The Unglamorous Parts of a Signal Chain: Between the Sensor and the ADC
Explainers · Published 2026-09-04 · Updated 2026-09-04
System-level accuracy is often decided by op-amps, references and isolators rather than the main processor — here is how to split the chain and rank the risk.
When a product misses its accuracy target, the processor is almost never the culprit. The usual suspects sit quietly between the sensor and the ADC: an op-amp offset that grows with temperature, a reference whose drift sets the floor for the whole chain, an isolation barrier checked for safety but not for common-mode noise. The right way to read a signal chain is segment by segment — conditioning, excitation and reference, conversion, isolation, then the trip into the main controller — assigning risk to each segment instead of blending every spec into one number. This article does two things: names the "looks fine, fails later" trap in every segment, and tells you which BOM lines should be marked non-substitutable.
Front-end conditioning: errors here cannot be rescued downstream
Sensors hand you a weak signal riding on common-mode voltage, and the first stage — op-amps, instrumentation amplifiers, programmable gain — squeezes it into the ADC's input range. Two classic traps. Instrumentation-amp gain is set by external resistors, so resistor drift and mismatch become gain error directly. And offset and drift matter most on the smallest signals, while prototype validation tends to poke at the largest ones — so the "verified" path is one the product will rarely travel. The benchmark parts here are ADI's AD-series op-amps and Texas Instruments' op-amps and interface parts, catalogued as high-accuracy signal-chain reference products; Microchip's MCP analog line covers the long tail of general-purpose needs. Scaling into range is the pass mark; controlling drift and common mode is the actual job.
Excitation and reference: drift sets the floor
Bridges and RTDs need excitation; the converter needs a reference. Reference drift and long-term stability define the accuracy floor of the entire chain — above that floor, more front-end gain just relocates the error. The most common shortcut is treating the ADC's internal reference as the measurement reference: convenient, but its initial tolerance and typically worse drift make it the wrong foundation. The second trap is supply noise arriving straight at the reference and the front end, which is why low-noise rails matter here; TI's TPS power management and ADI's LT power management earn their spots on this segment for exactly that reason.
Conversion: beyond resolution, sampling and drivers bite
Everyone starts an ADC selection at bit count, but architecture decides the job: SAR for medium-speed precise sampling, sigma-delta for slow high-resolution work, pipeline for speed — none of them eats the others' lunch. What actually fails on the real board is subtler. If the signal does not settle within the sampling window, you get nonlinearity that shifts with input conditions. If the driver op-amp cannot handle the ADC's switched-capacitor load, readings get modulated by the driver's output impedance. In real-time control the chain is short and unforgiving, and a large part of why TI's C2000 remains the real-time control MCU benchmark is how its on-chip sampling and conversion work together. For the high-accuracy conversion itself, the catalogue entry is ADI's ADC / DAC family.
Isolation: safety distance and common-mode rejection are two ledgers
Between power stage and control stage sit digital isolators; gate drivers get isolated drivers. Two books get mixed into one far too often: insulation and creepage belong to safety certification, while common-mode rejection decides how much noise from fast-switching nodes floods your ground. A substitute that matches one ledger and ignores the other is the textbook paper-swap accident. In our database, ADI's digital isolators are the incumbent mainline, while Novosense's NSI digital isolators and isolated drivers lead the domestic supply, its sensor signal conditioning covers pressure and temperature chains, and its automotive analog parts ship on AEC-Q100 platforms.
The last segment: it detonates at EMC testing
Level shifters, bus transceivers, ground loops — the run into the main controller usually fails jointly with EMC. Everything is polite on the bench, then bit errors appear once the whole unit enters certification. ST's STSPIN motor drivers, which fold power stage and interface into one part, are a reminder that the more segments you integrate, the more carefully you must check each segment's boundary in the documentation. Flattening the five segments:
| Segment | Typical failure | Basis for prioritising |
|---|---|---|
| Front-end conditioning | Offset and drift amplified on small signals | Accept on the weakest signal, not the strongest |
| Excitation and reference | Internal ADC reference used as the measurement reference | Reference drift sets the chain's floor |
| Conversion and driver | Insufficient settling, driver mismatch | Judge by measured board nonlinearity, not bit count |
| Isolation | Creepage checked, common mode skipped | Safety and immunity are separate ledgers |
| Interface into the MCU | Only EMC testing exposes it | Close on certification results, not paper timing |
Substitution, in the right order
The same table gives the domestic-substitution sequence. Isolators and sensor signal conditioning — clear structure, weak ecosystem coupling — substitute well today. Precision references and high-resolution converters move slowly, and the delay is in models, evaluation material and validation records, not in wafers.
Notes for buyers
- Mark references, isolated drivers and precision op-amps as non-substitutable BOM positions; changing them requires full evaluation plus whole-unit retest.
- Compare substitutes beyond the first page of the datasheet: leakage current, settling time and driver stability belong on the same comparison table — they are the real constraints at segment boundaries.
- Validate segments on evaluation boards and the chain on the finished board; ADC nonlinearity and isolation common-mode margin must be measured under whole-unit conditions.
- Sequence substitutions by ecosystem coupling: isolators and conditioning first; references and high-resolution converters only after documentation and validation catch up. The analog and interface category directory lists vendors whose design-material depth you can check first.
Data caveats
Product lines and positioning statements come from the ICNAV database snapshot dated 2026-10-28, including TI as the top analog and power supplier worldwide, ADI as second in high-performance analog and signal chain, and Novosense as the domestic isolation and sensor-signal-chain vendor. No accuracy figures, test data or prices are quoted; all qualitative claims must be re-checked against datasheets and board-level measurement. This is editorial analysis, not procurement advice.
Written by ICNAV 编辑部 · ICNAV editorial team. References: the ICNAV company directory and public industry sources.