updates
This commit is contained in:
Logan
@@ -1,12 +1,10 @@
|
||||
# DRB Client (Edge Node)
|
||||
|
||||
Software stack for a DRB (Discord Radio Bot) edge node. Runs on the SDR hardware machine. Decodes P25/DMR/NBFM radio with OP25, streams decoded audio via Icecast, and streams live audio into Discord voice channels on demand.
|
||||
Software stack for a DRB (Discord Radio Bot) edge node. Runs on the SDR hardware machine. Decodes P25/DMR/NBFM radio with OP25, routes decoded audio through PulseAudio for low-latency Discord streaming and call recording, and secondarily streams to Icecast for browser listening.
|
||||
|
||||
## System Overview
|
||||
## Audio Architecture
|
||||
|
||||
An edge node is a machine (typically a small Linux box or Raspberry Pi) with an RTL-SDR dongle attached. It connects to the central DRB C2 server over MQTT and waits for a radio system config. Once configured, it decodes transmissions and streams audio. When a Discord user runs `/join`, the server sends the node a `discord_join` command and the node puts a Discord bot in the user's voice channel that plays the live Icecast stream.
|
||||
|
||||
The Discord bot's speaking ring (green animated border) **only lights up when radio is actively transmitting** — not while it's silently waiting in the channel.
|
||||
**PulseAudio is the primary audio path. Icecast is for browser listening only.**
|
||||
|
||||
```
|
||||
RTL-SDR dongle
|
||||
@@ -14,22 +12,49 @@ RTL-SDR dongle
|
||||
▼
|
||||
op25-container ←── GnuRadio + OP25 + Liquidsoap
|
||||
│ decodes P25/DMR/NBFM
|
||||
│ streams MP3 audio to Icecast on port 8000
|
||||
│ HTTP API on port 8001 (start/stop/generate-config)
|
||||
│ HTTP terminal on port 8081 (live talkgroup metadata)
|
||||
│
|
||||
▼
|
||||
icecast ←── Icecast2
|
||||
│ serves audio at http://localhost:8000/radio
|
||||
├──► PulseAudio (drb_sink null sink) ← PRIMARY audio path
|
||||
│ ~250ms latency, synchronized with OP25 call events
|
||||
│ shared via named Docker volume (pulse_socket:/run/pulse)
|
||||
│
|
||||
▼
|
||||
└──► Icecast on port 8000 ← BROWSER STREAMING ONLY
|
||||
serves http://localhost:8000/radio
|
||||
2-5s buffering lag — NOT suitable for recording or Discord
|
||||
```
|
||||
|
||||
```
|
||||
PulseAudio (drb_sink.monitor — default source)
|
||||
│
|
||||
├──► drb-edge-node: FFmpeg call recorder
|
||||
│ records each call to MP3, uploads to C2 server
|
||||
│ aligned with OP25 call_start / call_end events
|
||||
│
|
||||
└──► drb-edge-node: discord.py bot
|
||||
streams live audio into Discord voice channel
|
||||
continuous stream while bot is in channel
|
||||
```
|
||||
|
||||
> **Why not Icecast for recording/Discord?**
|
||||
> Icecast introduces 2-5 seconds of pipeline buffering that grows over time. Since
|
||||
> call_start and call_end events fire in real-time from OP25 RF detection, using
|
||||
> Icecast as the recording source causes recordings to miss the beginning and end
|
||||
> of transmissions — especially on short calls (< 8s). PulseAudio has ~250ms
|
||||
> latency and stays locked to the live audio position.
|
||||
|
||||
## System Overview
|
||||
|
||||
An edge node is a machine (typically a small Linux box or Raspberry Pi) with an RTL-SDR dongle attached. It connects to the central DRB C2 server over MQTT and waits for a radio system config. Once configured, it decodes transmissions and streams audio. When a Discord user runs `/join`, the server sends the node a `discord_join` command and the node puts a Discord bot in the user's voice channel that plays the live PulseAudio stream.
|
||||
|
||||
```
|
||||
drb-edge-node ←── FastAPI
|
||||
│ polls OP25 terminal every 0.5s — detects call start/end events
|
||||
│ publishes call metadata + status to C2 server over MQTT
|
||||
│ records each call with FFmpeg, uploads recording to C2
|
||||
│ records each call with FFmpeg from PulseAudio, uploads to C2
|
||||
│ receives discord_join/leave commands over MQTT
|
||||
│ starts Discord bot (discord.py), joins voice channel
|
||||
│ starts/stops FFmpeg→Discord stream on OP25 call events
|
||||
│ streams PulseAudio continuously into Discord while bot is connected
|
||||
│ serves local web dashboard on port 8080
|
||||
│
|
||||
▼
|
||||
@@ -43,9 +68,9 @@ C2 Server (remote — over MQTT + HTTP)
|
||||
|
||||
| Service | Container | Description |
|
||||
|---|---|---|
|
||||
| `icecast` | Icecast2 (Debian) | Audio streaming server — receives encoded audio from OP25/Liquidsoap, serves it over HTTP |
|
||||
| `op25` | OP25 + GnuRadio (Debian) | SDR decoder — tunes to a radio system, decodes P25/DMR/NBFM, streams via Liquidsoap to Icecast |
|
||||
| `edge-node` | Python 3.11 / FastAPI | Node agent — bridges OP25 metadata, C2 MQTT commands, call recording, and Discord voice |
|
||||
| `op25` | OP25 + GnuRadio (Debian) | SDR decoder — tunes to a radio system, decodes P25/DMR/NBFM, outputs audio to PulseAudio (primary) and Icecast (browser) via Liquidsoap |
|
||||
| `edge-node` | Python 3.11 / FastAPI | Node agent — bridges OP25 metadata, C2 MQTT commands, call recording from PulseAudio, and Discord voice streaming from PulseAudio |
|
||||
| `icecast` | Icecast2 (Debian) | Browser audio streaming only — serves the stream at `http://<node>:8000/radio` for local/remote listening. Not used for recording or Discord. |
|
||||
|
||||
All services run with `network_mode: host` so they share the host network and reach each other on `localhost` without any port mapping.
|
||||
|
||||
@@ -71,8 +96,8 @@ Client/
|
||||
│ │ ├── mqtt_manager.py # MQTT: checkin/status/metadata pub + command/config/apikey sub
|
||||
│ │ ├── metadata_watcher.py # Polls OP25 terminal at 0.5s — fires on_call_start/on_call_end
|
||||
│ │ ├── op25_client.py # HTTP client to op25-container API (start/stop/status/config)
|
||||
│ │ ├── discord_radio.py # discord.py bot — joins voice, streams Icecast, @mention commands
|
||||
│ │ ├── call_recorder.py # FFmpeg recording from Icecast per call, uploads to C2
|
||||
│ │ ├── discord_radio.py # discord.py bot — joins voice, streams PulseAudio, @mention commands
|
||||
│ │ ├── call_recorder.py # FFmpeg recording from PulseAudio per call, uploads to C2
|
||||
│ │ ├── credentials.py # Persists C2-issued API key to /configs/credentials.json
|
||||
│ │ ├── config_manager.py # Reads/writes /configs/node_config.json
|
||||
│ │ └── logger.py # Stdout logging setup
|
||||
@@ -176,10 +201,10 @@ The C2 server translates these into OP25's `active.cfg.json` format (with Hz fre
|
||||
2. Server sends a `discord_join` command to c2-core with the target guild ID, channel ID, and an optional preferred token.
|
||||
3. c2-core picks a free token from the pool, marks it `in_use`, and forwards everything to the edge node over MQTT.
|
||||
4. Edge node starts a `discord.py` bot with that token and connects it to the voice channel.
|
||||
5. The bot **sits silently** in the channel until radio is active (no speaking ring while idle).
|
||||
6. When OP25 detects an active call → `start_stream()` → FFmpegPCMAudio opens the Icecast URL → Discord speaking ring lights up.
|
||||
7. When the call ends (1s of inactivity on the talkgroup) → `stop_stream()` → FFmpeg stops → ring goes dark.
|
||||
8. On `/leave`, the bot disconnects and the token is released back to the pool.
|
||||
5. The bot immediately begins streaming from PulseAudio (`drb_sink.monitor`) — continuous stream while connected.
|
||||
6. On `/leave`, the bot disconnects and the token is released back to the pool.
|
||||
|
||||
**Audio source:** PulseAudio (`-f pulse "default"` → `drb_sink.monitor`). ~250ms latency, in sync with live RF. Icecast is **not** used for Discord streaming.
|
||||
|
||||
**Direct bot commands:**
|
||||
|
||||
@@ -212,11 +237,13 @@ The `configs` volume is critical — without it, the node forgets its API key an
|
||||
|
||||
## Call Recording
|
||||
|
||||
Every call is recorded via FFmpeg pulling from the Icecast stream. Recordings are:
|
||||
Every call is recorded via FFmpeg reading from PulseAudio (`drb_sink.monitor`, the default source). Recordings are:
|
||||
- Saved to `/recordings/YYYYMMDD_HHMMSS_<call_id>.mp3`
|
||||
- Uploaded to the C2 server (if `C2_URL` is set) using the node's API key as a Bearer token
|
||||
- Capped at 600 seconds for safety
|
||||
|
||||
**Audio source:** PulseAudio (`-f pulse -i default`). Recording starts the moment `call_start` fires from OP25, capturing the full transmission including short calls (< 8s). Icecast is **not** used for recording — its 2-5s buffering lag causes recordings to miss the start of calls and produce empty files for short transmissions.
|
||||
|
||||
## Makefile Targets
|
||||
|
||||
```
|
||||
|
||||
@@ -18,6 +18,8 @@ services:
|
||||
- /dev:/dev
|
||||
- ./op25-container/app:/app
|
||||
- pulse_socket:/run/pulse
|
||||
environment:
|
||||
PULSE_SERVER: unix:/run/pulse/native
|
||||
depends_on:
|
||||
- icecast
|
||||
|
||||
|
||||
@@ -27,15 +27,13 @@ class CallRecorder:
|
||||
self._current_file = self._recordings_dir / f"{ts}_{call_id}.mp3"
|
||||
self._current_call_id = call_id
|
||||
|
||||
# Read from the Icecast stream. burst-size=0 in icecast config ensures
|
||||
# each new connection starts at the live position with no buffered data.
|
||||
stream_url = f"http://{settings.icecast_host}:{settings.icecast_port}{settings.icecast_mount}"
|
||||
# Record from PulseAudio (drb_sink.monitor, set as the default source).
|
||||
# PulseAudio is the primary audio path — it has ~250ms latency vs 2-5s for Icecast,
|
||||
# so recordings align with OP25 call_start/call_end events.
|
||||
cmd = [
|
||||
"ffmpeg", "-y",
|
||||
"-reconnect", "1",
|
||||
"-reconnect_streamed", "1",
|
||||
"-reconnect_delay_max", "5",
|
||||
"-i", stream_url,
|
||||
"-f", "pulse",
|
||||
"-i", "default",
|
||||
"-acodec", "libmp3lame",
|
||||
"-ar", "22050",
|
||||
"-b:a", "32k",
|
||||
|
||||
@@ -101,11 +101,11 @@ class RadioBot:
|
||||
def _play_stream(self):
|
||||
if not self._voice_client:
|
||||
return
|
||||
from app.config import settings
|
||||
stream_url = f"http://{settings.icecast_host}:{settings.icecast_port}{settings.icecast_mount}"
|
||||
# Stream from PulseAudio (drb_sink.monitor, the default source).
|
||||
# ~250ms latency vs 2-5s for Icecast — stays in sync with live transmissions.
|
||||
source = discord.FFmpegPCMAudio(
|
||||
stream_url,
|
||||
before_options="-reconnect 1 -reconnect_streamed 1 -reconnect_delay_max 5 -probesize 32 -analyzeduration 0 -fflags nobuffer",
|
||||
"default",
|
||||
before_options="-f pulse",
|
||||
)
|
||||
self._voice_client.play(
|
||||
discord.PCMVolumeTransformer(source, volume=1.0),
|
||||
|
||||
@@ -27,18 +27,25 @@ input = compress(input, attack = 2.0, gain = 0.0, knee = 13.0, ratio = 2.0, rele
|
||||
input = normalize(input, gain_max = 6.0, gain_min = -6.0, target = -16.0, threshold = -65.0)
|
||||
|
||||
# ==========================================================
|
||||
# OUTPUT: Referencing the new variables
|
||||
# OUTPUTS
|
||||
# ==========================================================
|
||||
|
||||
# Primary: PulseAudio null sink — used by edge-node for Discord streaming and call recording.
|
||||
# The edge-node reads from the drb_sink.monitor source (set as PulseAudio default source).
|
||||
output.pulseaudio(mean(input))
|
||||
|
||||
# Secondary: Icecast — for browser-based listening only (local node UI / server stream page).
|
||||
# Do NOT use Icecast for recording or Discord; it introduces 2-5s buffering lag.
|
||||
output.icecast(
|
||||
%mp3(bitrate=16, samplerate=22050, stereo=false),
|
||||
description=ICE_DESCRIPTION,
|
||||
genre=ICE_GENRE,
|
||||
url="",
|
||||
fallible=false,
|
||||
host=ICE_HOST,
|
||||
port=ICE_PORT,
|
||||
mount=ICE_MOUNT,
|
||||
password=ICE_PASSWORD,
|
||||
%mp3(bitrate=16, samplerate=22050, stereo=false),
|
||||
description=ICE_DESCRIPTION,
|
||||
genre=ICE_GENRE,
|
||||
url="",
|
||||
fallible=false,
|
||||
host=ICE_HOST,
|
||||
port=ICE_PORT,
|
||||
mount=ICE_MOUNT,
|
||||
password=ICE_PASSWORD,
|
||||
mean(input)
|
||||
)
|
||||
"""
|
||||
@@ -4,6 +4,8 @@
|
||||
# The -D flag starts it as a daemon.
|
||||
# The --exit-idle-time=-1 prevents it from automatically shutting down.
|
||||
echo "Starting PulseAudio daemon..."
|
||||
mkdir -p /run/pulse
|
||||
chmod 777 /run/pulse
|
||||
pulseaudio -D --exit-idle-time=-1 --system
|
||||
|
||||
# Wait a moment for PulseAudio to initialize
|
||||
|
||||
@@ -1,16 +1,22 @@
|
||||
# PulseAudio system daemon configuration for DRB containers.
|
||||
# Allows anonymous connections so the edge-node container can read the audio monitor.
|
||||
#
|
||||
# Audio architecture:
|
||||
# Liquidsoap (op25 container) → output.pulseaudio() → drb_sink (null sink)
|
||||
# FFmpeg (edge-node container) → -f pulse -i default → drb_sink.monitor
|
||||
#
|
||||
# Icecast is a secondary output for browser listening only.
|
||||
# PulseAudio is the primary path for Discord streaming and call recording.
|
||||
|
||||
.nofail
|
||||
|
||||
# Native protocol with anonymous auth — required for inter-container access
|
||||
load-module module-native-protocol-unix socket=/run/pulse/native auth-anonymous=1
|
||||
|
||||
# Always provide a default sink even without audio hardware
|
||||
load-module module-always-sink
|
||||
# Named null sink — Liquidsoap outputs here; edge-node records from its monitor
|
||||
load-module module-null-sink sink_name=drb_sink sink_properties=device.description=DRB-Sink
|
||||
|
||||
# Restore device/stream settings across restarts
|
||||
load-module module-device-restore
|
||||
load-module module-stream-restore
|
||||
# Make the sink and its monitor the system defaults
|
||||
set-default-sink drb_sink
|
||||
set-default-source drb_sink.monitor
|
||||
|
||||
.fail
|
||||
|
||||
Reference in New Issue
Block a user