audiosock-broadcast: audio from a unix socket

Reads raw audio from a unix socket and puts it on the air, modulated through csdr and sent with rpitx’s sendiq. Point any audio source at the socket: a file decoded by ffmpeg, a microphone, a program that generates tones. Code: audiosock_broadcast.go, scripts/audiosock_broadcast.sh and scripts/modulation.sh.

This is a script module. It needs socat and csdr on the Pi, and sendiq in GORPITX_PATH.

Fields

JSON	Go field	Type	Required	Rules
socketPath	SocketPath	string	yes	Path of the unix socket to read from
frequency	Frequency	float64	yes	Hz, 5000 to 1500000000
sampleRate	SampleRate	*int	no	Audio sample rate in Hz, greater than 0; gorpitx sends 48000 when it is not set
modulation	Modulation	*string	no	AM, DSB, USB, LSB, FM or RAW (the gorpitx.Modulation* constants); gorpitx sends FM when it is not set
gain	Gain	*float64	no	Multiplier applied to the audio, 0 or more; gorpitx sends 1 when it is not set

gorpitx does not check that the socket exists. The script does, when it starts, and exits with status 1 and ERROR: Unix socket <path> does not exist if it does not.

The socket and the audio format

The script connects to the socket as a client (socat UNIX-CONNECT:<socketPath> STDOUT), so some other process has to be listening on it and writing audio. A named pipe made with mkfifo does not work: the script checks for a socket and refuses anything else.

The audio must be raw PCM with no header: signed 16-bit little-endian, mono, at sampleRate.

Start the source first, then call Exec. Stream a file with ffmpeg and socat:

ffmpeg -i music.mp3 -f s16le -ar 48000 -ac 1 - \
  | socat -u - UNIX-LISTEN:/tmp/audio.sock

Or a microphone with ALSA:

arecord -f S16_LE -r 48000 -c 1 -t raw \
  | socat -u - UNIX-LISTEN:/tmp/audio.sock

The transmission ends when the source closes the connection, or when Exec times out or is stopped.

Modulations

modulation.sh turns the audio into I/Q samples for sendiq with one csdr chain per mode:

Mode	csdr chain after convert_s16_f and gain_ff
FM	fmmod_fc
AM	dsb_fc, add_dcoffset_cc, agc_ff
DSB	dsb_fc, agc_ff
USB	dsb_fc, bandpass_fir_fft_cc 0.002 0.06 0.01, agc_ff
LSB	dsb_fc, bandpass_fir_fft_cc -0.06 -0.002 0.01, agc_ff
RAW	nothing more: conversion and gain only, no modulation, no AGC

USB and LSB are slow on a Pi Zero. The FFT bandpass filter adds latency, artifacts and dropouts there. Use DSB instead: it sends both sidebands, so a receiver in USB or LSB mode hears it.

Example

{
  "socketPath": "/tmp/audio.sock",
  "frequency": 14200000,
  "modulation": "DSB",
  "gain": 2
}

Runs:

stdbuf -oL /tmp/audiosock_broadcast.sh 14200000 /tmp/audio.sock 48000 DSB 2

with RPITX_PATH set to GORPITX_PATH. The script then runs:

socat UNIX-CONNECT:/tmp/audio.sock STDOUT \
  | /tmp/modulation.sh DSB 2 \
  | "$RPITX_PATH/sendiq" -i /dev/stdin -s 48000 -f 14200000 -t float

Side effects

• The script appends every run to /tmp/audiosock_broadcast.log as well as printing it.
• On SIGINT or SIGTERM it runs pkill -f sendiq and pkill -f csdr. That kills every process on the Pi whose command line contains those words, not only its own.