Lesson 50 of 78 · Power, Circuits & Signals
Signals, Noise, Sampling & Conditioning
A sensor produces a signal plus bias, noise, interference, quantization, drift, and delay. Signal conditioning makes that mixture compatible with the controller and the decision.
Scale and resolution
An ideal 12-bit converter over 0–10 V has \(10/4096=2.44\) mV per count. If a pressure transducer maps 0–10 V to 0–1 MPa, one count represents about 244 Pa. That is resolution, not accuracy. Offset, gain error, nonlinearity, noise, and sensor uncertainty remain.
Sampling and aliasing
Sampling at \(f_s\) cannot uniquely represent arbitrary content above \(f_s/2\). An anti-alias filter limits analog bandwidth before conversion. Choose sample rate from relevant plant and fault dynamics, not a slogan. A temperature loop may need seconds; motor current control may need microseconds.
Noise paths
Interference couples capacitively, inductively, conductively through shared impedance, or radiatively. Manage it with physical separation, twisted pairs, appropriate shielding and termination, differential measurement, isolation, short return loops, filtered power, and intentional grounding. “Ground” is not an abstract zero-voltage bucket; currents create voltage differences.
Filtering with consequence
A first-order low-pass filter reduces high-frequency noise but delays response. If a filter time constant is 100 ms, a real 50 ms impact may be strongly attenuated. Separate control filtering, display smoothing, and safety detection. Preserve raw data when possible for diagnosis.
Calibration chain
Trace physical input to sensor output, conditioning, conversion, scaling, displayed engineering unit, alarm, and control decision. Check zero, span, repeatability, hysteresis, and relevant intermediate points. Record as-found and as-left values.
Lab
With a safe low-voltage sensor, log raw and filtered signals at several sample rates. Inject a known slow change and a brief transient. Plot amplitude and delay. Your report must state which filter is acceptable for control, display, and fault detection—and why one setting should not automatically serve all three.
Source trail
References
Further reading
- Design of Electromechanical Robotic Systems. MIT OpenCourseWare. verifiedOpen textbook, assignments, laboratories, and projects for system design under measurement, environmental, and modeling uncertainty.
- Dynamics and Control II. MIT OpenCourseWare. verifiedModeling, parameter estimation, time and frequency response, feedback compensation, implementation, and experimental verification.
Check your understanding
- What can occur when sampling below twice the highest relevant signal frequency?
- Aliasing
- Infinite accuracy
- Zero noise
- Automatic filtering
Frequencies can fold into misleading lower-frequency components unless bandwidth is limited before sampling.
- Why is digital smoothing not a free improvement?
- It can add delay and hide real transients
- It removes units
- It increases sensor range
- It eliminates calibration
Filtering trades noise reduction against bandwidth, phase lag, and fault detectability.