Closed mhaberler closed 5 years ago
it seems this started after recompiling with NFM=1 - backing that out.
That was definitely the cause for the echo.
closing.
I have used the mixing feature many times, both on AM and NFM channels and I've never have this problem.
Did it show up right after you started the program? Or was it initially working fine and the effect popped up later?
Did a restart make it disappear or not?
I had the station working fine several months, feeding liveatc I decided to try some NFM reception and recompiled with NFM=1 but did not activate any NFM channels yet, and rebooted - however, did not check for reception quality thereafter and noticed the echo only some days later reboot did definitely not cure it
if you are interested I can reproduce and give you access to the box, haberlerm at gmail dot com
With four dongles on a RPi2 you are already at the edge of GPU performance even with NFM support disabled. It has to calculate 4 x 8000 = 32000 FFT transforms per second, which is close to its limit (at the default FFT size of 512). Enabling NFM doubles the processing rate = you need 4 x 16000 = 64000 transforms per second. This won't work. You'll get choppy audio even when not using mixers.
Please try the following:
Well I rather leave this box as "production" without NFM for liveatc - not that I really need NFM right there. I can experiment on a testbox though.
thanks for clarifying the performance issue!
curious: I had the sampling rate already reduced on some channels (1.4, 1.0, 1.0, and channel 4 disabled as I have dumpvdl2 running there)
is there a suggested "downscaling" ratio for samplerate like power of two?
what exactly is the effect of reducing FFT size (besides reducing GPU needs) - more aliasing? narrower bandwidth? (sorry, not up to speed on digital radio)
Downscaling ratio is not very important. The program will produce a 8000 (or 16000) sps stream from any input sampling rate.
Lower FFT size = less FFT bins (channels) = wider channels. But if you reduce FFT size by half and reduce the sampling rate by half at the same time, then the effective channel width won't change.
thanks!
have a listen at 16:10 of http://static.mah.priv.at/public/LOWG1-Twr-App-Jan-22-2019-0730Z.mp3
there's echo - and I do not understand why, used to work just fine
note this is a mixer channel - the underlying Approach and Tower mp3's seem fine, those are mixed with balance -1.0 and 1.0 respectively:
http://static.mah.priv.at/public/LOWG1-Twr-Jan-22-2019-0730Z.mp3 http://static.mah.priv.at/public/LOWG1-App-Jan-22-2019-0730Z.mp3
platform: rpi2, cpu governor: performance 4 NESDR dongles, 4 channels in total commit 6f9ffa
puzzled - any suggestions?