Why data sheets look more precise than they are
A sensor data sheet is a long list of numbers, and it is easy to read them as promises. Most of them are better read as conditions: this is what the part does under these particular circumstances. Knowing the vocabulary helps you compare parts fairly and spot when a headline figure hides something important.
Range
The measurement range is the span of values the sensor can report, for example a temperature sensor covering minus 40 to 85 degrees Celsius. Outside it, readings may be wrong or the part may be damaged.
Two ranges are often listed, and they are not the same:
- Measurement or operating range. Where the sensor performs as specified.
- Absolute maximum ratings. Limits beyond which damage can occur. A part may survive the absolute maximum yet not give valid readings there.
Accuracy often varies across the range. A sensor may be good in the middle and worse near the ends, so look for the range where the stated accuracy applies.
Resolution
Resolution is the smallest change the sensor or its converter can show. If a hypothetical temperature sensor covers 0 to 100 degrees with a 10-bit output, there are 1,024 steps, so each step is about 0.098 degrees (100 divided by 1,024). A 12-bit version would have 4,096 steps of about 0.024 degrees.
Resolution is about how finely the number is divided, not whether the number is right. A display showing three decimal places can still be wrong in the first decimal.
Accuracy
Accuracy describes how close a reading is to the true value, usually written as a plus-or-minus figure, such as plus or minus 0.5 degrees. Several details change what that means:
- It may apply only at a particular temperature or part of the range.
- It can be expressed as a percentage of the reading, of the full scale, or as a fixed amount. Percent of full scale is less forgiving at low readings. A sensor with a hypothetical accuracy of 1 percent of a 0 to 100 range is off by up to 1 unit even when measuring 5, which is 20 percent of the reading.
- It may include or exclude certain errors, such as temperature effects or long-term drift.
- It is often a typical value rather than a guaranteed maximum. Look for “typical” and “maximum” columns.
Accuracy is the number that decides whether a sensor is trustworthy for a task, and it is the one most affected by installation and age. The topic of sensor calibration covers how it can be improved.
Precision and repeatability
Precision, or repeatability, describes how closely repeated readings agree with each other under the same conditions. A sensor can be precise but inaccurate, consistently reading 1.5 degrees too high. That kind of fixed offset is often correctable. Random scatter, the noise, is harder to remove and is usually reduced by averaging.
Response time
Response time says how quickly the output follows a change. It is often given as the time to reach a certain portion, such as 63 percent, of a step change. This is sometimes called a time constant. After several time constants, the reading approaches the final value. Watch for:
- The test medium, since a temperature sensor responds faster in moving water than in still air.
- The effect of the housing. A sensor inside a plastic case or a thick probe reacts more slowly than the bare element.
- The difference between the sensor’s response and the sampling interval of your device. Reading every ten minutes cannot capture a change lasting ten seconds.
Other terms worth knowing
- Sensitivity. How much the output changes per unit of the measured quantity, for example millivolts per degree. A higher sensitivity makes small changes easier to see.
- Linearity. How closely the output follows a straight line. Nonlinear sensors need a correction formula.
- Hysteresis. The reading differs depending on whether the value is rising or falling. Humidity and pressure sensors often show it.
- Offset and gain error. A constant shift and a scaling error, respectively. Both can sometimes be fixed in software.
- Drift. Change over time with no change in the real quantity, usually listed per year or per thousand hours.
- Long-term stability. An indication of how often recalibration may be needed.
- Noise. Random variation, sometimes given as a root-mean-square value or per square root of frequency.
- Warm-up time. How long the sensor needs after power-up before readings are valid, which can range from instant to many minutes for chemical sensors.
- Supply current and sleep current. Key figures for battery operation, and tightly tied to battery life and duty cycling.
- Interface. Analog voltage, or digital buses such as I2C and SPI. This affects wiring, noise and which controllers can read it.
Common traps
- Confusing resolution with accuracy. More digits are not more truth.
- Reading the best case. Specifications often assume a stable lab temperature and a clean power supply.
- Ignoring the footnotes. Conditions and exceptions are frequently small print under a table.
- Comparing mismatched figures. Typical versus maximum, percent of reading versus full scale.
- Forgetting the system. The converter, power supply noise, wiring and enclosure all add error to the sensor’s own.
- Overlooking environmental effects. Temperature, humidity, vibration and exposure to dust or chemicals can shift performance.
A simple routine for comparing sensors
Write down what you need: the range you must cover, the error you can live with, how fast the quantity changes, and your power budget. Then find each of these in the data sheet under the right conditions, and note whether it is typical or guaranteed. If two parts look similar, prefer the one whose documentation is clearer about conditions. Where stakes are high, test a sample against a trusted reference before committing, because the data sheet describes the part and not your installation.