State estimator is the foundation of system awareness as well as a variety of other widely deployed applications. Indeed, the deployment of phasor measurement units (PMU) has highly advanced the accuracy and granularity of state estimators. However, deploying PMUs to all buses is neither physically necessary nor financially viable for granting full observability of the system. Consequently, by leveraging system observability and capital costs of PMUs, PMUs could be optimally placed in a system that would minimize total investment costs while also satisfying observability requirements. This paper proposes a novel integer linear programming (ILP) based optimal PMU placement (OPP) model, in which detailed PMU installation costs are considered. More importantly, different from existing OPP models that are based on network connectivity matrix, the proposed model defines bus observability indicators to explicitly describe how individual buses achieve their observability, namely propagation of the observability from one bus to another. In addition, contributions of zero injection buses for providing extra observability and consequently reducing total investment cost are accurately formulated via carefully crafted constraints. Numerical studies illustrate effects of the proposed approach.