Research Centre for Ubiquitous Computing
A. Location-Aware Computing:
|Description: Integrate machine learning for location estimation based on signal power levels and geometry model based methods. Focusing on inferring locations that are taken as the classes of grid-wise sampled locations, we investigates in details the k-nearest neighbor approach in RADAR under the location classification setting, and demonstrates that considering more neighboring signal strength measurements usually cannot help. Instead the orientation when the signal strength is taken should be more carefully treated. We developed a refinement step for RADAR, by building nearest neighbor classifiers to further clarify several top location estimates by RADAR. At a very economic cost, our refinement step can significantly boost the accuracy of location estimation.|
|Description: Our research focuses on location-aware computing based on signal information. We use several methods to make use of the signal information to provide the location service. We had built up a model to describe the relationship between the distance and the signal strength, and then we had chosen the fingerprint method to enhance the performance. In order to improve the performance, we try to combine the signals from downlink and uplink. Furthermore, we combine the GSM signal and the Wi-Fi signal to enhance the location service. In order to provide a stable and accurate positioning service, we investigate other factors such as the time variant of signal and patterns of the environment as our current work|
|Description: Most location estimation system based only on one positioning technology. Since each of the positioning technologies has its strength and weakness subject to different environment and limitations, we propose to develop a multi-channel location estimation system to get the best of both worlds and to investigate methods to merge all positioning technologies together and formed a more accurate and more stable location estimation system in the mobile community.|
|Description: Researchers have been proposing many location estimation algorithms each have its own strength and weakness subject to different environment. In this project, our approach is to construct an algorithm selector for location estimation. Utilizing AI and machine learning technique, the algorithm selector will select the best algorithm based on the observed environment and provide the best estimation for the user's location.|
|Description: User location data is now being actively recorded and is about to be published by mobile operators or service providers. It is thus an urge and critical problem to publish it without compromising user privacy. In this project, we show that traditional k-anonymity overprotects privacy and makes the published data less useful. Hence, we propose a local recoding algorithm as opposed to conventional hierarchy- or partition-based generalization adopted in k-anonymity.|
|Description: We investigate the location privacy issues involving mobile devices such as GPS-enabled PDAs, cell phones etc., in mobile environments, where the mobile users' movement can be traced easily by service providers. Their location privacy can be threatened if their moving path information is not handled carefully. We develop new database methods to safeguard the location privacy of mobile devices.|
|Description: In a cellular network, a cell-phone user constantly submits his/her location to the service provider in order to facilitate efficient location discovery (paging) for incoming calls. The phone service provider can correlate this information with a map to discover the user's location. We study how location privacy of users can be protected by proposing a method called "location cloaking", which blurs locations of the users.|
|Description: This project studies how to provide a secure service and communication in pervasive computing environments.|
Description: In the past few months, our research interest mainly focus on studying a state-of-art learning technique, named transfer learning, for indoor localization. Recall that learning-based localization techniques require a training process, whose underlying assumption is that the signal data distributions should be similar at training and testing. However, in practice, we find that this assumption is usually violated due to the signal variation over different environment factors such as time variation and device variation, etc. The intuitive solution by recollecting a large amount of new data for training is expensive in calibration effort. Therefore, we are trying to use transfer learning to help handle this problem.
|Description: We have studied various localization methods including the transceiver-free approach and RFID-based approach. We have proposed a real-time vehicle tracking system based on the RFID technology and a moving object counting system using ultrasonic sensors. We have also proposed an efficient and precise RFID tag counting method. A patent has been filed based on this research. Indoor localization based on RF signal strength still suffers from accuracy problem. We are still investigating if RF is truly useful to accurate indoor localization.|
B. Wireless LAN, RFID and Sensor Networks
|With the recent advance of wireless sensor networks (tiny intelligent devices with sensing, computation and communication capabilities) and embedded systems, we propose to make use of a sensor network to locate a useríŽs position or proximity and to estimate usersíŽ flow within a controlled environment. In this project, we also investigate the effects of temperature fluctuation and humidity towards location estimation within a wireless sensor network.|
|RFID (Radio Frequency Identification) is an important enabling technology for identification, retailing, and logistics and supply chain management systems. Effective deployment of RFID systems requires careful selection of RFID tags and readers, and their seamless integration into the target processes. Addressing the need, we propose to develop a set of sound RFID benchmarking methodologies and testing tools for the scientific classification and evaluation of RFID componentsíŽ performances. Traditional benchmarking technologies cannot be directly performed on RFID components due to several reasons. First, the performance of RFID components is affected by many variables. Their measurements are subject to natural environmental factors. Second, RFID performance is unlikely to be simulated or estimated algorithmically. Third, existing communications and radio frequency models were not originally developed for RFID. As such, we propose to conduct our RFID benchmarking research on three directions: combinatorial testing, machine learning and communications modeling.|
|Localization based on RF technologies has provided a much needed added-value to further expand the application domain of various RF-based applications. However, the dynamic of RF signals creates much uncertainty in location tracking. In this project, we study two RF-based platforms: wireless sensor networks and active RFID networks in the estimation of the location of moving objects. Both theoretical analysis and experimental test-beds are used to demonstrate the effectiveness of the proposed location tracking algorithms. .|
|Description: This project studies various power-saving issues in the design of wireless sensor networks, such as energy-efficient event detection methods, energy-efficient and reliable communications, and low-power constructions of self-organized sensor networks.|
|Description: Qiang Yang and his research group are working on data mining for sensor network, RFID and WiFi networks. One issue they are concerned about is how to determine the location of a mobile user in an online fashion through data mining and machine learning. A particular advancement of their group is to be able to reduce the calibration effort to a minimum while accurately determine the locations. Another issue is how to detect user activities at a high level from low level signals.|
|Description: We investigate how localize both the APs or routers and the client devices for radio signal strength based localization. We will study a variety of machine learning methods for this task.|
|Description: We have evaluated the performance of IEEE 802.11 infrastructure mode with intra-cell UDP traffic through reproducible experiments. We have developed an analytical model to precisely calculate the saturation throughput of IEEE 802.11 point-to-point link. We have designed a framework for the provisioning of parameterized QoS in IEEE 802.11e based wireless mesh networks. As for recent work, We will focus on the design of Voice over IEEE 802.11e based wireless mesh networks.|
|Description: In this project, we have so far focused on the formulation of important notions for micro/macroscopic analysis, as well as the behavioral characterization of a novel self-organizing algorithm for solving distributed, computationally hard problems, which are often found in an ad hoc distributed computing environment. In order to explore its performance, we have used some benchmark TSP instances, and have obtained promising results. We plan to extend the present algorithm further based on the Autonomy Oriented Computing (AOC) methodology, and in particular, to better understand and hence achieve its convergence, optimality, and efficiency in handling distributed, large-scale optimization problems.|
|Description: Despite their popularity, wireless sensor network systems are difficult to code. Developers of these systems need to deal with high-level application logics and low-level input/output operations simultaneously. These operations are usually implemented through interrupt mechanisms so that processors on motes can sleep to conserve energy. Explicitly handling interrupts in code makes programming such systems more error-prone. Researchers have reported data race, interface-contract violation, and memory safety problems due to the high degree of concurrency caused by interrupts. Interrupts also create challenges for testing these systems. An abrupt interrupt may occur at any time and interfere with the current code execution when the interrupt occurs through unforeseen dependencies. To detect failures due to interrupts, testers not only need to consider providing what values (sensing data and communication messages) to the system under test but also need to consider when to fire interrupts to produce undesired interleaving.
Based on the study carried out in this project, we observed that the scheduling policy for wireless sensor network systems is different from those for conventional programs. We have adapted existing concurrency testing techniques and made them tractable in testing wireless sensor network systems. We have formulated our observations as inter-context flow graphs, and further developed two test adequacy criteria for wireless sensor network systems. These two criteria extend conventional control-flow and data-flow criteria, respectively. We conducted experiments using a scalable sensor network simulator, Avrora, to evaluate our proposal. Empirical results showed that our extended criteria are more effective than their conventional counterparts in exposing faults, especially for those related to improper synchronization.
|Description: We have proposed a new opportunity-based topology control for wireless sensor networks. This paper received the best paper award in the 2008 IEEE International Conference on Distributed Computing Systems. We have studied a number of energy-efficient event detection methods in wireless sensor networks meeting different requirements. We are working on more generalized probabilistic topology control for wireless sensor networks. We are also working on if multiple channels can be used to further increase the network throughput while considering fairness among all sensor nodes.|
C. Middleware Support and Mobile Agent Platform
|Develop dedicated interfaces and methods for access to online forum and weblogs with mobile devices. The dedicated interfaces rely on machine learning modules to customize the required content for the mobile user. The learning modules involves: topic detection module, user interest tracking module and media summarization module. Using Markov logic network, user participation models can be developed to help us gain insights on the latent base topics of online discussions. Furthermore, joint non-negative matrix factorization model of participation and content data which can be viewed as a bipartite graph model between users and media. The factorizations allows simultaneous automatic discovery of leaders and sub-communities in the online forum as well as the core latent topics in the forum.|
|Description: Traditional Personal Response System was based on Infrared or RF technology with scale-up limitation. As the mobile phone has become a basic necessity for our daily life, we will develop a Mobile Personal Response System based on SMS technology for its convenient and flexibility for large-scale opinion polling.|
|Description: In this project, we design and implement platforms for ubiquitous access of real-time TV and interactive movies. In order to achieve user scalability and cost-effectiveness, we use the emerging peer-to-peer (P2P) technologies. In this project, we study and prototype the content distribution mechanisms, peer-to-peer TV and movie delivery to Internet and mobile users, multimedia codecs and presentation onto desktops and handhelds, peer-to-peer database, charging and security.|
|Description: Context-awareness is a key feature in pervasive computing whose environments keep evolving. The support of context-awareness requires comprehensive management of context consistency that includes inconsistency detection and repairing mechanism. In this project, we have developed a framework for realizing context consistency management. The framework consists of a context matching theory, an inconsistency triggering model and a proactive repairing mechanism. We incorporated our ideas by developing a pervasive context management middleware, Cabot. The feasibility of the framework and its performance were evaluated through a real case and a simulated experiment, respectively. The results were disseminated in our recent publications.|
|Description: We have been working on a model called UIO (Ubiquitous Intelligent Object) and a middleware framework based on the model. Our idea is that, contracting to traditional top-down design and programming approach, we argue that objects and entities in a pervasive computing environment are more autonomous and should combine both physical and behavior properties when being developed. So we are seeking for a bottom up approach to develop system from UIO's view of the pervasive computing environment - UIOs are capable of being identified, sensing their environments, discovering neighboring UIOs and interacting with them, cooperating and collaborating with UIOs and environments to provide user-specified services. Following this approach, we are developing algorithms and protocols for UIOs to discover services and collaborate in heterogeneous, wireless ad hoc networking environment. We are also implementing system prototypes of middleware functions. Based on the framework we are investigating various applications, including smart space and ubiquitous searching in both cyber and physical worlds.|
|Description: Mobile agents are units of computation that control where they execute. This form of mobility is supported by a software infrastructure called a mobile agent system. Mobile agent is being used for long since in large scale distributed system as a paradigm for code mobility. Due to its properties such as asynchronous execution, autonomy, being able to be personalized, and being able to support disconnected operations, MA has a great potential of being used as a mobile and pervasive computing technology. We have been developing MDAgent, a mobile agent-enabled platform and context-aware reflective middleware. The platform allows mobile agent-based applications to be dispatched to network servers and support the operations and coordination of mobile agents. The middleware is modeled as a component based middleware consisting of components which work together to achieve a common goal. The functional components of the middleware are - Context Providers, Context Manager, Context Consumers and Context-aware entities. We have been developing mechanisms and algorithms for these functional components, as well as mobile and pervasive computing applications run on the platform with the support of the middleware. We are also investigating other pervasive computing issues such as service discovery and application mobility.|
|Description: We have successfully defined the framework of an adaptive mobile computing platform that supports context-awareness for evolving mobile applications. The framework aims to provide a level of abstractions that facilitates ease of developing future mobile applications that seamlessly integrates context awareness, while requiring developers to specify only semantic requirements of the system. To date, we have developed a fuzzy model that provides appropriate control to balance between requested QOS and available resources available on the system.
Implementation of the system is underway and progress is satisfactory. On completing the implementation, we will conduct a series of tests and evaluations on the system to validate its performance and to study the complex interactions among the core modules of the system.
D. Mobile Data Management
|Description: This proposed research exploits the trade-off between data quality and communication cost to improve energy efficiency. An adaptive strategy has been developed to maximize the accuracy of data collected by the base station over the network lifetime. We have also proposed mobile filter that explores migration of filters to maximize overall traffic reduction for error-bounded data collection.|
Description: We extend the query processing techniques making use of client cache in mobile ubiquitous environments in two dimensions. First, we consider location-dependent spatial queries (LDSQs) where the current client locations are used as query points and the queried spatial objects are those around the query points. We look into the validity of the cache containing spatial objects by considering the data semantics of the cached result. Second, we investigate into the support of continuous query processing by extending existing query processing on one-time LDSQs.
|Description: The processing of kNN and continuous kNN queries on spatial network databases (SNDB) has been intensively studied recently. However, there is a lack of systematic study on the computation of network distances, which is the most fundamental difference between a road network and a Euclidean space. Since the online Dijkstra's algorithm has been shown to be efficient only for short distances, we propose an efficient index, called distance signature, for distance computation and query processing over long distances. Distance signature discretizes the distances between objects and network nodes into categories and then encodes these categories. To minimize the storage and search costs, we present the optimal category partition, and the encoding and compression algorithms for the signatures, based on a simplified network topology. By mathematical analysis and experimental study, we showed that the signature index is efficient and robust for various data distributions, query workloads, parameter settings and network updates.|
|Description: A continuous nearest neighbor (CNN) search, which retrieves the nearest neighbors corresponding to every point in a given query line segment, is important for location-based services such as vehicular navigation and tourist guides. It is infeasible to answer a CNN search by issuing a traditional nearest neighbor query at every point of the line segment due to the large number of queries generated and the overhead on bandwidth. Algorithms have been proposed recently to support CNN search in the traditional client-server systems but not in the environment of wireless data broadcast, where uplink communication channels from mobile devices to the server are not available. In this project, we develop a generalized search algorithm for continuous k-nearest neighbors based on Hilbert Curve Index in wireless data broadcast systems. A performance evaluation is conducted to compare the proposed search algorithms with an algorithm based on R-tree Air Index. The result shows that the Hilbert Curve Index-based algorithm is more energy efficient than the R-tree-based algorithm.|
|Description: We have been working on a personal privacy protection method for location-based services. The main idea is to transform location data before being sent to the service provider, so that the service provider can process the transformed dataset. Our technique not only prevents the service provider from knowing the exact locations of users, but also protects information about user movements and locations from being disclosed to other users who are not authorized to access this information. We also define a privacy model to analyze our framework. The effectiveness and scalability of our approach are validated through detailed experiments. We have published our results in the ACM GIS SPRINGLúż08 workshop in 2008 and the Transactions on Data Privacy Journal in 2009.|
|Description: Broadcast is an effective data dissemination technique for ubiquitous computing. Scheduling algorithm is the core any broadcast mechanism. Currently, we are designing a suite of scheduling algorithms that are aimed at providing quality services to mobile clients and conserving broadcast bandwidth at the server.
Existing scheduling solutions mostly focussed on a restricted scenario such as push-based broadcast, non-real-time, single-item request and single-channel architecture. When these restrictions are relaxed, existing solutions are suffering from a few performance problems. We are going to design a suite of scheduling algorithms that can perform satisfactorily in a realistic environment.
|Description: Developing methods for access to online forum and weblogs with mobile devices.|
|Counting the population of tags in the region of interest is one of the most important tasks in large-scale RFID systems. Due to the long latency, tag identification based counting schemes are often impractical. We propose two protocols, Lottery Frame (LoF), and Precise and Anonymous Counting (PAC). LoF scheme can estimate the tag population in a very short period with a high accuracy. PAC counts the tag cardinality precisely, and remarkably reduces the processing time and energy cost in comparison to identification protocols. We believe both of them will make the large-scale RFID systems more efficient and effective, especially in high mobility scenarios.|