Coded Modulation with Mismatched CSIT over Block-Fading Channels

Reliable communication over delay-constrained block-fading channels with discrete inputs and mismatched (imperfect) channel state information at the transmitter (CSIT) is studied. The CSIT mismatch is modeled as Gaussian random variables, whose variances decay as a power of the signal-to-noise ratio (SNR). A special focus is placed on the large-SNR decay of the outage probability when power control with long-term power constraints is used. Without explicitly characterizing the corresponding power allocation algorithms, we derive the outage exponent as a function of the system parameters, including the CSIT noise variance exponent and the exponent of the peak power constraint. It is shown that CSIT, even if noisy, is always beneficial and leads to important gains in terms of exponents. It is also shown that when multidimensional rotations or precoders are used at the transmitter, further exponent gains can be attained, but at the expense of larger decoding complexity.