Abstract We consider a multi-carrier joint processing (JP) coordinated multi-point (CoMP) cellular system wherein the beamformers are designed by a centralized controller (CC). Those are subsequently fed back to the respective base stations (BSs) via finite capacity backhaul link. Hence, we design transmit beamformers at the CC that associate users to BSs depending on their backhaul capacity and users channel state to address this problem. Upon identifying the active links, user multiplexing and scheduling over spatial and frequency dimensions are performed in conjunction with the beamformer design for all BSs subject to the objective of minimizing the number of backlogged packets at the CC. Since the beamformers are notified to the respective BSs along with the data, the backhaul utilization must include the overhead incurred by the quantized beamformers, which has higher precision as it is used by all data symbols transmitted over a coherence block. In this paper, we assume fixed quantization levels for each complex entry in the beamformers. Therefore, we select a subset of entries for each beamformer to reduce the overhead incurred by the active antennas, which when removed has the minimal impact on the user rate. With the above mentioned constraints and requirements, the beamformer design becomes a nonconvex combinatorial problem. Therefore, we provide a solution for the proposed formulation by employing successive convex approximation technique to handle the nonconvexity and the combinatorial search involving binary variables are replaced by a linear bounds along with a sparsity inducing term in the objective. The proposed method is useful only when the backhaul usage due to the beamformers is significantly larger when compared to the overhead incurred by the active users data. Numerical results are provided to illustrate the performance.