fix error of taking the absolute value of 'uint64_t'
This commit is contained in:
@@ -68,7 +68,7 @@ void rxAddInput( const c16_t *input_sig,
|
||||
// Energy in one sample to calibrate input noise
|
||||
// the normalized OAI value seems to be 256 as average amplitude (numerical amplification = 1)
|
||||
const double noise_per_sample = pow(10,channelDesc->noise_power_dB/10.0) * 256;
|
||||
const int dd = abs(channelDesc->channel_offset);
|
||||
const uint64_t dd = channelDesc->channel_offset;
|
||||
const int nbTx=channelDesc->nb_tx;
|
||||
|
||||
for (int i=0; i<nbSamples; i++) {
|
||||
|
||||
Reference in New Issue
Block a user