Trajectory representation learning (TRL) aims to map raw trajectories to d-dimensional vectors, which has been extensively studied on GPS-based trajectory and check-in sequences. The task of Cellular Trajectory Representation Learning (CTRL) is a new problem that focuses on analyzing trajectories derived from cellular-based data. To this end, we extend the task of TRL onto cellular trajectory by developing an enhanced pre-training model. Apart from the inherent sequential features involved in trajectories that have been widely exploited in existing work, the temporal characteristics and cellular-based network topological information are fully explored for better TRL in this paper. Specifically, the proposed method consists of two main components. Firstly, we build a time-aware encoding layer to incorporate the representations of travel preference and temporal periodic patterns in a cellular trajectory. Then, we build spatio-temporal dependencies encoding layer based on an enhanced pre-trained model to learn the representations of long-term dependencies, meanwhile incorporating spatial correlations with graph convolution networks. The experiments conducted on two real-world datasets demonstrate the effectiveness of our proposed model.