In signal processing, sampling is the act of taking regular and periodic values of a continuous-time signal to create a discrete-time signal. The opposite conversion, from a DT signal to a CT signal is interpolation. Where is defined as the sampling period:
The sampling function is defined as:
i.e., we have an infinite CT impulse train spaced out by seconds. Then, when we sample a function , we multiply to get a periodic weighted sum of impulses:
Some common sampling rates: audio (at 44 100 samples per second) or video (at 30 samples per second). Our key questions with sampling involve the following, both of which are answered with the Nyquist-Shannon theorem for a bandlimited signal:
- How small should be?
- How can we reconstruct the original CT signal from the DT samples ?
Fourier series
Since the sampling function is periodic, we can take the CTFS. The CTFS coefficients are given by (where is the fundamental angular frequency of ):
Since we have a bounded interval, and because of the nature of the delta function, the integrand is only zero if and equal to 1 for . So the coefficient reduces to:
Since the coefficients represent the strength of the harmonic frequencies present in the signal, we observe that the CTFT spectrum has equal contributions from all harmonics.
i.e., the CTFT of the sampling function is itself a periodic sampling function with period .1
Footnotes
-
Prof Pavel called this the “picket fence miracle”. ↩