FPGA Implementation of Complex-Valued Neural Network for Polar-Represented ECG Classification

Electrocardiogram (ECG) signal classification is essential for detecting cardiac abnormalities at an early stage. Traditional real-valued neural networks often struggle to represent phase and amplitude variations effectively. This project presents an FPGA-based implementation of a Complex-Valued Neural Network (CVNN) for ECG heartbeat classification using polar-represented spectrograms. ECG signals are transformed into time-frequency representations using Short-Time Fourier Transform (STFT) and then converted into polar coordinates to capture both magnitude and phase information. These complex-valued features are fed into a CVNN, which can inherently process and learn from complex inputs more effectively than conventional networks. The entire architecture is implemented on an FPGA using High-Level Synthesis (HLS), providing a low-latency, energy-efficient solution suitable for real-time embedded applications. Resource usage, execution time, and accuracy are optimized to meet hardware constraints without sacrificing performance. Experiments conducted using the Kaggle ECG Heartbeat Classification Dataset show high classification accuracy, validating the model’s effectiveness. This project demonstrates the feasibility of deploying CVNNs on FPGA for accurate, real-time cardiac monitoring systems.