TLDR: A new framework uses AI-powered “Channel Knowledge Maps” (CKM) to predict wireless network performance, enabling efficient scheduling for active intelligent reflecting surfaces (AIRSs) in multi-user systems. This approach, involving two neural networks (LPS-Net and SE-Net) and a scheduling algorithm (SM-IB), significantly improves data throughput and reduces computational complexity by bypassing the need for real-time channel information.
In the rapidly evolving landscape of wireless communication, technologies like Intelligent Reflecting Surfaces (IRSs) and their more advanced counterparts, Active IRSs (AIRSs), are poised to revolutionize next-generation networks. These surfaces can intelligently reflect wireless signals, significantly boosting network performance. However, deploying them effectively in multi-user systems presents significant challenges, including signal loss over long distances (known as double-pathloss) and the complex task of scheduling resources for many users, especially given hardware limitations and the need for real-time channel information.
Traditional methods for managing these systems often require precise, instantaneous knowledge of the wireless channel conditions for every user. As the number of users and the complexity of the network grow, acquiring this information becomes incredibly burdensome, leading to high overhead and delays that can render scheduling decisions obsolete before they are even implemented. This is where the new research, titled “Neural Channel Knowledge Map Assisted Scheduling Optimization of Active IRSs in Multi-User Systems,” steps in, proposing an innovative solution to these critical issues.
A Novel Approach: Neural Channel Knowledge Map (CKM)
The core of this research is a novel scheduling framework built upon a concept called the neural Channel Knowledge Map (CKM). Imagine a detailed map that doesn’t just show physical locations but also predicts wireless channel characteristics and performance metrics for any given user position. This CKM accumulates historical channel and data throughput measurements, tagged with user locations, to create a spatially correlated database. By leveraging this map, the system can infer channel conditions and predict performance without needing to acquire full, real-time channel state information (CSI) for every user, every time.
To make this CKM intelligent and adaptable, the researchers designed a system based on deep neural networks (DNNs), specifically using Transformer-based architectures. This neural CKM consists of two cascaded networks:
-
LPS-Net (Link Power Statistics Network): This network is responsible for predicting the statistical properties of power for different wireless links. For example, it can predict how strong the signal will be directly from the base station to a user, or from the base station to an AIRS and then to a user, or even the noise generated by the AIRS itself. It takes into account the user’s position and the system’s configuration.
-
SE-Net (Spectral Efficiency Network): Once LPS-Net provides the link power statistics, SE-Net takes over. It combines these statistics to accurately predict the ergodic spectral efficiency (SE) – a measure of how efficiently the wireless spectrum is used – for a user under a specific AIRS association. This means it can tell how well a user will perform if served by a particular AIRS.
By using these cascaded networks, the system can efficiently store and predict complex channel information, overcoming the challenges of high-dimensional data and the need for real-time measurements.
Optimizing Scheduling with SM-IB Algorithm
With the ability to predict spectral efficiency using the neural CKM, the next challenge is to optimize how users are assigned to AIRSs and how time and frequency resources are allocated. The goal is to maximize the minimum data throughput for all users, ensuring a fair and efficient distribution of resources. To achieve this, the paper introduces a low-complexity algorithm called Stable Matching-Iterative Balancing (SM-IB).
The SM-IB algorithm works in three stages:
-
Initial User Grouping: Users are first grouped based on their preference for different AIRSs, and then assigned to specific time slots. This initial step aims to roughly balance the expected performance across different groups.
-
Max-Min Throughput in Each Slot: Within each time slot, the algorithm refines the user-to-AIRS assignments and adjusts the allocation of resource blocks (portions of bandwidth and time). It iteratively balances the throughput among users in that slot, ensuring no single user is left with very low performance.
-
Max-Min Throughput Across Slots: Finally, the algorithm works to balance the throughput across all time slots. It identifies slots with the lowest and highest common throughput and strategically swaps users between them to further improve the overall minimum throughput.
This iterative process ensures that the system converges towards an optimal solution, providing excellent performance while keeping computational complexity significantly lower than traditional, more exhaustive optimization methods.
Also Read:
- Smart Privacy for Image Data: Optimizing Semantic Communication with Independent Jammers
- Building Adaptive and Covert UAV Networks with AI
Promising Results
Numerical evaluations of the proposed framework have shown highly encouraging results. The neural CKM significantly improves prediction accuracy and computational efficiency compared to other baseline models. Furthermore, the SM-IB algorithm effectively achieves a max-min throughput that is very close to the theoretical upper bound, but with a dramatically reduced running time. For instance, in scenarios with many users, the proposed algorithm completes its calculations in milliseconds, whereas traditional solvers can take hundreds or even thousands of seconds.
This research marks a significant step forward in making active IRSs a practical reality for future wireless networks. By leveraging AI to predict channel conditions and optimize scheduling, it addresses key challenges that have previously limited the full potential of these promising technologies. For more in-depth information, you can read the full research paper here.
