Every measurement is a translation
Up to now your circuits have lived in a comfortable world where signals arrive already as voltages — a function generator, a divider, the output of a previous stage. But a thermostat does not start with a voltage; it starts with warmth. A camera starts with light, a scale with force, a microphone with air pressure. The part that bridges that gap is a transducer (when its job is to measure, we usually call it a sensor): a device that translates some physical quantity of the world into an electrical one you can finally do circuit-things to. Turning the world into voltage is the very first link of every measurement chain, and it is what this whole rung is about.
Sensors come in two broad families, and telling them apart is the first thing to learn. Some sensors change a passive property — most often their resistance — in response to the world: they sit there inert until your circuit interrogates them, like a question that only gives an answer when you ask. A thermistor is just a resistor whose value slides with temperature. The other family generates its own signal: a thermocouple actually produces a voltage, a photodiode pushes out a current, all on their own. The first family needs a circuit to read it; the second needs a circuit to amplify it. Both feed the same downstream grooming.
A field guide to the usual suspects
The resistive family is the one your divider knowledge already prepares you for. A thermistor drops resistance as it warms; an RTD (a coil of platinum, the Pt100) rises gently and very predictably with temperature; a light-dependent resistor falls as light brightens; a strain gauge changes by a whisper when a metal surface stretches, and a load cell is just several strain gauges bonded to a flexing beam to weigh things; a potentiometer is a resistor with a sliding tap, so its wiper position is a built-in divider reporting an angle or a distance. For all of these, your job is to turn a resistance into a voltage.
SENSOR SENSES OUTPUT IS... WATCH OUT FOR ---------------- -------------- --------------------- ---------------------------- thermistor temperature resistance (falls) very nonlinear; self-heating RTD (Pt100) temperature resistance (rises) small change; wants a bridge thermocouple temperature voltage ~40 uV/degC microvolts; cold-junction ref photodiode light current (~nA to uA) wants a transimpedance amp LDR light resistance (falls) slow, loose tolerance strain gauge force / strain resistance ~0.1% FS buried in offset; use a bridge load cell weight a few mV per volt needs in-amp + excitation potentiometer position/angle voltage (a divider) wiper noise; loading microphone sound small AC voltage needs gain + a bias supply Hall sensor magnetic field voltage (ratiometric) offset drift over temp
The signal-generating family is more dramatic but more demanding. A thermocouple makes a voltage from the junction of two dissimilar metals — but only about 40 microvolts per degree C, so a 100-degree difference is a mere 4 mV that you must amplify perhaps 250 times. A photodiode answers light with a current, often nanoamps, begging for a transimpedance amplifier to convert it to a voltage. A microphone hands you a small AC voltage; a Hall-effect sensor reports a magnetic field as a voltage that helpfully tracks its supply. The common thread: these outputs are small, sometimes tiny, and the rest of this guide is about getting them safely up to a size an ADC can read.
From a tiny wiggle to a usable voltage
Reading a resistive sensor is, at heart, the divider you already own. Put a 10 k thermistor on top and a 10 k fixed resistor on the bottom across a 5 V reference, and the midpoint sits at 5 times 10/(10+10) = 2.5 V when the thermistor is at its 10 k temperature. Warm it so it falls to 5 k and the output swings to 5 times 5/(5+10) = 1.67 V. A comfortable, easy-to-read voltage that an ADC will happily digitize. For sensors that swing a lot, that is the whole story — a divider and you are done.
But now meet the cruel case. A strain gauge of 120 ohm changes by only about 0.1% at full load, so 120 ohm becomes 120.12 ohm. Drop it into that same divider against a 120 ohm fixed resistor on 5 V and the output moves from 2.500 V to 2.50125 V — a real signal of just 1.25 mV riding on a 2.5 V pedestal, a ratio of two thousand to one. To resolve that wiggle to even 1% you would need to claw it out of a huge unchanging offset, asking an absurd amount of an ordinary ADC. The fix is the Wheatstone bridge: a second divider beside the first, balanced so the two pedestals cancel and only the difference — the precious 1.25 mV — survives. That bridge is the entire subject of guide 2.
A bridge hands you a small differential signal — the difference between two nodes — but both of those nodes sit up near half the supply, a large common-mode voltage you do not care about. What you need is an amplifier that fiercely magnifies the difference while utterly ignoring the part both inputs share. That is the instrumentation amplifier, a refined cousin of the difference amplifier, and its merit is measured by its common-mode rejection ratio (CMRR): how many thousands-to-one it favours the wanted difference over the unwanted common pedestal. High gain and high CMRR together — that pairing is the whole of guide 3.
The conditioning chain before the ADC
Between the raw sensor and the converter sits a short assembly line called signal conditioning — a sequence of small, honest jobs that turn an awkward physical signal into one the ADC can read cleanly. None of these jobs is new to you: they are the dividers, op-amps, and filters of earlier rungs, now lined up with a purpose. Think of the whole thing as a signal chain, where each link must hand the next a voltage of the right size, the right offset, and the right cleanliness.
- Buffer and amplify. First lift the tiny sensor output to a sensible level with an instrumentation or non-inverting amp — a thermocouple's 4 mV at gain 250 becomes a healthy 1 V. Do the amplifying early, while the signal is still bigger than the noise that later stages add.
- Remove the offset. Many sensors carry a stubborn pedestal — a bridge's mid-rail, a sensor that reads 0.5 V at zero. Subtract it (with the in-amp's reference pin or a summing stage) so your full signal swing, not a dead pedestal, fills the ADC's input range.
- Scale and level-shift to fit. Match the swing to the converter's window: attenuate a signal that is too big, add gain to one too small, and shift the whole thing up or down so a bipolar +/-10 V signal lands inside a 0 to 3.3 V ADC. This scaling and level-shifting is guide 4's first half.
- Anti-alias filter, then sample. Pass the conditioned signal through a low-pass filter that removes everything above half the sampling rate, and only then digitize. This last filter is non-negotiable — and the second half of guide 4.
Practical realities the datasheet whispers
Few sensors are honest straight lines. A thermistor's resistance curves steeply, a thermocouple's voltage bends, a strain gauge is nearly linear but not perfectly. Linearity is the question of how far a sensor strays from the tidy proportionality you wish it had, and the cure is calibration: measuring the sensor against a known truth at several points and building a correction — a curve, a lookup table, a polynomial in software — that maps its raw reading back to a real-world value. Calibration is where a rough sensor becomes a trustworthy instrument, and it is the heart of guide 5.
Here is an elegant trick that costs nothing. Power a resistive sensor's divider or bridge from the same reference that the ADC uses to judge its input. Now if that reference sags or drifts, both the sensor's output and the ADC's yardstick move together, and the ratio between them — which is all the ADC actually reports — stays put. This is ratiometric measurement, and it quietly cancels the one error source (a wandering supply) that would otherwise haunt every reading. Wherever you can arrange it, the reading depends on a ratio, not on any absolute voltage being perfect.
Two more realities. First, loading: a high-impedance sensor handed to a low-impedance reader sags, exactly the loading effect from the foundations rung — a potentiometer wiper or a photodiode wants a buffer with a very high input resistance so the act of reading does not change what is read. Second, sending a signal down a long noisy cable: the industrial answer is the 4-20 mA current loop, which carries the measurement as a current rather than a voltage, so resistance in the wires cannot rob it, and a 'live zero' of 4 mA for the bottom of scale means a true 0 mA unmistakably signals a broken wire.