A Predictive Bandwidth Reservation Scheme Using Mobile Positioning and Road Topology Information

IEEE/ACM Transactions on Networking,Vol. 14, Iss. 5, Oct. 2006. pp:1078 - 1091 A Predictive Bandwidth Reservation Scheme Using Mobile Positioning and Road Topology Information Wee-Seng Soh andDepartment ofElectrical and Computer Engineering, National University of Singapore (NUS)Hyong S. KimCarnegie Mellon University, Pittsburgh, PA, where he is currently the Drew D. Perkin’s Chaired Professor of Electrical and Computer Engineering

Outline • Introduction • Mobility Prediction Module • Prediction database • Mobility prediction algorithm • Dynamic Bandwidth Reservation Module • System module • Logic behind our approach • Adjusting Tthreshold at each BS • Adjusting Rtarget at each BS • Simulations and Results • Conclusion

Introduction • A call may be dropped during a handoff attempt when the new cell does not have sufficient bandwidth. • The call blocking probability of new calls (PCB)vs. the handoff dropping probability (PHD) target. • Static approach [1] • a fixed portion of the radio capacity is permanently reserved for incoming handoffs • cannot handle variable load and mobility • a dynamic approach that adjusts the reservation according to anticipated handoffs may potentially result in a lower PCB for the same PHD target. [1] D. Hong and S. S. Rappaport, “Trafﬁc model and performance analysis for cellular mobile radio telephone systems with prioritized and non-prioritized handoff procedures,” IEEE Trans. Veh. Technol., vol. VT-35, no. 3, pp. 77–92, Aug. 1986.

Introduction (cont.) • The best tradeoff between PCB and PHD can only be achieved if the dynamics of every mobile terminal (MT) are known in advance. • such as its path and its arrival/departure times in each cell • However, such an ideal scenario is unlikely. • The next best option is to predict their mobility • Pattern matching techniques and a self-adaptive extended Kalman filter [3] • for next-cell prediction based on cell sequence observations, signal strength, and cell geometry assumptions • shadow cluster [4]: a set of BSs to which a MT is likely to attach in the near future • estimates each MT’s probability of being in any cell within the cluster for future time intervals, based on its dynamics and call holding patterns in the form of pdfs [3] T. Liu, P. Bahl, and I. Chlamtac, “Mobility modeling, location tracking, and trajectory prediction in wireless ATM networks,” IEEE J. Sel. Areas Commun., vol. 16, no. 6, pp. 922–936, Aug. 1998. [4] D. A. Levine, I. F. Akyildiz, and M. Naghshineh, “A resource estimation and call admission algorithm for wireless multimedia networks using the shadow cluster concept,” IEEE/ACM Trans. Networking, vol. 5, no. 1, pp. 1–12, Feb. 1997.

Introduction (cont.) • In the U.S., the FCC mandates that cellular-service providers must be able to pinpoint a wireless emergency call’s location to within 125 m. • According to[10], assisted GPS positioning methods could yield an accuracy of 20 m during 67% of the time. • During 2003–2009, a new batch of GPS satellites will be launched to include two additional civilian carrier frequencies that could potentially yield an accuracy of 1 m [11]. • The European Space Agency also plans to launch their own system (GALILEO), which targets an accuracy of 1 m (95% of the time within 10 m) [12]. • Give rise to better prediction accuracy and greater adaptability to time-varying conditions than previous methods, • which is crucial for more timely and efficient reservations [10] Y. Zhao, “Standardization of mobile phone positioning for 3G systems,” IEEE Commun. Mag., vol. 40, no. 7, pp. 108–116, Jul. 2002. [11] E. A. Bretz, “X marks the spot, maybe,” IEEE Spectrum, vol. 37, no. 4, pp. 26–36, Apr. 2000. [12] J. Benedicto, S. E. Dinwiddy, G. Gatti, R. Lucas, and M. Lugert, “GALILEO: Satellite System Design and Technology Developments,” European Space Agency, Tech. Rep., 2000.

Introduction (cont.) • has not addressed that the cell boundaryis fuzzy and irregularly shaped • previous attempts in [3], [6], [8] to perform positioning based predictive reservation • either hexagonal or circular boundaries are assumed. • no previous work has utilized road topology information for predictive reservation until we first proposed the idea in [13] and [14]. • only incoming handoff predictions at each BS are used to adjust its reservation. • more efficient tradeoffs between and may be achieved if bothincoming and outgoing handoff predictions are used. [3] T. Liu, P. Bahl, and I. Chlamtac, “Mobility modeling, location tracking, and trajectory prediction in wireless ATM networks,” IEEE J. Sel. Areas Commun., vol. 16, no. 6, pp. 922–936, Aug. 1998. [6] M.-H. Chiu and M. A. Bassiouni, “Predictive schemes for handoff prioritization in cellular networks based on mobile positioning,” IEEE J. Sel. Areas Commun., vol. 18, no. 3, pp. 510–522, Mar. 2000. [8] A. Aljadhai and T. Znati, “Predictive mobility support for QoS provisioning in mobile wireless environments,” IEEE J. Sel. Areas Commun., vol. 19, no. 10, pp. 1915–1930, Oct. 2001. [13] W.-S. Soh andH. S.Kim, “QoS provisioning in cellular networks based on mobility prediction techniques,” in Proc. World Telecommunications Congress (WTC), Paris, France, Sep. 2002. [14] W.-S. Soh andH. S.Kim, “QoS provisioning in cellular networks based on mobility prediction techniques,” IEEE Commun. Mag., vol. 41, no. 1, pp. 86–92, Jan. 2003.

Introduction (cont.) • In [9], a reservation scheme that utilizes mobility predictions based on mobile positioning information. • the first scheme that considers irregular cell boundaries. • the cell boundary is approximated as a series of points around the BS • uses linear extrapolation from a MT’s recent positions to predict its handoff cell and time • In this paper, we utilizes road topology knowledge, in addition to the MT’s positioning information. • for predicting the MT’s future handoff time and target cell, • so as to adjust the reservations • Not to decide whether a handoff should be initiated • more accurate predictions at the cost of increased complexity, • but the resulting improvement in reservation efficiency may justify this cost [9] W.-S. Soh andH. S. Kim, “Dynamic guard bandwidth scheme for wireless broadband networks,” in Proc. IEEE INFOCOM, Anchorage, AK, Apr. 2001, pp. 572–581.

Logic behind our approach

Mobility Prediction Module • Each MT in an active call reports its position to the serving BS every ΔT (say, 1 s) • The predictions are performed periodically by the BSs • Each BS maintains a database • During a prediction, the BS (expected to have sufficient computational and storage resources) • identifies a set of possible paths from every active MT’s current position that may lead to a handoff within a threshold time Tthreshold • generates a 4-tuple for each such path, in the form of • [ target cell, prediction weight, lower prediction limit (LPL), upper prediction limit (UPL)]

Prediction Database • The database stores the essential information • identity of neighboring segments at each junction; • a road segment = a junction pair (j1, j2) • bended road segment can be broken down into lines • two-way roads are stored separately • handoff-probable segment (HPS) • several hundred KB to several MB memory are required • transition probabilities between neighboring segments • computed from the paths taken by previous MTs • statistical data of time taken to transit each segment; • statistical data regarding possible handoffs along each segment • such as probability of handoff, time in segment before handoff, and handoff positions.

Prediction Database (cont.) • In reality, the transition probabilities between neighboring road segments would likely vary with time and traffic. • We model the transition between neighboring road segments as a second-order Markov process, • and assume stationarity between the database updates to simplify the computations. • MT1: P[sk+1 = EF| sk=BE,sk-1=CB] • MT2: P[sk+1 = EF| sk=BE,sk-1=AB] • The first-order conditional distribution is also needed, if sk-1 is unknown, • P[sk+1 = EF| sk=BE]

y1 x1 x1 b b a a y2 x2 x2 y3 x3 x3 只評估在這些segment上的MT是否需要reservation pdf of time taken to transit sab 最有可能handoff到哪個cell pdf of time spent on sab before handoff pdf of handoff時remaining distance to jb small X is enough (2 or 3) pdf of 過了sab還剩下多久會handoff, 0.5th quantile

Prediction Database (cont.)

Mobility Prediction Algorithm set of 4-tuple prediction for MT i 還有多遠會到jb 還有多久會到jb 最可能handoff過去的cell的Tthreshold

System Module • Preclude soft handoffs in CDMA systems • consider applications that require fixed bandwidth guarantees. • assume that the minimum bandwidth granularity that may be allocated to any call is 1 bandwidth unit (BU). • For example, a voice call may require 1 BU, while a CBR video call may require several (4) BUs. • each BS j has a fixed capacity of C(j) BUs • A new call is admitted if • A handoff call is admitted if

Logic Behind Our Approach • it is possible to reduce PHD by giving the BS earlier notice, which could be done by increasing Tthreshold. • But 預測精準度會降低, 較可能發生over-reservation Rtarget • it is undesirable when • anincoming handoff occurs earlier than its predicted time, andalso • an outgoing handoff occurs later than its predictedtime.

Adjusting Tthreshold at each BS • μ: a scaling factor close to 1 • ideally, μ=1 • but through simulations, the PHD obtained this way slightly deviates from the desired target • (about 1.2~1.25). • A possible explanation for this observation is that • handoffs are bursty and • the best that our adaptive algorithm could achieve is to allow the value of to fluctuate around its optimal value.

Adjusting Rtarget at each BS • Rtarget is adjusted when: • a predicted incoming handoff has taken place; • a predicted incoming handoff has either ended its call without handoff, or has handed off to a different cell. • The BS needs to be informed by the neighboring BS (of the predicted incoming handoff). • a predicted outgoing handoff has either handed off or ended its call; of its outgoing calls The overall peak discovered is then assigned to Rtarget

Simulation Model • handoff positions are randomly distributed • When a MT is between 1.1R and 1.2R from the BS, a handoff will occur during its transit through this region • The time at which the handoff occurs is a random variable • uniformly distributed over the total time spent in the region. • The target BS is assumed to be the nearest neighboring BS at the time when the handoff occurs • Traffic lights are introduced at the road junctions • change every 30 s • speed limit of each segment equally chosen from {40, 50, 60} km/h • The MT’s speed is Gaussian-distributed, • Its mean is the speed limit of that segment. • The standard deviation is assumed to be 5 km/h, and • the speed is truncated to a limit of three standard deviations from its mean.

Simulation Model (cont.) • Each cell has a fixed capacity C of 100 BUs • Tpredict = 5 s • New calls are generated according to Poisson distribution with rate λ (calls/second/cell) in each cell. • The initial position of a new call and its destination can be on any road segment with equal probability. • A voice and video call: 1 and 4 BU, • made with probability Rvo and 1- Rvo, (set to 0.5) • All MTs’ bandwidth requirements are symmetric • having the same PHD requirement • The holding times for both types of calls are assumed to be exponentially distributed, with mean 180 s • Normalized offered load per cell

Schemes Simulated for Comparison • Road topology based (RTB) scheme • Benchmark scheme • perfect knowledge about every MT • uses the same algorithms for adjusting Tthreshold and Rtarget • Reactive scheme • No predictions • adapt the BS’s Rtarget according to the number of handoffs dropped over handoff-requests • Linear Extrapolation (LE) scheme • proposed in [9] • No road topology information • RTB with path knowledge (RTB_PK) • a MT’s path is always known • In practice, it might be possible to predict a MT’s path using an adaptive algorithm [20] that could learn a user’s mobility profile • Did not know the exact time and position that the handoff might occur

Choi’s AC1 scheme [2] • by estimating the probability that a MT would hand off into a neighboring cell within an estimation time window Test, • based upon its previous cell, and its extant sojourn time. (footprint) • The neighboring cell’s Rtarget is then increased by the MT’s bandwidth requirement, • weighted by the estimated probability. • The Test of each cell is dynamically adjusted based on the measured handoff dropping ratio among a number of handoffs recently observed, so as to meet the desired . [2] S. Choi and K. G. Shin, “Adaptive bandwidth reservation and admission control in QoS-sensitive cellular networks,” IEEE Trans. Parallel Distrib. Syst., vol. 13, no. 9, pp. 882–897, Sep. 2002.

Simulation Results • LE RTB, RTB_PK • Positioning info. • Consider both incoming and outgoing handoff • RTB_PK • eliminates the uncertainty in predicting the MTs’ future paths • With Tpredict = 10 s • the PCB’s of the RTB_PK, RTB, and LE schemes increased by no more than 0.003 (or 2% of their original PCB’s), and • still outperform Choi’s AC1 scheme significantly

For the LE and RTB schemes, even at Lnorm = 1.5 , the PHD’s are approximately 0.0125 and 0.011, respectively. • the Benchmark scheme uses the same algorithms presented in Sections III-C and III-D to adapt its Tthreshold and Rtarget, • we can conclude that our proposed algorithms used work extremely well

This again demonstrates the advantages of using positioning information for predictions

using the RTB scheme

An exponentially correlated Gaussian random variable with zero mean, standard deviation σ, and correlation coefficient is e-1/τ added to the MT’s actual position • the degree of correlation is determined by τ, whereby τ≒ 0 second gives uncorrelated errors. • the MT’s speed is estimated using several of its most recently reported positions. • When the errors are uncorrelated, the RTB scheme’s performance deteriorates considerably as σ increases • when the errors are correlated, the RTB scheme’s performance does not deteriorate as much • the RTB scheme is more sensitive towards errors in speedestimation, rather than the actual positioning errors

Conclusion • A predictive bandwidth reservation scheme that utilizes mobile positioning and road topology information. • Did not assume that the cell boundaries are hexagonal or circular • More practical • The mobility prediction module defines the prediction tasks to be undertaken by the BSs. • The periodically generated 4-tuple predictions are used by the dynamic bandwidth reservation module for adjusting the reservations. • uses both incoming and outgoing handoff predictions to achieve more efficient reservations • Can be implemented to work in real-time

Conclusion (cont.) • Six schemes are simulated and compared • The huge jump in efficiency from Choi’s AC1 scheme to the LE scheme highlights the advantages of using positioning information for predictions. • With the advantage from using road topology information, the RTB scheme outperforms the LE scheme. • RTB_PK scheme shows limited improvement over the RTB scheme. • little incentive to implement an RTB-RTB_PK hybrid scheme • RTB only degrades modestly when the prediction time interval (Tpredict) increases from 5 s to 10 s. • the PHD only deviates slightly from its target when the normalized load (Lnorm) is 1.5 • The variants of the Benchmark, LE, and RTB schemes that do not consider outgoing handoffs • exhibit significant degradation when compared to their original schemes • still outperform Choi’s AC1 scheme • demonstrating the advantages of using positioning information for predictions • the RTB scheme is more sensitive to errors in speed estimation, rather than the actual positioning errors

A Predictive Bandwidth Reservation Scheme Using Mobile Positioning and Road Topology Information