Aliasing effect and how to minimize it

Aliasing is the distortion that occurs when a signal is sampled below its Nyquist rate ($f_s < 2W$), or when it contains components above $f_s/2$. The shifted spectral copies overlap, so high-frequency components fold back and appear as false low frequencies $|f - nf_s|$. Example: a 6 kHz tone sampled at 8 kHz appears as a 2 kHz tone after reconstruction, and cannot be removed afterwards.

  /‾‾\   /‾‾\
 /    \ /    \      overlap region = aliasing
/      X      \
0    fs/2     fs

Ways to minimize aliasing

1. Anti-aliasing (pre-alias) filter: a low-pass filter before the sampler removes components above $W$ (and noise) so the signal is strictly band-limited.
2. Sample above the Nyquist rate: use $f_s > 2W$ to create a guard band, so practical filters with finite roll-off can separate copies (e.g. speech limited to 3.4 kHz sampled at 8 kHz).
3. Sharp-cut-off filters / oversampling: high-order filters or oversampling with digital decimation reduce residual aliasing.