Improved NetArgus

A Suite of Wi-fi Positioning & SNMP Monitor

Tryfon Theodorou

, George Violettas

, Antigoni Polychroniadou

and Christos Georgiadis

University of Macedonia, 156, Egnatia Street, 54006, Thessaloniki, Greece

Royal Holloway University of London, Egham, U.K.

Keywords: Wi-fi Positioning, Kalman Filter, SNMP, FSPL, RTLS.

Abstract: In this paper, field research was conducted in order to enrich and strengthen the “NetArgus” application that

is able to monitor either a wired or a wireless network with all its possible aspects. Original algorithms were

implemented in an effort to find the position of a moving station utilizing the signal level of wireless local

area network (WLAN) devices. Exhausting sampling and research were done, in order to resolve various

practical problems associated with the signal level of the wireless stations as well as the involved hardware.

Various flaws and omissions of 802.11x protocols were discovered, such as the lack of definition of the

signal level and its width range. Moreover, a plethora of positioning practical problems has been adequately

managed. Hence, an application was constructed to draw paths and manipulate proper information from

various network moving clients, using the Simple Network Management Protocol (SNMP) as well as low

level operating system libraries. Last but not least, “Kalman Filter” was implemented and fully tested in

order to correct the positioning unavoidable mistakes.

1 INTRODUCTION

This paper is an extended-updated and improved

version of (Violettas et al., 2009). The aim of the

essay was to use the Wi-Fi available information for

finding the position of a station. SNMP protocol was

used for a central management and possible central

guidance of such devices. In addition, Kalman filter

(KF) is implemented in order to derive the correct

positioning estimates.

Practical problems of positioning were solved,

by creating the NetArgus application, which draws

information from network devices using various low

level operating system libraries like Microsoft WMI

and NDIS revealing several flaws and lacks. It was

proven that only NDIS library is functional.

The NetArgus application is open source, it

permits any kind of parameterization and it exploits

all the new technologies. This modular application

consists of 3 basic segments, providing the ability to

install every part of it to the most convenient part of

the network. The GUI user part is location

independent. This application manipulates all the

necessary information. Thus, an SQL database stores

in real time the SNMP available information

together with the signal level information sent by the

moving stations through the application created for

that purpose. A server application derives data from

the database and it calculates the position of the

moving stations in real time. It also stores and

manipulates all the provided SNMP information.

The client application pictures in a 2-axis graphical

interface the position of the network elements, in

real time too.

A theoretical barrier of the 802.11 signal level

strength measurement was discovered, due to

limitations of the implementation of this specific

variable as an integer (the gap between two

consecutive integer numbers in a logarithmic

function is resulting in a minimum error). For the

purposes of this updated version, extensive practical

research was done in order to add Kalman Filter

(KF) to the application. Hence, we describe the

theory of KF, its implementation and the extensive

testing of it.

2 RELATED WORK

The systems introduced to find the location of a

moving target are called Real Time Location System

(RTLS). Cisco (CISC Systems, 2006) has

289

Theodorou T., Violettas G., Polychroniadou A. and Georgiadis C..

Improved NetArgus - A Suite of Wi-ﬁ Positioning & SNMP Monitor.

DOI: 10.5220/0004077902890294

In Proceedings of the International Conference on Signal Processing and Multimedia Applications and Wireless Information Networks and Systems

(WINSYS-2012), pages 289-294

ISBN: 978-989-8565-25-9

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

implemented such a system addressed mainly to

closed spaces. It is a commercial (not open sourced)

system, cooperating with various hardware vendors,

which combines software and hardware (Wi-Fi

tags). Aeroscout is the first company introduced in

2003 an RFID tag compatible with Wi-Fi. As a

result, it is possible to find the position of the target

carrying this tag. The closest to this implementation

is the “Ekahau” system, a not open source

commercial system. It operates in open and close

spaces and it seems to dominate the market,

basically because of the accuracy, but also because it

is not using any hardware, being able to cooperate

with various systems and vendors.

Most of the papers investigated were vague

(probably on purpose, due to the future commercial

applications). A demanding job is described in

(Interlink Networks, 2002). It describes the

possibility of physically locating the position of a

wireless intruder into a corporate network. An

interesting job is also described in (Bahl et al.,

2000). They faced the same problems, such as the

extraction of the signal level info (without revealing

much), the statistical error, and the distance from the

broadcasting station. In (Seppänen et al., 2003) the

extraction of the signal level through SNMP is

described. It will be shown, that this is not always

possible. The essay in (Kotanen et al., 2003) has

some very interesting ideas, and it was the main

inspiration for the KF work. So far the authors are

not aware of such a work combining SNMP with

Wi-Fi positioning.

3 DISCRETE TIME OPTIMAL

FILTERS & PREDICTORS

The objective of such systems is to calculate “best”

estimates of a state vector sequence

in the

discrete interval

max

0 kk

. Those estimates are

based on all available (previous) information about

the system and measurement(s) taken during the

time interval. The time increment

1kk



is usually

constant, although not necessarily so.

The estimation problems fall into 3 categories:

1) If all the measurements are known prior to the

appliance of the filter, then a smoother can be applied

to the data.

2) If the time of calculation is synchronous with the

time of the computation, then the filter has to take

into consideration only previous data and is called a

filter, as it “filters out” the noise of the system.

3) If an estimate of the system’s position in future

time is needed, then the filter is called a predictor.

Usually the smother is producing better results due

to the usage of more data.

3.1 Kalman Filter

The purpose of a filter is to compute the state

estimate, while an optimal filter minimizes the

spread of the estimate- probability density in the

process. A recursive optimal filter will propagate the

conditional probability density function from one

sampling to the next, taking into account system

dynamics and inputs (Stengel,. 1994). Computing

the weighting factors (also called filter gains), which

optimally combine samples and future states, is the

crucial step of the computation and the only part that

needs extensive practical work.

“Kalman Filter opens the way to the maximum

likelihood estimation of the unknown parameters in

a model…”(Harvey, 1990). A KF may be applied

for prediction and smoothing on models put in state

space form. It is usually described by the following

linear difference equation (Kalman, 1960):

111kk kk

xAx Bu w





(1)

with a measurement

z 

that is

kkk

zHxv

(2)

The random variables

and

are process &

measurement error factors. They are independent,

white and with normal probability distributions, i.e.

we assume that

()~ (0, )pw N Q

and

()~ (0, )pv N R

. The

nn

matrix A in (1), relates the previous step

at k-1 to the current step k. The

1n

matrix B in (1)

relates the optional control input

u 

to the state

x. The

mn

matrix

in (2) relates the state to the

measurement

(2).

According to the above, applied KF needs only

the previous position (state) of the system to

calculate its current state. Combined with the

measured position (state), the system outputs the

current estimated position.

If the position of a moving station on a given

moment is

and this position is sampled every

time fraction t, then the position of the moving

station is described as follows:

1ktk tk

xTx Ra





(3)

Where α is the acceleration of the moving station (if

the speed of the moving station is constant, then

WINSYS 2012 - International Conference on Wireless Information Networks and Systems

290

obviously α=1) and











and















3.2 Kalman Filter in Practice

For the needs of the current paper, KF in its simple

form was used. The best practical explanation found

for KF, was the one in (G. Welch and G. Bishop,

1995). Eq. (1) and (2) are directly depicted from it.

In addition KF in practice needs to define the

following states:





to be the “a priori” state estimate at

state k given knowledge, and

x 

the “a

posteriori” state estimate at step k , given the

measurement

.Then the estimate errors can be

defined as follows:

kkk

exx





, as the a priory error, and its error

covariance



kkk

PEee





(4)

kkk

exx

, as the a posteriori error, and its error

covariance



kkk

PEee

(5)

Then, the final KF equation can be described as



ˆˆ ˆ

kk k k

xxKzHx



 

(6)

where

zHx



is the residual reflecting the

discrepancy between the predicted measurement



and the actual measurement

. If those two

are equal, then obviously the residual equals to zero,

or in other words there is no difference between the

predicted and the measured state of the system.

According to (Welch and Bishop, 1995). one form

of K that minimizes (5) is (7):



kk k

KPHHPHR







(7)

A practical representation of the two discrete

(“predict” & “update”) phases of the KF, are

described in Fig. 1. Notice that when initial

estimates

10k





and

10k





are known, they

can be inserted once, at the start of “predict” state.

Otherwise if the filter begins at the middle of the

process, x

k-1

and P

k-1

are set to an arbitrary high (or

respectively low) value, so the system will

immediately ignore them, taking into consideration

the first actual measurement(s).

As the filter was implemented,



in (1) was

ignored. Matrixes Q & R were first arbitrary set, and

there was a lot of discussion whether those values

can be set in general. Since more than one (different)

systems were not available, it remains an open

question. The wisest strategy would be to set those

matrixes after long observation, of each particular

system.

3.3 Kalman Filter Appliance Results &

Comparison Charts

For the implementation of the KF, there was a big

set of data available. This data was collected in an

open area system, roughly described as follows:

In an area of approximately 1,2Km

, there were

13 APs implemented, covering the area at a non

optimal way (there were “holes” – dead zones of

signal coverage in the area).

A moving station recorded its own movement

repeatedly in the area. Most of the research was

based on this set of data. Time sampling was

arbitrary set to 1 sec. This can be easily altered

depending on each particular system needs (speed of

movement is usually determining the time

sampling). The area was transferred to a digital map

of 600X600 pixels. Axes X,Y are mapped to each

dimension of the map, by a scale of 1 meter to 1

pixel (1:1).

As Fig.2-3 is demonstrating, the recorded

movement was far from being described as optimal.

There was a lot of distortion and noise. Most

important, there were several samples taken

indicating positions completely outside the area

drawn on the map. These are the result of the

weaknesses of the system described at (Violettas et

al., 2009) plus the high noise of the surrounding

area.

After the appliance of the KF as depicted in Fig.

2-3, an almost optimal movement is demonstrated,

in perfect accordance with the theory of KF.

Because KF is taking into consideration all the

data available to predict the position, if a really

distant point is inserted into KF it obviously affects

the prediction (although in a very small fraction due

to the nature of the filter). In order to avoid data that

is obviously useless, before the data is passed to the

KF there should be a simpler filter asking the

following question:

If a station is moving with a speed of 5m/s, is it

possible to alter its position by 10 meters from one

second to the next and then return to a normal speed

again? Obviously not, so this measurement should

be disqualified.

Improved NetArgus - A Suite of Wi-fi Positioning & SNMP Monitor

291

Because of this problem (i.e. when a much distanced

point suddenly “appears” inside the smooth-like

movement), an algorithm for disqualifying this kind

of data was implemented. More specifically, the

algorithm is examining each set of data checking if it

is different from the previous and the next set of

data, more than one threshold. If it is, then this set of

data is disqualified, and it is not taken into

consideration at all. The threshold of the algorithm is

still an open debate, as it highly relates to the nature

of the movement (e.g. the threshold should be much

smaller for a pedestrian walk than the one for a

moving car).

4 ARCHITECTHURE AND

IMPLEMENTATION OF THE

“NETARGUS” SYSTEM

The “NetArgus” application collects the available

SNMP information from the network devices in

order to construct the network structure and to

investigate its status. NetArgus is a 3-tier network

application that operates over TCP/IP networks. The

system uses Microsoft SQL Server 2005 for data

storage. The NetArgus System application is divided

into three separate applications, which communicate

using the common database and TCP/IP network

messages. In detail:

Figure 1: Kalman filter in action.

Figure 2: Movement trails along axes X, Y, before & after Kalman Filter.

100

200

300

400

500

600

13:17

13:18

13:19

13:20

13:21

13:22

13:23

13:24

13:25

13:26

13:27

13:28

13:29

13:30

Movement in betewwen [0.600] pixels

Time Fraction (Min)

Movement Trails Before Kalman Filter along X axis

100

200

300

400

500

600

13:17

13:18

13:19

13:20

13:21

13:22

13:23

13:24

13:25

13:26

13:27

13:28

13:29

13:30

Movement in betewwen [0.600] pixels

Time Fraction (Min)

Movement Trails Before Kalman Filter along Y axis

100

200

300

400

500

600

13:17

13:18

13:19

13:20

13:21

13:22

13:23

13:24

13:25

13:26

13:27

13:28

13:29

13:30

Movement in betewwen [0.600] pixels

Time Fraction (Min)

Movement Trails After Kalman Filter along X axis

100

200

300

400

500

600

13:17

13:18

13:19

13:20

13:21

13:22

13:23

13:24

13:25

13:26

13:27

13:28

13:29

13:30

Movement in betewwen [0.600] pixels

Time Fraction (Min)

Movement Trails After Kalman Filter along Y axis

WINSYS 2012 - International Conference on Wireless Information Networks and Systems

292

Figure 3 . Movement Trails along 2-D space maped, before & after Kalman Filter.

4.1 NetArgus Client, Server, Viewer

This is the part of the software that is installed in the

moving client’s PC. It sends in real time the signal

level of all the APs in its region thought the wireless

network card. The collected information is stored in

the database. NetArgus Client also collects

information about the AP’s Mac Address, SSID,

Security protocol, Network type, speed, etc. in order

to fill the database with as much information as it

can.

It should be noted, that the application can easily

carry SNMP information or any extra information

provided by another I/O device. E.g. in Fig. 13, the

NetArgus Client (at the upper right corner) is

transmitting to the NetArgus Server the details (id)

of a specific cargo named “Container CODE 1”,

inserted by an RFID reader, any other source or even

manually.

The application runs as a process in the

background and it can be started or stopped using a

tray icon. The signal level sampling time and the

database location can be configured by the user

through a windows form.

The second part of the application is the Server-

side application. It reads information from the

moving client (with the NetArgus application

installed) and by using the triangulation algorithm

described in (G.E. Violettas et al., 2009)., computes

the absolute position of the above client, storing that

information into the server database.

The server also does the entire SNMP related

job. It stores the SNMP collected information, in

addition to the collected information from the Wi-Fi

signaling.

This is the graphical representation module that

is used for presenting the network devices and the

network structure on a map. Moreover it shows in

real time the movement of the mobile clients in the

covered area (Fig. 4).

Figure 4: NetArgus Viewer, with Classroom map.

5 CONCLUSIONS

The aim of this paper was to create an application

for finding the position of a station. So, a three-tier

distributed application was created, following all the

modern design patterns. The positioning finding

algorithms are running in real time, finding the trail

of the moving station(s). However, due to the fact

that noise exists, the addition of the KF addressed

this problem with very accurate results. Thus, KF

corrects (optimizes) those trails. The NetArgus

Client is installed in every moving station we care of

finding its position. It continuously sends to the

database the received Signal level from all the APs

within its reach. The application was tested in

various harsh environments in real time. It did find

the position of the station with a relative accuracy.

Valuable results were gathered according to the

quality of the received signal, the distance between

station and the hardware used, specially the

100

200

300

400

500

600

0 100 200 300 400 500 600

Pixels in Y axis

Pixels in X axis

Movement Trails in 2-Dimension Space (X,Y) - Before Kalman filtering

100

200

300

400

500

600

0 100 200 300 400 500 6

Pixels in Y axis

Pixels in X axis

Movement Trails in 2-Dimension Space (X,Y) - Before Kalman filtering

Improved NetArgus - A Suite of Wi-fi Positioning & SNMP Monitor

293

differences between various vendors.

6 FUTURE WORK

Various additions can be easily made in SNMP, such

as exporting results of routing loads of a network

and the possible optimization through dynamic

routing protocols (RIP, BGP, IGMP), or possible

implementation of network watch applications e.g.,

clever water leakage sensors. As for 802.11

protocols, there is a wide area open. This is the

roaming protocol between the transmitting APs. The

802.11 protocol gives some general directions, but

every manufacturer implements its own methods

(most of the times without documentation) making

the possibility of roaming between devices of

different vendors, almost impossible. Moreover, the

positioning finding algorithms can be improved and

expanded specially on the direction of including

smart devices such as telephones, active tags etc.

Augmented Reality (AR) is another very

promising field of interest. For years, AR’s usage

has been restricted to high-end settings. Now,

though, AR is beginning to move into more

consumer settings, and the technology may be ready

to become more commonplace and commercially

successful. AR provides a digital enrichment of the

physical world. More specifically, it projects images

or information that augments what users see. For

example, when users view their surroundings via the

camera, an application displays information about

nearby shops, restaurants, and other landmarks.

According to “Computer” magazine (Vaughan-

Nichols, 2009) “location-based services (LBS)—

used with smart phones and other types of mobile

technology—are a major driving force behind AR's

entering the mainstream”.

The NetArgus can be firstly used for Finding the

Physical Position of a Transmitting 802.11 Station.

More specifically, with the use of any kind of vector

maps (e.g. Google Earth), or any kind of DEM maps

with Latitude & Longitude information, measuring

up the signal strength at three random spots (Fig. 5),

we can determine the co-ordinates of the

broadcasting station in question.

Figure 5: Triangulation of a known Target AP.

Secondly, NetArgus can be used for Mapping

and Tagging the Signal Level of all the Transmitting

Surrounding Stations in a given space (Fig.6), where

several stations are broadcasting. This can lead to an

optimization of the signal coverage through the

physical re-location of the APs towards an optimum

hexagon cell layout.

Figure 6: Empty space 802.11 signal coverage.

REFERENCES

G. E. Violettas, T. L. Theodorou, and C. K. Georgiadis.

(2009). “NetArgus: An SNMP Monitor & Wi-Fi

Positioning, 3-tier Application Suite”. La

Boca/Cannes, France: icwmc, pp. 346–351.

CISCO Systems. (2006) “Wi-Fi Based Real-Time

Location Tracking: Solutions and Technology”. C11-

346575-00, vol. White Paper.

AeroScout, “Technology Overview” Available: http://

www. aeroscout.com/content/technology.

Interlink Networks. (2002) Inc. “A Practical Approach to

Identifying and Tracking Unauthorized 802.11

Cardsand Access Points”.

P. Bahl, V. N. Padmanabhan, and A. Balachandran. .

(2000) “A software system for locating mobile users:

Design, evaluation, and lessons”. online document,

Microsoft Research, February.

A. Seppänen, J. Ikonen, and J. Porras. (2003) “Extracting

And Using Position Information In Wlan Networks”.

WAWC'03.

A. Kotanen, M. Hannikainen, H. Leppakoski, and T.D.

Hamalainen. (2003) “Positioning with IEEE 802.11 b

wireless LAN”. 14th IEEE Proceedings on Personal,

Indoor and Mobile Radio Communications, 2003.

PIMRC 2003, vol. 3.

R. F. Stengel. (1994) Optimal Control and Estimation.

New York, NY: Dover Publications.

A. C. Harvey. (1990) Forecasting, Structural Time Series

Models and the Kalman Filter. Cambridge: Cambridge

University Press.

R. E. Kalman. (1960) “A new approach to linear filtering

and prediction problems”. Journal of basic

Engineering, vol. 82, pp. 35–45.

G. Welch and G. Bishop. (1995) “An introduction to the

Kalman filter”. University of North Carolina at

Chapel Hill, Chapel Hill, NC.

P. A. Ruymgaart. (1988) Mathematics of Kalman-Bucy

Filtering, Berlin: Springer-Verlag.

Steven J. Vaughan-Nichols. (2009) “Augmented Reality:

No Longer a Novelty?” Computer, vol. 42, pp. 19-22.

WINSYS 2012 - International Conference on Wireless Information Networks and Systems

294