Limits of cache-aided wireless BC: Interplay between coded-caching and CSIT feedback

Building on the recent breakthrough by Maddah-Ali and Niesen that revealed substantial gains due to coded caching in single-link (wired) broadcast communications to K identically-linked users, the work here considers the K-user cache-aided wireless multi-antenna broadcast channel (BC) with random fading and imperfect feedback, and analyzes the throughput performance as a function of feedback statistics and cache size, often identifying the optimal cache-aided degrees-of-freedom (DoF) performance within a factor of 2. In our setting where a single transmitter communicates -- us-

ing non-timely and imperfect-quality channel state information (CSIT)--to K independent users with pre-filled caches, the work identifies near-optimal schemes that combine data caching, folding and precoding, to efficiently utilize caching and feedback resources. Interestingly in the large K setting, where the schemes are often DoF optimal, the derived limits reveal the surprising fact that full (perfect) CSIT can be completely substituted (without performance losses) by combining a vanishingly small portion of delayed CSIT, with a vanishingly small fraction of the files content per user's cache. The key lies in finding the right balance between cache-induced gains of multicasting common information, and CSIT-induced gains of broadcasting private information. It also builds on the retrospective nature shared by both coded caching and (communicating with) non-timely feedback, where in both cases the transmitter -which has timely knowledge of the information content -must act retrospectively to compensate for not knowing the 'destination' (channel and user identity) of this content.