IoT devices are often described as “detecting people”, “measuring distance”, “checking air quality”, “measuring body fat”, or “detecting leaks”. That language is fine at the product level, but it can be misleading when you design, install, or debug a device.
Most sensors do not measure those business concepts directly.
They first measure lower-level physical changes: acoustic flight time, electromagnetic echoes, infrared heat changes, light intensity, resistance, pressure, magnetic field, strain, or body impedance. Distance, presence, air quality, body fat percentage, and leak alarms are interpretations built on top of conversion, compensation, thresholds, calibration, and algorithms.
A safer model is:
Physical change
-> Sensitive element response
-> Analog front end and ADC
-> Filtering, compensation, calibration
-> Business state or numeric output
The key point is simple: a sensor reading is not a direct copy of reality.
First Ask What It Really Measured
Many mistakes come from treating the final output as the first measurement.
An ultrasonic ranging sensor does not directly “see a wall”. It measures acoustic round-trip time. A mmWave presence sensor does not understand a human body directly. It looks at echoes, phase, micro-motion, and background changes. A PIR sensor does not detect “a person”; it detects changes in infrared heat radiation across lens zones.
A temperature and humidity sensor measures the air around the sensor, not necessarily the whole room. A gas sensor measures a material, current, or optical-path response, not necessarily one clean gas species. A body fat scale measures body impedance, not fat mass directly.
Once the first measured quantity is clear, many field problems become easier to explain:
- Why ultrasonic readings jump on cloth, angled surfaces, and thin rods
- Why mmWave can be disturbed by fans, curtains, and moving plants
- Why PIR is good at motion but weak at still presence
- Why temperature and humidity readings do not match comfort
- Why a gas sensor can show
ppmwithout being truly selective - Why body fat scale values can change within a day
These are not always software bugs. They often come from the gap between the measurement chain and the output meaning.
Then Ask What Model Interpreted It
Sensor readings usually pass through a model.
A pressure sensor measures pressure; liquid level and altitude are inferred from pressure models. A scale measures deformation caused by force; mass is a calibrated result. A magnetometer measures a three-axis magnetic field; heading comes from tilt compensation, calibration, and a geomagnetic model. A light sensor outputting lux is weighted by human photopic vision, not by all optical energy in every application.
Models are necessary. Without models, many sensors would not produce useful data.
The problem is that every model has assumptions:
- Pressure-to-level conversion assumes the right density and pressure reference
- A scale assumes the force path and calibration remain valid
- A compass assumes the local magnetic field is not heavily polluted
- A lux sensor assumes the spectrum and placement match the use case
- BIA body-fat estimation assumes hydration, contact, and population model fit the user
When those assumptions fail, the sensor may still produce a number, but that number no longer means what the application expects.
Finally Check the Field Conditions
The same sensor can see a completely different world after installation.
Ultrasonic and ToF sensors depend on target material, angle, blind zones, and enclosure openings. Temperature, humidity, gas, and light sensors depend on enclosure design, airflow, self-heating, and window material. Hall sensors and magnetometers depend on magnet orientation, air gap, steel structures, and external magnetic fields. Water leak sensors depend on floor slope, probe height, dirt, and liquid conductivity.
Many field issues should not start with code:
Check the physical path first
Then check placement and mounting
Then check compensation and calibration
Then inspect algorithm thresholds
If the sensor never touches the right physical phenomenon, later software can only process an already-biased input.
A Useful Classification
IoT devices contain many sensors, but most misunderstandings fall into a few measurement models:
- Distance and presence: ultrasonic, mmWave, PIR, optical ToF
- Attitude and magnetic field: tilt, IMU, magnetometer, Hall
- Environment and home state: temperature, humidity, pressure, gas, light, water leak
- Electrical, force, and body impedance: current sensing, scale, BIA body fat scale
When you meet a new product, you do not always need a new mental model. A door contact can often be understood as a magnetic switch or Hall sensor. Smart-plug energy metering starts from current, voltage, and power. A touch key starts from capacitance change. A smoke alarm starts from particles, an optical path, or a gas response.
The reusable rule is:
Sensors measure physical changes.
Business states are interpretations based on models, calibration, installation, and algorithms.
Ask that first whenever you read a sensor value. It prevents many wrong conclusions.