Effects of finite Kernel word length in signal processing

When fast Fourier transform (FFT) is implemented on a real-time equipment with fixed point arithmetic, errors are inevitable. Computational error is investigated as a function of kernel word lengths under the condition of 12-bit A/D conversion. Computer simulation in terms of dynamic range and FFT/IFFT leads to optimum kernel word length. Real signals such as two sine waves, impulse and noise are applied to system, and performance tests are made for various kernel word length. Test result: dynamic range, transfer function and coherence function have a good performance with optimum kernel word length.