Technical Documentation

Deep dive into MRRC system architecture, DSP processing pipeline and advanced feature implementations

πŸ—οΈ System Architecture

Overall Architecture

MRRC System Architecture V5.6.5 Client Layer Mobile UI mobile_modern .html Desktop UI modern.html CW Decode cw_live.html (ONNX Runtime) FT8 Ultron ft8_ultron.html Voice Asst mobile_voice_ PTT Safety Timeout / Optical HTTPS / WebSocket (Opus / Int16) Service Layer MRRC Main (Tornado / Python 3.12+) /WSCTRX Control + audio (PTT/freq/mode/WDSP/recording) /WSaudioRX RX audio stream (Opus/Int16, RNNoise/WDSP) /WSaudioTX TX audio upload (Opus encode, EQ chain) /WSATR1000 Tuner power/SWR real-time, learning buffer /WSFT8 FT8 WebSocket bridge (UDP 2238) /WSCW CW backend decode (PyTorch, optional) rigctld (Hamlib) IC-M710 @ 4800 baud Transceiver (IC-M710) Hardware Device ATR-1000 Proxy LearningBuffer V5.6 ATR-1000 Tuner Power / SWR Meter FT8 / ULTRON Remote Console (V5.6+) Browser UI: waterfall, 6 TX slots, auto-sequencer, DXCC tracking Server: QSOSequencer, ADIF logging, DXCC DB lookup | Bridge: JTDX/WSJT-X/MSHV via UDP 2237 to WS Bridge 2238 PTT Safety System (V5.6+) L1 Browser: touch-move release, visibility-change stop, 30s watchdog L2 Protocol: ACK retry x3 + redundant audio-path fallback | L3 Server: 120s TOT hard limit, persistent 2s retry

Core Components

Component File Function
Main Program MRRC WebSocket server, handles audio streams and control commands
Audio Interface audio_interface.py PyAudio wrapper, supports multiple format decoding
DSP Processing wdsp_wrapper.py WDSP library wrapper, NR2/AGC/ANF
Transceiver Control hamlib_wrapper.py rigctld communication wrapper
Tuner Proxy atr1000_proxy.py ATR-1000 independent proxy process
Voice Assistant voice_assistant_service.py Whisper ASR + Qwen3-TTS
FT8 Integration ft8/ultron.py ULTRON automation tool, DXCC tracking

πŸ”Š Audio Processing Pipeline

Receive Audio Chain (RX)

RX Audio Processing Pipeline Transceiver Audio (48kHz Float32) DC Removal + Soft Clipping Protect downstream processing Int16 Convert 48kHz to 16-bit (Optional) WDSP Processing Spectral NR / Noise Blank / Auto Notch / AGC NR2 / NB / ANF / AGC Gain 15–20 dB noise reduction Opus Encode 16kHz / 20kbps WebSocket Transport Browser client delivery AudioWorklet Low-latency playback Audio Buffer

Transmit Audio Chain (TX)

TX Audio Processing Pipeline Microphone Input (48kHz) Anti-Alias Filter 6kHz lowpass Downsample 48kHz to 16kHz TX Equalizer 3-band EQ presets Opus Decode 16kHz to 48kHz WebSocket Transport Send audio to server Transceiver Input CAT PTT control

πŸŽ›οΈ WDSP Digital Signal Processing

WDSP (Warren Pratt's Digital Signal Processing) is a high-performance DSP library from the OpenHPSDR project, widely used in professional amateur radio software.

Feature Modules

Module Function Effect
NR2 Spectral Noise Reduction 15-20dB noise reduction depth
NB Noise Blanker Eliminates impulse interference
ANF Automatic Notch Filter Removes single-tone interference
AGC Automatic Gain Control 4 selectable modes

AGC Modes

  • LONG: Long delay, suitable for stable signals
  • SLOW: Slow response, balanced choice
  • MED: Medium response, default recommendation
  • FAST: Fast response, suitable for rapid QSY

πŸ“‘ CW Decoding System Architecture

MRRC V4.9 introduces a complete CW real-time decoding system with a Browser-First AI design philosophy.

Dual-Mode Decoding Architecture

CW Decoding Dual-Mode Architecture Frontend Mode (Recommended) Audio Stream 16kHz ONNX Runtime 2MB model CTC Decode QSO State Machine Smart Reply Latency: <50ms | Dependencies: None (pure browser) Backend Mode (Alternative) WebSocket Client /WSCW morseangel PyTorch backend Text Output Latency: <100ms | Dependencies: PyTorch + CUDA

Technical Specifications

Metric Frontend Mode Backend Mode
Model Size 2MB (ONNX) 160MB (PyTorch)
Inference Latency <50ms <100ms
Character Error Rate <3% <2%
Network Dependency None Required

Try It Live

CW Real-time Decode CW Test Page

πŸ“Š ATR-1000 Integration

System Architecture

ATR-1000 Integration Architecture Frontend mobile_modern.js WebSocket /WSATR1000 MRRC Main Message routing Unix Socket atr1000_proxy Independent Process WebSocket ATR-1000 Device 192.168.1.63:60001 Power / SWR / Relay Frontend (browser) connects via WebSocket to MRRC Main, which routes commands through Unix Socket to atr1000_proxy. atr1000_proxy communicates with ATR-1000 hardware over TCP and pushes real-time power/SWR/relay data back via WebSocket broadcasts.

Smart ATU Learning

  • Auto Learning: Records frequency-to-tuner parameter mapping during each transmission
  • Quick Tune: Automatically applies learned parameters when frequency changes
  • Persistence: Saves to atr1000_tuner.json

πŸ“» FT8/ULTRON Integration

ULTRON is MRRC's FT8 digital mode remote console (introduced V4.9, overhauled V5.6), providing a full-featured browser-based interface to JTDX/WSJT-X/MSHV via WebSocket bridge.

Core Features

Feature Description
Auto CQ Automatically call CQ when idle
Smart Response Automatically respond to incoming calls
DXCC Tracking Entity whitelist, intelligent selection
Band Strategy Different targets per band
Log Analysis Automatically analyze worked entities

Run Commands

# Enter FT8 directory
cd ft8

# Run ULTRON (Python version)
python run_ultron.py

# DXCC enhanced mode
python ultron_dxcc.py

# Analyze DXCC status
python dxcc_analyzer.py

Try It Live

FT8 Console

πŸ“ Communication Protocols

WebSocket Commands (WSCTRX)

Command Format Description
Set FrequencysetFreq:14074000In Hz, syncs to ATR-1000
Get FrequencygetFreqRead from rig, broadcast response
Set ModesetMode:USBUSB/LSB/CW/FM/AM
Get ModegetModeRead mode from rig
PTT TXsetPTT:trueKey transmitter (with retry)
PTT RXsetPTT:falseRelease (with ACK protocol, 3x retry)
Get PTTgetPTTRead actual PTT state from hardware
Tunetune:true / tune:falseStart/stop tune tone (5% volume)
CQcq:true / cq:falseStart/stop CQ playback (cq.wav)
WDSP TogglesetWDSPEnabled:trueEnable/disable WDSP processing
NR2 LevelsetWDSPNR2Level:20=OFF, 1=LOW, 2=MED, 3=HIGH, 4=MAX
Noise BlankersetWDSPNB:trueToggle NB (pulse interference)
Auto NotchsetWDSPANF:trueToggle ANF automatic notch filter
Add NotchaddWDSPNotch:1500,200Manual notch (center_hz, width_hz)
AGC ModesetWDSPAGC:30=OFF, 1=LONG, 2=SLOW, 3=MED, 4=FAST
BandpasssetWDSPBandpass:300,2700Low Hz, High Hz
RecordingstartRecording / stopRecordingStereo WAV β†’ MP3 recording
MemorymemLoadAll / memSaveAll:{json}Per-user memory channels (6 slots)
PingPINGLatency check, responds PONG

Need More Help?

Check the installation guide or join community discussion

Installation Guide GitHub Issues