What happens if I sample below the Nyquist rate?

If a signal contains frequency components above half the sampling rate (the Nyquist frequency), those components are aliased: they appear as phantom signals at incorrect frequencies in the digital domain. A 30 MHz signal sampled at 40 MSPS (Nyquist = 20 MHz) appears as a false 10 MHz signal (30 minus 20 = 10 MHz alias). This alias is indistinguishable from a real 10 MHz signal and cannot be removed by digital filtering after sampling. The anti-aliasing filter must be placed before the ADC to attenuate all signals above the Nyquist frequency. The filter must transition from passband to stopband within the gap between the signal bandwidth and the Nyquist frequency, requiring steeper roll-off when the sampling rate is only slightly above twice the signal bandwidth.

What is undersampling and when is it useful?

Undersampling (also called bandpass sampling or harmonic sampling) intentionally samples a bandpass signal at a rate much lower than twice its carrier frequency, but at least twice its bandwidth. A 2.4 GHz WiFi signal with 20 MHz bandwidth does not need a 4.8 GSPS ADC. If sampled at 80 MSPS, the signal aliases down to a frequency between DC and 40 MHz, effectively performing frequency conversion (mixing) in the sampling process. The ADC acts simultaneously as a digitizer and a downconverter. The requirements are: the ADC must have sufficient analog input bandwidth to handle the 2.4 GHz signal (track-and-hold bandwidth), and a bandpass anti-aliasing filter must ensure that only the desired Nyquist zone contains signal energy.

How does oversampling improve dynamic range?

Oversampling spreads the ADC's quantization noise across a wider frequency band. If the signal occupies only a fraction of the Nyquist bandwidth, the noise in the signal band is reduced. The SNR improvement from oversampling is 10 log10(fs / 2B) dB, where fs is the sampling rate and B is the signal bandwidth. A 14-bit ADC with 86 dB SNR sampling at 1 GSPS with a signal bandwidth of 10 MHz: oversampling gain = 10 log10(1000/20) = 17 dB, giving an effective SNR of 103 dB. This is equivalent to approximately 2.8 additional bits of resolution. Sigma-delta ADCs exploit this principle aggressively, oversampling by 64 to 256 times and using noise shaping to push quantization noise out of the signal band.

Signal Processing

Sampling Rate

A 5G NR base station digitizes a 100 MHz channel centered at 3.5 GHz. The traditional approach: mix down to IF, then sample at 250 MSPS (2.5× the bandwidth). The modern approach: use a 10 GSPS direct RF sampling ADC that digitizes the 3.5 GHz carrier directly, eliminating the mixer, LO, and IF filter entirely. The ADC's sampling rate determines the instantaneous bandwidth available to the digital signal processor: at 10 GSPS, the Nyquist bandwidth is 5 GHz, enough to capture the entire sub-6 GHz band simultaneously. Sampling rate is not just a spec; it is the gateway between the analog world and the digital processing engine that defines a modern receiver's capability.

Category: Signal Processing

Nyquist: f_s ≥ 2×f_max

Trend: Direct RF sampling replacing superhets

ADC Sampling Rates by Application

Application	Sampling Rate	Resolution	Nyquist BW	Strategy
Audio	44.1 / 48 kSPS	16 to 24 bits	22 kHz	Oversampled ΣΔ
HF SDR receiver	60 to 125 MSPS	14 to 16 bits	30 to 62 MHz	Direct sampling
Cellular IF	200 to 500 MSPS	12 to 14 bits	100 to 250 MHz	IF sampling
5G NR direct RF	4 to 10 GSPS	12 to 14 bits	2 to 5 GHz	Direct RF sampling
Oscilloscope	20 to 256 GSPS	8 to 10 bits	10 to 128 GHz	Interleaved ADCs
Radar digital beamforming	1 to 4 GSPS	12 to 14 bits	0.5 to 2 GHz	Undersampling or direct

Nyquist criterion:
f_s ≥ 2×f_max (for baseband signals)
f_s ≥ 2×BW (for bandpass undersampling)

Oversampling SNR gain:
ΔSNR = 10·log₁₀(f_s / 2B) dB
1 GSPS, 10 MHz BW: ΔSNR = 10·log(1000/20) = 17 dB (~2.8 extra bits)

Alias frequency:
f_alias = |f_signal − n×f_s| (choose n for result in 1st Nyquist zone)

Common Questions

Frequently Asked Questions

What if I sample too slowly?

Frequencies above f_s/2 alias to false positions. 30 MHz at 40 MSPS appears at 10 MHz. Indistinguishable from a real signal. Anti-aliasing filter before ADC must attenuate everything above Nyquist. Steeper filter needed when f_s is barely above 2×BW.

What is undersampling?

Sample a bandpass signal at ≥2×BW (not 2×carrier). A 2.4 GHz WiFi signal (20 MHz BW) at 80 MSPS: aliases to baseband. ADC acts as digitizer + downconverter. Requires wideband track-and-hold and bandpass anti-alias filter.

How does oversampling help?

Spreads quantization noise across wider band. Signal occupies only a fraction: SNR gain = 10·log(f_s/2B). 14-bit ADC at 1 GSPS, 10 MHz BW: +17 dB SNR = effective 16.8 bits. Sigma-delta ADCs oversample 64 to 256× with noise shaping.

Related Terms

← S-Parameters Skin Effect →

System Design

Sampling Rate Planner

Enter signal bandwidth, center frequency, and ADC resolution. Get the minimum sampling rate for Nyquist, oversampling, or undersampling strategies with aliasing analysis.

Plan Sampling