An anti-aliasing filter is an analog lowpass filter applied to a continuous-time signal before sampling, to enforce bandlimiting and prevent aliasing.
The cutoff frequency is below , where is the planned sampling rate. Any signal content above the cutoff is suppressed by the filter, so when sampling occurs, the replicas in the sampled spectrum don’t overlap.
Why it’s necessary
Real-world signals are never perfectly bandlimited — every real signal has high-frequency components from noise, harmonics, or sharp edges. Without anti-aliasing, those high frequencies would alias into the lower band during sampling and contaminate the part of the signal we care about.
The anti-aliasing filter does lose information — the parts of the signal above the cutoff are gone. But that information would have been corrupted by aliasing anyway. The filter doesn’t make things worse; it makes them honest. A signal that has been anti-alias-filtered then sampled gives faithful samples of the filtered version of the original signal — which is the best we can do.
The complete signal chain
A digital signal-processing chain (digital audio, image processing, etc.) typically looks like:
analog signal → anti-aliasing filter → sampler → digital processing → reconstruction filter → analog output
The anti-aliasing filter on the input side prevents aliasing during sampling. The reconstruction filter on the output side smooths the sampled output back into a continuous-time signal (approximating sinc interpolation). Both filters are essential — without them, the digital chain would mangle the signal.
Practical constraints
Anti-aliasing filters have to be analog (they operate on the continuous-time signal before sampling) and causal (they’re physical real-time circuits). This rules out ideal brick-wall lowpass filters, which are non-causal. Real anti-aliasing filters have a gradual rolloff, not a perfect cutoff.
The design tension:
- Sharper rolloff = better protection against aliases, but more complex filter (higher order, more components, more cost, more phase distortion).
- Gentler rolloff = simpler filter, but you have to oversample further to push the sampling rate higher and give the rolloff room to work.
A common compromise: oversample by 2× or 4× beyond the Nyquist rate, so the cutoff can be gentler without aliasing issues. Modern audio uses oversampling with low-order analog anti-aliasing filters, followed by digital decimation filters that have steeper cutoffs.
In audio
A 44.1 kHz CD-quality system has Nyquist frequency 22.05 kHz. The anti-aliasing filter has its passband up to ~20 kHz and stopband above ~22 kHz. The 2 kHz transition band is steep but achievable with reasonable filter complexity.
Higher sampling rates (48, 96, 192 kHz) move the Nyquist frequency further away from the audible band, allowing gentler and more transparent anti-aliasing filters. This is one of several arguments for high-rate audio formats — not because human hearing extends to 96 kHz, but because the filters don’t have to fight as hard.