1
0
mirror of https://github.com/f4exb/sdrangel.git synced 2024-11-04 16:01:14 -05:00

Implement variable cutoff frequency for audio filter

This commit is contained in:
f4exb 2019-02-17 01:31:59 +01:00
parent 7a16ccff06
commit 6d4cb53eb6
2 changed files with 178 additions and 6 deletions

View File

@ -14,11 +14,16 @@
// along with this program. If not, see <http://www.gnu.org/licenses/>. //
///////////////////////////////////////////////////////////////////////////////////
#include <math.h>
#include <algorithm>
#include <QDebug>
#include "audiofilter.h"
// f(-3dB) = 3.6 kHz @ 48000 Hz SR (w = 0.0375):
const float AudioFilter::m_lpa[3] = {1.0, 1.392667E+00, -5.474446E-01};
const float AudioFilter::m_lpb[3] = {3.869430E-02, 7.738860E-02, 3.869430E-02};
// f(-3dB) = 300 Hz @ 8000 Hz SR (w = 0.075):
// f(-3dB) = 300 Hz @ 8000 Hz SR (w = 0.0375):
const float AudioFilter::m_hpa[3] = {1.000000e+00, 1.667871e+00, -7.156964e-01};
const float AudioFilter::m_hpb[3] = {8.459039e-01, -1.691760e+00, 8.459039e-01};
@ -26,12 +31,166 @@ AudioFilter::AudioFilter() :
m_filterLP(m_lpa, m_lpb),
m_filterHP(m_hpa, m_hpb),
m_useHP(false)
{
}
{}
AudioFilter::~AudioFilter()
{}
void AudioFilter::setDecimFilters(int sr, uint32_t decim)
{
int downSR = sr / (decim == 0 ? 1 : decim);
double fcH = (0.45 * downSR) / (sr <= 0 ? 1 : sr); // high cut frequency normalized to SR
double fcL = 300.0 / downSR; // low cut frequency normalized to downsampled SR
calculate2(false, fcH, m_lpva, m_lpvb);
calculate2(true, fcL, m_hpva, m_hpvb);
m_filterLP.setCoeffs(m_lpva, m_lpvb);
m_filterHP.setCoeffs(m_hpva, m_hpvb);
}
void AudioFilter::calculate2(bool highPass, double fc, float *va, float *vb)
{
double a[22], b[22];
cheby(highPass, fc, 0.5, 2, a, b); // low-pass, 0.5% ripple, 2 pole filter
// Copy to the 2-pole filter coefficients
for (int i=0; i<3; i++) {
vb[i] = a[i];
va[i] = b[i];
}
va[0] = 1.0;
qDebug() << "AudioFilter::calculate2:"
<< " highPass: " << highPass
<< " fc: " << fc
<< " a0: " << va[0]
<< " a1: " << va[1]
<< " a2: " << va[2]
<< " b0: " << vb[0]
<< " b1: " << vb[1]
<< " b2: " << vb[2];
}
/*
* Adapted from BASIC program in table 20-4 of
* https://www.analog.com/media/en/technical-documentation/dsp-book/dsp_book_Ch20.pdf
*/
void AudioFilter::cheby(bool highPass, double fc, float pr, int np, double *a, double *b)
{
double a0, a1, a2, b1, b2;
double ta[22], tb[22];
std::fill(a, a+22, 0.0);
std::fill(b, b+22, 0.0);
a[2] = 1.0;
b[2] = 1.0;
for (int p = 1; p <= np/2; p++)
{
cheby_sub(highPass, fc, pr, np, p, a0, a1, a2, b1, b2);
// Add coefficients to the cascade
for (int i=0; i<22; i++)
{
ta[i] = a[i];
tb[i] = b[i];
}
for (int i=2; i<22; i++)
{
a[i] = a0*ta[i] + a1*ta[i-1] + a2*ta[i-2];
b[i] = tb[i] - b1*tb[i-1] - b2*tb[i-2];
}
}
// Finish combining coefficients
b[2] = 0;
for (int i=0; i<20; i++)
{
a[i] = a[i+2];
b[i] = -b[i+2];
}
// Normalize the gain
double sa = 0.0;
double sb = 0.0;
for (int i=0; i<20; i++)
{
if (highPass)
{
sa += i%2 == 0 ? a[i] : -a[i];
sb += i%2 == 0 ? b[i] : -b[i];
}
else
{
sa += a[i];
sb += b[i];
}
}
double gain = sa/(1.0 -sb);
for (int i=0; i<20; i++) {
a[i] /= gain;
}
}
/*
* Adapted from BASIC subroutine in table 20-5 of
* https://www.analog.com/media/en/technical-documentation/dsp-book/dsp_book_Ch20.pdf
*/
void AudioFilter::cheby_sub(bool highPass, double fc, float pr, int np, int stage, double& a0, double& a1, double& a2, double& b1, double& b2)
{
double rp = -cos((M_PI/(np*2)) + (stage-1)*(M_PI/np));
double ip = sin((M_PI/(np*2)) + (stage-1)*(M_PI/np));
// Warp from a circle to an ellipse
double esx = 100.0 / (100.0 - pr);
double es = sqrt(esx*esx -1.0);
double vx = (1.0/np) * log((1.0/es) + sqrt((1.0/(es*es)) + 1.0));
double kx = (1.0/np) * log((1.0/es) + sqrt((1.0/(es*es)) - 1.0));
kx = (exp(kx) + exp(-kx))/2.0;
rp = rp * ((exp(vx) - exp(-vx))/2.0) / kx;
ip = ip * ((exp(vx) + exp(-vx))/2.0) / kx;
double t = 2.0 * tan(0.5);
double w = 2.0 * M_PI * fc;
double m = rp*rp + ip*ip;
double d = 4.0 - 4.0*rp*t + m*t*t;
double x0 = (t*t)/d;
double x1 = (2.0*t*t)/d;
double x2 = (t*t)/d;
double y1 = (8.0 - 2.0*m*t*t)/d;
double y2 = (-4.0 - 4.0*rp*t - m*t*t)/d;
double k;
if (highPass) {
k = -cos(w/2.0 + 0.5) / cos(w/2.0 - 0.5);
} else {
k = sin(0.5 - w/2.0) / sin(0.5 + w/2.0);
}
d = 1.0 + y1*k - y2*k*k;
a0 = (x0 - x1*k + x2*k*k)/d;
a1 = (-2.0*x0*k + x1 + x1*k*k - 2.0*x2*k)/d;
a2 = (x0*k*k - x1*k + x2)/d;
b1 = (2.0*k + y1 + y1*k*k - 2.0*y2*k)/d;
b2 = (-(k*k) - y1*k + y2)/d;
if (highPass)
{
a1 = -a1;
b1 = -b1;
}
}
float AudioFilter::run(const float& sample)
{
return m_useHP ? m_filterLP.run(m_filterHP.run(sample)) : m_filterLP.run(sample);
@ -45,4 +204,4 @@ float AudioFilter::runHP(const float& sample)
float AudioFilter::runLP(const float& sample)
{
return m_filterLP.run(sample);
}
}

View File

@ -21,7 +21,7 @@
#include "dsp/iirfilter.h"
/**
* This is a 2 pole lowpass Chebyshev (recursive) filter at fc=0.075 using coefficients found in table 20-1 of
* By default this is a 2 pole lowpass Chebyshev (recursive) filter at fc=0.075 using coefficients found in table 20-1 of
* http://www.analog.com/media/en/technical-documentation/dsp-book/dsp_book_Ch20.pdf
*
* At the interpolated sampling frequency of 48 kHz the -3 dB corner is at 48 * .075 = 3.6 kHz which is perfect for voice
@ -36,6 +36,8 @@
*
* This one works directly with floats
*
* It can be generalized using the program found in tables 20-4 and 20-5 of the same document. This form is used as a
* decimation filter and can be set with the setDecimFilters method
*/
class SDRBASE_API AudioFilter {
@ -45,18 +47,29 @@ public:
void useHP(bool useHP) { m_useHP = useHP; }
bool usesHP() const { return m_useHP; }
void setDecimFilters(int sr, uint32_t decim);
float run(const float& sample);
float runHP(const float& sample);
float runLP(const float& sample);
private:
void calculate2(bool highPass, double fc, float *a, float *b); // two pole Chebyshev calculation
void cheby(bool highPass, double fc, float pr, int np, double *a, double *b);
void cheby_sub(bool highPass, double fc, float pr, int np, int stage,
double& a0, double& a1, double& a2, double& b1, double& b2);
IIRFilter<float, 2> m_filterLP;
IIRFilter<float, 2> m_filterHP;
bool m_useHP;
float m_lpva[3];
float m_lpvb[3];
float m_hpva[3];
float m_hpvb[3];
static const float m_lpa[3];
static const float m_lpb[3];
static const float m_hpa[3];
static const float m_hpb[3];
};
#endif // _SDRBASE_AUDIO_AUDIOFILTER_H_
#endif // _SDRBASE_AUDIO_AUDIOFILTER_H_