A traffic matrix is a necessary parameter fornetwork management functions,and itsupplies a flow-level view of a largescale IP-over-WDM backbone network.This paper studies the problem of traffic matrix estimationand proposes an exact traffic matrix estimation approach based on network tomography techniques.The traditional network tomography model is extended to make it compatible with compressive sensing constraints.First,a stochastic perturbation is introduced in the traditional network tomography inference model.Then,an algorithm is proposed to achieve additional optical link observations via optical bypass techniques.The obtained optical link observations are used as extensions for the perturbed network tomography model to ensure that the synthetic model can meetcompressive sensing constraints.Finally,the traffic matrix is estimated from the synthetic model by means of a compressive sensing recovery algorithm. A traffic matrix is ​​a necessary parameter for network management functions, and its applications. A flow-level view of a large scale IP-over-WDM backbone network. This paper studies the problem of traffic matrix estimation and proposes an exact traffic matrix estimation approach based on network tomography techniques The traditional network tomography model is extended to make it compatible with compressive sensing constraints. First, a stochastic perturbation is introduced in the traditional network tomography inference model. Chen, an algorithm is proposed to achieve additional optical link observations via optical bypass techniques. obtained optical link observations are used as extensions for the perturbed network tomography model to ensure that the synthetic model can meet thecompressive sensing constraints.Finally, the traffic matrix is ​​estimated from the synthetic model by means of a compressive sensing recovery algorithm.