Device-to-Device (D2D) communication allows for users placed in a cell to establish direct connections with each other using several connection modes. In this paper, we propose Multi-Path D2D (MPD2D), a mathematical optimization framework that accounts for the availability of D2D modes under the requirements dictated by a process of flow requests. MPD2D selects the combination of cellular and D2D links that boosts network performance as much as possible. We consider Underlay and Overlay as Inband D2D modes reusing cellular frequencies with scheduled resources and the Outband D2D mode exploiting WLAN frequencies and the 802.11 random access scheme to complement cellular resources. We model throughput, energy consumption, interference, and per-flow network requirements, so to define a network utility function that accounts for throughput and power efficiency. Moreover, we formulate a user satisfaction metric that accounts for the history of users within the cell. Integrating such a metric in a throughput optimization problem is lightweight yet very effective to drive towards almost complete fairness. Our optimization scheme is formulated as a Binary Non-Linear Program, which results in higher throughput performance in comparison to other state-of-the-art solutions we have tested. Finally, we propose two effective heuristics, whose performance is near-optimal, whereas their complexity scales polynomially with the number of users.

设备到设备（D2D）通信允许放置在小区中的用户使用几种连接模式彼此建立直接连接。在本文中，我们提出了Multi-Path D2D（MPD2D），这是一种数学优化框架，该框架考虑了流程请求过程所要求的D2D模式的可用性。MPD2D选择蜂窝和D2D链接的组合，以尽可能提高网络性能。我们将Underlay和Overlay视为带内D2D模式，以计划的资源和带外重用蜂窝频率D2D模式利用WLAN频率和802.11随机访问方案来补充蜂窝资源。我们对吞吐量，能耗，干扰和每流网络需求进行建模，以便定义一个考虑吞吐量和功率效率的网络实用程序功能。此外，我们制定了一个用户满意度指标，该指标说明了单元内用户的历史记录。在吞吐量优化问题中集成这样的指标是轻量级的，但是却非常有效地推动了几乎完全的公平性。我们的优化方案被公式化为二进制非线性程序，与我们测试过的其他最新解决方案相比，它具有更高的吞吐性能。最后，我们提出了两种有效的启发式算法，它们的性能接近最佳，