Why MuLab is Better!

[syntonica]

I've been developing a VSTi on Mac and testing it out on MuLab, Reaper and Tracktion. I just noticed something odd.

MuLab has a lower noise floor! I've been testing an antialiasing filter, so have been looking righteously at EQs to see how I'm doing. Good cutoff of aliasing, but still struggling with some oscillator noise.

Now, given the exact same code run on each DAW, all as Float, MuLab is a couple of dB quieter on the noise from a plain saw than Reaper and just a hair quieter than Tracktion.

I'm not sure why this is, but it's not me. It's the DAW playback engine.

So, well done, Jo!

Anyone else have a reason why MuLab is betterer?

[MuTools]

Some other DAWs add a very very small number to the audio to avoid denormalization. MuLab doesn't do that. Could that explain what you're measuring?

[syntonica]

It's just the noise when my VSTi is playing. When you play a saw, you can see all the tall spikes for the harmonics that are supposed to be there, along with the aliasing marching back down and then some oscillator noise as little -60dB spikes, or less.

I've gotten rid of the aliasing and I think I've done my best I can for now to eliminate the noise. Nothing formal, but the difference is obvious in the graphs.

Forgot about normalization, darn it! I'm not sure I turned it off on Reaper. Is that even really a thing anymore? Okay, turned it of and Reaper is still the noisiest of the three.

Face it. You're a supergenius!

[EvilDragon]

Some other DAWs add a very very small number to the audio to avoid denormalization. MuLab doesn't do that. Could that explain what you're measuring?

Preventing denormals is a good thing to do, though

Okay, turned it of and Reaper is still the noisiest of the three.

Might wanna adjust the resampler mode in project settings, assuming it actually gets used in your testing? That, and perhaps see what different track mixing modes do (Project Settings->Advanced).


I wonder, though, how can DAWs have different noise floors (denormalization notwithstanding), if they're all using 64-bit FP internally? 1+1=2, no (minus denormal protection difference, and perhaps different dithering algos)? Especially if what you're doing is a generated waveform. Curious.

[MuTools]

Some other DAWs add a very very small number to the audio to avoid denormalization. MuLab doesn't do that. Could that explain what you're measuring?

Preventing denormals is a good thing to do, though

Yes absolutely, MuLab also avoids denormals, but not by adding a tiny little bit of noise. Sorry for my confusing message.

[syntonica]

Preventing denormals is a good thing to do, though

What you got against us denirmals?

Might wanna adjust the resampler mode in project settings, assuming it actually gets used in your testing? That, and perhaps see what different track mixing modes do (Project Settings->Advanced).

I'll check there and see. I'm still learning DSP and I was surprised to see this happening. I mean, how can you add so much noise summing one note?

I wonder, though, how can DAWs have different noise floors (denormalization notwithstanding), if they're all using 64-bit FP internally? 1+1=2, no (minus denormal protection difference, and perhaps different dithering algos)? Especially if what you're doing is a generated waveform. Curious.

I'm using single-cycle waveforms and having to interpolate for values, so I expect a bit of noise. My test is a mathematically generated saw wave oversampled 4x to reduce noise and some aliasing and then run through a 20khz low pass filter at 8x oversampling just for antialiasing. Everything is done in 64-bit and only converted at the very end to 32-bit when summed to go out. I was trying to see where the noise was coming from--wave generation, DC, filter or DAW.

I guessed that different DAWs were using different dithering or summing methods. I haven't gotten to anything everything how hosts are designed, or even how Core Audio works. I'm dreading just wrapping my VST for an AU.

Thanks for the tips. I never really got on with Reaper, so I don't know the ins and outs of it.

[syntonica]

OK, checked everything. My fader wasn't zeroed in Mulab. Fixed that and now it's only a touch better than Reaper with Tracktion now in a very noticeable lead with all faders properly at zero. Which got me thinking, maybe they cheat by lowering their zero by a few dB?

Nothing I could find in Reaper made any difference in sound quality. Same with running processReplacing() and processReplacingDouble()--no discernable difference in quality.

Anyways, MuLab is STILL the best!

[EvilDragon]

Which got me thinking, maybe they cheat by lowering their zero by a few dB?

They wouldn't be the first ones to do so (E-MU Paris did that)!

[syntonica]

that's not cool!

I suppose that's the easy way to make your final summing come up right. But it's a problem I'm still wrestling with as to the best solution dir my VST. Limiters? Ratios? Let the user figure it out?

I meant this thread as a bit of tongue-in-cheek as both a why is this happening? and some of praise for Jo and all his long suffering hard work. Aside from a couple of minor quibbles where we don't see eye to eye feature-wise, MuLab is like just the essentials of a DAW with no fluff and no fillers.

But if this is the answer to my question--they're cheating!--then it was a bit of a bust in that regard...