Combining of Redundant Signal Strength Readingsfor an Improved RF Localization in Multipath

Indoor EnvironmentsAndreas Fink, Helmut Beikirch

Department of Computer Science and Electrical Engineering,Rostock University,Rostock, Germany.

Email: andreas.fink@uni-rostock.de, helmut.beikirch@uni-rostock.de

Abstract—Multipath signal fading is a fundamental problemwhen using received signal strength (RSS) readings as sensor fora range-based indoor localization. Redundant RF transceiverswith a simultaneous use of spatial and frequency diversity enablea compensation of multipath fading effects for more precisedistance and position estimations. A novel maximum probabilitycombining technique for a sensor fusion of the RSS readings ofup to eight redundant and decorrelated RF channels is proposed.Experimental results on a one-dimensional motion test track areused to evaluate the improved localization performance comparedto a maximum signal combining and the accuracy of a positionestimation without redundancy.

Index Terms—Indoor Localization, Multimodal Diversity Com-bining, Multipath Fading Compensation, Object Tracking, Real-Time Positioning, Received Signal Strength.

I. INTRODUCTION

The latest research activities regarding indoor local position-ing systems (ILPS) for the tracking and navigation of peopleand objects show a variety of applied sensor technologies(cf. Fig. 1). The use of directional sensors like infra-red(IR), ultrasound or optical systems are limited to line-of-sight (LOS) scenarios. Systems with artificial quasi staticmagnetic fields are more robust in multipath environments,although the high power consumption of coil-based magneticfields limit the field of application. Due to the latest advancesin micro-electro-mechanical systems (MEMS) the use of anminiaturized inertial navigation system (INS) seems interestingfor an ILPS. The main issue of an INS is the limited long-termstability of MEMS-based localization systems.

Hybrid localization systems with a sensor fusion of differenttypes of sensors are able to overcome the specific drawbacksof a single system. The fusion of RF-based systems with anINS is a common solution which offers both, a good short-term accuracy and a good long-term stability. An overviewof the latest positioning systems with a comparison of theirspecific coordinate accuracy and coverage is given in [1]. Amore general taxonomy of localization systems can be foundin [2].

Range-based systems with an evaluation of the received sig-nal strength (RSS) of narrow-band RF signals are commonlyused for obstructed environments with many non-line-of-sight(NLOS) conditions, although they are known to be error-prone

in multipath environments.[3],[4] The characteristics of fadingchannels inside a factory building are discussed in [5] and [6].The multipath propagation due to reflection, diffraction andscattering leads to signal interferences at the receiver. Thus,the RSS is affected by strong variations and frequent signaldrop outs which make it challenging to find a relationshipbetween the RSS and the distance between transmitter andreceiver.

One solution to overcome multipath signal fading is theuse of an ultra-wide band (UWB) technique. The short signalpulses lead to a relative low sensitivity to multipath fadingeffects. Otherwise, UWB has a limited communication rangeand needs expensive transceiver hardware. A possibility tocompensate fading with the use of narrow-band RF signalsis the implementation of multiple-antenna techniques. Thespatial multiplexing of different antennas is primarily used toincrease the data rate of RF communication systems, like itis implemented in the IEEE 802.15.4n WLAN standard withMIMO channels and multiple frequency bands.

Diversity with a combining of redundant and decorrelatedRF channels can be used to compensate signal fading andenhance the link reliability [8]. Thus, multiple-antenna tech-niques are not only exploiting the capabilities of data commu-nication, but also can be used in indoor localization systemsto reduce the error of the RSS-based distance estimation andincrease the accuracy of the associated position estimation.

Time diversity with a repeated transmission often is notan acceptable solution – especially when a real-time local-

Fig. 1. Taxonomy of localization techniques according to the used sensor,the signal processing and the location estimator (cf. [7]).

1308

ization is required. With an a priori limited transmissiondelay the simplest possibility to get redundancy into theradio transmission is the so-called spatial diversity. Thereforethe RF signal is transmitted and / or received using twodifferent antennas which are placed in a certain distance toeach other. In order to compensate the effects of multipathfading and especially the influences of noise from externalinterferer, the same signal might also be transmitted usingtwo different carrier frequencies. Therefor, two channels ofthe same frequency band or even of different frequency bandscan be used. Our latest research investigations focus on low-cost redundant transceiver solutions with spatial and frequencydiversity and efficient signal fusion strategies for redundantRSS measurements to reach an improved position estimationin obstructed indoor environments. The application of severalantennas with different radiation patterns and polarizations isinteresting for the localization with angle of arrival (AOA)measurements and therefore is not considered in the following.

In section II, a survey of the related work regardingmultiple-antenna techniques for an improved RSS-based po-sition estimation is given. The used system model for theRF propagation in multipath indoor environments and theprinciples of diversity combining are presented in section III.The system architecture of the used RF diversity transceiverhardware is described in section IV. The proposed maximumprobability combining algorithm for up to eight diversitychannels is described in section V. Results from a path lossmeasurement are used to evaluate the combining process. Insection VI, the position estimations using different combiningtechniques are compared by experimental results of a dynamictracking measurement on a one-dimensional motion test track.In the last section VII, the results are discussed and investi-gated in terms of an outlook for further system developments.

II. RELATED WORK

A localization in wireless sensor networks with a combi-nation of temporal and frequency diversity is presented in[9]. With a switchover between two different channels in the2.4 GHz WLAN band the rms error of the position estimationin an office building is reduced by about 30 %. For the signalcombining the results for the maximum, average and minimumRSS are compared without showing significant differences. In[10] the RSS values of two consequtive RF transmissions intwo adjacent frequency bands (868 MHz, 915 MHz) using thesame antenna are averaged to detect outliers and reach a moreprecise localization. According to the Jakes model [11] theredundant channels are not completely decorrelated due to thelow frequency separation. Also the signal combining with anaveraging of the redundant RSS is not an optimum solutionfor multipath fading channels.

In [12] an indoor location estimation and tracking basedon a selection combining of a 433 MHz and 2.4 GHz RSSmeasurement is proposed. With the large frequency separationthe channels can be assumed to be decorrelated. The resultsshow better distance approximations for 2.4 GHz in LOSscenarios and for 433 MHz in NLOS environments.

For all related techniques an additional time diversity iseffective since there is a time lag between the transmissionswith different carrier frequencies. With the mingling of dif-ferent diversity techniques the redundant channels can notbe assigned as decorrelated. Thus, the results might only berelevant for a position estimation of stationary devices but notfor a localization and tracking of moving devices.

In [13] we propose a multimodal diversity transceiverplatform with the simultaneous use of spatial and frequencydiversity. A digital selection combining of up to four redundantRSS indicator (RSSI) values is used to compute the maximumRSS and achieve a high available and precise two-dimensionallocalization in a factory hall. A miniaturized platform withchip antennas and tracking results for a one-dimensional setupare proposed in [14] together with a detailed path loss analysisof eight redundant RF channels.

III. SYSTEM MODEL

A. Multipath Fading Channel

To find a relationship for the distance-dependent RSSand to evaluate the impact of diversity in multipath fadingenvironments, it is necessary to have a look at the indoorsignal propagation model for a single channel. Without anydisturbances (free space propagation) the distance-dependingpath loss shows a logarithmic dropping of power with a linearincreasing distance according to the log-distance path lossmodel. With (1) the average path loss PL(d) (in dB) overa distance d is given by the reference path loss PL(d0) over areference distance d0 and the environment-specific propagationcoefficient n.

PL(d) = PL(d0) + 10n log

(d

d0

). (1)

The value of PL(d0) is influenced by the effective radiatedpower (ERP) of the RF transmitter and the gain of the trans-mitting and receiving antenna. The value of n is influencedby the specific environmental propagation conditions and theused frequency. In [6] and [15] values for n between 1.8and 3.2 are given for frequencies between 900 MHz and4.0 GHz in obstructed indoor environments where not onlyNLOS conditions but also multipath signal fading affects theRF signal propagation.

The distance dependent path loss from (1) is an averagevalue and therefore not suitable to entirely describe a realchannel. For obstructed indoor environments the log-normalshadowing model (LNS) is used. A zero-mean Gaussianrandom variable Xσ with standard deviation σ is added tothe average path loss according to (2).

PL(d) = PL(d0) + 10n log

(d

d0

)+Xσ (2)

To investigate the path loss in fading environments we havecarried out an indoor measurement with proprietary 2.4 GHzRF modules. The measured RSS over a distance of 11 mis shown in Fig. 2 where the small-scale fading due to themultipath propagation is pointed out. The determined reference

1309

‐100

‐90

‐80

‐70

‐60

0 2 4 6 8 10 12

Received

 Signal Strength (RSS) in dB

m

Distance between TX and RX antenna in m

Smin

PL(do)

n = 2.62 / σ = 5.70 dB

2.4 GHz RF Path Loss (TX speed=0.27 m/s)

Fig. 2. Path loss for 2.4 GHz over a distance of 11 m (150 RSS samples,TX speed = 0.27 m/s, 5 Hz update rate).

path loss PL(d0) at a distance of d0 = 1 m between TX andRX antenna is −67 dBm for the used IEEE 802.15.4 compliant2.4 GHz RF transceiver with an output power of +10 dBm.The resulting path loss coefficient n = 2.62 is a typicalvalue for fading environments. The destructive interferencesof different multipath signals lead to abrupt signal dropoutswhere the signal sometimes even falls below the receiver’ssensitivity level Smin and the transmitted information gets lost.

In Fig. 2 the multipath fading becomes apparent by asignificant short-term variation of the RSS values in the casethe TX antenna is moved in the order of the wavelength (e.g.a few centimeters). The experimental path loss shows deepfades of about 25 dB, according to [15] even larger dropoutsby more than 40 dB have to be assumed. In general, a Rayleighdistribution of the RSS is assumed for NLOS conditions. Whenthere is also a LOS component reaching the receiver the fadingis modeled with a Rician distribution. The probability densityfunction (PDF) of the Rician distribution is given by

f(x, ν, θ) =x

θ2exp

(−(x2 + ν2

)2θ2

)I0

(xνθ2

), (3)

where θ is the amplitude of the superimposed RSS from allmultipath components and has a scaling functionality. Theparameter ν is the peak amplitude of the LOS component.For ν = 0, the Rician distribution reduces to a Rayleighdistribution. Thus, the Rician distribution is a generalization ofthe Rayleigh distribution. For ν →∞ the distribution reducesto a Gaussian distribution.

The ratio of the LOS component compared to the su-perimposed signal can be expressed by the Rician K-factoraccording to (4).

K(dB) = 10 logν2

2θ2dB (4)

The environment-specific values of ν and θ can be evaluatedwith an empirical analysis of multiple path loss observationsusing a two moment-based estimator as described in [16].

A typical variation of the RSS values for the path loss inmultipath environments with a dominating LOS componentis shown in the histogram in Fig. 3. The values are takenfrom the path loss measurement from Fig. 2 using proprietary2.4 GHz RF transceivers. The variation of the RSS around theaverage value is modeled with the shown Rician distributionusing K ≈ 17.7 dB which completely specifies the Riciandistribution. In Fig. 4 the distribution of the RSS values for a868 MHz path loss measurement and the corresponding Riciandistribution with K ≈ 20.0 dB are shown.

2.4 GHz

RSS Samples Rice

x1.21.110.9

f(x)

0.11

0.1

0.09

0.08

0.07

0.06

0.05

0.04

0.03

0.02

0.01

0

Measured RSS

Rician fit

K≈17.7 dB

ν≈0.98θ≈0.09

Fig. 3. RSS distribution for 2.4 GHz path loss measurement and correspond-ing Rician distribution.

868 MHz

RSS Samples Rice

x1.21.110.9

f(x)

0.14

0.13

0.12

0.11

0.1

0.09

0.08

0.07

0.06

0.05

0.04

0.03

0.02

0.01

0

Measured RSS

Rician fit

K≈20.0 dB

ν≈0.99θ≈0.07

Fig. 4. RSS distribution for 868 MHz path loss measurement and corre-sponding Rician distribution.

1310

B. Diversity Channel Correlation

As statistical model for the correlation coefficient envelopeρ of redundant channels in a diversity configuration we usethe Jakes model [11] according to

ρ =J20 (k0vτ)

1 + (2π)2S2τ (f2 − f1)

2 , (5)

where J0 is the zeroth order Bessel function, k0 is thewavenumber, v is the velocity, τ is the temporal separation, S2

τ

is the rms delay spread and f2−f1 is the frequency separation.For the used Jakes model some assumptions need to be made,including the validity of the wide sense stationary uncorrelatedscatterer (WSSUS) model and omnidirectional antennas withisotropic power distributions.

Any correlation between the fading of the channels reducesthe improvement of diversity. If ρ is below a certain thresholdthe channels are assumed to be ”effectively” decorrelated [17].This can be reached by large frequency separations or largespacings between the antennas. Since the place on a transceivermodule is limited and the Bessel function has certain minimumvalues, the spatial separation with the minimum optimal an-tenna spacing is of interest and will be discussed in detail inthe next section.

C. Optimum Antenna and Frequency Separation

The requirements of a small correlation coefficient betweentwo radio channels are a large frequency separation or a largespatial antenna separation (at RX and / or TX). A large antennaseparation might not be possible on a miniaturized hardwareplatform. Since J0 has more than one local minimum thehardware design goal should be a preferably small correla-tion coefficient with respect to the limited spatial hardwareexpansion.

For a moving communication participant the spatial diver-sity is mathematically equivalent to the temporal diversity.Assuming a stationary transmitter, the distance d which amobile receiver moves between the first transmission and itsrepetition is equal to the distance of two spatial separatedantennas at a stationary receiver. For spatial diversity with∆f = (f2−f1) = 0, the envelope of the correlation coefficientfrom (5) can be written as

ρ(d/λ) = J20

(2πd

λ

), (6)

where d is the spatial antenna separation. The envelope isshown in Fig. 5 for the correlation of frequency and spatialdiversity channels.

For frequency diversity the correlation coefficient decreaseswith larger frequency separations. Thus, a maximum of possi-ble separation should be used. The minimum optimal distancefor spatial diversity can be derived from (6). The first root ofρ(d/λ) at d/λ ≈ 0.38 gives the minimum optimal antennaspacing. For a frequency of 2440.2 MHz this value of d/λ isreached with d ≈ 4.67 cm.

0

0.2

0.4

0.6

0.8

1

0 1 2 3 4 5

Envelope

 correlation coefficient ρ(x)

x

x=2π∆fSτ, d=0

x=d/λ, ∆f=0

Fig. 5. Envelope of the correlation coefficient as a function of antennaand frequency separation (normalization to d/λ and 2π∆fSτ for spatialseparation and frequency separation, respectively). Source: [17]

IV. DIVERSITY TRANSCEIVER ARCHITECTURE

For our new signal combining approach we use our latestmultimodal RF platform (cf. Fig. 6) which consists of aMSP430 16-bit MCU and four proprietary RF transceivers,two operating at 868 MHz (T1, T3) and two operating at2.4 GHz (T2, T4). Both transceivers use the TI Chipcon RFstack (CC1101 for 868 MHz, CC2500 for 2.4 GHz) and areconnected via SPI to the MCU. The distance between thetwo antennas of a frequency band is d = 5 cm. Accordingto the Jakes fading model and Fig. 5 the distance is nearthe optimum for the 2.4 GHz band (fg = 2440.2 MHz,channel #36). For the 868 MHz band (fg = 868.3 MHz,channel #0) the theoretical minimum optimal antenna spacingis d ≈ 13.13 cm. This distance can not be realized witha miniaturized hardware platform, thus we accept a highercorrelation coefficient ρ(d/λ) ≈ 0.65 for the 868 MHz band.

5.5 cm

T4T1

T2T3

MACμC

MACμC

RXTX

T1

T2

T3

T4

T1

T2

T3

T4

t1t1

t2t2

τ=t1-t2=1ms

Fig. 6. Miniaturized diversity module (right) and schematic communicationstructure with eight diversity branches (left).

In the following, the progress of a redundant RSS samplingwill be described in detail. First of all, the transmitter MCUsplits up the signal into four packets which are sent to the RFtransceivers. In the first step, the RF packet is transmitted overT1 and T2 at t = t1 simultaneously. Of course, the other twoRF transceivers (T3 and T4) have two be hold in IDLE modefor the duration of the first transmission to prevent the platformfrom receiving its own RF packet. Immediately afterwardsthese two transceivers send out a packet at t = t2 and the

1311

remaining transceivers are hold in IDLE mode. The time gaprespectively temporal separation from (5) between the first andthe second transmission is τ = t2−t1 = 1 ms. For the dynamicbehavior of the mobile platform we assume a maximum speedof 2 m · s−1. Thus, there is a spatial misalignment of 2 mmbetween the two transmissions. For both frequencies, thisdistance is much lower than the corresponding wavelength andthus, the time difference can be neglected. The receiver usesthe same hardware as the transmitter. Since two RF packetsare received for each transmission, overall eight packets canbe analyzed for their RSS values.

V. ADVANCED RSS-BASED COMBINING

In addition to our previous selection combining algorithmwe use the information of all RSSI values of the samefrequency – especially their statistical distribution – to getanother improvement of the system’s accuracy. The combininglogic which runs on the MCU at the RX is a two-stageprocessing according to the principle from Fig. 7.

T1(868 MHz)

T3(868 MHz)

RSS Ch. ARSS Ch. B

RSS Ch. CRSS Ch. D

MPC(868 MHz)

T2(2.4 GHz)

T4(2.4 GHz)

RSS Ch. ARSS Ch. B

RSS Ch. CRSS Ch. D

MPC(2.4 GHz)

RSS’ 868 MHz

RSS’2.4 GHz

MPC RSS’’

Fig. 7. Two-stage combining using spatial and frequency diversity channels,separate maximum probability combining (MPC) for each frequency andmaximum signal combining (MSC) for the output fusion.

In the first step a separated processing of the two frequencybands is realized. The maximum probability combining (MPC)unit computes an RSS value with the maximum probabilityaccording to a Rician RSS distribution of the spatial diversitychannels with (3). For both frequencies, we use the RicianK-factor from the path loss measurement according to Fig.3 and Fig. 4 (K868MHz = 20.0 dB, K2.4GHz = 17.7 dB).The two frequency-separated branches are combined in thesecond step by an additional MPC stage. In order to evaluatethe performance of the MPC combining, we performed a pathloss measurement on a motion test track (cf. Fig. 9).

In Fig. 8, the RF path loss for the different diversitybranches of the MPC logic are compared with the path lossof two single channels of each frequency and the MPC outputusing all channels. The given values for the standard deviationare an indicator for the quality of an RSS-based distanceapproximation. Thus, the MPC approach (σ = 1.84 dB)outperforms the MSC technique (σ = 2.43 dB) and MPCwill result in an increased accuracy of the RSS-based positionestimation. Of course, MSC has a slighter path loss droppingthan MPC (n = 1.29 compared to n = 1.66) but this mightonly be important for an optimization of the SNR at thereceiver when the availability of the communication is the onlycriterion.

‐100

‐90

‐80

‐70

‐60

0 2 4 6 8 10 12

Received

 Signal Strength (RSS) in dB

m

Distance between TX and RX antenna in m

Smin

PL(do)

n = 1.96 / σ = 3.31 dB

868 MHz RF Path Loss (P_max channel A,B,C,D)

‐100

‐90

‐80

‐70

‐60

0 2 4 6 8 10 12

Received

 Signal Strength (RSS) in dB

m

Distance between TX and RX antenna in m

Smin

PL(do)

n = 2.31 / σ = 5.57 dB

868 MHz RF Path Loss (channel A: T1‐T1)

‐100

‐90

‐80

‐70

‐60

0 2 4 6 8 10 12

Received

 Signal Strength (RSS) in dB

m

Distance between TX and RX antenna in m

Smin

PL(do)

n = 1.66 / σ = 1.84 dB

RF Path Loss (P_max 868 MHz + 2.4 GHz)

‐100

‐90

‐80

‐70

‐60

0 2 4 6 8 10 12

Received

 Signal Strength (RSS) in dB

m

Distance between TX and RX antenna in m

PL(do)

n = 1.72 / σ = 2.84 dB

Smin

PL(do)

2.4 GHz RF Path Loss (P_max channel A,B,C,D)

‐100

‐90

‐80

‐70

‐60

0 2 4 6 8 10 12

Received

 Signal Strength (RSS) in dB

m

Distance between TX and RX antenna in m

Smin

PL(do)

n = 1.29 / σ = 2.43 dB

RF Path Loss (Max 868 MHz + 2.4 GHz)

‐100

‐90

‐80

‐70

‐60

0 2 4 6 8 10 12

Received

 Signal Strength (RSS) in dB

m

Distance between TX and RX antenna in m

Smin

PL(do)

n = 2.62 / σ = 5.70 dB

2.4 GHz RF Path Loss (channel A: T2‐T2)

n=1.96 / σ=3.31 dB

b) 868 MHz (MPC ch. A,B,C,D)

-100

-90

-80

-70

-60

0 2 4 6 8 10 12TX-RX distance in m

PL(d0)

Smin

n=1.72 / σ=2.84 dB

d) 2.4 GHz (MPC ch. A,B,C,D)

-100

-90

-80

-70

-60

0 2 4 6 8 10 12TX-RX distance in m

PL(d0)

Smin

n=1.66 / σ=1.84 dB

f) MPC 868 MHz + 2.4 GHz

-100

-90

-80

-70

-60

0 2 4 6 8 10 12TX-RX distance in m

PL(d0)

Smin

n=1.29 / σ=2.43 dB

e) MSC 868 MHz + 2.4 GHz

rece

ived

sig

nal s

treng

th (R

SS

) in

dBm

-100

-90

-80

-70

-60

0 2 4 6 8 10 12TX-RX distance in m

PL(d0)

Smin

n=2.62 / σ=5.70 dB

c) 2.4 GHz (ch. A: T2-T2)

rece

ived

sig

nal s

treng

th (R

SS) i

n dB

m

-100

-90

-80

-70

-60

0 2 4 6 8 10 12TX-RX distance in m

PL(d0)

Smin

n=2.31 / σ=5.57 dB

a) 868 MHz (ch. A: T1-T1)

rece

ived

sig

nal s

treng

th (R

SS

) in

dBm

-100

-90

-80

-70

-60

0 2 4 6 8 10 12TX-RX distance in m

PL(d0)

Smin

Fig. 8. RF path loss with and without diversity for maximum selectioncombining (MSC) and maximum probability combining (MPC).

VI. LOCATION ESTIMATION RESULTS

The test bed for the tracking measurements is shown inFig. 9 on the left. The mobile blind node (BN) which shouldbe located performs periodic movements on a motion testtrack according to the motion profile shown on the right. The

 

‐0.4

‐0.2

0

0.2

0.4

0.6

0 25 50

v(t) in m/s

t in s

0

3

6

9

12

0 25 50

s(t) in m

t in s

Fig. 9. Measurement setup on a motion test track in an obstructed test hall(motion profiles show one A-B-A motion cycle, T = 65 s).

1312

duration of one movement from position A to B and back toA is 65 s. Seven reference nodes (RNs) are evenly distributednext to the track with a distance of 2.0 m between two nodes.For an explicit multipath propagation we installed metallicreflecting walls next to the track. Since there is always a LOSbetween the nodes, a Rician fading channel is assumed.

In Fig. 10 the trajectories of the BN on the motion test trackfor a complete motion cycle (A-B-A) are shown. Like we haveproceeded at the path loss measurement in section V, differentconfigurations are compared to point out the influence of thediversity and combining on the localization accuracy.

0

3

6

9

12

1 51 101 151 201 251

x in m

Iteration

channel A

MPC(A,B,C,D)

868MHz

0

3

6

9

12

1 51 101 151 201 251

x in m

Iteration

channel A

MPC(A,B,C,D)

2.4 GHz

0

3

6

9

12

1 51 101 151 201 251

x in m

Iteration

MSC

MPC

868 MHz + 2.4 GHz

a) 868 MHz

b) 2.4 GHz

c) 868 Mhz + 2.4 Ghz

Fig. 10. Estimated tracks for 868 MHz and 2.4 GHz measurements on amotion test track, qualitative comparison of different diversity and combiningtechniques.

A detailed comparison of the location estimation error(LEE) of different diversity and combining techniques is givenusing Table I and the cumulative distribution functions (CDFs)in Fig. 11. The MPC of the four spatial diversity channelsreduces the error for both, the 868 MHz and the 2.4 GHzband. The improvement is slightly higher for the 2.4 GHz bandwhich indicates a lower correlation coefficient according to (6).Looking at the CDFs, the performance of the MPC comparedto the MSC algorithm gets emphasized. The MPC reduces themaximum error for the combining of all eight RSS values bymore than 34 %. Even the MPC of four 2.4 GHz channelsenables a more accurate position estimation than the MSC ofall eight channels.

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0

prob

ability

estimation error in m

868_A

2.4_A

868_MPC

2.4_MPC

MSC

MPC

Fig. 11. Cumulative distribution functions for the location estimationerror (LEE) of a 9.6 m tracking measurement using different diversity andcombining techniques.

TABLE IPERFORMANCE COMPARISON OF DIFFERENT DIVERSITY AND COMBINING

TECHNIQUES (LEE - LOCATION ESTIMATION ERROR IN METERS)

868 A 868 MPC 2.4 A 2.4 MPC MSC MPC

LEEmed 0.74 0.34 0.53 0.22 0.29 0.21σLEE 0.67 0.41 0.48 0.32 0.34 0.23LEE95% 1.96 1.27 1.49 1.02 1.13 0.70LEEmax 4.54 2.25 2.78 1.56 1.70 1.12

VII. CONCLUSION AND FUTURE WORK

The tracking results show that RSS-based localization holdsa lot of promise when redundant signal strength readingsare used. Especially the path loss measurement indicates thegood conformity of the MPC combining approach with thetheoretical Rician RSS distribution in indoor LOS environ-ments with multipath fading. The simultaneous use of spatialand frequency diversity channels shows a low inter-channelcorrelation and a good fading compensation performance.Although not shown, the use of multiple frequency bands leadsto a more reliable localization in the presence of externalinterferer like WLAN, Bluetooth or other widespread RFstandards.

For a more sophisticated evaluation of the tracking systemwe are going to introduce a further measurement with thecontrolled inclusion of external interferences to point outthe influence of the frequency diversity. The tracking systemshould also be tested in more obstructed multipath envi-ronments to confirm the performance of the proposed MPCtechnique for an ILPS under real-life conditions.

REFERENCES

[1] T. Kohoutek, R. Mautz, and A. Donaubauer, “Real-time indoor position-ing using range imaging sensors,” in Proceedings of SPIE Photonics, vol.7724, no. 20, 2010.

[2] J. Hightower and G. Boriello, “A survey and taxonomy of location sys-tems for ubiquitous computing,” University of Washington, Departmentof Computer Science and Engineering, Tech. Rep., 2001.

[3] E. Elnahrawy, X. Li, and R. Martin, “The limits of localization usingsignal strength: A comparative study,” in IEEE Sensor and Ad HocCommunications and Networks (SECON), 2004, pp. 406 – 414.

1313

[4] K. Whitehouse, C. Karlof, and D. Culler, “A practical evaluation of radiosignal strength for ranging-based localization,” SIGMOBILE MobileComputing and Communications Review, vol. 11, pp. 41–52, January2007.

[5] T. S. Rappaport, “Characterization of UHF multipath radio channels infactory buildings,” IEEE Transactions on Antennas and Propagation,vol. 37, no. 8, pp. 1058–1069, Aug. 1989.

[6] T. S. Rappaport and C. D. McGillem, “UHF fading in factories,” IEEEJournal on Selected Areas in Communications, vol. 7, no. 1, pp. 40–48,Jan. 1989.

[7] I. Guvenc, “Enhancements to RSS based indoor tracking systems usingkalman filters,” in GSPx & International Signal Processing Conference,2003.

[8] J. Figueiras and S. Frattasi, Mobile Positioning and Tracking - FromConventional to Cooperative Techniques, 1st, Ed. John Wiley & Sons,2010.

[9] P. Kulakowski, F. Royo-Sanchez, R. Galindo-Moreno, and L. Orozco-Barbosa, “Can indoor RSS localization with 802.15.4 sensors be vi-able?” in 2nd COST IC1004 Management Committee Meeting, Lisbon,Portugal, 2011.

[10] F. Reichenbach and D. Timmermann, “Indoor localization with low com-plexity in wireless sensor networks,” in IEEE International Conferenceon Industrial Informatics, aug. 2006, pp. 1018 –1023.

[11] W. Jakes and D. Cox, Microwave Mobile Communications. Wiley-IEEEPress, 1994.

[12] T. E. Abrudan, L. M. Paula, J. Barros, J. P. S. Cunha, and N. B. Carvalho,“Indoor location estimation and tracking in wireless sensor networksusing a dual frequency approach,” in IEEE International Conference onIndoor Positioning and Indoor Navigation (IPIN), 2011.

[13] A. Fink and H. Beikirch, “RSSI-based indoor localization using antennadiversity and plausibility filter,” in 6th Workshop on Positioning Navi-gation and Communication. Hannoversche Beitraege zur Nachrichten-technik, HBN 0.5, 2009, pp. 159–165.

[14] ——, “Fading compensation for improved indoor RF tracking based onredundant RSS readings,” in Autonomous and Intelligent Systems, ser.Lecture Notes in Computer Science, M. Kamel, F. Karray, W. Gueaieb,and A. Khamis, Eds. Springer Berlin / Heidelberg, 2011, vol. 6752,pp. 384–395.

[15] T. S. Rappaport, Wireless Communications – Principles and Practice.Prentice Hall PTR, 2002.

[16] A. Abdi, C. Tepedelenlioglu, S. Member, S. Member, M. Kaveh, andG. Giannakis, “On the estimation of the K parameter for the Rice fadingdistribution,” IEEE Commun. Lett, vol. 5, pp. 92–94, 2001.

[17] A. Molisch, Wireless Communications. John Wiley & Sons, 2005.

1314