An FFT approximation technique suitable for on-chip generation and analysis of sinusoidal signals

TLDR: This paper presents an FFT approximation technique suitable for on-chip spectral BIST signal generation and analysis and shows that the noise produced by the approximation technique is under 24.74 dB for a 256 point FFT with a 32 point approximate kernel.

Abstract: Signal generation and analysis are an important part of BIST for analog and mixed-signal systems. An accurate analysis of the spectral content of signals produced by analog components can be accomplished with a digital implementation of a fast Fourier transform (FFT) algorithm. In the past, size and speed have limited the application of such a technique to off-chip test equipment or DSP chips (primarily due to the number of multiplication operations). In this paper, we present an FFT approximation technique suitable for on-chip spectral BIST signal generation and analysis. For signal generation, we show that the noise produced by the approximation technique is under 24.74 dB for a 256 point FFT with a 32 point approximate kernel. For signal analysis, we show that the instantaneous dynamic range (IDR) for the approximation technique is under 21.80 dB for a 256 point FFT with a 32 point approximate kernel. Our techniques have been implemented and demonstrated on a Xilinx Virtex-II FPGA using an off-chip ADC and DAC, and we are currently implementing the technique on an ASIC using a 0.13 /spl mu/m SiGe process for 2-16 GHz applications.