A 440Hz square wave from my MT-32 (v1.07) looks like this:

Somewhere in between the two, I think!

Thank you Sfryers for this detail ! It clearly shows that even the real MT32 has a DC filter !

I wasn't accurate with my previous image because it just served to show the DC filter effect I mean overall. I have now attached two plots below, one from my D20, one from Munt, both are square waves (max cutoff) at 440 Hz. As you can see, the D20 wave is pretty identical to the one you shown from the MT32, while Munt wave is perfectly flat. This simply tells that Munt is NOT emulating the DC filter !

I must specify that my interest for this matter is not merely picky or academical (after all a DC filter has no audible effect). Since I have plans to write a LA32 emulator VST NOT based on Munt (which *imhhho* still adopts a lot of guessworks), I am intrigued by this DC filter. I wonder if it is really an analog HPF put before the very audio output, or rather it is somehow done digitally and is part of the way analog waveforms are so creatively produced...
Since no real LPF for TVF is used but the filtered squared waveform is assembled using cosine edges, I wonder if the flat square wave portions are actually flat and then filtered in analog domain, or rather they are really curved to start with... Another guess is that the DC filter is needed for PWM, because that way assembling waves with different pulse width is simpler, maybe they are even rendered as positive waves, and the DC filter performs the centering at 0 V.
After all, if you press a key and sample the result and inspect it, anyone can easily see that the signal produced starts with a DC bias and soon stabilizes, which is the typical behavior of a DC filter. If the square waves with asymmetrical PW were synthesized as centered already (without DC i.e sum of all cycle samples == 0), we would NOT see this behavior, and no DC filter would be needed...

There's a circuit schematic of the MT-32 'new' on page 9 of the official service manual: https://dosdays.co.uk/media/roland/mt-32/rola … vice-manual.pdf

There do appear to be a number of capacitors and resistors in the analogue signal path, after the op-amp stages and immediately before the output jack sockets. I'm certainly no expert on audio circuits, but it looks to me as though these are acting as filters.

I am not expert with analog circuits either sadly. Thx for the link btw. Whatever it is, I must remove the effect of this filter, to better examine and rev-eng the waveforms. Luckily I have all means to do that with some phase shift curve.

Anyway, if somebody else could confirm those analog circuits act as DC blocker and why Munt is not emulating them (just out of curiosity !) he is welcome 😀

BadHellie wrote on 2025-06-21, 18:49:

Anyway, if somebody else could confirm those analog circuits act as DC blocker and why Munt is not emulating them (just out of curiosity !) he is welcome 😀

I don't know the reason but, if I had to guess, since the DC filter is supposed to be inaudible, as you mentioned, then it possibly was not seen as a priority (or even considered necessary at all) to emulate that characteristic. Again, that is just speculation on my part.

You could ask sergm here or on the MT-32 Github ( https://github.com/munt/munt ).

Indeed, chances are it was simply ignored because of little purpose. But for me it seems a clue of something nonetheless. I will ask Sergm out of curiosity, thx

Could also be the PCM54 DAC conversion process, which may or may not be emulated correctly in MUNT (I know the SC-55 emu does not reproduce this).

Yeah I was thinking something alike out of mere intuition. I inspected the DC filter response of my D20, it is pretty identical to an 1-pole HPF with cutoff set at about 10 Hz or so. The strange thing is that this filter (or a resulting combination with other electronics) has a slight ringing somewhere between 10 and 15 Khz. This can be both appreciated on a spectral analyzer and inspecting the waveform. As a fact, a plain square wave with max cutoff (cut=100 + TVF env maximally opened) and RES=0 still presents some resonance. I thought it was resonance simulation which was never completely abolished even when zero; that turned out an effect of the filter/hardware instead. As a proof, the same ringing and DC filtering are also identically noticeable with PCM waveforms !

However, to elaborate further, I am now pretty sure the DC filter is required to stabilize the analog waveforms when Pulse With is >0. It is evident that in these cases, asymmetrical waveforms are always centered around zero (i.e they have DC) and soon stabilize after few cycles when pressing a key, by effect of the DC filter. Without it, the output when PW >0 would have a DC bias.

Interesting. You could check the service manuals for the D-10, D-20, D-110 and MT-32 to compare output stages, much easier to do. Although the MT-32 has 3-4 revisions, worth nothing (MT-32 1, MT-32 2, CM-32L, E-30 keyboard). MT-32 first gen is probably the accurate comparison with the D-series, since there are no revisions of the D synths that I'm aware of.

All do use the same DAC though, the PCM54. Of course, MUNT may not correctly emulate this, but the hardwares should all have that in common at least.

EDIT: Might be some additional variants of the CM that use the PCM55 DAC as well.. but that caused bugs.