Unit 8: Magnetic Field Measurements: Search Coils, Fluxmeters, and Hall-Effect Sensors

Unit 8 — Noise, selective detection and automated measurements

Why this unit matters. Distinguish interference from random noise; define a measurement protocol that can be repeated.

Application setting. Reproducible software-only studies of detection, averaging, command sequencing and reporting. The examples are synthetic, deliberately low-energy, and do not establish safe procedures for handling energized high-power systems.

Concept map

Broadband Johnson noise, 1/f noise, clock interference and coherent pickup have different signatures and countermeasures. Repeated observations can characterize short-term repeatability, but if noise is correlated the usual √N improvement from averaging is optimistic. Analyze a time series in both time and frequency domains before deciding whether to filter, shield or redesign the circuit.

A lock-in identifies the part of a signal coherent with a known reference; this is not the same as reconstructing every component in a waveform. Its bandwidth is set by filtering and integration time. Automation can consistently repeat commands, but state checks, metadata, calibration and error handling are needed before results are scientifically reproducible.

Lessons and reading sequence

• Lesson 8.1 — [Thermal noise and interference coupling](../lessons/U08-L01-thermal-noise-and-interference-coupling.md)
• Lesson 8.2 — [Averaging, autocorrelation and effective sample size](../lessons/U08-L02-averaging-autocorrelation-and-effective-sample-size.md)
• Lesson 8.3 — [Lock-in detection and synchronous demodulation](../lessons/U08-L03-lock-in-detection-and-synchronous-demodulation.md)
• Lesson 8.4 — [Automated acquisition, SCPI and reproducibility](../lessons/U08-L04-automated-acquisition-scpi-and-reproducibility.md)

After reading the lessons, complete the [unit assessment](../assessments/unit-08.html), [research-style project](../projects/unit-08.html), and [interactive software model](../simulations/model-08.html). Follow the sequence for conceptual continuity, but revisit earlier examples when error analysis changes the meaning of a later result.

Five essential terms

• PSD: Power spectral density per unit bandwidth.
• Autocorrelation: Similarity of signal with delayed versions.
• Lock-in: Synchronous measurement against reference.
• SCPI: Common structured instrument-command language.
• Provenance: Recorded origin and processing history of data.

Cross-lesson example: competing explanations

Assume two engineers report conflicting results for the same notional specimen. One engineer repeats a measurement ten times and obtains a very narrow cluster. The second uses another range, reference, sample timing or wiring topology and obtains a systematic difference. The first engineer's repeatability does not prove that the second engineer is wrong. In this unit, list the physical differences between instrument chains and predict which could create a reproducible offset or scale change. Identify one controlled experiment that would separate the causes without relying on a higher number of decimal places.

Unit laboratory notebook template

Use fields for measurand, connection topology, symbol definitions, numeric input sources, model assumptions, instrument settings, raw observations, rejection criteria, derived values, uncertainty budget, limitations and conclusion. Include a citation for each manufacturer specification if a real instrument is discussed. Reserve a separate field for software version and simulator parameters; a teaching simulator is not a calibrated artifact.

Reader checkpoint

1. Explain the difference between an instrument's reported indication and the intended measurand.
2. Give one case in which making the display resolution finer does not improve the measurement result.
3. Identify one quantity in the unit simulator that must be experimentally calibrated before using the equation for a real system.
4. Define the most important limitation of the idealized model and a practical observation that could falsify it.

Answer guidance. A sound response links the observed quantity to the physical chain and names a bias or uncertainty source instead of vaguely saying 'human error.' It distinguishes a model prediction from independent evidence. Complete safety qualification is outside the scope of this book.

Safety note: The examples are synthetic and intended for analysis or software-only simulation. Do not connect measuring instruments to mains, high-energy equipment, CT secondaries or exposed conductors without authorized training, appropriately rated equipment and site-specific procedures. Instrument-category markings, isolation ratings and current limits must be verified in the manufacturer documentation.

8.1 — Thermal noise and interference coupling

Unit 8: Noise, selective detection and automated measurements Level: undergraduate, with prerequisite circuit theory, calculus and introductory probability. Estimated study: 70–100 minutes, including the numeric investigation. Tools: calculator, spreadsheet or the original offline simulator in this book. Evidence: examples are synthetic unless explicitly identified otherwise.

A problem that a measurement must answer

Johnson–Nyquist noise is a thermal fluctuation whose mean-square voltage grows with resistance, temperature and noise bandwidth. Interference from mains fields or digital clocks is not the same phenomenon: it may be coherent, correlated and removable with topology or shielding. Filtering always trades bandwidth for response time.

The application for this unit is: Reproducible software-only studies of detection, averaging, command sequencing and reporting. The meaningful output is not necessarily the number displayed by a meter. First decide whether it is the unperturbed circuit voltage, a time average, a reference-scale resistance, a physical temperature or a confidence-qualified estimate. That decision determines the topology, calibration approach and acceptable model. In an instrumentation report it is better to leave a result undecided than to replace the unmeasured quantity with a convenient proxy without disclosure.

Mathematical model and physical assumptions

Central relation: For ideal resistor in a bandwidth B, v_noise,rms=√(4 k_B T R B).

Derivation strategy

Start by writing the quantity and its reference state as a sentence. Identify every input that is actually measured and every value that is assumed. If a parameter came from a manufacturer specification, label it as a specification rather than calling it an independent observation. Use the equation to map those inputs onto the desired output. For a laboratory record, store instrument mode and environmental conditions alongside the number. In this case the key input evidence is the reading or physical parameter introduced above, not the final calculated answer. An experimenter should separate constants taken from a definition, instrument parameters taken from a datasheet and conditions measured on the day of use. If an assumed parameter is poorly characterized, a small model residual does not make the resulting prediction precise. The uncertainty of a quantity calculated from a product, quotient or nonlinear relation must follow its sensitivity to each contributing input, not a simple count of display digits.

Why the relation is useful

Random versus deterministic contamination Broadband Johnson noise, 1/f noise, clock interference and coherent pickup have different signatures and countermeasures. Repeated observations can characterize short-term repeatability, but if noise is correlated the usual √N improvement from averaging is optimistic. Analyze a time series in both time and frequency domains before deciding whether to filter, shield or redesign the circuit. The formula above is one of the operational links between a physical system and a numerical record. It is not an endorsement of any particular manufacturer's ratings, which must be checked independently. For instance, ideal resistance or sampling calculations disregard temperature-dependent and frequency-dependent behavior unless those are included explicitly.

Fully worked numerical case

At T=300 K, R=10 kΩ and B=1 kHz, v_noise,rms≈0.407 μV using k_B=1.380649×10⁻²³ J/K. This is a theoretical resistor-noise estimate, not total amplifier input noise.

Uncertainty, nonidealities and instrumentation bias

Amplifier current noise through source impedance and flicker noise can exceed this white-noise floor. A strong periodic 50/60 Hz pickup is not best modeled as independent Gaussian thermal noise. The first concern affects the credibility of the numerical parameter; the second suggests a rival explanation of the observed behavior. A useful uncertainty analysis therefore goes beyond listing a collection of percentages. It defines the output measurand, classifies major sources as additive or multiplicative, evaluates correlations when necessary, and identifies the sensitivity of the output to each one. An uncertainty budget should preserve documentary support for Type B inputs and raw observations for Type A estimates. If the model is an approximation, give a bounded validity statement in addition to any statistical interval.

A counterexample to overconfidence

Detection with references A lock-in identifies the part of a signal coherent with a known reference; this is not the same as reconstructing every component in a waveform. Its bandwidth is set by filtering and integration time. Automation can consistently repeat commands, but state checks, metadata, calibration and error handling are needed before results are scientifically reproducible. This contextual distinction provides a way to design a discriminating check. For an instrument-chain problem, change input range, source impedance, frequency or wiring to see whether the measurement follows the predicted dependence. For a data-analysis problem, change sample rate, observation length or independent reference while holding the physical input constant. If a conclusion persists only because the same assumption is reused in every run, that persistence supplies little additional evidence.

Engineering use and what would be required to trust it

Identify the required measurement bandwidth before picking an amplifier noise density. Calibration records, environmental limits and independent checks belong in the method, not just in appendices. Choose an experiment that isolates the main influence while staying within authorized, properly rated equipment and low-energy pedagogical conditions. A formula computed with fictional values is evidence that arithmetic was performed, not that a real device passed a standard. When the output feeds a safety, certification or financial decision, the appropriate reviewer and agreed acceptance rule must be identified before collecting measurements.

Study exercise 1 — reconstruct the baseline

Derive or justify For ideal resistor in a bandwidth B, v_noise,rms=√(4 k_B T R B). from the relevant conservation law, circuit relation, transfer model or statistical definition. Recompute the worked example, label every unit and distinguish the physical value from what an actual instrument could have displayed. Identify one data item whose uncertainty is not quantified by the example. Explain what would happen if that datum were systematically too high by a small amount.

Study exercise 2 — sensitivity experiment

Use the same physical configuration but change a controlled parameter by 10% in the direction that would expose a plausible failure mechanism. Compute or predict the response using the stated equation, and state whether the output sensitivity is linear, inverse, nonlinear or conditional. Now specify one nonideal effect, grounded in this lesson, that could cause a real observation to differ. A technically sufficient answer must compare two predicted numerical or symbolic outputs; a verbal statement that the error “increases” is not enough where values are available.

Study exercise 3 — measurement design

Design a low-energy or software-only verification that could distinguish the main mechanism from this competing explanation: A strong periodic 50/60 Hz pickup is not best modeled as independent Gaussian thermal noise. Specify the independent variable, the measured variable, the reference standard or alternate observation, the expected trend and a stop criterion for an invalid run. Describe what metadata would be saved for another engineer to reproduce the analysis without interviewing the experimenter.

Solution and grading guide

Exercise 2. The independent variable is changed while the model assumptions remain fixed. For a linear proportional dependency, +10% in the input produces +10% in the relevant output contribution; for a reciprocal dependence the output becomes 1/1.1=0.9091 of the original; for a squared dependence it becomes 1.1²=1.21. Apply only the form actually supported by For ideal resistor in a bandwidth B, v_noise,rms=√(4 k_B T R B).; do not choose a generic factor without checking the equation. Explain how amplifier current noise through source impedance and flicker noise can exceed this white-noise floor. could invalidate an extrapolation.

Exercise 3. The strongest design isolates one cause using a second evidence pathway. A strong periodic 50/60 Hz pickup is not best modeled as independent Gaussian thermal noise. should lead to a different prediction from the central mechanism. For example, changing source impedance while fixing source voltage is an effective loading test; changing sample rate while holding a physical sine frequency constant reveals aliasing; testing increasing and decreasing excitation reveals hysteresis. The appropriate case must match the instrument and phenomenon studied here.

Simulator and extension

Use [interactive Unit 8 model](../simulations/model-08.html) to explore the unit's central instrumentation mechanism. The simulator may focus on a related quantity rather than exactly this lesson's formula. Save the default model inputs and at least two modified cases; attach a screenshot or numerical result table. Compare one manually calculated point with the model output, and state a physical mechanism that the simulated equations omit. The original example dataset is not field evidence and must not be cited as such.

Safety note: The examples are synthetic and intended for analysis or software-only simulation. Do not connect measuring instruments to mains, high-energy equipment, CT secondaries or exposed conductors without authorized training, appropriately rated equipment and site-specific procedures. Instrument-category markings, isolation ratings and current limits must be verified in the manufacturer documentation.

8.2 — Averaging, autocorrelation and effective sample size

Unit 8: Noise, selective detection and automated measurements Level: undergraduate, with prerequisite circuit theory, calculus and introductory probability. Estimated study: 70–100 minutes, including the numeric investigation. Tools: calculator, spreadsheet or the original offline simulator in this book. Evidence: examples are synthetic unless explicitly identified otherwise.

A problem that a measurement must answer

For independent identically distributed zero-mean noise, averaging N observations reduces standard deviation of their mean by √N. Correlated observations do not offer that reduction; their effective sample size can be much smaller than the count displayed in software. Drift is another form of time-dependent bias.

The application for this unit is: Reproducible software-only studies of detection, averaging, command sequencing and reporting. The meaningful output is not necessarily the number displayed by a meter. First decide whether it is the unperturbed circuit voltage, a time average, a reference-scale resistance, a physical temperature or a confidence-qualified estimate. That decision determines the topology, calibration approach and acceptable model. In an instrumentation report it is better to leave a result undecided than to replace the unmeasured quantity with a convenient proxy without disclosure.

Mathematical model and physical assumptions

Central relation: u(mean)=σ/√N only for independent observations; for lag-one AR(1), N_eff≈N(1−ρ)/(1+ρ) for large N.

Derivation strategy

Begin with the lowest-complexity circuit or signal model that explains the key effect, then state the approximations it makes. For a loaded circuit, preserve current paths and terminal definitions; for a sampled signal, preserve sample timing and analog bandwidth; for a statistical model, preserve independence assumptions. A circuit may obey Kirchhoff laws perfectly while the inferred physical quantity is still wrong because the sensor transfer function is uncharacterized. In this case the key input evidence is the reading or physical parameter introduced above, not the final calculated answer. An experimenter should separate constants taken from a definition, instrument parameters taken from a datasheet and conditions measured on the day of use. If an assumed parameter is poorly characterized, a small model residual does not make the resulting prediction precise. The uncertainty of a quantity calculated from a product, quotient or nonlinear relation must follow its sensitivity to each contributing input, not a simple count of display digits.

Why the relation is useful

Random versus deterministic contamination Broadband Johnson noise, 1/f noise, clock interference and coherent pickup have different signatures and countermeasures. Repeated observations can characterize short-term repeatability, but if noise is correlated the usual √N improvement from averaging is optimistic. Analyze a time series in both time and frequency domains before deciding whether to filter, shield or redesign the circuit. The formula above is one of the operational links between a physical system and a numerical record. It is not an endorsement of any particular manufacturer's ratings, which must be checked independently. For instance, ideal resistance or sampling calculations disregard temperature-dependent and frequency-dependent behavior unless those are included explicitly.

Fully worked numerical case

For N=100 and sample standard deviation 2 mV, iid standard error is 0.2 mV. If lag-one correlation ρ=0.8 follows an approximate AR(1) model, effective N≈11.1 and the estimated standard error becomes about 0.60 mV.

Uncertainty, nonidealities and instrumentation bias

Autocorrelation at many lags and trends can invalidate this simple effective-size approximation. Averaging a drifting reference moves the estimate toward the time-average of the drift, not necessarily toward the desired truth. The first concern affects the credibility of the numerical parameter; the second suggests a rival explanation of the observed behavior. A useful uncertainty analysis therefore goes beyond listing a collection of percentages. It defines the output measurand, classifies major sources as additive or multiplicative, evaluates correlations when necessary, and identifies the sensitivity of the output to each one. An uncertainty budget should preserve documentary support for Type B inputs and raw observations for Type A estimates. If the model is an approximation, give a bounded validity statement in addition to any statistical interval.

A counterexample to overconfidence

Detection with references A lock-in identifies the part of a signal coherent with a known reference; this is not the same as reconstructing every component in a waveform. Its bandwidth is set by filtering and integration time. Automation can consistently repeat commands, but state checks, metadata, calibration and error handling are needed before results are scientifically reproducible. This contextual distinction provides a way to design a discriminating check. For an instrument-chain problem, change input range, source impedance, frequency or wiring to see whether the measurement follows the predicted dependence. For a data-analysis problem, change sample rate, observation length or independent reference while holding the physical input constant. If a conclusion persists only because the same assumption is reused in every run, that persistence supplies little additional evidence.

Engineering use and what would be required to trust it

Plot residuals against time before making a repeatability claim. Calibration records, environmental limits and independent checks belong in the method, not just in appendices. Choose an experiment that isolates the main influence while staying within authorized, properly rated equipment and low-energy pedagogical conditions. A formula computed with fictional values is evidence that arithmetic was performed, not that a real device passed a standard. When the output feeds a safety, certification or financial decision, the appropriate reviewer and agreed acceptance rule must be identified before collecting measurements.

Study exercise 1 — reconstruct the baseline

Derive or justify u(mean)=σ/√N only for independent observations; for lag-one AR(1), N_eff≈N(1−ρ)/(1+ρ) for large N. from the relevant conservation law, circuit relation, transfer model or statistical definition. Recompute the worked example, label every unit and distinguish the physical value from what an actual instrument could have displayed. Identify one data item whose uncertainty is not quantified by the example. Explain what would happen if that datum were systematically too high by a small amount.

Study exercise 2 — sensitivity experiment

Use the same physical configuration but change a controlled parameter by 10% in the direction that would expose a plausible failure mechanism. Compute or predict the response using the stated equation, and state whether the output sensitivity is linear, inverse, nonlinear or conditional. Now specify one nonideal effect, grounded in this lesson, that could cause a real observation to differ. A technically sufficient answer must compare two predicted numerical or symbolic outputs; a verbal statement that the error “increases” is not enough where values are available.

Study exercise 3 — measurement design

Design a low-energy or software-only verification that could distinguish the main mechanism from this competing explanation: Averaging a drifting reference moves the estimate toward the time-average of the drift, not necessarily toward the desired truth. Specify the independent variable, the measured variable, the reference standard or alternate observation, the expected trend and a stop criterion for an invalid run. Describe what metadata would be saved for another engineer to reproduce the analysis without interviewing the experimenter.

Solution and grading guide

Exercise 2. The independent variable is changed while the model assumptions remain fixed. For a linear proportional dependency, +10% in the input produces +10% in the relevant output contribution; for a reciprocal dependence the output becomes 1/1.1=0.9091 of the original; for a squared dependence it becomes 1.1²=1.21. Apply only the form actually supported by u(mean)=σ/√N only for independent observations; for lag-one AR(1), N_eff≈N(1−ρ)/(1+ρ) for large N.; do not choose a generic factor without checking the equation. Explain how autocorrelation at many lags and trends can invalidate this simple effective-size approximation. could invalidate an extrapolation.

Exercise 3. The strongest design isolates one cause using a second evidence pathway. Averaging a drifting reference moves the estimate toward the time-average of the drift, not necessarily toward the desired truth. should lead to a different prediction from the central mechanism. For example, changing source impedance while fixing source voltage is an effective loading test; changing sample rate while holding a physical sine frequency constant reveals aliasing; testing increasing and decreasing excitation reveals hysteresis. The appropriate case must match the instrument and phenomenon studied here.

Simulator and extension

Use [interactive Unit 8 model](../simulations/model-08.html) to explore the unit's central instrumentation mechanism. The simulator may focus on a related quantity rather than exactly this lesson's formula. Save the default model inputs and at least two modified cases; attach a screenshot or numerical result table. Compare one manually calculated point with the model output, and state a physical mechanism that the simulated equations omit. The original example dataset is not field evidence and must not be cited as such.

Safety note: The examples are synthetic and intended for analysis or software-only simulation. Do not connect measuring instruments to mains, high-energy equipment, CT secondaries or exposed conductors without authorized training, appropriately rated equipment and site-specific procedures. Instrument-category markings, isolation ratings and current limits must be verified in the manufacturer documentation.

8.3 — Lock-in detection and synchronous demodulation

Unit 8: Noise, selective detection and automated measurements Level: undergraduate, with prerequisite circuit theory, calculus and introductory probability. Estimated study: 70–100 minutes, including the numeric investigation. Tools: calculator, spreadsheet or the original offline simulator in this book. Evidence: examples are synthetic unless explicitly identified otherwise.

A problem that a measurement must answer

A lock-in multiplies a measured signal by a reference of known frequency and integrates or low-pass filters the product. The in-phase and quadrature terms identify a component coherent with the reference while reducing broadband noise over a chosen bandwidth. A wrong reference frequency or phase can suppress genuine signal.

The application for this unit is: Reproducible software-only studies of detection, averaging, command sequencing and reporting. The meaningful output is not necessarily the number displayed by a meter. First decide whether it is the unperturbed circuit voltage, a time average, a reference-scale resistance, a physical temperature or a confidence-qualified estimate. That decision determines the topology, calibration approach and acceptable model. In an instrumentation report it is better to leave a result undecided than to replace the unmeasured quantity with a convenient proxy without disclosure.

Mathematical model and physical assumptions

Central relation: For x(t)=A sin(ωt+φ), low-pass[2x(t) sinωt]=A cosφ; quadrature gives A sinφ using an appropriately signed reference.

Derivation strategy

Classify each influence by its pathway to the output. Consider bias corrections separately from dispersion. A standard uncertainty is not necessarily equal to a specified maximum tolerance; a rectangular bounded-error model, for example, converts half-width a into u=a/√3 only if that probability model is justified. In a multivariable equation, derivatives and covariance decide how input uncertainty propagates. In this case the key input evidence is the reading or physical parameter introduced above, not the final calculated answer. An experimenter should separate constants taken from a definition, instrument parameters taken from a datasheet and conditions measured on the day of use. If an assumed parameter is poorly characterized, a small model residual does not make the resulting prediction precise. The uncertainty of a quantity calculated from a product, quotient or nonlinear relation must follow its sensitivity to each contributing input, not a simple count of display digits.

Why the relation is useful

Random versus deterministic contamination Broadband Johnson noise, 1/f noise, clock interference and coherent pickup have different signatures and countermeasures. Repeated observations can characterize short-term repeatability, but if noise is correlated the usual √N improvement from averaging is optimistic. Analyze a time series in both time and frequency domains before deciding whether to filter, shield or redesign the circuit. The formula above is one of the operational links between a physical system and a numerical record. It is not an endorsement of any particular manufacturer's ratings, which must be checked independently. For instance, ideal resistance or sampling calculations disregard temperature-dependent and frequency-dependent behavior unless those are included explicitly.

Fully worked numerical case

For A=2.0 mV and relative phase 60°, the in-phase estimate is A cos60°=1.0 mV and the quadrature magnitude is A sin60°=1.732 mV. Combining them gives 2.0 mV, idealized without noise.

Uncertainty, nonidealities and instrumentation bias

The ideal equations presume sufficient averaging to reject the 2ω term and stable phase during measurement. Mains hum at nearby frequency can leak through a lock-in if the record is too short to resolve the separation. The first concern affects the credibility of the numerical parameter; the second suggests a rival explanation of the observed behavior. A useful uncertainty analysis therefore goes beyond listing a collection of percentages. It defines the output measurand, classifies major sources as additive or multiplicative, evaluates correlations when necessary, and identifies the sensitivity of the output to each one. An uncertainty budget should preserve documentary support for Type B inputs and raw observations for Type A estimates. If the model is an approximation, give a bounded validity statement in addition to any statistical interval.

A counterexample to overconfidence

Detection with references A lock-in identifies the part of a signal coherent with a known reference; this is not the same as reconstructing every component in a waveform. Its bandwidth is set by filtering and integration time. Automation can consistently repeat commands, but state checks, metadata, calibration and error handling are needed before results are scientifically reproducible. This contextual distinction provides a way to design a discriminating check. For an instrument-chain problem, change input range, source impedance, frequency or wiring to see whether the measurement follows the predicted dependence. For a data-analysis problem, change sample rate, observation length or independent reference while holding the physical input constant. If a conclusion persists only because the same assumption is reused in every run, that persistence supplies little additional evidence.

Engineering use and what would be required to trust it

Choose integration time from signal dynamics and interference spectrum, then log phase convention. Calibration records, environmental limits and independent checks belong in the method, not just in appendices. Choose an experiment that isolates the main influence while staying within authorized, properly rated equipment and low-energy pedagogical conditions. A formula computed with fictional values is evidence that arithmetic was performed, not that a real device passed a standard. When the output feeds a safety, certification or financial decision, the appropriate reviewer and agreed acceptance rule must be identified before collecting measurements.

Study exercise 1 — reconstruct the baseline

Derive or justify For x(t)=A sin(ωt+φ), low-pass[2x(t) sinωt]=A cosφ; quadrature gives A sinφ using an appropriately signed reference. from the relevant conservation law, circuit relation, transfer model or statistical definition. Recompute the worked example, label every unit and distinguish the physical value from what an actual instrument could have displayed. Identify one data item whose uncertainty is not quantified by the example. Explain what would happen if that datum were systematically too high by a small amount.

Study exercise 2 — sensitivity experiment

Use the same physical configuration but change a controlled parameter by 10% in the direction that would expose a plausible failure mechanism. Compute or predict the response using the stated equation, and state whether the output sensitivity is linear, inverse, nonlinear or conditional. Now specify one nonideal effect, grounded in this lesson, that could cause a real observation to differ. A technically sufficient answer must compare two predicted numerical or symbolic outputs; a verbal statement that the error “increases” is not enough where values are available.

Study exercise 3 — measurement design

Design a low-energy or software-only verification that could distinguish the main mechanism from this competing explanation: Mains hum at nearby frequency can leak through a lock-in if the record is too short to resolve the separation. Specify the independent variable, the measured variable, the reference standard or alternate observation, the expected trend and a stop criterion for an invalid run. Describe what metadata would be saved for another engineer to reproduce the analysis without interviewing the experimenter.

Solution and grading guide

Exercise 2. The independent variable is changed while the model assumptions remain fixed. For a linear proportional dependency, +10% in the input produces +10% in the relevant output contribution; for a reciprocal dependence the output becomes 1/1.1=0.9091 of the original; for a squared dependence it becomes 1.1²=1.21. Apply only the form actually supported by For x(t)=A sin(ωt+φ), low-pass[2x(t) sinωt]=A cosφ; quadrature gives A sinφ using an appropriately signed reference.; do not choose a generic factor without checking the equation. Explain how the ideal equations presume sufficient averaging to reject the 2ω term and stable phase during measurement. could invalidate an extrapolation.

Exercise 3. The strongest design isolates one cause using a second evidence pathway. Mains hum at nearby frequency can leak through a lock-in if the record is too short to resolve the separation. should lead to a different prediction from the central mechanism. For example, changing source impedance while fixing source voltage is an effective loading test; changing sample rate while holding a physical sine frequency constant reveals aliasing; testing increasing and decreasing excitation reveals hysteresis. The appropriate case must match the instrument and phenomenon studied here.

Simulator and extension

Use [interactive Unit 8 model](../simulations/model-08.html) to explore the unit's central instrumentation mechanism. The simulator may focus on a related quantity rather than exactly this lesson's formula. Save the default model inputs and at least two modified cases; attach a screenshot or numerical result table. Compare one manually calculated point with the model output, and state a physical mechanism that the simulated equations omit. The original example dataset is not field evidence and must not be cited as such.

Safety note: The examples are synthetic and intended for analysis or software-only simulation. Do not connect measuring instruments to mains, high-energy equipment, CT secondaries or exposed conductors without authorized training, appropriately rated equipment and site-specific procedures. Instrument-category markings, isolation ratings and current limits must be verified in the manufacturer documentation.

8.4 — Automated acquisition, SCPI and reproducibility

Unit 8: Noise, selective detection and automated measurements Level: undergraduate, with prerequisite circuit theory, calculus and introductory probability. Estimated study: 70–100 minutes, including the numeric investigation. Tools: calculator, spreadsheet or the original offline simulator in this book. Evidence: examples are synthetic unless explicitly identified otherwise.

A problem that a measurement must answer

Instrument automation should control state explicitly, verify responses, catch overloads and timeouts, and preserve raw records. A query after an incomplete configuration sequence may return stale or conditionally formatted data. Standard command patterns help interoperability, but specific syntax varies among instruments.

The application for this unit is: Reproducible software-only studies of detection, averaging, command sequencing and reporting. The meaningful output is not necessarily the number displayed by a meter. First decide whether it is the unperturbed circuit voltage, a time average, a reference-scale resistance, a physical temperature or a confidence-qualified estimate. That decision determines the topology, calibration approach and acceptable model. In an instrumentation report it is better to leave a result undecided than to replace the unmeasured quantity with a convenient proxy without disclosure.

Mathematical model and physical assumptions

Central relation: Result package = raw data + units + settings + calibration metadata + timestamp basis + analysis version.

Derivation strategy

Use both a numerical consistency check and a counterexample. Change one input to challenge sensitivity; change the topology or data-generating assumption to challenge validity. A result is only as robust as the least defensible assumption. When discussing acceptance, state a decision rule, uncertainty and consequences for an incorrect classification. In this case the key input evidence is the reading or physical parameter introduced above, not the final calculated answer. An experimenter should separate constants taken from a definition, instrument parameters taken from a datasheet and conditions measured on the day of use. If an assumed parameter is poorly characterized, a small model residual does not make the resulting prediction precise. The uncertainty of a quantity calculated from a product, quotient or nonlinear relation must follow its sensitivity to each contributing input, not a simple count of display digits.

Why the relation is useful

Random versus deterministic contamination Broadband Johnson noise, 1/f noise, clock interference and coherent pickup have different signatures and countermeasures. Repeated observations can characterize short-term repeatability, but if noise is correlated the usual √N improvement from averaging is optimistic. Analyze a time series in both time and frequency domains before deciding whether to filter, shield or redesign the circuit. The formula above is one of the operational links between a physical system and a numerical record. It is not an endorsement of any particular manufacturer's ratings, which must be checked independently. For instance, ideal resistance or sampling calculations disregard temperature-dependent and frequency-dependent behavior unless those are included explicitly.

Fully worked numerical case

Consider a simulated 10-point calibration sweep. If one point is missing, fitting nine values without preserving the missing-point flag can conceal a communications fault. The correct workflow marks the missing observation and records the device response or timeout.

Uncertainty, nonidealities and instrumentation bias

A passed automated test does not verify electrical safety, calibration validity or software scientific assumptions. A smooth plot is possible even if a script accidentally reuses one cached device response many times. The first concern affects the credibility of the numerical parameter; the second suggests a rival explanation of the observed behavior. A useful uncertainty analysis therefore goes beyond listing a collection of percentages. It defines the output measurand, classifies major sources as additive or multiplicative, evaluates correlations when necessary, and identifies the sensitivity of the output to each one. An uncertainty budget should preserve documentary support for Type B inputs and raw observations for Type A estimates. If the model is an approximation, give a bounded validity statement in addition to any statistical interval.

A counterexample to overconfidence

Detection with references A lock-in identifies the part of a signal coherent with a known reference; this is not the same as reconstructing every component in a waveform. Its bandwidth is set by filtering and integration time. Automation can consistently repeat commands, but state checks, metadata, calibration and error handling are needed before results are scientifically reproducible. This contextual distinction provides a way to design a discriminating check. For an instrument-chain problem, change input range, source impedance, frequency or wiring to see whether the measurement follows the predicted dependence. For a data-analysis problem, change sample rate, observation length or independent reference while holding the physical input constant. If a conclusion persists only because the same assumption is reused in every run, that persistence supplies little additional evidence.

Engineering use and what would be required to trust it

Separate software simulation from control of physical instruments. Use state machines and safe fail-closed behavior in real systems. Calibration records, environmental limits and independent checks belong in the method, not just in appendices. Choose an experiment that isolates the main influence while staying within authorized, properly rated equipment and low-energy pedagogical conditions. A formula computed with fictional values is evidence that arithmetic was performed, not that a real device passed a standard. When the output feeds a safety, certification or financial decision, the appropriate reviewer and agreed acceptance rule must be identified before collecting measurements.

Study exercise 1 — reconstruct the baseline

Derive or justify Result package = raw data + units + settings + calibration metadata + timestamp basis + analysis version. from the relevant conservation law, circuit relation, transfer model or statistical definition. Recompute the worked example, label every unit and distinguish the physical value from what an actual instrument could have displayed. Identify one data item whose uncertainty is not quantified by the example. Explain what would happen if that datum were systematically too high by a small amount.

Study exercise 2 — sensitivity experiment

Use the same physical configuration but change a controlled parameter by 10% in the direction that would expose a plausible failure mechanism. Compute or predict the response using the stated equation, and state whether the output sensitivity is linear, inverse, nonlinear or conditional. Now specify one nonideal effect, grounded in this lesson, that could cause a real observation to differ. A technically sufficient answer must compare two predicted numerical or symbolic outputs; a verbal statement that the error “increases” is not enough where values are available.

Study exercise 3 — measurement design

Design a low-energy or software-only verification that could distinguish the main mechanism from this competing explanation: A smooth plot is possible even if a script accidentally reuses one cached device response many times. Specify the independent variable, the measured variable, the reference standard or alternate observation, the expected trend and a stop criterion for an invalid run. Describe what metadata would be saved for another engineer to reproduce the analysis without interviewing the experimenter.

Solution and grading guide

Exercise 2. The independent variable is changed while the model assumptions remain fixed. For a linear proportional dependency, +10% in the input produces +10% in the relevant output contribution; for a reciprocal dependence the output becomes 1/1.1=0.9091 of the original; for a squared dependence it becomes 1.1²=1.21. Apply only the form actually supported by Result package = raw data + units + settings + calibration metadata + timestamp basis + analysis version.; do not choose a generic factor without checking the equation. Explain how a passed automated test does not verify electrical safety, calibration validity or software scientific assumptions. could invalidate an extrapolation.

Exercise 3. The strongest design isolates one cause using a second evidence pathway. A smooth plot is possible even if a script accidentally reuses one cached device response many times. should lead to a different prediction from the central mechanism. For example, changing source impedance while fixing source voltage is an effective loading test; changing sample rate while holding a physical sine frequency constant reveals aliasing; testing increasing and decreasing excitation reveals hysteresis. The appropriate case must match the instrument and phenomenon studied here.

Simulator and extension

Use [interactive Unit 8 model](../simulations/model-08.html) to explore the unit's central instrumentation mechanism. The simulator may focus on a related quantity rather than exactly this lesson's formula. Save the default model inputs and at least two modified cases; attach a screenshot or numerical result table. Compare one manually calculated point with the model output, and state a physical mechanism that the simulated equations omit. The original example dataset is not field evidence and must not be cited as such.

Safety note: The examples are synthetic and intended for analysis or software-only simulation. Do not connect measuring instruments to mains, high-energy equipment, CT secondaries or exposed conductors without authorized training, appropriately rated equipment and site-specific procedures. Instrument-category markings, isolation ratings and current limits must be verified in the manufacturer documentation.

Project 8 — Noise, selective detection and automated measurements: reproducible measurement review

Research question. Can two plausible measurement chains give meaningfully different estimates of the same physical measurand when one ideal-model assumption is violated?

Starting case. Reproducible software-only studies of detection, averaging, command sequencing and reporting. This is a software-only project. It must not be implemented by handling energized installations or using the simulation as a substitute for calibration.

Background and model selection

Broadband Johnson noise, 1/f noise, clock interference and coherent pickup have different signatures and countermeasures. Repeated observations can characterize short-term repeatability, but if noise is correlated the usual √N improvement from averaging is optimistic. Analyze a time series in both time and frequency domains before deciding whether to filter, shield or redesign the circuit.

A lock-in identifies the part of a signal coherent with a known reference; this is not the same as reconstructing every component in a waveform. Its bandwidth is set by filtering and integration time. Automation can consistently repeat commands, but state checks, metadata, calibration and error handling are needed before results are scientifically reproducible.

Select one formula from each of Lessons 8.1–8.4, but model one of them in depth. For the chosen case, clearly define independent input, measured indication, desired output, units, reference conditions and primary sensitivity. Explain why changing a software parameter is a model intervention, not a laboratory measurement.

Task A — baseline and synthetic evidence

Open [Unit 8 simulator](../simulations/model-08.html). Save the four slider values, the computed numerical output, the axes and the model description. Change one input to three different values and record the response. Independently calculate one point. Create a small table including percent deviations from the baseline and at least one limiting-case check. Where a plotted variable is not the same as an instrument's recorded output, document that distinction.

Task B — deliberately introduce error

Choose one missing mechanism from the lesson limitations. The candidates specific to this unit include: Amplifier current noise through source impedance and flicker noise can exceed this white-noise floor. Another candidate is: Autocorrelation at many lags and trends can invalidate this simple effective-size approximation. Write a small correction term, state its assumed magnitude and investigate how the inferred result changes. Do not assert that an invented error magnitude is representative of a commercial instrument or a deployed installation.

Task C — contrast two explanations

Two technical reports disagree. One report blames calibration; the other blames a physical effect: Mains hum at nearby frequency can leak through a lock-in if the record is too short to resolve the separation. Design a low-energy or observational check that would produce different predictions under those explanations. State the independent evidence source and explain how correlated calibration references could defeat the check. If both explanations predict the same observable under your selected settings, your proposed test is not discriminating and must be redesigned.

Task D — quantitative uncertainty and decision

Assign three clearly labeled synthetic standard uncertainties and propagate them through the selected measurement equation. Where appropriate, include a common-reference correlation. Present a result estimate, combined standard uncertainty and an optional expanded uncertainty with specified k. Introduce a fictional tolerance limit and a predeclared decision rule. Show a near-boundary example where the displayed reading by itself would give a different decision from a guard-banded approach. Do not conflate a rejected acceptance decision with proof of nonconformity.

Deliverables

• Reproducible spreadsheet or notebook with units, formulas and a synthetic-data flag.
• At least one parameter-sensitivity plot and one tabulated calculation.
• A short model-assumption statement and a distinguishable rival hypothesis.
• A traceable uncertainty budget with corrections separated from uncertainty contributions.
• A final one-page technical memo explaining what is supported and what remains unmeasured.

Evaluation and extension

A report is complete only when a reader can reproduce the model calculation, recognize synthetic inputs and identify the decisive missing evidence for a real-world conclusion. Additional complexity earns credit only if it addresses the dominant uncertainty, not simply because the diagram becomes more elaborate. For an extension, estimate how much improved reference accuracy or increased acquisition time would actually reduce the dominant combined uncertainty under the stated model.

Safety note: The examples are synthetic and intended for analysis or software-only simulation. Do not connect measuring instruments to mains, high-energy equipment, CT secondaries or exposed conductors without authorized training, appropriately rated equipment and site-specific procedures. Instrument-category markings, isolation ratings and current limits must be verified in the manufacturer documentation.