Unsupervised deep learning to solve power allocation problems in cognitive relay networks

Abstract

HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Robustness to imperfect CSI of power allocation policies in cognitive relay networks

“…Based on the above, we consider a relay-aided cognitive radio network, as in [1], [7], and study the maximization of the opportunistic rate when the relay performs either Decodeand-Forward (DF) or Compress-and-Forward (CF) in a fullduplex manner, while ensuring a predefined primary Quality of Service (QoS) constraint. Because of the non-linear and complex operations performed at the relay, the resulting resource allocation problems for cooperative cognitive networks are non-convex ones and cannot be solved in closed-form in general [7].…”

“…As opposed to our previous investigations [1], [7], which rely on a perfect and global channel state information (CSI), our main objective in this paper is to relax this assumption. Indeed, perfect CSI can be particularly difficult to obtain in cognitive networks, e.g., when estimating the channels from the secondary to the primary network, the full cooperation of the primary network may not be granted.…”

“…To the best of our knowledge, except for our previous study in [7], DNN-based techniques have not been used for resource allocation problems in cooperative cognitive networks. Compared to [7], in which we only focused on DF relaying, in this paper, we extend our investigation to include CF relaying, for which we derive the analytical solution under perfect CSI. We then shift gears and investigate the robustness of our solutions, either in closed form for CF or via deep learning for DF and CF, to imperfect CSI.…”

“…Main contributions: In this paper, we derive the optimal power allocation policy for full-duplex CF relaying under individual power constraints and a primary QoS constraint, assuming perfect and global CSI. Then, we investigate the robustness to imperfect CSI of this solution coupled with that of our existing DNN solution for DF in [7], which we exploit here also for CF. The training of the DNN is performed with perfect CSI as in [7].…”

“…Then, we investigate the robustness to imperfect CSI of this solution coupled with that of our existing DNN solution for DF in [7], which we exploit here also for CF. The training of the DNN is performed with perfect CSI as in [7]. The resulting deep learning predictions perform just as poor as the ideal benchmarks (i.e., the brute force for DF, and our closed-form analytical solution for CF) when tested with imperfect CSI and lead to prohibitive primary QoS degradation levels.…”