Everything downstream depends on the signal you started with.
Choosing what to measure and how: sensor selection, mounting, amplification, filtering, sampling and noise rejection, so the data arriving at your system is worth processing.
The quiet decision that determines the project
Most connected-product problems that surface late were created early, at the point where somebody chose a sensor from a catalogue and a sampling rate from habit. No amount of firmware, cloud processing or machine learning recovers information that was never captured. If the signal is wrong, everything built on it inherits the fault.
This is unglamorous work and it is where projects are quietly won or lost.
What choosing well involves
- The physical quantity. What actually changes when the condition you care about occurs, which is not always the obvious thing. Bearing wear may show in vibration, in current draw or in temperature, and one of those may be far easier to measure in your installation.
- Range and resolution. Enough range to avoid clipping on the worst case, enough resolution to see the change that matters.
- Bandwidth and sampling. The sampling rate has to capture the phenomenon, with anti-alias filtering ahead of the converter. Sampling too slowly does not produce a coarse version of the signal; it produces a misleading one.
- Mounting. For vibration in particular, how and where a sensor is attached can matter more than which sensor it is.
- Environment. Temperature drift, humidity, chemical exposure and mechanical loading over the service life.
- Electrical noise. Industrial environments are hostile. Motors, drives and switching supplies inject interference that can exceed the signal.
Conditioning between sensor and converter
Between the sensing element and the analogue-to-digital converter sits the work that decides whether the reading is usable: amplification with the right gain and input impedance, filtering to remove what you do not want before it is digitised, isolation where ground potentials differ, protection against surges and miswiring, and a reference stable enough that the measurement means the same thing next month.
Layout matters as much as topology here. Careful analogue routing, grounding and supply decoupling frequently make more difference than a more expensive part. See hardware and PCB design.
Calibration and drift
A measurement chain is only trustworthy if you know how it behaves over time and temperature. That means deciding early whether calibration happens in manufacturing, in the field or both, what reference is used, how coefficients are stored, and how a unit is recalibrated years later by someone who did not build it.
How we help
- Measurement strategy. Working out what to measure to answer your question, including cases where a cheaper indirect measurement does the job.
- Sensor evaluation. Candidate selection against range, bandwidth, environment, availability and cost, tested on the real signal where feasible rather than compared on paper.
- Front-end design. Amplification, filtering, isolation, protection and converter selection, laid out with the analogue constraints respected.
- Firmware acquisition. Sampling, timing, oversampling and digital filtering, matched to the analogue design.
- Verification. Measuring the noise floor and the real performance of the chain, so you know what your data is actually worth.
- Calibration approach. Practical and repeatable in production and in the field.
What you receive
A documented measurement approach, schematic and layout for the front end, acquisition firmware, and measured performance of the chain including noise floor and the limits of what can be resolved.
Related
Legacy machine retrofit, where external sensing is the whole job. Edge AI on microcontrollers, where signal quality caps model accuracy. IoT development for the system around it.
What are you trying to measure?
Tell us the quantity, the environment it sits in and the change you need to detect. If you already have data that looks wrong, that is a useful starting point too.
What clients ask before starting
Our data looks noisy. Can it be fixed in software?
Sometimes. Filtering can remove interference that sits outside the band of interest. What cannot be recovered is information that was never captured, for example where the sampling rate was too low for the phenomenon, so the honest answer depends on what the raw signal contains.
How do you decide the sampling rate?
From the phenomenon being measured rather than from convenience, with anti-alias filtering ahead of the converter. Sampling too slowly does not give a coarse version of the signal; it gives a misleading one.
Will the sensors need recalibrating?
That depends on the sensor, the required accuracy and the environment. The important thing is deciding early whether calibration happens in manufacturing, in the field or both, and how a unit is recalibrated years later by someone who did not build it.