Tensions in Practice: Common-offset rejection in tension with another noisy read¶
Paired sensor states · additive error model
A sensor’s reference state is 0 and its signal state is 10. Reading only the signal is simpler but includes its offset. Subtracting a reference read cancels an offset that is identical in both. The three rows separate that benefit from two limits: independent read errors enter the difference twice, and an offset that changes between reads does not fully cancel.
Avoid another read and its noise
Use one signal reading when the reference is known and offsets are negligible or separately controlled.
Remove a shared additive offset
Difference paired states so their identical unwanted component cancels.
Why these aims pull against each other
Subtraction rejects what the readings share. It also retains the difference between what they do not share, including read noise and offset drift.
Choose an arrangement to see what changes and what remains difficult.
Finite illustrative comparisons. Text states carry the meaning; color is not a measured score or universal preference.
What this choice protects
What it costs
When it fits
Compare the arrangements
Signal only
Report the signal reading, with the true reference level stipulated as zero.
| Reference | Signal | Report | |
|---|---|---|---|
| Shared offset | Not read | 13 | 13 |
| Read errors | Not read | 9 | 9 |
| Offset drifts | Not read | 14 | 14 |
- What it protects
- No reference acquisition, settling delay or reference-read noise is added.
- What it costs
- A signal-side offset is included in the report: 13 instead of 10 in the shared-offset row.
- When it fits
- Offset is negligible, calibrated separately, or less troublesome than the extra read and its uncertainty.
Illustration note: Shared-offset row: offset 3, no read errors. Read-errors row: no offset, signal error −1. Drift row: the signal offset is 4. These are illustrative settings, not samples used to estimate a noise distribution.
Paired difference
Report signal minus reference, using the paired readout states.
| Reference | Signal | Report | |
|---|---|---|---|
| Shared offset | 3 | 13 | 10 |
| Read errors | 1 | 9 | 8 |
| Offset drifts | 3 | 14 | 11 |
- What it protects
- The common offset 3 cancels exactly: 13 − 3 = 10.
- What it costs
- With reference error +1 and signal error −1, differencing gives 9 − 1 = 8. With offset drifting from 3 to 4, it gives 14 − 3 = 11.
- When it fits
- A sufficiently common offset dominates, the signal has settled, and reference acquisition is feasible.
Illustration note: The read-errors row has zero offset and independently specified read errors. For independent zero-mean errors of equal variance, differencing adds their variances; this row alone is not a statistical estimate.
What this illustration does—and does not—establish
Correlated Double Sampling: Correlated Suppression versus Uncorrelated Read Noise and its core define the shared-error requirement. The arithmetic isolates common error, independent read error and drift.
- The desired change is held at 10; each row is a different declared error setting, not a measured sequence.
- The method does not cancel arbitrary noise, drift or a signal that has not settled.
- Averaging the readings would answer a different question and would not perform this shared-offset cancellation.
Source entries
Correlated Double Sampling
Correlated Double Sampling: Correlated Suppression versus Uncorrelated Read Noise supplies the conflict examined here.
Correlated Suppression versus Uncorrelated Read Noise
Subtraction removes a common term but carries independent noise from both samples into the difference.
Core Idea
Correlated double sampling (CDS) is an electrical measurement method that pairs a reference reading with a signal reading and subtracts the two. If an unwanted offset or reset-noise component is substantially the same in both, it cancels in the difference while the desired change remains. The word correlated names this shared-error condition, not merely the fact that there are two samples.