A Location-determination Application in A Location-determination Application in WirelessHARTWirelessHART

Xiuming Zhu Xiuming Zhu11, Wei Dong, Wei Dong11,Aloysius K. Mok,Aloysius K. Mok11,Song ,Song HanHan11, Jianping Song, Jianping Song11, Deji Chen, Deji Chen22,Mark Nixon,Mark Nixon22

11University of Texas at AustinUniversity of Texas at Austin22Emerson Process Management Emerson Process Management

A Location-determination Application in A Location-determination Application in WirelessHARTWirelessHART

Xiuming Zhu Xiuming Zhu11, Wei Dong, Wei Dong11,Aloysius K. Mok,Aloysius K. Mok11,Song ,Song HanHan11, Jianping Song, Jianping Song11, Deji Chen, Deji Chen22,Mark Nixon,Mark Nixon22

11University of Texas at AustinUniversity of Texas at Austin22Emerson Process Management Emerson Process Management

OutlineOutline OutlineOutline

Location Awareness In WSN

WirelessHART

Localization in WirelessHART

Implementation

Experimental Results

Conclusion

Future Work

Summary

Location Awareness in WSNLocation Awareness in WSNLocation Awareness in WSNLocation Awareness in WSN Three kinds of distance indication information

– RSSI (Received Signal Strength Indicator)

It makes use of signal strength decay models to estimate the distance

– TDOA(Time Difference of Arrival)

It makes use of signal (usually sound) propagation speed

– AOA(Angle of Arrival)

It determines the direction with antenna array

Our solution is based on RSSI because we do not want to add extra devices to hardware

WirelessHART (I)WirelessHART (I)WirelessHART (I)WirelessHART (I) The first open wireless

standard for the process control industry

Four Types of devices

– Network Manager: control center

– Gateway: similar to AP

– Field Devices: Sensors

– Handheld Device (Badge): Carried by workers

A Typical WirelessHART Network

WirelessHART (II)WirelessHART (II)WirelessHART (II)WirelessHART (II)

WirelessHART Architecture

– Physical Layer : IEEE 802.15.4-compatible DSSS radios

– MAC Layer : A time synchronized and secure layer

– Network Layer: Supports mesh and star topology

– Application Layer: Command-oriented

WirelessHART (III)WirelessHART (III)WirelessHART (III)WirelessHART (III)

WirelessHART devices will be widely installed in thousands of factories. And it is valuable to know the locations of workers and assets because of the hazardous conditions in industrial environment, for example, a chemical leak, a tornado.

However, there is no location awareness support in WirelessHART

Localization in WirelessHART : QuestionLocalization in WirelessHART : QuestionLocalization in WirelessHART : QuestionLocalization in WirelessHART : Question

All field devices are attached to fixed locations

We need to locate the handheld device

The handheld device can sense the signal strength from neighboring field devices

Also, field devices can also sense the signal strength of the handheld device

However, two devices can not communicate with each other for security issues

Then, how to compute the location?

Localization in WirelessHART : SolutionLocalization in WirelessHART : SolutionLocalization in WirelessHART : SolutionLocalization in WirelessHART : Solution

Both field devices and the handheld device send neighbor health reports to the network manager periodically, which contain receive signal strength of their neighbors.

Thus, the network manager gets pairs of receive signal strength .e.g. field device 1 reports RSS of the handheld device is -50dbm and the handheld device report RSS of field device 1 is -51 dbm.

Then, the network manager can choose most trustable pairs of receive signal strength by comparison and compute the location of the handheld device by trilateration

Localization in WirelessHART: ExampleLocalization in WirelessHART: ExampleLocalization in WirelessHART: ExampleLocalization in WirelessHART: Example

Implementation : Hardware Platform Implementation : Hardware Platform Implementation : Hardware Platform Implementation : Hardware Platform Hardware : JM128 Board

– 48MHz 32-bit CodeFire V1 processor with a programmable 128KB flash and 16KB RAM

Three commands provided by WirelessHART

– Command 780 : neighbor receive signal strength report

– Command 787 : neighbor health report

– Command 797 : setting the transmitting power

A demo field device

Implementation : Propagation Model (I)Implementation : Propagation Model (I)Implementation : Propagation Model (I)Implementation : Propagation Model (I) Floor Attenuation Factor Model

n is the rate at which the path loss increases with distance, P(d0) is the signal power at certain reference distance d0 and d is the distance, nw is number of walls and WAF is wall attenuation factor

Except P(d0), every other parameter can be derived empirically. However, it is not realistic to get one value for all cases even within the same test scenario. Then,how to set these parameters?

WAF*nw- (d/d0) 10nlog-)[dBm] P(d0=P(d)[dBm]

Implementation : Propagation Model (II)Implementation : Propagation Model (II)Implementation : Propagation Model (II)Implementation : Propagation Model (II)

We formulate it as an optimization problem.

Before the localization, we can collect enough data about received signal strength and distance. These data can be used to train the model to get the best parameter tuple (n,nw,WAF). Later on, we use these parameters to estimate the distance.

Implementation: TrilaterationImplementation: TrilaterationImplementation: TrilaterationImplementation: Trilateration Trilateration

We have to calculate the point that minimizes

r0

p1 p2

p3

p0r1

r2

0p

n

iii rpp 2

0 )|(|

Implementation: Threshold (C)Implementation: Threshold (C)Implementation: Threshold (C)Implementation: Threshold (C) Choosing Threshold (C)

In our experiments, is 1.5 meters. That is, if the difference of two signal strength values will cause a distance error more than 1.5 meters, the pair will be discarded.

WAF*nwd)10nlog( C d

Experimental Results: Noisy, indoor(I)Experimental Results: Noisy, indoor(I)Experimental Results: Noisy, indoor(I)Experimental Results: Noisy, indoor(I)

Indoor, noisy with obstacle

A noisy office (Black dots are the possible locations for handheld devices)

Experimental Results: Noisy, indoor(II)Experimental Results: Noisy, indoor(II)Experimental Results: Noisy, indoor(II)Experimental Results: Noisy, indoor(II)

n nw WAF

2.91 0 0

Max Min Average Median

Error: (m) 11.72 0.39 4.56 3.96

It is a little surprising to see both nw ( number of walls)and WAF (wall attenuation factor) are zero, although there are obstacles. A possible explanation is that the fminsearch in matlab balances the effect of obstacle attenuation (WAF) with optimized attenuation factor (n) .

Figure4 Distance Error CDF(noisy)

Table 1

Table 2

Experimental Results: Quiet, indoor(I)Experimental Results: Quiet, indoor(I)Experimental Results: Quiet, indoor(I)Experimental Results: Quiet, indoor(I) Indoor, quiet, no obstacle

– A Discussion Area(4m*6m)

– Light-of-sight connection between devices

Experimental Results: Quiet, indoor(II)Experimental Results: Quiet, indoor(II)Experimental Results: Quiet, indoor(II)Experimental Results: Quiet, indoor(II)

n nw WAF

4.12 0 0

Max Min Average

Median

Error: (m) 2.49 0.26 1.52 1.67

Compared to the noisy case, it is much better . This can be attributed to the fact the training data and the test data are under the same uniform environment. Compared with the noisy case, there is much less uncertainties. Figure 5 Distance Error CDF(quiet)

Table 3

Table 4

Experimental Results: Quiet, outdoor(I)Experimental Results: Quiet, outdoor(I)Experimental Results: Quiet, outdoor(I)Experimental Results: Quiet, outdoor(I) Outdoor, no obstacle

– A Parking Area(10m*25m)

– Close to a busy road

Experimental Results: Quiet, outdoor(II)Experimental Results: Quiet, outdoor(II)Experimental Results: Quiet, outdoor(II)Experimental Results: Quiet, outdoor(II)

n nw WAF

3.58 0 0

Max Min Average Median

Error: (m) 11.87 1.00 5.03 3.16

Compared to the quiet indoor case, the distance errors are larger. This is because the training and testing for this experiment were done in different environments. However, the result is still promising. The median is below 4 meters, which is quite accurate for industrial use.

Figure 6 Distance Error CDF(outdoor)

Table 5

Table 6

ConclusionConclusionConclusionConclusion Our application is purely software-based and applicable

to all WirelessHART networks.

It is the responsibility of the network manager to compute the location of the handheld device because it can make use of all information in the network.

Accuracy is improved by the careful use of two techniques: comparison of two-way sensing distance and training parameters before localization.

Experimental results are very promising. All experimental results shows that median errors are blow 4 meters , which is quite enough for industrial use (10 meters)

Future WorkFuture WorkFuture WorkFuture Work Use fingerprint information : A thorough

investigation RSSI fingerprint can be done beforehand. Later on, a pattern-matching method will be employed to locate the device

Consider the historical location information and then, we can use it to estimate the present location

Finally, we hope we can deploy our application in real process control plants to validate our implementations

Thank you and Q&AThank you and Q&AThank you and Q&AThank you and Q&A