Comparing soft and hard handoffs

Comparing Soft and Hard Handoffs

Yi-Bing Lin, Senior Member, IEEE, and Ai-Chun Pang

Abstract—This paper studies the soft-handoff mechanism and

compares its performance with hard handoff. Our study indicates that although a handset may potentially consume extra radio links in soft handoff, the mechanism provides better opportunity to transfer the link successfully in the handoff procedure. Thus, by carefully planning the overlay areas of cells, soft handoff can outperform hard handoff.

Index Terms—Hard handoff, personal communications services,

radio channel allocation, soft handoff.

I. INTRODUCTION

I

area of a BS), the destination (or the originating) handset is

con-nected if a channel is available. Otherwise, the call is blocked (this is referred to as a new call blocking). When a communi-cating handset moves from one cell to another, the channel in the old BS is released and a channel is required in the new BS. This process is called handoff. In mobile systems such as AMPS [1], global system for mobile communication (GSM) without macrodiversity [2], DECT [3], D-AMPS [4], and PHS [5], hard

handoff is employed [6], [7]. In hard handoff, the old radio link

is broken before the new radio link is established, and a handset always communicates with one BS at any given time. In the handoff procedure, the network needs to set up the new voice path for the handoff call. This setup time is referred to as the

network response time If the old radio link is disconnected before the network completes the setup, the call is forced termi-nated. Thus, even if idle channels are available in the new cell, a handoff call may fail if the network response time for link transfer is too long. Note that a handoff failure may not neces-sarily cause a call drop. It is normally some time-out mechanism for the voice or signaling path which leads to a dropped call.

Some code-division multiple-access (CDMA) systems [8] and GSM with macrodiversity [2] utilize soft handoff where a handset may communicate with the outside world using multiple radio links through different BS’s at the same time. During handoff, the signaling and voice information from multiple BS’s are typically combined (or bridged) at the mobile switching center [9]. Similarly, voice and signaling information must be sent to multiple BS’s, and the mobile station must combine the results. In some soft-handoff systems, a handset may connect up to three or four radio links at the same time.

Manuscript received January 31, 1998; revised April 8, 1999. This work was supported in part by the National Science Council, R.O.C., under Contracts NSC-87-2213-E-009-013 and NSC88-2213-E009-079.

The authors are with the Department of Computer Science and Information Engineering, National Chiao Tung University, Hsinchu, Taiwan, R.O.C. (e-mail: liny@csie.nctu.edu.tw).

Publisher Item Identifier S 0018-9545(00)03667-7.

Fig. 1. The timing diagram for the hard-handoff model.

Thus, within the overlay area of cells, a handset can connect to multiple BS’s. During the process of dropping a failing link, the handset may communicate using other radio links. Thus, link transfer is not sensitive to the elapsed link-transfer time. Note that the soft-handoff link-transfer procedure may not be faster than that for hard handoff. However, soft handoff is not time critical as compared with hard handoff [10].

On the other hand, soft handoff degrades channel availability because a handset may consume multiple radio channels. Thus, it is desirable to investigate the performance of soft handoff and the tradeoff between hard and soft handoffs. In this paper, we propose analytic and simulation models to study soft handoff and compare its performance with hard handoff. To strengthen the theme of our study, we do not consider the handoff priori-tized schemes [11] that are seldom implemented in the commer-cial systems.

II. ANALYTICMODEL

With minor modifications to the two analytic models we developed in [12] and [11], we compare the performance for hard and soft handoffs. For the reader’s benefit, we reiterate the models in [12] (for hard handoff) and [11] (for soft handoff) with new notation and new interpretation.

A. The Hard-Handoff Model

Fig. 1 illustrates the timing diagram for the hard-handoff model. In this figure, represents the time that a handset can receive the signal from cell (i.e., the time that the handset resides in cell Since the cells may overlay, the handset will enter the overlay area before it moves from cell to cell Let be the overlay time. Then can be expressed as , where is the time that the handset stays in the nonoverlay area of cell In hard handoff, a communicating handset is switched from cell to cell at some point within

In Fig. 1, the handoffs occur at time

From the viewpoint of the hard-handoff scheme, the residence time of the handset at cell is Let be the

nonoverlay period. If and then

(1) 0018–9545/00$10.00 © 2000 IEEE

the model.

• The call arrivals to/from a handset are a Poisson process. The net new call arrival rate to a cell is

• The mobile residence time in a cell has an exponential distribution with the density function

This assumption will be relaxed to accommodate general residence time distribution in Appendix A.

• The call holding time is exponentially distributed with the mean

The output measures are:

handoff call arrival rate to a cell; new call blocking probability;

probability that a handoff call is blocked because no radio channel is available;

probability that a hard-handoff call is blocked because the network response time is too long;

forced termination probability or the probability that a handoff call is blocked because no radio channel is avail-able or because the network response time is too long; call incompletion probability.

As mentioned before, a handoff call is forced terminated if the network response time is too long (with probability or no channel is available (with probability Since a nonprioritized scheme is considered, and

(2) From [12], we have

(3) The channel occupancy time of a call in a cell is the minimum of the remaining call holding time (note that the call holding time for a handoff call has the same distribution as a new call because of the memoryless property of the exponential distribution) and the remaining cell residence time. Thus, the channel occupancy time is also exponentially distributed with rate The net traffic to the system is Let be the number of chan-nels in a cell. The hard-handoff scheme can be modeled by an

system and from the Erlang-B formula

(4)

The probability can be obtained by assigning an initial value for and by iterating (4) and (3) until the value converges. From [12], the call incompletion probability is derived as

(5)

Fig. 2. The timing diagram for the soft-handoff model.

B. The Soft-Handoff Model

For the demonstration purpose, we assume that a handset can connect up to two radio links in a CDMA system. Fig. 2 illus-trates the timing diagram for the soft-handoff model. The nota-tions and are the same as that in Fig. 1. In soft handoff, a communicating handset at cell utilizes one channel during the nonoverlay period and is looking for a second radio link from cell during Suppose that the second link is found at time then the channel occupancy time of the handset at cell is the minimum of (in Fig. 2) and the remaining call holding time. Assume that is exponentially distributed, then from the memoryless property, also has the same distribution as , i.e., it is exponentially distributed with mean

(nonexponential are considered in Appendix A). For a fixed period, the number of cells visited by a handset is independent of the handoff schemes and the moving rate of a handset in soft handoff is as expressed in (1). Let be the probability that a soft-handoff call is blocked because the network response time is too long. Unlike the hard handoff, it is apparent that

in this scheme. Following the same reasoning in the previous section, and for soft handoff are similar to (2), (3), and (5) and can be expressed as

(6)

(7) To compute and the soft-handoff scheme can be modeled by a Markov process with states where represents the number of busy channels as in [11]. Fig. 3 illustrates the Markov process. When the process is in state (for

channels are busy. The effective call traffic to a cell at

is [and the process moves from to with

this rate]. Since a busy channel is released with the rate

the process moves from to (for with

the rate

When the process is in where all channels are busy, and handoff calls are looking for the second links. When a call arrives at state the call is dropped immediately if it is a new call. On the other hand, if the call is a handoff call, then it is trying to connect to the second link before it leaves the overlay area. Thus, the process moves from to

Fig. 3. The Markov chain.

with rate for Since all channels are busy, the first completion (among the connected calls) releases its channel with rate For those handoff calls who look for the second links, before the second links are available, the calls may leave the system in two cases: either the handset leaves the overlay area (with rate and is forced terminated or the call is completed (with rate Thus, the first such call leaves the system with the rate and the process moves from

to with rate for

Let be the steady-state probability for Then

Since we have

Since a new call is blocked when the system is in state (where at its arrival, the originating call blocking prob-ability is

(8)

Following the technique we developed in [11], the probability is derived as follows. Suppose that a handoff call arrives at time when the cell is in state (where

and the call leaves the overlay area at time Let be the remaining call holding time of at time (i.e., the call will be completed at time From the memoryless property, has the same exponential distribution as Consider the outstanding calls that arrive at the cell earlier than Suppose that among these calls, the first call leaves the system (either completes, expires, or leaves the cell) at time Then the density function for is

(9)

If then at time sees handsets in

conver-sations and handoff calls looking for the second links. Now consider the first call that leaves the system among these

TABLE I

THEPROBABILITY  FOR VARIOUS 

( = 0:5 ;  = 100 )

calls (excluding Suppose that the call leaves the system at time Because of the memoryless prop-erty of the call occupancy distribution and the overlay time dis-tribution, has the density distribution as expressed in

(9). Let For a call arriving at state

the probability that is blocked is and

Thus, the probability (that no radio resource is available for a handoff call) is

and

(10)

By using the same iterative procedure described in the previous sections, and can be obtained.

C. Derivation for and

Suppose that and are exponentially distributed with rates and respectively (nonexponential distributions will be considered in Appendix A). In soft handoff, let be the pe-riod between the handset connects to the new cell and when the handset leaves the overlay area. Then from the Markov model in Fig. 3, and have the same exponential distribution. Thus, we have

Fig. 4. Comparing the analytic and the simulation results.

Fig. 5. The effect of the overlay time on soft handoff( = 0:5 ;  = 100 ):

Note that is independent of and Table I lists the values obtained from simulation experiments (described in

Fig. 6. The effect of the nonoverlay period [ = 6 ; E[t ] = 0:01  (exponential)].

Fig. 7. The effect of the variance of the nonoverlay period(E[t ] = 0:01 ; E[ ] = 0:05 ; and  = 6 ):

Appendix A) and the values computed from (11). The table in-dicates that (11) is consistent with the simulation experiments.

For hard handoff, the handoff procedure is initiated when the signal of the new link is better than the old link. Thus, we assume

that and

(12)

The analytic model is validated against a simulation model de-scribed in Appendix A.

Fig. 4 plots the curves obtained from the analytic model (the dashed curves) and the simulation model (the solid curves). The figure indicates that the analytic and the simulation results are consistent.

Fig. 8. The effect of network response time( = 6 ;  = ):

III. NUMERICALEXAMPLES

This section uses some numerical examples to illustrate the effects of the overlay time the mobility and the network response time on output measures such as and

1) The Effect of the Overlay Time: Fig. 5 illustrates the

soft-handoff output measures and as functions of the ex-ponential overlay time where and the network response time is exponentially distributed with mean 0.01/ Fig. 5(b) shows that decreases as increases (the longer the overlay time, the higher the probability that the second radio link is successfully connected to the handset). Fig. 5(a) shows that increases as increases (since handoff calls have better op-portunity to obtain radio channels as the overlay time increases, the new call attempts are more likely to be blocked). Fig. 5(c) plots the curves. We first note that [in Fig. 5(a)] and [in Fig. 5(b)] are two major factors [see (7)] that determine For slightly increases as increases. On the other hand, significantly decreases as increases. Thus, de-creases as the overlay time inde-creases. On the other hand, when the offered load is large (e.g., ), significantly in-creases as increases. Since significantly decreases as increases, the net effect is that decreases then increases as

increases.

2) The Effect of the Nonoverlay Period: Fig. 6 plots

against the nonoverlay period where

and This figure indicates that is more

sensitive to for large than small In other words, when the user mobility is large, the cell overlay area layout significantly affects the performance of soft handoff. Fig. 7 plots against the variance of with the normal distribution. We ob-serve that when is insensitive to the change of the variance of the nonoverlay period. On the other hand, when

is very sensitive to

3) The Effect of Network Response Time: Fig. 8 plots

as a function of the network response time In this figure, and Fig. 8(a) shows the effect of the exponen-tial with various mean values. If the network response time is zero, then and handoff always fails. In this case, increases as the overlay time increases. However, when is nonzero, decreases then increases as increases [this phe-nomenon was explained in Fig. 5(a)]. The effect of on is similar to the effect of That is, is more sensitive to

for large than small Fig. 8(b) demonstrates how the variance of with the normal distribution

af-fects the system performance, where and

The curves indicate that decreases as decreases. When the variance of the

Fig. 9. Soft versus hard handoff.

distribution only has an insignificant effect on On the

other hand, when significantly decreases

as decreases.

4) Soft Versus Hard Handoffs: Fig. 9 compares the call

in-completion probabilities for soft handoff (the solid curves) and for hard handoff (the dashed curves). In this figure, (11) and (12) are used to compute and respectively. The figure indicates that in the ranges of the input parameters we consid-ered, soft handoff outperforms hard handoff.

IV. CONCLUSION

This paper proposed an analytic model and a simulation model to study the performance of soft handoff. Our study indicated that the handoff network response time, the mobility of the user and the overlay time significantly affect the per-formance of soft handoff. Furthermore, we observed that the call incompletion probability can be significantly affected by the variances of the network response time and the nonoverlay time. Under the ranges of the input parameters we considered, soft handoff may significantly outperform hard handoff. Our study provides guidelines to determine the degree of the overlay among cells.

APPENDIX I THESIMULATIONMODEL

This Appendix describes a simulation model to investigate the performance of CDMA soft handoff. In the simulation ex-periments, the PCS system consists of 64 BS’s connected as an 8 8 wrapped mesh [13]. In the simulation model, a handset resides in the nonoverlay area of a cell for a period , then moves to the overlay area of one of the four neighboring cells (selected with equal routing probabilities) for a period , and finally moves to the nonoverlay area of the new cell. In the sim-ulation, a cell is modeled as acellobject with data structure

C(i)to represent the number of idle channels at the cell.

In this case,C(i)is decremented by one, and the call holding time and the nonoverlay time for this call are generated. If then generate a

COMPLETION event with timestampclock Oth-erwise, generate an OVERLAY event with timestamp

clock (with the destination cell where the handset is moving, and the overlay period

The call is blocked. Update the call blocking statistics and

The simulation generates the next ARRIVAL event ac-cording to the call arrival rate

• OVERLAY event represents that a handset in conversa-tion moves into the overlay area between the old cell and the new cell

(the soft-handoff procedure is exercised). Let be the remaining call holding time and be the overlay period. DecrementC(j)by one. If , then generate a COMPLETION event with timestamp

clock Otherwise, generate the handoff network response time Generate a RELEASE event with timestampclock

The handset should continue to try until a radio channel in the new cell is available. This part can be implemented by generating another OVERLAY event if the handset is still in the overlay area at the next try time, or a RELEASE event if the next try occurs after the handset leaves the overlay area is set to indicate soft failure due to shortage of radio channel). • RELEASE event represents various situations described

below.

: The call is forced terminated because no radio channel is available.C(i)is incremented by one, and the output statistics and are updated.

: The call is forced terminated because long network response time. C(i) and C(j) are incremented by one, and the output statistics and

are updated.

: The second link is successfully connected. Let be the remaining call holding time. Generate the nonoverlay period at cell If then generate a COMPLETION event with timestamp clock Otherwise, generate an OVERLAY event as described before.

• COMPLETION event represents the completion of a call at cell If the completion occurs at the nonoverlay area of cell thenC(i)is incremented by one. If the completion occurs at the overlay area between cell and cell then bothC(i)andC(j)are incremented by one.

In the simulation experiments, the PCS system consists of 64 BS’s connected as an 8 8 wrapped mesh [13]. The call arrivals to a cell form a Poisson process with arrival rate The call holding times are exponentially distributed with mean 1/ The periods and are generated from exponential or

normal random number generators. In the experiments, every BS has ten channels. In each simulation experiment, 500 000 incoming calls are simulated to ensure that the simulation results are stable.

ACKNOWLEDGMENT

The authors would like to thank the anonymous reviewers for their valuable comments.

REFERENCES

[1] “Mobile station–land station compatibility specification,” ANSI/EIA/TIA Tech. Rep. 553, 1989.

[2] M. D. Yacoub, Foundations of Mobile Radio Engineering. Boca Raton, FL: CRC, 1993.

[3] “Digital European cordless telecommunications services and facilities requirements specification,” ETSI DI/RES Tech. Rep. 3002, 1991. [4] “Cellular system dual-mode mobile station–base station compatibility

standard,” EIA/TIA Tech. Rep. IS-54, 1992.

[5] T. Kobayashi, “Development of personal handy-phone system,” in

ITS’94.

[6] W. C. Y. Lee, Mobile Cellular Telecommunications Systems. New York: McGraw-Hill, 1995.

[7] B. Jabbari, G. Colombo, A. Nakajima, and J. Kulkarni, “Network issues for wireless communications,” IEEE Commun. Mag., Jan. 1995. [8] “Mobile station–base station compatibility standard for dual-mode

wideband spread spectrum cellular system,” EIA/TIA Tech. Rep. IS-95, 1993.

[9] V. K. Garg, K. F. Smolik, and J. E. Wilkes, Applications of CDMA in

Wireless/Personal Communications. Englewood Cliffs, NJ: Prentice-Hall, 1997.

[10] A. R. Noerpel and Y.-B. Lin, “Handover management for a PCS net-work,” IEEE Personal Commun. Mag., vol. 4, no. 6, pp. 18–26, 1997. [11] Y.-B. Lin, S. Mohan, and A. Noerpel, “Queueing priority channel

as-signment strategies for handoff and initial access for a PCS network,”

IEEE Trans. Veh. Technol., vol. 43, no. 3, pp. 704–712, 1994.

[12] Y.-B. Lin, “Impact of PCS handoff response time,” IEEE Commun. Lett., vol. 1, no. 6, pp. 160–162, 1997.

[13] Y.-B. Lin, Y.-J. Lin, and V. W. K. Mak, “Allocating resources for soft requests—A performance study,” Information Sciences, vol. 85, no. 1, pp. 39–65, 1995.

Yi-Bing Lin (S’80–M’96–SM’96) received the

B.S.E.E. degree from National Cheng Kung Uni-versity, Taiwan, R.O.C., in 1983 and the Ph.D. degree in computer science from the University of Washington, Seattle, in 1990.

From 1990 to 1995, he was with the Applied Research Area at Bell Communications Research (Bellcore), Morristown, NJ. In 1995, he was appointed Professor at the Department of Computer Science and Information Engineering (CSIE), National Chiao Tung University (NCTU), Hsinchu, Taiwan. In 1996, he was appointed Deputy Director of the Microelectronics and Information Systems Research Center, NCTU. Since 1997, he has been Chairman of the CSIE, NCTU. His current research interests include design and analysis of a personal communications services network, mobile computing, distributed simulation, and performance modeling. He was a Guest Editor for the IEEE TRANSACTIONS ONCOMPUTERSSpecial Issue on Mobile Computing. He is an Associate Editor of IEEE NETWORKS.

Dr. Lin is a Subject Area Editor of the Journal of Parallel and Distributed

Computing, an Associate Editor of the International Journal in Computer Sim-ulation, an Associate Editor of SIMULATION Magazine, an Area Editor of ACM Mobile Computing and Communication Review, a Columnist of ACM Simula-tion Digest, a Member of the Editorial Board of the InternaSimula-tional Journal of Communications Systems, a Member of the Editorial Board of ACM/Baltzer Wireless Networks, a Member of the Editorial Board of Computer Simulation Modeling and Analysis, and Guest Editor for the ACM/Baltzer MONET Special

Issue on Personal Communications. He was the Program Chair for the 8th Work-shop on Distributed and Parallel Simulation, General Chair for the 9th WorkWork-shop on Distributed and Parallel Simulation, Program Chair for the 2nd International Mobile Computing Conference, and Publicity Chair of ACM Sigmobile.

Ai-Chun Pang received the B.S.C.S.I.E. and

M.S.C.S.I.E. degrees from National Chiao Tung University (NCTU), Hsinchu, Taiwan, R.O.C., in 1996 and 1998, respectively. She is currently working toward the Ph.D. degree at NCTU.

Her current research interests include personal communications services, computer telephony integration, and mobile computing.