14 Jan
2018
14 Jan
'18
2:22 p.m.
FWIW, I've been running some benchmarks on the RPI3 using multi_rx. As a preliminary
result it looks like the PI can demod four DMR frequencies (times two slots/frequency =
eight voice channels). I'd say that's right at the margin and there were a few
underruns reported (more or less expected once the threshold is approached given that all
four are running at full blast). There were also a couple of crashes noted, intermittent
of course. It's not yet clear what the fix for them might be. Most of the CPU time
seems to be in the voice codec which is appropriate given that's what the whole point
of the app is. There are two codecs in OP25 and currently multi_rx only supports the
floating point one, Pavel's fixed point implementation *might* be moire efficient.
It should be noted that the codecs have never been a target of any optimization work;
it's likely there is room for improvement - not without effort of course...
73
Max