Abstract-In order to mitigate spectrum scarcity, the cognitive radio (CR) paradigm has been invoked for improving the overall exploitation of the licensed spectrum by identifying and filling the free spectrum holes without degrading the transmission of primary users (PUs). Hence, we conceive a CR communication scheme, which enables a cognitive user (CU) to sense the activity of the PUs over a primary radio (PR) channel, which is exploited to transmit data using the modified Go-Back-N hybrid automatic repeat request (GBN-HARQ) protocol, when PR channel is free from the PUs. This arrangement is termed as the cognitive GBN-HARQ (CGBN-HARQ), where the activity of PUs on the PR channel is modelled as a two-state Markov chain having 'ON' and 'OFF' states. However, the CU may wrongly detect the 'ON'/'OFF' activity of the PUs in the channel, hence resulting in false-alarm or misdetection. Therefore, the two-state Markov chain is extended to four states by explicitly considering all the wrong sensing decisions. In this paper, we analytically modelled the CGBN-HARQ scheme with the aid of a Discrete Time Markov Chain (DTMC). Explicitly, an algorithm is developed for deriving all the legitimate states and for eliminating the illegitimate states, which assists us in reducing both the dimensionality of the state transition matrix and the associated computational complexity. Furthermore, based on DTMC modelling, we derive closed-form expressions for evaluating the throughput, the average packet delay and the end-to-end packet delay of CGBN-HARQ in realistic imperfect sensing environment. The results are also validated by our simulations. Our performance results demonstrate that both the achievable throughput andThe authors are with the