Abstract-We address the problem of rapidly discovering spectrum opportunities for seamless service provisioning in cognitive radio networks (CRNs). In particular, we focus on multichannel communications via channel-bonding with heterogeneous channel characteristics of ON/OFF patterns, sensing time, and channel capacity. Using dynamic programming (DP), we derive an optimal online sensing sequence incurring a minimal opportunity-discovery delay, and propose a suboptimal sequence that presents a nearoptimal performance while incurring significantly less computational overhead than the DP algorithm. To facilitate fast opportunity discovery, we also propose a channel-management strategy that maintains a list of backup channels to be used at building the optimal sequence. A hybrid of maximum likelihood (ML) and Bayesian inference is introduced as well for flexible estimation of ON/OFF channel-usage patterns, which selectively chooses the better between the two according to the frequency of sensing and ON/OFF durations. The performance of the proposed schemes, in terms of the opportunity-discovery delay, is evaluated via in-depth simulation, and for the scenarios we considered, the proposed suboptimal sequence achieves a near-optimal performance with only an average of 0.5 percent difference from the optimal delay, and outperforms the previously proposed probabilistic scheme by up to 50.1 percent. In addition, the backup channel update scheme outperforms the no-update case by up to 49.9 percent.