Frequency content of discrete-time signals


During lecture, I attempted to illustrate, through music (a snippet of Chinese opera), that when we sample a signal at some sampling frequency fs, the only frequencies present in the reconstructed signal lie in the range [-fs/2, fs/2]. I took a piece of music from a CD (sampled at 44.1kHz) and subsampled it at 22.05kHz, 11.025kHz, 8kHz, 4kHz and 2kHz. According to sampling theory, the highest frequency components that can be reconstructed for each sound file are half of each sampling frequency. For example, for the music snippet with sampling rate fs = 8kHz, we should not be able to hear any frequencies greater than 4kHz; we saw that this is indeed the case, by playing each subsampled piece of music in an mp3 player, and manipulating different frequency bands in the equalizer of the mp3 player. As an example, an 8kHz sampled signal should sound exactly the same for the two equalizer configurations pictured below, and, indeed, it did.

A perceptible difference in music sampled at lower frequencies is that it sounds muffled, or completely unrecognizable (as is the case for fs = 2kHz). To listen for yourself, click on the music sound files below.

Cantonese Opera (1 min.)

Sampling rate Music files
44.1kHz
(mp3, 2.3 Mb)

(wav, 10.1 Mb)
22.05kHz
(mp3, 2.3 Mb)

(wav, 5 Mb)
11.025kHz
(mp3, 2.3 Mb)

(wav, 2.5 Mb)
8kHz
(mp3, 2.3 Mb)

(wav, 1.8 Mb)
4kHz
(mp3, 2.3 Mb)

(wav, 948 kb)
2kHz
(mp3, 2.3 Mb)

(wav, 476 kb)

Last updated February 4, 2003 by Michael C. Nechyba