Intelligent Wireless Multi-Beam Directional Routing with Software-Defined Network Implementation

摘要

Wireless mesh networks (WMNs) have drawn lots of attention in the past decades due to its scalability, robustness, and flexibility. However, the performance of WMNs is still limited by the wireless bandwidth, radio frequency interference, etc. To deal with the limitation, our research first focuses on a particular radio frequency technology, called Multi-beam directional antennas (MBDAs). The MBDA allows a node to simultaneously send out packets in multiple directions without interference among the beams. Thus it significantly improves the throughput and elongates the transmission distance compared with omni-directional or single-beam directional antennas. Our goal in this study is to come up with a series of Artificial Intelligent (AI) strategies to explore MBDAs during routing implementation. Then, the AI method is expanded to a more general WMNs with heterogeneous wireless devices, in which we use centralized network management structure for the purpose of traffic engineering optimization.;First, we present a novel routing scheme for WMNs with MBDAs. It has the following 3 features: (1) Ripple-Diamond-Chain (RDC) shaped routing: to explore the multi-direction transmission capability of the MBDAs, we propose to use rateless codes to obtain loss-resilient symbols for original packets. Those symbols can go through each beam in the same time. Then we propose to use ripples to differentiate each hop of nodes in the tree topology of the WMN, which consists of mesh routers (tree roots) and a large number of mesh clients (tree nodes). The main path consists of the nodes with the best link quality and reliability. The symbols are divergent into multiple paths but converge into the main path node in the next hop. The entire routing topology looks like a diamond chain. By using such RDC style routing, we can fully explore the MBDA benefits. (2) Systematic link quality modeling: Our research targets the highly dynamic radio conditions in the WMN. The directional antennas can cause node capture issue. The link could have deep fading in each hop. The rateless codes need to adjust the transmission pause time. We propose to integrate all these factors together to determine the link quality in dynamical network conditions. (3) AI-Augmented path link selection: Our routing scheme is augmented via Artificial Intelligence (AI) algorithms. Especially, we use two AI techniques to enhance the routing performance: Fuzzy Logic (FL): To adapt different QoS requirements, we propose to use Fuzzy Logic to define the weighted link quality. Thus we know which link should be selected for different QoS flows. Reinforcement Learning (RL): Since the dynamic radio conditions need a long-term consideration of the throughput performance within multiple phases of routing path control, we propose to use RL to select the main path based on the cumulative throughput rewards in all links. In the simulations, we use real-time video as well as other types of traffic types to validate the high-throughput, QoS-differentiated, multi-beam routing efficiency, as well as its intelligent path determination in dynamic WMN environment.;Second, we expand the routing/TE problem from conventional WMNs to Software-Defined Networking based WMNs (SD-WMNs) and propose a novel TE structure on SD-WMN called "Prediction-based Link Uncertainty Solution in SD-WMNs" (PLUS-SW). The SDN aims to realize a centralized monitor and control upon a network by detaching the control module from data plane, in which an independent control plane is employed for the network management and all the routers on data plane are simplified to the dummy packet forwarding devices. Although the centralized control achieved by SDN is promising on the significant improvement of traffic engineering via network-wide management, it is naturally inadequate when responding to the uncertainty of WMNs in terms of latency reduction and coarse control panel management. We thus propose PLUS-SW to overcome these shortages. The PLUS-SW possesses a centralized traffic engineering and wireless channel scheduling on WMNs according to the paradigm of SDN in order to efficiently arrange the network traffic and omit wireless interference in a global manner. Moreover, PLUS-SW employs double-layer supervised learning model to predict unexpected wireless link failure in the sense that the central controller can notice the potential link failure threat and send back the backup solution to affected routers ahead the link failure. The rerouting calculation of PLUS-SW on congested traffic is based on the network-wide observation while keeping the overhead of centralized control at a low level.;Finally, a wireless network platform is also introduced. This platform is built with Software Defined Radio hardware, called USRP. In this platform, we achieved some preliminary functions of WMNs, such as real-time video transmission, cross-layer design and etc. The platform can work as a test bed to estimate the performance of proposed design of traffic engineering.