Aliasing in sampled systems is dangerous precisely because it produces results that look entirely reasonable. Any frequency above half the sampling rate does not vanish and does not appear as obvious corruption; it reappears somewhere below half the sampling rate as a perfectly plausible signal. Once the samples exist, nothing downstream can tell the difference or undo it.
Why a high frequency comes back as a low one
Sampling captures the signal’s value at discrete instants and discards everything between them. If the signal completes more than half a cycle between samples, the sequence of captured values is identical to what a much slower signal would have produced. The classic visual analogy is a spoked wheel filmed at a fixed frame rate appearing to rotate slowly backward. The mathematics is the same: every frequency above half the sampling rate maps onto a specific frequency below it, and the mapping is not reversible because multiple input frequencies map to the same output. The threshold at half the sampling rate is the limit that defines the usable band, and exceeding it does not degrade gracefully.
The filter has to be before the converter
This is the point that gets missed most often. Digital filtering applied after sampling operates only on the sampled data, in which the alias is already indistinguishable from a real signal at that frequency. Removing it digitally would require removing the genuine content at the same frequency too. The only place to act is in the analog domain ahead of the sampler, where the offending frequencies still exist separately and can be attenuated before they are folded down. That is what an anti-aliasing filter does, and it is a required part of any sampled measurement rather than an optional refinement. A converter with a built-in digital filter still needs attention to what reaches its actual sampling instant.
Designing the filter with the right margin
A practical filter does not cut off abruptly, so the design has to account for its transition band. The passband must cover the signal of interest with acceptable flatness, and the stopband must provide enough attenuation by the time the frequency reaches half the sampling rate. How much attenuation is enough depends on the resolution you are trying to achieve, since an alias only needs to be below the noise floor to be harmless. Sampling faster than the signal strictly requires is a powerful technique here, because it widens the gap between the signal band and the folding frequency and allows a gentler and less expensive filter. Filter type matters too: sharper responses come with phase distortion and ringing that may be unacceptable for transient measurements.
Where the offending frequencies come from
Designers often reason that the signal has no high-frequency content, which may be true of the signal but is never true of everything present at the converter input. Switching converters inject content at their switching frequency and its harmonics. Digital circuitry couples clock energy into analog nodes. Radio transmitters nearby can be rectified by nonlinearities in the front end and appear as a slowly varying offset. Mains-frequency interference and its harmonics appear on long sensor cables. Any of these can be above the folding frequency and will alias down into the measurement band, appearing as a drifting offset or a slow oscillation that looks like a real physical effect. Chasing such an artifact as though it were genuine can consume a great deal of time.
Recognizing an alias when you see one
The diagnostic that settles it is changing the sampling rate. A genuine signal component stays at the same frequency; an alias moves, because its apparent frequency depends on the relationship between the input frequency and the sampling rate. Changing the rate slightly and observing whether a suspicious component shifts is quick and decisive. Other clues include a component that appears only when a particular subsystem is active, or one whose frequency drifts with the temperature of a nearby oscillator. Once identified, the fix is upstream: improve the analog filter, increase the sampling rate, or reduce the interference at its source through better layout and shielding.
What to check before you act
- Place the anti-aliasing filter in the analog path, before the sampler.
- Set stopband attenuation from your target resolution, not a habit.
- Sample faster than strictly required to ease the filter requirement.
- Change the sampling rate to test whether a suspect component is an alias.
Common questions
Can I remove aliasing with a digital filter?
No. After sampling, the alias occupies the same frequency as legitimate content and cannot be separated from it. The filtering has to happen before the converter samples.
Is sampling at exactly twice the highest frequency sufficient?
In theory it is the limit, but no practical filter is ideal and no real signal is perfectly band-limited. Sampling comfortably faster is standard practice.
The practical takeaway
Put a real analog filter ahead of every sampler and choose its stopband attenuation against the resolution you need. Sample faster than the minimum to make that filter achievable, look beyond the intended signal for sources of high-frequency content, and confirm any suspicious component by changing the sampling rate.
References and further reading
- FCC: electromagnetic compatibility and radio frequency safety questions answered
- FCC: radio frequency safety background and exposure guidance
- Department of Energy: how electronics losses turn into wasted energy and heat
Numerical scenarios are illustrative unless identified otherwise. Follow the exact product instructions; component ratings and local installation requirements can differ.
