Reducing Power and Cycle Requirement for FFT of ECG Signals through Low Level Arithmetic Optimizations for Cardiac...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Computer Science

Background:

• The Fast Fourier Transform (FFT) is crucial for spectrum analysis in biomedical applications.
• Floating-point operations in FFT are power-intensive and cycle-costly for embedded systems like pacemakers.
• Electrocardiogram (ECG) data possesses repetitive characteristics that can be exploited for computational efficiency.

Purpose of the Study:

• To reduce arithmetic operations for FFT computation in ECG analysis on power-constrained embedded systems.
• To improve the performance and energy efficiency of FFT algorithms for biomedical signal processing.
• To leverage the repetitive nature of ECG signals for a more efficient FFT implementation.

Main Methods:

• Developed a novel algorithm utilizing lookup tables to exploit ECG signal repetitiveness.
• Implemented a 128-point FFT routine optimized for a 32-bit embedded platform.
• Tested the algorithm using actual ECG data from PhysioNet.

Main Results:

• Achieved a 9.22% increase in computational speed for the FFT routine.
• Demonstrated a 10.1% improvement in battery life on the embedded platform.
• Validated the effectiveness of the lookup table approach for ECG data.

Conclusions:

• The proposed lookup table-based FFT algorithm significantly enhances performance and energy efficiency for ECG analysis.
• This method offers a practical solution for implementing efficient signal processing in power-constrained biomedical devices.
• Exploiting signal-specific characteristics like ECG repetitiveness is key to optimizing embedded FFT computations.