In this guide
How to Choose an Op-Amp for Sensor Circuits: Gain, Bandwidth, Offset and Supply
Most sensor projects fail at the amplifier, not the sensor. A temperature, pressure, light or current sensor usually produces a small, awkward signal: a few millivolts, a tiny current, or a voltage that sits close to a supply rail. The op-amp in between has to lift that signal to a level an ADC can read cleanly, without adding so much error that the reading is meaningless. Choosing the part by price or by familiarity alone is how a circuit ends up with a flatlined output, a noisy reading, or a gain that collapses at the frequency that matters.
The problem a sensor amplifier has to solve
A sensor rarely hands you a convenient signal. A thermocouple produces tens of microvolts per degree. A photodiode produces current, not voltage. A bridge sensor produces a small difference between two voltages that are both near half the supply. A shunt resistor produces a small voltage sitting on top of a large common voltage. Each case asks the amplifier for something different: gain, a difference measurement, a current-to-voltage conversion, or rejection of a shared voltage. The selection process is therefore not "find the best op-amp" but "find the op-amp whose weaknesses do not matter for this sensor".
How selection works, step by step
- Describe the sensor signal first: its range, its source resistance, whether it is single-ended or differential, and whether it is DC, slow, or fast.
- Decide what the ADC needs: the full-scale voltage, the resolution you actually need, and the sampling rate.
- Work out the required gain from the sensor range and the ADC range, and check it against bandwidth at that gain.
- Check the error budget: offset voltage, input bias current, noise, and drift against the smallest change you need to resolve.
- Check the supply reality: single or dual supply, how close the output must swing to the rails, and what the input range must cover.
- Shortlist two or three parts, compare them on the same datasheet conditions, then test one on a breadboard with the real sensor before committing to a board layout.
Gain and bandwidth: the pair that trades against each other
Gain is the easy number, and it is the one that misleads most often. An op-amp's gain-bandwidth product tells you the bandwidth available at a given closed-loop gain: roughly, bandwidth equals gain-bandwidth product divided by gain. A part with a 1 MHz gain-bandwidth product gives about 100 kHz of bandwidth at a gain of 10, but only about 10 kHz at a gain of 100. If your sensor signal changes at 20 kHz and you need a gain of 100, that part is already out of its depth, and the output will look rounded and late even though the DC gain is correct.
Slew rate is the second speed limit, and it matters for large, fast signals. It is the fastest rate at which the output voltage can change, in volts per microsecond. A small signal can be fast without much slew, but a full-scale sine wave at a high frequency needs a surprising amount of it. Check both limits: bandwidth for small signals, slew rate for large ones. For slow signals such as temperature, humidity or weight, almost any general-purpose part is fast enough, and speed should not drive the choice.
Offset voltage, bias current and drift: the quiet errors
Offset voltage is the small voltage the amplifier adds to its own output even with zero input difference. A 5 mV offset after a gain of 100 becomes 500 mV at the output, which can swallow a sensor signal that only spans a few hundred millivolts. If the smallest change you need to measure is small, offset matters more than almost anything else, and a precision part with a sub-millivolt offset is worth its price.
Input bias current is the tiny current that flows into each input pin. Through a high source resistance it creates a voltage error: bias current multiplied by resistance. A sensor behind a 100 k resistor with a part drawing 100 nA of bias current picks up 10 mV of error before any gain. High-impedance sensors such as pH probes and photodiodes therefore need parts with very low bias current, typically FET-input types, while low-resistance sources such as shunts barely notice it.
Drift is offset changing with temperature. A circuit calibrated on a cool bench can read differently in a warm enclosure if the offset drifts. Datasheets quote this in microvolts per degree. For a project graded in a lab at one temperature it may not matter; for anything measured outdoors or over a day, it belongs in the error budget alongside offset itself.
Single supply, dual supply and rail-to-rail
Many student circuits run from a single 5 V or 3.3 V supply, and this is where rail behaviour bites. A classic op-amp designed for dual supplies may not accept inputs near ground, and its output may not swing closer than a volt or two to either rail. A sensor signal that lives near ground then gets clipped, and the bottom of the measurement range simply disappears.
Rail-to-rail input means the inputs work across the full supply range; rail-to-rail output means the output can swing almost to the rails. You may need one without the other: a bridge sensor centred at half supply may need neither, while a single-supply photodiode amplifier reading near ground needs both. Read the datasheet's input common-mode range and output swing figures at your actual supply voltage and load, not the headline claim. "Rail-to-rail" on the front page sometimes means only the output, or only under light load.
Matching the amplifier to the sensor type
| Sensor situation | What matters most | What to watch |
|---|---|---|
| Thermocouple or bridge, small differential voltage | Low offset, good common-mode rejection | Offset after gain, drift with temperature |
| Photodiode, current output | Very low bias current, low noise | Bias current through the feedback resistor, stability with large feedback resistance |
| Shunt resistor, small voltage on a large common voltage | Common-mode range, rejection | Whether the part tolerates the common voltage at all |
| High-impedance probe (pH, piezo) | FET input, low bias current | Leakage on the board itself, guarding and cleanliness |
| Fast signal (ultrasonic, vibration) | Gain-bandwidth product, slew rate | Bandwidth at the actual gain, not at unity gain |
Common-mode rejection ratio (CMRR) deserves a note for bridge and shunt work: it measures how well the amplifier ignores a voltage shared by both inputs. A bridge output is a small difference riding on a large shared voltage, so poor rejection turns supply wobble into measurement error. Resistor matching in a discrete difference amplifier often limits rejection more than the op-amp itself, which is why instrumentation amplifiers exist for this job.
Noise: when it matters and when it does not
Noise is the random fuzz on the output, quoted as a voltage density in nanovolts per root hertz. It matters when the sensor signal is small and the bandwidth is wide, because noise grows with the square root of bandwidth. A wideband, high-gain photodiode amplifier can be noise-limited; a temperature logger sampled once a second almost never is, because averaging over slow samples removes most of it. Before paying for a low-noise part, work out whether your error budget is dominated by noise or by offset and bias, which are usually larger in student circuits. Filtering and averaging are cheaper than a premium part when the signal is slow.
A worked selection example
Suppose a bridge pressure sensor gives a 0 to 20 mV differential output, centred at 2.5 V, and the ADC accepts 0 to 3.3 V. A gain of about 100 turns 20 mV into 2 V, which fits with headroom. The signal changes slowly, so bandwidth is not the constraint. The error budget says the smallest pressure step of interest is 0.1 mV at the sensor, which becomes 10 mV after gain, so an offset of a few millivolts at the input is tolerable but not negligible. The 2.5 V common voltage sits at mid-supply, so ordinary input range is fine, but the output must swing from near 0.5 V to near 2.5 V on a single 3.3 V supply, which a non-rail-to-rail output may not reach. The shortlist therefore favours a low-offset, single-supply part with a rail-to-rail output, and rejects a fast, high-offset part even though its headline speed looks better.
Common mistakes
- Picking a part for its unity-gain bandwidth, then running it at a gain of 200 where the real bandwidth is two hundred times smaller.
- Ignoring bias current with a high-resistance source, then wondering why the reading has a fixed offset that changes when the resistor value changes.
- Assuming a dual-supply classic part works on a single 5 V supply near ground. It often does not, and the symptom is a dead zone at the bottom of the range.
- Trusting the headline "rail-to-rail" without checking the output swing at your load current and supply voltage.
- Adding gain to fix a noisy reading when the noise is coming from the sensor wiring or the supply, not the amplifier. Gain amplifies noise exactly as it amplifies signal.
- Calibrating offset out at one temperature and assuming it stays calibrated. Offset drift is a real error source, not a calibration failure.
Frequently asked questions
Is a more expensive precision op-amp always better?
No. Precision parts trade speed, supply range or output drive for low offset and drift. A slow, precise part can be the wrong choice for a fast signal, and a part with a narrow supply range can be unusable on a single 3.3 V rail. Better means better matched to this sensor, not better on every datasheet line.
Should I use an instrumentation amplifier instead?
For bridge and shunt measurements where the difference is small and the common voltage is large, usually yes. An instrumentation amplifier packages three op-amps with matched resistors, giving high input impedance and strong common-mode rejection in one part, set by a single gain resistor. It costs more than a single op-amp, but it removes the resistor-matching problem that limits a discrete difference amplifier.
Can I just use the part I already know?
For a first prototype, yes, and that is often the fastest way to learn what the circuit actually needs. But treat the familiar part as a test, not a decision: measure the offset, the clipping points and the noise with the real sensor, and be ready to change parts when the measurements show a limit. A known part with a known weakness beats an unknown part with an unknown one, provided you check.
How do I compare two datasheets fairly?
Compare figures at the same supply voltage, temperature and load. Offset, bias current and noise all move with conditions, and a part that looks best at 15 V dual supply may look ordinary at 3.3 V single supply. The conditions column is part of the specification, not fine print to skip.
Limitations
This guide covers selection for common student sensor circuits, not high-frequency design, where layout, parasitic capacitance and stability dominate, nor precision metrology, where calibration chains and reference standards take over. Stability with capacitive loads, such as long cables or large feedback capacitors, is a separate topic: a part that is stable on the bench can oscillate when a cable is added. Always prototype with the real sensor, the real cable length and the real supply before finalising a board, because datasheet figures describe the part under test conditions, not your circuit.