You are on page 1of 5

Load Balancing Effect of Inter-Frequency Handover

with Pilot Power Tuning in UTRAN

Tsuyoshi KASHIMA, Houtao ZHU


Nokia Research Center
Tokyo, Japan
Tsuyoshi.Kashima@nokia.com, Houtao.Zhu@nokia.com

Abstract—Load balancing is an important technique in radio in soft handover (SHO) case and in IFHO case. In SHO, a
resource managements (RRM). In future, when several paper [6] concludes that CPICH EcNo achieves better
frequencies are required to cover increasing number of users and performance in outage probability and call blocking probability.
increasing amount of traffic, inter-frequency handover (IFHO) A paper [7] studies IFHO in hierarchical-cell-structured-
will perform an important role in load balancing between WCDMA network and concludes that RSCP outperforms
frequencies. Meanwhile, there is usually geographical load EcNo in call dropping probability and blocking probability.
imbalance in real network and changing coverage by tuning
common pilot channel (CPICH) power can mitigate this
geographical load imbalance. In this context, we evaluated the
B. CPICH power tuning
RRM techniques combining IFHO and CPICH power tuning for CPICH power tuning has been investigated as one of the
balancing load and for achieving better performance by using a techniques to control the cell coverage and to achieve the load
dynamic WCDMA simulator. First, we compare the IFHO balancing [1][2]. By reducing the CPICH power, the operator
performances in terms of the IFHO trigger types. In this can reduce the cell coverage and cell load. In addition to that,
comparison, we will show the CPICH EcNo (code energy divided the reduction of CPICH power directly leads to the decrease of
by noise energy)-triggered IFHO can achieve 30-40 % downlink DL load. According to [4], the adjustment of CPICH power can
(DL) load decreases in hotspot area. Next, we analyze the relation improve the network stability and call quality by reducing the
between the IFHO and CPICH power tuning, by comparing the BTS power. Generally speaking, the impact of CPICH power
effect of CPICH power tuning with and without IFHO function tuning is broad. For example, when reducing CPICH power,
and show its performance by using the simulator. Then finally, the problem of coverage hole needs to be considered. Then the
we will propose an example procedure to utilize the combination
paper [3] uses cost functions to evaluate the whole
of IFHO and CPICH power tuning based on the revealed relation
between them.
performance.

Keywords: WCDMA, Inter-frequency handover, CPICH power II. ANALYTICAL APPROACH


tuning, Pilot power tuning, Radio resource management, Load
Analytical approach can provide an idea about what is
balancing, Dynamic simulator
expected. It is possible to apply the following discussion to
both UL and DL. Assume that there are two frequencies, freq1
I. BACKGROUNDS and freq2 in the system. The definition of UL load ηUL and DL
In this paper, the combination of IFHO and CPICH power load η DL in this paper are as follows:
tuning technique will be discussed. For that purpose, the
backgrounds of IFHO function and CPICH power tuning are ηUL = ( Interferenceown + Interferenceother ) Pr x total ,
separately explained in this section.
η DL = Ptx total Ptx max . (1)
A. Inter-Frequency Handover
One advantage of IFHO is the trunking gain, which can be Those loads fluctuate due to fading, noise, mobility, traffic,
obtained from the load balancing between different system action, and other factors. In addition, if there are packet
frequencies. One main disadvantage of IFHO is the users, the fluctuation of the load becomes larger. Thus, loads
compressed mode, in which UE abandons its transmission and need to be analyzed as statistical values. Assuming that η freq1
reception periodically to measure another frequency. This and η freq 2 are stochastic variables, η freq1 and η freq 2 are the mean
mode reduces the performance of power control [5], and leads values, ∆η freq1 and ∆η freq 2 are the standard deviation of the
to the increase of interference.
In terms of IFHO trigger, there are several choices, such as loads on freq1 and freq2. If there is no IFHO function or no
whether to use uplink (UL) trigger or downlink (DL) trigger, load balancing between freq1 and freq2, those load
and whether to use code power or EcNo (code energy divided distributions do not change. If there is IFHO function and, if
by noise energy). The comparison of CPICH EcNo and RSCP the perfect load balancing effect is available, the stochastic
(Received Signal Code Power) has been already discussed both variables are balanced to the same value η as,

0-7803-8256-0/04/$20.00 (C) 2004 IEEE


η = (η freq1 + η freq 2 ) 2 . (2)

Then, the mean value and the standard deviation can be


easily calculated as
η = (η freq1 + η freq 2 ) 2 , (3)
hotspot
2 2
∆η = ∆η freq1 + ∆η freq 2 2. (4) freq.2

Thus, if the frequency 1 and frequency 2 originally have the


same statistical distribution, in other words, if
η freq1 = η freq 2 ≡ η freq and ∆η freq1 = ∆η freq 2 ≡ ∆η freq , the above
equation can be written as,
freq.1
η = η freq , (5)

∆η = ∆η freq 2. (6)
Figure 1. cell configuration and hotspot area
These equations indicate that thanks to the load balancing
effect of IFHO, the fluctuation of the load can be suppressed.
As a result, the system becomes more stable, and the B. Simulation results
probability that the system goes into the overload state Because CPICH power tuning has a direct impact on the
becomes lower. Therefore the system can accommodate more DL IFHO triggers and our final target is developing a
users, and the capacity becomes more than double. This is controllable combination of IFHO and CPICH power tuning,
called as a trunking gain. It can be concluded that the important DL IFHO triggers are more important than UL IFHO trigger in
thing to achieve a trunking gain is how perfect the load this research. Then, we compared the performances between
balancing is performed. CPICH RSCP and CPICH EcNo triggers. Note that in the
following, the definition of “IFHO rate” used in figures is
From the above analysis, it becomes clear that IFHO can
provide not only the trunking gain but also the stability of the IFHO rate =number of IFHO / number of calls. (7)
load. Usually, when tuning CPICH power, too high Figure 2 and Figure 3 show call success rate (CSR). In
interference and coverage hole need to be considered. If the Figure 4 showing DL loads, Cell 0, 1, 2, 7, 8 and Cell 19, 20,
load is more stable, the probability that the system goes into 26, 27, 30 are in different frequencies and, Cell 7 and Cell 20
those bad conditions becomes lower. Then the range that the are two hotspot-closest cells. The threshold of IFHO trigger
CPICH power can be changed becomes larger. From this can control IFHO rate.
consideration, at least, it is expected that using the combination
of IFHO and CPICH power tuning can provide better load 100
99
balancing ability than using only one of them. 98
97
CSR [%]

96
III. IFHO EVALUATION BY DYNAMIC SIMULATIONS 95
U L , cp ich E cN o
D L , cp ich E cN o
94 U L , cp ich R S C P

A. Simulation condition 93 D L , cp ich R S C P


92
Different from the paper [7], we used the uniformly 91

distributed two-layered macro cell configuration shown in 0 0 .1 0 .2 0 .3 0 .4 0 .5 0 .6


IFH O r a te
Figure 1. Both frequencies have almost 100 % coverage by
using two kinds of BTSs, which are 120-degree-separated 3- Figure 2. speech CSR for DL IFHO triggers
sector BTS and at the same location 60-degree-rotated to each 10 0
other. This is for fully utilizing the available frequencies and
for simplifying evaluation. Terminals generate calls by 98

randomly choosing either frequency 1 or frequency 2 with 96


CSR [%]

equal probabilities for each frequency. They move at 20 km/h 94 U L, c pic h E cN o

and, in addition, there is a hotspot, at which user density is 92


D L, c pic h E cN o
U L, c pic h R S C P
higher than other area. The other dynamic simulator settings 90
D L, c pic h R S C P

are listed in table 1 in the last page of this paper. Our


88
evaluations are done on ten sectors surrounding this hotspot, 0 0.1 0.2 0.3 0.4 0.5 0 .6
namely, sectors 0, 1, 2, 7, 8, 19, 20, 26, 27 and 30 in Figure 1. IF H O rate

In simulations, the negative effect of compressed mode is Figure 3. packet CSR for DL IFHO triggers
not included due to the limitation of the simulator. However,
that effect will be evaluated manually in the analysis section of
this paper using the values from the literature [8].

0-7803-8256-0/04/$20.00 (C) 2004 IEEE


0.8
Cell 0 compressed mode and the gain is still positive. In addition,
0.7 Cell 1 because IFHO rate determines the negative effect of
Cell 2
0.6 Cell 7
compressed mode and the type of IFHO triggers does not have
DL loading

0.5
Cell 8 any influences on that negative effect, our conclusion, that
Cell 19
Cell 20
EcNo trigger is better, is still valid, even if the compressed
0.4
Cell 26 mode is considered.
0.3 Cell 27
Cell 30
0.2
0 0.1 0.2 0.3 0.4 0.5 0.6
C. Difference between UL and DL performance
IFHO rate
As explained in the previous subsection, the DL
Figure 4. DL load balancing effect of CPICH EcNo-triggered IFHO performance is improved better than UL performance. In some
cases, UL performance becomes worse in too high IFHO rate
The simulation results reveal that region. Even using UL IFHO triggers such as UL transmission
1. CSR in Figure 2 and Figure 3, packet delay, and power and UL quality, DL performance is also improved and
throughput (not shown) are better with higher IFHO. UL performance has a peak at a certain IFHO rate. This
The different behavior between UL and DL is different behavior of UL and DL performance can be explained
discussed in the next subsection. from the behavior of UL load. Figure 5 shows the UL load for
CPICH RSCP-triggered IFHO case. The UL load increases if
2. Both DL IFHO triggers improve the DL performance the IFHO rate is over 10 %. In CPICH EcNo-triggered IFHO
better than UL performance. One reason is that DL case, the UL load is almost stable or slightly increases. One
IFHO trigger is based on DL load imbalance and UL important lesson from those results is that even without the
only receives an indirect benefit from this DL- negative effect of compressed mode, too high IFHO rate could
triggered IFHO. Detail discussion is done in the next cause higher load and worse performance in UL.
sub section.
Next, we explain the increase of UL load shown in Figure 5.
3. CPICH EcNo trigger achieves better performance Generally speaking, handovers done at improper timing are
than CPICH RSCP trigger. Figure 2 and Figure 3 located far from the middle point between two BTSs (base
show that difference of achievable performance. In transceiver stations), in other words handover is done too early
addition, comparing to the load balancing effect of or too late when mobiles move. This kind of improper timing
CPICH EcNo triggered IFHO in Figure 4, the load handover causes higher UL interference to the closer BTSs
balancing effect of RSCP triggered IFHO is smaller. because the UE (user equipment) is connected to the farther
This is because EcNo includes the impact of BTS. If the IFHO threshold is too loose and IFHO rate is too
interference in the trigger decision and RSCP does high, many improper timing handovers would occur. This can
not. As an example, this difference can be seen also cause the UL load increase and worsen the UL performance.
in the reaction to the hotspot, or load imbalance. In This discussion can be also applied to DL. However, because
IFHO occurrence distribution map, although CPICH generally UL suffers from interference more than DL in
EcNo triggered IFHO is concentrated on hotspot area, CDMA network, UL load increase is more serious in the
CPICH RSCP triggered IFHO scattered indifferent to results.
hotspot.
4. CPICH EcNo trigger achieves 30-40 % DL load IV. IFHO AND CPICH POWER TUNING
decrease as shown in Figure 4. The paper [7]
concludes that CPICH RSCP-triggered IFHO A. Simulation condition
achieves better performance than CPICH EcNo- To evaluate the combination of IFHO and CPICH power
triggered IFHO, which is opposite to our conclusion tuning in load balancing, we choose CPICH EcNo trigger. One
from these results. There are two main reasons for reason for this choice is that CPICH EcNo trigger proves to be
this difference. One reason is the difference of the a better trigger in the performance analysis done in section II.
performance indicator used for the evaluation. The The other reason is that CPICH EcNo trigger has better load
paper [7] only mentions the call block and call drop, balancing ability. This can be seen in the results that CPICH
although in our simulation cases, call blocks and call EcNo triggered-IFHO occurrence is concentrated on hotspot.
drops are almost zero because the coverage is almost
100 % and load is not too high. Then, we used other 0.9
Cell 0
indicators, such as load, CSR, throughput, and packet 0.85
0.8
Cell 1

delay. The other reason is the difference of the cell 0.75


Cell 2
Cell 7
UL loading

configuration. They used hierarchical structure and 0.7


0.65
Cell 8

our simulations use uniformly-distributed cell 0.6


Cell 19
Cell 20
structure. 0.55
Cell 26
0.5
Cell 27
According to the literature [8], the negative effect of 0.45
0.4
Cell 30

compressed mode can be estimated to be roughly 20 % if all 0 0.1 0.2 0.3 0.4 0.5

users are in compressed mode and 2 % if a tenth of users are in IFHO rate

compressed mode. Therefore, we can conclude that 30-40 % Figure 5. UL load balancing effect of CPICH RSCP-triggered IFHO
gain in DL capacity is larger than the negative effect of

0-7803-8256-0/04/$20.00 (C) 2004 IEEE


DL loading / (7 20)/ no IFHO
0.8 Cell 0
0.8 Cell 1
0.7
0.7 Cell 2
0.6
0.6 Cell 7

DL loading
0.5
0.5 Cell 8
IFHO rate

0.4 Cell 19
0.4
0.3 Cell 20
0.3
0.2 Cell 26
0.2
Cell 27
0.1
0.1 Cell 30
0 0
1 1.5 2 2.5 3
1 1.5 2 2.5 3
Pilot pow e r of Ce ll 7 a nd 20 [W ]
Pilot power of Cell 7 and 20 [W]
Figure 8. CPICH power impact on DL load without IFHO
Figure 6. The impact of CPICH power on IFHO rate
Another thing to be mentioned about Figure 7 and Figure 8
Generally speaking, the purpose of CPICH tuning is to
is that the DL loads of frequency 1 and frequency 2 are almost
mitigate the load imbalance, such as the situation occurring in
the same when IFHO function is active but the DL loads of
hotspot area, by changing the coverage of each sector defined
frequency 2 is higher than that of frequency 1 when the IFHO
by CPICH power. Then in this simulation experiment, we
function is not active. Although there in no clear reason about
changed the CPICH power of two hotspot-closest sectors,
which frequency should have higher load, this difference is, at
which are sector 7 and sector 20 in Figure 1, from 1 W to 3 W
least, because of the load balancing effect of IFHO function.
keeping both CPICH powers to be the same. Other simulation
Without IFHO function, the load difference between frequency
settings are the same as explained in subsection II-A.
1 and frequency 2 is mainly determined at the call initiation
As shown in Figure 6, CPICH power tuning has a large phase and then call blocks have an effect on this load
impact on IFHO rate triggered by EcNo. This is because by difference. In addition, call blocks are also affected by the load
decreasing the CPICH power and reducing coverage of situation. In this sense, the system load behavior without IFHO
hotspot-closest sectors, other sectors than sector 7 and sector function is more complicated than that with IFHO load
20 need to cover the hotspot. This situation causes more IFHO balancing effect. In other words, if IFHO function is used, its
around hotspot area. This CPICH power tuning simulation is load balancing effect makes the system load behavior simpler,
done with very low IFHO threshold, namely with very high and makes the CPICH power tuning easier and more effective.
IFHO rate. The rightmost point in Figure 4 corresponds to 2 W
in Figure 6. C. Analysis on CPICH power tuning with IFHO
The comparison between with IFHO and without IFHO
B. Comparison between with IFHO and without IFHO shows that the CPICH power tuning is easier and more
To clarify the effect of IFHO on the performance of CPICH effective with IFHO than without IFHO. Then hereafter, the
power tuning, we first show the DL load behavior caused by analysis is done only for the CPICH power tuning with IFHO
the change of CPICH power with and without IFHO function. function. In Figure 7, the DL load of hotspot-closest cells can
Figure 7 shows the result with IFHO and Figure 8 shows the be decreased to the same level as the load of other cells by
result without IFHO. The comparison of those results suggests reducing their CPICH power to 1 W. This load decrease is very
that the CPICH power tuning with IFHO has better load clear and the rule to achieve this load balancing from CPICH
balancing ability than without IFHO. This results can be power tuning is very simple. Another important thing is that
explained as follows. Without IFHO, when decreasing CPICH along with this DL load balancing, CSR, packet delay and
power, other cells, which need to cover that hotspot area, throughput are also improved and UL load does not increase.
should be of the same frequency. With IFHO, covering cells Later in conclusion, we propose an example procedure to
can be of any frequencies. Then this means that the distance achieve this load balancing by IFHO and CPICH power tuning.
between hotspot and these alternative cells covering the hotspot
Figure 9 shows the UL load behavior, which does not have
is larger without IFHO than with IFHO. Then, without IFHO,
any clear tendency and is almost stable. Even without IFHO,
the transmission powers in UL and DL are larger, causing the
the UL load behaves in almost the same way. In this sense,
increase of loads.
performance analysis should be on DL.
0.7
U L lo a d in g / (7 =2 0 )/ c p ic h E c N o 12
0.6 0.8
Cell 0 Cell 0
0.75
0.5 Cell 1 Cell 1
0.7
Cell 2
DL loading

Cell 2
0.4 0.65
Cell 7 Cell 7
UL loading

0.6
Cell 8 Cell 8
0.3
0.55 Cell 19
Cell 19
0.2 0.5 Cell 20
Cell 20
0.45 Cell 26
0.1 Cell 26
0.4 Cell 27
Cell 27
0.35 Cell 30
0 Cell 30
1 1.5 2 2.5 3 0.3
1 1.5 2 2 .5 3
Pilot power of Cell 7 and 20 [W]
P ilo t p o w e r o f Ce ll 7 a n d 20 [W ]
Figure 7. CPICH power impact on DL load with IFHO Figure 9. CPICH power impact on UL load with IFHO

0-7803-8256-0/04/$20.00 (C) 2004 IEEE


100 TABLE I. DYNAMIC SIMULATOR SETTING
98
Cell 18 macro cells (frequency 1)
96 environment 23 macro cells (frequnecy 2)
CSR [%]

94
UL P a c k e t C S R Multi path Vehicular A
92 D L Packet C SR model
90 UL S p e e c h C S R Fast fading Jakes model
88
D L Speech C SR
Log normal distribution with 8 dB
86
std.dev.
1 1 .5 2 2 .5 3 Slow fading Site correlation = 0.5, sector
P ilo t P o w e r o f C e ll 7 a n d C e ll 2 0 [ W ] correlation = 0.8
Figure 10. CPICH power impact on CSR with IFHO
Correlation distance = 50 m
Code 0.5
In Figure 7, another topic needs to be analyzed. The loads orthogonality
of hotspot-closest cells have a peak at CPICH power = 2.5 W. Propagation Okumura-Hata model
model
The behavior of CSR shown in Figure 10 can explain this DL
UL target: noise rise = 4 dB
load behavior. An important result from this figure is that the Load target DL target: transmission power = 10
DL speech CSR decreases dramatically at the CPICH power W
of 3W. The increases of CPICH power leads to the increase of CPICH power 2 W, (1-3 W in CPICH tuning)
coverage and load. The higher load triggers admission control BTS max
(AC) actions more and leads to lower CSR and lower load. power 20 W
But due to the larger coverage, the number of packet users Terminal
increases, which leads to the larger load fluctuation. Therefore velocity 20 km/h
the number of AC actions is still high although the average Number of 10000 UEs including active and
load becomes lower. terminals inactive
70 % speech (12.2 kbps) and
V. CONCLUSION Proportion of
traffic 30 % packet (8, 12, 16, 32, 64, 144
In this paper, we reported the benefit of a combination of kbps)
IFHO and CPICH power tuning. Analytical approach reveals FER target 1 % for speech and 10 % for packet
that IFHO can stabilize the load fluctuation and helps the
CPICH power tuning. Using a dynamic simulator, a
comparison of IFHO trigger types is done first. The results ACKNOWLEDGMENTS
indicate that CPICH EcNo is better than CPICH RSCP for This work is supported by Nokia Research Center. The
balancing load in the uniformly overlapped multi-frequency authors would like to thank Rin Nagaike, Dr. Sivakumar and
WCDMA network. CPICH EcNo achieves 30-40 % DL load the developers of WCDMA dynamic simulator.
decrease of the hotspot-closest cells. Although compressed
mode could make a certain amount of loss, according to the
REFERENCES
literature, the loss is less than 20 % capacity loss, which is less
that the gain we obtained above. Next, we did simulations on
the combination of IFHO and CPICH power tuning. CPICH [1] Dongwoo Kim., et al, “Pilot Power Control and Service Coverage
Support in CDMA Mobild Systems”, in proceedings of the 49th IEEE
power tuning can almost perfectly remove the load imbalance VTC. Pp. 1464-1468, 1999.
at DL left by IFHO, while CPICH power tuning without IFHO
[2] Valkealahti K, et al, “WCDMA common pilot power control for load
cannot achieve any load balance by simple CPICH power and coverage balancing”, Personal, Indoor and Mobile Radio
tuning. Other performance indicators do not become worse in Communications, the 13th IEEE, pp 1412-1416, 2002
the tuning of CPICH power with IFHO. Therefore, we [3] Valkealahti K, et al, “WCDMA common pilot power control with cost
concluded that their combination is useful for DL load function minimization”, Proceedings of the 56th IEEE VTC pp 2244-
balancing. 2247, 2002
[4] Houtao Zhu et at. “Load balancing in WCDMA systems by adjusting
An example of simple optimization procedure is as follow: pilot power” Wireless Personal Multimedia Communications, IEEE The
First turn on IFHO to stabilize the load fluctuation and help the 5th international Sym. Vol. 3, pp 936-940, 2002
CPICH tuning. Next, tune CPICH power to the optimum. Then, [5] Maria Custafsson et al. “Compressed Mode Techniques for Inter-
tune IFHO threshold considering the balance of trunking gain Frequency Measurements in a Wide-band DS-CDMA System” IEEE
and the negative effect of compressed mode. There are reasons pp231-235, 1997
for this procedure. Because CPICH power has a large impact [6] X Yang, et al. “Performance of Power-Triggered and Ec/No-Triggered
Soft Handover Algorithms for UTRA”, 3G Mobile Com, IEEE 2001
on IFHO rate, it needs to be tuned first. But without IFHO,
[7] Wang Ying et al. “Performance of RSCP-Triggered and Ec/No-
CPICH tuning shows totally different performance, CPICH Triggered Inter-Frequency Handover Criteria for UTRA”, IEEE 2002
tuning needs to be done with IFHO. The quantitative
[8] “Radio Network Planning and Optimization for UMTS” book, WILEY
evaluation of the compressed mode needs to be done in future
work.

0-7803-8256-0/04/$20.00 (C) 2004 IEEE

You might also like