Location Estimation and Trajectory Prediction for Cellular Networks ...

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 53, NO. 6, NOVEMBER 2004

1903

Location Estimation and Trajectory Prediction for Cellular Networks With Mobile Base Stations Pubudu N. Pathirana, Andrey V. Savkin, and Sanjay Jha

Abstract—This paper provides mobility estimation and prediction for a variant of the GSM network that resembles an ad hoc wireless mobile network in which base stations and users are both mobile. We propose using a Robust Extended Kalman Filter (REKF) to derive an estimate of the mobile user’s next mobile base station from the user’s location, heading, and altitude, to improve connection reliability and bandwidth efficiency of the underlying system. Our analysis demonstrates that our algorithm can successfully track the mobile users with less system complexity, as it requires measurements from only one or two closest mobile base stations. Further, the technique is robust against system uncertainties caused by the inherent deterministic nature of the mobility model. Through simulation, we show the accuracy of our prediction algorithm and the simplicity of its implementation. Index Terms—Ad hoc networks, CarNet, location tracking, mobility modeling, robust extended Kalman filter (REKF).

I. INTRODUCTION

A

MOBILE Ad Hoc Network (MANET) is a wireless network consisting of mobile nodes capable of communicating with each other without the help of any fixed infrastructure. MANETs date from the 1970s when they were known as DARPA packet radio networks. Recently, there has been renewed interest in such networks because of the availability of smaller, smarter and cheaper portable computers, inexpensive wireless technology, small wireless sensors, and mobile users’ demand for information “anywhere any time.” The IETF has established a special working group for developing standard protocols for such networks [1]. MANETs are self-organizing networks built dynamically in the presence of nodes equipped with radio interface devices. The nodes are capable of movement in an arbitrary fashion. All the functionality of routing and switching are carried out by the nodes themselves. When two nodes are within communicating range of each other, they can exchange information directly. However, when two nodes are not within each other’s communicating range, they can still communicate with each other provided there are nodes in between who can pass the data packets for the communicating nodes. Communication, in this later case, occurs in a multihop

Manuscript received January 26, 2003; revised December 4, 2003, April 21, 2004, and May 25, 2004. This work was supported by the Australian Research Council. P. N. Pathirana is with the School of Engineering and Technology, Deakin University, VIC 3217, Australia (e-mail [email protected]). A. V. Savkin is with the School of Electrical Engineering and Telecommunications, University of New South Wales, Sydney 2052, Australia (e-mail: [email protected]). S. Jha is with the School of Computer Science and Engineering, University of New South Wales, Sydney 2052, Australia (e-mail: [email protected]). Digital Object Identifier 10.1109/TVT.2004.836967

fashion [2]. These networks are designed for temporary and special use, such as on battlefields or in emergency rescue operations where there may not be any established infrastructure for networking. Research in MANETs has given rise to many new network architectures. The Multihop cellular Architecture [3] provides localized ad hoc networking within a cell, where mobile hosts within the cell help each other to forward packets to the base station. By using multihopping, the cellular architecture can expand the cell coverage while maintaining the transmission range of the base station. As a result, reduced numbers of base stations are possible. Lin et al. [3] used WLAN (802.11) for their experimentation and have shown that the resulting throughput can be higher than for single-hop-based networks. The Terminode project [4] looks at developing a wide-area, autonomous, self-organized, wireless multimedia network that is totally independent of any fixed infrastructure. The Grid project at the MIT Laboratory for Computer Science deployed a test bed network composed of cars – CarNet [5]. The CarNet project attempts to equip all test bed cars with an IEEE 802.11 radio, Linux box and a GPS receiver, to demonstrate their Grid architecture. The sample services for CarNet include traffic congestion monitoring, fleet tracking, and highway chat (similar to CB radio). The High Altitude Aeronautical Platform (HAAP) [6] uses undedicated aircrafts to support mobile coverage. The aircrafts (e.g., commercial planes) act as satellites to provide city-wide coverage while they are appropriately positioned (location and altitude). The HAAP architecture uses redundancies to provide continuous coverage while planes enter or leave the city, or prepare for landing. Location tracking (also known as mobility tracking or mobility management) is the set of mechanisms by which location information is updated in response to mobility of a communication endpoint [7]. Many approaches such as Mobile IP try to hide changes in access point by redirecting packets but maintaining the same IP address. There are several situations where knowledge of location could be beneficial. Development of location-aware services is a very active area of research in both cellular wireless and ad hoc/sensor network communities. It is, therefore, necessary to manage the mobility of the terminals in a cellular network for smooth operation of real-time applications. Mobility tracking based on signal strength measurements is solved by treating it as on-line estimation in a nonlinear dynamic system. For example, the extended Kalman filter has been used to solve this problem in [8] and [9]. Traditionally, in GSM-type networks, the base stations are statically located. In this paper, we relax this assumption by assuming that the base stations are free to move randomly and

0018-9545/04$20.00 © 2004 IEEE

1904


organize themselves arbitrarily; thus, the network’s wireless topology may change rapidly and unpredictably. This paper presents the use of the Robust Extended Kalman Filter (REKF) to predict a mobile user’s arrival in the next cell, based on new theoretical results presented in [10], [11]. Our first implementation, with a single mobile base station, uses only the measurement from the closest neighboring station and, hence, improves computational efficiency. Our second proposal further improves the prediction performance by using measurements from only the two base stations that are closest to the current cell. This eliminates the need to sample with six GSM base stations as required in [8], and hence considerably reduces the network traffic while improving the computational efficiency in using this algorithm. Further, in this paper we propose a much more realistic model in which vehicle acceleration may be any bounded function of time as opposed to the stochastic model given in [8]. This model is not realistic in practice, as acceleration of a real vehicle cannot be represented by a Gaussian stochastic process. Moreover, our model incorporates significant uncertainty and measurement errors. Simulation results are provided to demonstrate the superiority of the proposed algorithm. Our work assumes availability of on-board GPS and acceleration of the wireless base stations but not of the mobile terminals. Beside the example scenarios given above, our paper can be useful in tactical environments for accurate prediction of location of various objects and targets. Sensor networks can benefit from this work where a few powerful robot-based sensors (acting as base stations) can roam around and collect data from small static (or mobile) sensor devices scattered in the environment. The rest of this paper is organized as follows. Section II presents related work with Section III emphasizing the system dynamic model that we consider with the nonlinear measurement model. With this model, in Section IV we present the theoretical background for the robust extended Kalman filter as a state estimator with reference to set-valued state estimation ideas. We give implementation details and simulation results in Section V with the conclusions given in Section VI.

II. RELATED WORK Most of the recent work on location management in ad hoc networks has been in the sensor network area where the term used is localization (determining the position of a sensor device in some coordinate system). Examples of indoor localization are the Cricket project at MIT [12], [13], and work by Savvides et al. [14]. Bulusu et al. [15] provide a good overview of such work and their own scheme for outdoor localization for very small devices. However, requirements for localization for these small devices are different from the application scenarios that we have discussed. Most of the work close to ours in the networking community has been in the area of wireless cellular networks (robotics and military applications may use these techniques). We provide a brief overview of a few sample studies. The location management approach has two components: location update and location prediction. As a passive strategy, in

location updates the system periodically records the current location of the mobile in some database that it maintains. Location update algorithms can either be static or dynamic depending on whether the location updating is triggered from network topology or users’ call and mobility patterns. Location prediction is a dynamic strategy in which the system proactively estimates the mobile’s location based on a user movement model. Most recent studies have focused mainly on the update method [16]–[18], and less attention has been given to the prediction side. Accurate prediction of a mobile terminal based on its previous location will improve the efficiency of the location management task even from the update and systems perspective. The task of location management and resource reservation will become easier if the user’s movement pattern is known in advance. However, even if the destination and proposed trajectory are known, a user may choose a different route while in transit because of traffic congestion. Tabbane [19] proposed that a mobile terminal’s location can be derived from its quasideterministic mobility behavior and can be represented as a set of movements in a user profile. Mobile Motion Prediction (MMP), which uses pattern matching and pattern recognition, has been proposed as an enhancement of Tabbane’s method [20]. Bhattacharya et al. [21] used an information-theoretic approach to characterize the complexity of the mobility tracking problem in a cellular network. Shannon’s entropy measure is identified as a basis for comparing user mobility models. By building and maintaining a dictionary of individual users’ path updates, the proposed adaptive on-line algorithm can learn subscribers’ profiles. These and several other similar schemes do not perform well when random factors are re-introduced or assumptions such as those regarding rectilinear movement patterns are removed. Extended Kalman filter technique has been applied in [8], [22]. Yang et al. [23] proposed application of sequential Monte Carlo (SMC) methodology to the problem of joint mobility tracking and hard hand-off detection. This is computationally expensive, although it is based on mobile user dynamic model assumptions. As it has been demonstrated that further improvements can be achieved via efficient prediction, in this paper we propose using a REKF as a state estimator in predicting a mobile user’s expected trajectory for efficient allocation of resources. These robust state estimation ideas emerged from the work of Savkin and Petersen [10]. This approach not only provides satisfactory results [24], but also eliminates the requirement of the knowledge or modeling of the user mobility pattern and measurement noise as required by the standard Kalman filter implementation presented in [8]. III. MOBILE BASE STATION/MOBILE USER DYNAMIC MODEL We use the terminology Car for a mobile base station in this paper (other mobile vehicles or robots fitted with a base station would fit into the same category). The dynamic model for the th car, and the mobile user to be used in this approach can be given in two-dimensional (2-D) Cartesian coordinates as [25] (1)

PATHIRANA et al.: CELLULAR NETWORKS WITH MOBILE BASE STATIONS

1905

is a constant determined by the transmitted power, where . is a slope index (typiwavelength, and antenna gain of cally two for highways and four for microcells in the city), and is the logarithm of the shadowing component, which is repreconsidered as an uncertainty in the measurement. , sents the distance between the mobile and base station of which can be further expressed in terms of the mobile’s position with respect to the location of the th car, i.e., (4)

Fig. 1.

Network geometry.

(2) The

dynamic

vector , where and represent the position of the user with respect to the base station (the th car) at time , and their and represent the relative first-order derivatives and directions. In other words, if speed along the represents the absolute state (position and velocity in order in the and directions, respectively) of the mobile user, and denotes the absolute state of the th car in the same order, then . denote the 2-D driving and acceleration Furthermore, let commands of the car from the respective accelerometer denote the unknown 2-D driving and readings and let acceleration commands of the mobile user.

In [8], three independent distance measurements are used to locate a moving user in a 2-D domain, as GSM systems sample the forward link signal levels of six neighboring cells. Here, we propose a twofold implementation scheme. Our first implementation algorithm uses a single base station measurement that is closest to the user (i.e., the highest value of the sampled six neighboring stations) as opposed to three. In our second implementation, further improvements can be made by using progressive measurement from two base stations. In this case we use the two closest cars. For the first implementation, the measurement equation is

state

A. Measurement Model In cellular systems, the distance between the mobile and a known base station is practically observable. Such information is inherent in the forward link received signal strength indication (RSSI) of a reachable base station. Measured in decibels at the mobile station, RSSI can be modeled as having two components: one from path loss and one from shadow fading [8]. Fast fading is neglected, assuming that a low-pass filter is used to attenuate Rayleigh or Rician fade. Denoting the th car as (Fig. 1), the RSSI from , , can be formulated as [26] (3)

(5) with denoting the number of cars in the network, and for the second implementation, we use two measurements to form the observation vector

They are chosen progressively as the user moves in the coverage area. We use measurements from the two closest base stations and therefore the measurement equation is in the form (6) where with (see (7) at the bottom of the page) where refers to the car nearest to the mobile user and corresponds to the next nearest car. IV. SET-VALUE STATE ESTIMATION WITH A NONLINEAR SIGNAL MODEL We consider a nonlinear uncertain system of the form

(8)

(7)

1906


as a general form of the system given by (1) with its measurement equation in the form of (6), and defined on the finite time . Here, denotes the state of the system, interval is the measured output, and is the unand certainty output. The uncertainty inputs are . In addition, is the known control input. We assume that all of the functions appearing in (8) have continuous and bounded partial derivatives. Additionally, we asis bounded. This was assumed to simplify the sume that mathematical derivations and can be removed in practice [11], is assumed to be independent of , and is [27]. The matrix of full rank. The uncertainty in the system is defined by the following nonlinear integral constraint [10], [11], [28]–[30] : (9)

B. Robust Extended Kalman Filter Here we consider an approximation to the PDE (11), which leads to a Kalman filter-like characterization of the set . Petersen and Savkin in [11] presented this as an extended Kalman filter version of the solution to the set-value state estimation problem for a linear plant with the uncertainty described by an Integral Quadratic Constraint (IQC). This IQC is also presented as a special case of (9). We consider the uncertain system described by (8) and an IQC of the form

(13) where , can be written as

and

. For (8), (13), the PDE (11)

where is a positive real number. Here, , and are bounded nonnegative functions with continuous partial derivatives satisfying growth conditions of the type (10) is the Euclidean norm with , and . Uncertainty inputs satisfying this condition are called admissible uncertainties. We consider the problem of characterizing the set of all posof (8) at time which are consistent with sible states a given control input and a given output path ; i.e., if and only if there exists admissible uncertainties such is the control input and and are resulting that if and , for all . trajectories, then

(14)

where

We obtain an approximate solution [11] to the PDE (14) by approximating with a function of the form (15) The resulting (16),(17), and (18) define our approximate solution to the PDE (14)

A. State Estimator is characterized in terms of level The state estimation set sets of the solution of the PDE [see (11) at the bottom of the page]. The PDE (11) can be viewed as a filter, taking observations and producing the set as output. The state of this filter is the function ; thus is an information state for the state estimation problem. Theorem 4.1: Assume the uncertain system (8), (9) satisfies the assumptions given above. Then the corresponding set of possible states is given by

(16) is defined as the solution to the Riccati differential equation (RDE)

(17) and

(12) where

is the unique viscosity solution of (11) in . Proof: see [11].

(18)

(11)


1907

TABLE I SIMULATION PARAMETERS

TABLE II ACCELERATIONS OF DYNAMIC ENTITIES

higher weightings). The initial state ( ) is the estimated state of the respective systems at acquisition or handover time. This initial state is essentially derived from the terminal state of the previous system together with other data available in the network (i.e., vehicle position and speed) to be used as the initial state for the next system taking over the tracking. With an uncertainty relationship of the form of (19), the inherent measurement noise [see (6)], the unknown mobile user acceleration and driving command, and the uncertainty in the initial condition are considered as bounded deterministic uncertain inputs. In particular, the measurement equation with the standard norm bounded uncertainty can be written as [(6)] (20) is a constant indicating the upper bound of the where norm-bounded portion of the noise. By choosing and (21) Hence, it follows from Theorem 4.1 that an approximate foris given by mula for the set

Considering the form of

and the uncertainty

satisfying the bound in (22)

This amounts to the so-called REKF generalization presented in [11]. In the application of REKF to CarNet, the th system (Car and the mobile user) tracking the mobile user during a corresponding time interval is represented by the nonlinear uncertain system in (8) together with the following IQC [from (13)]

(19) Here , and with are the weighting matrices for system . For each system , these parameters represent the relative weightings among the unknowns: the mobile user’s acceleration, measurement noise and the uncertainty in the initial condition. Therefore, the relative effect of these inputs on system performance can be controlled by adjusting these parameters (i.e., more dominant inputs can have

it is clear that this uncertain system leads to the satisfaction of the condition in inequality (9) and, hence, (13) (see [11]). This more realistic approach removes any noise model assumptions in the development of the algorithm and guarantees its robustness. C. Robust Versus Optimal State Estimation The REKF was introduced in the previous subsection. It tends to increase the robustness of the state estimation process and reduce the chance that a small deviation from the Gaussian process in the system noise causes a significant negative impact on the solution. However, we lose optimality and our solution will be just suboptimal. To explain the connection between REKF and the standard extended Kalman filter, consider (8) with (23) is some bounded function, and is a pawhere rameter. Then, the REKF estimate for (8), (23), (13) defined as tends to 0. Here is the by (16), (17) converges to

1908


Fig. 2.

Trajectories of the cars and the mobile user with single base station measurements.

Fig. 3.

Actual and estimated velocities of the mobile user with single base station measurement.

extended Kalman state estimate for (8) with the Gaussian noise satisfying

See, e.g., [31]. The parameter in (23) describes the size of uncertainty in the system and measurement noise. For small our robust state estimate approaches the Kalman state estimate with Gaussian noise; for larger we achieve more robustness

but less optimality. Hence, we always have some tradeoff between robustness and optimality. We will show below via computer simulations that with larger uncertainty (which is quite realistic) our robust filter still performs well whereas the standard extended Kalman estimate diverges. V. SIMULATION To examine the performance of the REKF implemented on a CarNet-type network, simple simulations were carried out for


1909

Fig. 4. Trajectories of the cars and the mobile user with dual base station measurements.

Fig. 5.

Actual and estimated velocities of the mobile user with dual base station measurements.

a mobile user in a three car coverage area. The network is assumed to have location and acceleration information for the mobile base stations via GPS and accelerometer readings, while no such information is available with respect to the mobile user of an arbitrary kind. We simulated the two scenarios that we introduced earlier. • Measurement from the closest base station is used – the strongest signal measured from the three neighboring base stations. The tracking is performed by the closest car.

• The two base stations closest to the user are used for measurement. Tracking is performed by the closest car while the network also makes measurements from the second closest car. The simulated service area contains three cars for illustrative purposes and can obviously be scaled for as many mobile base stations and users as required. Identical parameters (Table I) were used for our simulations in each case for comparison purposes.

1910

Fig. 6.


Error in location estimation for single and dual car measurements.

Fig. 7. REKF and the standard Kalman filter for location estimation for single car measurement.

In the simulation of the dynamic system, we chose the functions given in Table II for arbitrary car accelerations ( s) and and uniunknown mobile user acceleration ( ), with . formly distributed random variables in the interval is simply a constant used in the simulation indicating Here the magnitude of the deterministic component of the unknown ). mobile user acceleration (

and For the measurement noise we use (20). It is evident that there exists a constant such that

in

(24) is satisfied and, hence, (22), which leads to the satisfaction of the inequality (13).


Fig. 8.

1911

Comparison of REKF and the standard Kalman filter for location estimation error for single car measurement.

The equation for the state estimation and the corresponding Riccati differential equation obtained from (16) and (17) are as follows:

TABLE III CARS TRACKING THE MOBILE USER DURING THE TIME PERIOD (SINGLE BASE STATION MEASUREMENT)

(25)

(26) where for single base station measurement (27) with corresponding to the closest base station, and for two base station measurements

as shown in (7). Here (28) is the last state before the handover. A. Discussion of Results The simulation of mobility modeling and trajectory tracking of a mobile user purely based on signal strength measurements by one or two cars was performed successfully in a 15 40 km suburban area. We have restricted the number of mobile agents

to a minimum to ensure the clarity and simplicity in demonstration but the approach can obviously be scaled to as many agents as required. We chose realistic values for the simulation parameters (i.e., velocities and accelerations of vehicles) to depict a real application. In the first scenario, a single car (the closest car) measures the forward link signal in the GSM system, tracks the mobile user’s location and predicts the velocity, as shown in Figs. 2 and 3. The respective handover times and corresponding vehicles are given in Table III. The second implementation we propose ensures further improvements by additionally involving the second closest car to measure the forward link signal. Using identical parameters as for single base station measurement, our second scenario produced improved tacking of position and velocity of the mobile user. The significant improvement in location estimation is shown in Fig. 4 and the performance comparison/improvement in location estimation of the two scenarios is shown in Fig. 6. Comparison of the estimated velocity in the first case (Fig. 3) with the estimated velocity in Fig. 5 for the second case shows the improvement in estimation. Estimation error comparison with the standard Kalman filter for single car measurements is shown in Figs. 7 and 8 and shows the efficiency in using this algorithm as compared with the standard extended ) to Kalman filter. Here we used noisier measurements ( demonstrate that, while the standard Kalman filter diverges, the REKF continues to track the actual trajectory in spite of these

1912


TABLE IV CARS TRACKING THE MOBILE USER DURING THE TIME PERIOD (DUAL BASE STATION MEASUREMENT)

large noise inputs. The tracking performed by the closest car and the forward link measurements by the two closest cars are shown in Table IV with the corresponding tracking/measuring and handover times. VI. CONCLUSION We have provided a scheme for mobility estimation and prediction for an ad hoc network consisting of mobile base stations and mobile users. To the best of our knowledge there have been no other studies of such networks. The desire was to develop an effective, robust and easily implementable algorithm with less burden on the system resources while effectively using readily available mobile media such as cars. We proposed using a REKF-based state estimation algorithm. It is evident from our research that a single mobile base station can successfully be used to track mobile users in the wireless network. Further improvements in overall system performance can be achieved by our second proposed technique, which additionally uses the next closest car for measurement purposes. Emerging from recent theoretical developments, REKFs can successfully be used in the prediction of a mobile user’s location in a wireless ad hoc network with trackers and measurers switching appropriately. As our implementation with a single base station uses only the measurement from the closest neighboring station, the computational efficiency of the overall network is significantly improved. Our second proposal further improves system performance by using measurements from only the two base stations that are closest to the mobile user, eliminating the need to sample with more (six) base stations, which is the approach used with immobile base stations. This considerably reduces the network traffic while improving the computational efficiency in using this algorithm. This algorithm is clearly more computationally efficient than the extended Kalman filter implementation provided in many PCS networks. It can also be implemented within the mobile user, if necessary, rather than in the base station to reduce the signaling traffic. Further, as no assumptions were made on the measurement noise and uncertain user acceleration component, the robustness of this algorithm is ensured. In future, we will consider extending this work to outdoor localization of small sensor devices. REFERENCES [1]

(2002) Mobile Ad-Hoc Networks MANET Working Group, Mobile Ad-Hoc Networks Charter. IETF. [Online]. Available: http://www.ietf.org/html.charters/manet-charter.html [2] E. Royer and C. Toh, “A review of current routing protocols for ad-hoc mobile wireless networks,” IEEE Pers. Commun. Mag., vol. 6, pp. 46–55, Apr. 1999.

[3] Y. Lin and Y. Hsu, “Multihop cellular: A new architecture for wireless communications,” in Proc. Conf. Computer Commun. (IEEE INFOCOM), July 2000, pp. 1273–1282. [4] J. Hubaux, J. Boudec, S. Giordano, and M. Hamdi, “The terminode project: Toward mobile ad-hoc WANs,” in Proc. MOMUC’99, San Diego, CA, 1999. [5] R. Morris, J. Jannotti, F. Kaashoek, J. Li, and D. Decouto, “Carnet: A scalable ad hoc wireless network system,” in 9th ACM SIGOPS European Workshop – Beyond the PC: New Challenges for the Operating System, Kolding, Denmark, Sept. 2000. [6] V. Pandiarajan and L. Joiner, “Undedicated HAAP based architecture for cellular data transfers,” in Proc. IEEE SoutheastCon 2000, Apr. 2000, pp. 23–26. [7] I. F. Akyildiz and J. S. M. Ho, “On location management for personal communications networks,” IEEE Comun. Mag., vol. 34, pp. 138–145, Sept. 1996. [8] T. Liu, P. Bahl, and I. Chlamtac, “Mobility modeling, location tracking, and trajectory prediction in wireless ATM networks,” IEEE J. Select. Areas Commun., vol. 16, pp. 922–936, Aug. 1998. [9] M. Hellebrandt and R. Mathar, “Location tracking of mobiles in cellular radio networks,” IEEE Trans. Veh. Technol., vol. 48, pp. 1558–1562, Sept. 1999. [10] A. Savkin and I. Petersen, “Recursive state estimation for uncertain systems with an integral quadratic constraint,” IEEE Trans. Automat. Contr., vol. 40, pp. 1080–1083, 1995. [11] I. Petersen and A. Savkin, Robust Kalman Filtering for Signals and Systems with Large Uncertainities. Boston, MA: Birkhauser, 1999. [12] N. B. Priyantha, A. Chakraborty, and H. Balakrishnan, “The cricket location-support system,” in Proc. 6th Annual Int. Conf. Mobile Computing and Networking (MobiCom), Boston, MA, Aug. 2000, pp. 32–43. [13] N. B. Priyantha, A. K. L. Miu, H. Balakrishnan, and S. Teller, “The cricket compass for context-aware mobile applications,” in Proc. 7th Annual Int. Conf. Mobile Computing and Networking (MobiCom), Rome, Italy, July 2001, pp. 1–14. [14] A. Savvides, C. Han, and M. B. Strivastava, “Dynamic fine-grained localization in ad-hoc networks of sensors,” in Proc. 7th Annual Int. Conf. Mobile Computing and Networking (MobiCom), Rome, Italy, July 2001, pp. 166–179. [15] N. Bulusu, J. Heidemann, and D. Estrin, “GPS-less low-cost outdoor localization for very small devices,” IEEE Pers. Commun. Mag., vol. 7, pp. 28–34, Oct. 2000. [16] V. Wong and V. Leung, “An adaptive distance-based location update algorithm for next-generation PCS networks,” IEEE J. Select. Areas Commun., vol. 19, pp. 1942–1952, Oct. 2001. [17] B. Akyol and D. Cox, “Signalling alternatives in a wireless ATM,” IEEE J. Select. Areas Commun., vol. 16, pp. 35–49, Jan. 1997. [18] M. Mouly and M. Pautet, The GSM System for Mobile Communications, 1992, ISBN 2-9 507 190-0-7. [19] S. Tabbane, “An alternative strategy for location tracking,” IEEE J. Select. Areas Commun., vol. 13, pp. 880–892, June 1995. [20] G. Liu and G. Maguire, “A predictive mobility management scheme for supporting wireless mobile computing,” in Proc. IEEE Int. Conf. Universal Personal Communication, 1995. [21] A. Bhattacharya and S. K. Das, “LeZi-update: An information-theoretic approach to track mobile users in PCS networks,” in Proc. ACM/IEEE Int. Conf. Mobile Computing and Networking (MobiCom’99), Seattle, WA, Aug. 1999, pp. 1–12. [22] J. Ho, Y. Lin, and I. Akylidiz, “Movement-based location update and selective paging for PCS networks,” IEEE/ACM Trans. Networking, vol. 4, pp. 629–639, Aug. 1996. [23] X. Wang and Z. Yang, “Joint mobility tracking and hard handoff in cellular networks via sequential monte carlo fitlering,” in Proc. Conf. Computer Communications (IEEE INFOCOM), New York, June 2002. [24] P. Pathirana and A. Savkin, “Precision missile guidance with angle only measurements,” in Proc. Information Decision and Control Conf., Adelaide, Australia, Feb. 2002. [25] A. Savkin, P. Pathirana, and F. Faruqi, “The problem of precision missile guidance LQR and H control frameworks,” IEEE Trans. Aerosp. Electron. Syst., vol. 39, pp. 901–910, July 2003. [26] H. Xia, “An analytical model for predicting path loss in urban and suburban environments,” in PIRMC, 1996. [27] M. James and I. Petersen, “Nonlinear state estimation for uncertain systems with an integral constraint,” IEEE Trans. Signal Processing, vol. 46, pp. 2926–2937, Nov. 1998. [28] A. Savkin and I. Petersen, “A connection between control and the absolute stabilizability of uncertain systems,” Syst. Contr. Lett., vol. 23, no. 3, pp. 197–203, 1994.

H


[29]

, “Nonlinear versus linear control in the absolute stabilizability of uncertain linear systems with an integral quadratic constraint,” IEEE Trans. Automat. Contr., vol. 40, pp. 122–127, 1995. [30] A. Savkin and R. Evans, Hybrid Dynamical Systems :Controller and Sensor Switching Problems. Boston: Birkhäuser, 2002. [31] B. Anderson and J. Moore, Optimal Filtering. Englewood Cliffs, NJ: Prentice-Hall, 1979.

Pubudu N. Pathirana was born in 1970 in Matara, Sri Lanka. He was educated at Royal College Colombo. He received the B.E. (first class honors) degree in electrical engineering, the B.Sc. degree in mathematics in 1996, and the Ph.D. degree in electrical engineering in 2000 from the University of Western Australia, sponsored by the government of Australia on EMSS and IPRS scholarships, respectively. During 1997–1998, he was a Research Engineer in industry in Singapore and Sri Lanka. He was a Postdoctoral Research Fellow with Oxford University, U.K., in 2001, a Research Fellow with the School of Electrical Engineering and Telecommunications, University of New South Wales, Sydney, and a consultant to the Defence Science and Technology organization (DSTO) Australia, 2002. Currently, he is with the School of Engineering and Technology, Deakin University Australia. His current research interests include missile guidance, autonomous systems, target tracking, computer integrated manufacturing, vision based navigation systems, quality of service (QoS) management, and mobile/wireless internet.

1913

Andrey V. Savkin was born in 1965 in Norilsk, U.S.S.R. He received the M.S. degree in mathematics and the Ph.D. degree in applied mathematics in 1987 and 1991, respectively, from Leningrad State University, Leningrad, U.S.S.R. From 1987 to 1992, he was with the All-Union Television Research Institute, Leningrad. From 1992 to 1994, he held a Postdoctoral position with the Department of Electrical Engineering, Australian Defence Force Academy, Canberra. From 1994 to 1996, he was a Research Fellow with the Department of Electrical and Electronic Engineering and the Cooperative Research Center for Sensor Signal and Information Processing, University of Melbourne, Australia. In 1996, he was a Senior Lecturer, and then an Associate Professor with the Department of Electrical and Electronic Engineering, University of Western Australia, Perth. Since 2000, he has been a Professor with the School of Electrical Engineering and Telecommunications, University of New South Wales, Sydney. Since 2002, he has also been the Director for the Centre of Excellence in Guidance and Control. His current research interests include robust control and filtering, hybrid dynamical systems, missile guidance, networked control systems and control of networks, computer-integrated manufacturing, control of mobile robots, and application of control and signal processing to biomedical engineering and medicine. He has published four books and numerous journal and conference papers on these topics. He served as an Associate Editor for several international journals and conferences.

Sanjay Jha received the Ph.D. degree from the University of Technology, Sydney, Australia. He is an Associate Professor of Networks with the School of Computer Science and Engineering (CSE), University of New South Wales (UNSW), Sydney, Australia. He is the Head of the Network Research Group at CSE, and a project leader for the Smart Internet CRC. His research activities cover a wide range of topics in including Wireless Adhoc and sensor networking, quality of service (QoS), and active/programmable network. Currently, he is supervising several Ph.D. degree candidates at UNSW. He has authored papers/articles in high-quality forum and holds two provisional patents. He is the principal author of the book Engineering Internet QoS (Dedham, MA: Artech House, 2002) and is coeditor of the book Wireless Sensor Networks: A Systems Perspective. He has been working as an industry consultant for major organizations such as Canon Research Lab (CISRA), Lucent, and Fujitsu. He was a visiting scholar with the Distributed Computing and Communications Laboratory, Computer Science Department, Columbia University, NY, and Transmission Systems Development Department of Fujitsu Australia Ltd., Sydney.