0
1
mirror of https://github.com/radio95-rnt/fm95.git synced 2026-02-26 11:22:00 +01:00
Files
fm95/filter/bs412.c

73 lines
2.4 KiB
C

#include "bs412.h"
inline float dbr_to_deviation(float dbr) {
return 19000.0f * pow(10.0, dbr / 10.0);
}
inline float deviation_to_dbr(float deviation) {
if (deviation < 1e-6f) return -100.0f;
return 10*log10f(deviation/19000);
}
void init_bs412(BS412Compressor* mpx, uint32_t mpx_deviation, float target_power, float attack, float release, float max, uint32_t sample_rate) {
mpx->mpx_deviation = mpx_deviation;
mpx->average_counter = 0;
mpx->average = 0;
mpx->sample_rate = sample_rate;
mpx->attack = expf(-1.0f / (attack * sample_rate));
mpx->release = expf(-1.0f / (release * sample_rate));
mpx->target = target_power;
mpx->gain = 0.0f;
mpx->max = max;
#ifdef BS412_DEBUG
debug_printf("Initialized MPX power measurement with sample rate: %d\n", sample_rate);
#endif
}
static inline float soft_clip_tanh(float sample, float threshold) {
if (fabsf(sample) <= threshold) {
return sample; // Linear region
}
float sign = (sample >= 0) ? 1.0f : -1.0f;
float excess = fabsf(sample) - threshold;
return sign * (threshold + tanhf(excess) * (1.0f - threshold));
}
float bs412_compress(BS412Compressor* mpx, float sample) {
mpx->average += sample * sample * mpx->mpx_deviation * mpx->mpx_deviation;
mpx->average_counter++;
float avg_power = mpx->average / mpx->average_counter;
float avg_deviation = sqrtf(avg_power) * sqrtf(2);
float modulation_power = deviation_to_dbr(avg_deviation);
#ifdef BS412_DEBUG
if(mpx->average_counter % mpx->sample_rate == 0) {
debug_printf("MPX power: %.2f dBr with gain %.2fx (%.2f dBr)\n", modulation_power, mpx->gain, deviation_to_dbr(avg_deviation * mpx->gain));
}
#endif
if (mpx->average_counter >= mpx->sample_rate * 60) {
#ifdef BS412_DEBUG
debug_printf("Resetting MPX power measurement\n");
#endif
mpx->average = avg_power;
mpx->average_counter = 1;
}
float target_gain = powf(10.0f, (mpx->target - modulation_power) / 10.0f);
if (modulation_power > mpx->target) {
mpx->gain = mpx->attack * mpx->gain + (1.0f - mpx->attack) * target_gain;
} else {
mpx->gain = mpx->release * mpx->gain + (1.0f - mpx->release) * target_gain;
}
mpx->gain = fmaxf(0.0f, fminf(mpx->max, mpx->gain));
float output_sample = sample * mpx->gain;
float limit_threshold = dbr_to_deviation(mpx->target + 0.1f) / mpx->mpx_deviation;
output_sample = soft_clip_tanh(output_sample, limit_threshold);
return output_sample;
}