Signal Conditioning and Measurement Integrity

Introduction

A sensor rarely produces a signal that can be connected directly to a digital acquisition system. Its output may be only a few microvolts, possess high source impedance, contain common-mode interference, exceed the safe input range, or occupy a bandwidth incompatible with the sampling rate. Signal conditioning converts such a signal into a form that can be measured without materially altering the information it represents.

Measurement integrity is broader than signal quality. It requires that the reported result remain traceable to the measurand through a defined model, calibrated signal chain, defensible uncertainty statement and controlled data-processing procedure. High numerical resolution does not guarantee integrity: a 24-bit converter can still produce inaccurate results because of loading, aliasing, reference drift, electromagnetic interference or incorrect calibration.

Learning Objectives

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

• design gain, attenuation, buffering, filtering and isolation stages;
• evaluate loading, common-mode rejection, noise and bandwidth;
• allocate ADC span while preserving overload capability;
• quantify ADC errors arising from quantisation, INL, DNL, reference noise, settling and aperture jitter;
• construct uncertainty budgets using sensitivity coefficients and covariance;
• distinguish calibration, adjustment, verification and metrological traceability;
• develop signal-conditioning systems that preserve safety, data integrity and measurement confidence.

1. The Measurement Chain

A practical measurement result is produced by interacting physical, analogue and digital subsystems.

Figure: Calibration is injected at a defined electrical interface; uncertainty and traceability encompass the complete measurement chain rather than implying feedback to the measurand.

A reference plane is a defined point at which a signal or error is expressed. Noise, offset and uncertainty must not be combined unless they are first referred to a common plane, such as the sensor input or ADC input.

1.1 Principal conditioning functions

2. Loading, Gain and Common-Mode Behaviour

2.1 Loading error

For a Thévenin source voltage V_s with source resistance R_s connected to an instrument input resistance R_in, the measured voltage is

The relative loading error is

These expressions apply to resistive, steady-state circuits. For AC measurement, resistances must be replaced by complex impedances. A nominally high input resistance may still exhibit significant capacitive loading at high frequency.

2.2 Instrumentation amplification

An instrumentation amplifier ideally produces

where G is differential gain, V_+ - V_- is differential input voltage and V_ref establishes the output reference level. Real devices also exhibit offset, finite bandwidth, input bias current, noise and restricted input common-mode and output swing ranges.

Common-mode rejection ratio is

where A_d and A_cm are differential and common-mode gains. An approximate common-mode-induced output error is G V_cm/10^CMRR/20 when CMRR is specified relative to the desired differential gain. The applicable frequency, gain setting and source-impedance balance must be checked because CMRR generally deteriorates with frequency and impedance mismatch.

2.3 Bridge transducers

For a quarter bridge with one resistance changing from R to R+\Delta R, one common polarity convention gives the magnitude of the small-signal output as

where V_ex is bridge excitation. This first-order approximation requires |\Delta R| \ll R and depends on which arm changes and how output polarity is defined. Exact bridge equations should be used for large resistance changes.

Ratiometric conversion, in which the ADC reference is derived from the same source as the bridge excitation, can suppress excitation-scale error. It does not automatically eliminate excitation noise outside the measurement bandwidth, lead resistance, self-heating or reference-path mismatch.

3. Span Allocation and Overload Design

Signal gain should exploit converter range without clipping normal transients, tolerances or specified overloads. For maximum expected input magnitude V_max and usable ADC magnitude V_FS,

Worked Example 1: Bridge amplifier gain

A bridge produces ±20 mV under normal operation. It must accommodate a 25% overload, and a bipolar ADC accepts ±1.25 V.

The overload input is

Thus,

Selecting G=48 gives:

• normal output: ±0.960 V;
• specified-overload output: ±1.200 V;
• normal utilisation of one-sided ADC span: 0.960/1.25=76.8\%;
• overload utilisation: 1.200/1.25=96.0\%.

Relative to normal output, the ADC limit permits 1.25/0.96-1=30.2\% headroom. Thus the design accommodates more than the required 25% overload; a further 4.17% headroom remains relative to the specified-overload output. Offset, gain tolerance and output-swing limits must also fit within this remaining allowance.

4. Noise, Interference and Filtering

4.1 Noise mechanisms

Relevant mechanisms include resistor thermal noise, semiconductor voltage and current noise, shot noise, flicker noise, reference noise, power-supply coupling and electromagnetic pickup. The rms thermal-noise voltage of a resistor over an ideal rectangular bandwidth B is

where k is Boltzmann’s constant, T is absolute temperature, R is resistance and B is noise bandwidth. For a non-rectangular filter and input-referred noise spectral density e_n(f),

Uncorrelated rms noise sources combine by root-sum-square after referral to the same plane. Coherent interference, such as power-frequency pickup, must not automatically be treated as independent random noise.

4.2 Analogue filtering and alias control

A first-order RC low-pass filter has

It provides only 20 dB/decade asymptotic attenuation. Higher-order Butterworth filters offer a flat passband, Bessel filters preserve waveform shape through nearly constant group delay, and Chebyshev filters obtain sharper transitions at the cost of ripple or phase distortion.

For sampling frequency f_s, frequencies above f_s/2 can alias into the Nyquist band. The anti-alias filter must reduce out-of-band energy sufficiently before sampling; digital filtering after aliasing cannot reconstruct the original frequency.

Filter selection also affects settling. A first-order system subjected to a step has residual fraction

where \tau=RC. Narrow bandwidth reduces noise but increases response time, creating a fundamental design trade-off.

4.3 Grounding, shielding and isolation

A cable shield intercepts electric-field coupling but cannot correct poor return-current design. Low-level differential connections should use balanced routing, controlled shield termination and minimal loop area. Single-point grounding may be useful at low frequencies, while high-frequency systems require low-inductance bonding and ground-plane analysis. Galvanic isolation is required when safety, large common-mode voltage or ground-loop interruption demands it, but isolation components introduce gain, offset, bandwidth and transient-immunity limitations.

5. ADC Integrity

5.1 Resolution and quantisation uncertainty

For an ideal N-bit ADC with full input span V_FSR, the nominal code width is

If quantisation error is sufficiently decorrelated from the input and uniformly distributed over [-q/2,q/2], its standard uncertainty is

This statistical model is inappropriate for a constant or coherently sampled signal that remains in one code. Dither can randomise quantisation error, though it increases instantaneous noise.

For an ideal full-scale sinusoid and quantisation noise alone,

The relation assumes an ideal ADC, a full-scale sinusoid and appropriate spectral analysis. Effective number of bits derived from measured signal-to-noise-and-distortion ratio is

5.2 Differential and integral nonlinearity

For code width W_k, differential nonlinearity is

A code can be missing when its transition interval collapses; DNL\leq-1 LSB indicates a zero or nonphysical code width in the adopted model. Integral nonlinearity is the deviation of actual transition levels from a chosen straight reference line, expressed in LSB. Endpoint and best-fit definitions yield different INL values and must be identified.

DNL affects code uniformity and monotonicity. INL produces input-dependent systematic error that averaging does not remove. Calibration at two points corrects offset and gain but not general INL.

5.3 Reference error and noise

For an ideal unipolar converter,

apart from code limiting and rounding. Small reference uncertainty produces approximately

when input and reference uncertainties are independent and the input is treated as fixed. The equivalent input-referred standard uncertainty due to reference variation is

Reference noise therefore has a signal-dependent effect. Reference output impedance, decoupling, load transients and ADC reference-input current must be considered, not merely nominal reference accuracy.

5.4 Acquisition and settling

A switched-capacitor ADC must charge its sampling capacitor through source and switch resistance. For a first-order acquisition model and full-scale step, settling to within one-half LSB requires

where \tau=R_eqC_s. This is a simplified sufficient condition for a single-pole system; parasitic poles, charge kickback and multiplexer memory can demand longer acquisition time or a buffer amplifier. Buffer stability with capacitive ADC input loading must also be verified.

5.5 Aperture jitter

Sampling-time uncertainty converts input slew rate into voltage uncertainty:

For a sinusoid with peak amplitude A, frequency f_in and rms timing jitter \sigma_t, the rms voltage error is

The corresponding jitter-limited signal-to-noise ratio is

This assumes small, random, signal-independent jitter and a sinusoidal input. Clock phase noise, ADC aperture uncertainty and clock-distribution jitter should be combined consistently, accounting for correlation where known.

Worked Example 2: ADC resolution and jitter

An 18-bit bipolar ADC spans −1.25 V to +1.25 V. Therefore,

Under the uniform quantisation model,

With the gain-48 conditioner, these values refer to 0.1987\ \mu V per code and 0.0574\ \mu V standard quantisation uncertainty at the conditioner input.

If a 100 kHz sinusoid is sampled with 2 ps rms jitter,

Jitter is therefore unlikely to dominate an 18-bit system at this frequency, although analogue noise and distortion may dominate well before the theoretical resolution is achieved.

6. Measurement Models and Uncertainty

6.1 General propagation law

For a measurement model y=f(x₁,\ldots,x_n), the combined standard uncertainty under first-order propagation is

where c_i=\partial f/\partial x_i is a sensitivity coefficient. The approximation requires adequate local linearity; Monte Carlo propagation may be preferable for strong nonlinearity, bounded distributions or discontinuous processing.

Type A uncertainty is evaluated statistically from repeated observations. Type B uncertainty uses other information, including calibration certificates, specifications, resolution and prior data. These categories describe evaluation methods, not whether an effect is random or systematic.

6.2 Gain–offset model

For

and independent inputs,

If correlations exist, covariance terms must be added. Sensor-input uncertainty is multiplied by G, while absolute gain uncertainty is multiplied by V. An alternative is to refer every contribution explicitly to one common plane before root-sum-square combination.

Worked Example 3: Output-referred uncertainty budget

A conditioner has G=48 and receives V=20.000\ \text{mV}. Assume the following independent standard uncertainties:

• sensor/input voltage: u(V)=4.167\ \mu V, giving G u(V)=0.200\ \text{mV} at the output;
• absolute gain: u(G)=0.005\ \text{V/V}, giving V u(G)=0.100\ \text{mV};
• output offset: u(V_o)=0.080\ \text{mV};
• ADC and reference contribution already output-referred: 0.050 mV.

All terms are now expressed at the conditioner output. Therefore,

and

For coverage factor k=2, the expanded uncertainty is approximately U=0.485\ \text{mV}. Interpreting this as about 95% coverage requires an approximately normal output distribution and adequate effective degrees of freedom; k=2 is not universally an exact 95% statement.

6.3 Calibration, adjustment and verification

Calibration establishes a relationship between indication and reference values, including uncertainty. Adjustment changes the instrument to reduce error. Verification supplies evidence that requirements are met. These activities are distinct: calibration need not include adjustment, and a successful verification does not by itself establish complete traceability.

Metrological traceability requires an unbroken, documented chain of calibrations to a stated reference, each contributing uncertainty. Calibration intervals should be risk-based, using drift history, environmental exposure, criticality and required uncertainty rather than an arbitrary fixed period.

7. Illustrative Case: Structural Strain Monitoring System

Background

An illustrative engineering team must monitor a 350 Ω strain-gauge bridge on a fatigue-test frame. Normal bridge output is ±20 mV, while mechanical events can produce a 25% overload. Required signal bandwidth is 200 Hz, and nearby motor drives generate substantial common-mode and switching interference.

Problem

Direct ADC connection would underuse the converter range and expose it to common-mode interference. Excessive gain could clip overload events, while an inadequately settled multiplexer or poorly designed filter could distort peak strain. The result must also be defensible in an accredited laboratory environment.

Analysis and approach

1. Excitation: A low-noise 5 V bridge supply is used. The ADC reference is derived ratiometrically from the excitation after confirming compatible loading and bandwidth.
2. Input stage: A low-noise instrumentation amplifier with gain 48 is selected. Its input common-mode range, bias-current error, output swing and CMRR at motor-switching frequencies are verified.
3. Span: Normal output is ±0.960 V and specified overload output is ±1.200 V for the ±1.25 V ADC, preserving additional headroom.
4. Filtering: A differential low-pass anti-alias filter is designed with a passband covering 200 Hz. Its order and cutoff are chosen from the required stopband attenuation at the sampling frequency rather than from the Nyquist criterion alone.
5. Conversion: An 18-bit ADC is evaluated for noise-free resolution, INL, reference input behaviour and acquisition settling, not merely nominal bit count.
6. Wiring: Shielded twisted-pair bridge leads, balanced input resistances and a controlled shield termination reduce magnetic pickup and CMRR degradation.
7. Calibration: A traceable bridge simulator injects zero, positive and negative calibration points at the conditioner input. Calibration data, temperature, gain setting, firmware version and uncertainty are retained.
8. Validation: Tests include overload recovery, common-mode injection, filter step response, temperature cycling and comparison against a reference channel.

Outcome

The system measures the required bandwidth without clipping the specified overload. Uncertainty analysis identifies bridge calibration and analogue gain stability—not ADC quantisation—as the dominant contributions. This prevents unnecessary expenditure on a higher-resolution ADC and directs resources toward thermal control, calibration and wiring quality.

Learning points

• Span utilisation must be evaluated together with overload and tolerance limits.
• Converter resolution is only one component of measurement capability.
• Ratiometric design suppresses some excitation errors but does not replace noise and thermal analysis.
• Validation must challenge the complete chain under realistic electrical and environmental conditions.

8. Integrity Governance, Safety and Contemporary Practice

Measurement integrity includes both metrological and information-governance controls. Raw data, calibration coefficients, timestamps, units, range settings and processing versions should be retained with audit trails. Undocumented filtering, selective deletion or post hoc range changes can invalidate technically accurate measurements.

Relevant frameworks include the JCGM Guide to the Expression of Uncertainty in Measurement, ISO/IEC 17025 for competent testing and calibration laboratories, IEC 61010 for measurement-equipment safety and IEC 61326 for electromagnetic compatibility. Applicability depends on equipment type, jurisdiction and intended use.

AI methods can detect drift, anomalous channels and impending sensor failure, but their outputs should not silently replace traceable calibration. Training-data provenance, false alarms, model drift and explainability are material concerns. Emerging systems increasingly combine self-test injection, digitally assisted analogue correction, synchronised distributed sampling and machine-readable calibration metadata. Such capabilities improve maintainability only when their algorithms and configuration states are controlled.

9. Engineering Best Practices

• Define the measurand, bandwidth, operating range and permissible uncertainty before selecting hardware.
• Establish reference planes for all error and noise quantities.
• Reserve span for overload, offset, gain tolerance and transients.
• Analyse source impedance and ADC acquisition settling together.
• Specify anti-alias attenuation from the actual out-of-band spectrum.
• Evaluate INL, DNL, reference behaviour, jitter and noise-free resolution.
• Separate calibration, adjustment and verification records.
• Validate under temperature, common-mode, overload and electromagnetic conditions.
• Preserve raw data, units, timestamps and processing metadata.
• Review uncertainty budgets whenever hardware, firmware or calibration models change.

Chapter Summary

Signal conditioning aligns sensor outputs with acquisition-system requirements through amplification, attenuation, buffering, filtering, excitation, protection and isolation. Measurement integrity requires more than adequate resolution: loading, common-mode rejection, noise, aliasing, settling, nonlinearities, reference stability and timing uncertainty all influence the reported result.

ADC performance must be assessed through quantisation uncertainty, ENOB, INL, DNL, reference effects, acquisition settling and aperture jitter. Uncertainty contributions must be transformed through their sensitivity coefficients or explicitly referred to a common measurement plane before combination. Traceable calibration, realistic validation, controlled metadata and risk-based governance complete the measurement chain.

Key Terms

Aliasing: Irreversible mapping of frequency content above the Nyquist limit into the sampled band.

Aperture jitter: Random uncertainty in the instant at which an ADC samples its input.

Calibration: Establishment of the relationship between indication and reference values, including uncertainty.

CMRR: Ratio expressing rejection of common-mode signals relative to differential signals.

DNL: Difference between an actual ADC code width and one ideal LSB.

ENOB: Effective resolution inferred from measured signal-to-noise-and-distortion performance.

Expanded uncertainty: Combined standard uncertainty multiplied by a stated coverage factor.

INL: Deviation of ADC transition levels from a specified straight reference line.

Instrumentation amplifier: Precision differential amplifier designed for high input impedance and high common-mode rejection.

Measurement integrity: Assurance that a result remains valid, traceable, reproducible and appropriately documented.

Metrological traceability: Documented calibration chain linking a result to a stated reference with associated uncertainties.

Noise bandwidth: Equivalent rectangular bandwidth that passes the same noise power as the actual filter.

Quantisation: Mapping of a continuous-amplitude input to discrete digital levels.

Ratiometric measurement: Measurement in which the input and reference share a common scale source.

Reference plane: Defined location in a signal chain to which signals, errors or uncertainties are referred.

Sensitivity coefficient: Partial derivative describing how an input quantity affects a measurement result.

Settling time: Time required for a response to remain within a specified error band.

Type A evaluation: Uncertainty evaluation using statistical analysis of observations.

Type B evaluation: Uncertainty evaluation using information other than repeated-observation statistics.

Verification: Provision of evidence that specified requirements have been fulfilled.

Review Questions

Short Answer Questions

1. Why must noise contributions be referred to a common plane before combination?
2. Distinguish calibration, adjustment and verification.
3. State the conditions underlying the uniform quantisation-error model.
4. Why can source-impedance imbalance reduce practical common-mode rejection?
5. What information is conveyed by ADC integral nonlinearity?

Long Answer Questions

1. Explain the complete signal-conditioning chain for a low-level bridge transducer.
2. Discuss the trade-off among bandwidth, noise reduction, alias suppression and settling time.
3. Derive the uncertainty expression for y=GV+V_o, including covariance terms.
4. Critically compare ADC resolution, ENOB, DNL and INL as measures of converter performance.
5. Explain how traceability, metadata and validation procedures support measurement integrity.

Analytical / Application Questions

1. Design a gain stage for a ±12 mV sensor, a 30% overload requirement and a ±2.5 V ADC. Include gain tolerance and offset allowances.
2. An ADC samples a 2 MHz sinusoid with 8 ps rms timing jitter. Calculate the jitter-limited SNR and assess whether it is adequate for 14-bit ideal quantisation performance.
3. Construct an uncertainty budget for a bridge system that includes sensor calibration, gain, offset, reference, quantisation and temperature effects. Identify all sensitivity coefficients and reference planes.

Multiple Choice Questions with Answers

Question 1

A ±20 mV sensor is amplified by 48 before connection to a ±1.25 V ADC. What percentage of the ADC one-sided span is used during normal operation?

A. 60.0%

B. 76.8%

C. 80.0%

D. 96.0%

Correct Answer: B. 76.8%

Question 2

For an ideal 16-bit ADC with a 20 V full input span, what is the nominal code width?

A. 76.3 μV

B. 152.6 μV

C. 305.2 μV

D. 610.4 μV

Correct Answer: C. 305.2 μV

Question 3

Under the uniform quantisation-error model, the standard uncertainty associated with ADC code width q is:

A. q/2

B. q/\sqrt{12}

C. q/\sqrt{2}

D. q²/12

Correct Answer: B. q/\sqrt{12}

Question 4

For y=GV+V_o, which term represents the output uncertainty contribution from absolute gain uncertainty u(G) when inputs are independent?

A. G u(G)

B. V_o u(G)

C. V u(G)

D. u(G)/V

Correct Answer: C. V u(G)

Question 5

Which ADC specification primarily describes input-dependent deviation of code-transition levels from a straight reference line?

A. Aperture jitter

B. Integral nonlinearity

C. Sampling frequency

D. Input resistance

Correct Answer: B. Integral nonlinearity

Question 6

What is the principal reason an analogue anti-alias filter must precede sampling?

A. Digital filtering cannot reverse frequency components that have already aliased

B. It increases ADC differential nonlinearity

C. It removes all quantisation error

D. It guarantees zero phase delay

Correct Answer: A. Digital filtering cannot reverse frequency components that have already aliased

Question 7

For a first-order acquisition system, settling a full-scale step to within one-half LSB of an ideal N-bit ADC requires at least:

A. \tau\ln(2^N)

B. (N+1)\tau\ln 2

C. \tau/2^N

D. 2\pi N\tau

Correct Answer: B. (N+1)\tau\ln 2

Question 8

A 100 kHz sinusoid is sampled with 2 ps rms aperture jitter. Approximately what is the jitter-limited SNR?

A. 78 dB

B. 98 dB

C. 118 dB

D. 138 dB

Correct Answer: C. 118 dB

Question 9

Four independent, output-referred standard uncertainties are 0.20, 0.10, 0.08 and 0.05 mV. What is their combined standard uncertainty?

A. 0.243 mV

B. 0.330 mV

C. 0.430 mV

D. 0.590 mV

Correct Answer: A. 0.243 mV

Question 10

Which statement correctly distinguishes calibration from adjustment?

A. Calibration always changes instrument gain; adjustment only records error

B. Calibration establishes an indication–reference relationship; adjustment changes the instrument

C. Calibration and adjustment are identical when a traceable standard is used

D. Adjustment establishes traceability without requiring calibration

Correct Answer: B. Calibration establishes an indication–reference relationship; adjustment changes the instrument