Quantum teleportation is a foundational protocol for sending quantum information through entanglement distribution and classical communication. Assuming ideal classical communication, the reliability of quantum teleportation is limited by the fidelity of the shared EPR pairs. This reliability can be improved through two mechanisms: entanglement purification and quantum error correction (QEC). Using both techniques in concert requires flexible QEC rates, since purification alters the structure of errors induced by imperfect-EPR teleportation, and fixed-rate codes cannot be uniformly effective across purification regimes or reliability targets. In this work, we supplement purification with punctured QEC codes, providing a family of code variants that can be adapted to error-channel characteristics and reliability targets. Punctured codes improve teleportation reliability across a broader range of purification regimes, enabling target reliability to be met without hardware-level code switching. This is corroborated by numerical results, showing that different punctured codes achieve the lowest logical error probability in different operating regimes, and that selecting among them reduces logical error relative to fixed-rate encoded teleportation. This reduction relaxes the requirement on the initial EPR fidelity or purification needed to achieve a target reliability. Overall, puncturing enables adaptation to varying entanglement conditions and reliability requirements while reusing a single stabilizer structure.

This work develops a learning-based framework that directly exploits noisy pilots to optimize reconfigurable intelligent surface (RIS) systems while accommodating different service priorities and fairness via user weights. First, an adaptive beamforming configuration problem is formulated to generate the base station active beamforming vectors and RIS passive beamforming reflection coefficients that optimize the weighted sum-rate. Under mild regularity conditions, this problem is shown to attain a maximum. To learn approximate solutions, a novel hypernetwork-based beamforming (HNB) framework is proposed. Particularly, a beamforming network (BFN) exploits available information, including noisy pilots, to generate optimized beamforming configurations. Rather than learning one BFN, a hypernetwork is trained to dynamically generate BFN learning parameters from an input conditioning vector. When the conditioning vector is chosen as the user weights, the trained HNB can tune the BFN to the user weights without the need for retraining. Numerical experiments demonstrate that tuning allows the proposed HNB to perform close to an optimistic block-coordinate descent with perfect CSI benchmark and significantly outperform static learning where a BFN is directly trained to optimize beamforming configurations. Additionally, employing the HNB to also tune the BFN to location information considerably reduces the pilots needed to generate optimized beamforming configurations.

Flexible resource allocation in intelligent reflecting surface (IRS)-assisted systems is necessary to account for fairness as well as time-varying channel behavior and user service priorities. An IRS-assisted system can achieve this flexibility by assigning different weights to each user when optimizing resources and the IRS configuration. In this paper, for the first time, we propose a hypernetwork-based beamforming (HNB) framework to dynamically leverage pilot information and user weights to generate the beamforming vectors and IRS configuration that maximize the weighted sum-rate (WSR) in a multi-user IRS-assisted system. As opposed to a traditional learning approach where a beamforming network (BFN) is trained once to optimize the WSR for every possible set of user weights, in a hypernetwork approach, a hypernetwork is trained to generate the learning parameters of the BFN conditioned on the input user weights, i.e., the BFN parameters are now adapted to the user weights without the need for any retraining. Numerical experiments corroborate the effectiveness of the proposed HNB framework to provide performance close to (within approximately 15−17% of) the optimistic benchmark produced by a numerical block-coordinate descent (BCD) algorithm that assumes perfect channel state information (CSI) knowledge. Moreover, the HNB trained with only a few epochs outperforms traditional fully-trained deep learning methods such as fully connected neural networks (FCNN) and graph neural networks (GNN). For example, in one considered scenario, the HNB nearly halves the gap to the BCD-with-CSI performance to 16% compared to gaps of 31% and 28% associated with FCNN and GNN schemes, respectively.

We investigate the joint uplink-downlink configuration of an intelligent reflecting surface (IRS) for multi-user frequency-division-duplexing (FDD) and time-division-duplexing (TDD) systems. This is motivated in FDD since uplink and downlink transmissions occur simultaneously and hence an IRS must be jointly configured for both transmissions. In TDD, while a joint design is not strictly necessary, it can significantly reduce feedback overhead, power consumption, and configuration periods associated with updating the IRS. To compute the trade-off between uplink and downlink rates achieved by a joint design, a weighted-sum problem is formulated and optimized using a developed block-coordinate descent algorithm. The resulting uplink-downlink trade-off regions are investigated by numerical simulation to gain insights into different scenarios. In all FDD scenarios and some TDD scenarios, the jointly optimized design significantly outperforms the fixed-uplink (fixed-downlink) heuristic of using the IRS configuration optimized for uplink (downlink) to assist downlink (uplink) transmissions. Moreover, the joint design substantially bridges the gap to the individual design upper bound of allowing different IRS configurations in uplink and downlink. Otherwise, in the remaining TDD scenarios, the fixed-uplink and fixed-downlink designs nearly achieve the individual design performance and substantially reduce overhead and/or complexity compared to the optimized joint design and individual design.