Noise, Distortion, and Precision Design

Introduction

Analog systems do not process ideal signals. Every practical amplifier, sensor interface, converter driver, and communication front end is affected by random electrical fluctuations, nonlinear signal transformation, parameter variation, and environmental disturbances. These effects determine the smallest detectable signal, the accuracy of measurement, the fidelity of reproduction, and the long-term reliability of a design.

Noise and distortion are related but fundamentally different. Noise is generally an unwanted random fluctuation that exists even when the signal path is linear. Distortion is a systematic alteration of waveform caused primarily by nonlinearity, bandwidth limitation, or memory effects. Precision design addresses both, together with offset, drift, finite gain, component tolerances, interference, and layout-related errors.

At postgraduate level, the important design task is not merely to calculate a noise figure or total harmonic distortion. It is to establish an error budget, identify dominant mechanisms, select appropriate architectures and devices, and verify performance under realistic operating conditions.

Learning Objectives

After studying this chapter, the reader should be able to:

1. distinguish thermal, shot, flicker, burst, avalanche, and interference noise;
2. calculate noise spectral density, integrated noise, signal-to-noise ratio, and noise figure;
3. analyse noise in cascaded networks and refer noise to the input or output;
4. explain harmonic, intermodulation, slew-rate, crossover, and quantization-related distortion;
5. use transfer functions, intercept points, and error budgets for precision design;
6. design low-noise, low-distortion signal-conditioning circuits;
7. evaluate grounding, shielding, layout, thermal effects, and calibration;
8. interpret measurement results critically and identify limitations of simplified models.

1. Noise as a Stochastic Circuit Phenomenon

Noise is usually represented as a random process. Its instantaneous waveform cannot be predicted exactly, but statistical quantities such as mean value, variance, autocorrelation, and power spectral density can be measured and modelled.

For a zero-mean voltage-noise process, the mean-square value over bandwidth B is

where S_v(f) is the one-sided voltage-noise power spectral density in V²/Hz. For white noise, S_v(f) = e_n², where e_n is the amplitude spectral density in V/√Hz. Thus,

This expression assumes white noise over the entire measurement bandwidth and an ideal rectangular noise bandwidth. Real filters have an equivalent noise bandwidth (ENBW), defined as the bandwidth of an ideal rectangular filter passing the same noise power. Therefore, the practical expression is v_n,rms = e_n√B_ENBW.

Noise power is additive for statistically uncorrelated sources. Voltages themselves should not normally be added directly; their mean-square values are added after referring them to a common point in the circuit.

1.1 Thermal noise

Thermal noise, or Johnson–Nyquist noise, is produced by random carrier motion in a resistive element. The open-circuit voltage spectral density of a resistor R at absolute temperature T is

where k = 1.380649 × 10⁻²³ J/K is Boltzmann’s constant. The corresponding noise-voltage variance over bandwidth B is

The equivalent short-circuit current spectral density is

These two representations are equivalent for a resistor. Thermal noise is white over the frequency range in which the lumped resistor model is valid. It increases with resistance and temperature, and therefore precision circuits often avoid unnecessarily large resistors and excessive operating temperature.

1.2 Shot noise

Shot noise arises from the discrete nature of charge flow across a potential barrier, particularly in semiconductor junctions. For a direct current I,

where q = 1.602176634 × 10⁻¹⁹ C. The formula gives current-noise density in A/√Hz and assumes statistically independent carrier arrivals. Junction leakage, photodiode current, transistor collector current, and diode reference currents can all contribute shot noise.

1.3 Flicker noise

Flicker noise, also called 1/f noise, increases as frequency decreases. A simplified model is

where K is a device-dependent coefficient and α is commonly close to 1, although its value depends on technology, bias, area, and operating conditions. Flicker noise is especially important in instrumentation amplifiers, bridge sensors, biomedical systems, and precision DC amplifiers. Chopper stabilization and auto-zero techniques translate or periodically remove low-frequency offset and flicker-noise effects, but introduce switching artifacts and ripple that must also be filtered.

1.4 Other noise mechanisms

Burst or popcorn noise consists of abrupt transitions between discrete voltage or current levels and is associated with material defects. Avalanche noise results from carrier multiplication in a strongly reverse-biased junction and can be useful in some noise-source applications. Generation–recombination noise is associated with trapping and release of carriers. Electromagnetic interference (EMI), radio-frequency pickup, ground loops, and power-supply coupling are not always intrinsic device noise, but they are common practical noise sources.

2. Noise Models and Noise Figure

A two-port amplifier can be represented at its input by an equivalent voltage-noise source e_n in series and current-noise source i_n in parallel. If the source impedance is Z_S, the total input-referred noise density is approximately

when the sources are uncorrelated. If voltage and current noise are correlated, a cross-spectral term must be included. The simplified uncorrelated model can therefore be inaccurate for advanced transistor-level optimisation.

The source impedance creates a design trade-off. Voltage noise dominates for low source resistance, whereas current noise becomes important for high source resistance. The optimum device is consequently application-dependent rather than universally defined by the lowest voltage-noise figure.

Signal-to-noise ratio

Signal-to-noise ratio is

or in decibels,

For equal reference impedance, a voltage ratio may be written as SNR_dB = 20 log₁₀(V_S/V_N). The power and voltage forms must not be mixed unless the impedance conditions justify doing so.

Noise factor and noise figure

For a two-port network, noise factor is

where the input and output SNR values refer to the same signal. Noise figure is NF = 10 log₁₀(F) dB. A noiseless network has F = 1 and NF = 0 dB. Noise figure is conventionally specified at a reference source temperature, usually T₀ = 290 K.

For cascaded stages, Friis’ formula is

where F_i is the noise factor of stage i and G_i is its available power gain expressed as a linear ratio. The formula assumes matched power relationships and consistent reference impedances. It shows why the first stage usually dominates system noise and why a low-noise preamplifier should be placed before lossy passive elements whenever possible.

Worked Example 1: Input-referred resistor noise

A 1 kΩ resistor operates at 300 K over a 20 kHz rectangular bandwidth. Using k = 1.380649 × 10⁻²³ J/K,

v_n = √(4kTRB)

= √[4 × 1.380649 × 10⁻²³ × 300 × 1000 × 20,000]

≈ 0.182 μV rms.

The result is an rms voltage. It is not a peak voltage and should not be compared directly with a peak-to-peak signal without specifying the signal statistics and crest factor.

3. Distortion in Analog Systems

Distortion occurs when the output is not a scaled and delayed replica of the input. It may arise from nonlinear device characteristics, limited frequency response, slew-rate constraints, switching transitions, magnetic saturation, dielectric absorption, or thermal modulation.

A memoryless nonlinear amplifier can be approximated by a power series:

The coefficients a₁, a₂, and a₃ describe the local linear gain and nonlinear terms. This model is valid only over a limited operating range where higher-order terms can be neglected.

3.1 Harmonic distortion

For a sinusoidal input v_i = A cos(ωt), the square and cube terms generate components at DC, 2ω, and 3ω. Total harmonic distortion is

where V₁ is the rms fundamental component and V₂, V₃, … are rms harmonic components. In percent, THD is multiplied by 100; in decibels, THD_dB = 20 log₁₀(THD). Measurement bandwidth and the number of included harmonics must be stated.

3.2 Intermodulation distortion

With two input tones at frequencies f₁ and f₂, nonlinear terms generate frequencies such as 2f₁ − f₂ and 2f₂ − f₁. These third-order products are particularly troublesome because they may fall close to the wanted signals and cannot be removed easily by filtering.

The third-order intercept point (IP3) is an extrapolated figure obtained by extending the fundamental and third-order output curves until they intersect. It is not a safe operating level. A higher IP3 generally indicates better large-signal linearity, but it may require greater current, power dissipation, or device area.

For a two-tone test, if the fundamental output power is P_F and the third-order product is P_IM3, both in dBm, the approximate difference is related to output intercept point by

This relation assumes the small-signal cubic region and equal-tone conditions.

3.3 Gain compression and clipping

As input amplitude increases, transistor bias, output swing, or current drive may become limiting. The 1 dB compression point is the input or output level at which gain is 1 dB below its small-signal value. Clipping occurs when the waveform reaches a voltage or current boundary and produces substantial harmonics.

3.4 Slew-rate distortion

An amplifier with maximum output slew rate SR cannot reproduce a sinusoid of peak amplitude V_P and angular frequency ω unless

Equivalently, the maximum undistorted frequency is f_max = SR/(2πV_P). Slew-rate limitation produces triangular or visibly distorted waveforms and is distinct from small-signal bandwidth limitation.

3.5 Crossover and switching distortion

Push–pull output stages may exhibit a dead region near zero crossing if neither transistor conducts sufficiently. Class-AB biasing reduces this crossover distortion. In switched-capacitor and chopper circuits, charge injection, clock feedthrough, settling error, and aliasing are analogous transition-related problems.

4. Precision Design Architecture

Precision design begins with a specification expressed in measurable quantities: signal range, bandwidth, input impedance, gain accuracy, offset, drift, noise, linearity, common-mode range, power, temperature range, and lifetime stability. A circuit is precise only relative to these requirements.

4.1 Error budget

A useful error budget separates independent random contributions from systematic contributions. If independent rms errors are referred to the same output, the combined random error is

Systematic errors, such as a known gain error or calibration offset, should generally be summed algebraically or corrected through calibration rather than combined by root-sum-square. Correlated errors also require covariance terms:

4.2 Offset and bias-current error

An operational amplifier with input offset voltage V_OS and input bias currents I_B produces an output error determined by the circuit’s noise gain and resistive network. Offset voltage is multiplied by noise gain, not necessarily by signal gain. Bias-current mismatch flowing through source resistance creates an additional differential error.

Precision selection therefore requires attention to offset drift, input-current drift, common-mode rejection ratio (CMRR), power-supply rejection ratio (PSRR), open-loop gain, and temperature coefficients—not only nominal offset voltage.

4.3 Resistor and reference accuracy

Resistor ratio matching is often more important than absolute resistor accuracy in differential amplifiers. Temperature tracking between matched resistors can be excellent when they are fabricated together, whereas separately purchased precision resistors may have smaller initial tolerance but poorer long-term tracking.

A voltage reference should be evaluated for initial accuracy, temperature coefficient, output noise, load regulation, line regulation, long-term drift, and dynamic response. Filtering a reference reduces noise but may impair transient performance.

4.4 Feedback and linearity

Negative feedback reduces sensitivity to open-loop gain variation. For loop gain T(s) = A(s)β(s), the closed-loop transfer is

When |Aβ| is large, closed-loop gain is primarily set by the feedback network. Feedback can also reduce distortion and broaden bandwidth, but it cannot remove errors generated after the feedback sensing point. Stability, phase margin, noise gain, and loading must be assessed together.

4.5 Common-mode rejection and instrumentation

A differential amplifier ideally responds only to the difference between its inputs. Its CMRR is

and in decibels, CMRR_dB = 20 log₁₀(A_d/A_cm). High CMRR requires accurately matched ratios, symmetrical layout, adequate common-mode range, and protection against capacitive imbalance. Instrumentation amplifiers are valuable because they combine high input impedance, accurately defined gain, and high common-mode rejection.

5. Practical Noise and Distortion Reduction

Noise reduction is most effective when the coupling path is identified rather than treated by indiscriminate filtering. Important practices include:

• use the smallest necessary bandwidth;
• place low-noise gain before noisy lossy stages;
• choose device voltage and current noise for the actual source impedance;
• avoid unnecessarily high resistor values;
• use clean bias and reference supplies;
• separate high-current return paths from sensitive analog grounds;
• use shielding, guarding, differential signalling, and twisted pairs where appropriate;
• minimize high-impedance conductor length;
• control leakage and contamination in low-current circuits;
• provide thermal symmetry and allow adequate warm-up time.

Filtering must be considered in the time and frequency domains. A low-pass filter reduces integrated noise but introduces settling time and phase shift. Anti-alias filtering is essential before sampling, while reconstruction filtering is required after digital-to-analog conversion.

Distortion reduction usually involves operating devices in a more linear region, applying local or global feedback, increasing output-headroom margin, controlling signal amplitude, using degeneration, and selecting appropriate class-A, class-AB, or switching architectures. Higher bias current may improve linearity but increases power and thermal stress.

6. Measurement and Verification

A noise measurement is meaningful only when the instrument’s own noise, bandwidth, detector type, grounding, and averaging method are known. A spectrum analyser may display noise density in dBm/Hz, whereas an oscilloscope may report time-domain rms noise over an implicit bandwidth. These results cannot be compared without bandwidth conversion.

For distortion measurements, the source generator’s distortion must be substantially below the device under test. The test fixture should prevent power-supply ripple, ground loops, and electromagnetic pickup from contaminating the result. A two-tone test should specify tone spacing, tone power, termination, bandwidth, and whether products are measured at the input or output.

Monte Carlo analysis is useful for component tolerances and mismatch. Worst-case analysis is more conservative but may combine statistically unlikely extremes. Temperature sweeps, supply variation, ageing tests, and vibration or mechanical-stress tests are necessary when the product environment demands them.

7. Applied Case: Precision Bridge-Sensor Interface

Background

Consider an industrial pressure sensor producing a differential output of 2 mV/V at full scale from a bridge excited by 5 V. The full-scale differential signal is therefore 10 mV. The required measurement bandwidth is 100 Hz, and the system must resolve 0.1% of full scale over 0–50 °C.

Problem

The interface must provide high input impedance, reject excitation-related common-mode variation, contribute less than 10 μV rms input-referred noise, and maintain gain accuracy after temperature changes. A conventional single-ended amplifier is unsuitable because the bridge output is small compared with its common-mode voltage.

Analysis

The signal is 10 mV full scale, so 0.1% corresponds to 10 μV. This total error allowance includes sensor nonlinearity, amplifier offset, drift, noise, reference error, resistor mismatch, and interference. An error budget assigning the entire allowance to amplifier noise would be poor engineering; each contributor must receive a fraction of the available error.

At 100 Hz, a white input voltage-noise density of 50 nV/√Hz produces approximately

v_n = 50 nV/√Hz × √100 Hz = 0.5 μV rms

before accounting for the actual filter ENBW. This is comfortably below 10 μV, but low-frequency flicker noise and offset drift may dominate. The selected instrumentation amplifier should therefore have low 1/f noise, low offset drift, high CMRR over the relevant frequency range, and an input common-mode range compatible with bridge excitation.

Approach and solution

The bridge is driven by a low-noise reference, and the differential signal is amplified at the sensor location. A first-stage gain of 32 is selected so that the 10 mV signal becomes 320 mV, leaving headroom for overload and transients. A second stage provides additional gain and a low-pass filter with a 100 Hz signal bandwidth. Kelvin connections are used for bridge excitation, and analogue and digital return currents are separated until a controlled star or partition point.

A ratiometric measurement can reduce sensitivity to excitation-voltage variation when the same reference supplies both the bridge and converter reference. Calibration at two or more points corrects gain and offset, while temperature characterization determines whether residual drift meets the specification.

Outcome

The design meets the noise requirement only after the complete signal chain is evaluated. The major remaining risks are bridge thermal drift, cable interference, excitation noise, input protection leakage, and common-mode pickup. The case demonstrates that a low-noise amplifier alone does not guarantee a precision measurement.

Learning points

1. Precision is a system property, not a component label.
2. Common-mode rejection must be maintained in the complete layout.
3. Error budgets should include drift, reference, sensor, and environmental effects.
4. Filtering and calibration trade bandwidth, response time, and accuracy.

8. Limitations, Contemporary Developments, and Responsible Practice

Simple white-noise models are inadequate at very low frequencies, near device corners, during overload, or in circuits with strong correlation. Small-signal polynomial distortion models do not describe memory effects, hysteresis, dielectric absorption, thermal feedback, or switching spectra completely. Simulation models must therefore be validated by measurement.

Modern precision systems increasingly combine low-noise analogue front ends with digital calibration, oversampling, digital filtering, and adaptive interference cancellation. Chopper amplifiers and auto-zero amplifiers achieve excellent low-frequency performance, but designers must manage ripple, aliasing, input-current modulation, and electromagnetic emissions. Advanced processes integrate capacitors, resistors, MEMS sensors, and data converters, while machine-learning methods may assist fault diagnosis and calibration. Such methods do not eliminate the need for traceable measurements and physically interpretable error budgets.

Responsible engineering requires honest specification. Noise should not be quoted without bandwidth, temperature, source impedance, and detector conditions. Distortion should not be reported without signal level and measurement bandwidth. Calibration data should be retained with uncertainty estimates, and safety-critical systems should include independent verification, fault detection, and protection against single-point failures.

Best-Practice Design Workflow

1. Define signal range, bandwidth, environment, and allowable total error.
2. Draw the complete signal path, including sensor, protection, references, converters, supplies, and loads.
3. Establish an input-referred noise and distortion budget.
4. Select architecture and first-stage devices according to source impedance and dynamic range.
5. Analyse feedback stability, noise gain, common-mode range, and large-signal limits.
6. Design grounding, shielding, decoupling, thermal paths, and physical placement.
7. Simulate tolerances, temperature, interference, and overload conditions.
8. Build a controlled prototype and measure noise and distortion with a verified setup.
9. Calibrate only after systematic errors are understood.
10. Document assumptions, bandwidths, uncertainty, and acceptance criteria.

Chapter Summary

Noise is a random process described through spectral density, variance, and statistical correlation. Thermal noise, shot noise, and flicker noise are intrinsic mechanisms, while EMI, ground loops, and supply coupling are important practical disturbances. Uncorrelated noise powers combine by root-sum-square, and cascaded noise is analysed using input-referred models and Friis’ formula.

Distortion is a systematic alteration caused by nonlinearity, bandwidth, slew-rate limitation, clipping, switching, or memory. Harmonic distortion describes products of a single tone, while intermodulation distortion describes products generated by multiple tones. THD, compression point, IP3, and slew-rate limits are complementary specifications.

Precision design combines low noise, low distortion, low offset, low drift, high CMRR and PSRR, stable feedback, accurate references, matched components, careful layout, controlled thermal conditions, and verified measurement. A complete error budget must distinguish random, systematic, and correlated errors. Calibration and digital processing can improve performance, but neither replaces sound analogue architecture or traceable verification.

Key Terms

Noise: Unwanted random fluctuation in a signal or circuit variable.

Thermal noise: Random noise generated by thermal carrier motion in a resistive element.

Shot noise: Noise caused by discrete, statistically varying charge transport.

Flicker noise: Low-frequency noise whose spectral density increases approximately as frequency decreases.

Noise spectral density: Noise power per unit bandwidth.

Equivalent noise bandwidth: Rectangular bandwidth passing the same noise power as a real filter.

Signal-to-noise ratio: Ratio of signal power to noise power.

Noise factor: Degradation of SNR produced by a network.

Noise figure: Noise factor expressed in decibels.

Input-referred noise: Noise represented as an equivalent source at the circuit input.

Distortion: Systematic alteration of a waveform from its ideal form.

Harmonic distortion: Distortion producing integer multiples of a fundamental frequency.

Intermodulation distortion: Distortion producing sums and differences of input frequencies.

Total harmonic distortion: RMS harmonic content divided by RMS fundamental content.

Intercept point: Extrapolated intersection of fundamental and nonlinear-product characteristics.

Slew rate: Maximum rate of change of an amplifier output voltage.

Offset voltage: Differential input voltage required to produce zero output in an idealized measurement.

CMRR: Ratio of differential gain to common-mode gain.

Error budget: Allocation and combination of permitted inaccuracies.

Ratiometric measurement: Measurement using a common reference for excitation and conversion.

Review Questions

Short Answer Questions

1. Why are uncorrelated noise sources combined using mean-square values rather than algebraic voltage addition?
2. Define noise figure and explain the significance of 0 dB noise figure.
3. Distinguish harmonic distortion from intermodulation distortion.
4. Why can a high input impedance increase the importance of amplifier current noise?
5. Explain why slew-rate limitation is a large-signal effect.

Long Answer Questions

1. Derive the integrated rms noise of a white-noise source and discuss the role of equivalent noise bandwidth.
2. Explain the input-referred voltage/current noise model of an amplifier and its dependence on source impedance.
3. Discuss Friis’ formula for cascaded noise and explain why the first stage is usually critical.
4. Explain how negative feedback affects gain accuracy, bandwidth, and distortion, including its stability limitations.
5. Develop a complete precision-design methodology for a low-level bridge-sensor interface.

Analytical / Application Questions

1. A 2 kΩ resistor is operated at 300 K over 10 kHz. Calculate its rms thermal noise and state the assumptions used.
2. An amplifier has a slew rate of 0.5 V/μs and must reproduce a 2 V peak sine wave. Determine the highest frequency that can be reproduced without slew-rate distortion.
3. A two-stage amplifier has F₁ = 2, G₁ = 100, F₂ = 5, and G₂ = 10, all as linear power ratios. Calculate the total noise factor and noise figure, and interpret which stage dominates.

Multiple Choice Questions with Answers

Question 1

Which expression gives the rms voltage noise of a white-noise source with density e_n over equivalent noise bandwidth B?

A. e_n/B

B. e_n√B

C. e_nB²

D. √(e_n/B)

Correct Answer: B. e_n√B

Question 2

For a resistor, thermal-noise voltage density is proportional to:

A. √(4kTR)

B. 4kT/R²

C. 2qI

D. R/T

Correct Answer: A. √(4kTR)

Question 3

The main purpose of Friis’ cascaded-noise formula is to determine:

A. Total harmonic distortion

B. Closed-loop phase margin

C. Overall noise factor of cascaded stages

D. Maximum slew rate

Correct Answer: C. Overall noise factor of cascaded stages

Question 4

Which distortion product is a third-order intermodulation product for two tones f₁ and f₂?

A. f₁ + f₂

B. 2f₁

C. 2f₁ − f₂

D. f₁ − f₂ only

Correct Answer: C. 2f₁ − f₂

Question 5

An amplifier with slew rate SR can reproduce a sine wave of peak amplitude V_P without slew-rate distortion when:

A. SR ≤ 2πfV_P

B. SR ≥ 2πfV_P

C. SR = V_P/f²

D. SR ≥ f/V_P

Correct Answer: B. SR ≥ 2πfV_P

Question 6

If two independent rms error contributions are 3 μV and 4 μV, their combined rms error is:

A. 1 μV

B. 5 μV

C. 7 μV

D. 12 μV

Correct Answer: B. 5 μV

Question 7

Noise figure is related to noise factor by:

A. NF = F/10

B. NF = 20 log₁₀F

C. NF = 10 log₁₀F

D. NF = F²

Correct Answer: C. NF = 10 log₁₀F

Question 8

In a precision amplifier, offset voltage is generally multiplied by the circuit’s:

A. Noise gain

B. Common-mode voltage

C. Power-supply voltage

D. Noise bandwidth only

Correct Answer: A. Noise gain

Question 9

For a high source impedance, which amplifier noise component generally becomes more important?

A. Input current noise

B. Output clipping noise only

C. Crossover noise only

D. Reference drift only

Correct Answer: A. Input current noise

Question 10

A principal benefit of a ratiometric bridge measurement is reduced sensitivity to variation in:

A. The common excitation/reference voltage

B. The sensor’s physical quantity

C. The measurement bandwidth

D. The amplifier’s input impedance

Correct Answer: A. The common excitation/reference voltage