Вы находитесь на странице: 1из 5

Coherence in Signal Level Measurements between GSMQOO

and GSM1800 Bands and its application to Single BCCH


Operation
T.B. S@rensen,P.E. Mogensen, C. Posch’, N.H. Moldt+
Center for Personkommunikation (CPK), Aalborg University
Fredrik Bajers Vej 7A-5, DK-9220 Aalborg Ost, Denmark
e-mail: { tbs, pm} @cpk.auc.dk; { cpo, nhm} @dmt.sonofon.dk+

Abstract - In this paper we investigate the network capacity. The basic idea is described in
possibility to operate a dualband GSM network Section II, and a description of the pilot
having co-located GSM900 and GSM1800 cells, experimental study that we performed follows in
with a single broadcast channel (BCCH). Our Section III. After a presentation of the results of
investigations are based on simultaneous analysis in Section IV we end up with a discussion
dualband signal strength measurements using a in Section V and conclusion in VI.
test measurement system, neighbour channel
measurements from a dual band GSM phone, 11. SINGLE BAND BCCH OPERATION
and Abis trace data. Our results indicate that Mobile station signal strength measurements on the
base station antennas with dissimilar radiation BCCH beacon frequencies are used to assist initial
patterns for the two bands will effect a change in cell assignment and hand over between cells. The
mean signal level difference. The characteristics GSM specifications allow the MS (Mobile Station)
of the dualband antenna on the mobile station, to report signal strength measurements on the
especially when interacting with the hand and serving and the six strongest neighbouring cells (the
head of the user, tend to strongly de-correlate neighbouring cell’s BCCH beacon frequency). In
the mean signal level variations in the two the case of separate BCCH beacon frequencies each
bands. Both issues, and in particular the latter, pair of co-located dual band cells will likely
may negatively influence the performance of “occupy” two indexes in the neighbour channel list
single BCCH operation of co-located GSM900 (out of only six indexes). For this reason the
and GSM1800 cells. efficiency of the hand over mechanism, especially
when having a multi-layer network structure,
I. INTRODUCTION reduces significantly. The end result is a reduction
Digital cellular services (GSM1800) at 1800 MHz in capacity and quality.
are used extensively to complement existing GSM
(GSM900) cellular networks in city areas, where Single BCCH operation of co-located GSM900 and
the user density is high. The two networks GSM 1800 cells can be implemented by deriving the
inherently rely on individual system broadcast 1800 MHz signal strength from the 900 MHz
channels (BCCH). The BCCH is broadcast on a measurements (or vice versa). In this case, the MS
beacon frequency which is transmitted continuously can restrict its measurement reporting to RXLevgm
at maximum power. The BCCH channels require a (signal strength on GSM900 BCCH beacon
much larger frequency reuse distance than the frequency), and the radio network should be able to
traffic channel carriers, which can benefit from predict RXLev is00 given RXLevgoo This operating
power control, discontinuous transmission, and scheme relies heavily on coherence (i.e. strong
frequency hopping. The need to transmit a BCCH correlation) in mean signal level between the two
beacon for both bands in the case of co-located bands.
GSM900 and GSM 1800 cells implies a significant
reduction in the net network capacity. The basic operating principle of single BCCH is
illustrated in Figure 1. A hand over to GSM1800
In the following we investigate the possibility of can be made for the threshold setting in case B
single band BCCH operation to increase the because, given RXLevgm, we’have more than 90 %

’ Dansk Mobil Telefon (Sonofon), Denmark

O-7803-5435-4/99/$10.000 1999 IEEE 2243 VTC ‘99


confidence that RXLevlsW is above the required A pilot experiment has been conducted in order to
threshold level. validate these observations, and further to resolve
the apparent ambiguities. Initially, an urban area
: Signal strength cell was selected to be of particular relevance for
dualband operation.

III. DUALBANDEXPERIMENT
The experiment was conducted in one sector of an
existing tri-sectored base station (small urban
e- macro cell) in Aalborg, Denmark. The urban area is
characterised by 3 to 5 story apartment buildings
A _
with street width varying between 10 and 15 m.
The measurement area is similar to the area used in
B
[31.
10 % probability area
I
The base station uses two single band antennas
Figure 1 Single band BCCH operating principle. placed 4 m apart (horizontal spacing). The
Threshold setting A: hand over from GSM900 to GSM1800 antenna (18.0 dBi) is aligned vertically,
GSM1800 will not be attempted; Threshold setting B: whereas the GSM900 antenna (17.0 dBi) has a 5.5”
hand over takes place. down tilt relative to the vertical. The position of the
antennas is 35 m above median ground level.
Halfway in-between the two existing base station
The possibility to predict the 1800 MHz signal antennas we placed a dualband (wideband log-
level from 900 MHz measurements has already periodic) reference antenna (aligned vertically).
been exemplified in the COST231 Hata extensions
The radiation patterns for this antenna are almost
[ 11. The COST23 1 extension adds extra
identical for the two frequency bands with a
compensation (path loss) terms to the frequency horizontal beamwidth of 90” and a (wide) vertical
dependent term of the Hata model. Originally, the beamwidth of 65”. The two single band antennas
idea was that these additional path loss terms,
differ primarily in having a different vertical
together with the observed large scale signal beamwidth (GSM1800 6.5” and GSM900 9”) with
coherence between GSM900 and GSM1800
multiple sidelopes (-15 dB). Vertical (E-plane)
frequency bands, would allow the extensive
radiation patterns can be seen in Figure 2 with tilt
knowledge base of 900 MHz signal propagation to
of the GSM900 antenna included.
be extended to the 1800 MHz band. A summary of
the factors contributing to the path loss difference is
given in [2]. goO
120/A- / 5\ 60
According to results input to COST231 [3] the path
loss difference between the two bands was
measured to be within the range of 8.7 - 11 dB
(urban area) depending on base station height and
position. Further, the standard deviation was found
to be as low as 3.3 - 3.6 dB and the correlation
between slow fading signal variations in the two
bands was high (above 0.9). Slightly different 180 0
results were reported in [4]: 6.4 - 7.0 dB mean
difference with a standard deviation of 3.1 dB. Also
in this case the correlation was above 0.9. These
results suggest that prediction is feasible.
However, recent measurement results, obtained
using dual band GSM test mobiles, do not support
the observation of a fixed mean signal strength
2401 1 A300
difference and a small standard deviation; hence
270
these GSM network results are in contradiction to
the more ideal propagation measurements that Figure 2 E-plane antenna radiation patterns for BTS
supported the COST23 1 modelling. single band antennas.

O-7803-54354/99/$10.OOO1999IEEE 2244 VTC'99


In the experiment we used the two BCCH beacon The van drove a route of 13 km to cover most of
frequencies (1806.4 MHz and 958.2 MHz) plus two the streets within the half-power beamwidth of the
CW signals (1812.4 MHz and 959.0 MHz) which base station antennas. At the farthest distance the
were transmitted on the reference antenna. van was approximately 2 km from the base station.

For the data recording we used a four-frequency IV. COHERENCE ANALYSIS


measurement system mounted in a van. The system The signal level measurements provided by the
has two separate receiving branches for the 900 and measurement system were processed to determine
1800 MHz bands. Each of the two receiving the median level Psow over 12.7 m sections as an
branches is sequentially switched in frequency in estimate of the local mean signal level. The RXLev
order to measure on both the GSM BCCH beacon measurements, on the other hand, represent
frequency and the CW frequency. Separate band temporal averaging (in dB) over approximately 7
(end-fed) dipole antennas were placed on the roof samples (determined from the size of the BCCH
of the van to receive the four transmitted signals. Allocation list) and were used as is.
All signals were (log) envelope detected in a
bandwidth of 100 kHz and recorded at a constant 1 I , /
I
spatial sample rate of 10 samples per m.

I Power I

Figure 4 shows the empirical distribution of the


level difference between CW 900 and 1800 MHz
signals, transmitted from the dualband reference
antenna and received on the dipole antennas. The
Measurement
distribution is approximate log-normal with 72.5 %
System
and 94.5 % of the samples having a level difference
within fo and f2o ((3 is standard deviation),
Figure 3 The setup used in the van. respectively. For a normal distribution the
Alongside the measurement system, as indicated in respective values are 68.3 % and 95.4 %. The mean
Figure 3, we recorded the measurement reports of a level difference is +11.6 dB, and therefore
dual-band mobile station (DB-MS) at 1s intervals. comparable to the observation in [3]. For the single
From the frequency carrier numbers and the BSIC band antenna signals the level difference is only
(base station and network colour code) +6.9 dB.
identification, we were able to track synchronously
the signal level measurements (RXLev) on the two All the results for the mean and standard deviation
BCCH frequencies. have been summarised in Table 1. We note that the
In one set of measurements a passive power- standard deviation is comparable to the values
splitting network provided identical signals for the referenced in Section II. The double entries refer to
DB-MS and the test measurement system (Figure the different setups mentioned previously (one is
3), whereas in a second setup, the DB-MS used its shown in Figure 3) and have been obtained during
own whip-antenna. The DB-MS was placed in a different times of the day. Therefore, we attribute
fixture at a slight slant angle, just behind the no significance to the small deviations in the mean
windscreen. During both measurements a call level difference.
connection was established in order to trace the The data has been analysed with respect to the
neighbour channel measurement reports. radial distance from the base station, but we found
no significant dependence on distance. Also we

O-7803-5435-4/99/$10.000 1999 IEEE 2245 VTC ‘99


noted that the standard deviation of the (log- and therefore is sensitive not only to signal strength
normal) local mean variations was no different but also channel dispersion and co-channel
from one frequency band to the other. interference. Co-channel interference was most
dominant at 900 MHz.
When we exclude the erroneous measurements the
level difference is calculated to be +6.1 dB with the
external antenna signal and +I.7 dB when the DB-
MS uses its own antenna (Table 1). As before, we
observed that the mean difference is constant (no
dependence on distance), but the standard deviation
has increased to approximately 5 dB. This is in part
due to the different, and less accurate, measurement
procedure in the DB-MS evidenced by the increase
from 3.0 dB to 4.8 dB (standard deviation in Table
1) and, with less confidence, the influence from the
mobile antenna (4.8 dB to 5.1 dB).

To further characterise the coherence in signal


strength between the two frequency bands we
investigated the correlation properties for the Pso%
Table 1 Statistical results of analysis for the level measurements.
difference between GSM900 and GSMlSOO. Grey
shaded italic numbers were obained with the setup in
Figure 3, whereas the other results were obtained 0.9
with separate antennas.
The evaluation and comparison of the DB-MS data
is not as straightforward. Figure 5 shows a sample
plot of the RXLev difference from which it is clear
that the measurements fail occasionally (sample
points below -10 dB).

cc

Figure 6 Empirical Distribution Function (EDF) of


signal correlation.
Figure 6 shows the correlation between the slow
fading processes at 900 and 1800 MHz. The result
has been obtained by analysing the total 13 km
measurement route in sparse sampled segments of
1000 1200 14% 1600 1800
length 240 m (one sample every 16 m). This allows
Observation number us to obtain a 90 % confidence interval estimate
using the bootstrap procedure [.5]. The three curves
Figure 5 Sample plot of the level difference calculated
in Figure 6 should be considered individually and
from RXLev reporting (external antenna).
Observations are taken along the measurement route. not in comparison; the confidence limit
distributions serve only to illustrate the estimation
From a comparison with the PSO%measurements accuracy.
(BCCH beacon frequency signal strength If instead we consider the whole data set as a single
measurements) we concluded that failures are sample the correlation turns out to be in the range
caused by the RXLevgOo measurement reports. 0.87 - 0.89. Clearly, based on Figure 6 we may
Supposedly, this is due to the fact that the MS likelv exnerience a different local mean behaviour
requires frequency and time synchronisation for a between GSM900 and GSM 1800.
signal level measurement (it must derive the BSIC)

O-7803-5435-4/99/$10.00 0 1999 IEEE 2246 VTC ‘99


Finally, Table 1 contains a result derived from an VI. CONCLUSION
Abis trace on the same cell as we used for the In this paper, we have reported our investigations
measurements. We see that when the mobile phone on signal coherence between GSM9OO and
users are included along with a mixture of different GSM1800 frequency bands, which is of major
types of DB-MS the situation changes radically. importance for the operation of a single band
The mean level difference has actually changed in BCCH network.
favour of GSMl800, and the standard deviation This investigation shows that despite of quite
confirms the trend observed earlier between PSO% favourable propagation conditions for the
and RXLev derived measurements - a significant prediction of the mean level difference between
increase in standard deviation, and hence less GSM900 and 1800 bands, the influence of mobile
coherence. It must be emphasised that we cannot station antennas, specifically the interaction with
make firm conclusions based on this result, but the user, tends to de-correlate the signal variations.
increased variability is evident. This necessitates high hand over margins for the
cells in the band without BCCH. We therefore
V. DISCUSSION suggest that further investigations be done to
We conjecture that the observed discrepancy in evaluate the influence on network performance.
mean level difference for different base station Also, we point out that base station antennas have
antennas is due to differences in the effective some influence on the operation of single BCCH
antenna radiation patterns. There will be a small dual band cells.
influence from the antennas themselves (Figure 2)
amplified by the difference in downward tilt and VII. ACKNOWLEDGEMENTS
structures in close proximity to the antenna(s). The work has been co-sponsored by Nokia
This does not prevent signal level prediction for Telecommunications. Their financial support is
single BCCH operation. It merely requires that the very much appreciated.
radio network obtains a preliminary measurement
of the mean level difference so as to characterise VIII. REFERENCES
the cell (environment and BTS antenna ill COST telecommunications, Action 23 1, “Digital
configuration). We suspect that a dualband antenna mobile radio towards future generation systems”,
tends to equalise propagation conditions and Final report EUR 18957, ISBN 92-828-5416-7,
therefore will be our preferred choice. European Communities, 1999
PI T.-S, Chu, Larry J. Greenstein, “A Quantification
At the mobile end of the link the mobile phone user of Link Budget Differences Between the Cellular
seems to have a large influence, and most and PCS Bands”, IEEE Transactions on
importantly may possibly be the cause of non- Vehicular Technology, Vol. 48, No. 1, January
1999, pp. 60-65
predictable level differences between the two
[31 P.E. Mogensen, C. Jensen, J. Bach Andersen,
bands. This has not been studied in detail in this “1800 MHz mobile net planning based on
experiment, but we infer from other results that it is 900 MHz measurements”, COST231 TD(91)-08,
a likely cause. ln [6] it has been shown that the Firenze, 22-24 January, 199 1
local mean variation in received signal strength c41 L. Melin, M. RBnnlund, R. Angbratt, “Radio
caused by different users may vary 8 dB at the Wave Propagation, A Comparison Between 900
median outage level for the same mobile station. and 1800 MHz”, 43rd Vehicular Technology
The impact of these observations is that the hand Conference, Denver USA, 1993, pp. 250-252
over margin shown in Figure 1 needs to be set high [51 P. Hall, M. A. Martin, “Better Nonparametric
Bootstrap Confidence Intervals for the
in order to be confident that a hand over is safe.
Correlation Coefficient”, Journal of statistical
This will effectively introduce a gap in the
computation and simulation, Vol. 33, No. 16,
coverage area of GSMl800 (assuming RXLevgm 1989, pp. 161-172
reporting only). Eventually, when the margin G.F. Pedersen, J.O. Nielsen, K. Olesen, I.Z.
[cl
becomes very large we may jeopardise the potential Kovacs, “Antenna Diversity on a UMTS
gain that we initially expect from single band Handheld Phone”, To be published in the
BCCH operation. A simple calculation based on the proceedings of Personal Indoor and Mobile Radio
DB-MS (own antenna) RXLev statistics in Table 1 Communications, Osaka, Japan, September 12-
(assuming log-normal distribution) gives a margin 15, 1999
of 6.5 dB at a 90 % confidence level.

O-7803-5435-4/99/$10.00 0 1999 IEEE 2247 VTC ‘99

Вам также может понравиться