Hi Avinash

Due to the Nyquist-Shannon Theorem, the sampling rate needs to be at least twice the highest desired frequency component (or bandwidth) of your signal. Not to get off track, but you will need to LPF your input signal so that it does not contain any frequency components higher than twice your sampling frequency. So now, back to the sampling/bandwidth rates, for example, if you wish to have a bandwidth of 1khz, you will need a sampling rate of at least 2khz (there's actually a rule of thumb to actually sample a little higher than Nyquist, but that's for another question). Here's Shannon's paper discussing the Sampling Theorem:

http://www.stanford.edu/class/ee104/shannonpaper.pdf

You'll also find in Shannon's paper that there is a quantization factor associated with a digital sampling system that has to with the number of bits/sample that tells you the maximum signal/noise ratio you can expect. This to I think should make an interesting discussion probably in another question.

Hopefully someone well versed in signal processing will expand upon this.

HTH

Mike

To recover the original signal from the sampled data, a "demodulator" can apply the procedure of modulation in reverse. After each sampling period, the demodulator reads the next value and shifts the output signal to the new value. As a result of these transitions, the signal has a significant amount of high-frequency energy caused by aliasing. To remove these undesirable frequencies and leave the original signal, the demodulator passes the signal through analog filters that suppress energy outside the expected frequency range (greater than the Nyquist frequency ).[note 1] The sampling theorem shows PCM devices can operate without introducing distortions within their designed frequency bands if they provide a sampling frequency twice that of the input signal. For example, in telephony, the usable voice frequency band ranges from approximately 300 Hz to 3400 Hz. Therefore, per the Nyquist–Shannon sampling theorem, the sampling frequency (8kHz) must be at least twice the voice frequency (4kHz) for effective reconstruction of the voice signal.

Here is a good sources