Title:

AI-Based Signal Analyzer & Equalizer

This research addresses the challenges of signal degradation in high-speed PCB environments (2.5 GSPS) caused by impedance mismatches and parasitic effects. To replace expensive traditional instrumentation, a Multi-Task Learning (MTL) neural network was developed using a shared ResNet backbone to simultaneously analyze signal health and reconstruct distorted waveforms. The system was trained on automated datasets from Keysight ADS and validated using a physical Bit Error Rate Tester (BERT) and oscilloscope setup. Results indicate that the AI Equalizer successfully recovered distorted signals, achieving a Normalized Mean Square Error (NMSE) of less than 5%. However, the Signal Analyzer requires further refinement, as it currently falls short of the 5 MAE (Mean Absolute Error) accuracy target for impedance prediction. Future work will focus on optimizing the model for FPGA integration to enable real-time, self-correcting signal integrity management.

Name

Comment*

Comments are viewable only by submitter

Captcha*