Validation of an Improved Location-Based Handover Algorithm Using ...
Validation of an Improved Location-Based Handover Algorithm Using GSM Measurement Data
Hsin-Piao Lin, Rong-Terng Juang, and Ding-Bing Lin Institute of Computer, Communication and Control, National Taipei University of Technology, Taipei, Taiwan, +886-2-27712171 EXT. 2248 [email protected], [email protected], [email protected]
Abstract–When a mobile station moves, path loss and shadow fading contribute to large-scale variation in the received signal strength. The variation of signal strength caused by shadow fading is a random process, and handover decision mechanisms based on measurements of signal strength induce the “ping-pong effect”. This paper proposes an improved handover algorithm based on the estimates of location and velocity of the mobile station, to suppress the ping-pong effect in cellular systems. A practical approach (not a GPS technique) with reasonable errors is used to estimate the location and velocity to identify a correlation among shadowing effects. The impact of location errors on handover performance was investigated, and the proposed handover algorithm was applied to a real GSM system in urban Taipei city. The results reveal that the improvement in handover number, which indicates performance, over the conventional method ranges from 18 to 26% while the signal outage probability remains similar, indicating similar quality of services. Besides, the computational complexity of the proposed algorithm is low and the algorithm does not use a database or lookup table.
Key Words: Handover, mobile location, mobile velocity estimation, shadow fading.
I.
INTRODUCTION
Handover refer to the mechanism by which an ongoing call is transferred from one base station (BS) to another. The performance of the handover mechanism is extremely important in mobile cellular networks, in maintaining the desired quality of service (QoS). In a cellular system, when a mobile user travels from one cell to another, the serving BS changes according to the planning of the system. Frequent handovers influence the QoS, increase the signaling overhead on the network, and degrade throughput in data communications. Thus, network operators should emphasize the optimization of handover decisions. Many metrics have been used to support handover decisions, including received signal strength (RSS), signal to interference ratio (SIR), distance between mobile and BS, traffic load and mobile velocity, where RSS is the most commonly used one. The conventional handover decision compares the RSS from the serving BS with that from one of the target BSs, using a constant handover threshold value (handover margin). The selection of this margin is crucial to handover performance. If the margin is too small, numerous unnecessary handovers may be processed. Conversely, the QoS may be low and calls may be dropped if the margin is too large. The fluctuations of signal strength associated with shadow fading cause a call sometimes to be repeatedly handed over back and forth between neighboring BSs, in what is called the “ping-pong effect”. Over recent years, many investigations have addressed handover algorithms for cellular communication systems. A local averaging technique removing fast fading component from the received signal strength was proposed in [1] to allow the conventional handoff decision reacting more quickly to corner effects. A timer-based hard handover algorithm was presented in [2] to prevent unnecessary handovers caused by fluctuations due to shadowing. A dynamic handover margin decision based on a traffic balancing rule was proposed in [3] to resize the cells according to the spatial variability of traffic. A speed-sensitive handover algorithm in a
hierarchical cellular system was described in [4], in which micro-cells serve the slowly-moving mobiles, and macro-cells serve fast-moving mobiles. In [5] and [6], RSS, mobile location and velocity were used as metrics for making handover decisions using fuzzy logic. A table lookup approach, proposed in [7], determines handover margins based on the mobile location, the mean of the intensity of the signal, and the standard deviation. Distance hysteresis for mitigating the effect of fading on handover performance was presented in [8]. Making handover decisions in various handover scenarios was presented in [9], in which a suitable handover decision is selected when mobile is located in an area with a pre-defined handover scenario. However, most handover algorithms presented in the literature based on information about mobile location suffer from a lack of practicability. The computational complexity of making a handover decision using fuzzy logic is excessive, and establishing and updating a lookup table to support a handover margin decision is time-consuming. The selection of a handover algorithm based on the handover scenario only succeeds in cases that the mobile environment is similar to one of the pre-classified environments, and involves complicated processes to define the handover scenarios. It also relies on an updated database when applied in a new mobile user environment. Furthermore, most studies assume that location of the mobile can be perfectly determined using the GPS (Global Positioning System), which is not available for most mobile telephones. In reality, the performances of available location estimators are far from that obtainable using GPS technique. This paper proposes an improved handover algorithm based on the estimates of mobile location (not using GPS) and velocity. The proposed handover algorithm was applied to a real GSM system in urban Taipei city. The signal level measured in decibels is the sum of two propagation terms - one associated with path loss and the other associated with shadow fading; fast fading is ignored because it is averaged out. The variation of the signal caused by shadow fading depends on the location and velocity of the mobile user. The proposed
algorithm outperforms the conventional method in making handover decisions for cellular systems by using location and velocity to identify the correlation among shadowing effects. Moreover, the computational complexity of the proposed algorithm is low and the algorithm does not employ a database or lookup table. The rest of this paper is organized as follows. Section Ⅱ describes the system model for analyzing handover performance and proposes the handover algorithm. Section Ⅲ examines the effect of location errors on handover performance. Section Ⅳ presents the performance verification using GSM measurements data. Section Ⅵ draws conclusions.
II.
PROPOSED HANDOVER ALGORITHM
a) System Model In a GSM system, when a mobile moves from BS1 to BS2, the signal strength is measured and reported to the network in a constant 480ms time interval [10] to support a handover decision. The signal level in decibels is the sum of two propagation terms, namely path loss and shadowing; fast fading is ignored because it is averaged out. Accordingly, the signal power levels received from BS1 and BS2 at discrete time instants t k k (where represents the time interval of 480ms within which the signal strength is measured), are given by P1[k ] and P2 [k ] , respectively.
P1[k ] m1[k ] u1[k ]
(1)
P2 [k ] m2 [k ] u 2 [k ]
(2)
where m1 and m2 are the received signal powers from BS1 and BS2, respectively, in terms of only path loss, and u1 and u 2 are respective the shadow fadings. The auto-correlation coefficient, ii , of the shadow fadings is commonly assumed to be an exponential function [11][12],
ii
E ui [k1 ]u i [k 2 ] exp( d d ), i 1, 2 i2
(3)
where i is the standard deviation of shadow fadings; d V k 2 k1 ( V is mobile velocity, non-negative number), and d is the decay distance (or correlation distance), which ranges from around 25 to 100m over urban, light urban, and suburban terrain [13]. The cross-correlation coefficient, ij , of shadow fadings is called the “site-to-site correlation” [14] and calculated as
ij
E ui [k ]u j [k ]
i j
, i, j 1, 2, i j
(4)
The correlation depends on 1) the angle between the two paths along the mobile to BS1 and BS2, and 2) the relative values of the two path lengths. Jay Weitzen et. al. verified that the shadow fading components are slightly correlated even at small angles [13]. b) Proposed Handover Algorithm Define P21[k ] as the difference between signal powers received from BS2 and BS1 at time index k:
P21[k ] P2 [k ] P1[k ] m2 [k ] m1[k ] u 2 [k ] u1[k ] m21[k ] u 21[k ]
(5)
where m21 represents the difference between signal powers received from BS2 and BS1 in terms only of path loss, u 21 represents the difference between the shadow fadings along the two paths. A handover from BS1 to BS2 occurs at time index k if the following two criteria are satisfied. Criterion 1: P21[k ] h Criterion 2: P21[k ] P21[k ] h where h is the handover margin, and is a positive non-zero integer, which needs to be carefully decided. In fact, criterion 1 is applied in making conventional handover decisions.
Because of shadowing, unnecessary handovers may be performed if a handover decision is based only on this criterion. Therefore, criterion 2 is imposed to improve the handover performance by determining whether path loss dominates the variation in the signal. Assume u 21[k ] and u 21[k ] are highly correlated, such that the correlation coefficient approaches unity; then, the difference between P21[k ] and P21[k ] can be approximated as P21[k ] P21[k ] m2 [k ] m2 [k ] m1[k ] m1[k ] m2(up ) m1( down )
(6)
where m2(up ) and m1( down) are the increase and degradation of the signal powers received from BS2 and BS1 in terms of path loss, due to motion of the mobile. Consequently, the difference between signal powers is always chiefly a function of path loss but not of shadow fading. Restate, the proposed algorithm ensures that the signal power received from target BS is h dB higher than that received from the serving BS (criterion 1), and that the difference between the signal powers is dominated by path loss associated with motion of the mobile (criterion 2). Hence, unnecessary handovers caused by fluctuations in shadow fadings are avoided. In the proposed algorithm, is critical to handover performance. The decided must guarantee high correlation between u 21[k ] and u 21[k ] , and sufficient space for signal variation caused by path loss. If is too large, criterion 2 is always met and does not support handover decisions. Conversely, the signal dose not vary if is too small. How to decide a suitable value of is explained below. Given u1[k ] , if the standard deviations of shadow fadings are assumed to be equal, such that 1 2 u , then u 2 [k ] , u1[k ] and
u 2 [k ] can be expressed as follow, based on the Gauss-Markov process. 2
u 2 [k ] 12u1[k ] 1 12 X 1
(7)
2
u1[k ] 11u1[k ] 1 11 X 2 2
u 2 [k ] 22u 2 [k ] 1 22 X 3
(8) (9)
where X 1 , X 2 and X 3 are identical independent Gaussian processes with zero-mean and variance u2 and Eu1[k ] X 1 Eu1[k ] X 2 Eu 2 [k ] X 3 0 . Assume 1) 12 21 c and 2) 11 22 a . The following can be proven. E u1[ k ]u 2 [ k ] Eu1[ k ]u2 [ k ] c u2
(10)
E u1[ k ]u1[ k ] Eu 2 [ k ]u 2 [ k ] a u2
(11)
E u1[ k ]u 2 [ k ] E u 2 [ k ]u1[ k ] c a u2
(12)
The correlation between u 21[k ] and u 21[k ] is
Eu 21 [k ]u 21 [k ] Eu 2 [k ] u1[k ]u 2 [k ] u1[k ] a (1 c )(2 u2 )
(13)
exp( V d )(1 c )(2 u2 ) The correlation coefficient between u 21[k ] and u 21[k ] must exceed a threshold, T ; that is, exp( V d )(1 c ) T , such that ln T 1 c
d V
(14)
Having describing the determination of , Fig. 1 displays the flowchart of the proposed handover decision. The mobile provides the measurement reports to network to support handover decisions within constant time intervals, . This data is buffered in the memory for the mobile location estimation proposed in [15]. The handover alarm is triggered, when the signal power received from the serving BS is below a threshold, then the availability of the target BS is verified according to criterion 1. If target BS meets criterion 1, the data buffered in the memory are fetched to estimate the location and velocity of the mobile, saving the overhead cost of calculating location since the mobile is not continuously tracked. The value
of can be obtained from mobile velocity using Eq. (14) to confirm criterion 2. Consequently, a handover occurs if the target BS satisfies criteria 1 and 2 simultaneously, otherwise the serving BS remains unchanged and the handover decision is made again at the next time.
Figure 1. Flowchart of proposed handover decision.
The results of the simulations using the proposed handover algorithm are compared with those obtained using the conventional method. A software package, SignalPro by EDX Engineering, was used to help the simulation. It includes a set of planning tools for wireless communication systems. Figure 2 reveals the simulation environment that covers an area of 1.6 x 1.4Km2. The solid blue trajectory from “A” to “B” represents a route through which the mobile moves. Polygons of different colors represent buildings of different heights. Seven BSs with omni-directional antennae are designated by encircled crosses ( ). The height of each BS is 35m and the mean and standard deviation of their transmitting power (EIRP) are 42.6dBm and 3.5dB, respectively. Walfisch-Ikegami model was applied to simulate path loss. It is a hybrid model, which considers diffraction down to street level and empirical correction
factors. The model was verified to predict accurately propagation path loss in urban areas with small cells [16]. Shadow fadings were simulated according to the model proposed in [17], where d 65 m, and c 0.1 . The mobile moved along the trajectory in Fig. 2 at a constant speed of 30Km/h. The sampling interval for reporting measurements was 0.48s. The handover alarm threshold, handover margin, and correlation threshold were set to –80dBm, 6dB and T 0.85 , respectively. Figure 3 is a typical comparison between the received signal time series obtained by the conventional method and that obtained by the proposed handover algorithm when the mobile moves along the beginning of the trajectory in Fig. 2. The top time series was simulated using the conventional method while the bottom one was simulated using the proposed algorithm. In the simulations, the mobile velocity was assumed to be perfectly estimated and the standard deviation of shadow fading was set to 9dB. The result indicates that the conventional method involves more handovers whereas the proposed algorithm prevents unnecessary handovers.
Figure 2. The simulation environment.
Figure 3. Comparison of signals received according to the conventional method (top plot) and those received according to the proposed handover algorithm (bottom plot).
III.
ANALYSIS OF HANDOVER PERFORMANCE WITH LOCATION ERRORS
Knowledge of the velocity of the mobile is required to determine the value of in the proposed algorithm. The GPS receiver is not yet available in mobile devices, so the effect of the estimation errors of velocity must be considered. The velocity of the mobile was estimated based on Doppler frequency shift in [18]. However, the estimated Doppler frequency is unreachable in most standards of the mobile cellular systems. This paper presents a means of estimating mobile velocity based on the mobile locations. For simplicity, the problem is reduced to the one-dimensional case. The mobile location estimate at time index k is modeled as Lˆ[ k ] L[ k ] n L
(15)
where L[k ] is the actual mobile location, and n L represents the location error, which is modeled as a zero-mean Gaussian process with variance L2 , as in [19]. Previous location information is used to estimate the current velocity. The size of the estimation window is M,
so the estimated locations
Lˆ[k ], Lˆ[k 1], , Lˆ[k M 1]
are used to estimate mobile
velocity. An adequate integer m (1 m M 2 and ( M m) is even) is chosen such that the mean
Lˆ[k ], , Lˆ[k (M m) 2 1]
Lˆ[ k ( M m) 4 0.5] ,
is used as a more accurate version of
which
is
represented
Lˆ[k (M m) 2], , Lˆ[k M 1]
is
used
as
Lˆ [i] ,
by a
more
and
accurate
the
mean
version
of
Lˆ[ k (3M m) 4 0.5] , which is represented by Lˆ [ j ] . Then, the estimated mobile velocity
at time index k has the form
Vˆ vˆ[k ]
Lˆ [i] Lˆ [ j ] m 1
v[k ] nv
(16)
where nv is the error in the estimated velocity and is also a zero-mean Gaussian process with variance v2 [ 4 ( M 2 m 2 )] 2 ( M m) L2 . Given suitably chosen M and m, the estimate of mobile velocity is highly accurate. As a simple example, M=7 and m=3 are chosen, as presented in Fig. 4, where Lˆ [3] and Lˆ [5] are
(17)
(18)
Vˆ vˆ[7] Lˆ [5] Lˆ [3] 2 v[7] nv
(19)
Lˆ [3] Lˆ[1] Lˆ[2] Lˆ[3] Lˆ[4] Lˆ[5] / 5 Lˆ [5] Lˆ[3] Lˆ[4] Lˆ[5] Lˆ[6] Lˆ[7] / 5 The estimated mobile velocity is
where the variance of nv is (1 / 25) L2 .
Figure 4. An example of estimation of mobile velocity from location. (M=7, m=3)
However, the mobile location is a two-dimensional problem in reality. The estimates of location on the horizontal-axis and the vertical-axis at time index k are respectively expressed as
xˆ[k ] x[k ] n L( x ) ( y) yˆ[k ] y[k ] n L
(20)
where x[k ] and y[k ] are the actual mobile location of the mobile, and n L( x ) and n L( y ) represent the location errors, which are modeled as zero-mean Gaussian processes with variance L2 . Similar to Eq. (16), the estimated velocity at time index k is Vˆ
vˆ x [k ]2 vˆ y [k ]2
(21)
where vˆ x [k ] and vˆ y [k ] are the estimations of velocity on the horizontal-axis and the vertical-axis, respectively. Denote the actual velocity of the mobile as V and assume the variances of the error terms, in vˆx [k ] and vˆ x [k ] , are the same to v2 , the probability density function (p.d.f.) of Vˆ is a Rice distribution with Rice factor K V 2 (2 v2 ) [14][20],
Vˆ 2 K Vˆ 2 Vˆ fVˆ (Vˆ ) 2 exp exp K I 0 2 v 2 v v
(22)
where I 0 ( z ) is a modified Bessel function of the first kind and zeroth-order. For a Rice distribution, a larger K which corresponds to a faster mobile or a lower v yields a more accurate estimate of velocity because the p.d.f. curve is sharper. Moreover, redefine Eq. (14) as ˆ ln T (1 c ) d (Vˆ ) , ˆ distributes as
f ˆ (ˆ) 2 f Vˆ ˆ ˆ
(23)
where ln T (1 c ) d . Figure 5 plots the p.d.f. curves of ˆ given various location errors, which indicate the 67th percentiles of errors. Given the parameter settings
ρ
c
0.1, ρT 0.85, d 65m, τ 0.48s , mobile velocity 30 Km/h , the actual is 0.929.
The accuracy of the estimate of is very high because 1) ˆ is run off during handover decisions, and 2) ˆ is a positive non-zero integer, which resulting in ˆ 1 with very high probability.
Figure 5. Probability density of ˆ given different location errors.
Number of handovers and handover delay (referred also as crossover point) are often used as performance measures [1][2][17]. In our study, however, signal outage probability, defined as the rate that the received signal from serving BS is less than a threshold, is substituted for handover delay because the cell shape of each BS is an irregular polygon instead of a regular hexagonal structure, which resulting in difficulty in defining handover delay. Meanwhile, a larger signal outage probability implies a larger handover delay. With reference to the same case as simulated in Section Ⅱ, Fig. 6 compares the performances of the proposed and the
conventional method averaged over 2000 iterations. In this case, the standard deviations of shadow fadings are set to 3, 6 and 9dB, and the location errors are set to 0, 30, 60 and 90m. In estimating the velocity, M and m are chosen as 9 and 3. The horizontal and vertical axes in Fig. 6 respectively represent the total number of handovers and the signal outage probability associated with the outage threshold of –95dBm. The signal outage probability is obtained by recognizing how much percentage of signal received from serving BSs is below the outage threshold when mobile moves along a testing route. Symbols “○”, “＋” and “□” indicate the performance in 3, 6 and 9dB shadow fading environments, respectively. The solid blue line and the dotted red line represent the results obtained using the conventional method and the proposed algorithm, respectively. Using the proposed algorithm reduces the number of handovers and only slightly increases in signal outage probability. The handover performance improvement measured as the decrease in the number of handovers number over that obtained using the conventional method ranges from 9 to 17%.
Figure 6. Comparison of handover performance.
IV.
VERIFYING PERFORMANCE USING GSM MEASUREMENTS DATA
The proposed handover algorithm was applied to a real GSM system (1800MHz) in urban Taipei city. GSM provides measurement reports for managing radio resources. These reports include the received signal levels, RXLEV, of the serving cell and of the strongest neighboring cells. These values of RXLEV are reported from a mobile within a period of 0.48s. “TEMS Investigation GSM”, provided by Ericsson, was used to make the measurements along three selected routes. Calls were made from a car, driving along the routes plotted in Fig. 7 - route 1 from A to B, route 2 from C to D and route 3 from E to F. The area where measurements were made was 2.1 x 1.6 Km2. The blue triangles indicate sector cells with an averaged cell radius of around 330m. The mean and standard deviation of
building heights are 20.3m and 14.4m, respectively. The average and standard deviation of BS heights are 26.4m and 10.2m, respectively. As the measurements are being made, the GPS coordinates of the mobiles, and the power levels received from serving cell and the strongest neighboring cells are recorded over a period of 0.48s.
Figure 7. Measurement Environment.
Before the handover performance can be evaluated, the propagation characteristics (shadowing components) must be investigated. This work is down on a point-by-point basis. The first step is to remove the path loss (local-mean) to leave the desired shadowing component. The local-mean of each data point was determined from the ensemble of signal points located within a circle with a radius 100m. Figure 8 plots the p.d.f. of the residual shadowing components. A zero-mean Gaussian with a standard deviation 5.55 is plotted as a blue line in this figure, to ensure the shadowing components exhibit Gaussian properties. The second step is to estimate the cross-correlation coefficient of shadowing fadings. The signals received from two BSs are grouped or binned according to the angle between the two paths from the mobile to the BSs. A group must include at least 25 samples from each BS to be considered statistically significant and used in the analysis. Each bin encompasses 10 degrees
of the 180 o angular range. Figure 9 presents the average cross-correlation coefficient of the shadowing fadings, which are slightly correlated even at small angles. The final step is to estimate the correlation distance of shadowing fadings. The correlation distances are computed based the measurements from each single BS. Figure 10 presents the average auto-correlation plot, on which the red line is the fitted curve with a correlation distance 64.97m.
Figure 8. Probability density function of the shadowing components extracted from measurements.
Figure 9. Average cross-correlation coefficient of shadowing fadings.
Figure 10. Average auto-correlation coefficient of shadow fadings. The red line is the fitted curve with correlation distance 64.97 m.
The measurements data were applied for simulation of the proposed handover algorithm. The proposed algorithm estimates the mobile location based on the differences among signal attenuations, as proposed in [15]. The differences among signal attenuations were used to determine circles that encompass possible mobile locations. The actual mobile location was then estimated from the intersections of these circles. Assuming the GSP coordinates as the actual mobile position, Table Ⅰ presents the statistics on location errors, of which 67% are less than 149.5m, 194.6m and 142.1m for route 1, route 2 and rout 3, respectively [15]. In the evaluation of handover performance, the correlation distance and cross-correlation coefficient of shadow fadings are set to d 65 m and c 0.1 , respectively. The handover alarm threshold, handover margin, signal outage threshold and correlation threshold are respectively set to –70dBm, h 6 dB, -80dBm and T 0.85 . Table Ⅱ compares the handover performance of the proposed algorithm and the conventional method. The results indicate that the proposed handover algorithm reduces the number of handovers and only slightly increases the signal outage probability. The improvement over the conventional
method in the number of handovers, which indicates performance, ranges from 18 to 26 %. Table Ⅰ Summary Statistics of Location Errors (m) In Real System [15] Measurement Routes Route 1 Route 2 Route 3 (1246 pts.) (1647 pts.) (2033 pts.) mean 128.0 157.6 135.9 std 79.2 101.2 73.1 Proposed median 114.6 129.8 122.2 Method 67% 149.5 194.6 142.1 98% 333.8 433.7 326.4 th th Rows 67% and 98% represent the 67 and 98 percentiles, respectively. Table Ⅱ Handover performance of proposed algorithm and conventional method in a real environment. Conventional Proposed Method Method Route 1
Handover Number
35
26
Signal Outage Probability (%)
0.0
0.16
Handover Number
38
31
1.34
1.58
92
68
0.94
1.08
Route 2
Signal Outage Probability Handover Number Route 3 Signal Outage Probability (%)
V.
CONCLUSIONS
An improved handover algorithm is proposed, it is based on the estimated location and velocity of the mobile user to suppress the ping-pong effect in cellular systems. The computational complexity of the proposed algorithm is low and no database or lookup table is required. A method based on non-GPS location techniques is presented to estimate the velocity of the mobile user. The effect of location errors on handover performance was examined. The performance of the proposed handover algorithm is compared with that of the conventional method in a scenario with reasonable location errors. The simulations indicate
that the number of handovers is reduced and that the signal outage probability is only slightly increased. The handover performance of the proposed method, which uses a practical means of estimating the location of a mobile instead of GPS techniques, is 18 to 26% better, in terms of number of handovers, than that of the conventional method, while the signal outage probability remains similar, indicating similar quality of services.
REFERENCES [1] A. E. Leu and B. L. Mark, “Modeling and Analysis of Fast Handoff Algorithms for Microcellular Networks,” in Proc. 10th IEEE Symp. Modeling, Analysis and Simulation of Computer and Telecommun. Syst., Oct. 2002, pp. 321-328. [2] A. E. Leu and B. L. Mark, “An Efficient Timer-Based Hard Handoff Algorithm for Cellular Networks,” in Proc. IEEE Wireless Commun. and Networking Conf., vol. 2, Mar. 2003, pp. 1207-1212. [3] M. Toril, S. Pedraza, R. Ferrer, and V. Wille, “Optimization of Handover Margins in GSM/GPRS Networks,” in Proc. IEEE Veh. Technol. Conf., vol. 1, Apr. 2003, pp. 150-154. [4] D. H. Kim, W. Noh and S. An, “Speed-Sensitive Handoff under Dynamic Hierarchical Cellular Systems,” The 9th Int’l conf. Comput., Commun. and Contr. Technol., CCCT ’03, July 2003. [5] B. Honman and W. Benjapolakul, “A Handover Decision Procedure for Mobile Telephone Systems Using Fuzzy Logic,” in Proc. IEEE Asia-Pacific Conf. Circuits Sys., Nov. 1998, pp. 503-506. [6] A. Majlesi, and B. H. Khalaj, “An Adaptive Fuzzy Logic Based Handoff Algorithm for Hybrid Networks,” in Proc. IEEE Conf. Signal Processing, vol. 2, Aug. 2002, pp. 1223-1228. [7] S. S. Wang, M. Green and M. Malkawi, “Adaptive Handoff Method Using Mobile Location Information,” IEEE Emerging Technol. Symp. Broadband Commun. for the Internet Era Symp. digest, Sept. 2001, pp. 97-101. [8] W. Teerapabkajorndet and P. Krishnamurthy, “Comparison of Performance of Location-Aware and Traditional Handoff-Decision Algorithms in CDPD Networks,” IEEE Conf. Veh. Technol., vol. 1, pp. 212-216, 2001.
[9] A. Markopoulos, P. Pissaris, S. Kyriazakos, Ch. Dimitriadis, G. Karetsos and E.D. Sykas, “Increased Handover Performance in 2G and 3G Wireless Systems Based on Combined Mobile-Location and Area,” in Proc. IEEE Int’l Symp. Wireless Personal Multimedia Commun., vol. 1, Oct. 2002, pp. 47-51. [10] Jorg Eberspacher and Hans-Jorg Vogel, “GSM-Switching, Services and Protocols,” John Wiley & Sons, 1998. [11] M. Gudmundson, “Correlation Model for Shadow Fading in Mobile Radio Systems,” Electronics Letters, vol. 27, no. 23, pp. 2145-2146, Nov. 1991. [12] D. Giancristofaro, “Correlation Model for Shadow Fading in Mobile Radio Channels,” Electronics Letters, vol. 32, pp. 958-959, May. 1996. [13] Jay Weitzen and Terri J. Lowe, “Measurement of Angular and Distance Correlation Properties of Long-Normal Shadowing at 1900 MHz and Its Application to Design of PCS Systems,” IEEE Trans. Veh. Technol., vol. 51, no. 2, Mar. 2002. [14] S. R. Saunders, “Antennas and Propagation for Wireless Communication Systems,” John Wiley & Sons, 1999. [15] D. B. Lin, R. T. Juang, H. P. Lin and C.Y. Ke, “Mobile Location Estimation Based on Differences of Signal Attenuations for GSM Systems,” in Proc. IEEE Society Int’l Conf. Antennas and Propagat., vol. 1, Jun. 2003, pp. 77-80. [16] H. P. Lin, D.B. Lin and R. T. Juang, “Performance Enhancement for Microcell Planning Using Simple Genetic Algorithm”, in Proc. IEEE Int’l Conf. Antennas and Propagat., vol. 4, Jun. 2002, pp. 664-667. [17] F. Graziosi, M. Pratesi, M. Ruggieri and F. Santucci, “A Multicell Model of Handover Initiation in Mobile Cellular Networks,” IEEE Trans. Veh. Technol., vol. 48, no. 3, May. 1999. [18] G. Azemi, B. Senabji and B. Boashash, “A Novel Estimator for the Velocity of a Mobile Station in a Micro-Cellular System,” in Proc., Int’l Symp. Circuits and Sys., vol. 2, May 2003, pp. 212-215. [19] K. I. Itoh, S. Watanabe, J. S. Shih, and T. Sato, “Performance of Handoff Algorithm Based on Distance and RSSI Measurements,” IEEE Trans. Veh. Technol., vol. 51, no. 6, Nov. 2002. [20] S. O. Rice, “Statistical Properties of a Sine Wave plus Random Noise,” Bell Systems Technical Journal, vol. 27, pp 109-157, Jan. 1948.
Hsin-Piao Lin, Rong-Terng Juang, and Ding-Bing Lin Institute of Computer, Communication and Control, National Taipei University of Technology, Taipei, Taiwan, +886-2-27712171 EXT. 2248 [email protected], [email protected], [email protected]
Abstract–When a mobile station moves, path loss and shadow fading contribute to large-scale variation in the received signal strength. The variation of signal strength caused by shadow fading is a random process, and handover decision mechanisms based on measurements of signal strength induce the “ping-pong effect”. This paper proposes an improved handover algorithm based on the estimates of location and velocity of the mobile station, to suppress the ping-pong effect in cellular systems. A practical approach (not a GPS technique) with reasonable errors is used to estimate the location and velocity to identify a correlation among shadowing effects. The impact of location errors on handover performance was investigated, and the proposed handover algorithm was applied to a real GSM system in urban Taipei city. The results reveal that the improvement in handover number, which indicates performance, over the conventional method ranges from 18 to 26% while the signal outage probability remains similar, indicating similar quality of services. Besides, the computational complexity of the proposed algorithm is low and the algorithm does not use a database or lookup table.
Key Words: Handover, mobile location, mobile velocity estimation, shadow fading.
I.
INTRODUCTION
Handover refer to the mechanism by which an ongoing call is transferred from one base station (BS) to another. The performance of the handover mechanism is extremely important in mobile cellular networks, in maintaining the desired quality of service (QoS). In a cellular system, when a mobile user travels from one cell to another, the serving BS changes according to the planning of the system. Frequent handovers influence the QoS, increase the signaling overhead on the network, and degrade throughput in data communications. Thus, network operators should emphasize the optimization of handover decisions. Many metrics have been used to support handover decisions, including received signal strength (RSS), signal to interference ratio (SIR), distance between mobile and BS, traffic load and mobile velocity, where RSS is the most commonly used one. The conventional handover decision compares the RSS from the serving BS with that from one of the target BSs, using a constant handover threshold value (handover margin). The selection of this margin is crucial to handover performance. If the margin is too small, numerous unnecessary handovers may be processed. Conversely, the QoS may be low and calls may be dropped if the margin is too large. The fluctuations of signal strength associated with shadow fading cause a call sometimes to be repeatedly handed over back and forth between neighboring BSs, in what is called the “ping-pong effect”. Over recent years, many investigations have addressed handover algorithms for cellular communication systems. A local averaging technique removing fast fading component from the received signal strength was proposed in [1] to allow the conventional handoff decision reacting more quickly to corner effects. A timer-based hard handover algorithm was presented in [2] to prevent unnecessary handovers caused by fluctuations due to shadowing. A dynamic handover margin decision based on a traffic balancing rule was proposed in [3] to resize the cells according to the spatial variability of traffic. A speed-sensitive handover algorithm in a
hierarchical cellular system was described in [4], in which micro-cells serve the slowly-moving mobiles, and macro-cells serve fast-moving mobiles. In [5] and [6], RSS, mobile location and velocity were used as metrics for making handover decisions using fuzzy logic. A table lookup approach, proposed in [7], determines handover margins based on the mobile location, the mean of the intensity of the signal, and the standard deviation. Distance hysteresis for mitigating the effect of fading on handover performance was presented in [8]. Making handover decisions in various handover scenarios was presented in [9], in which a suitable handover decision is selected when mobile is located in an area with a pre-defined handover scenario. However, most handover algorithms presented in the literature based on information about mobile location suffer from a lack of practicability. The computational complexity of making a handover decision using fuzzy logic is excessive, and establishing and updating a lookup table to support a handover margin decision is time-consuming. The selection of a handover algorithm based on the handover scenario only succeeds in cases that the mobile environment is similar to one of the pre-classified environments, and involves complicated processes to define the handover scenarios. It also relies on an updated database when applied in a new mobile user environment. Furthermore, most studies assume that location of the mobile can be perfectly determined using the GPS (Global Positioning System), which is not available for most mobile telephones. In reality, the performances of available location estimators are far from that obtainable using GPS technique. This paper proposes an improved handover algorithm based on the estimates of mobile location (not using GPS) and velocity. The proposed handover algorithm was applied to a real GSM system in urban Taipei city. The signal level measured in decibels is the sum of two propagation terms - one associated with path loss and the other associated with shadow fading; fast fading is ignored because it is averaged out. The variation of the signal caused by shadow fading depends on the location and velocity of the mobile user. The proposed
algorithm outperforms the conventional method in making handover decisions for cellular systems by using location and velocity to identify the correlation among shadowing effects. Moreover, the computational complexity of the proposed algorithm is low and the algorithm does not employ a database or lookup table. The rest of this paper is organized as follows. Section Ⅱ describes the system model for analyzing handover performance and proposes the handover algorithm. Section Ⅲ examines the effect of location errors on handover performance. Section Ⅳ presents the performance verification using GSM measurements data. Section Ⅵ draws conclusions.
II.
PROPOSED HANDOVER ALGORITHM
a) System Model In a GSM system, when a mobile moves from BS1 to BS2, the signal strength is measured and reported to the network in a constant 480ms time interval [10] to support a handover decision. The signal level in decibels is the sum of two propagation terms, namely path loss and shadowing; fast fading is ignored because it is averaged out. Accordingly, the signal power levels received from BS1 and BS2 at discrete time instants t k k (where represents the time interval of 480ms within which the signal strength is measured), are given by P1[k ] and P2 [k ] , respectively.
P1[k ] m1[k ] u1[k ]
(1)
P2 [k ] m2 [k ] u 2 [k ]
(2)
where m1 and m2 are the received signal powers from BS1 and BS2, respectively, in terms of only path loss, and u1 and u 2 are respective the shadow fadings. The auto-correlation coefficient, ii , of the shadow fadings is commonly assumed to be an exponential function [11][12],
ii
E ui [k1 ]u i [k 2 ] exp( d d ), i 1, 2 i2
(3)
where i is the standard deviation of shadow fadings; d V k 2 k1 ( V is mobile velocity, non-negative number), and d is the decay distance (or correlation distance), which ranges from around 25 to 100m over urban, light urban, and suburban terrain [13]. The cross-correlation coefficient, ij , of shadow fadings is called the “site-to-site correlation” [14] and calculated as
ij
E ui [k ]u j [k ]
i j
, i, j 1, 2, i j
(4)
The correlation depends on 1) the angle between the two paths along the mobile to BS1 and BS2, and 2) the relative values of the two path lengths. Jay Weitzen et. al. verified that the shadow fading components are slightly correlated even at small angles [13]. b) Proposed Handover Algorithm Define P21[k ] as the difference between signal powers received from BS2 and BS1 at time index k:
P21[k ] P2 [k ] P1[k ] m2 [k ] m1[k ] u 2 [k ] u1[k ] m21[k ] u 21[k ]
(5)
where m21 represents the difference between signal powers received from BS2 and BS1 in terms only of path loss, u 21 represents the difference between the shadow fadings along the two paths. A handover from BS1 to BS2 occurs at time index k if the following two criteria are satisfied. Criterion 1: P21[k ] h Criterion 2: P21[k ] P21[k ] h where h is the handover margin, and is a positive non-zero integer, which needs to be carefully decided. In fact, criterion 1 is applied in making conventional handover decisions.
Because of shadowing, unnecessary handovers may be performed if a handover decision is based only on this criterion. Therefore, criterion 2 is imposed to improve the handover performance by determining whether path loss dominates the variation in the signal. Assume u 21[k ] and u 21[k ] are highly correlated, such that the correlation coefficient approaches unity; then, the difference between P21[k ] and P21[k ] can be approximated as P21[k ] P21[k ] m2 [k ] m2 [k ] m1[k ] m1[k ] m2(up ) m1( down )
(6)
where m2(up ) and m1( down) are the increase and degradation of the signal powers received from BS2 and BS1 in terms of path loss, due to motion of the mobile. Consequently, the difference between signal powers is always chiefly a function of path loss but not of shadow fading. Restate, the proposed algorithm ensures that the signal power received from target BS is h dB higher than that received from the serving BS (criterion 1), and that the difference between the signal powers is dominated by path loss associated with motion of the mobile (criterion 2). Hence, unnecessary handovers caused by fluctuations in shadow fadings are avoided. In the proposed algorithm, is critical to handover performance. The decided must guarantee high correlation between u 21[k ] and u 21[k ] , and sufficient space for signal variation caused by path loss. If is too large, criterion 2 is always met and does not support handover decisions. Conversely, the signal dose not vary if is too small. How to decide a suitable value of is explained below. Given u1[k ] , if the standard deviations of shadow fadings are assumed to be equal, such that 1 2 u , then u 2 [k ] , u1[k ] and
u 2 [k ] can be expressed as follow, based on the Gauss-Markov process. 2
u 2 [k ] 12u1[k ] 1 12 X 1
(7)
2
u1[k ] 11u1[k ] 1 11 X 2 2
u 2 [k ] 22u 2 [k ] 1 22 X 3
(8) (9)
where X 1 , X 2 and X 3 are identical independent Gaussian processes with zero-mean and variance u2 and Eu1[k ] X 1 Eu1[k ] X 2 Eu 2 [k ] X 3 0 . Assume 1) 12 21 c and 2) 11 22 a . The following can be proven. E u1[ k ]u 2 [ k ] Eu1[ k ]u2 [ k ] c u2
(10)
E u1[ k ]u1[ k ] Eu 2 [ k ]u 2 [ k ] a u2
(11)
E u1[ k ]u 2 [ k ] E u 2 [ k ]u1[ k ] c a u2
(12)
The correlation between u 21[k ] and u 21[k ] is
Eu 21 [k ]u 21 [k ] Eu 2 [k ] u1[k ]u 2 [k ] u1[k ] a (1 c )(2 u2 )
(13)
exp( V d )(1 c )(2 u2 ) The correlation coefficient between u 21[k ] and u 21[k ] must exceed a threshold, T ; that is, exp( V d )(1 c ) T , such that ln T 1 c
d V
(14)
Having describing the determination of , Fig. 1 displays the flowchart of the proposed handover decision. The mobile provides the measurement reports to network to support handover decisions within constant time intervals, . This data is buffered in the memory for the mobile location estimation proposed in [15]. The handover alarm is triggered, when the signal power received from the serving BS is below a threshold, then the availability of the target BS is verified according to criterion 1. If target BS meets criterion 1, the data buffered in the memory are fetched to estimate the location and velocity of the mobile, saving the overhead cost of calculating location since the mobile is not continuously tracked. The value
of can be obtained from mobile velocity using Eq. (14) to confirm criterion 2. Consequently, a handover occurs if the target BS satisfies criteria 1 and 2 simultaneously, otherwise the serving BS remains unchanged and the handover decision is made again at the next time.
Figure 1. Flowchart of proposed handover decision.
The results of the simulations using the proposed handover algorithm are compared with those obtained using the conventional method. A software package, SignalPro by EDX Engineering, was used to help the simulation. It includes a set of planning tools for wireless communication systems. Figure 2 reveals the simulation environment that covers an area of 1.6 x 1.4Km2. The solid blue trajectory from “A” to “B” represents a route through which the mobile moves. Polygons of different colors represent buildings of different heights. Seven BSs with omni-directional antennae are designated by encircled crosses ( ). The height of each BS is 35m and the mean and standard deviation of their transmitting power (EIRP) are 42.6dBm and 3.5dB, respectively. Walfisch-Ikegami model was applied to simulate path loss. It is a hybrid model, which considers diffraction down to street level and empirical correction
factors. The model was verified to predict accurately propagation path loss in urban areas with small cells [16]. Shadow fadings were simulated according to the model proposed in [17], where d 65 m, and c 0.1 . The mobile moved along the trajectory in Fig. 2 at a constant speed of 30Km/h. The sampling interval for reporting measurements was 0.48s. The handover alarm threshold, handover margin, and correlation threshold were set to –80dBm, 6dB and T 0.85 , respectively. Figure 3 is a typical comparison between the received signal time series obtained by the conventional method and that obtained by the proposed handover algorithm when the mobile moves along the beginning of the trajectory in Fig. 2. The top time series was simulated using the conventional method while the bottom one was simulated using the proposed algorithm. In the simulations, the mobile velocity was assumed to be perfectly estimated and the standard deviation of shadow fading was set to 9dB. The result indicates that the conventional method involves more handovers whereas the proposed algorithm prevents unnecessary handovers.
Figure 2. The simulation environment.
Figure 3. Comparison of signals received according to the conventional method (top plot) and those received according to the proposed handover algorithm (bottom plot).
III.
ANALYSIS OF HANDOVER PERFORMANCE WITH LOCATION ERRORS
Knowledge of the velocity of the mobile is required to determine the value of in the proposed algorithm. The GPS receiver is not yet available in mobile devices, so the effect of the estimation errors of velocity must be considered. The velocity of the mobile was estimated based on Doppler frequency shift in [18]. However, the estimated Doppler frequency is unreachable in most standards of the mobile cellular systems. This paper presents a means of estimating mobile velocity based on the mobile locations. For simplicity, the problem is reduced to the one-dimensional case. The mobile location estimate at time index k is modeled as Lˆ[ k ] L[ k ] n L
(15)
where L[k ] is the actual mobile location, and n L represents the location error, which is modeled as a zero-mean Gaussian process with variance L2 , as in [19]. Previous location information is used to estimate the current velocity. The size of the estimation window is M,
so the estimated locations
Lˆ[k ], Lˆ[k 1], , Lˆ[k M 1]
are used to estimate mobile
velocity. An adequate integer m (1 m M 2 and ( M m) is even) is chosen such that the mean
Lˆ[k ], , Lˆ[k (M m) 2 1]
Lˆ[ k ( M m) 4 0.5] ,
is used as a more accurate version of
which
is
represented
Lˆ[k (M m) 2], , Lˆ[k M 1]
is
used
as
Lˆ [i] ,
by a
more
and
accurate
the
mean
version
of
Lˆ[ k (3M m) 4 0.5] , which is represented by Lˆ [ j ] . Then, the estimated mobile velocity
at time index k has the form
Vˆ vˆ[k ]
Lˆ [i] Lˆ [ j ] m 1
v[k ] nv
(16)
where nv is the error in the estimated velocity and is also a zero-mean Gaussian process with variance v2 [ 4 ( M 2 m 2 )] 2 ( M m) L2 . Given suitably chosen M and m, the estimate of mobile velocity is highly accurate. As a simple example, M=7 and m=3 are chosen, as presented in Fig. 4, where Lˆ [3] and Lˆ [5] are
(17)
(18)
Vˆ vˆ[7] Lˆ [5] Lˆ [3] 2 v[7] nv
(19)
Lˆ [3] Lˆ[1] Lˆ[2] Lˆ[3] Lˆ[4] Lˆ[5] / 5 Lˆ [5] Lˆ[3] Lˆ[4] Lˆ[5] Lˆ[6] Lˆ[7] / 5 The estimated mobile velocity is
where the variance of nv is (1 / 25) L2 .
Figure 4. An example of estimation of mobile velocity from location. (M=7, m=3)
However, the mobile location is a two-dimensional problem in reality. The estimates of location on the horizontal-axis and the vertical-axis at time index k are respectively expressed as
xˆ[k ] x[k ] n L( x ) ( y) yˆ[k ] y[k ] n L
(20)
where x[k ] and y[k ] are the actual mobile location of the mobile, and n L( x ) and n L( y ) represent the location errors, which are modeled as zero-mean Gaussian processes with variance L2 . Similar to Eq. (16), the estimated velocity at time index k is Vˆ
vˆ x [k ]2 vˆ y [k ]2
(21)
where vˆ x [k ] and vˆ y [k ] are the estimations of velocity on the horizontal-axis and the vertical-axis, respectively. Denote the actual velocity of the mobile as V and assume the variances of the error terms, in vˆx [k ] and vˆ x [k ] , are the same to v2 , the probability density function (p.d.f.) of Vˆ is a Rice distribution with Rice factor K V 2 (2 v2 ) [14][20],
Vˆ 2 K Vˆ 2 Vˆ fVˆ (Vˆ ) 2 exp exp K I 0 2 v 2 v v
(22)
where I 0 ( z ) is a modified Bessel function of the first kind and zeroth-order. For a Rice distribution, a larger K which corresponds to a faster mobile or a lower v yields a more accurate estimate of velocity because the p.d.f. curve is sharper. Moreover, redefine Eq. (14) as ˆ ln T (1 c ) d (Vˆ ) , ˆ distributes as
f ˆ (ˆ) 2 f Vˆ ˆ ˆ
(23)
where ln T (1 c ) d . Figure 5 plots the p.d.f. curves of ˆ given various location errors, which indicate the 67th percentiles of errors. Given the parameter settings
ρ
c
0.1, ρT 0.85, d 65m, τ 0.48s , mobile velocity 30 Km/h , the actual is 0.929.
The accuracy of the estimate of is very high because 1) ˆ is run off during handover decisions, and 2) ˆ is a positive non-zero integer, which resulting in ˆ 1 with very high probability.
Figure 5. Probability density of ˆ given different location errors.
Number of handovers and handover delay (referred also as crossover point) are often used as performance measures [1][2][17]. In our study, however, signal outage probability, defined as the rate that the received signal from serving BS is less than a threshold, is substituted for handover delay because the cell shape of each BS is an irregular polygon instead of a regular hexagonal structure, which resulting in difficulty in defining handover delay. Meanwhile, a larger signal outage probability implies a larger handover delay. With reference to the same case as simulated in Section Ⅱ, Fig. 6 compares the performances of the proposed and the
conventional method averaged over 2000 iterations. In this case, the standard deviations of shadow fadings are set to 3, 6 and 9dB, and the location errors are set to 0, 30, 60 and 90m. In estimating the velocity, M and m are chosen as 9 and 3. The horizontal and vertical axes in Fig. 6 respectively represent the total number of handovers and the signal outage probability associated with the outage threshold of –95dBm. The signal outage probability is obtained by recognizing how much percentage of signal received from serving BSs is below the outage threshold when mobile moves along a testing route. Symbols “○”, “＋” and “□” indicate the performance in 3, 6 and 9dB shadow fading environments, respectively. The solid blue line and the dotted red line represent the results obtained using the conventional method and the proposed algorithm, respectively. Using the proposed algorithm reduces the number of handovers and only slightly increases in signal outage probability. The handover performance improvement measured as the decrease in the number of handovers number over that obtained using the conventional method ranges from 9 to 17%.
Figure 6. Comparison of handover performance.
IV.
VERIFYING PERFORMANCE USING GSM MEASUREMENTS DATA
The proposed handover algorithm was applied to a real GSM system (1800MHz) in urban Taipei city. GSM provides measurement reports for managing radio resources. These reports include the received signal levels, RXLEV, of the serving cell and of the strongest neighboring cells. These values of RXLEV are reported from a mobile within a period of 0.48s. “TEMS Investigation GSM”, provided by Ericsson, was used to make the measurements along three selected routes. Calls were made from a car, driving along the routes plotted in Fig. 7 - route 1 from A to B, route 2 from C to D and route 3 from E to F. The area where measurements were made was 2.1 x 1.6 Km2. The blue triangles indicate sector cells with an averaged cell radius of around 330m. The mean and standard deviation of
building heights are 20.3m and 14.4m, respectively. The average and standard deviation of BS heights are 26.4m and 10.2m, respectively. As the measurements are being made, the GPS coordinates of the mobiles, and the power levels received from serving cell and the strongest neighboring cells are recorded over a period of 0.48s.
Figure 7. Measurement Environment.
Before the handover performance can be evaluated, the propagation characteristics (shadowing components) must be investigated. This work is down on a point-by-point basis. The first step is to remove the path loss (local-mean) to leave the desired shadowing component. The local-mean of each data point was determined from the ensemble of signal points located within a circle with a radius 100m. Figure 8 plots the p.d.f. of the residual shadowing components. A zero-mean Gaussian with a standard deviation 5.55 is plotted as a blue line in this figure, to ensure the shadowing components exhibit Gaussian properties. The second step is to estimate the cross-correlation coefficient of shadowing fadings. The signals received from two BSs are grouped or binned according to the angle between the two paths from the mobile to the BSs. A group must include at least 25 samples from each BS to be considered statistically significant and used in the analysis. Each bin encompasses 10 degrees
of the 180 o angular range. Figure 9 presents the average cross-correlation coefficient of the shadowing fadings, which are slightly correlated even at small angles. The final step is to estimate the correlation distance of shadowing fadings. The correlation distances are computed based the measurements from each single BS. Figure 10 presents the average auto-correlation plot, on which the red line is the fitted curve with a correlation distance 64.97m.
Figure 8. Probability density function of the shadowing components extracted from measurements.
Figure 9. Average cross-correlation coefficient of shadowing fadings.
Figure 10. Average auto-correlation coefficient of shadow fadings. The red line is the fitted curve with correlation distance 64.97 m.
The measurements data were applied for simulation of the proposed handover algorithm. The proposed algorithm estimates the mobile location based on the differences among signal attenuations, as proposed in [15]. The differences among signal attenuations were used to determine circles that encompass possible mobile locations. The actual mobile location was then estimated from the intersections of these circles. Assuming the GSP coordinates as the actual mobile position, Table Ⅰ presents the statistics on location errors, of which 67% are less than 149.5m, 194.6m and 142.1m for route 1, route 2 and rout 3, respectively [15]. In the evaluation of handover performance, the correlation distance and cross-correlation coefficient of shadow fadings are set to d 65 m and c 0.1 , respectively. The handover alarm threshold, handover margin, signal outage threshold and correlation threshold are respectively set to –70dBm, h 6 dB, -80dBm and T 0.85 . Table Ⅱ compares the handover performance of the proposed algorithm and the conventional method. The results indicate that the proposed handover algorithm reduces the number of handovers and only slightly increases the signal outage probability. The improvement over the conventional
method in the number of handovers, which indicates performance, ranges from 18 to 26 %. Table Ⅰ Summary Statistics of Location Errors (m) In Real System [15] Measurement Routes Route 1 Route 2 Route 3 (1246 pts.) (1647 pts.) (2033 pts.) mean 128.0 157.6 135.9 std 79.2 101.2 73.1 Proposed median 114.6 129.8 122.2 Method 67% 149.5 194.6 142.1 98% 333.8 433.7 326.4 th th Rows 67% and 98% represent the 67 and 98 percentiles, respectively. Table Ⅱ Handover performance of proposed algorithm and conventional method in a real environment. Conventional Proposed Method Method Route 1
Handover Number
35
26
Signal Outage Probability (%)
0.0
0.16
Handover Number
38
31
1.34
1.58
92
68
0.94
1.08
Route 2
Signal Outage Probability Handover Number Route 3 Signal Outage Probability (%)
V.
CONCLUSIONS
An improved handover algorithm is proposed, it is based on the estimated location and velocity of the mobile user to suppress the ping-pong effect in cellular systems. The computational complexity of the proposed algorithm is low and no database or lookup table is required. A method based on non-GPS location techniques is presented to estimate the velocity of the mobile user. The effect of location errors on handover performance was examined. The performance of the proposed handover algorithm is compared with that of the conventional method in a scenario with reasonable location errors. The simulations indicate
that the number of handovers is reduced and that the signal outage probability is only slightly increased. The handover performance of the proposed method, which uses a practical means of estimating the location of a mobile instead of GPS techniques, is 18 to 26% better, in terms of number of handovers, than that of the conventional method, while the signal outage probability remains similar, indicating similar quality of services.
REFERENCES [1] A. E. Leu and B. L. Mark, “Modeling and Analysis of Fast Handoff Algorithms for Microcellular Networks,” in Proc. 10th IEEE Symp. Modeling, Analysis and Simulation of Computer and Telecommun. Syst., Oct. 2002, pp. 321-328. [2] A. E. Leu and B. L. Mark, “An Efficient Timer-Based Hard Handoff Algorithm for Cellular Networks,” in Proc. IEEE Wireless Commun. and Networking Conf., vol. 2, Mar. 2003, pp. 1207-1212. [3] M. Toril, S. Pedraza, R. Ferrer, and V. Wille, “Optimization of Handover Margins in GSM/GPRS Networks,” in Proc. IEEE Veh. Technol. Conf., vol. 1, Apr. 2003, pp. 150-154. [4] D. H. Kim, W. Noh and S. An, “Speed-Sensitive Handoff under Dynamic Hierarchical Cellular Systems,” The 9th Int’l conf. Comput., Commun. and Contr. Technol., CCCT ’03, July 2003. [5] B. Honman and W. Benjapolakul, “A Handover Decision Procedure for Mobile Telephone Systems Using Fuzzy Logic,” in Proc. IEEE Asia-Pacific Conf. Circuits Sys., Nov. 1998, pp. 503-506. [6] A. Majlesi, and B. H. Khalaj, “An Adaptive Fuzzy Logic Based Handoff Algorithm for Hybrid Networks,” in Proc. IEEE Conf. Signal Processing, vol. 2, Aug. 2002, pp. 1223-1228. [7] S. S. Wang, M. Green and M. Malkawi, “Adaptive Handoff Method Using Mobile Location Information,” IEEE Emerging Technol. Symp. Broadband Commun. for the Internet Era Symp. digest, Sept. 2001, pp. 97-101. [8] W. Teerapabkajorndet and P. Krishnamurthy, “Comparison of Performance of Location-Aware and Traditional Handoff-Decision Algorithms in CDPD Networks,” IEEE Conf. Veh. Technol., vol. 1, pp. 212-216, 2001.
[9] A. Markopoulos, P. Pissaris, S. Kyriazakos, Ch. Dimitriadis, G. Karetsos and E.D. Sykas, “Increased Handover Performance in 2G and 3G Wireless Systems Based on Combined Mobile-Location and Area,” in Proc. IEEE Int’l Symp. Wireless Personal Multimedia Commun., vol. 1, Oct. 2002, pp. 47-51. [10] Jorg Eberspacher and Hans-Jorg Vogel, “GSM-Switching, Services and Protocols,” John Wiley & Sons, 1998. [11] M. Gudmundson, “Correlation Model for Shadow Fading in Mobile Radio Systems,” Electronics Letters, vol. 27, no. 23, pp. 2145-2146, Nov. 1991. [12] D. Giancristofaro, “Correlation Model for Shadow Fading in Mobile Radio Channels,” Electronics Letters, vol. 32, pp. 958-959, May. 1996. [13] Jay Weitzen and Terri J. Lowe, “Measurement of Angular and Distance Correlation Properties of Long-Normal Shadowing at 1900 MHz and Its Application to Design of PCS Systems,” IEEE Trans. Veh. Technol., vol. 51, no. 2, Mar. 2002. [14] S. R. Saunders, “Antennas and Propagation for Wireless Communication Systems,” John Wiley & Sons, 1999. [15] D. B. Lin, R. T. Juang, H. P. Lin and C.Y. Ke, “Mobile Location Estimation Based on Differences of Signal Attenuations for GSM Systems,” in Proc. IEEE Society Int’l Conf. Antennas and Propagat., vol. 1, Jun. 2003, pp. 77-80. [16] H. P. Lin, D.B. Lin and R. T. Juang, “Performance Enhancement for Microcell Planning Using Simple Genetic Algorithm”, in Proc. IEEE Int’l Conf. Antennas and Propagat., vol. 4, Jun. 2002, pp. 664-667. [17] F. Graziosi, M. Pratesi, M. Ruggieri and F. Santucci, “A Multicell Model of Handover Initiation in Mobile Cellular Networks,” IEEE Trans. Veh. Technol., vol. 48, no. 3, May. 1999. [18] G. Azemi, B. Senabji and B. Boashash, “A Novel Estimator for the Velocity of a Mobile Station in a Micro-Cellular System,” in Proc., Int’l Symp. Circuits and Sys., vol. 2, May 2003, pp. 212-215. [19] K. I. Itoh, S. Watanabe, J. S. Shih, and T. Sato, “Performance of Handoff Algorithm Based on Distance and RSSI Measurements,” IEEE Trans. Veh. Technol., vol. 51, no. 6, Nov. 2002. [20] S. O. Rice, “Statistical Properties of a Sine Wave plus Random Noise,” Bell Systems Technical Journal, vol. 27, pp 109-157, Jan. 1948.
