Wireless systems are deployed in environments like buildings, bridges and forests - each of which has very different wireless properties. In this paper, we propose a technique to customize simulation for a new RF environment, before a physical layer model has been created for it. Our approach is to create physical layer traces or PhyTraces that capture the RF environment of a single node. These PhyTraces can then be composed in multiple ways to simulate protocol-layer performance in the target environment. This approach offers an alternative to conventional trace-based network simulation that is very realistic, but does not permit protocol-level changes after the traces are collected. We show that PhyTraces can be collected very quickly and can be used in simulation with low computational overhead to model networks of various sizes and densities. To evaluate this approach, we collect PhyTraces in 3 different physical environments using 2 different low-power radios at 4 different transmission power levels. Our analysis indicates that the PhyTraces can predict the performance of neighbor discovery algorithms, distance vector routing algorithms, and flooding algorithms in real deployments, including some effects of complex dynamics due to packet loss and wireless collisions.

Off-the-shelf home automation technology is making it easier than ever for people to convert their own homes into smart homes. However, manual configuration is tedious and error-prone. In this article, we present and compare a family of solutions that automatically generate a map of the home and the devices within it using data from the smart home sensors themselves (e.g., light and motion sensors). These solutions can be used to automatically configure home automation systems or to automatically produce an intuitive map-like interface for visualizing sensor data and interacting with controllers. We call our approach Smart Blueprints because it automatically maps out the unique configuration of each smart home. We demonstrate the Smart Blueprints using a variety of sensor combinations, including light sensors, motion sensors, and magnetometers deployed on the doors and/or windows of the home. For evaluation of each combination on sensor-map generation, we deployed more than 200 sensors in seven different houses at different locations and compared the ability to use a variety of techniques to map out the configuration. We show that, in almost all houses, our system can automatically narrow the configuration down to 1--5 candidates per home using only one week of collected data.

Daylight harvesting is the use of natural sunlight to reduce the need for artificial lighting in buildings. The key challenge of daylight harvesting is to provide stable indoor lighting levels even though natural sunlight is not a stable light source. In this paper, we present a new technique called SunCast that improves lighting stability by predicting changes in future sunlight levels. The system has two parts: 1) it learns predictable sunlight patterns due to trees, nearby buildings, or other environmental factors, and 2) it controls the window transparency based on a quadratic optimization over predicted sunlight levels. To evaluate the system, we record daylight levels at 39 different windows for up to 12 weeks at a time, and apply our control algorithm on the data traces. Our results indicate that SunCast can reduce glare by 59% over a baseline approach with only a marginal increase in artificial lighting energy.

Homes are rich with information about people's energy consumption, medical health, and personal or family functions. In this paper, we present our experiences deploying large-scale residential sensing systems in over 20 homes. Deploying small-scale systems in homes can be deceptively easy, but in our deployments we encountered a phase transition in which deployment effort increases dramatically as residential deployments scale up in terms of 1) the number of nodes, 2) the length of time, and 3) the number of houses. In this paper, we distill our experiences down to a set of guidelines and design principles to help future deployments avoid the potential pitfalls of large-scale sensing in homes.

Heating, ventilation and cooling (HVAC) is the largest source of residential energy consumption. In this paper, we demonstrate how to use cheap and simple sensing technology to automatically sense occupancy and sleep patterns in a home, and how to use these patterns to save energy by automatically turning off the home's HVAC system. We call this approach the smart thermostat. We evaluate this approach by deploying sensors in 8 homes and comparing the expected energy usage of our algorithm against existing approaches. We demonstrate that our approach will achieve a 28% energy saving on average, at a cost of approximately $25 in sensors. In comparison, a commercially-available baseline approach that uses similar sensors saves only 6.8% energy on average, and actually increases energy consumption in 4 of the 8 households.

We present the Flash flooding protocol for rapid network flooding in wireless sensor networks. Traditional flooding protocols can be very slow because of neighborhood contention: nodes cannot propagate the flood until neighboring nodes have finished their transmissions. The Flash flooding protocol avoids this problem by allowing concurrent transmissions among neighboring nodes. It relies on the capture effect to ensure that each node receives the flood from at least one of its neighbors, and introduces new techniques to either recover from or prevent too many concurrent transmissions. We evaluate the Flash flooding protocol on both a 48-node wireless sensor network testbed and in a trace-based simulator. Our results indicate that the Flash flooding protocol can reduce latency by as much as 80%, achieving flooding latencies near the theoretical lower bound without sacrificing coverage, reliability or power consumption.

We demonstrate MetroNet, which is an application that illustrates collaborative data sharing on the Wide Wide Web. MetroNet has two parts. First, our sensors gather data about pedestrian foot traffic in front of and into stores, which is made available online to shopkeepers. Second, this sensor data can be made public, at the discretion of the shopkeeper, for use by city planners, other shopkeepers, or residents of the city. MetroNet is an application through which we study fundamental problems of sharing data on the Web, such as search, data fusion, and privacy.

In this paper, we present the Flash flooding protocol that exploits the capture effect to produce low-latency network floods. The capture effect is the ability of some radios to correctly receive one of several concurrently transmitted messages, even if the received strengths of the two messages are almost the same. We exploit this phenomenon in a network flooding scenario by allowing nodes to propagate the flooding message concurrently, thus reducing delays due to neighborhood contention. Our experimental results indicate that Flash can reduce latency by 75-80 percent without sacrificing flooding reliability or coverage.