BSARC 20-Minute Technical Session

AI & Modern DSP In The Radio Shack

Extracting Weak Signals & Overcoming Local RFI with Neural DSP

Brunswick Sciences & Amateur Radio Club • Presented 12-Aug-2026
1. The Shifting RF Noise Landscape
Background & Problem
Traditional DSP Limits
  • Fixed Filtering: Standard IIR/FIR and crystal filters rely on fixed bandwidths and mathematical notches.
  • Static Assumptions: Legacy noise blankers work well on steady ignition pulses but fail against complex, modern household EMI.
  • Audio Artifacts: Heavy hardware DSP creates hollow, "underwater" audio artifacts on weak voice signals.
The AI & Neural DSP Paradigm
  • Data-Driven Recognition: Neural networks are trained on thousands of hours of speech and RF noise samples.
  • Non-Stationary RFI Removal: Instantly adapts to shifting solar inverters, LED switching supplies, and HVAC blowers.
  • Formant Reconstruction: Subtracts the non-speech noise background while preserving human vocal clarity.
2. Neural Audio Suppression
Voice Isolation Engines
How Neural Filters Work
  • Deep Learning Models: Tools like RM Noise and DeepFilterNet use real-time spectral masking.
  • Sub-20ms Latency: Runs locally on standard shack PCs without delay, enabling full-duplex SSB ragchewing.
  • 25–30 dB Attenuation: Drops broadband hash to zero while leaving buried voice formants intelligible.
Live Audio Path Setup
[Radio Audio Out]
↓ (USB / 3.5mm Aux Cable)
[PC Sound Input]
↓
[AI DSP Neural Filter Software]
↓
[Shack Headphones / Speaker]
3. Spectrum Computer Vision
Waterfall Intelligence
Automated Signal Detection
  • CNN Spectrum Scanners: Convolutional Neural Networks analyze SDR waterfall images in real-time.
  • Mode Classification: Instantly identifies FT8, VaraC, CW, RTTY, and SSB traces with >90% accuracy.
  • Automated Routing: Automatically tunes receivers and launches corresponding decoders (WSJT-X, FLdigi).
Benefits for the Shack
  • Wideband Monitoring: Continuously monitors entire bands for rare DX without manual scanning.
  • Accessibility Aid: Assists hams with visual or hearing limitations by highlighting faint signals visually.
4. ML-Powered Propagation Models
Dynamic Predictions
Legacy VOACAP vs. AI Forecasting
  • Legacy Limits: Classic models rely on smoothed monthly sunspot averages and miss 15-minute band openings.
  • Real-Time Data Streams: AI aggregators pull live telemetry from RBN (Reverse Beacon Network), PSKreporter, and WSPR.
  • Space Weather Integration: Blends live geomagnetic (Kp and Ap) indices and solar wind satellite feeds.
Hyper-Local Openings
  • Calculates exact propagation probability for specific station-to-station paths.
  • Tells operators precisely when and where to move bands for 10m–15m openings or nighttime 40m low-band DX.
5. Case Study: HOA Attic Antennas
Overcoming Indoor EMI
The HOA Attic Antenna Challenge
  • Submerged in Noise: Attic EFHW/dipoles live inside the home's RFI envelope (LEDs, solar, HVAC).
  • The S8-S9 Hash: Weak S3–S4 DX signals are completely buried under broadband switching noise.
The AI Neural Solution
  • 25–30 dB Noise Reduction: Neural network drops non-speech hash to total silence.
  • Result: Extracts the S3 voice signal clearly without altering the radio's RF performance.
6. Shack Integration: FTDX10 vs. Legacy Rig
Hardware Interface Guide
Modern Rig (e.g., Yaesu FTDX10)
Single USB Cable ➔ USB AUDIO CODEC
  • Connection: Single USB cable handles CAT control and 24-bit audio stream.
  • Menu Config: FUNC ➔ RADIO SETTING ➔ USB OUT LEVEL (Set to 30%–40%).
  • Layered DSP: Combine FTDX10 roofing filters/DNR with PC AI processing.
Legacy Analog Rig (e.g., TS-450)
3.5mm Aux Cable ➔ PC Line In / Sound Card
  • Connection: 3.5mm cable from Headphone jack or rear ACC port to PC Line In.
  • Gain Staging: Set radio volume to 9–10 o'clock to avoid clipping PC input.
  • Replaces Missing DSP: Brings modern AI noise reduction to 30-year-old radios.
7. Quick-Start Shack Implementation
3-Step Setup
Software Workflow
  • Step 1 (Audio Input): Select Line (USB AUDIO CODEC) or Line In as input in AI software.
  • Step 2 (AI Engine): Launch RM Noise or DeepFilterNet and set slider to 100% filtered.
  • Step 3 (Listen): Route clean output to PC headphones or external shack speaker.
Operating Strategy for RFI
  • 1. ATT / RF Gain: Back off RF Gain slightly if local noise pegs the S-meter.
  • 2. Clean Levels: Ensure audio reaching the PC peaks between -12 dB and -6 dB.
  • 3. Engage AI: Enjoy whisper-quiet background audio on weak SSB contacts.
8. Club Discussion & Q&A
Wrap-Up
Prompts for the Room
  • 1. Local Noise Sources: What indoor/neighborhood RFI sources are giving your shack the most trouble right now?
  • 2. Operator Skill vs. Automation: Where do we draw the line between operator filtering skill and software automation?
  • 3. Compromised Antennas: How many members are currently running stealth/attic antennas in HOA areas?
Slide 1 of 9