Question 1 of 7: Instrumentation Fundamentals — Sensitivity, Noise, and Calibration

Question 1: Instrumentation Fundamentals — Sensitivity, Noise, and Calibration (20 marks)

Question text not reproduced: the examination questions are © Engineers and Geoscientists BC. Open the official past paper (linked at the top of this page) to read the question, then follow the worked solution below.

a) Sensitivity is the ratio of the change in an instrument's output to the change in the input (measurand) that produced it — the slope of the calibration curve, $S=\Delta(\text{output})/\Delta(\text{input})$. A highly sensitive instrument produces a large output change for a small input change.

b) Raising an instrument's gain to respond more strongly to the desired input almost always raises its response to other, unwanted inputs of a similar nature just as strongly — sensitivity is a statement about response magnitude, while selectivity is a statement about discriminating the desired quantity from similar interferents. Because the same high-gain front end amplifies both together, chasing more sensitivity typically broadens what the instrument responds to and erodes selectivity, and vice versa; the two must usually be traded off against each other rather than both maximized at once.

c) Noise is the random, unpredictable fluctuation superimposed on a signal (e.g. thermal/Johnson noise, shot noise) — statistical in nature, it cannot be predicted from its own past values and is only ever described probabilistically (RMS value, standard deviation, power spectral density). Electrical interference is a deterministic, externally coupled signal from an identifiable source (mains hum, a motor, an RF transmitter) — often periodic or otherwise structured, and therefore, unlike true noise, removable by targeted filtering, shielding, or subtraction once its source and character are known.

d) Measurement quality is limited by how well the genuine signal can be distinguished from whatever noise rides along with it, not by the signal's raw size: a large signal buried in even larger noise is unusable, while a small signal riding on very low noise can be resolved precisely. Amplifying the raw signal level alone does not help if the noise is amplified by the same factor, so it is the ratio of the two — SNR — that actually governs how accurately the true value can be extracted.

e) Accuracy describes systematic bias — how far the AVERAGE of many readings sits from the true value — while noise is, by definition, a random, zero-mean fluctuation superimposed on each individual reading. Noise adds scatter (degrades precision/repeatability) around the true mean, but because its average contribution over many readings tends to zero, it does not shift where that mean sits; averaging enough repeated readings removes noise's effect on the reported value. Accuracy is instead compromised by systematic errors — offset, calibration drift, non-linearity — which noise, being random, does not introduce.

f) A single calibration reading at each point is itself corrupted by random noise, so the fitted calibration curve from one pass could be biased by whatever noise happened to occur at each point. Repeating each calibration point and averaging reveals the true underlying input–output relationship with the noise averaged out, and the spread between repeats at a given point directly quantifies the instrument's own repeatability/precision — information a single reading can never provide — while also giving an early warning of any drift between repeats.

g) Apply the input over the full calibration range twice: once in an ascending (increasing) sequence and once in a descending (decreasing) sequence, and compare the output recorded at the SAME input value on the way up versus on the way down. If the ascending and descending calibration curves do not coincide — if there is a gap or loop between them — the instrument shows hysteresis: its output depends on the direction of approach (the input's recent history), not on the input value alone.

h) A one-point calibration is valid only when the instrument's zero (offset) is already known and stable and its response is already established (from prior, more complete calibration) to be linear with a fixed, non-drifting slope — the single point then serves only to re-confirm or re-set that already-known span, not to characterize an unknown curve. It is appropriate for routine re-verification of a mature, well-characterized instrument, not for a new instrument or one suspected of non-linearity or zero drift.

i) RMS (root-mean-square) error squares every residual before averaging and then takes the square root, so it always weights larger errors more heavily than an average of raw or absolute residuals would, and it is always positive — unlike a simple signed average, where a large positive error and a large negative error can cancel and misleadingly suggest a small overall error. RMS therefore gives a single, statistically meaningful measure of the typical SIZE of the measurement error that cannot be hidden by error cancellation.

j) The lowest detectable limit (LDL) is set by comparison against the instrument's own noise/blank fluctuation, not by its nominal resolution: it is conventionally the smallest input whose response exceeds the standard deviation of the noise/blank measurement by a fixed statistical margin (commonly $3\sigma_{blank}$). Below that input level, a genuine response cannot be reliably distinguished from random baseline fluctuation.