In this paper, we present an improved detector for ACK/NACK message detection in the LTE uplink control channel with imperfect channel state information at the receiver. The detector is based on the generalized likelihood-ratio test (GLRT) paradigm. We derive detection metrics for the cases when the noise variances at the receiver are known and unknown. Noise here may comprise both thermal noise and interference. Simulation results show remarkable performance gains of the GLRT-based detector with unknown noise variances compared to the training-based maximum-likelihood detector with unknown noise variances when the noise variances in two slots are different. Furthermore, the performance of the GLRT-based detector with unknown noise variances is nearly the same as that of the training-based maximum-likelihood detector with known noise variances.