Energy-Efficient Sparse FFT and Compressed Transpose Memory for mmWave FMCW Radar Sensor System

Millimeter-wave (mmWave) frequency-modulated continuous wave (FMCW) radar sensors play a big role in autonomous vehicle target detection. However, the sensor data processing system consumes significant power and area. Fast Fourier transform (FFT) is the main computation used in these systems. This article presents an energy-efficient sparse FFT (sFFT) accelerator using zero-skipping and compressed transpose memory for mmWave FMCW radar sensor target detection. Using dynamic thresholding and zero skipping, the computation is much reduced without losing accuracy significantly. Using compressed transposed memory saves the sparse compressed data after the 1D FFT and enhances the energy saving of the system. Analysis shows that the proposed sFFT with compressed memory results in lower latency and obtained peak signal-to-noise ratio (SNR) as compared to existing sFFTs. The proposed system achieved 33.6% total area saving with half-sparsity and 45.8% with quarter-sparsity as compared to dense transpose memory. Compared with other sparse memory, our proposed work makes it easier to build a high-consistency mmWave radar sensor system. The overall 2-D sFFT system using the transpose memory is two times (
$2\times $
) energy efficient and achieves up to 2.3 times (
$2.3\times $
) less latency compared with existing works.