A unified deep learning framework addresses narrowband interference (NBI) in OFDM systems by jointly performing interference cancellation and robust soft demodulation. The approach replaces conventional compressed-sensing methods with two specialized networks: NBI-CNet for parameter estimation and removal, and LLR-CNet for calibrating residuals.
- NBI-CNet uses a physics-informed convolutional architecture to remove multi-tone interference in a single forward pass without prior knowledge of the interferer count.
- It reduces computational complexity by up to 60% compared to the EOMP-IDS algorithm for configurations with N=2048 and Q=64.
- LLR-CNet acts as a structural whitener, mapping non-Gaussian post-mitigation residuals onto well-calibrated soft metrics to ensure reliable log-likelihood ratios.
- Simulations show the framework eliminates error floors inherent to traditional baselines across dense grids.
- Under severe interference (SIR=-10 dB), the pipeline operates within a 0.2 to 0.5 dB SNR margin of the optimal iterative baseline at a block error rate of 10^-4.
- Under mild interference (SIR=10 dB) with heavy spectral overlap, NBI-CNet avoids signal-peak confusion to deliver a coding gain exceeding 3 dB.
The architecture circumvents error floors triggered by interferer-estimation errors and enables robust generalization across arbitrary FFT sizes without retraining due to its scale-invariant design.