AI-Powered Signal Recognition in Software Defined Radio

Artificial Intelligence and Machine Learning are transforming the world of Software Defined Radio. From automatic modulation recognition to real-time spectrum monitoring, AI is enabling capabilities that were previously impossible.

The Convergence of AI and SDR

The electromagnetic spectrum is becoming increasingly crowded. Traditional signal processing methods struggle to keep pace with the growing complexity and volume of RF signals. This is where AI and machine learning step in, offering powerful tools for:

Automatic Modulation Recognition (AMR) - Identifying signal types without prior knowledge
Signal Classification - Categorizing transmissions in real-time
Spectrum Monitoring - Detecting anomalies and unauthorized transmissions
RF Fingerprinting - Identifying specific devices by their unique RF signatures

Radio Frequency Spectrum Overview

The radio spectrum spans from 3 kHz to 3000 GHz, divided by the ITU into distinct bands:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
┌─────────────────────────────────────────────────────────────────────────────┐
│                    RADIO FREQUENCY SPECTRUM ALLOCATION                       │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                             │
│  VLF        LF         MF         HF         VHF        UHF        SHF      │
│  3-30kHz    30-300kHz  300kHz-3MHz 3-30MHz   30-300MHz  300MHz-3GHz 3-30GHz │
│  ─────────────────────────────────────────────────────────────────────────  │
│  │         │          │          │          │          │          │        │
│  │ Naval   │ AM Radio │ Amateur  │ FM/TV    │ Cellular │ Radar    │        │
│  │ Comms   │ Navigation│ Shortwave│ Air Band │ WiFi     │ Satellite│        │
│  │         │          │          │          │ GPS      │ 5G       │        │
│                                                                             │
└─────────────────────────────────────────────────────────────────────────────┘

Common Frequency Allocations

Band	Frequency Range	Common Uses
HF	3 - 30 MHz	Amateur radio, shortwave broadcast, maritime
VHF	30 - 300 MHz	FM radio, TV broadcast, aviation, marine VHF
UHF	300 MHz - 3 GHz	TV, cellular, WiFi, GPS, Bluetooth
SHF	3 - 30 GHz	Radar, satellite, 5G, microwave links
EHF	30 - 300 GHz	Radio astronomy, advanced radar, 5G mmWave

How AI Signal Recognition Works

1. Signal Capture

SDR hardware captures raw IQ (In-phase/Quadrature) data from the antenna across a wide bandwidth.

2. Feature Extraction

The AI system extracts features from the signal:

Time-domain characteristics
Frequency-domain features (FFT)
Cyclostationary features
Higher-order statistics

3. Neural Network Classification

Deep learning models (CNN, RNN, Transformers) process the features to classify:

Modulation type (AM, FM, PSK, QAM, OFDM, etc.)
Protocol identification
Signal source fingerprinting

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
┌──────────────┐    ┌─────────────┐    ┌────────────────┐    ┌──────────────┐
│   Antenna    │───▶│  SDR/ADC    │───▶│ Neural Network │───▶│ Classification│
│              │    │  IQ Data    │    │   Processing   │    │    Output    │
└──────────────┘    └─────────────┘    └────────────────┘    └──────────────┘
                           │                    │
                           ▼                    ▼
                    ┌─────────────┐    ┌────────────────┐
                    │   Feature   │    │  CNN/RNN/      │
                    │  Extraction │    │  Transformer   │
                    └─────────────┘    └────────────────┘

Companies & Solutions in AI-SDR

Commercial Leaders

DeepSig

DeepSig is a pioneer in AI-powered RF sensing. Their OmniSIG platform is described as “the world’s most advanced wireless signal detection and classification technology.”

OmniSIG - Signal detection and classification
OmniSIG Localization - Direction of arrival estimation
OmniSIG Studio - Build custom models from captured RF signals

Recently partnered with Epiq Solutions to deploy AI/ML software directly onto SDRs.

Deepwave Digital

Deepwave Digital offers the AIR-T product family - SDRs with integrated NVIDIA GPUs for real-time AI processing.

Their spectrum sensing solution can classify:

LTE / LTE Uplink
5G
WiFi
WCDMA, CDMA2K, GSM
P25, FM, Bluetooth

National Instruments / Ettus Research

NI highlights AI in SIGINT - Using COTS SDR solutions from NI and Ettus Research for signals intelligence applications.

BrainChip Holdings

Neuromorphic computing solutions using Akida technology for SDR devices - enabling signal detection, classification, and anomaly detection with ultra-low power consumption.

Open Source Projects

ATAKRR

ATAKRR on GitHub - Open-source platform for passive spectrum monitoring with:

Automatic Modulation Classification using deep learning
RF fingerprinting for device identification
Integration with ATAK (Android Team Awareness Kit)
Uses HackRF hardware (1MHz - 6GHz)

RadioML Dataset & Models

RF Modulation Classification - Different neural architectures for RF classification
Deep Learning Radio Signal Classification - Using DeepSig’s RadioML 2018.01A dataset

Panoradio SDR

Panoradio SDR offers resources on wireless signal recognition with deep learning.

Modulation Types Recognized by AI

Modern AI systems can classify numerous modulation schemes:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
┌─────────────────────────────────────────────────────────────────┐
│                  MODULATION CLASSIFICATION                       │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  ANALOG                    DIGITAL                              │
│  ────────                  ───────                              │
│  • AM (Amplitude Mod.)     • ASK (Amplitude Shift Keying)       │
│  • FM (Frequency Mod.)     • FSK (Frequency Shift Keying)       │
│  • SSB (Single Sideband)   • PSK (Phase Shift Keying)           │
│                              - BPSK, QPSK, 8PSK                 │
│                            • QAM (Quadrature Amplitude Mod.)    │
│                              - 16QAM, 64QAM, 256QAM             │
│                            • OFDM (Orthogonal FDM)              │
│                            • Spread Spectrum                    │
│                              - DSSS, FHSS                       │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

Applications

Spectrum Management

AI enables real-time monitoring of spectrum usage, detecting:

Unauthorized transmissions
Interference sources
Spectrum efficiency optimization

Electronic Warfare / SIGINT

Military and defense applications use AI-SDR for:

Threat detection and classification
Signal of interest identification
Real-time situational awareness

The U.S. Army SBIR program is actively seeking AI/ML capabilities for SDR-based spectrum characterization.

Amateur Radio

Ham radio operators can use AI tools for:

Automatic mode detection
Weak signal extraction
Band activity monitoring

IoT Security

RF fingerprinting can identify and authenticate IoT devices, detecting spoofing attempts or unauthorized equipment.

Hardware for AI-SDR

Device	Frequency Range	AI Capability
RTL-SDR	24 MHz - 1.7 GHz	External processing
HackRF One	1 MHz - 6 GHz	External processing
Epiq Matchstiq	Various	Integrated NVIDIA GPU
Deepwave AIR-T	300 MHz - 6 GHz	Integrated NVIDIA GPU
USRP (Ettus)	Various models	External/Integrated