colorchord/embeddedcommon/DFT12Small.c

347 lines
10 KiB
C

//NOTE DO NOT EDIT THIS FILE WITHOUT ALSO EDITING DFT8TURBO!!!
#include <stdint.h>
#include <stdlib.h>
#include "DFT12Small.h"
#include <math.h>
#include <stdio.h>
#define MAX_FREQS (12)
#define OCTAVES (4)
/*
General procedure - use this code, with uint16_t or uint32_t buffers, and make sure none of the alarms go off.
All of the paths still require no more than an 8-bit multiply.
You should test with extreme cases, like square wave sweeps in, etc.
*/
//#define TWELVEBIT
#define EIGHTBIT
#ifdef TWELVEBIT
//No larger than 12-bit signed values for integration or sincos
#define FRONTEND_AMPLITUDE (0)
#define INITIAL_DECIMATE (2)
#define INTEGRATOR_DECIMATE (8)
#define FINAL_DECIMATE (4)
#elif defined( EIGHTBIT )
//No larger than 8-bit signed values for integration or sincos
#define FRONTEND_AMPLITUDE (2)
#define INITIAL_DECIMATE (5) //Yurgh... only 3 bits of ADC data. That's 8 unique levels :(
#define INTEGRATOR_DECIMATE (8)
#define FINAL_DECIMATE (1)
#endif
//4x the hits (sin/cos and we need to do it once for each edge)
//8x for selecting a higher octave.
#define FREQREBASE 8.0
#define TARGFREQ 10000.0
/* Tradeoff guide:
* We will optimize for RAM size here.
* INITIAL_DECIMATE; A larger decimation: {NOTE 1}
+) Reduces the bit depth needed for the integral map.
If you use "1" and a fully saturted map (highest note is every sample), it will not overflow a signed 12-bit number.
-) Increases noise.
With full-scale: 0->1 minimal 1->2 minimal 2->3 significantly noticable, 3->4 major.
If sound is quieter, it matters more. Not sure with other changes in system. (2) seems ok.
-) If you make it (1) or (0) You can't do an 8-bit multiply and keep the output in a signed range.
Also, other things, like frequency of hits can manipulate the maximum bit depth needed for integral map.
* If you weight the bins in advance see "mulmux", you can: {NOTE 2}
+) potentially use shallower bit depth but
-) have to compute the multiply every time you update the bin.
* You can use a modified-square-wave which only integrates for 1/2 of the duty cycle. {NOTE 3}
+) uses 1/2 the integral memory.
-) Not as pretty of an output. See "integral_at"
*TODO: Investigate using all unsigned (to make multiply and/or 12-bit storage easier)
*TODO: Consider a mode which has 16-bit integrals, but still 8-bit cossin data.
So, the idea here is we would keep a running total of the current ADC value, kept away in a int16_t.
It is constantly summing, so we can take an integral of it. Or rather an integral range.
Over time, we perform operations like adding or subtracting from a current place. It basically is
a DFT where the kernel is computed using square waves (or modified square waves)
*/
//These live in RAM.
int16_t running_integral; //Realistically treat as 12-bits on ramjet8
int16_t integral_at[MAX_FREQS*OCTAVES]; //For ramjet8, make 12-bits
int32_t cossindata[MAX_FREQS*OCTAVES*2]; //Contains COS and SIN data. (32-bit for now, will be 16-bit, potentially even 8.)
uint8_t which_octave_for_op[MAX_FREQS]; //counts up, tells you which ocative you are operating on. PUT IN RAM.
uint8_t actiontableplace;
#define NR_OF_OPS (4<<OCTAVES)
//Format is:
// 255 = DO NOT OPERATE
// bits 0..3 unfolded octave, i.e. sin/cos are offset by one.
// bit 4 = add or subtract.
uint8_t optable[NR_OF_OPS]; //PUT IN FLASH
#define ACTIONTABLESIZE 256
uint16_t actiontable[ACTIONTABLESIZE]; //PUT IN FLASH // If there are more than 8 freqbins, this must be a uint16_t, otherwise if more than 16, 32.
//Format is
uint8_t mulmux[MAX_FREQS]; //PUT IN FLASH
static int Setup( float * frequencies, int bins )
{
int i;
printf( "BINS: %d\n", bins );
float highestf = frequencies[MAX_FREQS-1];
for( i = 0; i < MAX_FREQS; i++ )
{
mulmux[i] = (uint8_t)( highestf / frequencies[i] * 255 + 0.5 );
printf( "MM: %d %f / %f\n", mulmux[i], frequencies[i], highestf );
}
for( i = bins-MAX_FREQS; i < bins; i++ )
{
int topbin = i - (bins-MAX_FREQS);
float f = frequencies[i]/FREQREBASE;
float hits_per_table = (float)ACTIONTABLESIZE/f;
int dhrpertable = (int)(hits_per_table+.5);//TRICKY: You might think you need to have even number of hits (sin/cos), but you don't! It can flip sin/cos each time through the table!
float err = (TARGFREQ/((float)ACTIONTABLESIZE/dhrpertable) - (float)TARGFREQ/f)/((float)TARGFREQ/f);
//Perform an op every X samples. How well does this map into units of 1024?
printf( "%d %f -> hits per %d: %f %d (%.2f%% error)\n", topbin, f, ACTIONTABLESIZE, (float)ACTIONTABLESIZE/f, dhrpertable, err * 100.0 );
if( dhrpertable >= ACTIONTABLESIZE )
{
fprintf( stderr, "Error: Too many hits.\n" );
exit(0);
}
float advance_per_step = dhrpertable/(float)ACTIONTABLESIZE;
float fvadv = 0.5;
int j;
int countset = 0;
//Tricky: We need to start fadv off at such a place that there won't be a hicchup when going back around to 0.
// I believe this is done by setting fvadv to 0.5 initially. Unsure.
for( j = 0; j < ACTIONTABLESIZE; j++ )
{
if( fvadv >= 0.5 )
{
actiontable[j] |= 1<<topbin;
fvadv -= 1.0;
countset++;
}
fvadv += advance_per_step;
}
printf( " countset: %d\n", countset );
}
//exit(1);
int phaseinop[OCTAVES] = { 0 };
int already_hit_octaveplace[OCTAVES*2] = { 0 };
for( i = 0; i < NR_OF_OPS; i++ )
{
int longestzeroes = 0;
int val = i & ((1<<OCTAVES)-1);
for( longestzeroes = 0; longestzeroes < 255 && ( ((val >> longestzeroes) & 1) == 0 ); longestzeroes++ );
//longestzeroes goes: 255, 0, 1, 0, 2, 0, 1, 0, 3, 0, 1, 0, 2, ...
//This isn't great, because we need to also know whether we are attacking the SIN side or the COS side, and if it's + or -.
//We can actually decide that out.
if( longestzeroes == 255 )
{
//This is a nop. Emit a nop.
optable[i] = 255;
}
else
{
longestzeroes = OCTAVES-1-longestzeroes; //Actually do octave 0 least often.
int iop = phaseinop[longestzeroes]++;
int toop = longestzeroes;
int toopmon = (longestzeroes<<1) | (iop & 1);
//if it's the first time an octave happened this set, flag it. This may be used later in the process.
if( !already_hit_octaveplace[toopmon] )
{
already_hit_octaveplace[toopmon] = 1;
toop |= 1<<5;
}
if( iop & 1 )
{
toop |= 1<<6;
}
//Handle add/subtract bit.
if( iop & 2 ) toop |= 1<<4;
optable[i] = toop;
//printf( " %d %d %d\n", iop, val, longestzeroes );
}
//printf( "HBT: %d = %d\n", i, optable[i] );
}
//exit(1);
return 0;
}
void Small12BitRun( int8_t adcval )
{
int16_t adcv = adcval;
adcv *= FRONTEND_AMPLITUDE;
if( adcv > 127 ) adcv = 127;
if( adcv < -128 ) adcv = -128;
running_integral += adcv>>INITIAL_DECIMATE;
uint32_t action = actiontable[actiontableplace++];
int n;
for( n = 0; n < MAX_FREQS; n++, action>>=1 )
{
if( !( action & 1 ) ) continue;
int ao = which_octave_for_op[n];
ao++;
if( ao >= NR_OF_OPS ) ao = 0;
which_octave_for_op[n] = ao;
int op = optable[ao];
if( op == 255 )
continue;
//int octaveplace = op & 0xf;
//Tricky: We share the integral with SIN and COS.
//We don't need to. It would produce a slightly cleaner signal. See: NOTE 3
uint8_t octave = op & 0xf;
uint8_t intindex = octave * MAX_FREQS + n;
//int invoct = OCTAVES-1-octaveplace;
int16_t diff;
if( op & 0x10 ) //ADD
{
diff = integral_at[intindex] - running_integral;
}
else //SUBTRACT
{
diff = running_integral - integral_at[intindex];
}
integral_at[intindex] = running_integral;
#ifdef TWELVEBIT
if( diff > 2000 || diff < -2000 ) printf( "!!!!!!!!!!!! %d !!!!!!!!!!!\n", diff );
#elif defined( EIGHTBIT )
if( diff > 124 || diff < -124 ) printf( "!!!!!!!!!!!! %d !!!!!!!!!!!\n", diff );
#endif
//uint8_t idx = ( intindex << 1 );
intindex<<=1;
if( op&(1<<6) )
{
intindex |= 1;
}
//printf( "%d: %d + %d * %d >> 8 - %d\n", intindex, cossindata[intindex], diff, mulmux[intindex/2], cossindata[intindex]>>4 );
uint8_t mulmuxval = mulmux[n];
//Do you live on a super lame processor? {NOTE 4}
//If you do, you might not have good signed multiply operations. So, an alternative mechanism is found here.
// +) Able to more cleanly crush to an 8-bit multiply.
// +) Gets extra bit of precision back, i.e. the sign bit is now used as a data bit.
// -) More than 1 line of C code. Requires possible double invert.
#if 1
//Terrible processor, i.e. PMS133
if( 0 && diff < 0 )
{
diff *= -1;
diff >>= (OCTAVES-1-octave);
if( diff > 250 ) printf( "!!!!!!!**** %d ****!!!!!!!\n", diff );
diff = (uint16_t)diff * (uint16_t)mulmuxval;
diff >>= INTEGRATOR_DECIMATE;
diff *= -1;
}
else
{
diff >>= (OCTAVES-1-octave);
if( diff > 250 ) printf( "!!!!!!!**** %d ****!!!!!!!\n", diff );
diff = (uint16_t)diff * (uint16_t)mulmuxval;
diff >>= INTEGRATOR_DECIMATE;
}
#else
//Decent processor, i.e. ATTiny85.
diff = ((diff>>(OCTAVES-1-octave)) * mulmuxval ) >> 6;
#endif
cossindata[intindex] = cossindata[intindex]
+ diff
- (cossindata[intindex]>>4)
;
#ifdef EIGHTBIT
if( cossindata[intindex] > 0 ) cossindata[intindex]--;
if( cossindata[intindex] < 0 ) cossindata[intindex]++;
#endif
}
}
void DoDFT12BitSmall( float * outbins, float * frequencies, int bins, const float * databuffer, int place_in_data_buffer, int size_of_data_buffer, float q, float speedup )
{
static int is_setup;
if( !is_setup ) { is_setup = 1; Setup( frequencies, bins ); }
static int last_place;
int i;
for( i = last_place; i != place_in_data_buffer; i = (i+1)%size_of_data_buffer )
{
int16_t ifr1 = (int16_t)( ((databuffer[i]) ) * 4095 );
Small12BitRun( ifr1>>5 ); //5 = Actually only feed algorithm numbers from -128 to 127.
}
last_place = place_in_data_buffer;
static int idiv;
idiv++;
#if 1
for( i = 0; i < bins; i++ )
{
int iss = cossindata[i*2+0]>>FINAL_DECIMATE;
int isc = cossindata[i*2+1]>>FINAL_DECIMATE;
int mux = iss * iss + isc * isc;
if( mux <= 0 )
{
outbins[i] = 0;
}
else
{
outbins[i] = sqrt((float)mux)/50.0;
#ifdef TWELVEBIT
if( abs( cossindata[i*2+0] ) > 1000 || abs( cossindata[i*2+1] ) > 1000 )
printf( "CS OVF %d/%d/%d/%f\n", i, cossindata[i*2+0], cossindata[i*2+1],outbins[i] );
#elif defined( EIGHTBIT )
if( abs( cossindata[i*2+0] ) > 120 || abs( cossindata[i*2+1] ) > 120 )
printf( "CS OVF %d/%d/%d/%f\n", i, cossindata[i*2+0], cossindata[i*2+1],outbins[i] );
#endif
}
}
#endif
}