This work addresses scheduling in a cognitive radio scenario where a minimum throughput for the downlink primary network (PN) is guaranteed to each user with an associated violation probability (probability of not obtaining the guaranteed throughput). The primary network is surrounded by multiple downlink secondary networks, each aiming to maximize its network throughput. Scheduling in PN is performed independent of the secondary networks. Some information about the PN is available at the central scheduler that is responsible for scheduling the secondary networks. The contribution of this work is to apply a novel scheduler to the PN which is more robust to QoS degradations resulting from the secondary networks than other state of the art schedulers. This is validated by numerical simulations of the cognitive radio network.