Wideband compressed sensing for cognitive radios

Efficient utilization of spectral bands in a dynamic environment with a continuously changing occupation rate is challenging. Static spectral allocations preclude the use of unoccupied spectrum, unless the spectrum manager has released the allocation to another mission. Standard spectral sensing techniques employ swept narrowband receivers. While effective in creating a composite, these techniques are ineffective at identifying short duration signals. Real-time spectral analysis techniques are effective at capturing short duration transmissions, but usually have narrow band capabilities, limited dynamic range and are relatively expensive. Timely accurate sensing of wide spectral band using traditional spectral estimations at the Nyquist rate (or higher) is another challenge due to the high data rate. Compressed sensing (CS) techniques utilize signal-to-information rate processing when signals are sparse in a specific domain. Dynamic spectrum access (often considered a critical component of a cognitive radio (CR)) can reutilize temporally unoccupied spectrum. An accurate estimate of the current state of spectral occupancy is critical to the autonomous decision processes involved in dynamic spectrum access. From this perspective, CS is being studied as an enhancement to dynamic spectrum access strategies. This study addresses feasibility issues for the development of autonomous CS-CR systems that are capable of performing spectrum sensing and recovery without a priori information about the spectral occupancy. Since recovery requires computationally intense non-linear optimization, we perform a single platform trade study on CS methodologies (BP-PD, BP-SPG, ROMP, Edge Detection, and Sequential Recovery) for efficient wideband recovery relative to execution time and reconstruction error. We propose adaptive coarse detection, and exact recovery based on adaptive edge detection. We present the results using continuous waves and pulses (MATLAB or GRC generated, or USRP me- sured), and show favorable conditions for BP-PD, BP-SPG, ROMP, adaptive recovery performance, and discuss potential for application of CS in a CR architecture.