Message faulting is an increasing problem in 5G and future 6G due to network crowding, receiver motion, signal fading at higher frequencies, and greater phase-noise sensitivity. Disclosed herein are methods for analyzing waveform features of the received signal using artificial intelligence, and identifying the likely faulted message elements according to correlations of those waveform features. For example, after demodulating, the receiver can identify a subset of message elements that are all demodulated according to the same modulation level, and can measure a signal parameter for each message element in the subset. The processor can then average the deviations in the subset, and compare those message elements to the average for the subset. If one of the message elements shows an anomalously large deviation from the average, that message element is likely faulted.