This little script runs multiple FM or AM demodulators and feeds the results to individual icecast MP3 streams. It can be used to feed all the 2-meter VHF repeaters to an icecast server with a single rtlsdr USB stick.

It's a bit unpolished, but it sort of works for me. I'm a bit busy right now, so if you have a bug to report, please fix the bug first and do a pull request on github, I'll be happy to merge your fix. Thanks!

• An user who is familiar with python and gnuradio.
• A working gnuradio installation, with rtlsdr drivers working.
• The 'lame' MP3 encoder needs to be installed, too.

python gr-multistream.py --freqset RNET-OH2RCH=145787.5,APRS-144800=144800 \
    --icecast icecast.example.com --icepw passwordhere

--freqset defines a comma-separated list of name-frequency pairs. Frequency (144800 in the above example) is given in kilohertz, floating point values should be accepted. The name (APRS-144800) is the stream name used for the Icecast server - you might want to embed the frequency in the stream name to make it visible on Icecast.

Icecast TCP server port is currently hardcoded as 8000. Sorry for that. But it's a script, so it's not really hardcoded.

Other parameters:

• --dev N specifies the N-th (0-based) device to be used. --dev 0 is the first hardware device, --dev 1 is the second stick.
• --gain 20 sets hardware frontend gain to 20 dB.
• --mode am switches to AM.
• --audio-output "Audio device name" enables mixdown of all demodulated streams, and output of the whole mess to an audio device.

The squelch isn't quite right or optimal. For FM, it really should set up a proper traditional high-pass-filter + level metering squelch.

It eats a lot of CPU. The filters could probably be adjusted to do a "good enough" job with "low enough" CPU consumption.

The rtlstr stick now runs at 2 MHz / 2 Msamples/sec rate all the time. Could use a lower sample rate if the actual required bandwidth is lower. My stick runs really hot (measured 80C with a fluke IR meter!), maybe lowering the sample rate helps.

It creates a named pipe for every stream, it's used for feeding the audio samples to the MP3 encoder. That probably only works nicely on Unix-like systems. I was in a hurry, and didn't get around to figuring out how to implement a custom sink block in Python, which in turn could be used to implement a pipe to stdin of an external program. A 'gr-pipe' module example exists, but it's not in the main gnuradio distribution. So I just used the file sink block to write in the named pipe.

The named pipes are created in current working directory, and they're not even deleted automatically when the program quits.