INL testing requires a very low-noise source to avoid introducing errors into the measurement. An 8.5-digit DMM is required to accurately measure the INL because this is a highly linear signal chain. The DMM measures the signal chain input, and the GUI measures the ADC output. The output and scaled input are compared. Measurements are taken along the full input range.

The INL can be less than the noise floor because the measurement signal chain is designed to have low nonlinearity. Averaging N samples reduces the thermal noise by a factor of √N. With sufficient averaging, the INL can be detected and measured. Each measured point is an average of 1024 measurements.

Figure 4-9 shows that the measured INL for the reference design is approximately 1.1ppm for the 10V range. TIDA-010970 is not a full DMM design, so a full DMM design includes other components in the signal chain that add nonlinearities, such as input protection. However, TIDA-010970 includes the most critical components of the signal measurement path.

Figure 4-9 TIDA-010970 Linearity Error vs Input Full-Scale Range

Calculate INL using a three-endpoint calibration. This method requires three measurements: zero, positive full-scale, and negative full-scale. Calculate two sets of coefficients from these three measurements: one set provides a linear correction for the negative range; the second provides a linear correction for the positive range. These linear corrections create a piece-wise linear fit. Then, the INL is calculated by comparing the actual measured output with the expected calibrated piece-wise linear output. In theory, more than three points can be measured to create a piece-wise function with more than two sets of coefficients. However, this system has been designed for low linearity such that two sets of coefficients are sufficient.