| Wireless networking is envisioned to provide a truly ubiquitous computing environment whereby a large number of pervasive computing devices are interconnected in any time, any where and any form, reliably and efficiently. Today's wireless application scenarios lie in a wide spectrum of data communication systems, ranging from personal area networks (e.g., Bluetooth) to local area networks (e.g., IEEE 802.11) to wide-area networks (e.g., 3G networks). Beyond these domains, new industrial trends call for the application of wireless technologies into two emerging networking areas: Network-on-Chip (NoC) over a multi-core processor by chip manufacturers and 4G Long-Term Evolution (LTE) standardization by cellular operators. This dissertation explores these two extreme domains and realizes high-performance wireless communication systems, further stretching today's wireless networking spectrum: from the microarchitecture level (i.e., wireless NoC) to the macro-scale (i.e., 4G LTE cellular networks).;While wireless networking offers great opportunities for those brand-new domains, it brings new design challenges for each component in the network architecture and every layer in the protocol stack. In particular, designing high-diversity, and we demonstrate that the exploitation of all potential dimensional diversity boosts up the overall system performance. Second, we identify redundant feedback information and make the control messages be suppressed or even unexchanged whenever its impact is negligible to the system performance. Such an on-demand feedback approach is based on our observation of infrequent packet loss events in wireless NoC and the redundant feedback information in LTE.;Following the above design principles, we conduct a systematic study on high-performance wireless on-chip interconnection design and wireless resource management for LTE networks. In the micro-scale NoC study, we make a case for using a two-tier wireless and wired architecture to interconnect hundreds to thousands of cores for future CMPs. To this end, we propose a recursive, wireless interconnect structure WCube via FDMA technique, which features a single transmit antenna and multiple receive antennas at each micro wireless router and offers scalable performance in terms of latency and connectivity. We further devise a new wormhole-based, two-tier routing algorithm that is deadlock free and ensures minimum-latency route. In order to achieve reliable communication over wireless channel while minimizing feedback overhead, we develop a lightweight on-demand minimum-signaling-overhead loss management scheme by exploiting the WCube interconnection property. Detailed simulation-based evaluation shows the scalability and effectiveness of our design for high-performance wireless on-chip interconnect, compared to the current 2D wired mesh designs.;In the macro-scale 4G LTE study, we address the problem of wireless data scheduling to incorporate new cellular access technologies on the LTE downlink and uplink. Our proposed scheduling framework fully exploits frequency and spatial dimensions on top of time domains, with OFDMA and MIMO techniques. Modeling it as an optimization framework and analyzing the complexity of the problem, we propose different types of resource allocation solutions with respect to the feedback resolution. We verify through theoretical analysis that 1) our proposed scheduling solutions all have the proven constant-factor performance guarantees, and 2) our practical feedback reduction schemes offer the same performance bounds as the full feedback cases. The LTE system model evaluations demonstrate the superior performance of our solutions over other algorithms that do not exploit the potential 3--dimensional diversity gain. |
