Energy-Efficient Rate-Adaptive GPS-based Positioning for Smartphones

by Jeongyeup Paek, Joongheon Kim, and Ramesh Govindan

EECE354Kyoungho An

Contents

• Introduction• Problem• Rate-Adaptive Positioning System– Movement– Space-Time History– Celltower-RSS Blacklisting– Bluetooth-based Positioning Synchronization

• Discussion• Evaluation

Introduction

• Smartphone applications require position information– GPS is preferred over GSM/WiFi

• more accurate but power hungry• duty-cycle GPS

– Trades off positioning accuracy for lower energy

• RAPS (Rate-Adaptive Positioning System)– GPS is less accurate in urban areas

• Turn on GPS only necessary to achieve the accuracy

– Use location-time history of the uses to estimate velocity and uncertainty

– Use Bluetooth communication to reduce position uncertainty– Use celltower-RSS blacklisting to detect GPS unavailability

• Evaluation of RAPS– Increase phone life-times by more than a factor of 3.8 over an

approach where GPS is always on

Introduction

• GPS is accurate– WPS (WiFi-based positioning system) is less accurate than GPS– GSM-based positioning has an error as high as 300meters

• GPS is extremely power hungry– Nokia N95 smartphones

• Internal GPS uses around 0.37 Watt• 0.06 Watt idle power

• GPS activated would drain the 1200mAh battery on an N95 smartphones in less than 11 hours

Introduction

• The key insight that motivates the work– GPS can exhibit errors in the rage of 100m when

used in urban areas

• Location-based applications– deal with the errors using application-specific

methods• map-matching• map-snapping.

– If applications can tolerate the position error, why not trade off some position accuracy for reduced GPS energy usage?

• Periodic duty-cycle GPS– The key challenge is to decide on a time period

Problem

• GPS is not accurate in urban areas– GPS trace collected for a week using a smartphone that con-

tinuously logged GPS positions every 1 second.– ground truth path is labeled as A– phanthom path labeled as B– GPS was available for only 11.2% of the time– GPS may provide inaccurate positioning as high as 100 me-

ters or more across a wide range of urban locations

Problem

• Where do these errors come from?– A GPS receiver requires signals from 4 satellites– However, when GPS receiver do not have complete signals from all satellites,

then the errors can increase significantly– When it guesses a position, a GPS receiver provides accuracy estimates, but low con-

fidence in its position results. – Moreover, smaller antennae, carried in clothing or bags, indoors, frequent power-off

• Can we cleverly activate GPS only when necessary and sacrifice a little accuracy in exchange for reductions in energy usage by GPS?

Problem

• Simple solution is duty-cycle GPS– Periodic duty-cycled GPS (3 mins)

• Less than 40 meters for 28%• Often exceed 100 meters and go as high as 300 meters• Uncertainty depends upon the movement pattern of the user

– Interval increases enabling lower energy usage, the uncertainty increase linearly with no sweet spot

Problem

• Summary– GPS is generally less accurate in urban

areas– Periodic duty-cycling with a fixed inter-

val can introduce significant, potentially unbounded error without energy bene-fits

Rate-Adaptive Positioning System (RAPS)

• User movement detection using a duty-cycled accelerometer

• Velocity and uncertainty estimation using space-time history

• GPS unavailability detection using celltower-RSS blacklisting

• Position uncertainty reduction using Bluetooth

Movement Detection

• Modern smartphones are equipped with an accelerometer, and can be used to detect whether the user is moving or not

• Use of the accelerometers in RAPS– It is used to measure the activity ratio• Activity ratio: the fraction of a given time

window during which the user is in motion

– Duty-cycle to save energy

Movement Detection

• Onset detection technique– It is possible to detect whether a user is stationary or not– It does this by maintaining running estimates of the signal envelope and compares these against the

noise mean (1g)• Signal envelope: dynamic upper and lower bounds of the signal

• RAPS uses this onset detection method and calculates activity ratio• However, RAPS cannot use the accelerometer continuously because of significant energy

usage.– 30 seconds intervals: consumes 0.08 Watt which means that turning on the accelerometer for 5 min-

utes consumes more energy than activating GPS for 1 minute. – The method for deriving a good duty-cycling parameter is collecting continuous acceleration mea-

surements for five different human activities: stationary, frequent walking and stopping, fast walking, driving in a car, and milling about in a coffee shop

Movement Detection

• For each trace, calculate activity ratio using the onset detector• Performed offline analysis, and calculated the error in the activity

ratio respect to the reference always-on case. • Duty-cycle parameter of 12.5% with 2 and 14 seconds ON/OFF peri-

ods is selected

Movement Detection

• Could we have used a similar accelerometer duty cycling technique to esti-mate user speed (distance), rather than just activity?

• It is not that attractive– Calculation is roughly correct when the orientation of the phone does not change too

frequently and the acceleration is greater than the noise level during the period of the movement.

– A duty cycle of 50% or more is required to estimate the distance within 10% error

• RAPS uses only the activity if there is sufficient history anddistance estimation and the activity ration only if there is insufficient history

Space-Time History

• A key component of RAPS is the space-time table– Space-time table: records the past history of user movements

• Whenever RAPS needs to decide to active GPS– Looks up the history of average user velocity and activity ratio– Calculate the current position uncertainty based on the estimated current

velocity and the activity ratio

• RAPS updates and uses the space-time history as follows– For scaling reasons, it quantizes both the space and time dimensions– RAPS associates with each grid box in this quantized coordinate system two

quantities: a history of average velocity and activity ration

Space-Time History

• Example– If a user moves into a position A, the user is stationary for 4 units of time and

moves out of position A in the next unit of time at velocity of 10m/s and activ-ity ratio of 0.5.

– Velocity is {0, 0, 0, 0, 10}• Average 2m/s

– Activity ratio is {0, 0, 0, 0, 5}• Average 0.1

– If the user is stationary, • the current activity ratio R(t) = 0• estimated velocity V(t) = 0

– If the user is moving out of position A with R(t) close to 0.5, V(t) becomes 2*0.5/0.1 = 10m/s.

• This approach allows RAPS to cheaply estimate user movement, activat-ing GPS only when user movement may have exceeded the accuracy bound

Celltower-RSS Balcklisting

• GSM data can be retrieved without additional energy cost. • However, in practice, GSM data from all visible cell towers is not available to third

party application developers– Only one cell tower information is visible at a time

• Needs to examined whether information from a single cell-tower can reliably detect user movement

• Distance between two consecutive GPS locations when there was a change in cellID. • Maximum distance between two positions within the same cellID• 58.3% less than 10m difference in GPS positions

28.2% greater than 100m• These results imply that simply using the cellID itself provides insufficient information

about whether the user has moved a significant distance or not

Celltower-RSS Balcklisting

• Signal strength difference with a cellID can be an indicator of change in po-sition?– The average distance plot shows that increasing RSS difference correlates with

increasing distance, but the variance of the distance is too high

– A single cell-tower cannot reliably indentify user movement

Celltower-RSS Balcklisting

• Since we are merely interested in determining whether and when GPS should be acti-vated, instead of using cell towers to detect motion, we directly detect whether uses are in an environment or not. – GPS availability probability as a function of signal strength for two different cellId’s

• Maintaining a history of GPS availability per cellId can help accurately predict GPS is avaiable. – Whenever a GPS reading is successfully obtained or the request has time-out, it stores this in-

formation in a celltower-RSS blacklist. – Update RSS Good Thresh and RSS Bad Thread

• When a getting GPS update, checks the blacklist.– If it is in the good and variable region, turns on GPS. Otherwise, wait until there is a change in

GSM data or maximum timeout

Bluetooth-based Positioning Syn-chronization

• Use of Bluetooth to synchronize position information between neighboring devices

• Consider a scenario where there are two smar-phones (A and B)– If A has recently activated GPS, then– B can get the position information from A without activat-

ing GPS

• Bluetooth is a good for a positioning synchronization– Communication range is less than 10 meters

• Uncertainty is less than 10 meters

– Energy cost is less than GPS– Available on almost all mobile phones

Bluetooth-based Positioning Syn-chronization

• Using Bluetooth, most of the power is consumed on the master node• During one synchronization cycle depicted in the figure, the slave node used 0.09J,

and the master node used 3.07J, averaging 1.58J per node.• For comparison, if a GPS receiver was turned on and stayed on for 60 seconds, it

would spend around 0.37W * 60sec = 22.2J.• 43%reduction in energy usage• If there are 5 nodes,

– GPS: 5*22.2J = 111J– BPS: 22.2 + 3.07 + 4*0.9 = 28.87J– 74% reduction in energy

Bluetooth-based Positioning Syn-chronization

• BPS works as follows– Every node becomes slave node and stays in idle– A node decides(master) to transmit its position and un-

certainty information.– If any slave nodes exist, the master connects to all of

them, and broadcasts its position information. – Each slave compares its uncertainty and updates its own

position if the received position has lower uncertainty.– If the received uncertainty is higher than its own, the

slave replies to the master with better position and un-certainty values.

– The uncertainty value of all connected devices are syn-chronized.

Summarization

• Three details of RAPS1. The user space-time history and the celltower-

RSS blacklist must be populated for RAPS to work efficiently

2. Velocity estimation based on activity ratio can be misled by handset activity not related to hu-man motion

3. Accelerometers on smartphones may need a one-time per-device calibration of the offset and scaling before running RAPS

• Bluetooth-based position synchronization requires user co-operation.

Evaluation - Setting

Evaluation - Lifetime

Evaluation – BPS

Evaluation – RSS blacklist

Evaluation – GPS Interval

Evaluation – Average Power

Evaluation – Median Dis-tance

Evaluation – Median Dis-tance

Evaluation – Comparison with fixed duty-cycle

Evaluation – Comparison with WPS

Conclusion

• Presented RAPS, rate-adaptive positioning sys-tem for smartphones– GPS is generally less accurate in urban areas,

so it suffices to turn on GPS as often as necessary to achieve this accuracy

– Used a collection of techniques that could be im-plemented on current generation of smartphones to cleverly determine when to turn on GPS

• Evaluated RAPS through real-world experi-ments– Increases lifetime by more than a factor of 3.8 rel-

ative to the case when GPS is always on.