Is Rack minBLEP implementation correct?

leodardo · June 30, 2021, 4:29pm

Please take a look at the issue I submitted.

MinBLEP table code may not be correct

opened 07:59AM - 29 Jun 21 UTC

I suspect there is something wrong in the minblep table generation code. I open…ed Befaco EvenVCO, which uses Rack built-in minblep code. I printed the values from the table and this thing comes out (see screenshot). As you can see, there is a little step which I do not expect from the minBLEP step. I'd rather expect the ripples to oscillate around 1, which it does not. ![image](https://user-images.githubusercontent.com/20128168/123759684-6c772480-d8c0-11eb-8710-04c086817f01.png) The minblep template uses 16 and 32 as Z and O: dsp::MinBlepGenerator<16, 32> triSquareMinBlep; ### Details Operating system and version: Linux 5.4.0-77-generic #86~18.04.1-Ubuntu SMP Fri Jun 18 01:23:22 UTC 2021 x86_64 Rack version (vX.Y.Z):VCV Rack v1.dev.e334902 Hardware relevant to your issue (e.g. graphic card model, audio/MIDI device): N/A ### Summary ### Steps to reproduce Oepn EvenVCO.cpp and edit the constructor as follows. Then launch Rack, open the module and copy from the command line into a Matlab/Octave variable for plotting. EvenVCO() { config(NUM_PARAMS, NUM_INPUTS, NUM_OUTPUTS); configParam(OCTAVE_PARAM, -5.0, 4.0, 0.0, "Octave", "'", 0.5); configParam(TUNE_PARAM, -7.0, 7.0, 0.0, "Tune", " semitones"); configParam(PWM_PARAM, -1.0, 1.0, 0.0, "Pulse width"); printf("MINBLEP TABLE WITH z=16, o=32\n"); for (int i = 0; i <= 2*16*32-4; i+=4) { printf("%.9G, %.9G, %.9G, %.9G, ...\n", triSquareMinBlep.impulse[i], triSquareMinBlep.impulse[i+1], triSquareMinBlep.impulse[i+2], triSquareMinBlep.impulse[i+3]); } printf("%f\n", triSquareMinBlep.impulse[2*16*32]); }

I am not sure that the current minBLEP table is correct. Using same parameters and implementing it the proper way in Matlab does not yield the same result. Beware of this.

Squinky · June 30, 2021, 8:45pm

It seems to work fine. I did log a bug ages ago that it has a DC offset. Could you be more specific about your issue?

Squinky · July 2, 2021, 4:57am

Oh, sorry, I didn’t read your issue. Yes, I see your illustration. It’s not clear to me it that’s a “step”, or if it’s a small DC offset which is then windowed. I can tell you from using it that is does a very good job ad getting rid of aliasing at high frequencies. I guess one could quibble about the phase of the highs, but it’s quite effective, and much better than most other VCO implementations in VCV.

The only issue I’ve observed with it is a DC offset, which is easily removed.

I’ve looked at this in depth quite a bit over here in this demo repo for VCOs: GitHub - squinkylabs/Demo