Modern wireless networks face fundamental challenges in terms of throughput, delay and complexity. First, as the volume of data traffic continues to explode, it is imperative to extract the highest throughput out of limited spectrum resources. Second, emerging new applications increasingly require not only high throughput, but also low delay, to be functional. Third, as these networks employ more complex heterogeneous topologies and advanced physical-layer techniques, the complexity to control them also continues to grow. However, these three dimensions of performance metrics (i.e., throughput, delay and complexity) are often not aligned with each other. For general topolgoies, existing wireless control algorithms suffer at least one of the three performance dimensions, i.e., they suffer from either low throughput guarantees, exponentially-large delay, or exponential-high complexity. Indeed, in the literature, there have been conjectures that there may exist a fundamental tradeoff between these performance dimensions that no practical algorithms can overcome. However, our preliminary work suggests that such pessimism may be unnecessary. Specifically, we have developed new algorithms that exploit the flexibility offered by multiple physical or virtual channels to overcome this difficulty. For the first time in the literature, our proposed algorithms can achieve arbitrarily close-to-optimal throughput utility as well as provably-low delay (that does not grow with the network size), with low complexity that only grows logarithmically with the network size. Our ongoing work aims to generalize these ideas to more complex physical layers (e.g., MIMO) and network structures (e.g., Cloud-RAN), to energy-efficient operations, and to highly-dynamic settings.
Selected publications: