On Tue, 23 Nov 2004 13:38:18 -0200, Xtian Xultz <[EMAIL PROTECTED]> wrote: > I Understand what you say. And what the mega samples per second means? I cant > understand why having 20 mega samples per second I can measure onle 250k Hz > of a sine wave...
The Nyquist theorem (20 MS/sec would give you a theoretical upper bandwidth of just under 10 MHz) assumes ideal sampling, and probably a bunch of other caveats that I can't remember now. In reality, you are going to get quantization noise (+/- full scale divided by 2^N, where N is your number of bits), sampling jitter, and coupled noise on the input. Setting your sampling frequency to 10x the highest frequency component of your input is not unreasonable. Also, real-life anti-aliasing filters (the ones that prevent higher frequencies from affecting your sampled data) can't have sharp cutoffs without distorting the input signal (usually manifested as ringing). If the analog anti-aliasing filter is only 10 or 20 dB/decade, that still allows a lot of higher-frequency energy in to your sampler. The oversampling lets you filter out the higher frequencies in the digital domain. I can't remember the app note number, but Microchip had a good paper on their website explaining analog-to-digital converter parameters, and IIRC they discussed some of these very issues. -- - Charles Lepple
