Transcript 14-More-Sensingx

More Security and Programming
Language Work on SmartPhones
Karthik Dantu and Steve Ko
WiSee: Intro
• Pu et al., “Whole-Home Gesture Recognition Using
Wireless Signals” from MobiCom’13
• An in-air gesture recognition system
• Think Xbox Kinect without any sensor
• Uses Wi-Fi signals and their Doppler shifts
• Works in line-of-sight, non-line-of-sight, and
through-the-wall conditions (94% best-case
accuracy)
Gestures
Two Questions
• How to detect gestures using Wi-Fi?
• Short answer: Doppler shifts & pattern matching
• How to deal with other humans in the environment?
• Short answer: Repeated gesture to detect the
user using MIMO
Gesture Detection
• Doppler shifts
• Frequency change between two objects when
moving
• E.g., Train coming towards you (higher observed
frequency) & moving away from you (lower
observed frequency)
• Wi-Fi Doppler shifts
• Humans reflect Wi-Fi signals, thus can be treated
as signal sources
• Different gestures exhibit different patterns.
Gesture Detection
• Doppler shifts with Wi-Fi and human gestures
• Problem: the frequency variation is too little.
• With 5 GHz transmission, 17 Hz Doppler shift (but
Wi-Fi channel is typically 20 Mhz).
• Question: how to amplify this frequency variation?
• Short answer: combine variation from multiple,
identical OFDM symbols to amplify
Gesture Detection
• A bit of background
• OFDM is the modulation scheme for Wi-Fi.
• It divides a Wi-Fi channel into multiple subchannels.
• One OFDM symbol (i.e., bits to send) gets sent
over the channel by sending one bit for each subchannel.
• If the sender sends the same OFDM symbols multiple
times, we can amplify the Doppler shift.
Gesture Detection
Gesture Detection
• One more issue: Wi-Fi signals are not coming from
exact same data repeatedly sent.
• Solution
• Data (bits) don’t matter, we’re only interested in
Doppler shifts.
• Data equalizer: the receiver re-generates OFDM
symbols with the same data from the “first”
OFDM symbol.
Patterns for Gestures
Patterns for Gestures
Evaluation
• Software radio receiver
• Scenarios
• Office building
• Two-bed apt
• Many conditions
• Line-of-sight, non-line-of-sight, through-the-wall,
through-the-corridor, through-the-room
Scenarios
Evaluation
Evaluation
ACE: Intro
• Nath, “ACE: Exploiting Correlation for EnergyEfficient and Continuous Context Sensing” from
MobiSys’12
• Context sensing
• Can a phone detect what a user’s context is?
• Walking, driving, at home, in a meeting, etc.
• ACE is a system that provides this in an energyefficient way.
Questions for Context Detection
• How accurate can we be when detecting a context?
• Not a focus for ACE
• Much work has been done
• Various sensors are used.
• How can context detection shared by multiple apps?
• Focus of ACE
• ACE does it in an energy-efficient way.
Key Insight
• Context inference is possible.
• Examples:
• If walking, then not driving.
• If at home, then not at office.
•
If an app wants to know one attribute, it can
infer other attributes when a different app wants
to know other things.
Using the Key Insight
• Contexters: context detection modules
• Rule miner: keeps the history of contexts and
discovers relationships between rules.
• Context cache: keeps recently discovered contexts
and other inferred contexts (using the rule miner).
• Sensing planner: given a context to discover, find
the cheapest sensors to get the context.
Architecture
Workflow
Contexters
Rule Miner
• Mines per-user rules using Apriori association rule
mining algorithm
Inference Cache
• Each entry has an attribute and expiration time.
• E.g., IsDriving, 5 minutes
• Inference cache uses relationship rules to return
inferred attributes as well.
Sensing Planner
• Due to the relationship rules, some attributed do
not need to discovered directly.
• Optimize energy by using less energy consuming
sensors
Evaluation
• Sensing planner
Evaluation
• Inference cache
Evaluation
• Per user power consumption
Summary
• WiSee: a gesture detection using Wi-Fi
• ACE: a energy-efficient context detection system