lba.gitea/WLED
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

AR: optimization for small MCUs, part2

Browse Source 
Only run FFT when the result will be used.
Please note that this also means that `FFTTime` shown in Info is only meaningful when there is sound input and not silence.

--> To get exact FFT times, the optimization can be disabled by compiling with `-D SR_DEBUG`
This commit is contained in:
Frank 2022-09-19 14:09:45 +02:00
parent 79776ae8bc
commit d7bc6b1be7
1 changed files with 29 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
usermods/audioreactive/audio_reactive.h
View File

@ -222,7 +222,13 @@ void FFTcode(void * parameter)
// early release allows the filters (getSample() and agcAvg()) to work with fresh values - we will have matching gain and noise gate values when we want to process the FFT results. micDataReal = maxSample;
micDataReal = maxSample;
// run FFT (takes 3-5ms on ESP32)
#ifdef SR_DEBUG
if (true) { // this allows measure FFT runtimes, as it disables the "only when needed" optimization
#else
if (sampleAvg > 1) { // noise gate open means that FFT results will be used. Don't run FFT if results are not needed.
#endif
// run FFT (takes 3-5ms on ESP32, ~12ms on ESP32-S2)
#ifdef UM_AUDIOREACTIVE_USE_NEW_FFT
FFT.dcRemoval(); // remove DC offset
FFT.windowing( FFTWindow::Flat_top, FFTDirection::Forward); // Weigh data using "Flat Top" function - better amplitude accuracy
@ -247,6 +253,12 @@ void FFTcode(void * parameter)
#endif
FFT_MajorPeak = constrain(FFT_MajorPeak, 1.0f, 11025.0f); // restrict value to range expected by effects
} else { // noise gate closed - only clear results as FFT was skipped. MIC samples are still valid when we do this.
memset(vReal, 0, sizeof(vReal));
FFT_MajorPeak = 1;
FFT_Magnitude = 0.001;
}
for (int i = 0; i < samplesFFT; i++) {
float t = fabsf(vReal[i]); // just to be sure - values in fft bins should be positive any way
vReal[i] = t / 16.0f; // Reduce magnitude. Want end result to be scaled linear and ~4096 max.