In this paper, we propose a new decentralized control for heavy haul trains by explicitly considering input constraints and time varying communication time delays. Through introducing a set of state and input transformations, the complicated longitudinal dynamical model of the heavy haul train is first converted into the double integrator form for ease of control design. Then, a decentralized cooperative control is designed to regulate the individual car's speed to the desired speed profile in order to achieve the objective of maintaining the minimum in-train forces for the safety consideration. It is rigorously proved that the closed loop system is asymptotically cooperative stable under the least communication conditions among individual cars in the heavy haul train system. Extensive simulations are conducted to validate the performance of the proposed new design.