Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Footfall – GPS Polling Scheduler for Power Saving on Wearable Devices
Kent W. Nixon, Xiang Chen, Yiran Chen
University of Pittsburgh
January 28, 2016
Department of Electrical & Computer Engineering
Talk Outline
•Background
•Current GPS Usage Model
•Footfall Design and Implementation
•Experiment and Evaluation
•Conclusions
Department of Electrical & Computer Engineering
Smart Wearables
•Computing device that is worn on the body• Smartwatch, fitness tracker, eyewear, etc.
•Growing segment of electronics market• 350 million devices installed by 2020
*http://www.ccsinsight.com/press/company-news/2332-wearables-market-to-be-worth-25-billion-by-2019-reveals-ccs-insight
Department of Electrical & Computer Engineering
Wearable Use
•Most common use of smartwatches is fitness/activity tracking
•Fitness/activity trackers projected to make up more than 50% wearable sales in 2019
Department of Electrical & Computer Engineering
Tracking Fitness
•Accomplished with embedded sensors• Accelerometer, gyroscope, heartrate sensors, etc.
•Logged and later transferred to another processing/visualizing unit
*http://www.sonymobile.com/global-en/products/smartwear/smartwatch-3-swr50/
Department of Electrical & Computer Engineering
Role of GPS in Wearable
•Wearables begin to include GPS in order to directly sample location information
•Goal: Track distance traveled in real time, reconstruct traveled route for later viewing
•Useful for both calorie tracking and training
*http://www.sonymobile.com/global-en/products/smartwear/smartwatch-3-swr50/
Department of Electrical & Computer Engineering
Weaknesses of GPS
•Accurate distance measurement is difficult
•High power consumption when enabled
GPS
Display
CPU
Average Current Draw (mA)
0 50 100 150
System
Department of Electrical & Computer Engineering
Existing Applications
•Existing fitness apps request high sample rate when recording route
•Forces GPS unit into high power mode
•30 minutes of GPS = 20% of battery capacity
Department of Electrical & Computer Engineering
Research Question
•Can we provide target features at lower power by…• Using existing sensors instead of GPS to
accurately estimate distance traveled?
• Reconstructing the traveled route in a more intelligent manner?
Department of Electrical & Computer Engineering
Feature #1: Distance Traveled
•Smartwatch already records step count• Very mature technology
•Combine with stride length to get distance
Step Length
Stride Length
Department of Electrical & Computer Engineering
Feature #2: Route Traveled
•Can utilize map-matching algorithms (MMA’s)
•Modern MMA’s can function with very sparse data
Department of Electrical & Computer Engineering
How It Works
•Road network is edges and nodes
•Location sample compared to candidate points on underlying network
•True location is considered candidate point that most closely matches observed
Department of Electrical & Computer Engineering
Footfall Design
Temporal Component
Spatial Component
Await Step
Increment Steps Count
Evaluate Cadence
Estimate Stride Length
Update Distance Traveled
Movement Database
Apply MMA
GPS Scheduler
Evaluate Velocity
Determine Sampling Period
Identify Scheduling Method
Wait Sample Period
Sample GPS Location
Department of Electrical & Computer Engineering
Android GPS Management
•GPS power mode is managed by system
•Hardware supports 3 power modes: navigate, hibernate, and off
Power Mode Description
NavigatePowered on,
actively sampling
HibernatePowered on,
not actively sampling
Off Completely powered off
Department of Electrical & Computer Engineering
Periodic Samples, <10 Seconds
RequestLocation
LocationReturned
EnableSampling
Data StreamAcquired
RequestLocation
LocationReturned
Navigate
Hibernate
Department of Electrical & Computer Engineering
Periodic Samples, ≥10 Seconds
RequestLocation
LocationReturned
EnableSampling
Data StreamAcquired
RequestLocation
Hib
ern
ate
Wake
LocationReturned
Hib
ern
ate
Navigate
Hibernate
Department of Electrical & Computer Engineering
GPS Scheduler
•Existing scheduler is optimized for higher sampling rate
•Can update the scheduler to more efficiently utilize the time between samples
Sample Period Power Model
T<2 𝑮𝑵𝑨𝑽 ∗ 𝒕
2≤T≤13 𝒕
𝑻∗ 𝑮𝑵𝑨𝑽 ∗ 𝑻𝑻𝑨𝑨 + 𝑮𝑯𝑰𝑩 ∗ (𝑻 − 𝑻𝑻𝑨𝑨)
T>13𝒕
𝑻∗ 𝑮𝑵𝑨𝑽 ∗ 𝑻𝑻𝑭𝑭 + 𝑻𝑻𝑨𝑨
Department of Electrical & Computer Engineering
Periodic Samples, <2 Seconds
RequestLocation
LocationReturned
EnableSampling
Data StreamAcquired
RequestLocation
LocationReturned
Navigate
Hibernate
Department of Electrical & Computer Engineering
Periodic Samples, 2≤T≤13 Seconds
RequestLocation
LocationReturned
EnableSampling
Data StreamAcquired
RequestLocation
Hib
ern
ate
Wake
LocationReturned
Hib
ern
ate
Navigate
Hibernate
Department of Electrical & Computer Engineering
Periodic Samples, >13 Seconds
RequestLocation
LocationReturned
EnableSampling
Data StreamAcquired
RequestLocation
Disa
ble
En
ab
le
LocationReturned
Navigate
Hibernate
Data StreamAcquired
Department of Electrical & Computer Engineering
What Does This Allow?
• Intelligently scheduling GPS power allows significant reduction in power per sample
Original Footfall
{ }
{ , }
{ , }
600
Aver
age
Pow
er
per
Sam
ple
(mW)
400
200
0
Sampling Period (s)
0 20 40 600 20 40 60
Department of Electrical & Computer Engineering
Footfall Design
Temporal Component
Spatial Component
Await Step
Increment Steps Count
Evaluate Cadence
Estimate Stride Length
Update Distance Traveled
Movement Database
Apply MMA
GPS Scheduler
Evaluate Velocity
Determine Sampling Period
Identify Scheduling Method
Wait Sample Period
Sample GPS Location
Department of Electrical & Computer Engineering
Experimental Evaluation
•Captured information from routes traveled on local roads by 5 users
•Varied in length, elevation, and underlying road network complexity
Department of Electrical & Computer Engineering
Accuracy vs Existing
•Distance traveled estimation remained high
•Route reconstruction proved robust to low sample rates
1.00
0.75
0.50
0.25
0.000 240 5800 240 580
Acc
ura
cy (
−
)
Sampling Period, (s)
Route 1
Route 2
Route 3
Department of Electrical & Computer Engineering
Power Picture
200
100
00 540 1080 0 540 1080
Po
wer
Dra
w (mA)
Time (s) Time (s)
Department of Electrical & Computer Engineering
Reductions in Power
•Updated scheduler reduces power consumption by up to ~75%
•Obvious relationship demonstrated between power utilization and route accuracy
0.75
0.50
0.25
0.000
1.00
0.90
0.80
0.7030 60 0 100 200
Sampling Period, (s) Avg. Power per Sample (mW)
Po
wer
Red
uct
ion
Fac
tor
Rou
te A
ccura
cy (
−
)Route 1
Route 2
Route 3
Department of Electrical & Computer Engineering
Conclusions
•Existing wearable fitness application inefficiently utilize GPS capabilities
•GPS power overhead can be greatly reduced while still providing target features
•A updated GPS scheduler can provide significant power savings at low sample rate
Department of Electrical & Computer Engineering
Thank You!