Lock-in Amplifier Measurement Setup and Noise Rejection Project

This project creates a defensible lock-in amplifier measurement setup for a weak periodic signal. The deliverable is not only a recovered number. It is a measurement package that proves the signal is coherent with the intended physical reference, the analog front end is not saturated, the bandwidth is appropriate, coherent artifacts are controlled, and the reported result has a documented uncertainty.

The scenario uses a chopped optical measurement, but the same workflow applies to bridge excitation, impedance measurements, resonator readout, magnetic susceptibility, vibration response, biomedical sensors, and low-level electronic measurements that use synchronous detection.

Project Objective

Build a lock-in measurement setup and release dossier that answers:

1. What physical quantity is modulated or naturally periodic?
2. What reference frequency and phase convention define the measurement?
3. Does the sensor and analog front end remain linear before demodulation?
4. What time constant, filter setting, and equivalent noise bandwidth are used?
5. How much SNR improvement is expected from bandwidth reduction?
6. Do in-phase and quadrature data agree with the expected physical signal?
7. Which reference-off, source-off, phase-sweep, frequency-sweep, and time-constant-sweep checks reject coherent artifacts?
8. What uncertainty and operating restrictions accompany the released measurement?

Baseline Scenario

A laboratory test fixture measures a weak chopped optical signal from a sample. A photodiode and transimpedance amplifier feed a digital lock-in amplifier. The light source is modulated by a chopper reference.

Quantity	Value
reference frequency	f_0=1.20\ \text{kHz}
measured signal peak amplitude at front-end output	A=18.0\ \mu\text{V}
signal phase relative to reference	\phi=35^\circ
broadband noise density near measurement band	e_n=1.2\ \mu\text{V}/\sqrt{\text{Hz}}
wideband observation bandwidth	B_{wide}=1000\ \text{Hz}
selected lock-in time constant	\tau=2.0\ \text{s}
first-order equivalent noise bandwidth screen	B_{ENBW}\approx 1/(4\tau)
measurement dwell per point	T_{dwell}=12\ \text{s}
ADC sample rate for digital demodulation	f_s=48\ \text{kS/s}
analog front-end valid output range	-1.0\ \text{V} to +1.0\ \text{V}
measured DC offset at front-end output	0.42\ \text{V}
largest observed noncoherent pickup	30\ \text{mV peak}
residual coherent output with optical path blocked	0.55\ \mu\text{V} at magnitude channel

The target is to release the setup if the recovered amplitude uncertainty is below 10 percent expanded uncertainty with coverage factor k=2, and if artifact checks show the coherent residual is less than 10 percent of the measured signal magnitude.

Measurement Boundary

The measurement boundary includes:

• the modulated source, chopper or electrical excitation, and reference pickup point;
• optical path, sample, detector aperture, and detector bias condition;
• analog front end, shielding, grounding, cables, and anti-alias filtering;
• digitizer, demodulation convention, low-pass filter, time constant, and averaging;
• calibration factor from lock-in output to reported physical quantity;
• validation tests that show the recovered coherent signal is physical.

Anything outside this boundary must be declared as an assumption. Examples include sample drift, optical alignment stability, detector temperature, source intensity drift, and chopper blade phase jitter.

Worked Check 1: Wideband SNR Before Lock-In Detection

The RMS noise over a wide observation bandwidth is approximated as:

Substitute the data:

A simple amplitude-to-noise screen is:

The signal is smaller than the broadband noise displayed over the wide measurement bandwidth.

Engineering comment: this does not mean the signal is absent. It means the measurement must avoid judging a coherent periodic signal using a bandwidth that includes much more uncorrelated noise than the final result will use.

Worked Check 2: Lock-In Components and Recovered Amplitude

For the peak-amplitude convention used here:

With A=18.0\ \mu\text{V} and \phi=35^\circ:

The lock-in magnitude channel is:

The recovered peak amplitude is:

Engineering comment: using both X and Y avoids under-reading when the phase is not exactly zero. The report must still state the sign convention, reference phase, scaling convention, and whether the instrument reports peak, RMS, or magnitude.

Worked Check 3: Equivalent Noise Bandwidth and Demodulated SNR

For a first-order low-pass screen:

With \tau=2.0\ \text{s}:

The demodulated RMS noise scale is:

Using the magnitude channel R=9.00\ \mu\text{V}:

The SNR improvement compared with the wideband screen is large because the effective bandwidth was narrowed from 1000\ \text{Hz} to about 0.125\ \text{Hz}.

Engineering comment: this is not a universal guarantee. The equivalent noise bandwidth depends on filter order and implementation. The setup sheet must record the instrument setting, filter roll-off, digital averaging rule, and settling policy.

Worked Check 4: Settling Time and Dwell

A first-order output is commonly treated as settled for many engineering screens after about five time constants:

The measurement dwell is:

Since 12\ \text{s}>10\ \text{s}, the dwell is acceptable for this first-order screen.

Engineering comment: a shorter dwell would make the display look stable before it is physically settled. If the filter order is higher, if overload recovery occurred, or if the source was moved immediately before acquisition, the required settling rule may be longer.

Worked Check 5: Sampling and Aliasing Margin

The digital demodulator samples at:

The reference frequency is:

Samples per reference period are:

This gives adequate timing resolution for a basic digital lock-in implementation.

Engineering comment: this does not replace an anti-alias check. The analog front end and digitizer must still reject high-frequency content that could fold into the demodulated band. A clean FFT before demodulation is useful evidence when switching supplies, motors, LEDs, choppers, or RF sources are nearby.

Worked Check 6: Front-End Headroom Before Demodulation

The front-end output includes a DC offset:

and a largest observed noncoherent pickup:

The approximate positive peak is:

This is inside the valid range of -1.0\ \text{V} to +1.0\ \text{V}.

Engineering comment: the weak coherent signal cannot be recovered if the analog input clips before demodulation. Background subtraction and lock-in processing are not cures for saturation. A front-end headroom screenshot, overload flag record, or oscilloscope capture should be part of the release evidence.

Worked Check 7: Coherent Artifact Residual

With the optical path blocked and the reference electronics still active, the residual lock-in magnitude is:

The measured signal magnitude is:

Residual fraction:

The residual is below the 10 percent screen, so the setup passes this artifact check.

Engineering comment: this is one of the most important tests. A reference-synchronous electrical, optical, thermal, or mechanical leak can create a convincing lock-in output even when the physical signal path is blocked. The artifact test must leave reference electronics running while disabling only the intended physical signal.

Worked Check 8: Uncertainty Budget for Released Amplitude

The released amplitude is:

The random standard uncertainty of one amplitude estimate is approximated from the magnitude-channel noise:

Relative to amplitude:

The project averages four independent settled readings, so the random contribution is reduced by:

Use the following standard relative uncertainty contributions:

Source	Standard relative uncertainty
calibration source amplitude	2.0 percent
lock-in scale factor and gain	1.0 percent
repeatability after four settled readings	2.36 percent
residual coherent artifact correction	0.7 percent
phase and magnitude convention	0.5 percent
temperature and alignment drift during run	1.0 percent

Combined standard relative uncertainty:

Expanded uncertainty with k=2:

The target is 10 percent expanded uncertainty, so the measurement passes the uncertainty screen.

Engineering comment: the uncertainty is valid only for the stated setup, dwell, bandwidth, source level, reference configuration, and environment. If the time constant changes, if the source amplitude changes, or if artifact residual increases, the uncertainty budget must be updated.

Validation Matrix

Check	Acceptance evidence	Result
reference frequency	1.20\ \text{kHz} selected away from dominant interference and within sensor bandwidth	pass
phase convention	X, Y, magnitude, scaling, and sign convention recorded	pass when documented
front-end headroom	output remains between -1.0\ \text{V} and +1.0\ \text{V} with source on and blocked	pass
equivalent noise bandwidth	\tau=2.0\ \text{s} gives B_{ENBW}\approx0.125\ \text{Hz} screen	pass
settling	12\ \text{s} dwell exceeds 5\tau=10\ \text{s}	pass
artifact residual	blocked-path coherent residual is 6.1 percent of signal magnitude	pass
phase sweep	X and Y rotate while magnitude remains stable within stated tolerance	required evidence
time-constant sweep	amplitude remains consistent while noise decreases with narrower bandwidth	required evidence
frequency sweep	recovered response follows expected sample and front-end frequency response	required evidence
uncertainty	expanded uncertainty is 6.70 percent against 10 percent target	pass

Required Plots and Records

The release package should include:

• wiring and optical layout with reference pickup point marked;
• raw time record or oscilloscope capture showing no front-end clipping;
• FFT or spectral record before demodulation;
• X, Y, R, and phase trend during settling;
• time-constant sweep showing amplitude consistency and noise reduction;
• phase sweep showing rotation of X and Y without magnitude collapse;
• blocked-path, source-off, and reference-off records;
• uncertainty budget and calibration traceability;
• release limits for reference frequency, time constant, dwell, source level, detector bias, cable routing, shielding, and environmental conditions.

Common Failure Modes

• the input clips before demodulation, so the lock-in output is precise-looking but invalid;
• the reference leaks electrically into the sensor cable;
• the chopper or source modulation heats the sample and creates a delayed thermal artifact;
• the time constant is long enough for low noise but too slow for the measurand being tracked;
• the setup reports only X, causing phase error to look like lower amplitude;
• a mains harmonic or switching supply lands near the reference frequency;
• a digital demodulator aliases high-frequency interference into the low-pass output;
• the blocked-path test disables the reference as well as the physical signal, hiding pickup;
• uncertainty is copied from instrument specifications without including setup repeatability and coherent residuals.

Release Decision

The setup can be released for the stated weak-signal measurement if the front end remains linear, the reference frequency is documented, the time constant and dwell satisfy the settling rule, the blocked-path coherent residual remains below 10 percent of signal magnitude, and the expanded uncertainty remains below 10 percent for the intended amplitude range.

The release is conditional. It does not cover different reference frequencies, shorter dwell, stronger background, different optical alignment, changed cable routing, or measurements where the sample response has a time dependence comparable to the lock-in filter time constant. A lock-in amplifier is powerful because it asks a narrow correlation question. It is trustworthy only when the engineer proves that the correlation belongs to the intended physical signal.