Downclocking WiFi to Improve Energy Efficiency in Mobile Devices /

As manufacturers continue to improve the energy efficiency of battery-powered wireless devices, WiFi has become one of the most significant power consuming components on them. Hence, modern devices fastidiously manage their radios, shifting into low-power listening or sleep states whenever possible. Unfortunately, recent studies have shown that a wide variety of popular applications make frequent and persistent use of the network, frustrating attempts to keep the WiFi chipset in a power-efficient state. Even a less chatty application like email could still constitute a significant source of energy consumption. A fundamental limitation with current power- saving approaches is that the radio is incapable of transmitting or receiving unless fully powered. In this dissertation, I introduce a widely-used technique for energy savings in CMOS, namely dynamic frequency scaling (DFS), to standards-compliant WiFi radios. A fundamental limit that prevents decreasing the clock frequency in WiFi chipsets is the Nyquist sampling theorem, which states that the sampling rate must be twice the signal bandwidth. Leveraging the inherent sparsity in direct sequence spread spectrum (DSSS) modulation in 802.11b, this dissertation proposes a transceiver design based on compressive sensing that allows WiFi devices to operate their radios below the Nyquist rate, thus consuming less power. Recognizing that modern WiFi chipsets are shifting towards orthogonal frequency-division multiplexing (OFDM), I further extend the benefits of DFS to OFDM communication systems. Based on my observations of OFDM signaling, I show that the aliasing effect, resulting from undersampling, essentially transforms the original per-subcarrier quadrature amplitude modulation (QAM) into a more dense, but still decodable, QAM modulation. implement both the 802.11b DSSS-based (SloMo) and 802.11a/g OFDM-based design (Enfold) on the Microsoft Sora platform, a full-stack programmable software-defined radio. Both implementations are standards-compliant, and SloMo is also fully backwards compatible with existing WiFi deployments. My experiments and evaluation demonstrate that downclocking WiFi is both practical and beneficial, even when clock rate is reduced by a factor of five. Specifically, my downclocked design can reduce energy consumption by over 30% for many popular smartphone apps even with a deliberately conservative power mode