Methods, apparatus, computer program products and other embodiments model a system's channel and noise characteristics using one or more Hidden Markov Models (HMMs) by modeling the state dynamics for one or more lines and associated profiles and by learning from the line history. Besides channel and noise characteristics, user data activity also can be modeled using an HMM, where all channel, noise and user data activity can be modeled jointly using HMMs as long as the joint states are properly defined. In multi-line systems, multiple lines (for example, a set of lines in a DSL system binder) may be modeled jointly. After learning channel, noise, user data activity and/or other dynamics using HMM techniques and methods, system operation and/or configuration parameters can be controlled based on the estimated HMM, providing flexible solutions for virtually any channel, noise, user data activity and/or other dynamics.