Opportunistic access to temporarily unused licensed frequency bands has been recognized as a cost-effective solution to mitigate the scarcity of radio spectrum incurred by the proliferation and evolution of wireless devices. In order to accomplish a more efficient spectrum utilization across time, space and frequency, the functionality of spectral awareness is expected to be incorporated into transceivers of cognitive radios (CR) in the foreseeable future. However, the agility requirements imposed on an individual CR might be very stringent since the spectrum has to be monitored on a continuous basis and the spectral holes must be detected reliably in the presence of channel fading and/or hardware failure, especially in a broadband wireless network with multiple primary users. To remedy this problem, light-weight collaboration among CRs becomes attractive since distributed signal processing can exploit the spatial diversity through network cooperation to improve the sensing reliability. To this end, we extend the classical framework for distributed detection of a single user to the case of multiuser detection, and introduce a Tanner graph description of the network topology. We designate the transmission pattern of primaries by a binary space-frequency signature sequence and view all possible patterns collectively as a low-density parity check (LDPC) codebook. Then, we invoke belief propagation message passing algorithm across the entire CR network to infer the unknown pattern iteratively from the noisy observations of all spectrum sensors. Based on the pattern inferred, the fusion centers can determine the spectral and spatial access pattern of secondary users, without interfering with the normal functioning of incumbent primaries.