The Internet Engineering Task Force recommends the use of a group of well-defined protocols to support Internet of ThingsLow-Power Low-Rate Networks (LLNs). These mechanisms range from physical and media access layer technologies like IEEE 802.15.4 and Bluetooth Low Energy to session and application layer protocols like the Constrained Application Protocol (CoAP). Specifically, CoAP provides, by means of two different modes of operation, great flexibility to deal with the power requirements of wireless LLNs. One mode supports fire-and-forget packet transmission while the other, through retransmissions, guarantees delivery. The trade-off between these mechanisms is the exchange of high packet loss by high latency and increased power consumption. In this paper we introduce an algorithm that dynamically predicts these parameters, by means of supervised learning, based on network conditions that result from Maximum Likelihood Estimation. This prediction, in turns, can be used for on-the-fly CoAP mode selection that accomplishes quality goals.

在互联网工程任务组建议使用一组明确定义的协议来支持物联网低功耗低速率网络（LLN的）。这些机制的范围从物理和媒体访问层技术（如IEEE 802.15.4和低功耗蓝牙）到会话和应用层协议（如约束应用协议）（CoAP）。具体而言，CoAP通过两种不同的操作模式提供了极大的灵活性来处理无线LLN的功率需求。一种模式支持即发即弃的数据包传输，而另一种则通过重传来保证传递。这些机制之间的权衡是通过高等待时间和增加的功耗来交换高分组丢失。在本文中，我们介绍了一种算法，该算法根据最大似然估计产生的网络条件，通过监督学习动态预测这些参数。该预测又可以用于实现质量目标的实时CoAP模式选择。