A Hierarchical Position Prediction Algorithm for Efficient Management of Resources in Cellular Networks Tong Liu, Paramvir Bahl, Imrich Chlamtac Tellabs Wireless Systems Division Microsoft Research Erik Jonsson School of Engineering and Computer Science, The University of Texas at Dallas GLOBECOM 97, November 1997
Main Messages … mobility prediction is a promising technique for improving resource efficiency and connection reliability in cellular networks. … theoretical richness of stochastic signal processing field makes it feasible for predicting random intercell movement… Bi-level stochastic movement model Approximate pattern matching Extended, self-learning Kalman Filtering Intercell movement prediction
Outline Mobility prediction - Problem and Framework Related work in literature Hierarchical Position Prediction User Mobility Model - A Global View Approximate Pattern Matching User Mobility Model - A Local View Extended, Self-learning Kalman Filtering Prediction Performance Conclusions
Mobility Prediction - Problem Description Global Prediction: Next-cell(s) Crossing ? Local Prediction: Dynamic State ?
Mobility Prediction - Problem Description ?
Mobility Prediction - Framework Movement Observation Movement Model Cell Geometry Prediction Algorithm Position Speed Cell siteTime Improve lifetime connectivity and radio resource efficiency - Bandwidth Reservation - QoS Control - Optimal Routing - Position/velocity Based Handoff Global Local Global Local
Related Work in Literature Recently Crossed Cells Pattern Matching Next Cell Tabbane (JSAC, 1995) Liu and Maguire (ICUPC,1995) Liu, Munro and Barton (ICUPC, 1996) Next-cell prediction based on movement pattern Prediction Performance Historical Movement Pattern
Related Work in Literature Cell Transition Probability Matrix Look Up Table Next Cell Bar-Noy, Kessler and Sidi (Jour. Of Wireless Networks, 1995) Akyildiz and Ho(Proc. ACM SIGCOMM, 1995) Liu and Maguire (ICUPC,1995) Prediction Performance Next-cell prediction based on Markov Chain model Current Cell ID
Prediction of Random Intercell Movement Cell Geometry Position Speed Prediction Performance Pattern Template Linear Dynamic System Approximate Pattern Matching Extended Self-Learning Kalman Filter Recent Crossed Cells RSS Measurement Next Cells
User Mobility Model - A Global View User Mobility Pattern Editing Process inserting changing deleting User Actual Path UAP Editing Operation UMP b1b1 b2b2 b3b3 b1b1 b2b2 b3b3 a UAP Spatial Cost Insertion deletion change b1b1 b3b3 b4b4 b2b2 b1b1 b3b3 b4b4 b1b1 b2b2 b3b3 b3b3 b4b4 b4b4 b1b1 b2b2 a
User Mobility Model - A Global View Spatial Cost
Approximate Pattern Matching
User Mobility Model - A Local View S 1 S m S 2 Moving Dynamics Commands Time correlated random acceleration r(t) + U(t) a(t) F( ) Measurement noise Nonlinear measurement -A max S1S1 S2S2 SmSm A max P( a(t)/S 1 )P( a(t)/S 2 ) P( a(t)/S m ) +
Dynamic Equations Continuous-time dynamic equation: Discrete-time dynamic equation:
Observation Model d0d0 d1d1 d2d2
Adaptive Dynamic State Estimator
Recursive Algorithm
Prediction of Next Cell Trajectory Direction cell 0 cell 6 cell 1 cell 2 cell 3 cell 4 cell 5
Hierarchical Position Prediction User Profile Approximate Pattern Matching Global Prediction User Mobility Buffer size:L UAP Forming Optimum Adaptive Filtering Local Prediction of Next Cell Dynamic state RSS measurement Local Prediction Global Prediction
Significance of Local Prediction Movement Pattern 2 Movement Pattern 1 ---Crossed Cell ---Uncrossed Cell A practical situation necessitates looking-ahead mode for movement pattern identification
Prediction Performance - Simulation Parameters
Prediction Performance - Trajectory Tracking
Prediction Performance - Speed Estimation
Local Prediction of Next Cell
Parametric Behavior of Next-cell Prediction
Local Prediction of Next Cell
Global Prediction d(UAP,UMP1) Current Cell C8C8 C9C9 C 10 d(UAP,UMP2) 3 Global Prediction C9C9 C 10 C 18 C 17 C 16 C 17 C 16 Determine Edit Distance: Prediction Result: UAP UMP UAP UMP
Conclusion Hierarchical Movement Model Top level: Movement Pattern subject to random editing operations Bottom level: A linear dynamic system driven by the combination of a semi-Markovian process and Color Gaussian Noise. Hierarchical Position Prediction Algorithm Approximate Pattern Matching Extended Self-learning Kalman Filter