Abstract:Signal integrity analysis for loopback links on high-speed digital automated test equipment (ATE) interface boards aims to achieve rapid prediction of wideband scattering parameters and end-to-end performance evaluation while ensuring physical compliance. Addressing the complex interconnect structure and vast parameter combinations of interface boards, the high-speed differential loopback link is physically partitioned into five modules: transition, stripline 1, Alternating Current coupling capacitor, stripline 2, and transition. Module-level deep learning surrogate models are constructed for each, with end-to-end channel response reconstructed through cascaded scattering parameter reconstruction. During model training, phase continuity representation and angular consistency loss functions were incorporated to enhance the stability and consistency of wideband phase prediction, achieving a global complex-form root mean square error (RMSE) of 0.024 5 at the module level. Building upon this, cascaded reconstruction of module-level four-port scattering parameters was completed, yielding an end-to-end complex-form prediction RMSE of 0.012 2. Comparison with time-domain simulation results shows that the predicted cascaded channel maintains consistent trends in primary reflection locations and reflection amplitude variations with the reference cascaded results in the time-domain reflectometer response. In eye diagram analysis under non-return-to-zero conditions, the eye width of the predicted results aligns closely with the reference results, while the relative deviation in eye height is controlled within 0.5%~5%. The results demonstrate that modular prediction and scattering parameter cascading method effectively preserves the characterization properties of critical discontinuities in the link. It can serve as a rapid evaluation tool in the early stages of interface board design, helping shorten simulation iteration cycles and assisting in structural scheme selection.