The IEEE 802.22 is a new cognitive radio standard that is aimed at extending wireless outreach to rural areas. Known as wireless regional area networks, and designed based on the not-to-interfere spectrum sharing model, WRANs are channelized and centrally-controlled networks working on the under-utilized UHF/VHF TV bands to establish communication with remote users, so-called customer premises equipment (CPEs). Despite the importance of reliable and interference-free operation in these frequencies, spectrum sensing fusion mechanisms suggested in IEEE 802.22 are rudimentary and fail to satisfy the stringent mandated sensing requirements. Other deep-rooted shortcomings are performance non-uniformity over different signal-to-noise-ratio regimes, unbalanced performance, instability and lack of flexibility. Inspired by these observations, in this paper we propose a distributed spectrum sensing technique for WRANs, named multi-channel learning-based distributed sensing fusion mechanism (MC-LDS). MC-LDS is demonstrated to be self-trained, stable and to compensate for fault reports through its inherent reward-penalty approach. Moreover, MC-LDS exhibits a better uniform performance in all traffic regimes, is fair (reduces the false-alarm/misdetection gap), adjustable (works with several degrees of freedom) and bandwidth efficient (opens transmission opportunities for more CPEs). Simulation results and comparisons unanimously corroborate that MC-LDS outperforms IEEE 802.22 recommended algorithms, i.e., the AND, OR and VOTING rules.