Efficient IIR filter bank design? about esp-dsp

[drewandre]

It has been a long time since I've worked on this project but I pulled up the function where I averaged the filter output. Hopefully this helps.

void calculate_iir_filter_bank() {
  float avg, x;
  for (int i = 0; i < NUM_AUDIO_ANALYSIS_BANDS; i++) { // NUM_AUDIO_ANALYSIS_BANDS was 7 in my case to match MSGEQ7
    avg = 0.0f;

    dsps_biquad_f32(buffer, filter_output, N_SAMPLES, filter_coefficients[i], filter_delay_lines[i]);

    for (int t = 0; t < N_SAMPLES; t++)
    {
      x = fabs(filter_output[t]) * 4.0f;
      avg += x;
    }
    avg /= (float)N_SAMPLES;

    avg *= filters_amp_adjustments[i];

    if (avg > 1) {
      avg = 1;
    }

    averages[i] = (SMOOTHING_FACTOR * averages[i]) + ((1 - SMOOTHING_FACTOR) * avg);
  }
}

[dmitry1945]

Hi @drewandre
For this purpose better if you will use IIR filter bank.
You can look to the esp-dsp/examples/iir.

There you will find how to generate a single IIR filter:
ret = dsps_biquad_gen_lpf_f32(coeffs_lpf, freq, qFactor);

The frequency is just: your_frequency/sample_frequency. In your case it will be 163/44100, and so on. You will have 7 filters.
The Q factor you have to play with this parameter.
In this example you can tune your filters.

At the output of your filters you make:
average[i] = K*average[i] + (1 - K)*absf(output[i]);

Where:
i - bank number from 0 to 6
K - smoothing coefficient 0.9..0.9999, 0.9999 - very smooth.
output[i] - output of i'th filter.
average[i] - result of your equalizer for i'th bank
absf - module operation.

Regards,
Dmitry

[drewandre]

@dmitry1945 thanks, this was very helpful!

[generic-beat-detector]

@dmitry1945

Hello sir. I applaud the effort behind this wonderful project!

I'd like some clarification please -- assuming the following for a single filter instance:

dsps_biquad_f32(x, y, N, coeffs_lpf, w_lpf);

How did the output y (array of floats) compute to output[i] (single float value)? I'm I missing something?

Regards.