CSE 620 Advanced Networks Concepts Location Management in Mobile Networks Ramesh Ravipati Veeraiah P. Gorantla Department of Computer Science and Engineering State University of New York at Buffalo Contents • Introduction • Location Management Techniques • Location Prediction Schemes - Traveling Demand Model - Mobility Prediction through Neuro-Fuzzy Theory Introduction Problem: How do we keep track of the location of the Mobile Host (MH)s in a mobile network? Why ? To provide better communication services in a mobile environment and provide better performance for such networks. We try to keep track of the Mobile Hosts through location management techniques. A number of techniques are being used for this purpose. This is to locate the user in a mobile network rapidly so as to improve the performance of the communications network. Location Management • Attempt to keep track of the mobile hosts in real time, as the hosts move through the network. • The location records have to be updated frequently so as to keep the location management in real-time. Location Prediction • Attempt to predict the location of the mobile host at a future instant so as to provide services to the user without break or interruption. • Good mobility models have to be used for accurate predictions. Location Management Techniques • Updating - Time-based updating - Distance-based updating - Movement-based updating • Paging - Random updating - Sequential updating • Caching location records (centralized) • Distributed location strategies Advantages • Simple to implement • Intuitively easy to understand Disadvantages • Limited resources of bandwidth, power etc. • Brings down the performance of communications network due to constant updates to the location records Location Prediction Schemes • We attempt to predict the future location of a mobile host or user • The future location of a host can be predicted from the present location based on given specified conditions • To predict the future location of a user, a mobility model has to be utilized • This mobility model simulates the movement of the hosts through the network • The mobility model to be selected has to be realistic enough to model the behavior of the hosts Mobility Models • Random Walk - The direction and velocity of the hosts are random variables • Fluid Flow - The users can move in a direction uniformly distributed between [0,2Π] with a given average velocity • Rule based Models - The movement of the users is defined by a given specified set of rules In this presentation, we present two location prediction techniques in mobile environments: • User Mobility Prediction by Traveling Demand Model - Uses a rule-based mobility model (traveling demand model) • Mobility Prediction Scheme based on Neuro-Fuzzy Theory - Uses neuro-fuzzy theory to create a mobility model User Mobility Prediction by Traveling Demand Model • The key point : Users have regular mobility patterns in their motion behavior • Users can move through several cells in their mobility patterns • Hence user mobility should not be parameterized independently for each cell • The characteristics of the mobility behavior have to be determined over the entire network (or a relevant part of the network) Traveling Demand Model In this model, the individual hosts make choices or decisions pertaining to the question of their mobility : 1. Given the present geographical location i, a time period t, and an activity A, the host first decides whether to travel or not 2. Given the choice made at the first level of decision, the host then chooses a location j for the conduct of the given activity 3. Given the outcomes of the first two decisions, the host then decides the transportation mode m to use, among the various alternative modes available between i and j 4. Given the outcomes of all the preceding decisions, the traveler chooses a route r among those available for the trip Traveler Decision Process Structure Formulating the Model The event that mobile users move from one location to another follows the exponential distribution given as ƒ(t) = (1/D’) • exp(-t/D’) where D’ is the mean duration time at an arbitrary location The mean duration time at an arbitrary location can be given by D(ρ) = - D’ • log(1 - ρ) where ρ Є [0,1] is the probability that mobile users stay in an arbitrary location and is selected using a uniform distribution Given the Φ destinations by which mobile users pass, the following parameter is required to obtain a mean duration time at i th destination ( i Є{1,2,3,….., Φ) Di is the mean duration time at i th destination In a real mobile environment, however,an equivalent mean duration time is not observed To approximate real-world behaviors sufficiently closely, every destination should have a different mean duration time. We now consider the traveling mode which users decide between two destinations • Mobility modeling should include changes in both the movement velocity and direction of mobile users A movement direction, θij, between two destination,Di and Dj , is calculated as θij = θ’ + (2П/m) • k where θ’ is the angle for the major direction m is the number of movement directions in [0,2П] k Є {1,2,3,…..,m-1} Let us consider the movement velocity of mobile users Mobile users have the following velocity between two destinations, Di and Dj vij = vij’ + v where vij is the actual velocity of the mobile user vij’ is the mean velocity between the two locations and v is the variation in the mean velocity Pij = probability vector for the movement between two locations Di and Dj Rij = route probability vector for choosing among k routes between Di and Dj Example of Movement Direction Simulation Simulation Factors region size Sx x Sy cell radius Cr initial location D0 number of destinations Φ mean duration time at each destination D i’ number of directions m Simulation Factors (contd…) direction probability vector between two destinations mean velocity and velocity variation between two destinations Pij vij’ , v number of routes k route probability vector Rij Simulation Environment We assume 5 locations D0 to D4 in the network Sx = 900 , Sy = 850 Cr = 1.0 km Hence we get a cell network approximately 25 x 28 Mean velocity of the mobile user = 10 km/h We assume Unit of time is hour Unit of velocity is km/h Trajectory of Mobile Users for Various Change of Velocity Trajectory of Mobile Users for m = 6 Trajectory of Mobile Users for m = 12 Simulation Results Features of the Traveling Demand Model • We can see from the simulation that mobile users follow a regular pattern • By these patterns, if the present position of the user is known, we can find the future location at a given instant • Such location prediction strategies improve the performance of the communication network by decreasing call-blocking probability and call-dropping Mobility Prediction Scheme based on Neuro-Fuzzy Theory • The key point : Use of neuro-fuzzy theory to estimate the future position of a mobile user • The system autonomously learns about the motion behavior of the user • Locations are obtained by approximate reasoning of the neuro-fuzzy inference system • High accuracy for prediction of various user patterns • Also can be used with considerable accuracy to users whose history has not been learned by the system Movement paths of mobile users Components of a mobility prediction scheme 1. Pre-processing component - user mobility data generation and classification 2. The Inference system - mobility learning and neuro-fuzzy inference 3. Mobility Prediction component - location prediction analysis Pre-processing Inference System Location Prediction As in the previous model, we consider the following factors • Given the present geographical location i, a time period t, and an activity A, the host first decides whether to travel or not • Given the choice made at the first level of decision, the host then chooses a location j for the conduct of the given activity • Given the outcomes of the first two decisions, the host then decides the transportation mode m to use, among the various alternative modes available between i and j • Given the outcomes of all the preceding decisions, the traveler chooses a route r among those available for the trip Once we have a history of movements of the users, we can then use the inference system to predict the future locations of the users The Inference System consists of 4 parts 1. Neuro-fuzzy Model - simplified fuzzy inference model 2. Fuzzy Rulebase - set of rules in which mobility history of users is represented as linguistic values 3. Mobility data Clustering - grouping of data into clusters for easy access and retrieval 4. Mobility Learning - procedures for learning mobility patterns using neural networks Motivation for Neuro-fuzzy theory • Existing methods are based on direct observation in the current mobile situation • Such methods have poor prediction performance, especially if the current location of the user is previously unknown • This approach to mobility prediction attempts to derive a future location based on the knowledge of previously known mobile information • The intelligent prediction scheme promises a more appropriate approach Control flow for simulation of the user mobility prediction scheme Neuro-fuzzy logic • Fuzzy logic provides an inference morphology that enables approximate human reasoning capabilities to be applied to knowledgebased systems • Fuzzy systems are suitable for uncertain or approximate reasoning, especially for the system with a mathematical model that is difficult to derive • Fuzzy logic allows decision making with estimated values under incomplete or uncertain information • Fuzzy Logic is a departure from classical two-valued sets and logic, that uses "soft" linguistic (e.g. large, hot, tall) system variables and a continuous range of truth values in the interval [0,1], rather than strict binary (True or False) decisions and assignments Mobility Prediction Scheme Expresses user mobility in terms of direction and velocity of the mobile user • It predicts a future location based on the movement factors of past and current locations of a mobile user Let Skt be the state k at time t of the movement of the user Skt is defined as Skt = (vkt,θkt) such that k < K and t Є T where K is the number of states T is the set of time vkt is the velocity of the user at time t in state k θkt is the direction of the user at time t in state k Using the movement states defined above, the movement function which maps current and past movement states into a future movement state can be where NFS is the neuro-fuzzy inference system h is the number of current and past movement states The current and past movement states represent location history, by which movement patterns can be determined. • To construct a fuzzy rulebase, we use cluster-based techniques to automatically generate fuzzy rules according to the degree of similarity of movement patterns • Similar states get grouped into clusters • Each cluster has a cluster center which is used as the reference value • The similarity(distance) between the present state of a user and a cluster is given as Di = | Skt – Ri | where Ri is the center of the cluster i. • We define a value r such that if Di <= r the similarity is high, otherwise it is low • Each cluster is associated with an element in the fuzzy rulebase • Based on the above criteria we build our fuzzy rulebase • But unnecessary fuzzy rules should be eliminated from the rulebase according to an appropriate mobile situation • It is necessary for a criterion to determine the elimination of an unnecessary fuzzy rule from the fuzzy rulebase. where agei is the age of the i th fuzzy rule α is the parameter which controls the age of the fuzzy rule β is the parameter determining the elimination of the fuzzy rule The age of a fuzzy rule is updated as follows: • If the distance between a current movement state vector and a cluster center is less than the radius, the age of a fuzzy rule becomes zero. • On the other hand, if the distance is larger than the radius, the age of a fuzzy rule is increased by one. • According to the value of β fuzzy rules are subdivided into two groups: young fuzzy rules and old fuzzy rules • Since the young rules suitably reflect the latest movement state, it is desirable that they be used as fuzzy rules • The old rules should be eliminated from the rulebase Learning Model Learning Model (cont…) This process is to be run iteratively from Step 2 to Step 5 This process will enable the learning model to recognize the movement patterns in the location history of the mobile users by clustering the movement states Simulation Environment Parameters • The number of movement data (days) • The number of current and past movement states (h) • The change period of movement states (ζ) • Radius (r) • Learning rate (η) • The number of learning iterations (m) • The control parameter for the age of the fuzzy rule (α) • The remove parameter for the unnecessary fuzzy rules (β) Conclusions • In this presentation, we discussed the mobility prediction schemes based on two criteria • The main issue addressed is the prediction of user mobility in order to improve QoS for mobile applications • It was motivated from the fact that the location history, plays an important role in user mobility • The predicted location information by the scheme could be used for supporting the pre-connection of services and the pre-allocation of resources • This provides an efficient solution for the service delay problem that occurs in a hand-off. • Mobility prediction on existing mobile networks will play a role as a coordinator for smooth communication References [1] Joon-Min Gil, Chan Yeol Park, Youn-Hee Han, Chong-Sun Hwang “User Mobility Simulation by Traveling Demand Model in Mobile Environments”, Nov 2000 [2] Joon-Min Gil “Simulation of a Mobility Prediction Scheme Based on Neuro-Fuzzy Theory in Mobile Computing”, July 2000 [3] John Scourias,Thomas Kunz “An activity-based mobility model and location management simulation framework”, Oct 1999 [4] Kuochen Wang, Jung Huey “A cost effective distributed location management strategy for wireless networks”, Sep 1999 [5] Sanjoy K. Sen, Amiya Bhattacharya, Sajal K. Das “A selective location update strategy for PCS users”, Jun 1999 [6] Atif A. Siddiqi, Thomas Kunz “The Peril of Evaluating Location Management Proposals through Simulations”, May 1999 [7] Guang Wan, Eric Lin “Cost reduction in location management using semi-realtime movement information”, Feb 1998 [8] Derek Lam, Yingwei Cui, Donald C. Cox, Jennifer Widom “A Location Management Technique To Support Lifelong Numbering in Personal Communications Services”, Oct 1997