A Survey of Methods and Architectures for Deep Learning-based Area Efficient 1024-Point Pipelined Radix-4 FFT Processor for Biomedical Application

Abstract

The Fast Fourier Transform (FFT) is a fundamental algorithm widely used in biomedical signal processing, including electroencephalogram (EEG), electrocardiogram (ECG), and medical imaging systems. The demand for real-time, low-power, and high-throughput processing has led to the development of area-efficient FFT architectures, particularly the 1024-point pipelined Radix-4 FFT processor. This survey presents a comprehensive analysis of recent methods and architectures integrating deep learning with FFT processors to enhance biomedical applications. FFT reduces computational complexity from to , making it highly efficient for large-scale signal processing. Recent studies have focused on optimizing hardware architectures using pipelining, parallel processing, and memory-efficient techniques such as single-path delay feedback (SDF) and mixed-radix approaches. Additionally, deep learning models such as convolutional neural networks (CNNs) leverage FFT-based feature extraction to improve classification accuracy in biomedical systems. Furthermore, emerging technologies such as processing-in-memory (PIM) and FFT-based neural network acceleration have demonstrated significant improvements in energy efficiency and computational speed. This survey highlights key advancements, comparative analysis, and research challenges, providing insights into future directions for efficient biomedical FFT processor design.