Abstract
The practical performance in terms of throughput, robustness, and scalability of traditional Wireless Multihop Networks (WMNs) is limited. The key problem is that such networks do not allow for advanced physical layers, which typically require (a) spatial diversity via multiple antennas, (b) timely Channel State Information (CSI) feedback, and (c) a central instance that coordinates nodes. We propose Corridor-based Routing to address these issues. Our approach widens traditional hop-by-hop paths to span multiple nodes at each hop, and thus provide spatial diversity. As a result, at each hop, a group of transmitters cooperates at the physical layer to forward data to a group of receivers. We call two subsequent groups of nodes a stage. Since all nodes participating in data forwarding at a certain hop are part of the same fully connected stage, corridors only require one-hop CSI feedback. Further, each stage operates independently. Thus, Corridor-based Routing does not require a network-wide central instance, and is scalable. We design a protocol that builds end-to-end corridors. As expected, this incurs more overhead than finding a traditional WMN path. However, if the resulting corridor provides throughput gains, the overhead compensates after a certain number of transmitted packets.
We adapt two physical layers to the aforementioned stage topology, namely, Orthogonal Frequency-Division Multiple Access (OFDMA), and Interference Alignment (IA). In OFDMA, we allocate each subchannel to a link of the current stage which provides good channel conditions. As a result, we avoid deep fades, which enables OFDMA to transmit data robustly in scenarios in which traditional schemes cannot operate. Moreover, it achieves higher throughputs than such schemes. To minimize the transmission time at each stage, we present an allocation mechanism that takes into account both the CSI, and the amount of data that each transmitter needs to transmit. Further, we address practical issues and implement our scheme on software-defined radios. We achieve roughly 30% average throughput gain compared to a WMN not using corridors. We analyze OFDMA in theory, simulation, and practice. Our results match in all three domains.
Further, we design a physical layer for corridor stages based on IA in the frequency domain. Our practical experiments show that IA often performs poorly because the decoding process augments noise. We find that the augmentation factor depends only on the channel coefficients of the subchannels that IA uses. We design a mechanism to determine which transmitters should transmit to which receivers on which subchannels to minimize noise. Since the number of possible combinations is very large, we use heuristics that reduce the search space significantly. Based on this design, we present the first practical frequency IA system. Our results show that our approach avoids noise augmentation efficiently, and thus operates robustly. We observe that IA is most suitable for stages with specific CSI and traffic conditions. In such scenarios, the throughput gain compared to a WMN not using corridors is 25% on average, and 150% in the best case.
Finally, we design a decision engine which estimates the performance of both OFDMA and IA for a given stage, and chooses the one which achieves the highest throughput. We evaluate corridors with up to five stages, and achieve roughly 20% average throughput gain. We conclude that switching among physical layers to adapt to the particular CSI and traffic conditions of each stage is crucial for efficient and robust operation.