# TranSIC-Net: An End-to-End Transformer Network for OFDM Symbol Demodulation with Validation on DroneID Signals

## TL;DR

TranSIC-Net is a Transformer-based neural network that improves OFDM signal demodulation by combining channel estimation and symbol detection, outperforming traditional and deep learning methods.

## Contribution

Introduces TranSIC-Net, a unified Transformer architecture for OFDM demodulation that implicitly learns channel dynamics and inter-subcarrier correlations.

## Key findings

- TranSIC-Net outperforms LS+ZF and ProEsNet in bit error rate and estimation accuracy.
- The model shows strong generalizability across various OFDM systems with minimal adaptation.
- It performs robustly in real-world UAV signals and synthetic OFDM waveforms.

## Abstract

Demodulating Orthogonal Frequency Division Multiplexing (OFDM) signals in complex wireless environments remains a fundamental challenge, especially when traditional receiver designs rely on explicit channel estimation under adverse conditions such as low signal-to-noise ratio (SNR) or carrier frequency offset (CFO). Motivated by practical challenges in decoding DroneID—a proprietary OFDM-like signaling format used by DJI drones with a nonstandard frame structure—we present TranSIC-Net, a Transformer-based end-to-end neural network that unifies channel estimation and symbol detection within a single architecture. Unlike conventional methods that treat these steps separately, TranSIC-Net implicitly learns channel dynamics from pilot patterns and exploits the attention mechanism to capture inter-subcarrier correlations. While initially developed to tackle the unique structure of DroneID, the model demonstrates strong generalizability: with minimal adaptation, it can be applied to a wide range of OFDM systems. Extensive evaluations on both synthetic OFDM waveforms and real-world unmanned aerial vehicle (UAV) signals show that TranSIC-Net consistently outperforms least-squares plus zero-forcing (LS+ZF) and leading deep learning baselines such as ProEsNet in terms of bit error rate (BER), estimation accuracy, and robustness—highlighting its effectiveness and flexibility in practical wireless communication scenarios.