Reducing the number of users serviced by congested cellular towers given an offered load and a minimum level of acceptable user quality is a major challenge in the operation of LTE networks. In this paper, we utilize a supervised Deep Learning (DL) technique to predict the LTE and LTE-A loading of connected users and then dynamically predict the congestion threshold of each cellular tower under offered load. We then use the predicted congestion thresholds together with quality constraints to fine-tune cellular network operating parameters leading to minimizing overall network congestion. We propose two sets of optimization algorithms to solve our formulated congestion optimization problem. Those are, namely, a variant of Simulated Annealing (SA) algorithm to which we refer as Block Coordinated Descent Simulated Annealing (BCDSA) and Genetic Algorithm (GA). We first compare the performance of integrated DL-BCDSA and DL-GA algorithms and then show that our integrated DL-BCDSA can outperform existing state-of-the-art commercial self organizing tool already deployed in actual cellular networks.

中文翻译：

---

LTE 网络拥塞最小化：深度学习方法


在给定负载和可接受的用户质量最低水平的情况下，减少拥塞蜂窝塔所服务的用户数量是 LTE 网络运营的主要挑战。在本文中，我们利用监督深度学习 (DL) 技术来预测连接用户的 LTE 和 LTE-A 负载，然后动态预测每个蜂窝塔在所提供负载下的拥塞阈值。然后，我们使用预测的拥塞阈值和质量约束来微调蜂窝网络操作参数，从而最大限度地减少整体网络拥塞。我们提出了两组优化算法来解决我们制定的拥塞优化问题。这些是模拟退火 (SA) 算法的变体，我们将其称为块协调下降模拟退火 (BCDSA) 和遗传算法 (GA)。我们首先比较集成 DL-BCDSA 和 DL-GA 算法的性能，然后表明我们的集成 DL-BCDSA 可以超越实际蜂窝网络中已部署的现有最先进的商业自组织工具。