mirror of
https://github.com/f4exb/sdrangel.git
synced 2024-11-21 23:55:13 -05:00
751 lines
21 KiB
C++
751 lines
21 KiB
C++
/* rmatch.c
|
|
|
|
This file is part of a program that implements a Software-Defined Radio.
|
|
|
|
Copyright (C) 2017, 2018, 2022 Warren Pratt, NR0V
|
|
Copyright (C) 2024 Edouard Griffiths, F4EXB Adapted to SDRangel
|
|
|
|
This program is free software; you can redistribute it and/or
|
|
modify it under the terms of the GNU General Public License
|
|
as published by the Free Software Foundation; either version 2
|
|
of the License, or (at your option) any later version.
|
|
|
|
This program is distributed in the hope that it will be useful,
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
GNU General Public License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with this program; if not, write to the Free Software
|
|
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
|
|
|
|
The author can be reached by email at
|
|
|
|
warren@wpratt.com
|
|
|
|
*/
|
|
|
|
#include <chrono>
|
|
#include <thread>
|
|
|
|
#include "comm.hpp"
|
|
#include "varsamp.hpp"
|
|
#include "rmatch.hpp"
|
|
|
|
namespace WDSP {
|
|
|
|
MAV* MAV::create_mav (int ringmin, int ringmax, float nom_value)
|
|
{
|
|
auto *a = new MAV;
|
|
a->ringmin = ringmin;
|
|
a->ringmax = ringmax;
|
|
a->nom_value = nom_value;
|
|
a->ring = new int[a->ringmax];
|
|
a->mask = a->ringmax - 1;
|
|
a->i = 0;
|
|
a->load = 0;
|
|
a->sum = 0;
|
|
return a;
|
|
}
|
|
|
|
void MAV::destroy_mav (MAV *a)
|
|
{
|
|
delete[] a->ring;
|
|
delete a;
|
|
}
|
|
|
|
void MAV::flush_mav (MAV *a)
|
|
{
|
|
memset (a->ring, 0, a->ringmax * sizeof (int));
|
|
a->i = 0;
|
|
a->load = 0;
|
|
a->sum = 0;
|
|
}
|
|
|
|
void MAV::xmav (MAV *a, int input, float* output)
|
|
{
|
|
if (a->load >= a->ringmax)
|
|
a->sum -= a->ring[a->i];
|
|
if (a->load < a->ringmax) a->load++;
|
|
a->ring[a->i] = input;
|
|
a->sum += a->ring[a->i];
|
|
|
|
if (a->load >= a->ringmin)
|
|
*output = (float)a->sum / (float)a->load;
|
|
else
|
|
*output = a->nom_value;
|
|
a->i = (a->i + 1) & a->mask;
|
|
}
|
|
|
|
AAMAV* AAMAV::create_aamav (int ringmin, int ringmax, float nom_ratio)
|
|
{
|
|
auto *a = new AAMAV;
|
|
a->ringmin = ringmin;
|
|
a->ringmax = ringmax;
|
|
a->nom_ratio = nom_ratio;
|
|
a->ring = new int[a->ringmax];
|
|
a->mask = a->ringmax - 1;
|
|
a->i = 0;
|
|
a->load = 0;
|
|
a->pos = 0;
|
|
a->neg = 0;
|
|
return a;
|
|
}
|
|
|
|
void AAMAV::destroy_aamav (AAMAV *a)
|
|
{
|
|
delete[] a->ring;
|
|
delete[] a;
|
|
}
|
|
|
|
void AAMAV::flush_aamav (AAMAV *a)
|
|
{
|
|
memset (a->ring, 0, a->ringmax * sizeof (int));
|
|
a->i = 0;
|
|
a->load = 0;
|
|
a->pos = 0;
|
|
a->neg = 0;
|
|
}
|
|
|
|
void AAMAV::xaamav (AAMAV *a, int input, float* output)
|
|
{
|
|
if (a->load >= a->ringmax)
|
|
{
|
|
if (a->ring[a->i] >= 0)
|
|
a->pos -= a->ring[a->i];
|
|
else
|
|
a->neg += a->ring[a->i];
|
|
}
|
|
if (a->load <= a->ringmax) a->load++;
|
|
a->ring[a->i] = input;
|
|
if (a->ring[a->i] >= 0)
|
|
a->pos += a->ring[a->i];
|
|
else
|
|
a->neg -= a->ring[a->i];
|
|
if (a->load >= a->ringmin)
|
|
*output = (float)a->neg / (float)a->pos;
|
|
else if (a->neg > 0 && a->pos > 0)
|
|
{
|
|
float frac = (float)a->load / (float)a->ringmin;
|
|
*output = (1.0 - frac) * a->nom_ratio + frac * ((float)a->neg / (float)a->pos);
|
|
}
|
|
else
|
|
*output = a->nom_ratio;
|
|
a->i = (a->i + 1) & a->mask;
|
|
}
|
|
|
|
void RMATCH::calc_rmatch (RMATCH *a)
|
|
{
|
|
int m;
|
|
float theta;
|
|
float dtheta;
|
|
int max_ring_insize;
|
|
a->nom_ratio = (float)a->nom_outrate / (float)a->nom_inrate;
|
|
max_ring_insize = (int)(1.0 + (float)a->insize * (1.05 * a->nom_ratio));
|
|
if (a->ringsize < 2 * max_ring_insize) a->ringsize = 2 * max_ring_insize;
|
|
if (a->ringsize < 2 * a->outsize) a->ringsize = 2 * a->outsize;
|
|
a->ring = new float[a->ringsize * 2];
|
|
a->rsize = a->ringsize;
|
|
a->n_ring = a->rsize / 2;
|
|
a->iin = a->rsize / 2;
|
|
a->iout = 0;
|
|
a->resout = new float[max_ring_insize * 2];
|
|
a->v = new VARSAMP(1, a->insize, a->in, a->resout, a->nom_inrate, a->nom_outrate,
|
|
a->fc_high, a->fc_low, a->R, a->gain, a->var, a->varmode);
|
|
a->ffmav = AAMAV::create_aamav (a->ff_ringmin, a->ff_ringmax, a->nom_ratio);
|
|
a->propmav = MAV::create_mav (a->prop_ringmin, a->prop_ringmax, 0.0);
|
|
a->pr_gain = a->prop_gain * 48000.0f / (float)a->nom_outrate; // adjust gain for rate
|
|
a->inv_nom_ratio = (float)a->nom_inrate / (float)a->nom_outrate;
|
|
a->feed_forward = 1.0;
|
|
a->av_deviation = 0.0;
|
|
a->ntslew = (int)(a->tslew * (float) a->nom_outrate);
|
|
if (a->ntslew + 1 > a->rsize / 2) a->ntslew = a->rsize / 2 - 1;
|
|
a->cslew = new float[a->ntslew + 1];
|
|
dtheta = (float) PI / (float) a->ntslew;
|
|
theta = 0.0;
|
|
for (m = 0; m <= a->ntslew; m++)
|
|
{
|
|
a->cslew[m] = 0.5f * (1.0f - cos (theta));
|
|
theta += dtheta;
|
|
}
|
|
a->baux = new float[a->ringsize / 2 * 2];
|
|
a->readsamps = 0;
|
|
a->writesamps = 0;
|
|
a->read_startup = (unsigned int)((float)a->nom_outrate * a->startup_delay);
|
|
a->write_startup = (unsigned int)((float)a->nom_inrate * a->startup_delay);
|
|
a->control_flag = 0;
|
|
// diagnostics
|
|
a->underflows = 0;
|
|
a->overflows = 0;
|
|
}
|
|
|
|
void RMATCH::decalc_rmatch (RMATCH *a)
|
|
{
|
|
delete[] (a->baux);
|
|
delete[] (a->cslew);
|
|
MAV::destroy_mav (a->propmav);
|
|
AAMAV::destroy_aamav (a->ffmav);
|
|
delete a->v;
|
|
delete[] (a->resout);
|
|
delete[] (a->ring);
|
|
}
|
|
|
|
RMATCH* RMATCH::create_rmatch (
|
|
int run, // 0 - input and output calls do nothing; 1 - operates normally
|
|
float* in, // pointer to input buffer
|
|
float* out, // pointer to output buffer
|
|
int insize, // size of input buffer
|
|
int outsize, // size of output buffer
|
|
int nom_inrate, // nominal input samplerate
|
|
int nom_outrate, // nominal output samplerate
|
|
float fc_high, // high cutoff frequency if lower than max
|
|
float fc_low, // low cutoff frequency if higher than zero
|
|
float gain, // gain to be applied during this process
|
|
float startup_delay, // time (seconds) to delay before beginning measurements to control variable resampler
|
|
int auto_ringsize, // 0 specified ringsize is used; 1 ringsize is auto-optimized - FEATURE NOT IMPLEMENTED!!
|
|
int ringsize, // specified ringsize; max ringsize if 'auto' is enabled
|
|
int R, // density factor for varsamp coefficients
|
|
float var, // initial value of variable resampler ratio (value of ~1.0)
|
|
int ffmav_min, // minimum feed-forward moving average size to put full weight on data in the ring
|
|
int ffmav_max, // maximum feed-forward moving average size - MUST BE A POWER OF TWO!
|
|
float ff_alpha, // feed-forward exponential averaging multiplier
|
|
int prop_ringmin, // proportional feedback min moving average ringsize
|
|
int prop_ringmax, // proportional feedback max moving average ringsize - MUST BE A POWER OF TWO!
|
|
float prop_gain, // proportional feedback gain factor
|
|
int varmode, // 0 - use same var for all samples of the buffer; 1 - interpolate from old_var to this var
|
|
float tslew // slew/blend time (seconds)
|
|
)
|
|
{
|
|
auto *a = new RMATCH;
|
|
a->run = run;
|
|
a->in = in;
|
|
a->out = out;
|
|
a->insize = insize;
|
|
a->outsize = outsize;
|
|
a->nom_inrate = nom_inrate;
|
|
a->nom_outrate = nom_outrate;
|
|
a->fc_high = fc_high;
|
|
a->fc_low = fc_low;
|
|
a->gain = gain;
|
|
a->startup_delay = startup_delay;
|
|
a->auto_ringsize = auto_ringsize;
|
|
a->ringsize = ringsize;
|
|
a->R = R;
|
|
a->var = var;
|
|
a->ff_ringmin = ffmav_min;
|
|
a->ff_ringmax = ffmav_max; // must be a power of two
|
|
a->ff_alpha = ff_alpha;
|
|
a->prop_ringmin = prop_ringmin;
|
|
a->prop_ringmax = prop_ringmax; // must be a power of two
|
|
a->prop_gain = prop_gain;
|
|
a->varmode = varmode;
|
|
a->tslew = tslew;
|
|
calc_rmatch(a);
|
|
return a;
|
|
}
|
|
|
|
void RMATCH::destroy_rmatch (RMATCH *a)
|
|
{
|
|
decalc_rmatch (a);
|
|
delete a;
|
|
}
|
|
|
|
void RMATCH::reset_rmatch (RMATCH *a)
|
|
{
|
|
a->run = 0;
|
|
std::this_thread::sleep_for(std::chrono::seconds(10));
|
|
decalc_rmatch(a);
|
|
calc_rmatch (a);
|
|
a->run = 1;
|
|
}
|
|
|
|
void RMATCH::control (RMATCH *a, int change)
|
|
{
|
|
{
|
|
float current_ratio;
|
|
AAMAV::xaamav (a->ffmav, change, ¤t_ratio);
|
|
current_ratio *= a->inv_nom_ratio;
|
|
a->feed_forward = a->ff_alpha * current_ratio + (1.0f - a->ff_alpha) * a->feed_forward;
|
|
}
|
|
{
|
|
int deviation = a->n_ring - a->rsize / 2;
|
|
MAV::xmav (a->propmav, deviation, &a->av_deviation);
|
|
}
|
|
a->var = a->feed_forward - a->pr_gain * a->av_deviation;
|
|
if (a->var > 1.04) a->var = 1.04f;
|
|
if (a->var < 0.96) a->var = 0.96f;
|
|
}
|
|
|
|
void RMATCH::blend (RMATCH *a)
|
|
{
|
|
int i;
|
|
int j;
|
|
for (i = 0, j = a->iout; i <= a->ntslew; i++, j = (j + 1) % a->rsize)
|
|
{
|
|
a->ring[2 * j + 0] = a->cslew[i] * a->ring[2 * j + 0] + (1.0f - a->cslew[i]) * a->baux[2 * i + 0];
|
|
a->ring[2 * j + 1] = a->cslew[i] * a->ring[2 * j + 1] + (1.0f - a->cslew[i]) * a->baux[2 * i + 1];
|
|
}
|
|
}
|
|
|
|
void RMATCH::upslew (RMATCH *a, int newsamps)
|
|
{
|
|
int i;
|
|
int j;
|
|
i = 0;
|
|
j = a->iin;
|
|
while (a->ucnt >= 0 && i < newsamps)
|
|
{
|
|
a->ring[2 * j + 0] *= a->cslew[a->ntslew - a->ucnt];
|
|
a->ring[2 * j + 1] *= a->cslew[a->ntslew - a->ucnt];
|
|
a->ucnt--;
|
|
i++;
|
|
j = (j + 1) % a->rsize;
|
|
}
|
|
}
|
|
|
|
void RMATCH::xrmatchIN (void* b, float* in)
|
|
{
|
|
auto *a = (RMATCH*) b;
|
|
if (a->run == 1)
|
|
{
|
|
int newsamps;
|
|
int first;
|
|
int second;
|
|
int ovfl;
|
|
float var;
|
|
a->v->in = a->in = in;
|
|
|
|
if (!a->force)
|
|
var = a->var;
|
|
else
|
|
var = a->fvar;
|
|
|
|
newsamps = a->v->execute(var);
|
|
a->n_ring += newsamps;
|
|
|
|
if ((ovfl = a->n_ring - a->rsize) > 0)
|
|
{
|
|
a->overflows += 1;
|
|
a->n_ring = a->rsize; //
|
|
|
|
if ((a->ntslew + 1) > (a->rsize - a->iout))
|
|
{
|
|
first = a->rsize - a->iout;
|
|
second = (a->ntslew + 1) - first;
|
|
}
|
|
else
|
|
{
|
|
first = a->ntslew + 1;
|
|
second = 0;
|
|
}
|
|
|
|
std::copy(a->ring + 2 * a->iout, a->ring + 2 * a->iout + first * 2, a->baux);
|
|
std::copy(a->ring, a->ring + second * 2, a->baux + 2 * first);
|
|
a->iout = (a->iout + ovfl) % a->rsize; //
|
|
}
|
|
|
|
if (newsamps > (a->rsize - a->iin))
|
|
{
|
|
first = a->rsize - a->iin;
|
|
second = newsamps - first;
|
|
}
|
|
else
|
|
{
|
|
first = newsamps;
|
|
second = 0;
|
|
}
|
|
|
|
std::copy(a->resout, a->resout + first * 2, a->ring + 2 * a->iin);
|
|
std::copy(a->resout + 2 * first, a->resout + 2 * first + second * 2, a->ring);
|
|
|
|
if (a->ucnt >= 0)
|
|
upslew(a, newsamps);
|
|
|
|
a->iin = (a->iin + newsamps) % a->rsize;
|
|
|
|
if (ovfl > 0)
|
|
blend (a);
|
|
|
|
if (!a->control_flag)
|
|
{
|
|
a->writesamps += a->insize;
|
|
if ((a->readsamps >= a->read_startup) && (a->writesamps >= a->write_startup))
|
|
a->control_flag = 1;
|
|
}
|
|
|
|
if (a->control_flag)
|
|
control (a, a->insize);
|
|
}
|
|
}
|
|
|
|
void RMATCH::dslew (RMATCH *a)
|
|
{
|
|
int i;
|
|
int j;
|
|
int k;
|
|
int n;
|
|
int zeros;
|
|
int first;
|
|
int second;
|
|
if (a->n_ring > a->ntslew + 1)
|
|
{
|
|
i = (a->iout + (a->n_ring - (a->ntslew + 1))) % a->rsize;
|
|
j = a->ntslew;
|
|
k = a->ntslew + 1;
|
|
n = a->n_ring - (a->ntslew + 1);
|
|
}
|
|
else
|
|
{
|
|
i = a->iout;
|
|
j = a->ntslew;
|
|
k = a->n_ring;
|
|
n = 0;
|
|
}
|
|
while (k > 0 && j >= 0)
|
|
{
|
|
if (k == 1)
|
|
{
|
|
a->dlast[0] = a->ring[2 * i + 0];
|
|
a->dlast[1] = a->ring[2 * i + 1];
|
|
}
|
|
a->ring[2 * i + 0] *= a->cslew[j];
|
|
a->ring[2 * i + 1] *= a->cslew[j];
|
|
i = (i + 1) % a->rsize;
|
|
j--;
|
|
k--;
|
|
n++;
|
|
}
|
|
while (j >= 0)
|
|
{
|
|
a->ring[2 * i + 0] = a->dlast[0] * a->cslew[j];
|
|
a->ring[2 * i + 1] = a->dlast[1] * a->cslew[j];
|
|
i = (i + 1) % a->rsize;
|
|
j--;
|
|
n++;
|
|
}
|
|
|
|
if ((zeros = a->outsize - n) > 0) //
|
|
{ //
|
|
if (zeros > a->rsize - i)
|
|
{
|
|
first = a->rsize - i;
|
|
second = zeros - first;
|
|
}
|
|
else
|
|
{
|
|
first = zeros;
|
|
second = 0;
|
|
}
|
|
std::fill(a->ring + 2 * i, a->ring + 2 * i + first * 2, 0);
|
|
std::fill(a->ring, a->ring + second * 2, 0);
|
|
n += zeros;
|
|
}
|
|
a->n_ring = n;
|
|
a->iin = (a->iout + a->n_ring) % a->rsize;
|
|
}
|
|
|
|
|
|
void RMATCH::xrmatchOUT (void* b, float* out)
|
|
{
|
|
auto *a = (RMATCH*) b;
|
|
if (a->run == 1)
|
|
{
|
|
int first;
|
|
int second;
|
|
a->out = out;
|
|
|
|
if (a->n_ring < a->outsize)
|
|
{
|
|
dslew (a);
|
|
a->ucnt = a->ntslew;
|
|
a->underflows += 1;
|
|
}
|
|
|
|
if (a->outsize > (a->rsize - a->iout))
|
|
{
|
|
first = a->rsize - a->iout;
|
|
second = a->outsize - first;
|
|
}
|
|
else
|
|
{
|
|
first = a->outsize;
|
|
second = 0;
|
|
}
|
|
|
|
std::copy(a->ring + 2 * a->iout, a->ring + 2 * a->iout + first * 2, a->out);
|
|
std::copy(a->ring, a->ring + second * 2, a->out + 2 * first);
|
|
a->iout = (a->iout + a->outsize) % a->rsize;
|
|
a->n_ring -= a->outsize;
|
|
a->dlast[0] = a->out[2 * (a->outsize - 1) + 0];
|
|
a->dlast[1] = a->out[2 * (a->outsize - 1) + 1];
|
|
|
|
if (!a->control_flag)
|
|
{
|
|
a->readsamps += a->outsize;
|
|
|
|
if ((a->readsamps >= a->read_startup) && (a->writesamps >= a->write_startup))
|
|
a->control_flag = 1;
|
|
}
|
|
|
|
if (a->control_flag)
|
|
control (a, -(a->outsize));
|
|
}
|
|
}
|
|
|
|
|
|
void RMATCH::getRMatchDiags (void* b, int* underflows, int* overflows, float* var, int* ringsize, int* nring)
|
|
{
|
|
auto *a = (RMATCH*) b;
|
|
*underflows = a->underflows;
|
|
*overflows = a->overflows;
|
|
a->underflows &= 0xFFFFFFFF;
|
|
a->overflows &= 0xFFFFFFFF;
|
|
*var = a->var;
|
|
*ringsize = a->ringsize;
|
|
*nring = a->n_ring;
|
|
}
|
|
|
|
|
|
void RMATCH::resetRMatchDiags (void*)
|
|
{
|
|
}
|
|
|
|
|
|
void RMATCH::forceRMatchVar (void* b, int force, float fvar)
|
|
{
|
|
auto *a = (RMATCH*) b;
|
|
a->force = force;
|
|
a->fvar = fvar;
|
|
}
|
|
|
|
|
|
void* RMATCH::create_rmatchV(int in_size, int out_size, int nom_inrate, int nom_outrate, int ringsize, float var)
|
|
{
|
|
return (void*)create_rmatch (
|
|
1, // run
|
|
nullptr, // input buffer, stuffed in other calls
|
|
nullptr, // output buffer, stuffed in other calls
|
|
in_size, // input buffer size (complex samples)
|
|
out_size, // output buffer size (complex samples)
|
|
nom_inrate, // nominal input sample-rate
|
|
nom_outrate, // nominal output sample-rate
|
|
0.0, // fc_high (0.0 -> automatic)
|
|
-1.0, // fc_low (-1.0 -> no low cutoff)
|
|
1.0, // gain
|
|
3.0, // startup delay (seconds)
|
|
1, // automatic ring-size [not implemented yet]
|
|
ringsize, // ringsize
|
|
1024, // R, coefficient density
|
|
var, // initial variable ratio
|
|
4096, // feed-forward moving average min size
|
|
262144, // feed-forward moving average max size - POWER OF TWO!
|
|
0.01f, // feed-forward exponential smoothing
|
|
4096, // proportional feedback min moving av ringsize
|
|
16384, // proportional feedback max moving av ringsize - POWER OF TWO!
|
|
4.0e-06f, // proportional feedback gain
|
|
1, // linearly interpolate cvar by sample
|
|
0.003f ); // slew time (seconds)
|
|
}
|
|
|
|
|
|
void RMATCH::destroy_rmatchV (void* ptr)
|
|
{
|
|
auto *a = (RMATCH*) ptr;
|
|
destroy_rmatch (a);
|
|
}
|
|
|
|
|
|
void RMATCH::setRMatchInsize (void* ptr, int insize)
|
|
{
|
|
auto *a = (RMATCH*) ptr;
|
|
a->run = 0;
|
|
std::this_thread::sleep_for(std::chrono::seconds(10));
|
|
decalc_rmatch(a);
|
|
a->insize = insize;
|
|
calc_rmatch (a);
|
|
a->run = 1;
|
|
}
|
|
|
|
|
|
void RMATCH::setRMatchOutsize (void* ptr, int outsize)
|
|
{
|
|
auto *a = (RMATCH*) ptr;
|
|
a->run = 0;
|
|
std::this_thread::sleep_for(std::chrono::seconds(10));
|
|
decalc_rmatch(a);
|
|
a->outsize = outsize;
|
|
calc_rmatch (a);
|
|
a->run = 1;
|
|
}
|
|
|
|
|
|
void RMATCH::setRMatchNomInrate (void* ptr, int nom_inrate)
|
|
{
|
|
auto *a = (RMATCH*) ptr;
|
|
a->run = 0;
|
|
std::this_thread::sleep_for(std::chrono::seconds(10));
|
|
decalc_rmatch(a);
|
|
a->nom_inrate = nom_inrate;
|
|
calc_rmatch (a);
|
|
a->run = 1;
|
|
}
|
|
|
|
|
|
void RMATCH::setRMatchNomOutrate (void* ptr, int nom_outrate)
|
|
{
|
|
auto *a = (RMATCH*) ptr;
|
|
a->run = 0;
|
|
std::this_thread::sleep_for(std::chrono::seconds(10));
|
|
decalc_rmatch(a);
|
|
a->nom_outrate = nom_outrate;
|
|
calc_rmatch (a);
|
|
a->run = 1;
|
|
}
|
|
|
|
|
|
void RMATCH::setRMatchRingsize (void* ptr, int ringsize)
|
|
{
|
|
auto *a = (RMATCH*) ptr;
|
|
a->run = 0;
|
|
std::this_thread::sleep_for(std::chrono::seconds(10));
|
|
decalc_rmatch(a);
|
|
a->ringsize = ringsize;
|
|
calc_rmatch (a);
|
|
a->run = 1;
|
|
}
|
|
|
|
|
|
void RMATCH::setRMatchFeedbackGain (void* b, float feedback_gain)
|
|
{
|
|
auto *a = (RMATCH*) b;
|
|
a->prop_gain = feedback_gain;
|
|
a->pr_gain = a->prop_gain * 48000.0 / (float)a->nom_outrate;
|
|
}
|
|
|
|
|
|
void RMATCH::setRMatchSlewTime (void* b, float slew_time)
|
|
{
|
|
auto *a = (RMATCH*) b;
|
|
a->run = 0;
|
|
decalc_rmatch(a); // deallocate all memory EXCEPT the data structure holding all current parameters
|
|
a->tslew = slew_time; // change the value of 'slew_time'
|
|
calc_rmatch(a); // recalculate/reallocate everything in the RMATCH
|
|
a->run = 1; // InterlockedBitTestAndSet(&a->run, 0); // turn ON the dataflow
|
|
}
|
|
|
|
|
|
void RMATCH::setRMatchSlewTime1(void* b, float slew_time)
|
|
{
|
|
auto *a = (RMATCH*) b;
|
|
float theta;
|
|
float dtheta;
|
|
a->run = 0;
|
|
delete[] a->cslew;
|
|
a->tslew = slew_time;
|
|
a->ntslew = (int)(a->tslew * (float) a->nom_outrate);
|
|
if (a->ntslew + 1 > a->rsize / 2) a->ntslew = a->rsize / 2 - 1;
|
|
a->cslew = new float[a->ntslew + 1]; // (float*)malloc0((a->ntslew + 1) * sizeof(float));
|
|
dtheta = (float) PI / (float)a->ntslew;
|
|
theta = 0.0;
|
|
for (int m = 0; m <= a->ntslew; m++)
|
|
{
|
|
a->cslew[m] = 0.5f * (1.0f - cos(theta));
|
|
theta += dtheta;
|
|
}
|
|
a->run = 1;
|
|
}
|
|
|
|
|
|
void RMATCH::setRMatchPropRingMin(void* ptr, int prop_min)
|
|
{
|
|
auto *a = (RMATCH*) ptr;
|
|
a->run = 0;
|
|
decalc_rmatch(a);
|
|
a->prop_ringmin = prop_min;
|
|
calc_rmatch(a);
|
|
a->run = 1;
|
|
}
|
|
|
|
|
|
void RMATCH::setRMatchPropRingMax(void* ptr, int prop_max)
|
|
{
|
|
auto *a = (RMATCH*) ptr;
|
|
a->run = 0;
|
|
decalc_rmatch(a);
|
|
a->prop_ringmax = prop_max; // must be a power of two
|
|
calc_rmatch(a);
|
|
a->run = 1;
|
|
}
|
|
|
|
|
|
void RMATCH::setRMatchFFRingMin(void* ptr, int ff_ringmin)
|
|
{
|
|
auto *a = (RMATCH*) ptr;
|
|
a->run = 0;
|
|
decalc_rmatch(a);
|
|
a->ff_ringmin = ff_ringmin;
|
|
calc_rmatch(a);
|
|
a->run = 1;
|
|
}
|
|
|
|
|
|
void RMATCH::setRMatchFFRingMax(void* ptr, int ff_ringmax)
|
|
{
|
|
auto *a = (RMATCH*) ptr;
|
|
a->run = 0;
|
|
decalc_rmatch(a);
|
|
a->ff_ringmax = ff_ringmax; // must be a power of two
|
|
calc_rmatch(a);
|
|
a->run = 1;
|
|
}
|
|
|
|
|
|
void RMATCH::setRMatchFFAlpha(void* ptr, float ff_alpha)
|
|
{
|
|
auto *a = (RMATCH*) ptr;
|
|
a->run = 0;
|
|
std::this_thread::sleep_for(std::chrono::seconds(10));
|
|
a->ff_alpha = ff_alpha;
|
|
a->run = 1;
|
|
}
|
|
|
|
|
|
void RMATCH::getControlFlag(void* ptr, int* control_flag)
|
|
{
|
|
RMATCH const *a = (RMATCH*) ptr;
|
|
*control_flag = a->control_flag;
|
|
}
|
|
|
|
// the following function is DEPRECATED
|
|
// it is intended for Legacy PowerSDR use only
|
|
|
|
void* RMATCH::create_rmatchLegacyV(int in_size, int out_size, int nom_inrate, int nom_outrate, int ringsize)
|
|
{
|
|
return (void*) create_rmatch(
|
|
1, // run
|
|
nullptr, // input buffer, stuffed in other calls
|
|
nullptr, // output buffer, stuffed in other calls
|
|
in_size, // input buffer size (complex samples)
|
|
out_size, // output buffer size (complex samples)
|
|
nom_inrate, // nominal input sample-rate
|
|
nom_outrate, // nominal output sample-rate
|
|
0.0, // fc_high (0.0 -> automatic)
|
|
-1.0, // fc_low (-1.0 -> no low cutoff)
|
|
1.0, // gain
|
|
3.0, // startup delay (seconds)
|
|
1, // automatic ring-size [not implemented yet]
|
|
ringsize, // ringsize
|
|
1024, // R, coefficient density
|
|
1.0, // initial variable ratio
|
|
4096, // feed-forward moving average min size
|
|
262144, // feed-forward moving average max size - POWER OF TWO!
|
|
0.01f, // feed-forward exponential smoothing
|
|
4096, // proportional feedback min moving av ringsize
|
|
16384, // proportional feedback max moving av ringsize - POWER OF TWO!
|
|
1.0e-06f, // proportional feedback gain ***W4WMT - reduce loop gain a bit for PowerSDR to help Primary buffers > 512
|
|
0, // linearly interpolate cvar by sample ***W4WMT - set varmode = 0 for PowerSDR (doesn't work otherwise!?!)
|
|
0.003f); // slew time (seconds)
|
|
}
|
|
|
|
} // namespace WDSP
|