← Sensing 2026 hub
Anchor 1 · core · the trade-off triangle

You can't have it all

Sensitivity, dynamic range and bandwidth pull against each other. A sensor is a choice of corner, not a wish-list — and in this toy model the three weights sum to one fixed design budget. (Real hardware moves the whole triangle, not just the point inside it — see Explain.)

The instrument — drag the budget P around the triangle
Jump to:
target: high bandwidth P
Drag P (mouse or touch). Corner = pure focus; centre = an even split. Coloured dots = the seven magnetometers in the table below (colour = best corner); grey dots = other sensor types, for scale.
Sensitivity 33%
Dynamic range 33%
Bandwidth 34%
Σ = 100% — a fixed toy budget. You re-allocate; only the hardware sets the total.

Try this Drag the budget around the triangle

The question: why can't one detector be exquisitely sensitive, cover a huge dynamic range, and respond at high bandwidth all at once? Drag the handle P above and watch the three weights — they always add to 100%. Push toward one corner and the other two collapse: that is the proof you can't max all three.

Got a feel for it? Step to 2 · Notice

Notice this The three always sum to 100%

The thing to notice: every weight you pour into one corner is drained straight out of the other two. The full seven-magnetometer comparison — the evidence for the exercise — is in the table below the triangle.

Explain this A fixed budget you can only re-allocate

Wherever P sits, the three weights sum to one. Slide P toward the Bandwidth corner and watch Sensitivity and Dynamic range drain away in lock-step. Why is the total conserved — why can't a clever design just buy back the corners it spent?

Reveal a one-paragraph answer

The three corners draw on the same physical pool: a finite count of quanta (photons, atoms, spins) read out in a finite time. Spend that pool on resolving a tiny change and you have used up the headroom for a large one (range) and the speed of repeated reads (bandwidth). You can move the allocation — that is engineering — but you cannot create more total without changing the hardware (more atoms, longer integration, lower temperature), which moves the whole triangle, not P inside it. Real sensors are points pinned to a corner because their physics already chose the split for them.

Show the math — barycentric coordinates & the conserved budget

Put the three corners at vertices vS, vD, vB. Any point P inside the triangle has a unique barycentric coordinate P = wSvS + wDvD + wBvB with wS + wD + wB = 1 and every w ≥ 0. Each weight is the relative area of the sub-triangle opposite that corner (divide P-to-edge distance by the triangle's height). At a vertex one weight is 1 and the others are 0 — the visual proof you cannot max all three. The constraint Σw = 1 is the conserved quantity: the readout's percentages are just 100·w.

Same shape elsewhere on the hub:

  • Squeezing (anchor 5): the quantum uncertainty area Δx·Δp is conserved — squeeze one quadrature and the conjugate one must widen. Same conservation law, a different pair of corners.
  • The Allan / GHZ "bathtubs" (anchors 2 & 4): averaging buys sensitivity as 1/√N only until drift takes over; entanglement buys 1/N only until decoherence does — a resource that stops paying past a point. Here, weight you pour into one corner stops paying because it is drained straight out of the others.

Now connect it Pick a corner, defend it

You need to detect a 10 nT magnetic field at 1 kHz. Drag P into the red target zone near the bandwidth corner — see how much sensitivity and range you spend to get there. Then read the seven-magnetometer table below the triangle, pick two candidates, and decide between them.

Fill the judgement (Claim · Evidence · Assumption · Limitation · Decision):

Claim(which technology, in one line)
Evidencewhere each sits on sensitivity / range / bandwidth for the 10 nT @ 1 kHz target
Assumptionwhat about the task you're taking for granted (field strength stable? single-shot? environment?)
Limitationwhich corner your choice gives up — and whether that matters here
Decisionyour pick, and what would change it

What good looks like:

  • Names the 1 kHz bandwidth as the binding constraint — not sensitivity (10 nT is a large field, so every row except the Hall probe has ample sensitivity). The pick is decided by which rows actually clear 1 kHz and at what practical cost (room-T vs cryogenics, size, integration) — which is why SERF/OPM drops out and NV-diamond stands out.
  • Uses the triangle as evidence — not "sensor X is better" but "X trades range for the bandwidth I need".
  • States one thing that would flip the decision (e.g. if the signal were DC, the slow sensitive option wins).

Logbook (one entry): Tried · Stuck (where/how long) · Hub resource used · Changed my mind because… · Still unclear.

Where this goes next

Exercise check. Drag P to the target zone. In one sentence: which magnetometer would you actually build for 10 nT @ 1 kHz, which corner did you spend to get there, and what single change to the task would make the opposite corner win?
Step 1 of 4

Seven magnetometers — the evidence for the exercise

The dots highlighted in the triangle above are these seven (greyed dots are non-magnetometers, for scale). • Sensitivity — smallest field you can resolve (noise floor). • Dynamic range — smallest to largest before it saturates. • Bandwidth — how fast a change you can follow.

TechnologySensitivity (noise floor)Dynamic rangeBandwidthBest corner
Hall probe~0.1–1 µT/√HzmT – tens of TDC – MHzrange
MR · magnetoresistive (AMR/GMR/TMR)~0.1–10 nT/√HznT – mTDC – MHzbandwidth
Coil · search / induction coil~fT–pT/√Hz (rises at low f)wideAC only, ~Hz – MHzbandwidth
FGM · fluxgate~5–10 pT/√Hz±~100 µT (Earth field)DC – ~1 kHzrange
NV · NV-diamond~1 pT – nT/√HzµT – T (vector)DC – MHz+bandwidth
SQUID~1–10 fT/√Hzwide (flux-locked loop)DC – kHz+sensitivity
OPM · SERF / optically-pumped~1–15 fT/√Hzsmall (near-zero field)DC – ~100–200 Hzsensitivity

Order-of-magnitude, after Bennett et al., Sensors 21, 5568 (2021) (open: arXiv:2106.15843); the conventional sensors after Lenz & Edelstein, IEEE Sens. J. 6, 631 (2006). Abbreviations & how each one works: the reference list below.

For 10 nT @ 1 kHz: 10 nT is a large field, so sensitivity is not the binding constraint — only the Hall probe (µT floor) is too coarse to see it. 1 kHz bandwidth is: it rules out SERF/OPM (fades by ~200 Hz) and strains the fluxgate. Which room-temperature rows actually clear 1 kHz with field to spare? Deciding that is the Connect step (4).

Reference — how each magnetometer works (lookup only, not part of the exercise)

Operating principle — one line each

Abbreviations

The seven graph/table labels: Hall Hall probe · MR magnetoresistive (AMR / GMR / TMR = anisotropic / giant / tunnelling) · Coil search / induction coil · FGM fluxgate magnetometer · NV nitrogen-vacancy (centre in diamond) · SQUID superconducting quantum interference device · OPM optically-pumped magnetometer. Also: SERF spin-exchange-relaxation-free · ODMR optically-detected magnetic resonance · /√Hz per-root-hertz (amplitude noise density) · T tesla (1 T = 10⁹ nT).

The same triangle beyond magnetometers

It governs sensors at every scale — the clock buys 10¹⁸ sensitivity by surrendering range and speed:

SensorSensitivityDynamic rangeBandwidthCorner
Human eye (dark-adapted rod)≈ a single photon~10⁶ (with adaptation)~10–60 Hzsensitive & wide, but slow
Smartphone MEMS accelerometer~10⁻³ m/s²0 – 80 m/s² (≈ ±8 g)< 100 Hzwide & fast, modest sensitivity
Trapped-ion optical clock~1 part in 10¹⁸one transition ± few MHz~1 reading / hourultimate sensitivity, narrow & slow

From the 2025 Sensing I notes.