Engineering adaptive communication protocols for upcoming WSN applications

随着物联网（IoT）的出现，越来越多的设备开始交换信息。虽然这带来了新的有趣的应用，但它也对底层通信基础设施提出了很高的要求，特别是无线传感器网络（wsn），它促进了物联网设备与环境之间的交互。这些必须能够实现可靠、节能和及时的通信，随着这种网络的动态和异质性的增加，通信变得越来越复杂。在这种情况下，自适应通信协议似乎是一种有趣的解决方案，因为它们可以在不断变化的条件下自动重新配置自己以满足指定的性能目标。不幸的是，现有的解决方案仍然存在限制其性能和适用性的主要缺陷。例如，它们只支持单一的性能目标（例如，可靠性或能源，而不是两者都支持），并且不考虑其他目标之间的异构性。为了解决这一问题，本文提出了一种新的自适应通信协议。与文献中的现有解决方案相比，所提出的方法基于一个专门的MAC（介质访问控制）协议，该协议本质上允许自适应，并通过ODA（观察、决定、行动）循环进行扩展，以实现控制和自动化。在此背景下，将研究不同的自适应技术，如控制理论方法和机器学习（例如Q-Learning）。目的是评估它们的有效性以及它们对嵌入式节点的内存/计算性能的影响。这种情况下的有效性是根据不同的适应技术在寻找对立的性能目标之间的折衷方面有多好来衡量的。例如，可靠性和能耗是两个相互矛盾的要求，即增加一个会使另一个恶化。因此，找到合适的权衡是至关重要的。此外，考虑到wsn是分布式的，具有异构（即特定于节点）限制的嵌入式系统，这将是一项非常具有挑战性的任务。最后，提出的自适应通信协议将在仿真和真实硬件上进行评估，例如在T-Mote Sky节点上。这提供了有价值的见解，有助于纠正当今解决方案的缺点，从而提高它们的适用性和性能——目前的自适应协议已经比经典的非自适应协议实现了80%的性能，但这还可以进一步提高。总之，本研究项目有助于设计和开发能够应对现代和未来应用中日益增加的复杂性、动态性和异质性的自适应无线传感器网络。

With the advent of the Internet of Things (IoT), a growing number of devices start exchanging information. While this enables new interesting applications, it also places high demands on the underlying communications infrastructure, in particular, wireless sensor networks (WSNs), which facilitate interaction between IoT devices and the environment. These must enable reliable, energy-efficient and timely communication, which becomes increasingly complex as dynamics and heterogeneity of such networks increase. In this context, adaptive communication protocols seem to be an interesting solution, since these can automatically reconfigure themselves to meet specified performance goals under changing conditions. Unfortunately, existing solutions still have major drawbacks limiting their performance and applicability. For example, they only support single performance goals (e.g., reliability or energy, but not both) and do not account for heterogeneity among others. In this research work, to overcome this problem, a novel adaptive communication protocol is proposed. In contrast to existing solutions from the literature, the proposed approach is based on a specialized MAC (medium access control) protocol that intrinsically allows for adaptivity, which is extended by an ODA (observe, decide, act) loop for control and automation. In this context, different adaption techniques will be investigated such as control-theoretic approaches and machine learning (e.g., Q-Learning). The purpose is to assess their effectiveness as well as their impact on memory/computational performance of the embedded nodes. Effectiveness in this context is measured with respect to how good different adaption techniques are in finding a compromise between opposing performance goals. For example, reliability and energy consumption are two contradictory requirements, i.e., increasing one worsens the other. As a result, it is paramount to find suitable tradeoffs. In addition, considering that WSNs are distributed, embedded systems with heterogeneous (i.e., node-specific) limitations, this turns to be a very challenging task.Finally, the proposed adaptive communication protocol will be evaluated in simulation and on real hardware, e.g., on T-Mote Sky nodes. This provides valuable insights that help correct shortcomings by today's solutions and thus increase their applicability and performance – current adaptive protocols already achieve >80% higher performance than classic, non-adaptive ones, but this can be further increased. In summary, this research project thus contributes to the design and development of adaptive WSNs able to cope with the increasing complexity, dynamics and heterogeneity of modern and upcoming applications.