When perfect channel state information (CSI) is available at the transmitter, adaptive coding and modulation (ACM) algorithms which maximize the goodput (GP) corresponding to the actual channel realization give rise to the maximum long-term average goodput. However, in realistic circumstances the available CSI is imperfect due to estimation errors and/or feedback delays, in which case the actual GP cannot be computed. In this paper we show that the long-term average goodput becomes maximum by optimizing the expectation of the goodput, conditioned on the available imperfect CSI. In order to make this expected goodput (EGP) tractable, we propose to apply an accurate modeling approximation, yielding the approximate EGP (AEGP). Next, we consider a cognitive communication system where a secondary-user (SU) orthogonal frequency division multiplexing (OFDM) system using bit-interleaved coded modulation (BICM) operates in the frequency band of a primary user (PU) network. Assuming only imperfect CSI is available, we derive various ACM algorithms (uniform and non-uniform bit allocation, uniform and non-uniform energy allocation) for the SU system which maximize the AEGP under a constraint on the interference caused to the PU network. We point out that, in spite of the imperfect nature of the available CSI, the resulting ACM algorithms achieve a signicantly larger goodput, compared to a non-adaptive selection of the transmission parameters.

Keywords:  Eective SNR mapping (ESM); Orthogonal frequency division multiplexing (OFDM); Adaptive coding and modulation (ACM); Imperfect channel state information; Goodput