Aliasing and Anti-Alias Filtering
Frequencies above half the sampling rate masquerade as lower ones; an analog filter before the sampler is the only way to prevent this corruption.
What Aliasing Is
Aliasing is the false representation of a high-frequency signal as a lower frequency after sampling. When a signal contains content above the Nyquist frequency, half the sampling rate, that content folds down and appears as a spurious low-frequency component indistinguishable from a genuine one. The classic visual analogy is a wagon wheel that appears to spin backward in film because its rotation is undersampled by the frame rate.
Why It Is Irreversible
Once a signal is aliased, the samples contain the false low frequency added to any true low-frequency content, with no way to separate them. Digital processing after the sampler cannot undo it, because the information distinguishing the true signal from its alias was destroyed at the instant of sampling. Prevention is the only remedy.
The Anti-Alias Filter
An anti-alias filter is an analog low-pass filter placed before the sampler. It attenuates frequencies above the Nyquist frequency so that little energy remains to fold down. Because real filters roll off gradually rather than cutting sharply, the sampling rate is usually set above the strict minimum, leaving a transition band in which the filter can attenuate before the Nyquist frequency is reached.
- The filter must be analog and before the ADC, not digital and after
- Sharper cutoff needs higher filter order, adding phase distortion and cost
- Oversampling relaxes the filter's steepness requirement
- Any noise or interference above Nyquist must be removed, not just the signal band
Interference Aliasing
Aliasing threatens not only wanted high-frequency content but also unwanted interference. High-frequency noise, switching-supply ripple, and radio pickup all fold into the measurement band if unfiltered, appearing as noise a controller cannot distinguish from real signal. The anti-alias filter therefore protects data integrity against sources you may not have anticipated, which is why it is a standard, non-negotiable part of any data-acquisition front end.