In this paper, we address the dynamic spectrum allocation problem in next generation cellular networks under the coordinated dynamic spectrum access (CDSA) model. Firstly, considering spectrum demands of base stations, we formulate spectrum allocation under physical interference models as a nonlinear optimization problem. Then, we propose a demand-driven dynamic spectrum allocation algorithm by transforming this nonlinear optimization problem into linear programming problem. The proposed algorithm first computes initial spectrum allocation and then applies iterative enhancement. Detailed simulations show that the proposed algorithm outperforms the existing algorithms in effective spectrum utilization efficiency and in average satisfaction degree.