mikec@ux1.lbl.gov (Mike Chin) (01/17/90)
I'm playing with a 16-bit ADC. Both the manafacturer and some HP Appnotes recommend inputting a spectrally pure sine wave, and taking the FFT. ADC nonlinearities show up as spikes at harmonics of the fundamental. No problem. However, they also claim that a signal/noise figure can be derived by taking the ratio of the fundamental(S) to the sum of the rest of the spectra (N) and computing 20*log10(S/N). They say that an M-bit ADC should have a S/N of (6*M + 1.8)dB When I cook up a 1024pt 16-bit ADC data simulation, I window the data with a Blackman-Harris window, take the FFT, and find the fundamental where I expect, spread over maybe 5 bins. If the FFT mags are in an array C[1...1024] and the fundamental is in C[30], I do: E1=sum(C1:C512), E2=sum(C28:33), S/N=20*log10((E1-E2)/E2). Problem: I get ~75dB instead of 96dB for this ideal 16-bit simulation. Mike Chin Lawrence Berkeley Labs MJChin@lbl.gov