A DL-based High-Precision Positioning Method in Challenging Urban Scenarios for B5G CCUAVs

Unmanned aerial vehicles (UAVs) facilitate services in civilian and industrial fields but suffer from a limited direct link operating range and unreliable satellite positioning in urban canyons. Fortunately, cellular-connected UAVs (CCUAVs) overcome these shortcomings, benefitting from the beyond 5th generation (B5G) network's city-level coverage and high-precision positioning capabilities , and are considered a paradigm of 5G-advanced and beyond. However, in a challenging airspace (e.g., urban canyon), the CCUAV localization accuracy deteriorates due to low signal-to-interference-plus-noise (SINR) air-ground channels and strong multipath effects . To solve these problems, we first construct channel amplitude-phase response (CAPR) images to characterize the cellular channel in a challenging airspace for CCUAV positioning. In particular, the effect of down-tilted antennas and high-dimensional channel features are embedded into CAPR images, to meet the relevant cellular communication criteria. Subsequently, a deep learning (DL) model, the scale-shared quarter network (SSQ-Net), is devised for CAPR image-based positioning, along with a robustness enhancement method. With this method, the multipath effects and interference in challenging environments are exploited to improve positioning accuracy and robustness, instead of being treated as detriments. Finally, the experimental results in a typical urban canyon show that our method outperforms state-of-the-art methods in terms of accuracy and robustness.