mirror of
https://github.com/saitohirga/WSJT-X.git
synced 2024-11-18 18:12:12 -05:00
76eed0ecc4
There are several object lifetime issues that can cause crashes on application close down. This change is the first phase of addressing these issues. git-svn-id: svn+ssh://svn.code.sf.net/p/wsjt/wsjt/branches/wsjtx@5719 ab8295b8-cf94-4d9e-aec4-7959e3be5d79
288 lines
8.1 KiB
C++
288 lines
8.1 KiB
C++
#include "Modulator.hpp"
|
|
#include <limits>
|
|
#include <qmath.h>
|
|
#include <QDateTime>
|
|
#include <QDebug>
|
|
#include "mainwindow.h"
|
|
#include "soundout.h"
|
|
|
|
#include "moc_Modulator.cpp"
|
|
|
|
extern float gran(); // Noise generator (for tests only)
|
|
|
|
#define RAMP_INCREMENT 64 // MUST be an integral factor of 2^16
|
|
|
|
#if defined (WSJT_SOFT_KEYING)
|
|
# define SOFT_KEYING WSJT_SOFT_KEYING
|
|
#else
|
|
# define SOFT_KEYING 1
|
|
#endif
|
|
|
|
double const Modulator::m_twoPi = 2.0 * 3.141592653589793238462;
|
|
|
|
// float wpm=20.0;
|
|
// unsigned m_nspd=1.2*48000.0/wpm;
|
|
// m_nspd=3072; //18.75 WPM
|
|
unsigned const Modulator::m_nspd = 2048 + 512; // 22.5 WPM
|
|
|
|
Modulator::Modulator (unsigned frameRate, unsigned periodLengthInSeconds,
|
|
QObject * parent)
|
|
: AudioDevice {parent}
|
|
, m_quickClose {false}
|
|
, m_phi {0.0}
|
|
, m_toneSpacing {0.0}
|
|
, m_fSpread {0.0}
|
|
, m_itone0 {0}
|
|
, m_frameRate {frameRate}
|
|
, m_period {periodLengthInSeconds}
|
|
, m_state {Idle}
|
|
, m_tuning {false}
|
|
, m_cwLevel {false}
|
|
{
|
|
}
|
|
|
|
void Modulator::start (unsigned symbolsLength, double framesPerSymbol,
|
|
double frequency, double toneSpacing,
|
|
SoundOutput * stream, Channel channel,
|
|
bool synchronize, double dBSNR)
|
|
{
|
|
Q_ASSERT (stream);
|
|
|
|
// Time according to this computer which becomes our base time
|
|
qint64 ms0 = QDateTime::currentMSecsSinceEpoch() % 86400000;
|
|
|
|
if (m_state != Idle)
|
|
{
|
|
stop ();
|
|
}
|
|
|
|
m_quickClose = false;
|
|
|
|
m_symbolsLength = symbolsLength;
|
|
m_isym0 = std::numeric_limits<unsigned>::max (); // big number
|
|
m_frequency0 = 0.;
|
|
m_addNoise = dBSNR < 0.;
|
|
m_nsps = framesPerSymbol;
|
|
m_frequency = frequency;
|
|
m_amp = std::numeric_limits<qint16>::max ();
|
|
m_toneSpacing = toneSpacing;
|
|
|
|
// noise generator parameters
|
|
if (m_addNoise) {
|
|
m_snr = qPow (10.0, 0.05 * (dBSNR - 6.0));
|
|
m_fac = 3000.0;
|
|
if (m_snr > 1.0) m_fac = 3000.0 / m_snr;
|
|
}
|
|
|
|
unsigned mstr = ms0 % (1000 * m_period); // ms in period
|
|
m_ic = (mstr / 1000) * m_frameRate; // we start exactly N seconds
|
|
// into period where N is the next whole second
|
|
|
|
m_silentFrames = 0;
|
|
// calculate number of silent frames to send
|
|
if (synchronize && !m_tuning) {
|
|
m_silentFrames = m_ic + m_frameRate - (mstr * m_frameRate / 1000);
|
|
}
|
|
|
|
initialize (QIODevice::ReadOnly, channel);
|
|
Q_EMIT stateChanged ((m_state = (synchronize && m_silentFrames) ?
|
|
Synchronizing : Active));
|
|
m_stream = stream;
|
|
if (m_stream) m_stream->restart (this);
|
|
}
|
|
|
|
void Modulator::tune (bool newState)
|
|
{
|
|
m_tuning = newState;
|
|
if (!m_tuning) stop (true);
|
|
}
|
|
|
|
void Modulator::stop (bool quick)
|
|
{
|
|
m_quickClose = quick;
|
|
close ();
|
|
}
|
|
|
|
void Modulator::close ()
|
|
{
|
|
if (m_stream)
|
|
{
|
|
if (m_quickClose)
|
|
{
|
|
m_stream->reset ();
|
|
}
|
|
else
|
|
{
|
|
m_stream->stop ();
|
|
}
|
|
}
|
|
if (m_state != Idle)
|
|
{
|
|
Q_EMIT stateChanged ((m_state = Idle));
|
|
}
|
|
AudioDevice::close ();
|
|
}
|
|
|
|
qint64 Modulator::readData (char * data, qint64 maxSize)
|
|
{
|
|
static int j0=-1;
|
|
static double toneFrequency0=1500.0;
|
|
double toneFrequency=1500.0;
|
|
|
|
if(maxSize==0) return 0;
|
|
Q_ASSERT (!(maxSize % qint64 (bytesPerFrame ()))); // no torn frames
|
|
Q_ASSERT (isOpen ());
|
|
|
|
qint64 numFrames (maxSize / bytesPerFrame ());
|
|
qint16 * samples (reinterpret_cast<qint16 *> (data));
|
|
qint16 * end (samples + numFrames * (bytesPerFrame () / sizeof (qint16)));
|
|
qint64 framesGenerated (0);
|
|
|
|
switch (m_state)
|
|
{
|
|
case Synchronizing:
|
|
{
|
|
if (m_silentFrames) { // send silence up to first second
|
|
framesGenerated = qMin (m_silentFrames, numFrames);
|
|
for ( ; samples != end; samples = load (0, samples)) { // silence
|
|
}
|
|
m_silentFrames -= framesGenerated;
|
|
return framesGenerated * bytesPerFrame ();
|
|
}
|
|
|
|
Q_EMIT stateChanged ((m_state = Active));
|
|
m_cwLevel = false;
|
|
m_ramp = 0; // prepare for CW wave shaping
|
|
}
|
|
// fall through
|
|
|
|
case Active:
|
|
{
|
|
unsigned isym (m_tuning ? 0 : m_ic / (4.0 * m_nsps)); // Actual fsample=48000
|
|
if (isym >= m_symbolsLength && icw[0] > 0) { // start CW condition
|
|
// Output the CW ID
|
|
m_dphi = m_twoPi * m_frequency / m_frameRate;
|
|
unsigned const ic0 = m_symbolsLength * 4 * m_nsps;
|
|
unsigned j (0);
|
|
|
|
while (samples != end) {
|
|
j = (m_ic - ic0) / m_nspd + 1; // symbol of this sample
|
|
bool level {bool (icw[j])};
|
|
|
|
m_phi += m_dphi;
|
|
if (m_phi > m_twoPi) m_phi -= m_twoPi;
|
|
|
|
qint16 sample=0;
|
|
float amp=32767.0;
|
|
if(m_ramp!=0) {
|
|
float x=qSin(float(m_phi));
|
|
if(SOFT_KEYING) {
|
|
amp=qAbs(qint32(m_ramp));
|
|
if(amp>32767.0) amp=32767.0;
|
|
}
|
|
sample=round(amp*x);
|
|
}
|
|
|
|
if (int (j) <= icw[0] && j < NUM_CW_SYMBOLS) // stop condition
|
|
{
|
|
samples = load (postProcessSample (sample), samples);
|
|
++framesGenerated;
|
|
++m_ic;
|
|
}
|
|
else
|
|
{
|
|
Q_EMIT stateChanged ((m_state = Idle));
|
|
return framesGenerated * bytesPerFrame ();
|
|
}
|
|
|
|
// adjust ramp
|
|
if ((m_ramp != 0 && m_ramp != std::numeric_limits<qint16>::min ()) || level != m_cwLevel)
|
|
{
|
|
// either ramp has terminated at max/min or direction has changed
|
|
m_ramp += RAMP_INCREMENT; // ramp
|
|
}
|
|
m_cwLevel = level;
|
|
}
|
|
|
|
return framesGenerated * bytesPerFrame ();
|
|
}
|
|
|
|
double const baud (12000.0 / m_nsps);
|
|
// fade out parameters (no fade out for tuning)
|
|
unsigned const i0 = m_tuning ? 9999 * m_nsps : (m_symbolsLength - 0.017) * 4.0 * m_nsps;
|
|
unsigned const i1 = m_tuning ? 9999 * m_nsps : m_symbolsLength * 4.0 * m_nsps;
|
|
|
|
for (unsigned i = 0; i < numFrames && m_ic <= i1; ++i) {
|
|
isym = m_tuning ? 0 : m_ic / (4.0 * m_nsps); //Actual fsample=48000
|
|
if (isym != m_isym0 || m_frequency != m_frequency0) {
|
|
if(itone[0]>=100) {
|
|
toneFrequency0=itone[0];
|
|
} else {
|
|
if(m_toneSpacing==0.0) {
|
|
toneFrequency0=m_frequency + itone[isym]*baud;
|
|
} else {
|
|
toneFrequency0=m_frequency + itone[isym]*m_toneSpacing;
|
|
}
|
|
}
|
|
m_dphi = m_twoPi * toneFrequency0 / m_frameRate;
|
|
m_isym0 = isym;
|
|
m_frequency0 = m_frequency; //???
|
|
}
|
|
|
|
int j=m_ic/480;
|
|
if(m_fSpread>0.0 and j!=j0) {
|
|
float x1=(float)qrand()/RAND_MAX;
|
|
float x2=(float)qrand()/RAND_MAX;
|
|
toneFrequency = toneFrequency0 + 0.5*m_fSpread*(x1+x2-1.0);
|
|
m_dphi = m_twoPi * toneFrequency / m_frameRate;
|
|
j0=j;
|
|
}
|
|
|
|
m_phi += m_dphi;
|
|
if (m_phi > m_twoPi) m_phi -= m_twoPi;
|
|
if (m_ic > i0) m_amp = 0.98 * m_amp;
|
|
if (m_ic > i1) m_amp = 0.0;
|
|
|
|
samples = load (postProcessSample (m_amp * qSin (m_phi)), samples);
|
|
++framesGenerated;
|
|
++m_ic;
|
|
}
|
|
|
|
if (m_amp == 0.0) { // TODO G4WJS: compare double with zero might not be wise
|
|
if (icw[0] == 0) {
|
|
// no CW ID to send
|
|
Q_EMIT stateChanged ((m_state = Idle));
|
|
return framesGenerated * bytesPerFrame ();
|
|
}
|
|
m_phi = 0.0;
|
|
}
|
|
|
|
m_frequency0 = m_frequency;
|
|
// done for this chunk - continue on next call
|
|
return framesGenerated * bytesPerFrame ();
|
|
}
|
|
// fall through
|
|
|
|
case Idle:
|
|
break;
|
|
}
|
|
|
|
Q_ASSERT (Idle == m_state);
|
|
return 0;
|
|
}
|
|
|
|
qint16 Modulator::postProcessSample (qint16 sample) const
|
|
{
|
|
if (m_addNoise) { // Test frame, we'll add noise
|
|
qint32 s = m_fac * (gran () + sample * m_snr / 32768.0);
|
|
if (s > std::numeric_limits<qint16>::max ()) {
|
|
s = std::numeric_limits<qint16>::max ();
|
|
}
|
|
if (s < std::numeric_limits<qint16>::min ()) {
|
|
s = std::numeric_limits<qint16>::min ();
|
|
}
|
|
sample = s;
|
|
}
|
|
return sample;
|
|
}
|