VOGONS


First post, by esher

User metadata
Rank Newbie
Rank
Newbie

Hi! Last year after modding my MU-128, SC-88 Pro and Yamaha AN200 with SPDIF output, i challenge myself to mod MT-32.
The complexity of the topic is that 6 channels feed in series to parallel DAC (clean stereo, 4 reverb channels), and then they demuxed to analog into 3 stereo pairs and then mixed to 1 stereo.
I started with DIT4192 and the very first FPGA Gowin 9K. I have lost 4 months in fights with noise and clicks, and only after making prototype pcb i realized that the noise was parasite from tens of wires on a breadboard.

Video of the first clean stereo capture from MT-32 (some clicks are still present, but now this is not big boss to fight with):

https://www.youtube.com/watch?v=VIkrG32c1l0

Upd: The project is published at https://github.com/vetal-esher/MT32-DIT. Will be updated on progress.

UPD: Project is done. All 6 channels were mixed into 1 stereo pair. Need some time to finish the article.

Last edited by esher on 2023-03-09, 20:01. Edited 2 times in total.

Reply 1 of 10, by mattw

User metadata
Rank Oldbie
Rank
Oldbie

indeed, very nice project!

I wonder, anyone with more in-depth knowledge, please, do give ideas - is it possible some simple digital captures from SC-88 with modification @esher made, to be done, which can answer how the ROM samples are processed, because currently to make them sound "properly" not once, but twice in a row RIAA filter is applied, I commented that many times, I believe one of the last times was here:

Re: About Roland Virtual Sound Canvas 3

which I find very strange and I even doubt is correct and what the real hardware is doing.

Also, same question about MT-32, i.e. simple digital captures made with @esher modification that can help evaluate how close the existing emulator is to the real thing, i.e. evaluate how precise is the emulator, any ideas how exactly such test could be performed?

Reply 2 of 10, by esher

User metadata
Rank Newbie
Rank
Newbie

I think that there will be no "digital reference sound" since the final mix in original MT32 is done in analog. Also i think that because of analog mixing, some sound artifacts were hidden behind the op-amp.
But i'll try different ways of digital mixing and interleaving to get as close as possible to "analog reference".

As for Roland SC series, the mod is the cleanest way to get digital reference, as all mixings are done before the DAC.

Reply 4 of 10, by mattw

User metadata
Rank Oldbie
Rank
Oldbie
esher wrote on 2023-02-09, 11:51:

As for Roland SC series, the mod is the cleanest way to get digital reference, as all mixings are done before the DAC.

any idea, which sound, etc is best to be captured? I mean we have all the sounds from the ROM, I posted SC-88 descrambling code here:

Re: About Roland Virtual Sound Canvas 3

and my question is how best we use digital capture, i.e. what note, without any effects applied, etc to get any idea how the ROM samples are processed. I am not expert on synths and that's why I hope my questions is not that unclear. basically, with @esher excellent mod and ability to digitally capture, how we can figure out how to process the ROM sounds in the right way and not how at the moment emulators like this one:

https://github.com/skjelten/emusc

is doing it with the assumption (without any proof) of doing RIAA filtering 2 times in a row. I think using @esher digital mod to help proper emulator to be made is one of the best ways to preserve those excellent synths forever.

Reply 5 of 10, by esher

User metadata
Rank Newbie
Rank
Newbie
mattw wrote on 2023-02-09, 18:00:
any idea, which sound, etc is best to be captured? I mean we have all the sounds from the ROM, I posted SC-88 descrambling code […]
Show full quote

any idea, which sound, etc is best to be captured? I mean we have all the sounds from the ROM, I posted SC-88 descrambling code here:

Re: About Roland Virtual Sound Canvas 3

and my question is how best we use digital capture, i.e. what note, without any effects applied, etc to get any idea how the ROM samples are processed. I am not expert on synths and that's why I hope my questions is not that unclear. basically, with @esher excellent mod and ability to digitally capture, how we can figure out how to process the ROM sounds in the right way and not how at the moment emulators like this one:

https://github.com/skjelten/emusc

is doing it with the assumption (without any proof) of doing RIAA filtering 2 times in a row. I think using @esher digital mod to help proper emulator to be made is one of the best ways to preserve those excellent synths forever.

i think the simplest sample is the best, e.g. square wave.

The SC mod is here https://github.com/vetal-esher/18bit-DIT, but people say that AK4103 is now obsolete and cannot be found anywhere. After finishing MT32-DIT mod, i'll return to 18bit-DIT rework, and may be
my verilog skills will improve and i'll make FPGA version without any DIT chip, just FPGA with all processing inside. Or, if i fail, there is always WM8804 cheap.

Reply 6 of 10, by mattw

User metadata
Rank Oldbie
Rank
Oldbie
esher wrote on 2023-02-09, 19:41:

i think the simplest sample is the best, e.g. square wave.

yes, square wave (or sinus wave), make sense to me too. I found here "Square Wave" midi file for SC-88:

http://www.takuichi.net/hobby/edu/sonic_wave/ … nstrument/midi/

http://www.takuichi.net/hobby/edu/sonic_wave/ … /midi/pc081.mid

even I am not sure if it's the best midi file for such test, because it contains several notes, as far as I can tell when listen to it.

So, please, if you have the hardware make few such captures and post it here. in the mean time I will look at the ROMs and see how to "extract" the same sound(s) - that will allow comparison and I believe definitely can help and give some ideas.

Reply 7 of 10, by esher

User metadata
Rank Newbie
Rank
Newbie

I'll try to capture on Sunday, and there's no problem to record any note at any pitch. As a reference i think it is, no doubt, A4 which has clean 440Hz base.

Reply 8 of 10, by Jo22

User metadata
Rank l33t++
Rank
l33t++
mattw wrote on 2023-02-09, 23:28:
esher wrote on 2023-02-09, 19:41:

i think the simplest sample is the best, e.g. square wave.

yes, square wave (or sinus wave), make sense to me too. I found here "Square Wave" midi file for SC-88:

+1

Square waves make good harmonics in analogue circuits (audio, radio frequency).
So they're good candidates for testing filters.

"Time, it seems, doesn't flow. For some it's fast, for some it's slow.
In what to one race is no time at all, another race can rise and fall..." - The Minstrel

//My video channel//

Reply 9 of 10, by mattw

User metadata
Rank Oldbie
Rank
Oldbie
esher wrote on 2023-02-10, 06:42:

I'll try to capture on Sunday, and there's no problem to record any note at any pitch. As a reference i think it is, no doubt, A4 which has clean 440Hz base.

thanks, from my side I used the code from here:

https://github.com/Kitrinx/SC55_Soundfont

to exact WAV from "square wave" instrument, it's supposed to be "E3" note and RIAA filtering is applied 2 times to it (the code can be patched and the filtering be removed, if that is necessary). Anyway, let we see with help from others here, if we can figure out what processing is applied to the original sample. I am attaching the file.

Reply 10 of 10, by esher

User metadata
Rank Newbie
Rank
Newbie

Here is the pc081.mid recorded via SC88 SPDIF: download/file.php?mode=view&id=157045
About E3 i tried record then realized that the default settings for square wave can be different on different devices (reverb, chorus, ADSR, etc), so the only way to be in equal conditions is recording the same midi.