Deep Learning-Based Modulation Classification Using Software-Defined Radios (USRP)

Öztürk G., Yilmaz R.

8TH INTERNATIONAL CONFERENCE ON APPLIED ENGINEERING AND NATURAL SCIENCES ICAENS 2025 August 22-23, 2025, KONYA, TURKEY, Konya, Türkiye, 22 - 23 Ağustos 2026, ss.83, (Özet Bildiri)

  • Yayın Türü: Bildiri / Özet Bildiri
  • Basıldığı Şehir: Konya
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.83
  • Dokuz Eylül Üniversitesi Adresli: Evet

Özet

– This study presents an end-to-end system for Automatic Modulation Classification (AMC) based

on deep learning techniques and Software-Defined Radios (SDRs). Using Universal Software Radio

Peripheral (USRP) devices, the study bridges the gap between simulation-based studies and real-world

signal environments. Three deep learning architectures—Convolutional Neural Networks (CNN), Long

Short-Term Memory networks (LSTM), and Multilayer Perceptrons (MLP)—are trained and evaluated on

cumulant-based and phase-based features. A real-time classification system is developed and deployed

across two synchronized USRP devices communicating over-the-air. Experimental results indicate robust

classification performance, achieving over 90% accuracy for digital modulation schemes such as QPSK,

16QAM, 64QAM, and 8PSK under various SNR and channel conditions.