Real Time Signal Processing with FPGA on AWS F1 in Wireless Communication

Abstract

Real-time audio signal processing requires high computational performance, low processing latency, and accurate frequency response, particularly in applications such as voice-enabled systems, hearing aids, and embedded audio platforms. This paper presents a cloud-based hardware acceleration framework for real-time audio denoising using a Finite Impulse Response (FIR) filter implemented on Amazon Web Services (AWS) EC2 F1 FPGA instances. The proposed system effectively suppresses unwanted noise while preserving speech quality within the critical 0–4 kHz voice frequency range.

The FIR filter is developed using fixed-point C++ to achieve a balance between computational accuracy and efficient hardware resource utilization. Filter coefficients are designed using standard digital signal processing techniques to ensure stable operation and a linear phase response. The design is synthesized into FPGA hardware using Xilinx Vivado High-Level Synthesis (HLS), enabling a deeply pipelined architecture for continuous audio sample processing. Compared with conventional CPU-based implementations, the FPGA-based solution significantly reduces processing latency while improving throughput.

By utilizing cloud-hosted FPGA resources, the proposed framework removes the need for dedicated hardware, providing scalability, remote accessibility, and cost-effective deployment through a pay-as-you-use model. Experimental results demonstrate that cloud-based FPGA acceleration offers an efficient and practical solution for meeting the demanding performance requirements of real-time digital audio processing in both research and industrial applications.