Distributed Dynamic Channel Allocation in 6G in-X Subnetworks for Industrial Automation

Abstract

In this paper, we investigate dynamic channel selection in short-range Wireless Isochronous Real Time (WIRT) in-X subnetworks aimed at supporting fast closed-loop control with super-short communication cycle (below 0.1 ms) and extreme reliability (>99.999999%). We consider fully distributed approaches in which each subnetwork selects a channel group for transmission in order to guarantee the requirements based solely on its local sensing measurements without the possibility for exchange of information between subnetworks. We present three fully distributed schemes: ϵ-greedy channel allocation, minimum SINR guarantee (minSINR) and Nearest Neighbor Conflict Avoidance (NNCA) based on measurements of the minimum SINR and interference power. We further apply a centralized graph coloring scheme as a baseline for evaluating performance of the proposed distributed algorithms. Performance evaluation considering subnetwork mobility and spatio-temporal correlated channel models shows that the dynamic allocation schemes results in significant performance improvement and a reduction in the bandwidth required for supporting such extreme connectivity by up to a factor larger than 2 relative to static channel assignment.

Multi-agent Reinforcement Learning for Dynamic Resource Management in 6G in-X Subnetworks

“…To tackle such computationally intractable problems, there have been many approaches, leveraging techniques in various fields, for example, geometric programming [5], weighted minimum mean square optimization [6], game theory [7] [8] [9], fractional programming [10], information theory [11] [12] and machine learning [13] [14]. However, in order to enable dynamic resource allocation optimization, these existing algorithms, no matter conventional Centralized Graph Coloring (CGC) algorithm [15] or machine learning-based methods, typically depend on sorts of hardly accessible information in a real-world network, such as channel gain between any two subnetworks as there are no direct communications between them in practice. In addition, the existing methods are difficult to reason the potential interference relationships between agents in multiple mobile subnetwork systems.…”

“…Among the centralized approaches, the work [15] lists some possible algorithms for subnetwork resource allocation, including the minimum SINR (signal to interference-plus-noise ratio) guarantee algorithm, the Nearest Neighbour Conflict Avoidance (NNAC) algorithm and the CGC algorithm. All these are centralized algorithms, on top of the issue that they can't access the unavailable channel gains between subnetworks, they also generate massive data traffic due to huge data exchange during the iterative resource allocation optimization.…”

“…To fully validate the effectiveness of the proposed approach, we compare our approach with other four approaches: i) Random Baseline, which selects the channel and transmission power for each subnetwork in a random manner at each time step; ii) Dynamic Greedy Algorithm (DGA) [15], which selects the channel with least RSSI every time interval, and the transmission power is always the maximal. iii) AC [51], which is the most classic deep RL method and can handle continuous action space; iv) MADDPG [32], which can obtain global knowledge and deal with discrete action spaces with Gumbel-Softmax; v) GA-Net w/o Hard Attn, which is GA-Net without hard attention; vi) GA-Net w/o Attn, which is GA-Net without attention and soft attention; vii) QMIX, which is a state-of-the-art multi-agent reinforcement learning algorithm based on value function decomposition.…”

Enhanced Interference Management for 6G in-X Subnetworks

“…Such interference levels may limit the possibility for supporting extreme communication requirements, necessitating novel measures beyond traditional reactive approaches. This has resulted in a number of recently published works on radio resource allocation methods for dense wireless networks with independent in-X subnetworks [34], [35]. In [34], distributed heuristic algorithms were evaluated and compared with a centralized graph coloring (CGC) baseline in dense deployments of in-X subnetworks.…”

“…This has resulted in a number of recently published works on radio resource allocation methods for dense wireless networks with independent in-X subnetworks [34], [35]. In [34], distributed heuristic algorithms were evaluated and compared with a centralized graph coloring (CGC) baseline in dense deployments of in-X subnetworks. Although the heuristics show better performance than random allocation, the high abstraction level in the utilized simulation approach limits the interpretability of the results.…”