Alan Marchiori

Using circuit-level power measurements in household energy management systems

The first requirement for any intelligent household energy management system is to be able to accurately measure energy usage in the home. Measuring energy usage is not difficult, however we must decide what to measure. Whole-home energy measurement is cheap and easy to setup because only one sensor is placed where the home connects to the power grid. The collected data can provide useful information for large appliances. However, the only way to monitor the energy usage of smaller devices is to install an energy meter on every device of interest. This creates a very detailed picture of household energy consumption, but requires a lot of additional hardware---one meter per device in the home. This paper explores an alternative, more practical, approach to monitor household energy usage including small devices. Our approach uses circuit-level power measurements and a new method to separate aggregate data into device-level estimates. Our initial evaluation resulted in an average error less than 5:35% for three devices with good response to changing device state. We therefore believe that this approach, coupled with a device-level control system, would create an ideal architecture for the next generation of household energy management systems.

Distributed wireless control for building energy management

Automated building energy management systems are essential to enabling the development of mass-market, low-energy buildings. In existing and future buildings, the impacts of occupant behaviors contribute significantly to the total energy efficiency. As building technologies and materials improve, the relative impact of behavioral factors is more significant. We propose a general framework where building systems can share information in order to optimize performance. To be successful, such a system must be responsive, intuitive, robust, and scalable. As a first step toward achieving these goals, we present a prototype distributed control system for building energy management that uses wireless sensor network-class nodes. Using protocol independent multicast, sensors and controllers are allowed to efficiently share information in a distributed peer-to-peer fashion. Our prototype system achieved an energy savings of 7.1% - 14.6% by implementing a relatively simple control policy. Based on the results of this this work we have identified three key areas for future work.

Realistic performance analysis of WSN protocols through trace based simulation

It is a difficult endeavor to realistically evaluate the performance of wireless sensor network (WSN) protocols. Generic network simulators are often used, but they tend to rely on synthetic models. Because WSN performance can be affected by many subtle features, these simulators lack a certain level of realism. The most realistic performance assessment is to implement the protocol in question and observe its performance in a real world deployment. This approach, however, is time consuming, costly, and makes the direct comparison of various protocols nearly impossible. We believe there exists a need to evaluate the real-world performance of WSN protocols in a controlled and repeatable fashion. To that end, we enable trace based WSN simulation by first enhancing an existing WSN profiler that automates the collection of connectivity traces and the generation of statistical link properties. We then present a trace based WSN simulator built on the discrete event simulator SimPy using the standard Python. The use of the high level language Python allows new WSN protocols to be rapidly prototyped and evaluated under the real-world conditions captured by the WSN profiler. To validate the premise that our simulation results closely model the real-world performance of the same protocol, we present a thorough performance analysis of the modern WSN collection tree protocol (CTP). Our approach enables the creation and use of a WSN trace database collected from various deployment environments. Such a database could be used to both fairly and more realistically benchmark existing WSN protocols and provide timely feedback on the real-world performance of protocols still in the development process.

PIM-WSN: Efficient multicast for IPv6 wireless sensor networks

We present PIM-WSN, a protocol independent multicast (PIM) protocol tailored for IPv6 wireless sensor networks (WSNs). Existing solutions for multicast in WSNs are limited because they either support multicast only from a single source node (usually the root node) or they limit the multicast group size to constrain memory usage. Our design allows any node to be a mulitcast source with an unlimited number of subscribers. We constrain the memory usage by approximating multicast group membership using a fixed sized Bloom filter. The efficiency of the protocol degrades as the false positive rate of the Bloom filter increases; however, correct operation is always maintained. Using detailed simulations we show that PIM-WSN achieves 1) high packet delivery rate (over 97%), 2) low latency per hop (less than 5 ms), and 3) lower radio utilization than all other comparable protocols (by more than 50%). Using a ten-hop testbed of TelosB motes we have verified our implementation of PIM-WSN for TinyOS 2.× with the Blip IPv6 networking stack which uses only 5,978 bytes of ROM and 235 bytes of RAM.

CBIR for medical images - an evaluation trial

Content-based image retrieval (CBIR) has the potential to provide medical doctors with a powerful resource to help make accurate diagnoses. To aid in diagnosis, a CBIR system must retrieve similar images from the same (unknown) disease class as the patient. We have implemented a CBIR system that first predicts the disease class of the query image and then retrieves the n images nearest to the query image from the pool of images with the predicted disease class. With the cooperation of residents/radiologists at Indiana University Medical Center and the Department of Radiology at the University of Wisconsin we have recently completed an evaluation of our system. The results show that when using our system, the diagnostic accuracy of the group increased on average by 32% over diagnosis without any reference materials.

Enabling distributed building control with wireless sensor networks

Energy is a precious resource and as electronic devices (e.g., tablets, smart phones, Internet-connected appliances) and electric cars become ubiquitous, the demand for energy will continue to increase. To meet this demand and enable the integration of renewable resources, electrical distribution networks are beginning to adopt smart-grid technologies by providing bidirectional communication between the distribution grid and the electric meter. The question of how appliances within the home will interface with the smart-grid is still an open question. We propose to use wireless sensor networks as a platform to enable detailed household monitoring, demand response, load scheduling, occupancy-based control, intelligent lighting, and other general-purpose control strategies. In this vision, the smart-grid interface is a single input to the building control system which consists of a wide variety of other sensors and supports many applications that greatly extend the capabilities of the smart-grid. To explore this concept we have: developed efficient and accurate energy disaggregation algorithms, developed reliable multicast and broadcast network protocols, evaluated 10 strategies for energy savings, designed a WSN-based building monitoring and control platform, and evaluated this approach in both offices and a residential home.

Crowdsourcing low-power wide-area IoT networks

The Internet of Things (IoT) promises to allow everyday objects to connect to the Internet and seamlessly interact with users and other machines. For this vital Internet connection, most current IoT devices use a personal gateway device such as a smartphone or a home WiFi access point. The necessity of configuring and maintaining these gateways presents an additional burden for both users and developers of IoT applications. Our vision for IoT connectivity is to eliminate the need for the personal gateway by developing crowdsourced low-power wide area networks (csLPWAN). Recent technologies such as RPMA, LoRa, and R-FDMA enable links to reach 15km using ISM-band transceivers, making csLPWANs an attractive option. In this paper, we investigate the practicality of csLPWANs and develop the first csLPWAN planning tool, PlanIt, which combines topography-aware RF signal analysis with demographic data to predict LPWAN coverage in specific geographic areas. Using PlanIt, we find that most cities achieve 99% coverage by deploying a single LPWAN base station within the city. To provide better service on the csLPWAN, we propose and evaluate D-QN, a near-optimal MAC protocol to enable efficient bandwidth sharing in highly utilized networks. In the future, csLPWANs could accommodate a heterogeneous set of IoT applications, simplifying the IoT application development cycle, reducing total system cost, improving application reliability, and enhancing the user experience.

A Two-Stage Bootloader to Support Multi-application Deployment and Switching in Wireless Sensor Networks

Wireless sensor networks are built from highly resource constrained embedded systems. Supporting multiple applications on the sensor network is a desirable goal, however, these constraints make supporting multiple concurrent applications on each node difficult. Therefore, we propose a dynamic application switching approach where only a single application is active on each sensor node at a time. In this paper we present a dynamic application switching framework that can automatically reprogram the sensor node in response to application requests. We implement our framework on the TelosB platform and evaluate its performance using a 52-node sensor network testbed. The implementation of a two-stage bootloader reduces the memory requirements to only 1KiB of program memory and 8 bytes of RAM on this platform. We evaluate the implementation using two different modes of application switching; asynchronous and synchronous. Extensive performance studies indicate that dynamic application switching using our two-stage bootloader is a useful approach to support multiple applications in wireless sensor networks.

PlanIt and DQ-N for Low-Power Wide-Area Networks: Demo Abstract

Low-Power Wide-Area networks promise to deliver limited IoT payloads reliably at distances in excess of 10 km raising the possibility of thousands of IoT devices connected to a single base station. PlanIt is a web application able to visualize connectivity in these large-scale deployments prior to deploying hardware or even making a single signal-strength measurement. DQ-N is an adaptation of distributed queuing, a hybrid MAC protocol, compatible with current LoRa packet radios that significantly increases the number of nodes that can be supported from a signal base station. This demo will allow users to visualize LoRa coverage for the entire USA and demonstrate DQ-N using several sensor nodes.

Maximizing coverage in low-power wide-area IoT networks

The Internet of Things (IoT) promises to allow everyday objects to connect to the Internet and seamlessly interact with users and other machines. Essential for the IoT to function is a reliable Internet connection. In 2016 the International Telecommunication Union reports 3.9 billion people - 53% of the worlds population are not using the Internet [1]. Projects like Loon (X) and Aquila (Facebook) aim to solve this connectivity gap using atmospheric satellites to deliver 4G-like signals to underserved regions. With the recent interest in low-power wide-area networks (LPWAN) in the license-free ISM bands, we consider using atmospheric satellites to improve coverage in LPWAN networks. We find that LPWAN technologies are compatible with atmospheric satellites and demonstrate that significant connectivity gains are possible by locating an LPWAN base station at altitude from 1 km – 28 km when compared to a typical ground-based base station.