Allowing an AC Coupled Sensor Channel to Settle After Startup

An AC coupled sensor channel can take much longer to produce a valid result than its amplifier takes to power up. A low high pass corner preserves slow signal content, but the associated time constants can also prolong recovery from bias changes. Define when measurements become usable rather than assuming the first conversion is representative.

Identify the charging paths

Draw the coupling capacitors, bias resistors, reference source, input protection, and amplifier stages. Determine the initial and final voltage across each capacitor during the intended power sequence.

The TI LMP91051 NDIR system report compares startup behavior of a DC coupled implementation and an AC coupled alternative, illustrating that a low frequency high pass network can dominate settling. Its measured times belong to the demonstrated circuits rather than every sensor channel.

Estimate a useful observation interval

For a simple first order response, the remaining step error falls exponentially with time constant RC. Reaching one percent of the initial error takes about 4.6 time constants. A hypothetical one second time constant therefore needs roughly 4.6 seconds for that particular relative error criterion.

Apply the criterion to the measurement requirement. One percent of a large startup excursion can still exceed the entire useful sensor signal. Use an absolute permitted error where that better represents the application.

Check saturation and abnormal recovery

Inspect intermediate outputs during startup. An amplifier that saturates may not follow the simple linear RC estimate until it returns to its operating range. Protection conduction or a changing bias reference can also alter the charging path.

Test brief power interruptions, sensor reconnection, and switching between measurement modes if those events occur in service. A fully discharged startup is only one possible initial condition. Residual capacitor charge can produce a different recovery trajectory.

Define the valid data condition

Capture sensor excitation, bias reference, relevant analog outputs, and converted data together. Compare the time when the waveform becomes repeatable with the actual error limit. If software discards initial samples, make that interval traceable to the measured requirement.

Save the initial conditions, RC assumptions, absolute error criterion, and measured settling over the relevant supply and temperature range. A faster startup modification should be checked for its effect on low frequency response and noise. The final interface should distinguish powered, converting, and settled states. This prevents a working converter from reporting a transient as a real sensor change simply because the analog channel has not yet reached its intended operating point.