An End-to-End Neural Network Transceiver Design for OFDM System with FPGA-Accelerated Implementation

TL;DR

This work tackles OFDM transceiver design for 6G-era demands by replacing the conventional DFT/IDFT and demodulation blocks with end-to-end trained neural networks (DFT-Net and Demod-Net). It introduces the DDNA FPGA accelerator to map these networks efficiently, using block matrix operations and pipelined execution to achieve low latency while preserving compatibility with legacy interfaces. Empirical results show about 1.5 dB BER gain and up to 66% reduction in execution time across multiple modulation schemes, with only modest hardware overhead compared to traditional FFT/DFT implementations. The approach demonstrates a practical path for DL-enabled transceivers with hardware-aware optimizations suitable for real-time 6G-like systems.

Abstract

The evolution toward sixth-generation (6G) wireless networks demands high-performance transceiver architectures capable of handling complex and dynamic environments. Conventional orthogonal frequency-division multiplexing (OFDM) receivers rely on cascaded discrete Fourier transform (DFT) and demodulation blocks, which are prone to inter-stage error propagation and suboptimal global performance. In this work, we propose two neural network (NN) models DFT-Net and Demodulation-Net (Demod-Net) to jointly replace the IDFT/DFT and demodulation modules in an OFDM transceiver. The models are trained end-to-end (E2E) to minimize bit error rate (BER) while preserving operator equivalence for hybrid deployment. A customized DFT-Demodulation Net Accelerator (DDNA) is further developed to efficiently map the proposed networks onto field-programmable gate array (FPGA) platforms. Leveraging fine-grained pipelining and block matrix operations, DDNA achieves high throughput and flexibility under stringent latency constraints. Experimental results show that the DL-based transceiver consistently outperforms the conventional OFDM system across multiple modulation schemes. With only a modest increase in hardware resource usage, it achieves approximately 1.5 dB BER gain and up to 66\% lower execution time.

Figures (11)

• Figure 1: Block diagram of a conventional OFDM system.
• Figure 2: Network architecture. (a) DFT-Net learns the bidirectional mapping between the time-domain symbol $\boldsymbol{y}$ and the frequency-domain symbol $\boldsymbol{Y}$. (b) Demod-Net provides two structural variants and performs soft demodulation from $\hat{\boldsymbol{X}}$ to soft bits $\tilde{\boldsymbol{b}}$ within one OFDM frame. Each bit is represented by two real-valued scores. The notation [$\cdot$] denotes the input dimensionality, and $m$ indicates the modulation order, e.g., $m=2$ for QPSK and $m=4$ for 16-QAM.
• Figure 3: BER performance comparison between the conventional receiver and the proposed DL-Receiver in an AWGN channel.
• Figure 4: BER performance comparison between the conventional transceiver and the proposed DL-Transceiver in an AWGN channel.
• Figure 5: Frequency distribution of activations and weights. The left figure represents the activations, the other represents the weights.