The MSPM0G3507 only supports fixed-point math, and thus some consideration of number overflow is required to minimize error and/or loss of precision. The current code is optimized to maintain precision and accuracy where it matters most, which is near the fault and warning thresholds. The user can avoid this challenge by switching to an MCU with floating-point math support.

Component tolerances are specifically chosen to meet the design target accuracy. Some passives such as the RES60A resistors have a maximum absolute tolerance of 15%, affecting measurement accuracy when the Riso is in the MΩ range. However, this is not important since the highest accuracy closer to the fault trip point of 100Ω/V is desired. For a 1kV bus, that is 100kΩ. At this level, the RES60A tolerance is not significant for a 5% accuracy target. However, the equivalent resistance reading must be compensated for the paralleling of RES60A and other resistances.

So far, the IMD is assumed to work under exceptional conditions without noise, providing a good place to start understanding the fundamental concepts. In practice, TIDA-010985 performance is heavily dependent on how well noise is controlled. These are the steps to help mitigate noise for this design:

• HW filtering to limit input noise
• Follow PCB layout guidelines to optimize SNR
• ADC HW averaging (default is set to 128)
• SW averaging of multiple ADC readings or predictions to estimate the steady-state voltage. For non-prediction-based IMD calculations, the SW averages 50 adjacent ADC samples to estimate the settled voltage.
• For prediction-based IMD calculations, the SW averages 330 predictions to estimate the settled voltage (i.e. Vp1, Vn1, Vp2, Vn2). Longer time constants create flatter voltage settling curves and decrease prediction accuracy under noisy conditions. As mentioned previously, increasing the time spacing between the three samples can help improve prediction performance.