the different divisors the oscillator chip gets with portamento on or off.

I don’t quite understand how different divisors can influence the timbre. Do you mean that the internal waveforms (SAW, PWM, SUB) somehow react differently to different dividers? Or are they still synchronized with each other? It sounds like several waveforms are out of phase. Or something like formant shift ))

In addition, if the CPU is producing the integration (linear) portamento effect, this may be impacting the clock rate at which it can output address/data to the oscillator chip and strobe the data latch pin. This could dramatically impact the performance of the oscillator or even every module in the synthesizer due to changes in timing on the CPU alone.

I mean the timbre in the area when the frequency after portamento has stabilized. The note frequency is exactly the same as without portamento. I don’t understand how different dividers can affect the timbre itself if the oscillator reproduces exactly the same frequency. But something about the sound is clearly different.

[aciddose]

So we're hearing not just interference from other voices, but also interference from another division of the master clock before the final division and wave-shaping produces the output waveform.

Maybe. That's one hypothesis to add to the list with no experimental evidence to really support anything.

From the flac, I've isolated one cycle 48000/11731 and observing near the end of that cycle I see a very chaotic ramp at 48000/46. 11731/46 = 255.021~, so this could in fact imply either an 8-bit DAC, or that the ratio is near 1/256~ for some other reason. Hypothetically we might question why they'd use so few bits here. A division step would also be conveniently in an 8-bit ratio, it wouldn't make much sense to see this particular combo given the 32' which should only be 1/4 (?).

[aciddose]

I don’t quite understand how different divisors can influence the timbre. Do you mean that the internal waveforms (SAW, PWM, SUB) somehow react differently to different dividers? Or are they still synchronized with each other? It sounds like several waveforms are out of phase. Or something like formant shift ))

I mean the timbre in the area when the frequency after portamento has stabilized. The note frequency is exactly the same as without portamento. I don’t understand how different dividers can affect the timbre itself if the oscillator reproduces exactly the same frequency. But something about the sound is clearly different.

I'd need to test this myself, but, I would wager that my description of this effect as "interference" is a good description from my memory. Likewise with the effects at high notes. There is "an effect", but, the reasons for it can't be so easily determined based upon the available evidence of recordings or the output alone.

Ideally we'd have the inner workings of the chip via de-potting or de-capping the IC. Even then, it would take a massive amount of work to identify all the circuits and there would be very little certainty at a glance.

[aciddose]

With the portamento timbre effect, we shouldn't make any assumptions. We can't be certain whether the portamento is computed by the CPU or whether some other chip handles it according to the parameter data stored in registers (?). We also can't draw any direct conclusions from the effect alone apart from that we've observed an apparent change in timbre. These sorts of effects are referred to as "emergent" phenomena. They might not even exist if you isolate one aspect of the system such as the oscillator IC alone. The total effect may not occur in the same way or at all without many components of the complete system interacting together to produce it.

[TechnoManiac]

From the flac, I've isolated one cycle .... so this could in fact imply either an 8-bit DAC, or that the ratio is near 1/256~ for some other reason.

I looked at what the wave looks like on the descent and on the straight line, at the top of the wave... when the line is parallel to the X axis. The background sound seems to be of the same amplitude. Although if this were quantization noise, then it would be maximally manifested on the line at an angle, and it should not be there at all on the straight line, since there is no difference in levels. Or I'm wrong?

And besides, the shape of the noise signal in different sections (both on the ascent, and on the descent, and on the straight line) resembles a parallel falling saw that is simply quieter, but with a constant amplitude.

Screen:
https://i.ibb.co/gVmLrkZ/step-a-Juno-ramp.jpg
https://i.ibb.co/MpNB1Sb/step-a-Juno-ramp2.jpg

[aciddose]

https://en.wikipedia.org/wiki/Differential_nonlinearity
https://en.wikipedia.org/wiki/Integral_nonlinearity

The ~255 ratio may simply be error and actually be 256. It could also be INL from a DAC, which typically takes on these sorts of shapes.

I looked at what the wave looks like on the descent and on the straight line, at the top of the wave... when the line is parallel to the X axis. The background sound seems to be of the same amplitude. Although if this were quantization noise, then it would be maximally manifested on the line at an angle, and it should not be there at all on the straight line, since there is no difference in levels. Or I'm wrong?
Screen:

Yes that's just the high-pass filters in the signal path. Some aren't even part of the synthesizer but rather your audio interface. That's why to really measure things requires completely isolating elements of the circuit and measuring them correctly with high precision equipment.


Here is a fair example image:


What we're seeing here could be one or multiple of:

1. Integral or Differential Non-linearity from a DAC (error in data vs. voltage output.)
2. Cross-interference, pulses transmitted from other parts of the chip such as earlier higher-frequency signal's edges.
3. Power-supply or other noise and interference being contaminated onto the power supply rails.
4. High-frequency electromagnetic interference from the CPU or any other digital components, both radio-transmission and electrical transmission.

The only really solid observation we have is the precise frequency ratio ... but this is only one measurement and I'd typically prefer to take at least ten to verify if that ratio can change or not.

[aciddose]

Incidentally again, if you zoom into 1 - 10 MHz of the output of a "white noise generator" it looks exactly like this too

[TechnoManiac]

Yes that's just the high-pass filters in the signal path. Some aren't even part of the synthesizer but rather your audio interface.

I don't agree. This is exactly the high-frequency background tone we are talking about (not noise, namely tone, which depends on the frequency of the main pitch). On low notes it is clear as a saw. On high notes it is minimized and becomes round. If it were quantization noise, it would probably be of a constant shape on different notes (but I'm not sure about that). And also it would not manifest itself in those areas where there is no difference in signal levels, that is, on a straight line parallel to the X axis (but I'm not sure about that). If it were a background interference from the CPU or Power-supply, it would not be corresponding to the note (pitch).

This is the only reason left))

Cross-interference, pulses transmitted from other parts of the chip such as earlier higher-frequency signal's edges.

[aciddose]

If it were a background interference from the CPU or Power-supply, it would not be corresponding to the note (pitch).

That's correct, but, you're assuming a lot. Yes, this may be INL from a DAC, but, we're not actually looking at the quantization itself because any "steps" of the wave are buried below this and other noise and interference. So the only solid evidence we have is that there is a strong peak every N*256 (?) codes. We don't know the value of "N". It's hard to say exactly how they handled things internally ... but I am curious whether they'd produce 6 different identical DAC circuits. Also we know for a fact that there is interference present between voices at a lower level.

The absolute minimum frequency I get with the DCO env is:
48000/16904 = 12000000/4226000. It seems to be above C# somewhere but I haven't played with fine-tuning. C and C# produce the same lower limit frequency while D is higher at a different pitch.

[TechnoManiac]

So the only solid evidence we have is that there is a strong peak every N*256 (?) codes.

Can you take a screenshot in different places? I can’t quite see on the wave exactly where this peak is...

[aciddose]

I mean, in terms of INL there is a delta peak where the error hits a maximum extent and then goes in the opposite direction. So as in the 8-bit DAC INL example screenshot, and my image I've already posted of the .flac sample I took. I'm thinking in terms of integrals/deltas when I say "peak" or "spike" or "impulse". Not a frequency/spectral peak (probably what most people would think at first.)

It's quite common for most DACs to have an INL pattern trimmed depending upon the purpose the DAC was designed for. Some minimize DNL, some minimize peak INL, some balance things to minimize how audible the error is. We can say pretty much for certain that most likely, if this is INL from a DAC that it has more than 8-bits, since the INL seems to hit a peak every N*256 codes ... at least the ratio of the full ramp waveform cycle's extent is approximately 1/256 to these higher frequency oscillations.

It also appears to me as if the INL is waaaay over 1 LSB.

I know it's a pain to isolate the oscillator IC, because I've already done that and measured it years ago. At the time I gave up on trying to identify quantization, because the INL (or whatever it is) looks identical with the IC in isolation with a pin lifted. To really measure this accurately I'd need to pull the whole chip and build a test circuit ... and even then based upon my experience I don't feel confident I'd be able to get a nice DC-accurate "staircase" shape measured. It just seems like a huge and very frustrating waste of time to me.