Comparator Hysteresis Calculator

Give it the trip points you want and it returns real E-series resistors, the reference voltage those thresholds actually imply, and the thresholds you will measure with the reference you have — then checks the result against the input noise you expect.

Runs entirely in your browser. Nothing is uploaded, stored, or sent anywhere.

Why a comparator without hysteresis is a mistake

A bare comparator has enormous gain and no memory. When the input sits near the threshold — which is exactly where it sits during any slow transition — a few millivolts of noise crosses the trip point repeatedly, and the output produces a burst of edges instead of one. On a slow ramp that burst can be hundreds of transitions.

Downstream that is rarely harmless. A counter counts them, an interrupt fires on each, a relay buzzes, a motor controller sees a stream of contradictory commands. It is also intermittent and load-dependent, which makes it a genuinely unpleasant fault to chase after the boards are built.

Hysteresis fixes it by making the threshold depend on the current output state: once the output flips, the trip point moves away from the input, so noise has to be larger than the hysteresis band to flip it back.

The two knobs are independent, and that is the useful part

The width of the band depends only on the resistor ratio. The position of the band depends only on the reference. They do not interact, which means you can design them separately — set the reference for where you want to trip, then choose the ratio for how much noise immunity you need.

It also means scaling both resistors together changes nothing electrically. That freedom is what lets you move the network impedance up to stop loading the source or the output, or down to reduce the offset caused by input bias current flowing through it.

What this tool does that the others do not

• Real parts. The ideal ratio is never an E-series ratio. It solves for stock values and reports the thresholds you will actually measure, not the ones the algebra wanted.
• It tells you what reference your thresholds imply. Ask for trips at 2.0 and 1.8 V and there is exactly one reference voltage that centres them. If yours is different, the band still has the width you asked for but sits somewhere else — and the tool shows you where rather than silently assuming you can dial the reference to whatever suits.
• Source impedance is inside the feedback network. In the non-inverting topology, the signal source's own impedance is in series with R1 and therefore changes the ratio. Drive it from a 10 kΩ sensor and your carefully chosen hysteresis is not what you calculated. Most calculators ignore this entirely.
• The noise check. You tell it how much input noise you expect and it says plainly whether the hysteresis is enough. Hysteresis smaller than the noise does not reduce chatter; it just moves it.

Open-drain outputs need care

With an open-drain comparator, V_OH is not a property of the comparator at all — it is set by the pull-up resistor. And the feedback resistor is a second path to that same node, so the pull-up and the feedback network form a divider and the real V_OH sits below the rail. Feed the loaded value into the calculator, not the supply voltage, or the hysteresis will come out smaller than predicted.

The same node also has to charge whatever capacitance is on it through the pull-up, so an open-drain output is slower on the rising edge — which on a fast comparator can be the dominant limit on how quickly the feedback actually arrives.

How much hysteresis

Enough to exceed the noise, with margin. Two to three times the peak-to-peak noise is a common starting point, and the noise to use is the noise at the comparator input — including supply coupling and whatever the input filter did not remove, not just the sensor's own datasheet figure.

The cost of more hysteresis is accuracy: a 200 mV band means the trip point is uncertain by 200 mV depending on which way you arrived. For a threshold detector that is usually irrelevant; for a precision level measurement it may not be, and the answer there is a filter before the comparator rather than a wider band.

More free tools

Each of these runs entirely in your browser. Nothing is uploaded, stored or sent anywhere, and none of them needs an email address.

See all 49 engineering tools →