Recent Advances in Deep Learning-based Area Efficient 1024-Point Pipelined Radix-4 FFT Processor for Biomedical Application: A Systematic Review

Abstract

Fast Fourier Transform (FFT) plays a crucial role in biomedical signal processing, including electroencephalography (EEG), electrocardiography (ECG), and medical imaging applications. The increasing demand for real-time, low-power, and high-throughput processing has driven the development of optimized FFT architectures, particularly the 1024-point pipelined Radix-4 FFT processor. Recent advancements integrate deep learning techniques with FFT-based architectures to enhance feature extraction, noise reduction, and computational efficiency in biomedical systems. This review presents a comprehensive analysis of recent developments in deep learning-assisted FFT processors, focusing on area efficiency, power consumption, and processing speed. Modern architectures emphasize pipelined and parallel processing to achieve high throughput while minimizing hardware complexity. Additionally, hybrid approaches combining convolutional neural networks (CNNs) with FFT have demonstrated improved performance in biomedical signal analysis, including enhanced accuracy in classification and diagnosis.  Furthermore, emerging low-power FFT designs such as minimal architecture single-delay feedback (mSDF) structures show significant reductions in power consumption, making them suitable for wearable and implantable medical devices.  This systematic review highlights key advancements, identifies research gaps, and discusses future directions in designing efficient FFT processors integrated with deep learning for biomedical applications.