20-Bio-B6 Analytical Biochemistry · December 2019
Nivaar worked solution (AI-drafted; not reviewed by a licensed engineer)
Paper format: National Exams, December 2019 — 04-Bio-B6 Bioinstrumentation. Three hours, open book, non-communicating calculator permitted. Six questions of equal value (25 marks each); four constitute a complete paper and only the first four appearing in the answer book are marked. All six are solved here as a complete study resource. Every question is a design/essay question (block-diagram instrumentation-system design, or descriptive explanation).
Q3(iv)'s marks belong to Q4(i)'s 12-mark opening sub-part, not to Q3; Q5(ii) and (iii) each carry their own 5 marks rather than a combined total; Q6(ii) covers the instrumentation for the whole ICU bedside monitor, not the pulse oximeter alone.
Reference texts (the books a candidate should have reviewed for this subject):
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.
Approach. Thirty-two independently-amplified scalp channels (extended 10-20 montage) plus one core-temperature channel feed a common epoch-based windowing/spectral-analysis stage; the resulting band-power and temperature time series are combined by a correlation engine, while noise rejection is built into each analog front end and into the epoch-acceptance logic before any value reaches that engine.
Each of the 32 scalp channels is bandpass-filtered (0.5-32 Hz to match the stated EEG range) and digitised at 128 Hz per channel — comfortably above the 64 Hz Nyquist rate for a 32 Hz signal, leaving margin for a realisable anti-alias roll-off. For each successive 30-second epoch, a power spectral density estimate (Welch's averaged periodogram) is computed per channel, and the theta (4-7 Hz) and delta (1-3.5 Hz) band powers are summed and expressed as a percentage of total 0.5-32 Hz power. The core-temperature channel is sampled at 1 Hz (its thermal time constant is minutes) and averaged over the same 30-second epochs, so both streams are time-aligned sample-for-sample; a 30-s window contains at least 30 cycles of the slowest 1 Hz delta component, giving a statistically stable spectral estimate every epoch. At the end of each epoch (or continuously, as a running statistic) the system computes the Pearson correlation coefficient $r=\dfrac{\sum(x_i-\bar x)(y_i-\bar y)}{\sqrt{\sum(x_i-\bar x)^2\sum(y_i-\bar y)^2}}$ between the %power-in-band series and the temperature series. All raw per-channel epoch values (theta %, delta %, temperature, timestamp, artifact flags) are logged to onboard flash and/or streamed to a host database for the full 8 hours (960 epochs); a bedside display shows scrolling multichannel trend plots plus the running correlation coefficient, and an optional topographic (per-electrode) map of band power across the scalp.
A wired system carries all 32 channels down shielded, twisted-pair leads to a bedside amplifier chassis; this gives the lowest noise floor and the most straightforward electrical-isolation implementation (a single isolated front end, per Question 5), but the resulting cable bundle is heavy, restricts the patient's movement during sleep, and is itself a safety/comfort hazard (tangling, tension on the electrodes, entanglement/strangulation risk overnight). A wireless system instead digitises all 32 channels in a small, battery-powered head-worn unit and transmits the data over a short-range, low-power radio link (e.g. Bluetooth Low Energy or a dedicated medical body-area-network band) to a bedside receiver; because the patient-worn electronics are battery-powered and have no galvanic (conductive) connection at all to mains-referenced equipment, the wireless design achieves a stronger and simpler safety margin than even an isolated wired front end, and it removes the tangling/movement-restriction hazard entirely, improving comfort for an overnight sleep study. The trade-offs are: RF link reliability (packet loss requires local buffering and re-transmission or interpolation so no epoch is silently lost), battery safety (thermal/chemical hazard of the head-worn cell, and a defined low-battery/fail-safe behaviour), and coexistence with other hospital wireless equipment (channel/band selection to avoid interference).
EMG from scalp/facial muscles is broadband and extends well above 32 Hz, so the 0.5-32 Hz analog bandpass on every channel already removes most of it; residual in-band EMG is further suppressed by monitoring each epoch's RMS amplitude per channel and flagging/discarding any epoch whose amplitude exceeds a threshold (e.g. >100 µV, well above the stated 30 µV EEG level) as artifact-corrupted before it reaches the correlation calculation. Instrumentation artifact (electrode-lead movement, connector noise, amplifier saturation) produces characteristic large, often step-like or saturating excursions distinguishable from genuine EEG by exceeding the amplifier's linear range or by an abrupt DC-level shift; the system flags any epoch containing a saturated sample or a baseline step larger than a set threshold. A small accelerometer on the headband/cap provides an independent movement channel, so any epoch coincident with a physical-movement event is flagged and excluded regardless of its spectral content, rather than relying on amplitude thresholds alone. Patient safety and comfort are addressed with lightweight, flexible gel electrodes and (per (ii)) either a battery-powered isolated wired front end or, preferably for an overnight study, the wireless head-worn design.
Classifying the EEG in the frequency domain requires, first, that each digitised epoch be windowed (a Hann or similar taper) before transforming, to control spectral leakage from the epoch boundaries; a Welch-averaged periodogram (splitting the epoch into overlapping sub-segments, transforming each, and averaging the resulting spectra) gives a lower-variance power spectral density estimate than a single raw FFT. The resulting PSD is then integrated over each band of interest — not only theta and delta but typically also alpha (8-13 Hz) and beta (14-32 Hz) — to build a feature vector of %power-per-band for the epoch; with a 30-s window the frequency resolution is $\Delta f=1/30\approx0.033$ Hz, far finer than the narrowest band of interest (delta, 2.5 Hz wide), so band-power integration is not resolution-limited. This feature vector is then passed to a classifier — in the simplest form a set of thresholds/discriminant rules on relative band power (e.g. delta-dominant epochs classified as deep sleep, theta-dominant as light sleep/REM-adjacent, beta-dominant as wake), or in a more capable design a trained statistical classifier (linear discriminant analysis or a small neural network) using labelled polysomnography data — producing a sleep-stage label per epoch that both drives the trend display in (i) and forms the %power series correlated against core temperature.
