SlideShare a Scribd company logo

Slides for Ph.D. Thesis Defense of Dheryta Jaisinghani at IIIT-Delhi, INDIA

Dheryta Jaisinghani, profile picture
Dheryta Jaisinghani

Dheryta Jaisinghani defended her Ph.D. thesis on understanding the role of active scans in large-scale WiFi networks. She investigated three main problems: unnecessary active scans, WiFi indoor localization with minimal active scans, and using active scans for data transfer. Her research methodology involved analyzing real-world WiFi traffic datasets, designing solutions to address the problems, and implementing and evaluating the solutions. Some key contributions included techniques to detect growth in probe traffic, infer the causes of active scanning, and reduce unnecessary probe requests from client devices.

Education
1 of 97
Downloaded 34 times
1
2
Most read
3
Most read
4
5
6
7
8
9
10
Most read
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
Understanding the Role of Active Scans for their Better Utilization in Large-scale WiFi Networks Ph.D. Thesis Defense of Dheryta Jaisinghani On Monday April 22, 2019 Advised By: Vinayak Naik and Sanjit K. Kaul
Agenda 2 1 Background ● WiFi Networks ● Connecting with an Access Point ● Scanning Scope ● Problems Investigated ● Methodology Followed ● Challenges 2 Problems ● Unnecessary Active Scans ● WiFi Indoor Localization ● Data Transfer with Active Scans 3 Conclusion ● Summary ● Contributions to Existing and New Standards ● Future Work 4
WiFi Network Architecture ... Controller AP1 WiFi Clients AP2 APN WiFi Association Ethernet 3 AP- Access Point BSSID1 BSSID2 BSSIDN ...
How does a Client Connect with an Access Point? 4 Step 3 Data Transmission and Reception Step 1 Discover an Access Point Step 2 Associate and Authentication Handshake with the Access PointAn essential initial step that enables steps 2 and 3 and much more!
Discovering an Access Point 5 Listen Beacons ... ... Beacons AP Client Probe Requests ... ... Probe Responses AP Client Passive Scanning Active Scanning Slow and Not Preferred Default Discovery Protocol
Why Focus on Discovery? Modern WiFi networks provide more than just basic network access functionality 6 IoTs Localization Environment Sensing Access Point discovery mechanisms continue to be traditional *Images from Internet (Open Access) * *
Agenda 7 1 Background ● WiFi Networks ● Connecting with an Access Point ● Scanning Scope ● Problems Investigated ● Methodology Followed ● Challenges 2 Problems ● Unnecessary Active Scans ● WiFi Indoor Localization ● Data Transfer with Active Scans 3 Conclusion ● Summary ● Contributions to Existing and New Standards ● Future Work 4
Problem Statement ● Investigate the mechanisms of discovery in large-scale and dense WiFi networks ● Default protocol - Active Scanning is crucial ○ Excessive amounts ⇒ Performance⬇ ○ Arbitrary reduction ⇒ Performance⬇ ● A thorough relook needed 8
Problems Investigated 9 Harnessing Active Scans for Data Transfer in WiFi-based IoT Nodes 03 ● Best Use of Active Scans WiFi Indoor Localization using Existing Infrastructure in the presence of Minimal Active Scans 02 ● Minimal Active Scans Understanding and Mitigating the Impact of Unnecessary Active Scans in WiFi Networks 01 ● Excessive Active Scans
Empirical Methodology Followed Sniff WiFi Traffic Analyze and Infer Design a Solution Implement the Solution 10 Management Probe, Beacons, Associate, Authenticate Data Qos/Non-QoS Data Control RTS, CTS, ACK, Polling IEEE 802.11 Frames
Challenges ● Work with actual network deployments ○ Dynamic environments ● Real-time analysis and validation ○ Precision ● Multitude of devices [>15,000 vendors] ○ One solution does not fits all ● Large-scale infrastructures ○ Integrable solutions 11
Agenda 12 1 Background ● WiFi Networks ● Connecting with an Access Point ● Scanning Scope ● Problems Investigated ● Methodology Followed ● Challenges 2 Problems ● Unnecessary Active Scans ● WiFi Indoor Localization ● Data Transfer with Active Scans 3 Conclusion ● Summary ● Contributions to Existing and New Standards ● Future Work 4
Problem#1* Understanding and Mitigating the Impact of Unnecessary Active Scans in WiFi Networks 13 1. D. Jaisinghani, V. Naik, S. K. Kaul, R. Balan, and S. Roy, “Improving the Performance of WLANs by Reducing Unnecessary Active Scans,” ArXiv e-prints, Jul. 2018. [Under Review] 2. D. Jaisinghani, V. Naik, S. K. Kaul, and S. Roy, “Realtime detection of degradation in WiFi network's goodput due to probe traffic,” in WinMee -WiOpt, 2015, May 2015. 3. D. Jaisinghani, V. Naik, S. Kaul and S. Roy, “Sniffer-based Inference of the Causes of Active Scanning in WiFi Networks,” in NCC,Chennai, India, Mar. 2017. 4. (H. Fulara, G. Singh)*, D. Jaisinghani and M. Maity and T. Chakraborty and V. Naik, “Use of Machine Learning to Detect WiFi Networks from Unnecessary Active Scans,” WoWMoM, 2019. [*Equal Contributions] # Best Graduate Forum Presenter Award at COMSNETS 2018
Research Question What is the extent and the impact of active scans and how to deal with them? 14
The Extent of Active Scans ● Analyzed 3 real-world datasets ○ Sigcomm [5 GHz, 2 Days, 14 Hours] ○ IIT-B [2.4 GHz, 1 Day, 2.5 Hours] ○ IIIT-D [2.4 GHz, 9 Days, 63 Hours] ● Clients probe lesser in 5 GHz [Sigcomm] ● Client Density ↑, Probe Requests ↑ [IIT-B] ● Frame size depends on clients and BSSIDs [50 - 300 Bytes] ● Probe Requests and Responses can arrive as early as every 0.01 ms ● Probe Response arrival time depends on #BSSIDs ● 2.4 GHz has lowest bitrate=1 Mbps, 5 GHz has lowest bitrate=6 Mbps ● 99% Probe Responses fetch redundant network information for stationary WiFi clients 15
Active Scanning - 2.4 GHz vs 5 GHz 16 ● WiFi networks today operate on both 2.4 GHz and 5 GHz - Different channel characteristics ● Previous datasets could not be used because, ○ SIGCOMM is old (2008) with minimal evidence ○ IIIT-Delhi < 5 clients/day in 5 GHz ○ IIT-B - 2.4 GHz ● ∴ Venue with dual-band devices needed ○ A university in Singapore ○ Dual-Band Aruba WiFi Network ○ Comparable number of clients in both bands ○ 6 hours of traffic collected
Effect of RSSI drop on #Status ChangesEffect of RSSI drop on #Probe Requests Comparison Results - 2.4 GHz vs 5 GHz 17 5 GHz sees active scans when RSSI drops 2.4 GHz sees active scans at all RSSI Stationary clients scan minimally in 5 GHz 5 GHz sees status change when RSSI drops 2.4 GHz sees status changes at all RSSI
Importance of 2.4 GHz ● 2.4 GHz is common in developing countries like India ● Dual band WiFi devices are costly ● 85% WiFi devices still operate in 2.4 GHz ● 2.4 GHz will be the likely choice for large-scale IoT deployments 18 We focus on 2.4 GHz band
Impact from Client-side Perspective - Latency 19Mean latency ↑ from 3.57 ms to 21.72 ms [6x ↑] ● Stationary WiFi client ● Associated ● AP-RSSI - Good ● Logs ○ Time of Scan ○ Latency ● Duration: 5.5 hours
Impact from Client-side Perspective - Association Patterns 20 38% Discovered 11, Chose 3 to associate ● Stationary WiFi clients (34) ● Associated ● AP-RSSI - Good ● Logs ○ Discovered APs ○ Associated APs ● Duration: 10 days Clients discover more, associate to few
Impact from Network-wide Perspective - Controlled Setup 21 Growth in Probe Traffic results in drop in Bandwidth [ 20 Mbps -> 10 Mbps]
Impact from Network-wide Perspective - Uncontrolled Setup 22
Impact from Network-wide Perspective - Uncontrolled Setup - Results 23 ACK losses - 2.6x ⬆ Frame retries - 2.19x ⬆ Goodput - 50% ⬇ Probe traffic - ~80% ⬆ Channel Utilization >= 70%
Addressing the Problem ● Problem ○ Stationary clients do not benefit from active scans ○ Goodput drops in heavily utilized WiFi networks ● Challenges ○ Active Scanning is Crucial Process - Cannot Disable ○ Passive Scanning is slow - Not Preferred ○ Fast Solution - Strict timing constraints of WiFi protocol ○ ⬆Probe Traffic ⇔ ⬇Data Traffic - Detecting exact cause is hard 24
Current Solutions and their Limitations ● Problem Recognition [IMC ‘15, TMC ‘10, 802.11ax] ● Reduce scanning delays ○ Modified scanning algorithms in [ICC ‘12, PERCOM ‘10] ○ Multiple radios/MPTCP [NSDI ‘15, TVT ‘14] ○ Channel frequency response [MOBICOM ‘13] ● Partially reduce probe traffic ○ Passive scanning or disable probe response on APs [WN ‘10, WoWMoM ‘09] ● Systems for improving the performance of WiFi networks ○ PIE, JIGSAW [NSDI ‘11, SIGCOMM ‘06] 25Device Changes, Partial Reduction, Reduce Interference
Our Solutions 26 ● Detect the growth of probe traffic - Validates the cause ● Infer the cause of growth in probe traffic - Better Network Planning ● Reduce probe requests from client-side - Inhibits Probe Responses
Detect Growth of Probe Traffic 27 Real-time growth in probe traffic ● Assumption - There is always data to be sent ● Cumulative increase of P/D > 1 is plotted against time ● Slope >= 0.10 => Goodput drop [90% accurate] Case Metric D Frames P Frames Probe Traffic ⬆ 1 P/D < 1 ✔ ✖ ✖ 2 P/D = 1 ✔ ✔ ✖ 3 P/D > 1 ✖ ✔ ✔ Probe Frames Fresh Data Frames
How do Clients Decide When to Scan? 28 Discovery Connection Establishment Connection Maintenance1. Periodic Scan 2. Associated State: Unassociated to Associated 3. Power State: Low to High 4. Loss of Beacons 5. AP-side Procedures 6. Low RSSI 7. Data Frame Losses Causes of Active Scans Applications (wpa_supplicant, network manager) MAC Drivers (cfg80211, mac80211) Device Drivers (ath9k, iwlwifi) WiFi Chipset (AR9462, BCM4322) User space Kernel space Hardware At client-side, it's a non-trivial process It can be triggered at any of these layers!
Inferring the Causes of Active Scanning 29 ● Frames received in real-time ● Every cause has a distinguished signature ● Defined static rules and metrics for each cause ● DFA-based detection with states and events Sniffer-based Inference Mechanism
An Example 30 AP-side Procedures ● Connection Maintenance → AP-side Procedures ● Load balancing/Rogue client ● AP deauthenticates client ● Client Scans
Extent of Causes in Real Datasets 31 Periodic Scans -> Most Common Network Planning Intermittent user activity Network Contention
Applying Inference Mechanism to Real-world Networks ● Unassociated Periodic Scans ⇒ Unassociated clients ⇒ Rogue clients ● Associated Periodic Scans ⇒ Aggressive clients ● Connection Maintenance ⇒ ↓ Coverage or ↑ Interference ● AP-side Procedures ⇒ Faulty Controller Algorithm 32 Network administrators can improve network planning
Reduce Probe Requests From the Client-end 33 ● Revisit the causes of Active Scans ● Stationary WiFi clients ⇒ Active Scans ❌ ● Mobile WiFi clients ⇒ Passive Scans ❌ ● Time Constraints - Connection Establishment/Maintenance ● Dynamic Decision Associated? Handover No Stationary? Connection Maintenance Passive Scans Yes YesNo Active Scans Connection Establishment? Yes No
Client-end Solution: Experiment Setup 34 ● Live IIIT-D network ● Modified and Unmodified clients - 10 each ● Introduced causes of active scans every hour ● Client collects all connection logs ● Sniffer logs WiFi traffic ● Duration: 9 hours
Client-end Solution: Results 35 Metric Modified Clients Unmodified Clients #Probe Requests 500/hour 932/hour Time to Connect 0.25 seconds 4.35 seconds Connection Persistence 1 minute - 0.37 5 minutes - 0.55 15 minutes - 0.80 1 minute - 0.20 5 minutes - 0.43 15 minutes - 0.73 Simple client intelligence works
Key Takeaways ● Device-agnostic measures without special h/w or s/w needs ● Most APs come with sniffing antennas ● Easy to integrate with existing controller ● Sniffers may miss frames ● Multiple sniffers needed for comprehensive network view 36 Excessive Active scans affect performance in 2.4 GHz
Problem#2* WiFi Indoor Localization using Existing Infrastructure in the presence of Minimal Active Scans 37 5. D. Jaisinghani, R. Balan, V. Naik, Y. Lee, and A. Misra, “Experiences & Challenges with Server-Side WiFi Indoor Localization Using Existing Infrastructure,” in MobiQuitous, 2018, New York, USA..
Research Question How to localize several thousands of devices in real-world indoor deployments? 38
The Challenges ● Infrastructure, i.e., the controller and the access points changes ● Devices cannot be modified in any way ○ Explicit/implicit participation for data generation ○ App download ○ Chipset changes ○ Extra antennas 39
State-of-the-art Solutions ● Accuracy of few centimeters is possible with, a. Sophisticated techniques such as AoA, ToF/ToA, RToF, and PoA [NSDI ‘13, MobiCom ‘14, SIGCOMM ‘15] b. Require extra hardware such as specialized antennas and ultrasonic systems [MobiSys ‘05, NSDI ‘13, MobiCom ‘14, SIGCOMM ‘15] c. Special requirements such as Line-of-Sight, Visible Light, Acoustics [NSDI ‘14, Mobiquitous ‘15, MobiSys ‘15, MobiSys ‘18] Can they be applied to a large-scale network where several thousands of WiFi devices are to be localized? NO Customized hardware, Client application, Rooting the Client OS, High Energy Budget, High Cost of deployment - Limit Generalizability and Scalability 40
What Do We Have? ● Lots of access points ● Few controllers that manage these access points ● Access points can measure RSSI from the clients ● Only available data is RSSI measurements from access points - ○ We can at least attempt to localize ○ That's all we have 41
Solution Fingerprint-Based WiFi Localization ● Scalability ○ Albeit the manual effort for setup, fingerprinting-based localization is scalable ● Device Friendly ○ No client-side changes / No AP-side changes ● Integrable ○ Seamless integration with existing infrastructure is possible Premise of fingerprint-based solutions - Clients always scan! 42
Fingerprinting Process Offline Online Fingerprints ∀ Li ∈ L, Known landmark Li, RSSI Vector= < Li , B, AP1 : RSSIi ; ...; APn : RSSIn ; > Unknown landmark Lx, RSSI Vector= < B, AP1 : RSSI'i ; ...; APn : RSSI'n ; > Distance, Di in Signal Space between Lx and ∀ Li ∈ L, Di = sqrt((RSSIAP1 - RSSI' AP1 )2 + (RSSIAP2 - RSSI' AP2 )2 + … + (RSSIAPn - RSSI' APn )2 ) Predicted Landmark = Li with minimum Di Already stored in database Collected in real-time 43
System Architecture ... Controller AP1 WiFi Clients AP2 APN WiFi Association Ethernet 44 Location ServerRTLS Server
System Architecture (Contd.) ● 750+ dual-band access points ● Centrally controlled by 11 WiFi controllers ● 4000 associated clients per day ● RTLS data feed received every 5 seconds 45 Timestamp Age ChannelClient MAC AP Association Status Data Rate RSSI RTLS Data Feed
Characterizing the Landmarks 46 Landmarks = Water Sprinklers on the ceiling Landmarks across floors (Overlapping Region A) Landmarks across floors (Overlapping Region B) Landmarks = water sprinklers, deployed every 3 meters
Ground Truth Data Collection 47 3203 landmarks, 86 hours data ● 4 different WiFi states ○ Emulate real-world usage scenarios ○ Modulate Scanning Frequency ○ Disconnected → Inactive → Intermittent → Active WiFi On State: Disconnected WiFi On State: Connected and Inactive WiFi On State: Connected and Intermittent WiFi On State: Connected and Active Scanning Frequency Low High Timestamp at every landmark is recorded at granularity of seconds
Data Processing Challenges ● Traffic Distinction ○ Scanning Records ○ Non-Scanning Records ● Offline fingerprints ➡ Scanning records ● Online fingerprints ➡ Scanning + Non-scanning records ● RTLS data feeds does not distinguish ● Proposed inference ○ Scanning records - fixed and lowest data rates ○ Non-scanning records - variable data rates 48
Problems Discovered 49 ● Cardinality: Set of access points reporting for a client ● Fingerprinting-based Solutions - Effective and Efficient match ○ Clients always trigger active scans ○ Same set of APs respond ● Online and offline cardinalities are different - #Matches Affected ○ Cardinality mismatch ○ RSSI mismatch ○ High Client Scan Latency Result → Clients are teleported !!
Evidence 1 - Cardinality Mismatch 50 OnlineOffline 5 GHz - 80% 2.4 GHz - 40% Cardinality ⬇ from 16 ➡ 6 (Max) Minimal Access Points to Match
Evidence 2 - RSSI Mismatch 51 Client close to AP Client far from AP 50 dB 5 dB 30 dB 10 dB Non-Scanning Frames Scanning Frames Scanning Frames are Better
Evidence 3 - High Client Scan Latency 52 Very few scans Scans ⬆ 5 GHz Scans ⬇ Dual-Band Scans ❌ Stationary client do NOT scan
Cause 1 - Cardinality in Presence and Absence of Scans Scans PresentScans Absent No Access Points Reports Max. Access Points in 5 GHz No of Access Points ⬆ Max. Access Points in 5 GHz No Access Points Reports Scans + 2.4 GHz is Better 53
Cause 2 - RSSI variation 2.4 GHz and 5 GHz But, RSSI variation in 2.4 GHz > 5 GHz (Higher range and interference) 5 GHz is Better 54
Summary of the Problems 55 Distance of Transmission Variation in RSSI #Scanning Frames 2.4 GHz High High High 5 GHz Low Low Low Distance of Transmission Variation in RSSI Frequency Scanning Frames High Low Low Non-scanning Frames Low High High Improves accuracy Reduces accuracy Non-Trivial to solve
Recommendations to Reduce the Localization Errors 56 Baseline Maximum #APs AP with Maximum RSSI AP of Association No floor detection logic Floor from which maximum number of APs report the client Floor from which strongest RSSI is received Floor of AP to which client is associated with Detect floor before localizing Metrics: Floor Error and Same Floor Error
An Example 57 Floor 2 AP1 AP2 AP3 AP4 Floor 1 Client -50 dBm -60 dBm -70 dBm -52 dBm Floor 3 Maximum Number of APs Floor 2 AP2 AP3 AP4 Floor 1 -50 dBm -60 dBm -70 dBm -52 dBm Floor 3 Client AP1 AP with Maximum RSSI Floor 2 AP2 AP3 AP4 Floor 1 Client -50 dBm -60 dBm -70 dBm -52 dBm Floor 3 AP1 AP of Association
Reduction in Localization Errors ● 5 GHz band outperforms 2.4 GHz band ● Worst Heuristic - Maximum APs ○ Reason - Distant APs also respond ● Best Heuristic - AP of association ○ 17 meters (2.4 GHz), 20 meters (5 GHz) ○ Floor Errors- 78% ⬇ (2.4 GHz), 46% (5 GHz) ○ Same Floor Errors - 58% ⬇ (2.4 GHz) , 3.8% (5 GHz) ○ Reason - Clients associate with the nearest AP ● Compared with SignalSLAM [IPIN ‘13] ○ ~4 meters better, no client-side changes 58
Key Takeaways ● Few meters accuracy is OK ○ Contextual Location Aware Applications - Marketing/Business use-cases ○ Tracking assets - employees/students ● Simple floor-detection heuristics can combat the challenges ● Limited by RTLS information 59 Reduced active scans hamper services like Localization
Problem#3 Harnessing Active Scans for Data Transfer in WiFi-based IoT Nodes 60 6. D. Jaisinghani, H. Fulara, G. Singh and M. Maity and V. Naik, “Demo: Elixir - Sending Data in WiFi-based IoT without the Data Frames,” Mobicom, 2018, New Delhi, INDIA 7. D. Jaisinghani, H. Fulara, G. Singh, M. Maity and V. Naik, “Sending Data in WiFi-based IoT without the Data Frames,” Under Review. # Won First prize at nationwide WiFi competition - WiFi ThinkFest 2017
When WiFi Fails 61 A WiFi Association is mandatory to enable data transfer Case Signal Quality Associated? Cause Effect 1 Bad ✖ Sparse AP Deployment No Association 2 Bad ✔ Reduced RSSI Connection Maintenance/Handovers 3 Good ✖ Overloaded APs/Admin Policies No Association 4 Good ✔ High Channel Utilization and Losses Connection Maintenance/Handovers
Research Question Is it possible to transfer data in the absence of an ongoing WiFi association, for example, in challenged scenarios? 62
Default States of a WiFi Client 63 Wake UpSleep Transmit Data Interface ON Scan Associated Discover APs Check Association Successful Association Data Unavailable Data Available Failed Association Until this is done This is not possible
Proposed Solution - ViFi Protocol [Scan (V) comes before Associate (W)] 64 Probe Requests Probe Responses Exchange of Data Access Points IoT Nodes Wake Up, Scan & Transmit Data Data Available Idle Transmission Complete Interface ON Data Available + The Idea Reduction in States WiFi Clients always scan - Active Scan
Assumptions ● IoT nodes are sensors ● Data to be transmitted is small (few bits) and infrequent ● IoT nodes can trigger active scans ● At least 1 AP is available ● AP is open and configurable 65
Challenges to Implement ViFi ● Conformance to Existing Standards ○ No new packets should be introduced ● Simple to Deploy ○ Should be capable of operating on diverse set of devices ● Reliable Data Transfer ○ Should be able to transmit data at low/variable RSSI ● Co-exist Seamlessly with Default Protocols ○ Performance of nodes with default protocols should not be hampered 66
State-of-the-Art Solutions ● Management frame stuffing [HotMobile ‘07, MSR Patent ‘05] ○ Interference Management [ arXiv ‘17 ] ○ Traffic Management [ VTC ‘14, AINA ‘16 ] ○ Localization [ TENSYMP ‘16 ] ○ Disaster Recovery [ Mobisys ‘16 ] ○ Location-based Information Services [ WMNC ‘16] 67 Application-based solutions, performance impact?
Protocol Implementation 68 Userspace App Libnl Library Client Userspace App Libnl Library AP Probe Request Probe Response WiFi System Architecture MAC Header (22) Frame Body (2320) FCS (4) Fixed Fields Elements Element ID (1) Length (1) OUI (3/5) Content Default Frame Structure Sequence Number More Data Length Data Proposed Frame Structure
Proposed Protocol 69 Client AP1 Probe Response#1 AP2 APN Channel 1 Channel 6 Channel 11 Probe Response#6 Probe Response#11 Probe Request#1 Probe Request#6 Probe Request#11 Stuff Data in Probe Request Trigger Active Scan Implicit Acknowledgement Extract Data Extract Data Extract Data
Evaluation 70 Compare WiFi and ViFi Impact of ViFi on WiFi Experiments ● IoT Nodes - Raspberry Pi ● AP - OpenWRT and hostapd ● Network conditions - Good and Bad ● Data - 300 Bytes ● Data generation intervals - 1 second and 120 seconds
Compare WiFi and ViFi 71 IoT Nodes - WiFi Protocol Access PointSniffer IoT Nodes - ViFi Protocol Compare the effect
Number of Frames 72 WiFi transmits ~470% ⬆ frames, 32% retries WiFi continues to send discovery frames
Airtime Utilization 73 WiFi consumes at least 8x higher Airtime
Delivery Rate & Goodput 74 ViFi delivers better or at par with no retries ViFi maintains better or at par goodput
Impact of ViFi on WiFi 75 Non-IoT Nodes - WiFi Protocol Access PointSniffer IoT Nodes - ViFi Protocol Non-IoT Nodes - WiFi Protocol Access PointSniffer IoT Nodes - WiFi Protocol Compare the effect Unchanged
Contention Time ● When contention/collisions ⬆ ⇒ Inter-frame arrival time increase ⬆ ● Affects all upper layer metrics - throughput, jitter, latency ● ⬆Contention time ⇒ nodes are struggling 76 IoT nodes Contention faced by non-IoT nodes ViFi reduces contention
Key Takeaways ● ViFi serves as an Elixir - when WiFi fails ● ViFi eliminates network traffic ● Saves up to 18% energy ● Apt for low bit-rate applications - ambient data collection, emergency broadcasts, ultra dense environments like India's Kumbh Mela ● In Good network conditions, ViFi experiences higher delay ● Data generation interval needs to be well thought off ● Not suited for high-end data intensive transfers 77
Agenda 78 1 Background ● WiFi Networks ● Connecting with an Access Point ● Scanning Scope ● Problem Statement ● Methodology Followed ● Challenges 2 Problems ● Unnecessary Active Scans ● WiFi Indoor Localization ● Data Transfer with Active Scans 3 Conclusion ● Summary ● Contributions to Existing and New Standards ● Future Work 4
Conclusion ● Detailed analysis on active scanning protocol ● Methods to mitigate impact of active scans ● Practical challenges in localizing clients ● ViFi protocol for low-end IoT nodes ● Datasets and source codes are public 79 Active Scanning is default discovery protocol, all latest and upcoming standards will benefit from our device-agnostic approaches
Future Work ● Performance enhancements with Software Defined Networking and Machine Learning ● Device-agnostic WiFi-based indoor localization ● Evaluate Association-free WiFi (The proposed ViFi) for ○ Security ○ Recovery ○ Downlink data transfer ○ MAC address randomization 80
Publications 1. D. Jaisinghani, V. Naik, S. K. Kaul, R. Balan, and S. Roy, “Improving the Performance of WLANs by Reducing Unnecessary Active Scans,” ArXiv e-prints, Jul. 2018. [Under Review] 2. D. Jaisinghani, V. Naik, S. K. Kaul, and S. Roy, “Realtime detection of degradation in WiFi network's goodput due to probe traffic,” in WinMee -WiOpt, May 2015, Bombay INDIA. 3. D. Jaisinghani, V. Naik, S. Kaul and S. Roy, “Sniffer-based Inference of the Causes of Active Scanning in WiFi Networks,” Mar. 2017, NCC, Chennai, INDIA. 4. (H. Fulara, G. Singh)*, D. Jaisinghani, M. Maity and T. Chakraborty and V. Naik, “Learning to Rescue WiFi Networks from Unnecessary Active Scans,” WoWMoM, 2019. Washington DC, USA [Core A, *-Equal Contributions] 5. D. Jaisinghani, R. Balan, V. Naik, Y. Lee, and A. Misra, “Experiences & Challenges with Server-Side WiFi Indoor Localization Using Existing Infrastructure,” in MobiQuitous, 2018, New York, USA. [Core A] 6. D. Jaisinghani, H. Fulara, G. Singh, M. Maity and V. Naik, “Demo: Elixir - Sending Data in WiFi-based IoT without the Data Frames,” Mobicom, 2018, New Delhi, INDIA [Core A*] 7. D. Jaisinghani, H. Fulara, G. Singh, M. Maity and V. Naik, “Sending Data in WiFi-based IoT without the Data Frames,” Under Review. 81
Acknowledgements 82 Vinayak Naik IIIT-Delhi, INDIA Sanjit K. Kaul IIIT-Delhi, INDIA Sumit Roy U. Washington, USA Rajesh Balan SMU, Singapore Mukulika Maity IIIT-Delhi, INDIA Sambuddho Chakravarty IIIT-Delhi, INDIA Vivek Bohara IIIT-Delhi, INDIA Karthik Dantu U. Buffalo, USA Nirupam Roy U. Maryland, USA Srihari Nelakuditi U. South Carolina, USA Mythili Vutukuru IIT-Bombay, INDIA
Appendix 83
Salient Features of Active Scanning ● Probe Requests and Probe Responses are transmitted at lowest bit-rate ● Probe traffic is broadcast traffic ● Stationary WiFi clients trigger active scanning unintelligently ● Frequent active scans fetch same information ● WiFi clients consume ~44.3% higher energy and lose up to 16.7% throughput [IMC'15] ● Performance of clients suffer in dense WiFi networks due to unnecessary active scans [TMC'10] ● Recognized in IEEE 802.11ax - High Efficiency WiFi Network (HEW) status update meeting 84
The Protocol of Active Scanning Client AP1 Probe Response#1 AP2 APN Ch# 1 Ch# 6 Ch# 11 Probe Response#11 Probe Response#6 Probe Request#1 Probe Request#6 Probe Request#11 ACK#1 ACK#6 ACK#11 85
Effect of RSSI drop on #Probe Requests Comparison Results - 2.4 GHz vs 5 GHz 86 5 GHz sees active scans when RSSI drops 2.4 GHz sees active scans at all RSSI Stationary clients scan minimally in 5 GHz [Confirmed with #Status Changes Metric] 5 GHz sees more active scans? #Probe Requests Recorded Per Minute Per Client
Uncontrolled Network Setup - Channel Utilization 87
Devices Studies for Causes of Active Scans 88 Device Type Chipset Vendor Operating System Device Driver Laptop Atheros, Intel, Broadcam Ubuntu 14.04/12.10 and Windows 8.1/7 Ath9k, iwlwifi Tablets Qualcomm, Broadcom Android KitKat and iOS 8.1 OS provided Smartphones Broadcom, MediaTek, Qualcomm Android KitKat, Marshmellow, JellyBean, Cyanogen, and Windows OS Provided USB Adapter Realtek and Atheros Ubuntu 14.04/12.10, Window 8.1/7 rtl8812AU, ath9k_htc
Implementing Sniffer-Based Inference with a DFA 89
States and Events 90
Cause Detection 91
Other Empirical Observations ● ↑ History of SSIDs ⇒ ↑ Probe Requests ● WiFi clients search for non-existing APs ● APs respond to all clients ● Dense deployment of APs ⇒ ↑ Probe Responses 92
Problem Statement - Localization Highlight challenges and propose easy to integrate solutions to build a universal indoor localization system – one that can spot localize all WiFi enabled devices on campus without any modifications to the client or infrastructure-side 93
Improvement in Localization Errors 94 Baseline Maximum #APs AP with Maximum RSSI AP of Association 2.4 ghz 5 ghz 2.4 ghz 5 ghz 2.4 ghz 5 ghz 2.4 ghz 5 ghz Cardinal ity Same Floor Floor Error Same Floor Floor Error Same Floor Floor Error Same Floor Floor Error Same Floor Floor Error Same Floor Floor Error Same Floor Floor Error Same Floor Floor Error 1 >50 20.00 29.50 3.10 >50 48.00 29.60 03.49 49.20 14.70 28.30 02.09 32.30 04.90 27.60 01.28 2 >50 18.30 26.80 02.00 >50 33.33 27.00 05.60 24.20 05.60 27.00 03.46 21.60 00.70 25.80 01.48 3 >50 33.15 20.12 00.00 >50 40.00 24.00 00.00 36.00 14.70 20.12 00.00 22.80 05.50 20.12 00.00 4 >50 18.49 NA 00.00 33.00 13.40 NA NA 24.00 11.00 NA NA 18.90 04.40 NA NA 5 19.20 10.25 NA NA 26.00 14.60 NA NA 17.49 00.00 NA NA 17.49 00.00 NA NA 6 23.00 04.17 NA NA 22.00 00.00 NA NA 16.00 00.00 NA NA 22.00 00.00 NA NA
Sniffer-Based Inference Dataset 95
Large Dataset for Sniffer Based Inference 96
Problem Statement - ViFi We consider WiFi-enabled IoT nodes and APs. The IoT nodes are sensor nodes, who want to transmit few bits of infrequent data. They can trigger an active scan and at least one access point responds. In such a network, we want to enable the IoT nodes to transmit data to an AP, in the absence of a WiFi association with any AP. 97

More Related Content

PPT
Congestion control and quality of service
JUW Jinnah University for Women
 
PDF
20372748 case-study-on-computer-networking-kotak-mahindra-bank
Shivani Joshi
 
PPTX
Basic Relationships between Pixels- Digital Image Processing
Md.Shohel Rana ( M.Sc in CSE Khulna University of Engineering & Technology (KUET))
 
PPTX
WSN-Routing Protocols Energy Efficient Routing
ArunChokkalingam
 
PDF
Data aggregation in wireless sensor network
Shiwangi Thakur
 
PPTX
Noise filtering
Alaa Ahmed
 
PPTX
Connected component labeling algorithm
Manas Mantri
 
PDF
Lecture 13 (Usage of Fourier transform in image processing)
VARUN KUMAR
 
Congestion control and quality of service
JUW Jinnah University for Women
 
20372748 case-study-on-computer-networking-kotak-mahindra-bank
Shivani Joshi
 
Basic Relationships between Pixels- Digital Image Processing
Md.Shohel Rana ( M.Sc in CSE Khulna University of Engineering & Technology (KUET))
 
WSN-Routing Protocols Energy Efficient Routing
ArunChokkalingam
 
Data aggregation in wireless sensor network
Shiwangi Thakur
 
Noise filtering
Alaa Ahmed
 
Connected component labeling algorithm
Manas Mantri
 
Lecture 13 (Usage of Fourier transform in image processing)
VARUN KUMAR
 

What's hot (20)

PPTX
IoT & Smart City
Md Mizanur Rahman
 
PDF
Principal Component Analysis
Ricardo Wendell Rodrigues da Silveira
 
PPTX
Convolutional Neural Networks
Ashray Bhandare
 
PPTX
Cnn
Nirthika Rajendran
 
PPTX
Convolution Neural Network (CNN)
Suraj Aavula
 
PPTX
Pca ppt
Dheeraj Dwivedi
 
PPTX
PPT on Data Science Using Python
NishantKumar1179
 
PPT
Data preprocessing
ankur bhalla
 
PDF
Support Vector Machines ( SVM )
Mohammad Junaid Khan
 
PPTX
Wireless Sensor Networks ppt
Devdutta Chakrabarti
 
PDF
Large Scale Deep Learning with TensorFlow
Jen Aman
 
PDF
Introduction to Recurrent Neural Network
Knoldus Inc.
 
PPTX
Activation function
Astha Jain
 
PPTX
IOT - Design Principles of Connected Devices
Devyani Vasistha
 
PDF
Introduction to dataset
datamantra
 
PPTX
Introduction to Grad-CAM (complete version)
Hsing-chuan Hsieh
 
PDF
IoT case studies from india
IET India
 
PPT
K mean-clustering algorithm
parry prabhu
 
PDF
State Space Representation and Search
Hitesh Mohapatra
 
PPTX
Radial basis function network ppt bySheetal,Samreen and Dhanashri
sheetal katkar
 
IoT & Smart City
Md Mizanur Rahman
 
Principal Component Analysis
Ricardo Wendell Rodrigues da Silveira
 
Convolutional Neural Networks
Ashray Bhandare
 
Cnn
Nirthika Rajendran
 
Convolution Neural Network (CNN)
Suraj Aavula
 
Pca ppt
Dheeraj Dwivedi
 
PPT on Data Science Using Python
NishantKumar1179
 
Data preprocessing
ankur bhalla
 
Support Vector Machines ( SVM )
Mohammad Junaid Khan
 
Wireless Sensor Networks ppt
Devdutta Chakrabarti
 
Large Scale Deep Learning with TensorFlow
Jen Aman
 
Introduction to Recurrent Neural Network
Knoldus Inc.
 
Activation function
Astha Jain
 
IOT - Design Principles of Connected Devices
Devyani Vasistha
 
Introduction to dataset
datamantra
 
Introduction to Grad-CAM (complete version)
Hsing-chuan Hsieh
 
IoT case studies from india
IET India
 
K mean-clustering algorithm
parry prabhu
 
State Space Representation and Search
Hitesh Mohapatra
 
Radial basis function network ppt bySheetal,Samreen and Dhanashri
sheetal katkar
 
Ad

Similar to Slides for Ph.D. Thesis Defense of Dheryta Jaisinghani at IIIT-Delhi, INDIA (20)

PPTX
Wireless survey
ashrawi92
 
PPTX
Wireless Site Survey
Huda Seyam
 
PDF
Datasheet NETSCOUT AirCheck G2 Wireless Tester
PT. Siwali Swantika
 
PPT
Conducting site surveys for wlan performance and reliability
Ken Scott
 
PDF
The UNH-IOL New WiFi Testing & Measurement Capabilities Webinar
UNH InterOperability Lab
 
PPTX
Latest wireless technology
nurmeen1
 
PPTX
Wifi cannerddddddddddddddddddddddddddddddddddddddddddddd
rodainaelhady1
 
PPTX
WiFi 101
NorthCoastHDI
 
PDF
Sensorless sensing with wi fi
ashajuly20
 
PDF
Real-time Wi-Fi network performance evaluation
IJICTJOURNAL
 
PDF
AndrewLeeming_COSENERS-2014
Andrew Leeming
 
PDF
A framework for wireless lan monitoring and its applications
2012engg
 
PDF
A Framework for Wireless LAN Monitoring and Its Applications 
2012engg
 
PDF
WiFi Data Leakage by Solomon Sonya
EC-Council
 
PDF
Web performance prediction using geo statistical method
eSAT Journals
 
PDF
Web performance prediction using geo statistical method
eSAT Publishing House
 
PDF
The Locator Framework for Detecting Movement Indoors
TELKOMNIKA JOURNAL
 
DOCX
Insider (Network Tool)
akki_hearts
 
PDF
Deceptive Speed: How to Fix Fast, Unstable Wi-Fi
NETSCOUT
 
PPTX
Wi-Fi Sensing: Attack on Privacy & Countermeasures
Speck&Tech
 
Wireless survey
ashrawi92
 
Wireless Site Survey
Huda Seyam
 
Datasheet NETSCOUT AirCheck G2 Wireless Tester
PT. Siwali Swantika
 
Conducting site surveys for wlan performance and reliability
Ken Scott
 
The UNH-IOL New WiFi Testing & Measurement Capabilities Webinar
UNH InterOperability Lab
 
Latest wireless technology
nurmeen1
 
Wifi cannerddddddddddddddddddddddddddddddddddddddddddddd
rodainaelhady1
 
WiFi 101
NorthCoastHDI
 
Sensorless sensing with wi fi
ashajuly20
 
Real-time Wi-Fi network performance evaluation
IJICTJOURNAL
 
AndrewLeeming_COSENERS-2014
Andrew Leeming
 
A framework for wireless lan monitoring and its applications
2012engg
 
A Framework for Wireless LAN Monitoring and Its Applications 
2012engg
 
WiFi Data Leakage by Solomon Sonya
EC-Council
 
Web performance prediction using geo statistical method
eSAT Journals
 
Web performance prediction using geo statistical method
eSAT Publishing House
 
The Locator Framework for Detecting Movement Indoors
TELKOMNIKA JOURNAL
 
Insider (Network Tool)
akki_hearts
 
Deceptive Speed: How to Fix Fast, Unstable Wi-Fi
NETSCOUT
 
Wi-Fi Sensing: Attack on Privacy & Countermeasures
Speck&Tech
 
Ad

Recently uploaded (20)

PDF
PSYCHOLOGY IN EDUCATION.pdf ( nice pdf ...)
shekedeganizani2
 
DOCX
UPPER GASTRO INTESTINAL DISORDER.docx
BANDITA PATRA
 
PPTX
NOI Hackathon - Summer Edition - GreenThumber.pptx
MartinaBurlando1
 
PDF
Module 3: Health Systems Tutorial Slides S2 2025
Jonathan Hallett
 
PDF
Open folder Downloads.pdf yes yes ges yes
JaybeeMacadangdang
 
PPTX
The Healthy Child – Unit II | Child Health Nursing I | B.Sc Nursing 5th Semester
RAKESH SAJJAN
 
PPTX
Onica Farming 24rsclub profitable farm business
shivakumar N
 
PDF
From loneliness to social connection charting
giovannaqahal
 
PDF
Mga Unang Hakbang Tungo Sa Tao by Joe Vibar Nero.pdf
MariellaTBesana
 
PDF
O5-L3 Freight Transport Ops (International) V1.pdf
labyankof
 
PDF
grade 11-chemistry_fetena_net_5883.pdf teacher guide for all student
banteamlakmebratu738
 
PDF
Piense y hagase Rico - Napoleon Hill Ccesa007.pdf
Demetrio Ccesa Rayme
 
PDF
Mark Klimek Lecture Notes_240423 revision books _173037.pdf
pascalgumisiriza1
 
PDF
ANTIBIOTICS.pptx.pdf………………… xxxxxxxxxxxxx
HakHe
 
PDF
102 student loan defaulters named and shamed – Is someone you know on the list?
Kweku Zurek
 
PPTX
Open Quiz Monsoon Mind Game Prelims.pptx
Sourav Kr Podder
 
PPTX
Introduction to Child Health Nursing – Unit I | Child Health Nursing I | B.Sc...
RAKESH SAJJAN
 
PPTX
Open Quiz Monsoon Mind Game Final Set.pptx
Sourav Kr Podder
 
PDF
3rd Neelam Sanjeevareddy Memorial Lecture.pdf
Academy of Grassroots Studies and Research of India (AGRASRI)
 
PPTX
How to Manage Starshipit in Odoo 18 - Odoo Slides
Celine George
 
PSYCHOLOGY IN EDUCATION.pdf ( nice pdf ...)
shekedeganizani2
 
UPPER GASTRO INTESTINAL DISORDER.docx
BANDITA PATRA
 
NOI Hackathon - Summer Edition - GreenThumber.pptx
MartinaBurlando1
 
Module 3: Health Systems Tutorial Slides S2 2025
Jonathan Hallett
 
Open folder Downloads.pdf yes yes ges yes
JaybeeMacadangdang
 
The Healthy Child – Unit II | Child Health Nursing I | B.Sc Nursing 5th Semester
RAKESH SAJJAN
 
Onica Farming 24rsclub profitable farm business
shivakumar N
 
From loneliness to social connection charting
giovannaqahal
 
Mga Unang Hakbang Tungo Sa Tao by Joe Vibar Nero.pdf
MariellaTBesana
 
O5-L3 Freight Transport Ops (International) V1.pdf
labyankof
 
grade 11-chemistry_fetena_net_5883.pdf teacher guide for all student
banteamlakmebratu738
 
Piense y hagase Rico - Napoleon Hill Ccesa007.pdf
Demetrio Ccesa Rayme
 
Mark Klimek Lecture Notes_240423 revision books _173037.pdf
pascalgumisiriza1
 
ANTIBIOTICS.pptx.pdf………………… xxxxxxxxxxxxx
HakHe
 
102 student loan defaulters named and shamed – Is someone you know on the list?
Kweku Zurek
 
Open Quiz Monsoon Mind Game Prelims.pptx
Sourav Kr Podder
 
Introduction to Child Health Nursing – Unit I | Child Health Nursing I | B.Sc...
RAKESH SAJJAN
 
Open Quiz Monsoon Mind Game Final Set.pptx
Sourav Kr Podder
 
3rd Neelam Sanjeevareddy Memorial Lecture.pdf
Academy of Grassroots Studies and Research of India (AGRASRI)
 
How to Manage Starshipit in Odoo 18 - Odoo Slides
Celine George
 

Slides for Ph.D. Thesis Defense of Dheryta Jaisinghani at IIIT-Delhi, INDIA

  • 1. Understanding the Role of Active Scans for their Better Utilization in Large-scale WiFi Networks Ph.D. Thesis Defense of Dheryta Jaisinghani On Monday April 22, 2019 Advised By: Vinayak Naik and Sanjit K. Kaul
  • 2. Agenda 2 1 Background ● WiFi Networks ● Connecting with an Access Point ● Scanning Scope ● Problems Investigated ● Methodology Followed ● Challenges 2 Problems ● Unnecessary Active Scans ● WiFi Indoor Localization ● Data Transfer with Active Scans 3 Conclusion ● Summary ● Contributions to Existing and New Standards ● Future Work 4
  • 3. WiFi Network Architecture ... Controller AP1 WiFi Clients AP2 APN WiFi Association Ethernet 3 AP- Access Point BSSID1 BSSID2 BSSIDN ...
  • 4. How does a Client Connect with an Access Point? 4 Step 3 Data Transmission and Reception Step 1 Discover an Access Point Step 2 Associate and Authentication Handshake with the Access PointAn essential initial step that enables steps 2 and 3 and much more!
  • 5. Discovering an Access Point 5 Listen Beacons ... ... Beacons AP Client Probe Requests ... ... Probe Responses AP Client Passive Scanning Active Scanning Slow and Not Preferred Default Discovery Protocol
  • 6. Why Focus on Discovery? Modern WiFi networks provide more than just basic network access functionality 6 IoTs Localization Environment Sensing Access Point discovery mechanisms continue to be traditional *Images from Internet (Open Access) * *
  • 7. Agenda 7 1 Background ● WiFi Networks ● Connecting with an Access Point ● Scanning Scope ● Problems Investigated ● Methodology Followed ● Challenges 2 Problems ● Unnecessary Active Scans ● WiFi Indoor Localization ● Data Transfer with Active Scans 3 Conclusion ● Summary ● Contributions to Existing and New Standards ● Future Work 4
  • 8. Problem Statement ● Investigate the mechanisms of discovery in large-scale and dense WiFi networks ● Default protocol - Active Scanning is crucial ○ Excessive amounts ⇒ Performance⬇ ○ Arbitrary reduction ⇒ Performance⬇ ● A thorough relook needed 8
  • 9. Problems Investigated 9 Harnessing Active Scans for Data Transfer in WiFi-based IoT Nodes 03 ● Best Use of Active Scans WiFi Indoor Localization using Existing Infrastructure in the presence of Minimal Active Scans 02 ● Minimal Active Scans Understanding and Mitigating the Impact of Unnecessary Active Scans in WiFi Networks 01 ● Excessive Active Scans
  • 10. Empirical Methodology Followed Sniff WiFi Traffic Analyze and Infer Design a Solution Implement the Solution 10 Management Probe, Beacons, Associate, Authenticate Data Qos/Non-QoS Data Control RTS, CTS, ACK, Polling IEEE 802.11 Frames
  • 11. Challenges ● Work with actual network deployments ○ Dynamic environments ● Real-time analysis and validation ○ Precision ● Multitude of devices [>15,000 vendors] ○ One solution does not fits all ● Large-scale infrastructures ○ Integrable solutions 11
  • 12. Agenda 12 1 Background ● WiFi Networks ● Connecting with an Access Point ● Scanning Scope ● Problems Investigated ● Methodology Followed ● Challenges 2 Problems ● Unnecessary Active Scans ● WiFi Indoor Localization ● Data Transfer with Active Scans 3 Conclusion ● Summary ● Contributions to Existing and New Standards ● Future Work 4
  • 13. Problem#1* Understanding and Mitigating the Impact of Unnecessary Active Scans in WiFi Networks 13 1. D. Jaisinghani, V. Naik, S. K. Kaul, R. Balan, and S. Roy, “Improving the Performance of WLANs by Reducing Unnecessary Active Scans,” ArXiv e-prints, Jul. 2018. [Under Review] 2. D. Jaisinghani, V. Naik, S. K. Kaul, and S. Roy, “Realtime detection of degradation in WiFi network's goodput due to probe traffic,” in WinMee -WiOpt, 2015, May 2015. 3. D. Jaisinghani, V. Naik, S. Kaul and S. Roy, “Sniffer-based Inference of the Causes of Active Scanning in WiFi Networks,” in NCC,Chennai, India, Mar. 2017. 4. (H. Fulara, G. Singh)*, D. Jaisinghani and M. Maity and T. Chakraborty and V. Naik, “Use of Machine Learning to Detect WiFi Networks from Unnecessary Active Scans,” WoWMoM, 2019. [*Equal Contributions] # Best Graduate Forum Presenter Award at COMSNETS 2018
  • 14. Research Question What is the extent and the impact of active scans and how to deal with them? 14
  • 15. The Extent of Active Scans ● Analyzed 3 real-world datasets ○ Sigcomm [5 GHz, 2 Days, 14 Hours] ○ IIT-B [2.4 GHz, 1 Day, 2.5 Hours] ○ IIIT-D [2.4 GHz, 9 Days, 63 Hours] ● Clients probe lesser in 5 GHz [Sigcomm] ● Client Density ↑, Probe Requests ↑ [IIT-B] ● Frame size depends on clients and BSSIDs [50 - 300 Bytes] ● Probe Requests and Responses can arrive as early as every 0.01 ms ● Probe Response arrival time depends on #BSSIDs ● 2.4 GHz has lowest bitrate=1 Mbps, 5 GHz has lowest bitrate=6 Mbps ● 99% Probe Responses fetch redundant network information for stationary WiFi clients 15
  • 16. Active Scanning - 2.4 GHz vs 5 GHz 16 ● WiFi networks today operate on both 2.4 GHz and 5 GHz - Different channel characteristics ● Previous datasets could not be used because, ○ SIGCOMM is old (2008) with minimal evidence ○ IIIT-Delhi < 5 clients/day in 5 GHz ○ IIT-B - 2.4 GHz ● ∴ Venue with dual-band devices needed ○ A university in Singapore ○ Dual-Band Aruba WiFi Network ○ Comparable number of clients in both bands ○ 6 hours of traffic collected
  • 17. Effect of RSSI drop on #Status ChangesEffect of RSSI drop on #Probe Requests Comparison Results - 2.4 GHz vs 5 GHz 17 5 GHz sees active scans when RSSI drops 2.4 GHz sees active scans at all RSSI Stationary clients scan minimally in 5 GHz 5 GHz sees status change when RSSI drops 2.4 GHz sees status changes at all RSSI
  • 18. Importance of 2.4 GHz ● 2.4 GHz is common in developing countries like India ● Dual band WiFi devices are costly ● 85% WiFi devices still operate in 2.4 GHz ● 2.4 GHz will be the likely choice for large-scale IoT deployments 18 We focus on 2.4 GHz band
  • 19. Impact from Client-side Perspective - Latency 19Mean latency ↑ from 3.57 ms to 21.72 ms [6x ↑] ● Stationary WiFi client ● Associated ● AP-RSSI - Good ● Logs ○ Time of Scan ○ Latency ● Duration: 5.5 hours
  • 20. Impact from Client-side Perspective - Association Patterns 20 38% Discovered 11, Chose 3 to associate ● Stationary WiFi clients (34) ● Associated ● AP-RSSI - Good ● Logs ○ Discovered APs ○ Associated APs ● Duration: 10 days Clients discover more, associate to few
  • 21. Impact from Network-wide Perspective - Controlled Setup 21 Growth in Probe Traffic results in drop in Bandwidth [ 20 Mbps -> 10 Mbps]
  • 22. Impact from Network-wide Perspective - Uncontrolled Setup 22
  • 23. Impact from Network-wide Perspective - Uncontrolled Setup - Results 23 ACK losses - 2.6x ⬆ Frame retries - 2.19x ⬆ Goodput - 50% ⬇ Probe traffic - ~80% ⬆ Channel Utilization >= 70%
  • 24. Addressing the Problem ● Problem ○ Stationary clients do not benefit from active scans ○ Goodput drops in heavily utilized WiFi networks ● Challenges ○ Active Scanning is Crucial Process - Cannot Disable ○ Passive Scanning is slow - Not Preferred ○ Fast Solution - Strict timing constraints of WiFi protocol ○ ⬆Probe Traffic ⇔ ⬇Data Traffic - Detecting exact cause is hard 24
  • 25. Current Solutions and their Limitations ● Problem Recognition [IMC ‘15, TMC ‘10, 802.11ax] ● Reduce scanning delays ○ Modified scanning algorithms in [ICC ‘12, PERCOM ‘10] ○ Multiple radios/MPTCP [NSDI ‘15, TVT ‘14] ○ Channel frequency response [MOBICOM ‘13] ● Partially reduce probe traffic ○ Passive scanning or disable probe response on APs [WN ‘10, WoWMoM ‘09] ● Systems for improving the performance of WiFi networks ○ PIE, JIGSAW [NSDI ‘11, SIGCOMM ‘06] 25Device Changes, Partial Reduction, Reduce Interference
  • 26. Our Solutions 26 ● Detect the growth of probe traffic - Validates the cause ● Infer the cause of growth in probe traffic - Better Network Planning ● Reduce probe requests from client-side - Inhibits Probe Responses
  • 27. Detect Growth of Probe Traffic 27 Real-time growth in probe traffic ● Assumption - There is always data to be sent ● Cumulative increase of P/D > 1 is plotted against time ● Slope >= 0.10 => Goodput drop [90% accurate] Case Metric D Frames P Frames Probe Traffic ⬆ 1 P/D < 1 ✔ ✖ ✖ 2 P/D = 1 ✔ ✔ ✖ 3 P/D > 1 ✖ ✔ ✔ Probe Frames Fresh Data Frames
  • 28. How do Clients Decide When to Scan? 28 Discovery Connection Establishment Connection Maintenance1. Periodic Scan 2. Associated State: Unassociated to Associated 3. Power State: Low to High 4. Loss of Beacons 5. AP-side Procedures 6. Low RSSI 7. Data Frame Losses Causes of Active Scans Applications (wpa_supplicant, network manager) MAC Drivers (cfg80211, mac80211) Device Drivers (ath9k, iwlwifi) WiFi Chipset (AR9462, BCM4322) User space Kernel space Hardware At client-side, it's a non-trivial process It can be triggered at any of these layers!
  • 29. Inferring the Causes of Active Scanning 29 ● Frames received in real-time ● Every cause has a distinguished signature ● Defined static rules and metrics for each cause ● DFA-based detection with states and events Sniffer-based Inference Mechanism
  • 30. An Example 30 AP-side Procedures ● Connection Maintenance → AP-side Procedures ● Load balancing/Rogue client ● AP deauthenticates client ● Client Scans
  • 31. Extent of Causes in Real Datasets 31 Periodic Scans -> Most Common Network Planning Intermittent user activity Network Contention
  • 32. Applying Inference Mechanism to Real-world Networks ● Unassociated Periodic Scans ⇒ Unassociated clients ⇒ Rogue clients ● Associated Periodic Scans ⇒ Aggressive clients ● Connection Maintenance ⇒ ↓ Coverage or ↑ Interference ● AP-side Procedures ⇒ Faulty Controller Algorithm 32 Network administrators can improve network planning
  • 33. Reduce Probe Requests From the Client-end 33 ● Revisit the causes of Active Scans ● Stationary WiFi clients ⇒ Active Scans ❌ ● Mobile WiFi clients ⇒ Passive Scans ❌ ● Time Constraints - Connection Establishment/Maintenance ● Dynamic Decision Associated? Handover No Stationary? Connection Maintenance Passive Scans Yes YesNo Active Scans Connection Establishment? Yes No
  • 34. Client-end Solution: Experiment Setup 34 ● Live IIIT-D network ● Modified and Unmodified clients - 10 each ● Introduced causes of active scans every hour ● Client collects all connection logs ● Sniffer logs WiFi traffic ● Duration: 9 hours
  • 35. Client-end Solution: Results 35 Metric Modified Clients Unmodified Clients #Probe Requests 500/hour 932/hour Time to Connect 0.25 seconds 4.35 seconds Connection Persistence 1 minute - 0.37 5 minutes - 0.55 15 minutes - 0.80 1 minute - 0.20 5 minutes - 0.43 15 minutes - 0.73 Simple client intelligence works
  • 36. Key Takeaways ● Device-agnostic measures without special h/w or s/w needs ● Most APs come with sniffing antennas ● Easy to integrate with existing controller ● Sniffers may miss frames ● Multiple sniffers needed for comprehensive network view 36 Excessive Active scans affect performance in 2.4 GHz
  • 37. Problem#2* WiFi Indoor Localization using Existing Infrastructure in the presence of Minimal Active Scans 37 5. D. Jaisinghani, R. Balan, V. Naik, Y. Lee, and A. Misra, “Experiences & Challenges with Server-Side WiFi Indoor Localization Using Existing Infrastructure,” in MobiQuitous, 2018, New York, USA..
  • 38. Research Question How to localize several thousands of devices in real-world indoor deployments? 38
  • 39. The Challenges ● Infrastructure, i.e., the controller and the access points changes ● Devices cannot be modified in any way ○ Explicit/implicit participation for data generation ○ App download ○ Chipset changes ○ Extra antennas 39
  • 40. State-of-the-art Solutions ● Accuracy of few centimeters is possible with, a. Sophisticated techniques such as AoA, ToF/ToA, RToF, and PoA [NSDI ‘13, MobiCom ‘14, SIGCOMM ‘15] b. Require extra hardware such as specialized antennas and ultrasonic systems [MobiSys ‘05, NSDI ‘13, MobiCom ‘14, SIGCOMM ‘15] c. Special requirements such as Line-of-Sight, Visible Light, Acoustics [NSDI ‘14, Mobiquitous ‘15, MobiSys ‘15, MobiSys ‘18] Can they be applied to a large-scale network where several thousands of WiFi devices are to be localized? NO Customized hardware, Client application, Rooting the Client OS, High Energy Budget, High Cost of deployment - Limit Generalizability and Scalability 40
  • 41. What Do We Have? ● Lots of access points ● Few controllers that manage these access points ● Access points can measure RSSI from the clients ● Only available data is RSSI measurements from access points - ○ We can at least attempt to localize ○ That's all we have 41
  • 42. Solution Fingerprint-Based WiFi Localization ● Scalability ○ Albeit the manual effort for setup, fingerprinting-based localization is scalable ● Device Friendly ○ No client-side changes / No AP-side changes ● Integrable ○ Seamless integration with existing infrastructure is possible Premise of fingerprint-based solutions - Clients always scan! 42
  • 43. Fingerprinting Process Offline Online Fingerprints ∀ Li ∈ L, Known landmark Li, RSSI Vector= < Li , B, AP1 : RSSIi ; ...; APn : RSSIn ; > Unknown landmark Lx, RSSI Vector= < B, AP1 : RSSI'i ; ...; APn : RSSI'n ; > Distance, Di in Signal Space between Lx and ∀ Li ∈ L, Di = sqrt((RSSIAP1 - RSSI' AP1 )2 + (RSSIAP2 - RSSI' AP2 )2 + … + (RSSIAPn - RSSI' APn )2 ) Predicted Landmark = Li with minimum Di Already stored in database Collected in real-time 43
  • 45. System Architecture (Contd.) ● 750+ dual-band access points ● Centrally controlled by 11 WiFi controllers ● 4000 associated clients per day ● RTLS data feed received every 5 seconds 45 Timestamp Age ChannelClient MAC AP Association Status Data Rate RSSI RTLS Data Feed
  • 46. Characterizing the Landmarks 46 Landmarks = Water Sprinklers on the ceiling Landmarks across floors (Overlapping Region A) Landmarks across floors (Overlapping Region B) Landmarks = water sprinklers, deployed every 3 meters
  • 47. Ground Truth Data Collection 47 3203 landmarks, 86 hours data ● 4 different WiFi states ○ Emulate real-world usage scenarios ○ Modulate Scanning Frequency ○ Disconnected → Inactive → Intermittent → Active WiFi On State: Disconnected WiFi On State: Connected and Inactive WiFi On State: Connected and Intermittent WiFi On State: Connected and Active Scanning Frequency Low High Timestamp at every landmark is recorded at granularity of seconds
  • 48. Data Processing Challenges ● Traffic Distinction ○ Scanning Records ○ Non-Scanning Records ● Offline fingerprints ➡ Scanning records ● Online fingerprints ➡ Scanning + Non-scanning records ● RTLS data feeds does not distinguish ● Proposed inference ○ Scanning records - fixed and lowest data rates ○ Non-scanning records - variable data rates 48
  • 49. Problems Discovered 49 ● Cardinality: Set of access points reporting for a client ● Fingerprinting-based Solutions - Effective and Efficient match ○ Clients always trigger active scans ○ Same set of APs respond ● Online and offline cardinalities are different - #Matches Affected ○ Cardinality mismatch ○ RSSI mismatch ○ High Client Scan Latency Result → Clients are teleported !!
  • 50. Evidence 1 - Cardinality Mismatch 50 OnlineOffline 5 GHz - 80% 2.4 GHz - 40% Cardinality ⬇ from 16 ➡ 6 (Max) Minimal Access Points to Match
  • 51. Evidence 2 - RSSI Mismatch 51 Client close to AP Client far from AP 50 dB 5 dB 30 dB 10 dB Non-Scanning Frames Scanning Frames Scanning Frames are Better
  • 52. Evidence 3 - High Client Scan Latency 52 Very few scans Scans ⬆ 5 GHz Scans ⬇ Dual-Band Scans ❌ Stationary client do NOT scan
  • 53. Cause 1 - Cardinality in Presence and Absence of Scans Scans PresentScans Absent No Access Points Reports Max. Access Points in 5 GHz No of Access Points ⬆ Max. Access Points in 5 GHz No Access Points Reports Scans + 2.4 GHz is Better 53
  • 54. Cause 2 - RSSI variation 2.4 GHz and 5 GHz But, RSSI variation in 2.4 GHz > 5 GHz (Higher range and interference) 5 GHz is Better 54
  • 55. Summary of the Problems 55 Distance of Transmission Variation in RSSI #Scanning Frames 2.4 GHz High High High 5 GHz Low Low Low Distance of Transmission Variation in RSSI Frequency Scanning Frames High Low Low Non-scanning Frames Low High High Improves accuracy Reduces accuracy Non-Trivial to solve
  • 56. Recommendations to Reduce the Localization Errors 56 Baseline Maximum #APs AP with Maximum RSSI AP of Association No floor detection logic Floor from which maximum number of APs report the client Floor from which strongest RSSI is received Floor of AP to which client is associated with Detect floor before localizing Metrics: Floor Error and Same Floor Error
  • 57. An Example 57 Floor 2 AP1 AP2 AP3 AP4 Floor 1 Client -50 dBm -60 dBm -70 dBm -52 dBm Floor 3 Maximum Number of APs Floor 2 AP2 AP3 AP4 Floor 1 -50 dBm -60 dBm -70 dBm -52 dBm Floor 3 Client AP1 AP with Maximum RSSI Floor 2 AP2 AP3 AP4 Floor 1 Client -50 dBm -60 dBm -70 dBm -52 dBm Floor 3 AP1 AP of Association
  • 58. Reduction in Localization Errors ● 5 GHz band outperforms 2.4 GHz band ● Worst Heuristic - Maximum APs ○ Reason - Distant APs also respond ● Best Heuristic - AP of association ○ 17 meters (2.4 GHz), 20 meters (5 GHz) ○ Floor Errors- 78% ⬇ (2.4 GHz), 46% (5 GHz) ○ Same Floor Errors - 58% ⬇ (2.4 GHz) , 3.8% (5 GHz) ○ Reason - Clients associate with the nearest AP ● Compared with SignalSLAM [IPIN ‘13] ○ ~4 meters better, no client-side changes 58
  • 59. Key Takeaways ● Few meters accuracy is OK ○ Contextual Location Aware Applications - Marketing/Business use-cases ○ Tracking assets - employees/students ● Simple floor-detection heuristics can combat the challenges ● Limited by RTLS information 59 Reduced active scans hamper services like Localization
  • 60. Problem#3 Harnessing Active Scans for Data Transfer in WiFi-based IoT Nodes 60 6. D. Jaisinghani, H. Fulara, G. Singh and M. Maity and V. Naik, “Demo: Elixir - Sending Data in WiFi-based IoT without the Data Frames,” Mobicom, 2018, New Delhi, INDIA 7. D. Jaisinghani, H. Fulara, G. Singh, M. Maity and V. Naik, “Sending Data in WiFi-based IoT without the Data Frames,” Under Review. # Won First prize at nationwide WiFi competition - WiFi ThinkFest 2017
  • 61. When WiFi Fails 61 A WiFi Association is mandatory to enable data transfer Case Signal Quality Associated? Cause Effect 1 Bad ✖ Sparse AP Deployment No Association 2 Bad ✔ Reduced RSSI Connection Maintenance/Handovers 3 Good ✖ Overloaded APs/Admin Policies No Association 4 Good ✔ High Channel Utilization and Losses Connection Maintenance/Handovers
  • 62. Research Question Is it possible to transfer data in the absence of an ongoing WiFi association, for example, in challenged scenarios? 62
  • 63. Default States of a WiFi Client 63 Wake UpSleep Transmit Data Interface ON Scan Associated Discover APs Check Association Successful Association Data Unavailable Data Available Failed Association Until this is done This is not possible
  • 64. Proposed Solution - ViFi Protocol [Scan (V) comes before Associate (W)] 64 Probe Requests Probe Responses Exchange of Data Access Points IoT Nodes Wake Up, Scan & Transmit Data Data Available Idle Transmission Complete Interface ON Data Available + The Idea Reduction in States WiFi Clients always scan - Active Scan
  • 65. Assumptions ● IoT nodes are sensors ● Data to be transmitted is small (few bits) and infrequent ● IoT nodes can trigger active scans ● At least 1 AP is available ● AP is open and configurable 65
  • 66. Challenges to Implement ViFi ● Conformance to Existing Standards ○ No new packets should be introduced ● Simple to Deploy ○ Should be capable of operating on diverse set of devices ● Reliable Data Transfer ○ Should be able to transmit data at low/variable RSSI ● Co-exist Seamlessly with Default Protocols ○ Performance of nodes with default protocols should not be hampered 66
  • 67. State-of-the-Art Solutions ● Management frame stuffing [HotMobile ‘07, MSR Patent ‘05] ○ Interference Management [ arXiv ‘17 ] ○ Traffic Management [ VTC ‘14, AINA ‘16 ] ○ Localization [ TENSYMP ‘16 ] ○ Disaster Recovery [ Mobisys ‘16 ] ○ Location-based Information Services [ WMNC ‘16] 67 Application-based solutions, performance impact?
  • 68. Protocol Implementation 68 Userspace App Libnl Library Client Userspace App Libnl Library AP Probe Request Probe Response WiFi System Architecture MAC Header (22) Frame Body (2320) FCS (4) Fixed Fields Elements Element ID (1) Length (1) OUI (3/5) Content Default Frame Structure Sequence Number More Data Length Data Proposed Frame Structure
  • 69. Proposed Protocol 69 Client AP1 Probe Response#1 AP2 APN Channel 1 Channel 6 Channel 11 Probe Response#6 Probe Response#11 Probe Request#1 Probe Request#6 Probe Request#11 Stuff Data in Probe Request Trigger Active Scan Implicit Acknowledgement Extract Data Extract Data Extract Data
  • 70. Evaluation 70 Compare WiFi and ViFi Impact of ViFi on WiFi Experiments ● IoT Nodes - Raspberry Pi ● AP - OpenWRT and hostapd ● Network conditions - Good and Bad ● Data - 300 Bytes ● Data generation intervals - 1 second and 120 seconds
  • 71. Compare WiFi and ViFi 71 IoT Nodes - WiFi Protocol Access PointSniffer IoT Nodes - ViFi Protocol Compare the effect
  • 72. Number of Frames 72 WiFi transmits ~470% ⬆ frames, 32% retries WiFi continues to send discovery frames
  • 73. Airtime Utilization 73 WiFi consumes at least 8x higher Airtime
  • 74. Delivery Rate & Goodput 74 ViFi delivers better or at par with no retries ViFi maintains better or at par goodput
  • 75. Impact of ViFi on WiFi 75 Non-IoT Nodes - WiFi Protocol Access PointSniffer IoT Nodes - ViFi Protocol Non-IoT Nodes - WiFi Protocol Access PointSniffer IoT Nodes - WiFi Protocol Compare the effect Unchanged
  • 76. Contention Time ● When contention/collisions ⬆ ⇒ Inter-frame arrival time increase ⬆ ● Affects all upper layer metrics - throughput, jitter, latency ● ⬆Contention time ⇒ nodes are struggling 76 IoT nodes Contention faced by non-IoT nodes ViFi reduces contention
  • 77. Key Takeaways ● ViFi serves as an Elixir - when WiFi fails ● ViFi eliminates network traffic ● Saves up to 18% energy ● Apt for low bit-rate applications - ambient data collection, emergency broadcasts, ultra dense environments like India's Kumbh Mela ● In Good network conditions, ViFi experiences higher delay ● Data generation interval needs to be well thought off ● Not suited for high-end data intensive transfers 77
  • 78. Agenda 78 1 Background ● WiFi Networks ● Connecting with an Access Point ● Scanning Scope ● Problem Statement ● Methodology Followed ● Challenges 2 Problems ● Unnecessary Active Scans ● WiFi Indoor Localization ● Data Transfer with Active Scans 3 Conclusion ● Summary ● Contributions to Existing and New Standards ● Future Work 4
  • 79. Conclusion ● Detailed analysis on active scanning protocol ● Methods to mitigate impact of active scans ● Practical challenges in localizing clients ● ViFi protocol for low-end IoT nodes ● Datasets and source codes are public 79 Active Scanning is default discovery protocol, all latest and upcoming standards will benefit from our device-agnostic approaches
  • 80. Future Work ● Performance enhancements with Software Defined Networking and Machine Learning ● Device-agnostic WiFi-based indoor localization ● Evaluate Association-free WiFi (The proposed ViFi) for ○ Security ○ Recovery ○ Downlink data transfer ○ MAC address randomization 80
  • 81. Publications 1. D. Jaisinghani, V. Naik, S. K. Kaul, R. Balan, and S. Roy, “Improving the Performance of WLANs by Reducing Unnecessary Active Scans,” ArXiv e-prints, Jul. 2018. [Under Review] 2. D. Jaisinghani, V. Naik, S. K. Kaul, and S. Roy, “Realtime detection of degradation in WiFi network's goodput due to probe traffic,” in WinMee -WiOpt, May 2015, Bombay INDIA. 3. D. Jaisinghani, V. Naik, S. Kaul and S. Roy, “Sniffer-based Inference of the Causes of Active Scanning in WiFi Networks,” Mar. 2017, NCC, Chennai, INDIA. 4. (H. Fulara, G. Singh)*, D. Jaisinghani, M. Maity and T. Chakraborty and V. Naik, “Learning to Rescue WiFi Networks from Unnecessary Active Scans,” WoWMoM, 2019. Washington DC, USA [Core A, *-Equal Contributions] 5. D. Jaisinghani, R. Balan, V. Naik, Y. Lee, and A. Misra, “Experiences & Challenges with Server-Side WiFi Indoor Localization Using Existing Infrastructure,” in MobiQuitous, 2018, New York, USA. [Core A] 6. D. Jaisinghani, H. Fulara, G. Singh, M. Maity and V. Naik, “Demo: Elixir - Sending Data in WiFi-based IoT without the Data Frames,” Mobicom, 2018, New Delhi, INDIA [Core A*] 7. D. Jaisinghani, H. Fulara, G. Singh, M. Maity and V. Naik, “Sending Data in WiFi-based IoT without the Data Frames,” Under Review. 81
  • 82. Acknowledgements 82 Vinayak Naik IIIT-Delhi, INDIA Sanjit K. Kaul IIIT-Delhi, INDIA Sumit Roy U. Washington, USA Rajesh Balan SMU, Singapore Mukulika Maity IIIT-Delhi, INDIA Sambuddho Chakravarty IIIT-Delhi, INDIA Vivek Bohara IIIT-Delhi, INDIA Karthik Dantu U. Buffalo, USA Nirupam Roy U. Maryland, USA Srihari Nelakuditi U. South Carolina, USA Mythili Vutukuru IIT-Bombay, INDIA
  • 84. Salient Features of Active Scanning ● Probe Requests and Probe Responses are transmitted at lowest bit-rate ● Probe traffic is broadcast traffic ● Stationary WiFi clients trigger active scanning unintelligently ● Frequent active scans fetch same information ● WiFi clients consume ~44.3% higher energy and lose up to 16.7% throughput [IMC'15] ● Performance of clients suffer in dense WiFi networks due to unnecessary active scans [TMC'10] ● Recognized in IEEE 802.11ax - High Efficiency WiFi Network (HEW) status update meeting 84
  • 85. The Protocol of Active Scanning Client AP1 Probe Response#1 AP2 APN Ch# 1 Ch# 6 Ch# 11 Probe Response#11 Probe Response#6 Probe Request#1 Probe Request#6 Probe Request#11 ACK#1 ACK#6 ACK#11 85
  • 86. Effect of RSSI drop on #Probe Requests Comparison Results - 2.4 GHz vs 5 GHz 86 5 GHz sees active scans when RSSI drops 2.4 GHz sees active scans at all RSSI Stationary clients scan minimally in 5 GHz [Confirmed with #Status Changes Metric] 5 GHz sees more active scans? #Probe Requests Recorded Per Minute Per Client
  • 87. Uncontrolled Network Setup - Channel Utilization 87
  • 88. Devices Studies for Causes of Active Scans 88 Device Type Chipset Vendor Operating System Device Driver Laptop Atheros, Intel, Broadcam Ubuntu 14.04/12.10 and Windows 8.1/7 Ath9k, iwlwifi Tablets Qualcomm, Broadcom Android KitKat and iOS 8.1 OS provided Smartphones Broadcom, MediaTek, Qualcomm Android KitKat, Marshmellow, JellyBean, Cyanogen, and Windows OS Provided USB Adapter Realtek and Atheros Ubuntu 14.04/12.10, Window 8.1/7 rtl8812AU, ath9k_htc
  • 92. Other Empirical Observations ● ↑ History of SSIDs ⇒ ↑ Probe Requests ● WiFi clients search for non-existing APs ● APs respond to all clients ● Dense deployment of APs ⇒ ↑ Probe Responses 92
  • 93. Problem Statement - Localization Highlight challenges and propose easy to integrate solutions to build a universal indoor localization system – one that can spot localize all WiFi enabled devices on campus without any modifications to the client or infrastructure-side 93
  • 94. Improvement in Localization Errors 94 Baseline Maximum #APs AP with Maximum RSSI AP of Association 2.4 ghz 5 ghz 2.4 ghz 5 ghz 2.4 ghz 5 ghz 2.4 ghz 5 ghz Cardinal ity Same Floor Floor Error Same Floor Floor Error Same Floor Floor Error Same Floor Floor Error Same Floor Floor Error Same Floor Floor Error Same Floor Floor Error Same Floor Floor Error 1 >50 20.00 29.50 3.10 >50 48.00 29.60 03.49 49.20 14.70 28.30 02.09 32.30 04.90 27.60 01.28 2 >50 18.30 26.80 02.00 >50 33.33 27.00 05.60 24.20 05.60 27.00 03.46 21.60 00.70 25.80 01.48 3 >50 33.15 20.12 00.00 >50 40.00 24.00 00.00 36.00 14.70 20.12 00.00 22.80 05.50 20.12 00.00 4 >50 18.49 NA 00.00 33.00 13.40 NA NA 24.00 11.00 NA NA 18.90 04.40 NA NA 5 19.20 10.25 NA NA 26.00 14.60 NA NA 17.49 00.00 NA NA 17.49 00.00 NA NA 6 23.00 04.17 NA NA 22.00 00.00 NA NA 16.00 00.00 NA NA 22.00 00.00 NA NA
  • 96. Large Dataset for Sniffer Based Inference 96
  • 97. Problem Statement - ViFi We consider WiFi-enabled IoT nodes and APs. The IoT nodes are sensor nodes, who want to transmit few bits of infrequent data. They can trigger an active scan and at least one access point responds. In such a network, we want to enable the IoT nodes to transmit data to an AP, in the absence of a WiFi association with any AP. 97