Network-Informed Learning For Extremely Reliable Qos In 6g Ultra-High-Density Scenarios

To facilitate the digital convergence, 6G mobile technologies are designed with significantly improved performance indicators. In fact, Quality-of-Service (QoS) increases between one and two magnitude orders compared to previous 5G networks. This extreme QoS must be managed in two different planes. Horizontally, every 6G local station must ensure all connected devices are provided with the required QoS. While, vertically, all devices operating a common application must perceive a stable network performance regardless of their geographical position. Traditionally, teletraffic theory reduces the complexity of user populations by using concepts such as “busy hour” or “maximum demand”. But in ultra-high-density scenarios, this approach would cause resource underutilization and economic failure. For most authors, the response to this challenge is a new network architecture where intelligent models being able to discover hidden patterns and relations could be seamlessly integrated. Despite their numerical success, intelligence models still face some critical open questions. First, complex deep learning models require exhaustive datasets where all behaviors are homogenously represented and described by a large enough number of entries. Second, pure optimization algorithms can reach operation points that, while optimum, are not acceptable in legal, contractual, social or technological terms. In this paper, we propose a network-informed learning architecture, where intelligent algorithms and models are supervised, evaluated and corrected through network and social laws in feedback loops, can ensure an effective extremely reliable QoS in 6G ultra-high-density scenarios. Local models are later consolidated vertically and horizontally, allowing an optimum and network-acceptable operation, even when non-exhaustive datasets are used from training or when unexpected biases are introduced. Experimental simulations are used to evaluate the improvement and performance of the proposed solution. Results show QoS objectives are met with a 99.999% reliability with device densities up to ten million devices per square kilometer.