As more people access the global Internet through wireless networks, the limited capacity of wireless networks is becoming a major challenge of our information-based society. Building around the point-to-point architecture, where the basic transmission unit is from a single transmitter to a single receiver, current wireless networks are shown to be severely limited in throughput and do not scale well as they become large and dense. Thus, it is time for a fundamental revisit and change to the wireless network system architecture, in order to explore distributed cooperation and one-to-many encoding/decoding in wireless networks.
The MatrixNet project uses a component-based framework to design implementable, distributed concurrency algorithms and protocols, spanning routing, media access control, and physical layers. It bridges and goes beyond theoretical asymptotic analysis in information theory in order to identify fundamental, algorithmic and system challenges facing the emerging, increasingly important field of system concurrent wireless networks. The project integrates multiple experimental platforms, including Sora, GNU radio, and networked MIMO, and conducts systematic, realistic evaluations. The project team consists of both academic and industrial researchers to facilitate potential technology transfer and cross-institution collaboration.
As more people access the global Internet through wireless networks, the limited capacity of traditinoal wireless networks is becoming a major challenge of our information-based society. Specifically, since the traditional wireless architecture, which is based on the point-to-point abstraction or a small number of concurrent transmissions, is approaching its capacity, this project investigates a fundamental change to the wireless network system architecture: how to support simultaneous, large-scale, cooperative wireless transmissions and receptions, to achieve optimal scaling beyond the traditional wireless architecture. The key intellectual contribution of this project is the design and implementation of MatrixNet, a new component-based framework for a systematic design and realistic evaluation of concurrency control in wireless networks. The MatrixNet project has addressed substantial challenges from multiple aspects, including wireless network architecture, algorithms, protocols, interaction with communication theory, and software implementation. The final outcomes of the project have far exceeded initial expectations, with some highlights as follows. First, the project achieved the challenging goal of finishing an initial design and complete implementation of a prototype named Argos. Exploiting hierarchical and modular design, by properly partitioning baseband processing and holistically considering real-time requirements, Argos was the first platform that was capable of achieving concurrent transmissions at scale: between a 64-antenna transceiver array and 15 terminal nodes. The experimental results of the project demonstrated that the design can scale from 1 to 64 antennas, achieving up to capacity gains up to 6.7 fold, while using a mere 1/64th of the transmission power. As a contrast, before the project, the largest platform in related efforts was equipped with only a dozen or so transceivers. The project achieved its remarkable gain, by applying a novel design, and solving challenges including clock synchronization, transmission synchronization and scalable channel estimation. Second, the project made substantial progress in designing and implementing related software. It proposed and implemented Maple, a generic software framework that substantially simplifies network controller design. It also proposed and designed SoftRAN, based on a fundamental rethinking of software design for radio access networks. Specifically, using software defined centralized control design for radio access networks, SoftRAN abstracts all base stations in a local geographical area as a virtual big base station, creating a framework through which a local geographical network can effectively perform a wide variety of tasks, including load balancing, interference management, throughput and global utility maximization. The aforementioned intellectual advances allow the project to achieve substantial broader impacts: Argos can improve the access capacity of wireless networks by orders of magnitude, benefiting our information-based society; Maple and SoftRAN make it easy to coordinate and control dense and chaotically deployed small cell base stations, improving both the controllability and the reliability of our mobile Internet infrastructure. In the process of developing the aforementioned breakthrough technologies, the project has also trained graduate students at top US academic institutions and developed software systems that may be reused in US higher education broadly.
