A high-parallelism memory-based FFT processor with high SQNR and novel addressing scheme

This paper presents an area-efficient memory-based FFT processor for long FFT lengths. To achieve high throughput, radix-4
²
FFT algorithm is adopted to reduce number of FFT stages. For low-complexity realization of the main butterfly processing element, a folded-by-2 scheme along with an optimized scheduling is designed. Variable FFT lengths (i.e., 1024 ~ 32768 points) can be supported through flexible switch configurations. Moreover, a conflict-free memory addressing scheme is devised to support 16-way parallel and normal-order data input/output without re-ordering buffers. An optimized block floating-point (BFP) scheme is employed for long-length FFT operations. The EDA synthesis results with TSMC-90nm process show that the area of proposed FFT processor is 2.98 mm
²
, and the power consumption is 29 mW @160MHz clock frequency. The SQNR performance is over 70dB for all supported FFT lengths with 16-bit wordlength.