Machine learning empowered Modulation detection for OFDM-based signals

TL;DR

This work tackles blind modulation detection for OFDM-based systems under realistic impairments by using a ResNet to classify modulation from FFT scatter plots after CP and CFO effects are mitigated. The approach jointly estimates the number of subcarriers $N$, CP size, and coarse timing, then removes CFO before constructing 400×400 scatter plots of $\operatorname{Im}\{X\}$ vs $\operatorname{Re}\{X\}$ to feed a ResNet for modulation and CP-type labeling. Training data covers six modulations (BPSK, QPSK, 16/64/256/1024-QAM) across three CP scenarios, with $N$ from 128 to 2048, CFO$\in[100,500]$ ppm, and $10$–$25$ dB noise in a SUI I channel, using cross-entropy loss. Simulation results show high accuracy for low-to-moderate order modulations and progressively better performance as SNR increases, achieving >$80\%$ at 10 dB and >$95\%$ at 25 dB, while also handling CP location errors and symbol-by-symbol modulation changes. This method enables practical, environment-resilient blind modulation recognition for OFDM systems without prior transmitter information.

Abstract

We propose a blind ML-based modulation detection for OFDM-based technologies. Unlike previous works that assume an ideal environment with precise knowledge of subcarrier count and cyclic prefix location, we consider blind modulation detection while accounting for realistic environmental parameters and imperfections. Our approach employs a ResNet network to simultaneously detect the modulation type and accurately locate the cyclic prefix. Specifically, after eliminating the environmental impact from the signal and accurately extracting the OFDM symbols, we convert these symbols into scatter plots. Due to their unique shapes, these scatter plots are then classified using ResNet. As a result, our proposed modulation classification method can be applied to any OFDM-based technology without prior knowledge of the transmitted signal. We evaluate its performance across various modulation schemes and subcarrier numbers. Simulation results show that our method achieves a modulation detection accuracy exceeding $80\%$ at an SNR of $10$ dB and $95\%$ at an SNR of $25$ dB.

Figures (6)

• Figure 1: Constellation structure before and after the effects of CFO under multi-path fading and noise.
• Figure 2: CP scenarios.
• Figure 3: Network model architecture.
• Figure 4: Test process.
• Figure 5: Detection accuracy as a function of the SNR.