Detecting Node encounters through WiFi By: Karim Keramat Jahromi Supervisor: Prof Adriano Moreira Co-Supervisor: Prof Filipe Meneses Oct 2013
Motivation Analysis of Wi-Fi data for understanding Encounter Pattern can provide significant knowledge about human mobility patterns. Human Mobility Modeling can be used in many application domain: Urban Planning, Social Science, Epidemiology, Network Communications. In Network Communications, realistic human mobility models have an important role in simulations of wireless networks. Understanding of nodal encounter patterns have important role in design of protocols and efficient deployment of mobile networks.
Objectives Detecting Pairs Node Encounters. Solution for Detecting and Smoothing Ping-Pong Events. Analyzing Statistics of Pair Encounters.
Physical encounters Detecting Node Encounters
Observed Encounters Encounters observed through the Wi-Fi network, using usage logs (RADIUS ) Definition of Encounter: two or more devices connected to the same AP simultaneously. Direct and Indirect Encounters
Detecting Node Encounters Challenges Nodes aren’t necessarily associated with the geographically nearest AP. Different devices have different aggressiveness for changing association with different APs. Ping-Pong Events. Overlap among coverage areas of different APs.
Detecting Node Encounters Related Work “On Nodal Encounter Patterns in Wireless LAN Traces”, Nov 2010, by Ahmad Helmy: Analyzing multiple wireless LAN traces from university and corporate campuses Looking for understanding encounter patterns using graph analyzing approach “On Modeling of User Associations in Wireless LAN Traces on University Campuses”, April 2006, by Ahmad Helmy: Study Large scale data Trace Proposing metrics for describing Individual Mobile Node behavior
Detecting Node Encounters Wi-Fi Data Set idacessoTimestampAcct_Session_IdAPSTAAcct_Session_TimeAcct_Status_Type /1/2011 0: Mac 1Mac 7NULLStart /1/2011 0: B8Mac 2Mac 8NULLStart /1/2011 0: BBMac 3Mac 894Stop /1/2011 0:009554Mac 4Mac 959Stop /1/2011 0: CBMac 5Mac 10NULLStart /1/2011 0:000000BCD2Mac 6Mac 1061Stop Anonymized part of Wi Fi Trace
Detecting Node Encounters Workflow for finding Pairs Encounters
Detecting Node Encounters idacessoTimestampAcct_Session_IdAPSTA Access Session Time[s] Acct_Status_Type Start TimeEnd Time /1/2011 0:00: Mac 1Mac 6 94stop :58: :00: /1/2011 0:00: B8 Mac 2Mac 7 59stop :59: :00: /1/2011 0:00: BB Mac 3Mac 8 61Stop :59: :00: /1/2011 0:00: Mac 4Mac Stop :33: :00: /1/2011 0:00: CB Mac 5Mac stop :43: :00:25 Anonymzed Part of Wi-Fi Trace after filtering and adding start time
Detecting and Smoothing Ping-Pong Events Detecting Ping-Pong Events Frequent Change in APs association. Transition Time should be short ( as threshold). Short Access Session Time ( as threshold ).
Detecting and Smoothing Ping-Pong Events Detecting Ping-Pong Events Flag Ping-PongSTAAP_Primary Access Session Time[s] Start TimeEnd Time 0Mac 1Mac :38: :38:08 0Mac 1Mac :55: :55:47 1Mac 1Mac :55: :55:53 1Mac 1Mac :55: :56:04 1Mac 1Mac :56: :56:08 1Mac 1Mac :56: :56:22 0Mac 1Mac :56: :44:28
Detecting and Smoothing Ping-Pong Events Smoothing Ping-Pong Events Non Ping-Pong APs are kept unchanged. APs which involves in Ping-Pong Intervals, will be replaced by one of nearest Non Ping-Pong APs based on max Access Session Time and/or summation Access Session Time in Ping-Pong Interval.
Detecting Node Encounters Finding Pair Encounters APSTA_ASTA_BEncounter Start TimeEncounter End Time Mac-1Mac-3Mac :40: :02:17 Mac-1Mac-3Mac :02: :39:34 Mac-2Mac-4Mac :37: :38:41 Mac-2Mac-5Mac :14: :09:00 Mac-2Mac-6Mac :17: :17:19 Mac-2Mac-6Mac :42: :42:32 Anonymized part of Pair Encounter List Algorithm for finding pair encounters is based on common definition of Pairs Encounters but after Smoothing Ping-Pong. Usually the number of Pair Encounters is larger than the initial number of Wi-Fi observations (after filtering).
Detecting Node Encounters Merged Pairs Encounters APSTA_ASTA_BEncounter Start TimeEncounter End Time Merged Encounter Start Time Merged Encounter End Time MAC-1 MAC-2MAC :50: :50: :50: :50:43 MAC-1 MAC-2MAC :06: :07: :06: :09:02 MAC-1 MAC-2MAC :07: :08:01NULL MAC-1 MAC-2MAC :08: :08:27NULL MAC-1 MAC-2MAC :08: :09:02NULL MAC-1 MAC-2MAC :09: :10: :09: :46:19 MAC-1 MAC-2MAC :10: :10:37NULL MAC-1 MAC-2MAC :10: :11:28NULL MAC-1 MAC-2MAC :11: :28:30NULL MAC-1 MAC-2MAC :28: :36:04NULL MAC-1 MAC-2MAC :36: :46:19NULL MAC-1 MAC-2MAC :46: :52: :46: :53:15 Anonymized part of merged Pair Encounter List
Detecting Node Encounters Choosing values for time Thresholds Time Threshold[s]Number of Ping- Pong events Number of distinct APs in WiFi trace Number of distinct APs in encounters list Number of encounters / variation after smoothing (%) Without smoothing ( ) ,
Statistics of Pair Encounters Contact Time Inter- Contact Time an inter-contact time a contact time STA-a STA-b STA-a STA-b STA-a STA-b STA-a STA-b Time Comparison of Contact Time distributions for a specific pair of nodes. Comparison of Aggregate Contact Time distributions.
Statistics of Pair Encounters Inter- Contact Time Comparison of Inter Contact Time distributions for a specific pair of nodes. Comparison of Aggregate Inter Contact Time distributions.
Statistics of Pair Encounters Comparison of Aggregate Inter Contact Time distributions. Inter Contact Time Distributions for a few pairs of nodes with different number of encounter events (3 months). Aggregate Encounter Distribution isn’t always representative of Pair Encounter Distribution.
Scale Free behavior Main properties of human mobility indicate Scale behavior on temporal dimension. Aggregate Inter Contact Time Distributions on different observation periods.Aggregate Contact Time Distributions on different observation periods.
Next Step Analysis Periodicity in Pair Encounters Transform List of Pair Encounters into Time Series Analyze periodicity by applying power spectral analysis (autocorrelation (ACF)+ Fourier Analysis ) Calculating of Encounters by considering overlap coverage areas. Strategies for calculating number of encounters
Conclusion Power law trends illustrate heterogeneity in human movement characteristics. Human Mobility Connectivity properties show scale free behavior on temporal dimension. Aggregate Encounter Distribution isn’t always representative of Pair Encounter Distribution.
Questions and Suggestions Thank You