Closed-loop link adaptation in LTE involves a family of modulation and coding schemes constructed out of powerful turbo codes. Due to their particular structure, such codes cannot be optimally decoded except for very limited block length. In practice, a suboptimal iterative decoding is applied. The smooth introduction of linear minimum mean-square error interference cancellation (LMMSE-IC) based turbo equalization in LTE calls for a new physical layer abstraction for this non-trivial doubly iterative algorithm to accurately predict the performance of each global iteration. The abstraction is based on a stochastic modeling of the whole turbo equalization using EXIT charts (and variants). As the core of the contribution, we find that, even in the simplified case of Gray mapping, a bivariate information transfer function is needed to characterize the evolution of the joint demapper and turbo decoder embedded within the LMMSE-IC based turbo equalization. This is in contrast with previous contributions where simple convolutional codes were considered and univariate information transfer function sufficient. The effectiveness of the approach is demonstrated through Monte Carlo simulations in a variety of transmission scenarios.