williamp/osmocom-analog
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fixed performance issue with IIR filter

Browse Source 
Passing samples with zero value (0.0) causes very slow processing of
IIR filter on my test machine 'nuedel'. I don't know why.
To solve this, I add a very small number to each input sample.
This commit is contained in:
Andreas Eversberg
2019-08-23 11:31:11 +02:00
parent f86adf4bb5
commit 31fca59294
2 changed files with 5 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
src/libfilter/iir_filter.c
View File

@@ -104,7 +104,7 @@ void iir_notch_init(iir_filter_t *filter, double frequency, int samplerate, int
filter->b1 = filter->a1;
filter->b2 = (1 - K / Q + K * K) * norm;
}
void iir_process(iir_filter_t *filter, sample_t *samples, int length)
{
double a0, a1, a2, b1, b2;
@@ -126,7 +126,8 @@ void iir_process(iir_filter_t *filter, sample_t *samples, int length)
/* process filter */
for (i = 0; i < length; i++) {
in = *samples;
/* add a small value, otherwise this loop will perform really bad on my 'nuedel' machine!!! */
in = *samples + 0.000000001;
for (j = 0; j < iterations; j++) {
out = in * a0 + z1[j];
z1[j] = in * a1 + z2[j] - b1 * out;
@@ -163,7 +164,8 @@ void iir_process_baseband(iir_filter_t *filter, float *baseband, int length)
/* process filter */
for (i = 0; i < length; i++) {
in = *baseband;
/* add a small value, otherwise this loop will perform really bad on my 'nuedel' machine!!! */
in = *baseband + 0.000000001;
for (j = 0; j < iterations; j++) {
out = in * a0 + z1[j];
#ifdef DEBUG_NAN