What type of output DAC engine is used in MiniDexed?

[pik33]

Dexed on PC has 3 DAC engines to select from: Mark1, Modern 24bit and OPL. Which one of them is used in MiniDexed?

[dcoredump]

Modern is used. I never heared any difference between them.

If someone can convince me that there really are audible differences, I would put the other engines into Synth_Dexed.

The question is of course with which variant of the DX7 one compares...

[probonopd]

Thank you very much. Build is running. Fingers crossed! 👍

[probonopd]

...and we have a build:
MiniDexed_2023-03-21-4379b6f

Since we now have many experts who can tell apart the sound of the various engines, can someone please confirm that it works as intended now?

Everyone who has contributed to this: Thanks for the great teamwork to pull this off!

[Banana71]

I have tested Modern and Mark I with success. With Mark I, Algorythms 4 and 6 sound as expected. Overall, Mark I sounds airier and more open to me, but has this click percentage with hard attacks. If I have time tomorrow I'll do a direct comparison of the engines.

Thank you to everyone involved

[donluca]

Hats off to @dcoredump , he did it!

Just tried and it now sounds correctly.

[dcoredump]

Nice! Great that it is working now. Thanks to all who have patiently tested.

[Banana71]

Now comes the disappointment
The engines Mark I and OPL only work to a limited extent with the RPi2. Up to 6 TG's can be played with Mark I and only up to 4 TG's will work with the OPL Engine. If you exceed the limit, you get bit cracklers.
All 8TG's work with an RPi3, but there are compromises in the sound. What I described as airy yesterday is probably just noise. :-(
In the case of sounds with a short attack time (EG Rate 1 greater than 70), there are clicking noises). With the OPL Engine, it's actually more of a smacking noise. The only advantages I see with Engine Mark I are that algorithms 4 and 6 work.
I will continue to work with Engine Modern for the time being.

[dcoredump]

Sorry, ~5% for 4 simultaniously played notes... but you are right: this sums up....

[donluca]

Is it possible that the Modern engine is still working in the background while using the others?

I remember the enumeration for the engines was first Modern, then Mark I and then OPL and, strangely enough, OPL is the one that's more CPU intensive, while in theory it should be the less demanding.

Just a random thought, because it's a strange coincidence.

[dcoredump]

Is it possible that the Modern engine is still working in the background while using the others?

Hm, not from my understanding. Let's take a look at the calls:

1. getSamples() in minidexed.cpp finally calls getSamples() in dexed.cpp (inside Synth_Dexed library).
2. From there voices[note].dx7_note->compute(audiobuf.get(), lfovalue, lfodelay, &controllers); is called (from dx7note.cpp) for each note.
3. inside dx7note.cpp the compute() function calls ctrls->core->render(buf, params_, algorithm_, fb_buf_, fb_shift_);
   The pointer to core will be adjusted when changing the engine. The other engines are in the memory, but their render() function is no longer addressed. In this respect I think that everything runs correctly.

Also, before changing the address for the used engine, panic() is called, which immediately terminates all currently played notes.

I remember the enumeration for the engines was first Modern, then Mark I and then OPL and, strangely enough, OPL is the one that's more CPU intensive, while in theory it should be the less demanding.

Just a random thought, because it's a strange coincidence.

Yes, that does sound kind of strange, only I absolutely can't explain it. You could try it out by changing the order of the engines inside Synth_Dexed.

Maybe I don't see the error because I'm blind. But in Dexed it is done in the same way...

[probonopd]

Has anyone experimented with Dexed on the RPi to compare its CPU load for the various algorithms?

[donluca]

@probonopd that would be interesting, I don't think I have any spare SD cards to test it though... maybe @Banana71 can try it out on Armbian?

@dcoredump maybe at some point I'll give it a spin to see what happens, right now I really don't have any free time :(

[dcoredump]

After some reading I made a small and simple change to Synth_Dexed - but I have no idea if this will help. Can anyone try the latest commit to Synth_Dexed (18fee6cbe6)?

[donluca]

Unfortunately I'm not at home, I had to get a small vacation and I'm away from my MiniDexed.

@Banana71 can you test this out?
I can make a build for you if needed (I need to know which version of RPi you own)

[probonopd]

Updated it in #443, so hopefully we'll have builds for all models soon.

[probonopd]

MiniDexed_2023-05-04-a680e19

[diyelectromusic]

What was the end result of this - do we know?

Also, @dcoredump why are all the engine assignments in setEngineType cast to (EngineMkI *) - shouldn't they be (FmCore *)?

Kevin

[dcoredump]

Hi @diyelectromusic,

sorry for the late answer - I was on vacation. Yes, I think you are right. I fixed this and some other stuff at 54e543143f and 22a46adcce. So the last one is HEAD.

Regards, Holger