30
Energy Efficient GPS Acquisition with Sparse-GPS (1) Prasant Misra (2) Wen Hu (3) Yuzhe Jin (3) Jie Liu (4) Amanda Souza de Paula (5) Niklas Wirstrom (5) (6) Thiemo Voigt (1) Robert Bosch Centre for Cyber Physical Systems, Indian Institute of Science, Bangalore (2) CSIRO Computational Informatics, Brisbane, Australia (3) Microsoft Research, Redmond, USA (4) University of Sao Paulo, Sao Paulo, Brazil (5) SICS Swedish ICT, Stockholm, Sweden (6) Uppsala Universitet, Uppsala, Sweden IEEE/ACM IPSN 2014, Berlin 1 “Energy Efficient GPS Acquisition with Sparse-GPS” P. Misra, W. Hu, Y. Jin, J. Liu, A.S. de Paula, N. Wirstrom, and T. Voigt In Proceedings of the 13th ACM/IEEE Conference on Information Processing in Sensor Networks, IPSN '14 (IP Track) [ A.R.: 23/111 = 20% | CORE Ranking: A* ]

Energy Efficient GPS Acquisition with Sparse-GPS

Embed Size (px)

Citation preview

Energy Efficient GPS Acquisitionwith Sparse-GPS

(1) Prasant Misra(2) Wen Hu(3) Yuzhe Jin(3) Jie Liu(4) Amanda Souza de Paula(5) Niklas Wirstrom(5) (6) Thiemo Voigt

(1) Robert Bosch Centre for Cyber Physical Systems, Indian Institute of Science, Bangalore(2) CSIRO Computational Informatics, Brisbane, Australia(3) Microsoft Research, Redmond, USA(4) University of Sao Paulo, Sao Paulo, Brazil (5) SICS Swedish ICT, Stockholm, Sweden(6) Uppsala Universitet, Uppsala, Sweden

IEEE/ACM IPSN 2014, Berlin 1

“Energy Efficient GPS Acquisition with Sparse-GPS”P. Misra, W. Hu, Y. Jin, J. Liu, A.S. de Paula, N. Wirstrom, and T. VoigtIn Proceedings of the 13th ACM/IEEE Conference on Information Processing in Sensor Networks,IPSN '14 (IP Track)[ A.R.: 23/111 = 20% | CORE Ranking: A* ]

Motivating Application

IEEE/ACM IPSN 2014, Berlin 2

IEEE/ACM IPSN 2014, Berlin

Tracking & Monitoring Megabats: Flying Foxes

Motivating Application

Disease Vector

Agricultural pests

Behavior

Interaction

• Hendra Virus• Ebola• SARS-like Coronavirus

• Crop Damage

• Not well understood• Threatened species

• With other flying foxes and animals

3

IEEE/ACM IPSN 2014, Berlin

Monitoring PlatformPlatform details

Low power SoC- microcontroller core- IEEE 802.15.4 radio transceiver

Power Supply

Li-Ion Battery Li-Ion Charger Solar Panels

Multi-modal sensors

GPS

Flash

Wireless Sensor Node

R. Jurdak, B. Kusy, P. Sommer, N. Kottege, C. Crossman, A. McKeown, D. Westcott, "Camazotz: Multimodal Activity-based GPS Sampling," In IPSN 2013.

4

IEEE/ACM IPSN 2014, Berlin

Application Requirements and Challenges

Computing challenge• Operate within very tight energy budgets• Needs to use energy hungry GPS for tracking

Application requirements• Long-term, unsupervised operation• Delay tolerant

• Bats travel between different camps and other locations• Store sensed data locally in the flash; and upload the contents

at known camps

Base-station DB/Server

5

GPS is highly demanding in energy ?

IEEE/ACM IPSN 2014, Berlin 6

IEEE/ACM IPSN 2014, Berlin

GPS Overview-I: Infrastructure

• Constellation of 32 GPS satellites vehicles (SV), synced in time to the nanosecond level

• SVs, periodically, transmit signals containing time and trajectory parameters (Ephemeris)

7

IEEE/ACM IPSN 2014, Berlin

Unique sequence of 1023 bits transmitted at 1023 kbps, repeats every millisecond

50 bps

1.575 GHz

GPS Overview-II: GPS Signal

8

IEEE/ACM IPSN 2014, Berlin

GPS Receiver estimates its location by:• Computing time-of-flight (pseudo range) from each visible SV to itself• Combining with the respective SVs’ location at the time

GPS Overview-III: GPS Receiving Operation

Sense & Record

Acquisition Tracking DecodingLeast

Square

1 ms of data (4kB)Intensive computation

SV IDs Codephase (subMS) Doppler Shift

Continuous, every millisecond

Timestamp (MS) – 06 s Ephemeris – 30 s

TOF (Pseudo range) = MS + subMS Ephemeris

[Lat., Long.]

9

IEEE/ACM IPSN 2014, Berlin

Operational Impact

1. Standalone GPS receiver needs to be turned ON for up to 30 seconds in order to receive a completedata packet from the SV Difficult to duty-cycle

2. Requires significant amount of processing Needs a sophisticated CPU

Sense & Record

Acquisition Tracking DecodingLeast

Square

1 ms of data (4kB)Intensive computation

SV IDs Codephase (subMS) Doppler Shift

Continuous, every millisecond

Timestamp (MS) – 06 s Ephemeris – 30 s

[Lat., Long.]

10

Previous work:Delay tolerant + Constrained platforms (wireless sensor nodes)

IEEE/ACM IPSN 2014, Berlin 11

Sense & Record

Cloud-Offloaded GPS (CO-GPS)

Acquisition

Tracking

Decoding

Least-square

J. Liu, B. Priyantha, T. Hart, H. S. Ramos, A. A. F. Loureiro, and Q. Wang, “Energy efficient GPS sensing with cloud offloading,” in SenSys ’12.

IEEE/ACM IPSN 2014, Berlin 12

Scope for ImprovementThe task of offloading GPS data (raw samples) to the cloud (base-station)introduces an additional cost in energy.

Can this be limited ?

Sense & Record

Acquisition

Tracking

Decoding

Least-square

IEEE/ACM IPSN 2014, Berlin 13

Sparse-GPS (S-GPS)

S-GPS adopts the sample-and-process approach of CO-GPS, but reduces the energy cost of data offloading.

In our application, energy is a critical resource and its conservation is greatly valued.

Sense & Record Compressed data

Acquisition

Tracking

Decoding

Least-square

IEEE/ACM IPSN 2014, Berlin 14

Sparse-GPS (S-GPS)

IEEE/ACM IPSN 2014, Berlin 15

IEEE/ACM IPSN 2014, Berlin

Key Insight Leading to S-GPSConventional: GPS Acquisition by Cross-correlation

Key insight:The information content is sparse as the value of the code phase (i.e., time delay) and frequency bin (i.e., Doppler shift) corresponding to the correlation peak is only useful.

16

IEEE/ACM IPSN 2014, Berlin

S-GPS: Theoretical Foundation - IS-GPS is based on the theory of Sparse Approximation

Motivating insight of Sparse Approximation. One can accurately and efficiently recover the information of a highdimensional signal from only a small number of compressed measurements, when the signal-of-interest issufficiently sparse in a certain transform domain.

x s

sx

To recover s sxtsss

:..minarg:)(11

1

Dictionary

17

IEEE/ACM IPSN 2014, Berlin

S-GPS: Theoretical Foundation - IIDimensionality reduction with random ensembles ?

x s

x

y

ssxy )()(

211

1 :..minarg:)( ystsss

Dictionary

18

IEEE/ACM IPSN 2014, Berlin

Design of Dictionary - I

[-] Sparsely represents the code-phase, but not the frequency bin

[-] Breaks down the joint (codephase delay, Doppler) estimationproblem into smaller sub-problems global optimal cannot be guaranteed

Multi-channel Stacked

Properties:

1. Should sparsely represent both:(codephase delay, Doppler shift )

2. Should be incoherent with

Dictionary

19

IEEE/ACM IPSN 2014, Berlin

Design of Dictionary - IIMulti-channel Flattened

Single-channel Flattened

[+] Sparsely represents both code-phase and frequency bin

[+] Joint (Delay, Doppler) estimation per satellite

[-] Optimizer gets biased towards satellites with strongest signal[-] Requires a minimum of 32 times more memory

(easily scales into tens of gigabytes)

[+] Sparsely represents both code-phase and frequency bin

[+] Joint (Delay, Doppler) estimation over ALL satellites

[+] Good tradeoff : sparsity, performance and space complexity

20

IEEE/ACM IPSN 2014, Berlin

Search Algorithm: S-GPS

21

IEEE/ACM IPSN 2014, Berlin

Both the methods yield the same result for the position of the tallest peak(code phase = 515 chips and frequency bin index = 19)

Uncompressed 70% compressed (alpha=0.30)

Basic Performance Test

22

IEEE/ACM IPSN 2014, Berlin

Low peak sharpness (i.e., ratio ofthe first to the second tallestpeak), and high recovery noise

Incorrect recovery Low peak sharpness, but low recovery noise

S-GPS Acquisition Challenges with 2ms of dataAccepted practice is to use 2ms of data for GPS acquisition

Low SNR levels (18-25 dB), typical in GPS signals, severely impacts the recovery accuracy

23

IEEE/ACM IPSN 2014, Berlin

SNR Boost Stage:Accumulate K coefficient vectors (from the recovery process), and sum their intensities to derive the consolidatedestimation point. [each K corresponds to recovery result of 1ms of data]

Improving Acquisition Reliability

24

Evaluation and Results

IEEE/ACM IPSN 2014, Berlin 25

IEEE/ACM IPSN 2014, Berlin

Q) Does processing longer data lengths (such as 10/20 ms) promise better acquisition quality than the acceptedpractice of 2ms ?

A) There is a high probability of detecting > 1.5x more satellites by processing longer data segments.

Acquisition Quality

26

IEEE/ACM IPSN 2014, Berlin

For acquiring the same number of satellites astheir equivalent uncompressed data lengths,there is a 95% probability of success by using aslow as K = 10 for alpha = 0.30.

Median location error is less than 40 m.

Acquisition Quality and Location Accuracy

27

IEEE/ACM IPSN 2014, Berlin

2x

Energy Consumption

5x

Goodput reported for IEEE 802.15.4 complaint transceivers are between 42kbps and 93.6kbps

28

IEEE/ACM IPSN 2014, Berlin

Highlights and Way forward …

• We introduced Sparse-GPS, a new computing framework for energy efficient GPS acquisition via sparseapproximation• Proposed: A new dictionary that combines the information sparsity along all search dimensions

• Analyzed the dependency of the received SNR and satellite acquisition count on data length• Showed: Using 10-20 ms over 2ms of data, there is a high probability of acquiring 50% additional satellites

with both the conventional and S-GPS

• Demonstrated the GPS acquisition capability and energy gains by empirical evaluations on real GPS signals.• Showed: S-GPS is 2 times more energy efficient than offloading uncompressed data• Showed: S-GPS is 5-10 times more energy efficient than standalone GPS• Showed: S-GPS has a median positioning error of 40m

Highlights

Way forward …

• Develop mechanisms to enhance the received SNR (the most daunting task in realizing the S-GPS solution)• Explore hardware designs for combining [sense-compress] of the target platform• Optimize energy expenditure and execution time on the server side

29

Thanks !!!

IEEE/ACM IPSN 2014, Berlin 30