A Systematic Approach

ADC accuracy problems usually appear as noise, drift or an offset in the reading, and they are easy to misdiagnose because the converter is rarely the cause. This article presents a systematic procedure for diagnosing TI ADC measurement chains, using the ADS1115 and OPA333 as references, so that the cause is found by measurement rather than by guessing.

Step 1: Isolate the Source

Before changing anything, short the converter's input to a clean reference and read the result. If the reading is stable and near the expected value, the converter, reference and supply are sound, and the problem is in the signal source or the conditioning. If the reading is still noisy or offset, the problem is in the converter's reference, supply or layout. This single test divides the problem space in half.

Measuring the Noise Floor

Record the peak-to-peak spread of the reading with the input shorted, and compare it with the converter's expected noise. A spread much larger than the datasheet figure points to the reference, supply or layout rather than to the signal.

Step 2: Check the Reference

The converter compares the input to the reference, so any reference error appears directly in the result. Confirm that the reference is stable, low-noise and well decoupled, and that it is not shared with a noisy load. Where the converter uses its internal reference, confirm that the supply feeding it is clean, because the internal reference derives from the supply. In practice, the reference limits accuracy more often than the converter does.

Reference Decoupling

Place the reference decoupling close to the reference pin, and keep the reference trace away from digital switching. A reference that is shared with a digital circuit will carry that circuit's noise into every reading.

Step 3: Inspect the Supply and Layout

The analog supply must be clean, and the layout must keep the analog input away from digital noise. Decouple the supply close to the converter, use a solid ground reference beneath the analog path, and separate the analog supply from the digital rail where possible. Keep the input trace short and free of stubs, and route it away from clocks and switching regulators.

Grounding

Use a solid ground plane and avoid splitting the ground return under the analog path. Digital switching currents flowing through the analog ground appear directly as noise in the reading, which is one of the most common causes of a noisy measurement.

Step 4: Check the Conditioning

If the converter and its reference are sound, the problem is in the signal or the conditioning. Confirm that the amplifier's offset and noise are within budget, and that the gain is set correctly. A zero-drift amplifier such as the OPA333 removes offset and drift from the chain, which is valuable when the signal is small.

When to Escalate

If the reading is still inaccurate after the reference, supply, layout and conditioning are corrected, escalate with the measured data: the noise floor with a shorted input, the reference stability and the readings across temperature. BeiLuo's FAE team can review the design and characterize the measurement chain in our analog lab so the cause is identified before a board respin.