33 Proxy Sensing and Inductive Loops
33.1 Start With the Measurement Story
Proxy sensing turns indirect evidence into a measurement claim. Before using an inductive loop, Wi-Fi pattern, or infrastructure signal, ask what it really observes, what it misses, and how ground truth will check it.
Phoebe’s Why
The chapter below tracks the loop detector through its frequency shift, \(\Delta f/f\approx-\tfrac12\Delta L/L\), and that is the right first story: less inductance, faster oscillation. But eddy currents do not just change \(L\) for free – induced currents in a lossy conductor dissipate real power, obeying nothing more exotic than Ohm’s law, \(P=I^2R\). That dissipation shows up in the tank as an effective series loss resistance, and the loss resistance sets a second, independent property of the oscillator: how sharply it rings, measured by the quality factor \(Q=\omega L/R\). A high-\(Q\) tank holds its energy for many cycles before it decays; a lossier tank – one loaded by more eddy current, like a large steel vehicle body – rings down faster. This is not the same signal as the frequency shift the chapter already derives; it is a second, amplitude-domain channel the same eddy currents create, and it decays on its own characteristic time constant, the RL analogue of an RC discharge: \(\tau=L/R=Q/\omega\).
The Derivation
Power dissipated in the loss resistance (Ohm’s law form):
\[P = I^2 R\]
Quality factor of the tank, and the loss resistance it implies:
\[Q = \frac{\omega L}{R} \;\Rightarrow\; R = \frac{\omega L}{Q}\]
Energy-decay time constant, the RL analogue of an RC discharge:
\[\tau = \frac{L}{R} = \frac{Q}{\omega} = \frac{Q}{2\pi f}\]
Tank capacitance implied by the oscillation frequency:
\[C = \frac{1}{(2\pi f)^2 L}\]
Worked Numbers: This Chapter’s 40 kHz Loop
- Catalog-typical values for a loop tuned to this chapter’s own \(f_0=40.000\) kHz baseline: \(L=100\ \mu\text{H}\) (typical multi-turn pavement loop). Solving for the tank capacitor: \(C=1/[(2\pi\times40{,}000)^2\times100\mu\text{H}]=158\) nF.
- With a catalog-typical detector \(Q=10\): \(R=\omega L/Q=(2\pi\times40{,}000\times100\mu\text{H})/10=2.51\ \Omega\) – a small but nonzero loss resistance, exactly what the alternating field is dissipating into eddy currents in the pavement, coil wire, and (when present) the vehicle body.
- At a catalog-typical drive current of 10 mA rms, the tank dissipates \(P=I^2R=(0.01)^2\times2.51=251\ \mu\text{W}\) into that resistance – the electrical cost of the “hum” this chapter’s intuition box describes.
- Ring-down time constant: \(\tau=Q/(2\pi f_0)=10/(2\pi\times40{,}000)=39.8\ \mu\text{s}\), about \(1.59\) oscillation cycles. That sets how quickly the detector’s amplitude measurement settles after a vehicle changes the loading – a genuinely separate design constraint from the frequency-counting approach this chapter’s own worked example covers, and the reason some commercial loop detectors also watch signal amplitude, not only frequency, when classifying a borderline target like a motorcycle.
33.2 Learning Objectives
After this page, you should be able to:
- Explain how existing infrastructure signals become proxy sensing evidence rather than direct measurements.
- Describe how an inductive-loop detector turns conductive vehicle presence into an LC oscillator frequency shift.
- Set detection thresholds against drift, small-target sensitivity, adjacent-lane cross-talk, and false calls.
- Identify proxy drift and record the evidence needed before infrastructure sensing can support decisions.
33.3 After Infrastructure Sensing
Sensing with Existing Infrastructure introduces Wi-Fi, NILM, cellular, and campus-occupancy examples that reuse operational signals. This page focuses on the deeper boundary question: what does the proxy physically measure, what does it miss, and how fresh is the baseline that makes the inference credible?
Use it when an infrastructure-derived signal looks attractive because it is cheap and already deployed, but the decision still needs a defensible confidence label and a known operating envelope.
33.4 Overview: The Signal Is Already There
Infrastructure sensing starts from a practical question: what useful signal is the site already producing? A Wi-Fi access point already broadcasts and records signal strength, a smart meter already samples voltage and current, a cellular network already tracks handoff and load, and a traffic lane may already contain a buried loop. The sensing task is to turn those operational signals into a proxy for occupancy, location, traffic, appliance use, or equipment condition.
One of the most widely deployed infrastructure sensors is a loop of wire cut into the pavement at traffic lights. The inductive loop detector does not photograph or weigh a vehicle, and it needs no line of sight. It senses how a large mass of metal changes the magnetic behaviour of the buried coil. The same design pattern appears in Wi-Fi fingerprinting: the access point was installed for networking, but the radio signal map can also become a location or presence sensor.
The mechanism differs by infrastructure, but the evidence pattern is the same. First, identify the physical quantity that perturbs an existing signal. Second, record a baseline or fingerprint while the system is in a known state. Third, look for deviations large enough to support the decision you want to make. The result is usually a useful proxy, not a certified ground truth.
For the traffic loop, the mechanism is inductive transduction. The loop is a coil, and a coil has inductance. When a car — a large conductive body — sits over the loop, it changes the loop's effective inductance. Turn that inductance change into a measurable signal and you have a vehicle detector that can live under the road surface for years.
Intuition: the loop hums with an alternating magnetic field. A metal vehicle overhead soaks up and reshapes that field, and the loop "feels" the difference as a change in its own inductance — presence without ever seeing the car.
Overview Knowledge Check
33.5 Inductance to Frequency
The loop is wired as the inductor of an LC oscillator, so its inductance sets the oscillation frequency:
Oscillation frequency: f = 1 / (2π × sqrt(L × C)) Because f depends on 1/sqrt(L): Δf/f ≈ -(1/2) × ΔL/L
Worked example: a vehicle nudges the frequency
A vehicle's eddy currents lower the loop's effective inductance by about 0.1%: ΔL/L = -0.001 Since f rises when L falls: Δf/f ≈ -(1/2)(ΔL/L) = -(1/2)(-0.001) = +0.0005 = +0.05% So a car present RAISES the oscillator frequency by about 0.05%. Loop detectors are built to resolve these small, sub-0.1% inductance shifts reliably.
Reading a frequency shift is far easier and more robust than reading a tiny inductance directly, which is why the oscillator approach is standard: the vehicle becomes a small, clean change in a frequency the electronics can count precisely.
The important engineering move is the same for other infrastructure signals: convert a weak physical disturbance into a robust digital feature. In a loop detector that feature is frequency shift. In Wi-Fi sensing it might be RSSI from several access points, CSI subcarrier amplitude and phase, or associated-device count per access point. In a smart meter it might be a step in real power, a change in reactive power, or a harmonic signature.
A deployment should define the feature, the baseline, and the decision threshold before it is trusted. Suppose an empty lane oscillator sits at 40.000 kHz and normal temperature drift moves it by +/-4 Hz over an hour. A vehicle that shifts the frequency by +20 Hz is easy to separate from drift; a small motorcycle that shifts it by +3 Hz is not. Raising sensitivity catches the motorcycle but increases splash-over from adjacent lanes. The practitioner tradeoff is not "detect or not"; it is missed detections versus false calls.
The same threshold logic applies to Wi-Fi occupancy. If a conference-room access point reports 2-4 associated devices overnight and 18-24 during a scheduled meeting, the occupancy proxy is strong. If the count changes from 8 to 10, the evidence is weak unless it is combined with room booking data, CSI motion features, or a small number of dedicated PIR sensors. Infrastructure sensing works best when the decision only needs zone-level confidence.
Practitioner Knowledge Check
33.6 Eddy Currents and Baseline Drift
The physics that makes the loop work also defines exactly what it cannot see and why it must keep re-learning the empty road.
Eddy currents lower inductance
The loop's alternating field induces circulating eddy currents in the vehicle's conductive body; those currents create an opposing field that reduces the loop's effective inductance — the same effect an industrial inductive proximity sensor uses to detect metal.
It only sees conductive mass
Because detection needs eddy currents, the loop responds strongly to cars and trucks but weakly to non-conductive objects. A carbon-frame bicycle or a pedestrian may not trigger it at all — a well-known real-world blind spot.
Geometry and cross-talk
Detection depends on how much conductive area couples to the loop and how high it sits, so motorcycles are hard. Turn sensitivity up too far and adjacent loops interfere (splash-over); too low and small vehicles are missed.
Baseline drift and auto-tuning
Temperature, moisture, and cable changes slowly shift the empty-loop inductance. Detectors continuously re-baseline to the empty road and look for fast deviations, separating a real vehicle from slow environmental drift.
This is the general shape of infrastructure-based sensing: it is cheap, passive, and robust because it reuses something already there, but it inherits blind spots from its physics and needs constant baseline tracking to stay honest. Knowing that an inductive loop is deaf to carbon and plastic explains a great deal of real traffic-signal behaviour — including why some cyclists wait forever at a red light.
The hidden risk is proxy drift. A Wi-Fi fingerprint can degrade when access points move, furniture changes, or a new wall attenuates one path. A smart-meter NILM model can confuse appliances when two loads switch at the same time or when a variable-speed motor changes its signature. A cellular density estimate can change when carrier policy or handset behaviour changes. None of those failures look like a broken sensor; they look like a plausible but stale inference.
Good infrastructure-sensing systems therefore carry evidence metadata with every decision: which infrastructure signal was used, when the baseline was last refreshed, what confidence or residual check passed, and which cases fall outside the operating envelope. For safety-critical decisions, the proxy should trigger a follow-up sensor or human check rather than stand alone.
Under-the-Hood Knowledge Check
33.7 Release Checklist
Before relying on an infrastructure-sensing proxy, confirm these points:
- The physical feature is named explicitly: frequency shift, RSSI/CSI pattern, device count, power step, harmonic signature, or another measurable proxy.
- The baseline was collected under known empty, normal, and edge-case conditions and has a refresh schedule.
- Thresholds are tested against drift, adjacent-zone cross-talk, small targets, and ordinary environmental changes.
- Blind spots are documented, such as non-conductive vehicles for loops or access-point layout changes for Wi-Fi fingerprints.
- Each decision records source signal, baseline timestamp, confidence/residual check, and cases outside the operating envelope.
- Safety-critical actions trigger a dedicated sensor, human check, or other independent confirmation.
33.8 See Also
- Sensing with Existing Infrastructure for the main Wi-Fi, NILM, cellular, and campus occupancy workflow.
- Sensor Calibration Techniques for baselines, reference records, and drift control.
- Sensor Data Processing for filtering and confidence checks after acquisition.
- Sensor Selection Guide for deciding when infrastructure reuse is enough and when dedicated hardware is required.
33.9 Next
Return to Sensing with Existing Infrastructure once the proxy boundary is clear, then continue to Sensor Selection Guide to compare reuse against dedicated sensor deployments.