RBJ Peaking hi frecuency Problem

tor · 29 Oct 2010

Do you use the OS-kit? [old link]

tor · 30 Oct 2010

I was wondering if the same thing is possible without over sampling. David W. Gunnes, Ojas S. Chauhan and Sophocles J. Orfanidis has written about the subject. But I have yet to wrap my head around it properly. I am not a educated person so I lack some mathematical skills. But it is really only a matter at 44100 sample rate. At higher samplerates the warping is above the audible range. For critical recordings I asume most people operate at higest possible SR. So I went for a double precision floating solution in my EQ instead. I have shared my EQ here:

RBJ Filters ported UPDATED...

Dell · 31 Oct 2010

Hi.

The lift is based on having two frequencies working together.

One is tuned higher than the other with each one having it's own Q and, dB settings.

This is why the lift resembles so.

It was more than likely designed that way to give the 20 kHz a bit of an accent in the sound. I would estimate the first frequency is around 22 kHz.

Cheers!

datsound · 3 Nov 2010

It would be interesting to see how the peak profile develops through 14, 16 and 18 KHz, else it difficult to see what it's designed to do, maybe it's because frequencies over 16 are much less important to human ears, in fact many audio producers can't hear much over that level from using loudspeakers a lot.

So probably the filter is not designed for an equaliser or it is in oversight imported from the lowpass etc, you should try some other filters there are quite a lot of peak ones to see if you can get a better match, you can also try and see the code is there is a reference to frequencies above 16 and a formula next to it, and you can try a forum with people that do more filter programming. If the equaliser has fixed sliders, you can just use a different notch for the higher frequencies.

Disco_Steve · 3 Nov 2010

its mainly just Nyquist showing his head. the filter tends to zero as you approach infinite frequency but for sampled audio infinite frequency is brought right down to the finite Nyquist rate. so the filter must fall to zero. all digital filters which peek and then tend to zero do this. the two main ways to overcome it is to, add another peeking filter (but will still be wrong at very high frequencies, or add a high shelf filter so that at the Nyquist frequency the gain is as it would be if an infinite sample rate were used.

or use over sampling to move the Nyquist frequency out of the important range.

I personalty use , well that's a secret 😉

datsound · 4 Nov 2010

the code is pretty straightforwards, there should be a line in there where you can make a fast roll off over say 30khz.

sin = sin1(0.49*freq); cos = cos1(0.49*freq);

in this line you could be able to edit the freq value so that the formula thinks that freq is rolled offf when it is near infinity ... well i dont know but if you keep an eye on your specroscope and edit all the parameters in the rbj code, say devide each by 2, one at a time, well at least you would know what it is doing.

// Ins and Outs... polyin in; polyin freq; polyin q; polyin A; polyout out;

// Previous ins/outs storage and filter coefs... float in1,in2,out1,out2; float a0,a1,a2,b0,b1,b2;

// Get sin and cos values... float sin,cos,alpha; sin = sin1(0.49*freq); cos = cos1(0.49*freq); alpha = sin/(2*q);

// Calculate filter coefs... a0 = 1.0 + alpha/A; a1 = -2.0*cos; a2 = 1.0 - alpha/A; b0 = 1.0 + alpha*A; b1 = -2.0*cos; b2 = 1.0 - alpha*A;

//Scale filter coefs... a1 = a1/a0; a2 = a2/a0; b0 = b0/a0; b1 = b1/a0; b2 = b2/a0;

// Get our output... out = in*b0 + in1*b1 + in2*b2 - out1*a1 - out2*a2;

// Store old values... out2 = out1; out1 = out; in2 = in1; in1 = in;

edit... http://www.musicdsp.org/files/Audio-EQ-Cookbook.txt