LM75 sensor/The reading/05. How close to the truth
The reading · 05 of 9

How close to the truth

The chip counts in eighths of a degree and is allowed to be two degrees off. Those are different numbers: the step says how small a change it can see, the accuracy how far its answer may sit from the air's. This book promises NXP's ±2 °C, the looser of the two datasheets.

Two numbers that look alike

Resolution is the smallest step the chip reports: 0.125 °C, an eighth. Accuracy is how far the reported number may be from the real temperature: on NXP's datasheet, up to 2 °C either way from −25 to 100 °C, and 3 °C beyond that, to −55 and 125 °C. Both are maxima, not typical values.

How close to the truth
Datasheet
True temperature25 °C
One step
0.125 °C
Allowed error
2 °C either way
Steps inside it
32
Fine steps, coarse truth. The chip counts in eighths of a degree, and NXP allows the count to be 2 °C off at 25 °C: 32 steps of room. So a reading that moves by 0.125 means the temperature moved; a reading of 23.5 does not mean the room is 23.5. Two blocks side by side may disagree by up to 4 °C and both be in spec.

The step is sixteen times finer than the error. So a change of 0.125 means the temperature really changed, and the chip is good at showing a room warming by a fraction of a degree. The number itself, 23.5, may be the room's 22 or its 25.

Two datasheets

The BOM names UMW's LM75BD, and UMW's own datasheet describes a better part than NXP's: 12 bits, a sixteenth of a degree, and ±1 °C across the whole range. LCSC's listing for the same UMW part quotes NXP's numbers instead.

Which silicon a given reel holds cannot be checked from the outside. So this book promises what both would meet: ±2 °C from −25 to 100 °C, steps of 0.125 °C or finer, and a new reading every 100 ms or sooner. The sketch divides by 256, which gives the right answer at either resolution.

What it is for

A number to within a couple of degrees, every tenth of a second, on two wires shared with other devices. That is a room, a case, a heatsink, a greenhouse, a fridge. It is not a clinical thermometer, and a reading of 36.8 from it says nothing about a fever.

When it does not work

Two TK42s side by side disagree by a degree.

Both can be right. Each is allowed ±2 °C from −25 to 100 °C on NXP's figures, so two of them may differ by up to 4 °C and still be in spec. If they have to agree, read both at one known temperature and store the difference as an offset.

The reading is a degree above my room thermometer.

First check where it sits: next to a microcontroller, a regulator or a warm hand it measures that. Then remember the room thermometer has its own error. A difference inside two degrees is within what the datasheet allows.

Can I calibrate it?

An offset, yes: note the difference against a trusted thermometer at the temperature you care about and subtract it in the sketch. It corrects this one chip near that temperature; it does not make the other readings better than the datasheet.

Where this goes next

Where it sits on the bus, and what cannot share it.

One address, 0x48 →

Edit this page — content/books/lm75/how-close-to-the-truth.mdx

Community

Questions about this product

See what other owners have asked, and read their solutions.

Ask a question ↗

LM75 Temperature Sensor

Loading discussions…

Discuss this article

Ask about this page. The answer stays here, on the page it belongs to, for whoever hits the same wall next.

Browse Modules and blocks on the forum →