Most of today's data is time-series' from sensors, transaction systems, and production systems. However, much of this data is sensitive and consequently unusable. Generative models have shown promise in generating non-sensitive synthetic data, to share and drive applications with. However, current generative time-series's models are severely limited in their ability to capture the data distribution, limiting their usability. In this paper we propose a transformer-based diffusion model, TDDPM, for time-series which outperforms and scales substantially better than state-of-the-art. The focus is primarily on mobility data, such as trajectories of people's movement in cities, and we propose a conditional distribution approach which demonstrate out-of-distribution generalization to city-areas not trained on. We further propose a new comprehensive benchmark across several sequence lengths, standard datasets, and evaluation measures, considering key distribution properties.