Table of Contents
Low offset drift does not guarantee a quiet digitized signal. A zero-drift amplifier can correct slow DC errors while its internal switching interacts with source impedance and the sampling system.
The diagnostic task is to identify how the unwanted energy reaches the measurement band. A source short, a filter change and a sample-rate change answer different questions; using them in a controlled sequence is more useful than immediately buying a different amplifier.
How does source impedance turn internal switching into an error?
Charge injection associated with internal switching can create voltage disturbances across finite source impedance. The magnitude and balance of the impedances at both inputs affect the resulting artifact.
Analog Devices’ zero-drift amplifier explanation describes the switching, charge-injection and filtering mechanisms. Its detailed examples are device-specific; do not copy one amplifier’s switching frequency or noise peak into an AD8628 design.
Test first with the intended source or a faithful electrical equivalent, including capacitance and resistance imbalance. Then reduce the source impedance while preserving common-mode voltage and gain. If the disturbance changes, the result points to an interaction with the source rather than proving that the amplifier is defective.
A shorted-input noise trace answers a narrower question. It can establish the amplifier and measurement floor under that setup, but it does not represent a high-impedance sensor connected through a cable.
Why can a high-frequency artifact appear at low frequency?
Sampling can fold out-of-band energy into the signal band. For a discrete tone, the observed alias can be estimated by choosing an integer k that places |fartifact − k × fsample| inside the sampled baseband.
If a 101 kHz artifact is sampled at 20 kS/s, it can appear at 1 kHz because |101 − 5 × 20| = 1 kHz. Changing the sample rate to 21 kS/s moves that candidate alias to 4 kHz. This diagnostic example does not specify an AD8628 switching frequency.
Real switching can include harmonics, modulation and frequency variation, so one tidy tone is not guaranteed. Nevertheless, a peak that moves predictably when sample rate changes is useful evidence. Keep the analog input and amplifier configuration fixed during that comparison.
A digital filter after sampling cannot distinguish an aliased artifact from a genuine in-band signal. Suppression must begin before the alias occurs or through a validated change to the source, amplifier or sampling arrangement.
Choose a filter with the signal and converter load in view
An analog low-pass filter can reduce unwanted high-frequency energy, but its impedance and settling must remain compatible with the desired signal and ADC input. A lower cutoff is useful only if the application can tolerate its bandwidth and response time.
Use a small set of controlled comparisons:
| Trial | Hold constant | What the result can reveal |
|---|---|---|
| Lower source impedance | Gain, common mode, sample rate | Source/switching interaction |
| Balance input impedances | Signal level and bandwidth | Sensitivity to impedance mismatch |
| Add or change analog filtering | ADC rate and input waveform | Out-of-band energy reaching the sampler |
| Change sampling rate | Analog circuit | A possible alias relationship |
| Substitute a measurement source | Acquisition setup | Sensor or cable contribution |
Record bandwidth and windowing with any spectral plot. Comparing two traces with different acquisition bandwidths can make a device look quieter without improving the actual application.
When does the result justify a different amplifier?
Change architecture when the required source impedance, bandwidth and sampling conditions cannot meet the error allocation with a stable, practical circuit. The decision should follow the mechanism identified by the trials, not the word “zero-drift” on a product summary.
For a candidate AD8628 or another AD862x option, return to the AD8620/AD8622/AD8628 selection comparison for supply, channel and source constraints. Then apply the op-amp replacement validation process to the actual feedback and load network.
Save the accepted noise band, input impedance, sample rate, filter values and source configuration with the approved BOM. Those fields explain why an otherwise plausible alternate requires a new measurement, and prevent a later “equivalent” resistor or capacitor change from invalidating the original result.
Frequently Asked Questions (FAQ)
Will a DC SPICE model show chopping-related artifacts?
Not necessarily. Check the model documentation for switching, noise and overload behavior. A model that reproduces DC offset can omit the artifacts relevant to a sampled circuit.
Is every spectral peak caused by the zero-drift amplifier?
No. Power converters, digital clocks, mains pickup and the acquisition system can produce peaks. Use controlled changes and a measurement baseline to identify the source.
Can two samples of the same amplifier show different artifact levels?
Yes, unit and operating-condition differences can affect switching behavior. Qualification should use an application-appropriate sample and corner plan rather than treating one quiet device as a universal result.