jhs@MITRE-BEDFORD.ARPA (02/01/87)
The use of a 3-pole Chebyshev filter for anti-aliasing, with cutoff at 20 KHz and sampling rate of 50 KHz (Nyquist frequency at 25 KHz) strikes me as "underkill". The Chebyshev filter is only 3 dB down at "cutoff" and is dropping off only 36 dB per ** OCTAVE ** thereafter. Even assuming the maximum sampling rate of 50 KHz, the Nyquist frequency (half the sampling frequency or 25 KHz) is only 1.25 times the 3-dB cutoff frequency, whereas an octave is 2:1, so the filter would be down only about 9 dB at the Nyquist frequency. This means that frequency components just above 25 KHz would be attenuated only about 9 - 10 dB before being "folded" back down below 25 KHz. Maybe this A/D converter would work OK for signal sources not containing much signal power above 25 KHz, but on a really wideband source (e.g. direct recording with a GOOD condenser mic. and enthusiastic playing of cymbals, etc.) it could introduce very significant distortion of highs. Especially when compared to the theoretical performance of a 16-bit PCM system. By comparison, the high-fidelity A/D I am most familiar with, which was built by Stan Kriz of Three Rivers Computer back in the 70s (later called PERQ Systems), used as I recall an 11th order (like 5.5 "poles", or maybe it was really 11 pole) Elliptic Function filter. These filters have a notch just above the cutoff frequency and drop off to their full stopband attenuation almost immediately above cutoff. The Three Rivers converter was one of the first to gain the acceptance and even praise of musicians who had always objected to the "distorted" sound of digital audio systems even when the engineers couldn't measure the distortion. There are other factors to consider, because the use of a sharper filter might introduce artifacts of a different sort, e.g. "ringing" on squarewaves or sharp transients, but users of this product might want to look into the use of a better filter ahead of the unit if they are dealing with very wideband sources with lots of energy above 25 KHz. You may not be able to hear it BEFORE sampling, but you sure will AFTER sampling if it is not kept out of the sampler circuit. -John Sangster jhs@mitre-bedford.arpa