In the model example "ADC IBIS-AMI Model Based on COM", the convergence values in the statistical phase model and in the time domain model are calculated differently. The statistical phase model uses a "Hula-Hoop Alexander" search, while the time domain model uses "Early-Late Alexander" phase detection.
The statistical (SerDes "AMI_Init") model is a fast approximation of the time domain (SerDes "GetWave") behavior and therefore has several limitations:
1. "AMI_Init" does not handle non-linear systems with accuracy.
2. "AMI_Init" can only approximate the time-varying convergence behavior of the clock recovery and equalization adaptation. For systems with several control loops, determining the convergence point in "AMI_Init" is difficult.
3. "AMI_Init" primarily returns a modified impulse response, which the simulator will use to create an eye diagram and then derive eye metrics from. An Analogue-Digital-Converter (ADC) doesn't have such a single probe point which a comparable eye diagram could be derived from; the eye diagram can therefore be considered "virtual" and only a rough estimation of the system behavior.
These limitations explain why there is a difference in the convergence values calculated in the statistical model and the time domain model.
A Signal-to-noise (SNR) metric as calculated by the "GetWave" receiver model is a more accurate model of system behavior and can be output as a custom AMI parameter. However, such an SNR output is not standardized as part of the IBIS spec.
