Spatial Deep Learning for Wireless Scheduling (1808.01486v3)

Abstract: The optimal scheduling of interfering links in a dense wireless network with full frequency reuse is a challenging task. The traditional method involves first estimating all the interfering channel strengths then optimizing the scheduling based on the model. This model-based method is however resource intensive and computationally hard because channel estimation is expensive in dense networks; furthermore, finding even a locally optimal solution of the resulting optimization problem may be computationally complex. This paper shows that by using a deep learning approach, it is possible to bypass the channel estimation and to schedule links efficiently based solely on the geographic locations of the transmitters and the receivers, due to the fact that in many propagation environments, the wireless channel strength is largely a function of the distance dependent path-loss. This is accomplished by unsupervised training over randomly deployed networks, and by using a novel neural network architecture that computes the geographic spatial convolutions of the interfering or interfered neighboring nodes along with subsequent multiple feedback stages to learn the optimum solution. The resulting neural network gives near-optimal performance for sum-rate maximization and is capable of generalizing to larger deployment areas and to deployments of different link densities. Moreover, to provide fairness, this paper proposes a novel scheduling approach that utilizes the sum-rate optimal scheduling algorithm over judiciously chosen subsets of links for maximizing a proportional fairness objective over the network. The proposed approach shows highly competitive and generalizable network utility maximization results.

• The paper introduces a novel unsupervised deep learning framework that bypasses traditional CSI estimation by using only geographic location data.
• It employs spatial convolution layers with iterative feedback to efficiently compute interference and achieve near-optimal scheduling performance.
• The method generalizes across diverse network sizes and densities, balancing sum rate maximization with fairness in resource allocation.

Spatial Deep Learning for Wireless Scheduling

In the paper "Spatial Deep Learning for Wireless Scheduling," the authors address the challenges of scheduling interfering links in dense wireless networks with full frequency reuse through an innovative approach that leverages deep learning. The conventional method of estimating all channel strengths prior to optimizing the scheduling can be resource-demanding and computationally prohibitive, especially in dense networks where the sheer number of interfering channels escalates the complexity. This paper shifts the paradigm by introducing a novel way to bypass this traditional approach.

The proposed method harnesses geographic location data of the transmitters and receivers to perform scheduling without any explicit channel state information (CSI). The underlying hypothesis is based on the observation that wireless channel strength is largely influenced by distance-dependent path loss in many environments. The authors develop an unsupervised learning framework that trains a neural network on randomly deployed networks. This neural network features a distinctive architecture comprising novel spatial convolution layers specifically designed to compute the interference of neighboring nodes based on their geographical positions. Multiple feedback stages are employed to refine the solution iteratively.

The results demonstrate the neural network's capability to achieve near-optimal performance in maximizing the sum-rate. Moreover, its ability to generalize to networks of varying sizes and densities without requiring re-training signifies a considerable practical advantage. In terms of fairness, a novel scheduling approach is proposed that applies sum-rate optimal algorithms over carefully chosen link subsets, effectively facilitating proportional fairness within the network.

From a technical standpoint, the paper establishes the feasibility and effectiveness of using deep learning techniques to mitigate computational complexities inherent in classical model-based scheduling techniques. By training the neural network across various network topographies, the model can generalize across diverse channel conditions, yielding competitive results compared to traditional methods like greedy heuristics and FPLinQ in terms of sum rate, with a significant reduction in computational overhead.

This methodology opens several avenues for future research. The scalability of such neural networks can be explored further in real-time dynamic wireless networks. There is also potential for improving fairness without adversely affecting the network-wide sum rate, thereby striking a balance between efficiency and equity. Additionally, integrating such a framework with concepts like multi-agent systems or cooperative game theory could unlock new dimensions of resource allocation strategies in the burgeoning domain of device-to-device communication.

In essence, the paper presents a compelling case for the deployment of spatial deep learning techniques in wireless scheduling, potentially transforming traditional resource allocation strategies by easing the computational and resource burdens and offering robust generalization capabilities.