MEASUREMENTS OF BASE STATION TWO-BRANCH SPACE AND POLARIZATION DIVERSITY RECEPTION AND A COMPARISON OF THE DIVERSITY GAIN BASED

ON THE CDF OF SIGNAL LEVEL AND SIMULATIONS OF BER I A GSM SYSTEM N

Dieter Emmer*, Eduard Humburg", Peter Weber*, Martin Weckerle**

* Siemens AG, OEN MN ER 51 HofmannstraBe 51, D-8 1359 Munich, Germany

** University of Kaiserslautern, Research Group for RF Communications
P.O. Box 3049, D-67653 Kaiserslautern, Germany

Abstract This paper summarizes the results of a measurement campaign to evaluate the performance of different diversity schemes. The measurements have been carried out using the Siemens proprietary wideband channel sounder SIMOCS-2000. The key objective of the measurement campaign has been to evaluate the performance of polarization diversity as compared to space diversity. In addition, we contrasted the results for diversity gain based on signal statistics with results from link level simulations. The measurements have been carried out at 1815 MHz and GSM has been assumed for the standard dependent evaluations by means of link level simulations. Evaluating the measurements based on the well known and widely used cdf method according to [l], space diversity provides a maximum value of 6 dB diversity gain for a horizontal separation of 10 3L between two vertically polarized antennas. Using a k45" dual polarized antenna polarization diversity results in a diversity gain of 5.5 dB and performs as well as space diversity. The use of a horizontalkertical polarized antenna provides a diversity gain of 3 ... 4 dB. Comparing the figures for the diversity gain computed according to [l] with those obtained by GSM link level simulations we show that the cdf method results in diversity gain figures which are 2 ... 3 dB too optimistic. However, a modification of the method according to [l] which takes into account the effects of an interleaver in the transmission chain of the real communication system results in an almost perfect match. I. INTRODUCTION Multipath propagation causes fast fading in the mobile radio channel deteriorating the transmission quality. The combining of multiple receive branches mitigates this effect by reducing the variance of fast fading and by improving the mean signal-to-noise ratio. Diversity relies on having at least two independent receive paths. One way to obtain two independent fading signals is to use space diversity with sufficient separation between the antennas. Another way is the use of polarization diversity. Polarization diversity is very attractive since there is no need for spatial separation of the antennas. For mobile communications two different types of dual polarized antennas are employed, i.e. horizontalhertical (Wv) and -E- 45" slant dual polarized antennas.

In this paper, the performance of different diversity schemes is compared based on the diversity gain offered by these schemes. Usually, the diversity gain is determined by means of a system independent signal envelope statistics, called cdf method [ 11. For a certain mobile communication system, the resulting diversity gain can be determined exactly by means of link level simulations. In this paper, we contrast link level simulation results for the GSM system and diversity gains evaluated based on the cdf method. The paper is organized as follows. In section I1 the SIMOCS-2000 measurement system is described. Section I11 describes the measurement configuration and procedure. The data analysis is described in section IV for both evaluation techniques: signal statistics and data analysis by simulations. The description of the simulation tool is also part of section IV. The results are summarized in section V. 11. DESCRIPTION OF THE MEASUREMENT SYSTEM The propagation measurements were performed using the Siemens wideband channel sounder SIMOCS-2000. This measurement system consists of a transmitter unit mounted in a van and two tightly synchronized receiver stations for diversity reception at the base station site. SIMOCS-2000 measures a set of time variant complex impulse responses (IRs) of the mobile radio channel. This is achieved by transmitting a digitally generated periodic spread spectrum test signal and by using optimum unbiased estimation in the receivers [2]. The main technical data of SIMOCS-2000 are summarized in Tab. 1. Tab. 1 Main technical data of the Siemens wideband channel sounder SIMOCS-2000. frequency range bandwidth sequence Period frequency stability dynamic range measurement rate signal processing 700 MHz to 3 GHz 5 MHz, 10 MHz, 20 MHz 12.8 ps, 25.6 ps, 51.2 ps, 102.4 ps Rubidium-Standards in transmitter and receiver synchronized to GPS > 40 dB within impulse response > 100 dB overall selectable up to 1000 impulse responses per second on-line calculation of complex impulse
responses and monitoring

measurement modes time, route

0-78034320-4/98/$5.001998 IEEE 0

VTC '98

For more information the reader is requested to refer to http://www.siemens.de/oen/products/mn/simochso. html. In contrast to simple power level measurements, the measured complex channel impulse responses can directly be fed into link level simulations. Instead of using synthetic channel models, these complex impulse responses represent a realistic radio channel. SIMOCS-2000 data offers both postprocessing of signal envelopes and
0

simulations with measured channels.

According to Fig. 1, both methods for data evaluation may be applied on the same measured data from SIMOCS-2000.

is a k 45" slant dual polarized antenna. The decoupling of the two diversity branches exceeds 30 dB for both antennas. Each branch of these three different configurations of base station antennas can be seen as individual antennas having almost identical radiation patterns and gains. Their half-power beamwidths and gains are summarized in Tab 2. For the measurement runs two different orientations of the routes, radial and tangential with respect to the base site were chosen in such a way that they are crossing each other perpendicular at half of their run length. In addition, the performance of the different diversity schemes was assessed both in boresight and at the sector border. Boresight means that the routes are within the main radiation pattern of the antennas. For sector border measurements, the routes are located at the perimeter of the specified beamwidth of the antenna. The routes for boresight measurements are depicted in Fig. 2.

A. Description of the Measurement Locations

Measurements were carried out in a suburban, macrocellular environment on the outskirts of Munich. The base station antennas (Rx) were located on a rooftop, about 25 m above ground level and the surroundings consists of threestorey buildings. During the measurements the mobile (Tx) was driving along the two routes (radial and tangential) for each arrangement of receiver antennas. The average distance between mobile and base station was about 1350 m and there was no line of sight (NLOS) all along the routes.
/
,/

Fig. 1 Two ways of SIMOCS-2000 data evaluation: 1) by signal statistics and 2) by link level simulations The first method often used in publications [ l ] is independent of a special mobile radio system. The second method relies on a specific mobile radio system, e.g. GSM. 111. MEASUREMENT CONFIGURATION AND PROCEDURE The measurements were carried out at a carrier frequency of 1815 MHz using a bandwidth of 10 MHz. A resolution of 0.1 ps of the delay time is obtained within every single impulse response. The mobile transmitter unit was mounted in a van. On the rooftop of the van a vertically polarized antenna (transmit antenna) with an omnidirectional radiation pattern and a gain of 2.0 dBi was mounted at a height of about 2 m above ground level. For space diversity measurements, two vertically polarized antennas with different spatial separations of 5 h, 10 h, 15 h and 20 h were used at the base station site. Tab. 2 Gain and half-power beamwidth of the used base station antennas.
type of antenna
V

l~easurement

Fig. 2 Measurement area with boresight orientated antenna. (Dashed line denotes the 3 dB half-power beamwidth and the dotted line the main lobe direction of the receiver antennas.)

B. Measurement Technique
The length of the routes described in section A was about 300 m. The channel impulse responses were sampled every 5.12 ms. With a maximum vehicle speed of about 30 km/h, the spatial distance between two samples was less than 5 cm. Consequently, it is guaranteed that the chosen interval between two consecutive impulse responses in observation time fulfill the Nyquist criterion to get the full Doppler-information. Approximately 12000 samples per route were recorded for each of the two diversity branches simultaneously and stored in separate data files. According to Tab. 3 a total of 24 different sets of parameters were employed covering the different combinations of diversity type, direction of arrival and route orientation.

gain

half-power beamwidth

h/V

f45"

18 dBi 2 x 18 dBi 2 x 18 dBi

azimuth 65" 65" 65"

elevation
7,O"

6,5" 6,5"

For two-branch polarization diversity reception, two different types of commercially available dual polarized antennas were used. The first type has two branches which are horizontally and vertically polarized (WV) and the second one

0-7803-4320-4/98/$5.001998 IEEE 0

VTC '98

IV. DATA ANALYSIS OF THE MEASURED DATA

A. Data Analysis by Signal Statistics

The data analysis based on signal statistics has been performed as described in [ 11. In order to enable a comparison of the results with those from GSM link level simulations, the channel impulse responses (IRs) measured with a bandwidth of 10 MHz are filtered down to 234 kHz using an idealized filter with a rectangular transfer function. The signal power of the filtered IRs is calculated and the correlation properties of the two received signals are determined. For the computation of the cross-correlation coefficient, it is necessary to remove the variation of the local mean inherent in the recorded waveforms primarily because the local mean of both branches is expected to be highly correlated [ l , 31. This is accomplished by a moving average technique [4] with a window length of 20 A. If N is the number of IRs stored in one data file, the crosscorrelation coefficient p between the fast-fading components of the signals is calculated using the relationship

resultant signal level and finally determining the diversity gain in the same way as described above.

B. Data Analysis by Link Level Simulations

1) Description of the Link Level Simulation Tool The link level simulation tool has a modular structure shown in Fig. 3. The main components are transmitter chain, receiver chain and an universal channel model. The universal channel model consists of two branches for diversity simulations. The option SIMOCS-2000- stored channel offers the opportunity to perform simulations with measured SIMOCS-2000 channels. speech encoder

blockencT1

frame error counter

encoder

i=l

i=l

where T i l ,

ri2 represent the i-th samples of the powers of the normalized fast-fading signals for the two receivers and rl and T2 are their respective means [ 11. An overall mean signal

I +

m t
noise addition

combiner

level for each route and a cumulative probability distribution function (cdf) of the differences between the local means of the two signals is also calculated. Using diversity reception, the improvement in signal statistics depends on how the two received signals are combined. In the following, maximal-ratio combining is applied. Assuming equal noise levels on both branches, the resulting signal power with maximal-ratio combining is calculated by summing-up the signal power from the two diversity branches

PI.
I

The cdf of each diversity branch along with that of the combined signal is calculated. According to [ 11 the diversity gain is defined as the difference between the cdf curve of the combined signal and that of the stronger diversity branch at 10% probability. In modern communication systems like GSM a channel encoding employs some sort of an interIeaving scheme. Due to the use of an interleaving scheme the quality of the received signal depends on the quality of the interleaving frame. With an interleaving depth of 8 bursts specified in GSM a comparable diversity gain using the cdf method can be obtained by averaging 8 consecutive samples of Pl(t),P2(t),PR(t)prior to calculating the cdf of each

Fig. 3 Baseband transmitter and receiver chain of the simulation tool with universal channel model The link level simulations were carried out for the GSM system. The transmitter and receiver chains are built up according to [6,71. At the beginning of the simulations a vector with random bits represents the speech encoder output frame to be encoded by a block and a convolution encoder. After interleaving the bits are mapped to the GSM burst structure. This burst is modulated according to a GMSK modulation scheme. The modulator output is convolved with the measured and postprocessed IRs of the two branches and noise is added. The received signals are detected by a Viterbi equalizer including soft output combining. After demapping

0-7803-4320-4/98/$5.00 1998 IEEE 0

VTC 98

and deinterleaving the data is decoded. The various types of error rates which are defined in the GSM recommendations are determined by comparing the transmitted bits with the received ones at several stages in the chain. A frame is indicated bad, if an error in the 50 class la bits has occurred. It has to be mentioned that for the sake of comparability only general and not any vendor specific component or algorithm (e.g. for equalizer, bad frame indicator etc.) have been used through out the simulations. The channel sounder IRs h(.t,t) have been adapted to the GSM channel bandwidth. In GSM systems (e.g. GSM 900, DCS 1800), the system bandwidth is 270 kHz which corresponds to a bit period of 3.7 ps. For baseband simulations it is sufficient to use this bit period for sampling. Therefore, the IRs provided by SIMOCS-2000 are limited to the GSM system bandwidth. This is accomplished by Fourier transform (FFT) with respect to the 7-axis yielding H(f,t) with the same number of samples spaced over the frequency axis in intervals of about 40 kHz. Prior to performing the inverse transform (FFT') one can decide which physical channel frequency fo the transmission shall refer to. Only the spectral lines of the wanted frequency band [fo- 135 kHz ... fo + 135 kHz] are used for FFT' in the time domain producing the samples of the impulse response h'(7,t) at the chosen channel frequency with a resolution of 270 kHz. Within the measured bandwidth of 10 MHz 50 different GSM channels can be chosen. In order to investigate the influence of the frequency dependency on the diversity gain we chose the three different channels, c13, c30 and c45. The postprocessing of SIMOCS-2000 measurement data for use in link level simulations is shown in Fig. 4.
IRs from

total, we investigated a set of 24 different configurations according to the parameters: 1. diversity scheme: - space diversity [5 h, 10 h, 15 A,20 A] - polarization diversity [+ 45" slant, h/v] 2. route orientation [radial, tangential] 3. route location [boresight, cell border]. The following figures show results for each of the different 24 combinations. The assignment of the different configurations to the measurement numbers used in the figures is shown in Tab. 3. Tab. 3 Assignment of measurement number to the different measurement configurations route location route orientation tangential radial tangential radial tangential radial tangential radial tangential radial
I

2 3 4
5 6 7 8

h/V h/V h/V *450 *45" k45" *45 O 15hl 15hl

boresight boresight sector border sector border boresight

SIMOCS
10 MHz

BW = 270 kHz

postprocessed channel data

for simulation
200 W I Z

[c13, c30, c451

h(W
H Kt)

!;I 1
9
10

'

boresight sector border sector border boresight boresight sector border sectorborder boresight boresight sector border sectorborder boresight boresight sector border sector border boresight boresight sector border sector border

15x1

11

I
I

tangential radial tangential radial tangential radial tangential radial tangential radial tangential radial tangential radial

--I5Lr

n' (fs)

h (W) '

Fig. 4 Postprocessing of SIMOCS-2000 measurement data for use in link level simulations When antenna diversity is simulated this procedure has been run separately for each antenna branch.

16 17

I10hI IlOhI IlOhl I10hI I15 hl

1
I

I 18 I

19

I 15 h I 115hI

I I

I I

2 ) Data Analysis of the Link Level Simulations

Three simulation runs have been carried out with one fixed noise level: two runs with the single branch channels and one run with the combined channel. The error rates for each of the three cases are plotted against the noise at the receiver input. The quality criteria used for the simulations is 3% frame erasure rate as defined in the GSM specifications [SI. The diversity gain is defined as the difference in noise level between the best single branch and the combined branches. V. RESULTS In the following, the results of the measurements are presented and discussed. Four different configurations have been considered for each arrangement of antenna diversity. In

20 21 22 23
24

I 15 h I
120x1 I20h1 120hI I20hI

A Cross-Correlation and Mean Signal Level

The Figs. 5 and 6 show the cdf of the cross-correlation coefficients and the differences in signal level between the diversity branches. For each diversity scheme the four different configurations for one antenna arrangement are plotted into one cdf curve.

0-7803-4320-4/9B/$5.000 1998 JEEP

VTC '98

For space diversity, the cross-correlation is inversely proportional to the distance between the two antennas. As an example, if the antennas are separated by 10 h, the crosscorrelation coefficient is less than 0.7 in 90% of the cases. The values of the cross-correlation coefficient for the k 45" slant polarized antenna are comparable to those of space diversity having the branches separated by 10 ... 15 h. Wv polarization results in values for the cross-correlation coefficient which are even less than those for space diversity with an antenna separation of 20 h.
100

FERcl3: link level simulation result from a measured

channel with center frequency c13 and 3% FER used as quality criteria.
7 .

2 90
Y

5 % IO

m v ,

' 5 80

v)

10 12 14 16 18 20 22 24

number of measurement Fig. 7 Diversity gain based on cdf method, cdf-8 method and link level simulations of the FER Fig. 7 shows, that for space diversity with an antenna separation of 10 ... 20 h (measurement number 13 to 24) the diversity gain is almost constant, This means that 10 h of horizontal separation suffices for space diversity schemes at 1800 MHz. The performance of k 45" slant polarization (measurement number 5 to 8) also delivers results comparable to space diversity where the antennas are separated by 10 ... 20 h. This is due to the fact that the diversity branches for both f 45" polarization diversity and space diversity are de-correlated and in addition the branches do not differ in the mean received signal power. This is not true for h/v polarization (measurement number 1 to 4). Despite the fact that the diversity branches of h/v polarized antennas provide the least values for cross-correlation the branches significantly differ in the mean received signal power. As an consequence, the diversity gain provided by h/v polarization is approximately 2 dB less than for space diversity where the antennas are separated by 10 ... 20 h. Moreover, the Fig. 7 shows, that the cdf method according to [l] results in values for the diversity gain which are too optimistic compared eo the results from the link level simulations.
- et cdf - -13
-&

-1.0-0.8-0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0

cross-correlation Fig. 5 Cumulative probability of cross-correlation coefficients

100 90 80
70 60

50 40 30 20 10
n Y

-25 -20 -15

-10

-5

10

15

difference in signal level [dB] Fig. 6 Cumulative probability of difference in signal level There are no significant differences in the signal levels between the diversity branches of space diversity and f 45" slant polarization. With a probability of 90% the difference do not exceed +/- 3 dB. This is not true for the h/v-polarized antenna. The signal of the vertical branch is significantly stronger than that of the horizontal branch.

cdf-8 - FERcl3

B. Diversity Gain
Fig. 7 shows the results for the diversity gain evaluated according to three different methods:
0

number of measurement Fig. 8 Differences between the diversity gain based on link level simulations and the diversity gain based on the cdf method and cdf-8 method, respectively.

cdf:

method according to [ll;

cdf-8: modified method according to [ 11, where 8 consecutive samples are averaged ;

0-7803-4320-4/98/$5.00 0 1998 IEEE

VTC '98

If the cdf method is modified in a way that 8 consecutive samples are averaged (cdf-S), it results in a pretty good match with the figures from the simulation. This is also shown in Fig. 8, where the differences between the cdf and the cdf-8 method compared with the results from link level simulations. Finally, in Fig. 9 the results from link level simulations are shown for different physical center frequencies c13, c30 and c45, respectively.
--(f

on the same set of data gathered during various measurement runs. When contrasting link level simulations and the cdf method it is shown that the figures resulting from link level simulations are approximately 2 ... 3 dB lower than those from the cdf method. We modified the method presented in [ l ] in order to take into account that for mobile communication systems employing interleaver structures the quality of the transmission depends on the quality on a whole interleaving frame. With an interleaving depth of 8 bursts as for GSM a comparable diversity gain using the cdf method can be obtained by averaging 8 consecutive samples of the signal powers prior to calculating the respective cdfs. If interleaving is modeled there is a good match between the link level simulations and the signal statistics evaluation. The difference for the total set of parameters investigated is less than 1 dB.

FERcl3

-D-

Fwc30

. i

10 12 14 16 18 20 22 24

number of measurement Fig. 9 A comparison between the diversity gains of three different channels based on link level simulations. There is no dependency of the diversity gain on the physical frequency within the bandwidth under investigation. VI. CONCLUSION In this paper we presented the results of a channel sounder measurement campaign at 1815 MHz to compare the performance of different polarization diversity arrangements (h/v and f 45" polarization schemes) with that of space diversity with different spatial separations (5 ... 20 h) between the antennas. We evaluated the measured data and determined figures for the diversity gain. Two different methods were employed. Firstly, we determined the figures for the diversity gain based on the commonly used cdf method as described in [ 11. Our results show that 10 h of horizontal separation suffices for two-branch space diversity at 1800 MHz resulting in a diversity gain of 6 dB,
k 45" polarization provides a diversity gain of 5.5 dB

Finally, our results show that there is no dependency of the diversity gain due to the physical frequency within the bandwidth under investigation. ACKNOWLEDGMENTS The authors wish to thank the colleagues from Siemens AG, SI E SY A 4, for performing the measurements and Siemens AG, OEN MN ER, for fruitful discussions. Furthermore, the authors are grateful to Prof. Dr.-Ing. habil. P.W. Baier and Dipl. Phys. Ch. Math for their sustaining aid and their valuable contributions of this work. Finally, the authors wish to thank the staff of the computer center "Regionales Hochschulrechenzentrum Kaiserslautern (RHRK)" and acknowledge the usage of Silicon Graphics Compute Server SC900 and Cray Y-MP4E/464. REFERENCES A. M. D. Turkmani, A. A. Arowojolu, P.A. Jefford, C . J. Kellett ,,An experimental evaluation of the performance of two-branch space and polarization diversity schemes at 1800 MHz", IEEE Trans. Vehic. Technol., vol. 44, no. 2, May 1995, pp. 318 - 326. T. Felhauer, P.W. Baier, W. Konig and W. Mohr: Optimized Wideband System for Unbiased Mobile Radio Channel Sounding with Periodic Spread Spectrum Signals. Special Issue Spread Spectrum Techniques and Applications of IEICE Transactions on Communications (Japan), vol. E76-B, No. 8, August 1993, pp. 1016 1029. W. C . Y. Lee, Y. S. Yeh ,,Polarization diversity system for mobile communications", IEEE Trans. Commun., vol. COM-20, no. 5, pp. 921 - 923, Oct. 1972. W. C. Y. Lee ,,Mobile communications engineering", McGraw-Hill, New York 1982. D. G. Brennan, ,,Linear diversity combining techniques", Proc. IRE, vol. 47, June 1957, pp. 1075 - 1102. ETS 300 575:1995 (GSM 05.03 version 4.3.0) ETS 300 576: 1994 (GSM 05.04 version 4.0.3) ETS 300 577: 1996 (GSM 05.05 version 4.17.0)

which is comparable to that from space diversity and that h/v-polarization performance lags due to the large difference in the mean signal of the diversity branches resulting in 3 ... 4 dB diversity gain. Secondly, by means of link level simulations we determined the diversity gain achievable in GSM radio networks employing standard GSM quality criteria. Thanks to the unique features of the Siemens channel sounder SIMOCS2000 and the proprietary integrated simulation test bed we are able to process both, the cdf method and link level simulations

VTC '98
0-7803-4320-4/98/$5.000 1998 IEEE