Nyquist sampling explains how continuous-time signals can be measured at discrete time intervals without losing information. This cheat sheet helps engineering students connect time-domain sampling, frequency-domain replicas, and the practical limits of data acquisition systems. It is useful for signal processing, communications, controls, instrumentation, and audio or image systems.
Students need it because aliasing errors can look like valid data unless sampling and filtering are designed correctly.
The core rule is that a bandlimited signal with highest frequency fmax must be sampled at fs > 2fmax to avoid overlap of spectral copies. The Nyquist frequency is fs/2, which is the highest frequency that can be represented without aliasing after sampling. Frequencies above fs/2 fold back into the baseband and appear as lower false frequencies.
Practical systems use anti-alias filters before sampling and reconstruction filters after digital-to-analog conversion.
Key Facts
- The sampling period is Ts = 1/fs, where fs is the sampling frequency in hertz.
- The Nyquist rate for a bandlimited signal is fs,min = 2fmax, where fmax is the highest frequency present in the signal.
- The Nyquist frequency is fN = fs/2, which is the highest representable frequency after sampling.
- To avoid aliasing in practice, choose fs greater than 2fmax and allow transition bandwidth for the anti-alias filter.
- A sampled signal spectrum repeats at integer multiples of fs, so spectral copies are centered at 0, ±fs, ±2fs, and so on.
- A sinusoid at frequency f aliases to falias = |f - kfs| for the integer k that makes falias fall between 0 and fs/2.
- For ideal reconstruction of a bandlimited signal, use a low-pass reconstruction filter with cutoff between fmax and fs - fmax.
- Undersampling can be intentional for bandpass signals, but only when the sampling rate and signal bandwidth prevent overlapping aliases.
Vocabulary
- Sampling frequency
- The number of samples taken per second, written as fs and measured in hertz.
- Sampling period
- The time between consecutive samples, written as Ts and equal to 1/fs.
- Nyquist rate
- The minimum ideal sampling frequency needed to capture a bandlimited signal without aliasing, equal to 2fmax.
- Nyquist frequency
- Half the sampling frequency, fs/2, which is the highest frequency that can be uniquely represented in sampled data.
- Aliasing
- The effect where a sampled high-frequency component appears as a lower false frequency because the sampling rate is too low.
- Anti-alias filter
- An analog low-pass filter placed before the sampler to remove frequency components above the allowed bandwidth.
Common Mistakes to Avoid
- Using fs = fmax instead of fs > 2fmax is wrong because one sample per cycle cannot uniquely determine a sinusoid or prevent spectral overlap.
- Treating fs/2 as the required sampling rate is wrong because fs/2 is the Nyquist frequency, not the sampling frequency needed for the signal.
- Ignoring filter transition bands is wrong because real anti-alias filters do not cut off instantly at one exact frequency.
- Assuming aliased data can always be corrected later is wrong because once different frequencies overlap in sampled data, their original components are usually impossible to separate.
- Sampling only the desired signal frequency and ignoring noise is wrong because high-frequency noise can alias into the measurement band.
Practice Questions
- 1 A sensor signal contains frequencies up to 3.5 kHz. What is the ideal minimum sampling frequency, and what practical reason might make you choose a higher value?
- 2 A sinusoid at 7 kHz is sampled at fs = 10 kHz. What alias frequency appears between 0 and fs/2?
- 3 An ADC samples at 48 kHz. What is the Nyquist frequency, and what cutoff region should an anti-alias low-pass filter protect?
- 4 Explain why a clean-looking sampled waveform can still be wrong if the original analog signal contained frequency components above the Nyquist frequency.
Understanding Nyquist Sampling and Aliasing Reference
Sampling is more than recording a signal at regular moments. Each sample captures one amplitude value, so the original waveform must be inferred from a sequence of values. This works because a limited range of frequencies cannot change arbitrarily fast.
A smooth low-frequency wave leaves a clear pattern of samples. As frequency rises, fewer samples describe each cycle. Near the Nyquist frequency, a sine wave can be sampled at only two points per cycle.
Its measured amplitude then depends strongly on the timing of the sample clock. A tiny timing shift can make the same wave appear much smaller or change its apparent phase.
Aliasing is especially dangerous because it does not usually look like random noise. A high-frequency vibration, tone, or electrical interference can become a convincing low-frequency pattern in recorded data. For example, a rotating machine may seem to vibrate at a slow fault frequency when its actual vibration is above the measurement limit.
In video, rotating wheels can appear to slow down or reverse. In audio, unwanted ultrasonic content can become an audible false tone.
Once an alias has entered the sampled data, software cannot identify it reliably from the samples alone. The lost distinction existed before conversion, so it must be handled before conversion.
Real anti-alias filters cannot switch from pass to stop instantly. Every physical filter has a transition region. Engineers therefore first decide the highest frequency worth measuring, then leave room above it for the filter to reduce unwanted content before the Nyquist frequency is reached.
A sharper filter needs more circuit complexity, more delay, or a higher sampling rate. Oversampling makes this job easier because it moves the Nyquist frequency farther away from the useful signal band.
The digital system can then apply precise filters and reduce the sample rate later. This approach is common in audio converters, sensor interfaces, and communication receivers.
Students should separate aliasing from quantization and clock jitter. Quantization rounds each sample amplitude to one of a limited number of levels, creating amplitude error and noise. Jitter means sample times are not perfectly regular, creating timing error that becomes worse for faster signals.
Aliasing comes from frequency content that cannot be uniquely represented at the chosen rate. Sketching frequency copies is a useful habit when solving problems. Mark the desired band, place repeated copies at each sampling frequency interval, then check whether any copies overlap.
For bandpass sampling, the same sketch shows why a narrow high-frequency band can sometimes be sampled below twice its highest frequency. The allowed rates are limited, and the band must land in a separate frequency region after folding. Good designs state the signal bandwidth, filter rolloff, clock quality, and the frequencies that must be rejected.