Sensors, actuators and signal acquisition
Follow the complete physical signal chain—from the quantity being measured to the action taken—and design each conversion, assumption and failure response deliberately.
After this module, you should be able to:
- Describe a complete measurement and actuation chain
- Select sensors using meaningful performance characteristics
- Reason about conditioning, sampling and calibration
- Design diagnostic and verification coverage for physical interfaces
Study guide · Updated · About 10 minutes reading + 20–30 minutes practice
Before you start: Read TEA-102. Basic voltage, sampling and calibration concepts are useful.
The system does not receive reality. It receives a representation.
A sensor converts a physical quantity—the measurand—into an electrical or digital representation. Conditioning, conversion, timing and software then transform that representation again. Each stage adds limits, uncertainty, delay and possible failure modes.
Choose the sensor for the behaviour you must distinguish.
| Characteristic | What it means in practice | Design question |
|---|---|---|
| Range | Minimum and maximum input without unacceptable damage or loss of performance. | What normal, transient and fault values are credible? |
| Accuracy | Closeness to the reference value under stated conditions. | What total error is allowed across temperature and life? |
| Precision / repeatability | Agreement between repeated measurements. | Can the system reliably distinguish the required change? |
| Resolution | Smallest observable increment in the output chain. | Is useful resolution limited by noise rather than converter bits? |
| Bandwidth and response | How quickly changing input is represented. | Which dynamics must be captured and which should be rejected? |
| Drift and hysteresis | Change with time and dependence on previous input direction. | What calibration or compensation is needed? |
| Cross-sensitivity | Response to temperature, vibration, supply or another unwanted influence. | Can environment mimic the quantity being measured? |
Sensor selection includes installation. Mechanical stress, cable routing, heat sources, grounding, connector resistance and ingress protection can dominate laboratory data-sheet performance.
Sampling must preserve the information the decision needs.
Condition before conversion
The analogue front end may provide excitation, protection, impedance matching, amplification, level shifting and filtering. Its output must remain within the converter’s input and common-mode limits for normal and credible abnormal conditions.
Control aliasing
The theoretical Nyquist threshold requires a sampling frequency greater than twice the highest frequency to be represented. A practical design also needs an analogue anti-alias filter, realistic transition band and margin for interference and timing uncertainty. Sampling faster does not recover information already distorted or clipped upstream.
Use an honest error budget
ADC bit depth is not system accuracy. Include sensor error, reference tolerance, offset, gain error, noise, non-linearity, temperature effects, quantisation, timing and calibration residuals. Effective number of bits can be far lower than nominal resolution.
Make time part of the data
For multiple channels, define whether samples are simultaneous or sequential, when timestamps are assigned and how acquisition aligns with control outputs. Stale but plausible data can be more dangerous than clearly invalid data.
An actuator converts information into energy.
The command signal is only one part of the actuation chain. The driver, supply, load, mechanical transmission, thermal behaviour and feedback path determine what actually happens.
- Define voltage, current, duty cycle, force, torque, speed, travel and thermal limits.
- Consider saturation, stiction, backlash, wear, stall, blocked motion and loss of supply.
- Specify the de-energised condition and whether it is genuinely safe for every operating state.
- Separate command monitoring from outcome monitoring: reading back an output register does not prove physical action.
- Use interlocks or independent cut-offs where software response is too slow or cannot be trusted for the hazard.
Worked example: closed-loop temperature control
Useful diagnostics include open- and short-circuit thresholds, rate-of-change limits, disagreement with an independent sensor and a check that temperature responds after heater energy is applied. Each diagnostic needs a detection time and a defined system response.
Verify the chain under realistic conditions.
Range, units, transfer functions, filtering, sampling, latency and diagnostic thresholds.
Allocated contributors, assumptions, worst-case combination and acceptance limits.
Reference standards, points, coefficients, storage, validity and recalibration triggers.
Known physical or electrical stimuli at nominal, boundary and abnormal conditions.
Open, short, stuck, drift, noise, stale data, saturation, jam and feedback loss.
Performance across temperature, supply, EMC, vibration, ageing and relevant loads.
Common failure patterns
A component figure is treated as end-to-end accuracy without installation, conditioning or software errors.
A sensor returns a well-formed number that is stale, out of context or physically implausible.
The controller confirms its requested output but does not detect a disconnected or jammed actuator.
Calibration compensates offset while noise, saturation, drift or non-linearity remain uncontrolled.
Separate resolution, accuracy and uncertainty
Resolution describes the smallest representable step. Accuracy concerns closeness to the quantity being measured. Repeatability describes agreement between repeated measurements under stated conditions. More ADC bits can improve resolution while leaving reference drift, sensor offset or installation error unchanged.
For an ideal 12-bit ADC with a 3.3 V span, one nominal code step is about 0.806 mV. That number does not include noise, non-linearity, reference tolerance or the sensor’s error. A temperature conversion also needs the transfer function and its valid range. Carry units through the calculation and specify how saturation is represented.
Build an uncertainty budget for the decision the product makes. When bounded errors may align, a worst-case sum can be appropriate; statistical combination needs justified distributions and dependence assumptions. Calibration can correct some offsets and gains but does not automatically remove ageing, hysteresis, self-heating or errors outside the calibration range.
Choose sampling and filtering from physical behaviour
The acquisition rate must preserve the information required for control, diagnosis and recorded evidence. Frequency content above half the sampling rate can alias into the measured band; an analogue anti-alias filter acts before conversion. A digital filter applied after sampling cannot recover information already aliased into the signal.
Filtering also changes response time. An eight-sample moving average at 100 samples/s has a nominal group delay of 35 ms for its linear-phase frequency response. A threshold crossing can experience different delay depending on the signal and threshold, so measure the relevant event directly. Allocate this latency in the fault-response budget.
Attach validity and acquisition time to each value. A fresh message may contain an old conversion. Use plausibility checks appropriate to the physics: range, rate of change, disagreement and stuck-value behaviour. Stable temperature can be real; a stuck-sensor test needs enough context to distinguish stability from loss of observation.
Verify the actuator’s physical effect
A PWM command is an instruction to a power stage, not a measurement of torque, heat or flow. Feedback may need current, position, speed or another physical quantity. Define what happens when feedback contradicts the command, and distinguish a disconnected load from a mechanically jammed one.
Consider stored energy and the passive state of the mechanism. Removing power can leave a valve open, allow a load to fall or release stored pressure. Choose the fault response using the hazard, then specify the sensing and output path needed to achieve it.
| Observation | Possible explanations | Useful follow-up |
|---|---|---|
| Commanded off, current present | Shorted switch, alternative supply path or measurement error. | Measure load voltage and isolate the energy path. |
| Motor current high, speed zero | Jam, start-up transient or speed-sensor failure. | Compare time, position and current limits. |
| Temperature unchanged | Stable process, detached sensor or frozen conversion. | Use independent stimulus and acquisition diagnostics. |
Practical exercise
An illustrative engineering exercise; values are assumptions for this lesson.
Budget a temperature-measurement chain
A chamber needs a maximum ±0.5 °C measurement error over a defined range. Bounded contributions are sensor ±0.25 °C, electronics ±0.10 °C and calibration residual ±0.08 °C.
- Calculate the remaining worst-case allowance and identify unbudgeted effects.
- Decide what evidence is needed before statistically combining the errors.
- Design a test that distinguishes a frozen value from a genuinely stable chamber.
Show the worked answer
The stated worst-case sum is ±0.43 °C, leaving 0.07 °C of the allowed magnitude. Installation gradients, drift, reference variation and test uncertainty still need consideration. The result is a provisional budget, not evidence that production units meet it. A root-sum-square calculation would need justified distributions and correlations; bounded tolerance values alone do not provide them.
Apply an independent, controlled temperature stimulus and compare the reported value, raw conversion, timestamp and validity. Inject a frozen conversion while the physical temperature changes, then verify detection and the specified response. Also test true stability so the diagnostic does not create unjustified false faults. Retain reference-instrument uncertainty and the physical sensor placement.
Review criteria
- Resolution is not used as a substitute for accuracy.
- The budget includes margin and identifies missing effects.
- The fault test observes physical stimulus, data age and response.
Check your understanding
Can calibration remove every measurement error?
No. It can estimate corrections under defined conditions. Drift, hysteresis, installation effects and uncertainty remain to be evaluated.
Why is a fresh packet not proof of a fresh measurement?
The packet may repeat an old sample. Acquisition time and validity must survive the transport path.
Further learning
- Texas Instruments — High-Speed Analog-to-Digital Converter BasicsSampling theory, ADC performance, selection, aliasing, clock jitter and evaluation.
- NASA Systems Engineering HandbookSystem design, interfaces, verification and technical management across the lifecycle.
Engineer the physical loop end to end.
A dependable embedded system controls the complete chain from measurand to data, decision, energy and verified physical outcome—including uncertainty, timing and failure behaviour at every conversion.