Digital Instrumentation and Data Acquisition

Introduction and Context

Digital instrumentation is the use of sensors, signal-conditioning circuits, converters, processors, communication interfaces, and software to measure, analyse, store, display, and control physical variables. Data acquisition (DAQ) is the systematic process of converting physical phenomena into digital data suitable for computation and decision-making. Together, digital instrumentation and DAQ form the measurement layer of modern laboratories, industrial automation, power systems, biomedical equipment, transportation, and cyber-physical systems.

A digital instrument is not merely an analogue instrument with a numerical display. Its accuracy depends on the complete measurement chain: sensor behaviour, loading, analogue conditioning, sampling, quantisation, timing, computation, calibration, communication, and interpretation. Errors introduced at any stage may be invisible in the final numerical display. Consequently, postgraduate study must treat instrumentation as a system-design problem rather than as a collection of isolated circuits.

A typical measurement objective can be expressed as estimating a measurand, x, from an acquired sequence of samples. The instrument produces an estimate x̂ together with an uncertainty determined by sensor noise, calibration, resolution, timing, environmental effects, and model assumptions.

Learning Objectives

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

1. Explain the architecture and operating principles of digital instrumentation and DAQ systems.
2. Analyse sampling, aliasing, quantisation, resolution, dynamic range, and conversion errors.
3. Select sensors, signal-conditioning circuits, ADCs, DACs, multiplexers, and timing systems for a measurement task.
4. Design a basic multichannel DAQ system and calculate relevant performance parameters.
5. Evaluate uncertainty, calibration, grounding, isolation, electromagnetic compatibility, and data integrity.
6. Interpret industrial communication, synchronisation, storage, and cybersecurity requirements.
7. Critically assess contemporary developments such as edge processing, distributed sensing, and AI-assisted diagnostics.

1. Architecture of a Digital Measurement System

The general signal path is:

measurand → sensor/transducer → signal conditioning → anti-alias filter → sample-and-hold → ADC → processor/memory → communication, display, or control

A reverse path is used when a digital controller must generate a physical output:

processor → DAC or pulse-width modulator → reconstruction filter → power/interface stage → actuator

Figure: Functional architecture of a digital instrumentation and data-acquisition system.

1.1 Sensors and transducers

A sensor converts a physical quantity into an electrical variable. A thermocouple produces a voltage related to temperature; a strain gauge changes resistance; a photodiode produces current; and a piezoelectric element produces charge or voltage under mechanical stress. The sensor model should include sensitivity, offset, range, bandwidth, nonlinearity, hysteresis, repeatability, drift, loading, and noise.

For a linear sensor, the ideal relationship is:

where y is the electrical output, x is the measurand, S is sensitivity in output units per input unit, and b is offset. The inverse estimate is:

where b̂ and Ŝ are calibrated estimates. This model is applicable only over the range in which linearity is acceptable. For nonlinear sensors, a polynomial, lookup table, physical model, or piecewise calibration is required.

1.2 Signal conditioning

Conditioning adapts the sensor output to the converter input. Common functions include amplification, attenuation, buffering, filtering, linearisation, excitation, rectification, bridge completion, cold-junction compensation, and galvanic isolation.

An instrumentation amplifier is often used for differential low-level signals because it offers high input impedance, high common-mode rejection, and accurately controlled gain. For a differential amplifier, the desired signal is the differential voltage, while common-mode voltage is ideally rejected. The common-mode rejection ratio is:

where A_d is differential gain and A_cm is common-mode gain. This expression assumes sinusoidal or small-signal gains at the frequency of interest. CMRR generally decreases with frequency and with gain-setting or resistor mismatch.

2. Analogue-to-Digital Conversion

An analogue-to-digital converter (ADC) maps a continuous input voltage into a finite set of digital codes. Three concepts must be distinguished:

• Sampling: selecting values at discrete instants.
• Quantisation: mapping each sampled amplitude to a finite level.
• Coding: representing the quantised level as a binary word.

For an ideal unipolar n-bit ADC with input range 0 to V_FS and 2ⁿ codes, the nominal code width is often approximated by:

where q is the least significant bit (LSB) size in volts. Some datasheets define the transition boundaries so that the ideal step is V_FS/(2ⁿ−1); therefore, the converter’s specified transfer convention must be checked. The approximation q = V_FS/2ⁿ is standard for resolution and quantisation-noise calculations.

The quantisation error for a rounding model is bounded by:

If the quantisation error is treated as uniformly distributed and uncorrelated with the signal, its rms value is:

This statistical model is useful for sufficiently varying signals; it is not reliable for a slowly varying or coherent input that produces repetitive codes.

For an ideal ADC driven by a full-scale sine wave, the signal-to-quantisation-noise ratio is approximately:

The equation assumes ideal quantisation, no analogue noise, a full-scale sinusoid, and an observation bandwidth extending to the Nyquist frequency. Effective number of bits (ENOB) can be estimated from a measured signal-to-noise-and-distortion ratio, SINAD:

SINAD must be expressed in decibels and measured under a defined bandwidth and input condition.

2.1 ADC architectures

Successive-approximation-register (SAR) ADCs use a comparator, internal DAC, and binary search. They provide a useful balance of speed, resolution, power, and latency and are common in embedded DAQ systems.

Flash ADCs use many comparators in parallel. They are extremely fast but require approximately 2ⁿ−1 comparators for n-bit conversion, making high resolution expensive.

Dual-slope ADCs integrate the input for a fixed time and then deintegrate using a reference. They offer excellent rejection of periodic interference, especially mains frequency, but are relatively slow and are common in precision digital multimeters.

Sigma-delta ADCs oversample, shape quantisation noise, and digitally filter the result. They can achieve high resolution and strong low-frequency performance, but introduce conversion latency and bandwidth limitations.

2.2 ADC error terms

Resolution is not the same as accuracy. Important specifications include offset error, gain error, integral nonlinearity (INL), differential nonlinearity (DNL), missing codes, aperture uncertainty, reference error, temperature drift, input leakage, and noise. A high-resolution converter may produce poor measurements if the reference, layout, sensor, or calibration is inadequate.

The total error should be expressed in the measurand’s units. If voltage error is ΔV and sensor sensitivity is S volts per unit of x, the equivalent input error is:

This relation assumes local linearity and neglects correlation among error sources.

3. Sampling Theory and Anti-Aliasing

Sampling converts a continuous-time signal x(t) into x[k] = x(kT_s), where T_s is the sampling interval and f_s = 1/T_s is the sampling frequency. A sinusoid of frequency f sampled at f_s is indistinguishable from frequencies differing by integer multiples of f_s. The observed alias frequency can be written as the frequency folded into the interval 0 to f_s/2.

The Nyquist-Shannon sampling theorem states that a band-limited signal with highest frequency B can be reconstructed from uniform samples if:

The strict inequality reflects the practical need for a transition band and margin. The theorem assumes ideal band limitation, exact timing, noiseless samples, and an appropriate reconstruction process. Real systems therefore sample faster than the theoretical minimum and use an analogue anti-alias filter before the ADC.

Aliasing is irreversible after sampling. Digital filtering after the ADC cannot determine whether a sampled component originated within the desired band or folded into it. Anti-alias filters must therefore be analogue, although oversampling and subsequent digital decimation can relax their required transition sharpness.

For a first-order low-pass filter, the magnitude response is:

where f_c is the −3 dB cutoff frequency. A first-order filter attenuates at approximately 20 dB per decade above cutoff. Higher-order filters provide greater rejection but may introduce phase distortion, component sensitivity, or transient effects.

Worked Example 1: Sampling and aliasing

A vibration signal contains useful components up to 3 kHz. Environmental interference may exist at 7 kHz. The minimum theoretical sampling frequency for the useful band is greater than 6 kHz. Selecting f_s = 10 kHz provides a transition margin, but the 7 kHz interference would alias to:

Thus, the interference would appear inside the useful band unless the analogue anti-alias filter sufficiently attenuates frequencies near 7 kHz. Merely increasing digital processing after conversion would not remove this ambiguity.

4. Digital-to-Analogue Conversion and Reconstruction

A DAC converts a digital code into a voltage or current. Important DAC architectures include binary-weighted, R–2R ladder, current-steering, and sigma-delta designs. DAC performance is characterised by resolution, monotonicity, INL, DNL, settling time, glitch impulse, reference stability, and output drive capability.

A staircase DAC output contains images at multiples of the update frequency. A reconstruction low-pass filter is used when a smooth analogue waveform is required. In control applications, a zero-order hold is often a useful model: each digital output is held constant until the next update.

Pulse-width modulation (PWM) is an alternative to a voltage-output DAC. If a PWM waveform has amplitude V and duty ratio D, its ideal average value after low-pass filtering is:

where 0 ≤ D ≤ 1. This relationship assumes a sufficiently low-pass load and neglects switching losses and ripple.

5. Data-Acquisition System Design

A DAQ design begins with the measurand and the decision to be made, not with the ADC resolution. The principal design questions are:

1. What range, bandwidth, accuracy, and uncertainty are required?
2. Is the signal single-ended, differential, isolated, or referenced to a floating source?
3. What sampling synchronism is required among channels?
4. What latency, storage rate, and communication reliability are acceptable?
5. What environmental stresses and safety requirements apply?

5.1 Multiplexed and simultaneous sampling

A multiplexer connects several analogue channels to one ADC. It reduces cost but introduces channel-to-channel delay and settling requirements. After switching, the ADC input network must settle to the required accuracy before conversion. The acquisition time depends on source resistance, switch resistance, input capacitance, and the ADC’s specified settling requirement.

Simultaneous-sampling ADCs use separate conversion paths or sample-and-hold circuits. They are preferred where phase relationships are important, such as power-system protection, motor-control analysis, and modal testing.

5.2 Throughput and storage

For N channels sampled at f_s samples per second per channel, with b stored bits per sample, the uncompressed payload rate is:

If storage is expressed in bytes per second:

Protocol overhead, packet headers, timestamps, error correction, and file-system overhead must be added in practice.

Worked Example 2: DAQ data rate

A system records 8 channels at 20 ksample/s per channel with 16-bit samples. The raw rate is:

Therefore:

One hour requires approximately 1.152 × 10⁹ bytes, or about 1.15 GB, before metadata and overhead. A storage and network design should therefore provide margin above 320 kB/s.

5.3 Timing and synchronisation

Measurement quality depends on clock quality as well as amplitude resolution. Clock jitter produces sampling-time uncertainty. For a sinusoidal input of frequency f_in and rms timing jitter σ_t, the approximate jitter-limited SNR is:

This assumes small, random, uncorrelated timing errors. Jitter becomes especially important for high-frequency or high-dynamic-range measurements.

Distributed systems may use trigger lines, precision time protocols, GPS-disciplined clocks, or local oscillators. A timestamp is meaningful only when its clock source, accuracy, time scale, and synchronisation uncertainty are documented.

6. Noise, Grounding, Isolation, and EMC

Noise sources include sensor noise, resistor thermal noise, amplifier voltage and current noise, reference noise, quantisation noise, electromagnetic interference, ground loops, clock feedthrough, and digital switching transients. Noise analysis should identify its spectral distribution and its position in the signal chain.

For a resistor R at absolute temperature T, the rms thermal-noise voltage over bandwidth B is:

where k is Boltzmann’s constant in J/K. The formula assumes a resistor in thermal equilibrium and a bandwidth over which the noise density is approximately white.

Grounding must distinguish signal reference from protective earth and chassis connections. A ground loop can cause unwanted voltage drops when current flows through shared impedances. Differential inputs, twisted-pair wiring, shielding, single-point reference strategies, galvanic isolation, and careful cable routing are common remedies.

Isolation is essential when channels have dangerous common-mode voltages, when equipment is at different potentials, or when a ground connection could damage equipment. Isolation amplifiers, isolated ADCs, transformers, optocouplers, and digital isolators have limits for working voltage, transient withstand, creepage, clearance, bandwidth, and common-mode transient immunity.

Electromagnetic compatibility requires both emissions control and immunity. A design should consider enclosure, filtering, cable shielding, PCB return paths, decoupling, surge protection, and separation of power and sensitive analogue circuits. Compliance must be assessed against the applicable institutional, industrial, and safety standards rather than assumed from a circuit diagram.

7. Calibration, Uncertainty, and Traceability

Calibration compares an instrument with a reference of known uncertainty and establishes a relationship between indication and measurand. It does not automatically correct every error. A calibration record should state the method, reference standards, environmental conditions, points tested, corrections, uncertainty, date, and acceptance criteria.

A simple two-point calibration uses:

For reference points (y₁, x₁) and (y₂, x₂):

This model assumes a linear response and stable conditions between calibration points.

For independent standard uncertainties u₁, u₂, …, u_m, the combined standard uncertainty is commonly approximated by the root-sum-square expression:

This assumes the inputs are uncorrelated. If correlation exists, covariance terms must be included. An expanded uncertainty is often reported as:

where k is a coverage factor selected according to the desired coverage probability and statistical basis, often close to 2 for approximately 95% coverage under suitable conditions.

Digital records should preserve raw data, calibration version, sensor identity, time base, units, scaling, filtering, missing-data flags, and operator or software provenance. Metadata is part of measurement validity, not an administrative afterthought.

8. Embedded Processing and Data Interpretation

A DAQ system may perform averaging, decimation, digital filtering, feature extraction, event detection, compression, or anomaly detection at the edge. Processing near the sensor reduces bandwidth and latency, but it can also discard information and complicate verification.

For an arithmetic average of N independent measurements with equal variance σ², the standard deviation of the average is:

The improvement assumes independent, identically distributed noise and no significant drift. Correlated noise, interference, and sensor drift reduce the benefit.

Digital filters must be specified by passband, stopband, ripple, phase response, order, sampling frequency, and implementation precision. A moving average can suppress high-frequency noise but introduces delay and has a frequency response with nulls determined by its window length. In control and protection applications, delay may be more harmful than noise.

AI and machine learning can classify waveforms, detect faults, and estimate virtual sensors. However, a trained model does not replace calibration or physical validation. Dataset shift, unbalanced fault classes, adversarial inputs, explainability, and the consequences of false alarms must be addressed. Safety-critical decisions require independent safeguards and documented validation.

9. Communication, Integrity, and Cybersecurity

DAQ systems commonly use local buses, Ethernet-based protocols, fieldbuses, serial links, or wireless networks. Selection depends on distance, throughput, determinism, electromagnetic environment, interoperability, and safety requirements.

Data integrity mechanisms include range checks, sequence numbers, checksums or cyclic redundancy checks, timestamps, plausibility tests, watchdogs, redundant sensors, and audit logs. A checksum can detect many transmission errors but does not prove that the measurement itself is correct.

Cybersecurity controls include authentication, least privilege, secure boot, signed firmware, network segmentation, encryption where appropriate, patch management, vulnerability monitoring, and recovery procedures. Measurement systems connected to operational technology must balance security updates with availability and validation requirements. False or manipulated data can cause unsafe control action even when the communication link appears operational.

10. Illustrative Case: Condition Monitoring of an Induction Motor

Background

Consider a 75 kW induction motor in a manufacturing plant. The maintenance team wants early warning of bearing degradation and electrical imbalance without interrupting production.

Problem

The original monitoring arrangement records one overall vibration value once per shift. It cannot identify transient events, correlate vibration with motor current, or distinguish mechanical faults from load changes.

Analysis

The engineering team identifies useful vibration content up to 10 kHz and requires phase comparison among three accelerometers and two current channels. The system must operate near variable-speed drives, where common-mode voltages and switching interference are substantial.

Approach and solution

IEPE accelerometers are selected with suitable frequency response and constant-current excitation. Differential, isolated input modules are used for motor currents. Each vibration channel is sampled simultaneously at 25 ksample/s, giving a Nyquist frequency of 12.5 kHz. Analogue anti-alias filters provide strong attenuation above the measurement band. A tachometer pulse supplies speed information, and all channels share a hardware trigger.

The edge processor computes band-limited rms values, crest factor, spectra, and order-tracking features, while raw waveform segments are retained when an event threshold is exceeded. Calibration coefficients and sensor serial numbers are attached to every file. Shield termination, cable routing, isolated power, and enclosure design are verified through EMC testing.

Outcome

The system detects a gradual increase in a bearing-related spectral component while motor current and speed remain normal. Maintenance is scheduled during planned downtime rather than after catastrophic failure. The retained raw segments allow engineers to verify that the alert was not caused by aliasing, clipping, or a communication fault.

Learning points

The case demonstrates that useful diagnostics require bandwidth planning, simultaneous sampling, anti-alias protection, synchronised timing, environmental design, metadata, and a decision-oriented data strategy. More samples alone do not guarantee better diagnosis.

11. Design Best Practices and Limitations

A robust design should:

• define accuracy, bandwidth, latency, safety, and uncertainty before selecting hardware;
• match sensor range to the signal while preventing overload;
• use analogue anti-alias filtering and verify settling after multiplexing;
• separate resolution, precision, accuracy, repeatability, and traceability;
• validate timing, clock drift, trigger latency, and channel synchronism;
• protect inputs against overload, surge, electrostatic discharge, and incorrect wiring;
• document calibration, units, scaling, filtering, and missing-data behaviour;
• retain raw or sufficiently rich data when future reanalysis is likely;
• test the complete chain under realistic temperature, vibration, EMC, and load conditions;
• apply access control, secure updates, and recovery procedures to connected instruments.

Limitations remain. Quantisation cannot be removed by software, aliasing cannot be corrected after inadequate sampling, calibration cannot compensate for an unstable sensor indefinitely, and machine learning cannot infer information that the sensor never captured. Cost, power, latency, storage, safety, and maintainability must be considered together.

Chapter Summary

Digital instrumentation converts physical variables into dependable digital information through an integrated chain of sensing, conditioning, filtering, sampling, conversion, processing, communication, and interpretation. ADC resolution is governed by code width, but system accuracy also depends on reference stability, linearity, noise, calibration, and environmental conditions. Sampling must satisfy the Nyquist requirement with practical margin, and analogue anti-alias filtering is indispensable. DAQ architecture must account for multiplexing, simultaneous sampling, throughput, storage, timing jitter, synchronisation, grounding, isolation, and EMC. Calibration and uncertainty analysis establish traceability, while metadata preserves the meaning of recorded data. Embedded processing and AI can improve responsiveness and diagnosis, but they require validation, cybersecurity, and awareness of information loss. Effective instrumentation is therefore a systems-engineering discipline in which measurement quality, safety, integrity, and decision relevance are designed together.

Key Terms

Sensor: Device that responds to a physical quantity and produces a measurable output.

Transducer: Device that converts energy or information from one physical form to another.

Signal conditioning: Amplification, filtering, isolation, excitation, or linearisation applied before conversion.

ADC: Analogue-to-digital converter that maps analogue amplitude to a digital code.

DAC: Digital-to-analogue converter that produces an analogue output from a digital code.

Sampling: Measurement of a continuous-time signal at discrete time instants.

Quantisation: Mapping of continuous amplitude values to finite levels.

Aliasing: Misidentification of frequency caused by insufficient sampling.

Nyquist frequency: One-half of the sampling frequency.

Anti-alias filter: Analogue filter that attenuates unwanted frequency components before sampling.

ENOB: Effective number of bits representing practical converter performance.

Multiplexer: Switching device that connects one of several inputs to a common path.

Simultaneous sampling: Acquisition of multiple channels at effectively the same instant.

CMRR: Common-mode rejection ratio, expressing rejection of common-mode signals.

Calibration: Comparison with a reference to establish measurement relationship and uncertainty.

Traceability: Property linking a measurement result to recognised standards through documented calibrations.

Metadata: Information describing units, timing, calibration, identity, and context of data.

Jitter: Variation in sampling time from its ideal value.

Galvanic isolation: Electrical separation that prevents direct conductive current between circuits.

Data integrity: Assurance that acquired, transmitted, and stored data remain complete and unaltered.

Review Questions

Short Answer Questions

1. Distinguish sampling, quantisation, and coding in an ADC.
2. Why is an analogue anti-alias filter required before an ADC?
3. Define ENOB and state how it differs from nominal resolution.
4. What information should be included in a calibration record?
5. State two reasons for using galvanic isolation in a DAQ system.

Long Answer Questions

1. Explain the architecture of a digital instrumentation system from sensor to data storage, identifying the principal error sources at each stage.
2. Compare SAR, flash, dual-slope, and sigma-delta ADC architectures in terms of speed, resolution, latency, and applications.
3. Develop a sampling and filtering strategy for a multichannel vibration measurement system.
4. Explain how grounding, shielding, differential measurement, and isolation influence measurement quality in an industrial environment.
5. Discuss the role and limitations of calibration, uncertainty analysis, metadata, and traceability in professional digital measurements.

Analytical / Application Questions

1. A 12-bit unipolar ADC has a 0–5 V input range. Calculate its nominal LSB size and the maximum ideal rounding error in voltage. Convert that error to an input error for a sensor sensitivity of 25 mV/°C.
2. A DAQ system records 6 channels at 12 ksample/s using 24-bit samples. Calculate the raw bit rate and approximate storage required for 8 hours, ignoring overhead.
3. A signal band extends to 4 kHz, while a strong interferer occurs at 9 kHz. Evaluate whether sampling at 12 ksample/s is safe and determine the interferer’s first alias frequency. Recommend a filtering or sampling change.

Multiple Choice Questions with Answers

Question 1

For an ideal 12-bit ADC with a 0–4.096 V input range, the approximate LSB size is:

A. 0.1 mV

B. 1 mV

C. 10 mV

D. 100 mV

Correct Answer: B. 1 mV

Question 2

Which function must occur before an ADC to prevent out-of-band components from irreversibly aliasing into the measurement band?

A. Digital averaging

B. Analogue anti-alias filtering

C. Data compression

D. Numerical differentiation

Correct Answer: B. Analogue anti-alias filtering

Question 3

For a full-scale sinusoidal input, the ideal quantisation SQNR is approximately:

A. 3.01n dB

B. 6.02n + 1.76 dB

C. 20n dB

D. 1.76n + 6.02 dB

Correct Answer: B. 6.02n + 1.76 dB

Question 4

Which ADC architecture is particularly associated with high resolution, oversampling, noise shaping, and conversion latency?

A. Flash

B. Successive approximation

C. Sigma-delta

D. Binary counter only

Correct Answer: C. Sigma-delta

Question 5

If a DAQ system has 4 channels, each sampled at 10,000 samples/s with 16-bit samples, its raw data rate is:

A. 40,000 bits/s

B. 160,000 bits/s

C. 640,000 bits/s

D. 2,560,000 bits/s

Correct Answer: C. 640,000 bits/s

Question 6

The common-mode rejection ratio is defined in terms of differential and common-mode gains as:

A. 20 log₁₀(A_cm/A_d)

B. A_d + A_cm

C. 20 log₁₀(A_d/A_cm)

D. A_d/A_cm in ohms

Correct Answer: C. 20 log₁₀(A_d/A_cm)

Question 7

A signal with useful bandwidth 5 kHz should theoretically be sampled at a frequency:

A. Less than 2.5 kHz

B. Exactly 5 kHz

C. Greater than 10 kHz

D. Equal to 10 Hz

Correct Answer: C. Greater than 10 kHz

Question 8

Which statement about quantisation error in the rounding model is correct?

A. It is always larger than one full LSB

B. It is bounded between −q/2 and +q/2

C. It is independent of ADC resolution

D. It can always be removed after conversion

Correct Answer: B. It is bounded between −q/2 and +q/2

Question 9

For N independent measurements with equal standard deviation σ, the standard deviation of their arithmetic average is:

A. Nσ

B. σ²/N

C. σ/√N

D. √N/σ

Correct Answer: C. σ/√N

Question 10

Which item is most directly an example of measurement metadata?

A. The colour of the instrument enclosure

B. The sensor serial number and calibration coefficient

C. The operator’s preferred programming language

D. The number of unused laboratory chairs

Correct Answer: B. The sensor serial number and calibration coefficient