AdaFortiTran: An Adaptive Transformer Model for Robust OFDM Channel Estimation

TL;DR

This work tackles reliable OFDM channel estimation in fast-fading and low-SNR regimes by introducing AdaFortiTran, a compact architecture that fuses CNN-based locality with a patch-based transformer encoder, augmented by a Channel Adaptivity Module that conditions attention on channel statistics. The method processes LS pilot estimates through a learned upsampler, shallow feature extraction, and a transformer with patch-level embeddings before a final convolutional reconstructor to produce the full channel estimate. Empirical results on 3GPP TDLA channels show AdaFortiTran achieving up to $6\ \text{dB}$ MSE improvement over state-of-the-art DL-CE methods across a wide range of Doppler, SNR, and delay spreads, and demonstrated robustness to pilot density. The approach offers a practical, adaptable solution for real-world wireless systems by integrating domain knowledge with modern Transformer architectures in a compact design.

Abstract

Deep learning models for channel estimation in Orthogonal Frequency Division Multiplexing (OFDM) systems often suffer from performance degradation under fast-fading channels and low-SNR scenarios. To address these limitations, we introduce the Adaptive Fortified Transformer (AdaFortiTran), a novel model specifically designed to enhance channel estimation in challenging environments. Our approach employs convolutional layers that exploit locality bias to capture strong correlations between neighboring channel elements, combined with a transformer encoder that applies the global Attention mechanism to channel patches. This approach effectively models both long-range dependencies and spectro-temporal interactions within single OFDM frames. We further augment the model's adaptability by integrating nonlinear representations of available channel statistics SNR, delay spread, and Doppler shift as priors. A residual connection is employed to merge global features from the transformer with local features from early convolutional processing, followed by final convolutional layers to refine the hierarchical channel representation. Despite its compact architecture, AdaFortiTran achieves up to 6 dB reduction in mean squared error (MSE) compared to state-of-the-art models. Tested across a wide range of Doppler shifts (200-1000 Hz), SNRs (0 to 25 dB), and delay spreads (50-300 ns), it demonstrates superior robustness in high-mobility environments.

Figures (4)

• Figure 1: Architecture and Submodules of AdaFortiTran
• Figure 2: Effect of Channel Adaptivity Module and Number of Transformer Layers $L$
• Figure 3: Performance Analysis Across Diverse Channel Conditions
• Figure 4: MSE vs Pilot Placement