Nyquist Rate

Introduction

The Nyquist rate is the minimum sampling rate required to perfectly reconstruct a bandlimited continuous-time signal from its discrete samples. Named after Harry Nyquist, who established this fundamental limit in the 1920s, it provides the critical bridge between the analog and digital worlds. Every analog-to-digital converter, every digital recording system, and every digital communication receiver must respect the Nyquist rate to avoid information loss.

The concept is elegantly simple: if a signal contains no frequency components above $f_{max}$ Hz, then sampling at a rate of at least $2f_{max}$ samples per second captures all the information in the signal. Sample any slower, and you permanently lose information through aliasing. This single number — twice the maximum frequency — governs the design of every digital acquisition system.

Formal Definition

For a bandlimited signal $x(t)$ whose highest frequency component is $f_{max}$ (or equivalently, $\omega_{max} = 2\pi f_{max}$):

Nyquist Rate: $f_N = 2f_{max}$ (samples per second)

Nyquist Interval: $T_N = \frac{1}{2f_{max}}$ (maximum allowable time between samples)

Nyquist Frequency: $f_{Nyq} = f_s/2$ (the highest frequency representable at sampling rate $f_s$)

Note the distinction: the Nyquist rate is a property of the signal (depends on its bandwidth), while the Nyquist frequency is a property of the sampling system (depends on the sampling rate).

Mathematical Basis

The Sampling Theorem states that a bandlimited signal $x(t)$ with $X(j\omega) = 0$ for $|\omega| > \omega_{max}$ can be perfectly reconstructed from its samples $x(nT_s)$ if and only if:

$$f_s > 2f_{max} \quad \text{or equivalently} \quad \omega_s > 2\omega_{max}$$

When sampling at rate $f_s$, the spectrum of the sampled signal is:

$$X_s(j\omega) = \frac{1}{T_s}\sum_{k=-\infty}^{\infty} X(j(\omega - k\omega_s))$$

The spectral replicas are centered at multiples of $\omega_s$. If $\omega_s > 2\omega_{max}$, these replicas do not overlap, and the original spectrum can be extracted with an ideal low-pass filter. If $\omega_s \leq 2\omega_{max}$, the replicas overlap (aliasing) and information is irreversibly corrupted.

Nyquist Rate for Common Signals

Single Sinusoid

For $x(t) = A\cos(2\pi f_0 t)$: the highest frequency is $f_0$, so: $$f_N = 2f_0$$

Example: A 1 kHz tone requires sampling at $f_s > 2$ kHz.

Sum of Sinusoids

For $x(t) = \cos(2\pi \cdot 100t) + \cos(2\pi \cdot 500t) + \cos(2\pi \cdot 1200t)$:

The highest frequency is 1200 Hz, so $f_N = 2400$ Hz. The Nyquist rate depends only on the highest frequency present, regardless of how many lower frequencies exist.

AM Modulated Signal

For $x(t) = m(t)\cos(2\pi f_c t)$ where $m(t)$ has bandwidth $W$:

The modulated signal occupies $f_c - W$ to $f_c + W$, so $f_{max} = f_c + W$ and $f_N = 2(f_c + W)$.

However, if bandpass sampling is used (sampling at a rate related to the bandwidth $2W$ rather than the highest frequency), reconstruction is still possible under certain conditions.

Real-World Audio

Human hearing ranges from 20 Hz to 20 kHz, so: $$f_N = 2 \times 20{,}000 = 40{,}000 \text{ Hz}$$

CD audio uses $f_s = 44.1$ kHz, which exceeds the Nyquist rate with about 10% margin.

Practical Oversampling

In practice, systems always sample above the Nyquist rate for several reasons:

Anti-aliasing filter limitations: Ideal brick-wall filters do not exist. Real filters have transition bands, so a guard band between $f_{max}$ and $f_s/2$ allows the filter to roll off.

Typical oversampling ratios:

• CD audio: $f_s = 44.1$ kHz for 20 kHz bandwidth → oversampling ratio = 1.1
• Professional audio: $f_s = 96$ kHz or 192 kHz → ratio = 2.4× or 4.8×
• Telecommunications: typically 4× to 8× oversampling
• Delta-sigma ADCs: 64× to 256× oversampling

Benefits of oversampling:

• Relaxed anti-aliasing filter requirements (gentle rolloff acceptable)
• Reduced quantization noise in the signal band (noise spreading)
• Easier digital post-processing and interpolation

Worked Example

Problem: A signal $x(t) = 3\cos(600\pi t) + 2\sin(1800\pi t) + \cos(2400\pi t)$ is to be sampled. Determine the Nyquist rate and whether $f_s = 1500$ Hz is sufficient.

Solution:

Identify frequencies: $f_1 = 300$ Hz, $f_2 = 900$ Hz, $f_3 = 1200$ Hz

Maximum frequency: $f_{max} = 1200$ Hz

Nyquist rate: $f_N = 2 \times 1200 = 2400$ Hz

Since $f_s = 1500$ Hz $< 2400$ Hz, this sampling rate is insufficient. Aliasing will occur for the 1200 Hz component, which will fold back to $|f_s - f_3| = |1500 - 1200| = 300$ Hz, corrupting the legitimate 300 Hz component.

Nyquist Rate for Bandpass Signals

For a bandpass signal occupying the band $[f_L, f_H]$ with bandwidth $B = f_H - f_L$, the minimum sampling rate can be lower than $2f_H$ under specific conditions:

$$f_s = \frac{2f_H}{m}$$

where $m$ is the largest integer satisfying $m \leq f_H/B$. This allows sampling at a rate related to the bandwidth rather than the carrier frequency — a technique called bandpass sampling or undersampling.

Common Misconceptions

1. "Nyquist rate means sampling at exactly $2f_{max}$": Actually, the theorem requires $f_s > 2f_{max}$ (strict inequality). Sampling at exactly $2f_{max}$ can fail for certain signals (e.g., sampling a sine wave at its zero crossings).

1. "Higher sampling rate always means better quality": Beyond the Nyquist rate, oversampling provides diminishing returns in terms of signal fidelity (though it helps with practical filter design).

1. "The Nyquist rate depends on signal power": It depends only on the highest frequency present, regardless of amplitude.

Key Takeaways

• Nyquist rate = $2f_{max}$ — the minimum sampling rate for perfect reconstruction of a bandlimited signal
• Sampling below the Nyquist rate causes aliasing (permanent information loss)
• The rate depends only on the highest frequency component, not on signal power or complexity
• Practical systems use oversampling (10% to 256×) to accommodate real filter limitations
• Bandpass signals can sometimes be sampled below $2f_{max}$ using bandpass sampling techniques
• CD audio at 44.1 kHz samples human-audible signals (up to 20 kHz) just above the Nyquist rate