apply procedural approach to clipper+tube

This commit is contained in:
2025-10-24 01:59:55 +02:00
parent b7ef24d324
commit 8af07582ae
4 changed files with 313 additions and 332 deletions

194
clip/tube.h Normal file
View File

@@ -0,0 +1,194 @@
#pragma once
#include <algorithm>
#include <cmath>
#include <cstdlib>
#include <stdint.h>
namespace trnr {
// modeled tube preamp based on tube2 by Chris Johnson (MIT License)
struct tube {
double samplerate;
double prev_sample_a;
double prev_sample_b;
double prev_sample_c;
double prev_sample_d;
double prev_sample_e;
double prev_sample_f;
uint32_t fdp_l;
uint32_t fdp_r;
float input_vol;
float tube_amt;
void set_input(double value) { input_vol = std::clamp(value, 0.0, 1.0); }
void set_tube(double value) { tube_amt = std::clamp(value, 0.0, 1.0); }
};
inline void tube_init(tube& t, double samplerate)
{
t.samplerate = 44100;
t.input_vol = 0.5;
t.tube_amt = 0.5;
t.prev_sample_a = 0.0;
t.prev_sample_b = 0.0;
t.prev_sample_c = 0.0;
t.prev_sample_d = 0.0;
t.prev_sample_e = 0.0;
t.prev_sample_f = 0.0;
t.fdp_l = 1.0;
while (t.fdp_l < 16386) t.fdp_l = rand() * UINT32_MAX;
t.fdp_r = 1.0;
while (t.fdp_r < 16386) t.fdp_r = rand() * UINT32_MAX;
}
template <typename t_sample>
inline void tube_process_block(tube& t, t_sample** inputs, t_sample** outputs, long sampleframes)
{
t_sample* in1 = inputs[0];
t_sample* in2 = inputs[1];
t_sample* out1 = outputs[0];
t_sample* out2 = outputs[1];
double overallscale = 1.0;
overallscale /= 44100.0;
overallscale *= t.samplerate;
double input_pad = t.input_vol;
double iterations = 1.0 - t.tube_amt;
int powerfactor = (9.0 * iterations) + 1;
double asym_pad = (double)powerfactor;
double gainscaling = 1.0 / (double)(powerfactor + 1);
double outputscaling = 1.0 + (1.0 / (double)(powerfactor));
while (--sampleframes >= 0) {
double input_l = *in1;
double input_r = *in2;
if (fabs(input_l) < 1.18e-23) input_l = t.fdp_l * 1.18e-17;
if (fabs(input_r) < 1.18e-23) input_r = t.fdp_r * 1.18e-17;
if (input_pad < 1.0) {
input_l *= input_pad;
input_r *= input_pad;
}
if (overallscale > 1.9) {
double stored = input_l;
input_l += t.prev_sample_a;
t.prev_sample_a = stored;
input_l *= 0.5;
stored = input_r;
input_r += t.prev_sample_b;
t.prev_sample_b = stored;
input_r *= 0.5;
} // for high sample rates on this plugin we are going to do a simple average
if (input_l > 1.0) input_l = 1.0;
if (input_l < -1.0) input_l = -1.0;
if (input_r > 1.0) input_r = 1.0;
if (input_r < -1.0) input_r = -1.0;
// flatten bottom, point top of sine waveshaper L
input_l /= asym_pad;
double sharpen = -input_l;
if (sharpen > 0.0) sharpen = 1.0 + sqrt(sharpen);
else sharpen = 1.0 - sqrt(-sharpen);
input_l -= input_l * fabs(input_l) * sharpen * 0.25;
// this will take input from exactly -1.0 to 1.0 max
input_l *= asym_pad;
// flatten bottom, point top of sine waveshaper R
input_r /= asym_pad;
sharpen = -input_r;
if (sharpen > 0.0) sharpen = 1.0 + sqrt(sharpen);
else sharpen = 1.0 - sqrt(-sharpen);
input_r -= input_r * fabs(input_r) * sharpen * 0.25;
// this will take input from exactly -1.0 to 1.0 max
input_r *= asym_pad;
// end first asym section: later boosting can mitigate the extreme
// softclipping of one side of the wave
// and we are asym clipping more when Tube is cranked, to compensate
// original Tube algorithm: powerfactor widens the more linear region of the wave
double factor = input_l; // Left channel
for (int x = 0; x < powerfactor; x++) factor *= input_l;
if ((powerfactor % 2 == 1) && (input_l != 0.0)) factor = (factor / input_l) * fabs(input_l);
factor *= gainscaling;
input_l -= factor;
input_l *= outputscaling;
factor = input_r; // Right channel
for (int x = 0; x < powerfactor; x++) factor *= input_r;
if ((powerfactor % 2 == 1) && (input_r != 0.0)) factor = (factor / input_r) * fabs(input_r);
factor *= gainscaling;
input_r -= factor;
input_r *= outputscaling;
if (overallscale > 1.9) {
double stored = input_l;
input_l += t.prev_sample_c;
t.prev_sample_c = stored;
input_l *= 0.5;
stored = input_r;
input_r += t.prev_sample_d;
t.prev_sample_d = stored;
input_r *= 0.5;
} // for high sample rates on this plugin we are going to do a simple average
// end original Tube. Now we have a boosted fat sound peaking at 0dB exactly
// hysteresis and spiky fuzz L
double slew = t.prev_sample_e - input_l;
if (overallscale > 1.9) {
double stored = input_l;
input_l += t.prev_sample_e;
t.prev_sample_e = stored;
input_l *= 0.5;
} else t.prev_sample_e = input_l; // for this, need previousSampleC always
if (slew > 0.0) slew = 1.0 + (sqrt(slew) * 0.5);
else slew = 1.0 - (sqrt(-slew) * 0.5);
input_l -= input_l * fabs(input_l) * slew * gainscaling;
// reusing gainscaling that's part of another algorithm
if (input_l > 0.52) input_l = 0.52;
if (input_l < -0.52) input_l = -0.52;
input_l *= 1.923076923076923;
// hysteresis and spiky fuzz R
slew = t.prev_sample_f - input_r;
if (overallscale > 1.9) {
double stored = input_r;
input_r += t.prev_sample_f;
t.prev_sample_f = stored;
input_r *= 0.5;
} else t.prev_sample_f = input_r; // for this, need previousSampleC always
if (slew > 0.0) slew = 1.0 + (sqrt(slew) * 0.5);
else slew = 1.0 - (sqrt(-slew) * 0.5);
input_r -= input_r * fabs(input_r) * slew * gainscaling;
// reusing gainscaling that's part of another algorithm
if (input_r > 0.52) input_r = 0.52;
if (input_r < -0.52) input_r = -0.52;
input_r *= 1.923076923076923;
// end hysteresis and spiky fuzz section
// begin 64 bit stereo floating point dither
// int expon; frexp((double)inputSampleL, &expon);
t.fdp_l ^= t.fdp_l << 13;
t.fdp_l ^= t.fdp_l >> 17;
t.fdp_l ^= t.fdp_l << 5;
// inputSampleL += ((double(fpdL)-uint32_t(0x7fffffff)) * 1.1e-44l * pow(2,expon+62));
// frexp((double)inputSampleR, &expon);
t.fdp_r ^= t.fdp_r << 13;
t.fdp_r ^= t.fdp_r >> 17;
t.fdp_r ^= t.fdp_r << 5;
// inputSampleR += ((double(fpdR)-uint32_t(0x7fffffff)) * 1.1e-44l * pow(2,expon+62));
// end 64 bit stereo floating point dither
*out1 = input_l;
*out2 = input_r;
in1++;
in2++;
out1++;
out2++;
}
}
} // namespace trnr