Noise/distortion after doing filters with vDSP_deq22 (biquad IIR filter)

Question

I'm working on a DSP class (obj-c++) for Novocaine, but my filters only seem to cause noise/distortion on the signal.

I've posted my full code and coefficients here: https://gist.github.com/2702844 But it basically boils down to:

// Deinterleaving...
// DSP'ing one channel:
NVDSP *handleDSP = [[NVDSP alloc] init];
[handleDSP setSamplingRate:audioManager.samplingRate];
float cornerFrequency = 6000.0f;
float Q = 0.5f;
[handleDSP setHPF:cornerFrequency Q:Q];
[handleDSP applyFilter:audioData length:numFrames];

// DSP other channel in the same way
// Interleaving and sending to audio output (Novocaine block)

See the gist for full code/context.

The coefficients:

2012-05-15 17:54:18.858 nvdsp[700:16703] b0: 0.472029
2012-05-15 17:54:18.859 nvdsp[700:16703] b1: -0.944059
2012-05-15 17:54:18.860 nvdsp[700:16703] b2: 0.472029
2012-05-15 17:54:18.861 nvdsp[700:16703] a1: -0.748175
2012-05-15 17:54:18.861 nvdsp[700:16703] a2: 0.139942

(all divided by a0)

Since I presumed the coefficients are in the order of: { b0/a0, b1/a0, b2/a0, a1/a0, a2/a0 } (see: IIR coefficients for peaking EQ, how to pass them to vDSP_deq22?)

What is causing the distortion/noise (the filters don't work)?

Answer 1

Update: I recommend everyone to use my DSP class I released on github: https://github.com/bartolsthoorn/NVDSP It'll probably save you quite some work.

Got it working, woooo! Long live the japanese: http://objective-audio.jp/2008/02/biquad-filter.html

The applyFilter method had to be:

- (void) applyFilter: (float *)data frames:(NSUInteger)frames {
    /*
     The first two samples of data being passed to vDSP_deq22 have to be initialized from the previous call. So, you'd want to hold onto a float buffer and feed the tailing two samples after a vDSP_deq22 call back to the front of that array for the next time you call. (Alex Wiltschko)
     */

    // Thanks a lot to: http://objective-audio.jp/2008/02/biquad-filter.html

    // Provide buffer for processing
    float *tInputBuffer = (float*) malloc((frames + 2) * sizeof(float));
    float *tOutputBuffer = (float*) malloc((frames + 2) * sizeof(float));

    // Copy the data
    memcpy(tInputBuffer, gInputKeepBuffer, 2 * sizeof(float));
    memcpy(tOutputBuffer, gOutputKeepBuffer, 2 * sizeof(float));
    memcpy(&(tInputBuffer[2]), data, frames * sizeof(float));

    // Do the processing
    vDSP_deq22(tInputBuffer, 1, coefficients, tOutputBuffer, 1, frames);

    // Copy the data
    memcpy(data, tOutputBuffer, frames * sizeof(float));
    memcpy(gInputKeepBuffer, &(tInputBuffer[frames]), 2 * sizeof(float));
    memcpy(gOutputKeepBuffer, &(tOutputBuffer[frames]), 2 * sizeof(float));

    free(tInputBuffer);
    free(tOutputBuffer);
}

Full class: https://github.com/bartolsthoorn/NVDSP

Answer 2

Copy the data with:

memcpy(data, tOutputBuffer+2, frames * sizeof(float));

And it will work