12
IEEE TRANSACTIONS ON VEHICULAR TECHOLOGY, VOL. 49, NO. 3, MAY 2000 1017 Efficient Interactive Call Admission Control in Power-Controlled Mobile Systems Dongwoo Kim, Member, IEEE Abstract—When a mobile newly arrives and seeks admission in power-controlled cellular/personal communications services (PCS) systems, interactive call admission control guides the evolution of the power transmitted by the new mobile in order to protect the transmission quality of ongoing calls from dropping below a desired level. In addition, it eventually decides whether the new mobile should be accepted or rejected without errors. However, existing interactive call admission algorithms incur high network signaling costs and suffer from slow speed in the “accept/reject” decision, which often renders the implementation of these algorithms impractical. This paper presents efficient interactive algorithms that improve the admission speed as well as curtail the network coordination cost. The algorithms are numerically evaluated with respect to the admission speed, decision error, and performance of protecting ongoing call quality. The simulation result shows that the proposed algorithms are promising for practical implementation. Index Terms—Cellular/PCS mobile systems, mobile call admis- sion control, power control. I. INTRODUCTION I N PRESENT cellular and personal communications services (PCS) systems, both cell size and frequency reuse distance are getting smaller, and as a result, the receivers are exposed to more severe noise compared to conventional cellular systems. In order to provide each mobile an acceptable connection in such interfering situations, many power control algorithms have fo- cused on limiting the interference seen by other mobiles [1]–[5]. The power control algorithms converge to an effective power vector that attains predetermined carrier-to-interference (CIR) targets at the receivers, if such a power vector exists. Otherwise, certain mobiles cannot achieve their target CIR and are conse- quently suffering severe deterioration in transmission quality. In the latter case, unnecessary interference is produced and some system performance might be lost. The admission control of a newly arrived mobile, which we study here, is considered as a practical means of preventing this latter case where not all mo- biles can be supported. When a new mobile requires a channel, the admission control has to tradeoff between two types of errors: type I error where the mobile is erroneously accepted and the congestion occurs and type II error where the mobile is erroneously rejected while it can be supported along with other active mobiles. For our pur- pose, type I error should be minimized. In addition, even though Manuscript received March 31, 1998; revised July 20, 1999. The author is with the School of Electrical Engineering and Com- puter Science, Hanyang University, Kyunggi 425-791, Korea (e-mail: [email protected]). Publisher Item Identifier S 0018-9545(00)03694-X. the mobile would be properly accepted, the admission control should guide the evolution of the power level transmitted by the new mobile. Otherwise, the interference introduced by the new mobile usually results in dropping transmission quality of on- going calls for the time period that is needed by the transmitters to “reconverge” to a new effective power vector. A centralized linear/combinatorial programming approach could provide an optimal admission decision as well as ade- quate power levels letting all mobiles meet their CIR targets [6]–[8]. Its practical implementation, however, is normally prohibited since it requires to gather and process a large amount of data including all link-gain information in the system. Thus, in designing the proposed call admission algorithm, our main stress is placed on easy implementation with an acceptable level of performance. With the simplest form of admission control, all-admission control, which admits all mobiles if there are available channels, type II error is minimized, but type I error is maximized. From the user’s perspective, however, type I error is more harmful than type II error, since type I error will result in an outage and may cause the dropping of ongoing calls. Obviously, screening the arrived calls and admitting them in a selective way will re- duce type I error though increase type II error. Received CIR, re- ceived signal strength, and present system load are well-known examples of the screening criteria. According to the observation in [9], a tight admission policy can improve type I error perfor- mance without compromising too much of type II error. How- ever, the screening methods do not provide a systematic way for setting the new transmitter power. Power-controlled admission algorithms have been proposed in [10] and [11]. They admit a new mobile at the moment it arrives in a controlled manner where the CIR of existing calls is adequately protected, and make the final accept/reject deci- sion after numbers of power-updating iterations. That is, the new mobile is allowed to interact with the network before the final decision is made, in order to learn how the network responds to. Since the new mobile is accepted only if an effective power vector is obtained, the algorithms are error free. However, their slow convergence due to the large number of iterations required before the final decision often renders them impractical. More- over, the algorithms also need to deliver control data generated by the new mobile to all the existing mobiles in the system, which requires significant network communications cost. Thus, the performance improvement in these interactive methods is at the expense of the admission delay and network signaling ef- forts. It is the objective of this paper to provide efficient interactive admission control that may curtail the unfavorable costs. For 0018–9545/00$10.00 © 2000 IEEE

Efficient interactive call admission control in power-controlled mobile systems

Embed Size (px)

Citation preview

IEEE TRANSACTIONS ON VEHICULAR TECHOLOGY, VOL. 49, NO. 3, MAY 2000 1017

Efficient Interactive Call Admission Control inPower-Controlled Mobile Systems

Dongwoo Kim, Member, IEEE

Abstract—When a mobile newly arrives and seeks admissionin power-controlled cellular/personal communications services(PCS) systems, interactive call admission control guides theevolution of the power transmitted by the new mobile in order toprotect the transmission quality of ongoing calls from droppingbelow a desired level. In addition, it eventually decides whetherthe new mobile should be accepted or rejected without errors.However, existing interactive call admission algorithms incurhigh network signaling costs and suffer from slow speed in the“accept/reject” decision, which often renders the implementationof these algorithms impractical. This paper presents efficientinteractive algorithms that improve the admission speed aswell as curtail the network coordination cost. The algorithmsare numerically evaluated with respect to the admission speed,decision error, and performance of protecting ongoing call quality.The simulation result shows that the proposed algorithms arepromising for practical implementation.

Index Terms—Cellular/PCS mobile systems, mobile call admis-sion control, power control.

I. INTRODUCTION

I N PRESENT cellular and personal communications services(PCS) systems, both cell size and frequency reuse distance

are getting smaller, and as a result, the receivers are exposed tomore severe noise compared to conventional cellular systems. Inorder to provide each mobile an acceptable connection in suchinterfering situations, many power control algorithms have fo-cused on limiting the interference seen by other mobiles [1]–[5].The power control algorithms converge to aneffectivepowervector that attains predetermined carrier-to-interference (CIR)targets at the receivers, if such a power vector exists. Otherwise,certain mobiles cannot achieve their target CIR and are conse-quently suffering severe deterioration in transmission quality. Inthe latter case, unnecessary interference is produced and somesystem performance might be lost. The admission control of anewly arrived mobile, which we study here, is considered as apractical means of preventing this latter case where not all mo-biles can be supported.

When a new mobile requires a channel, the admission controlhas to tradeoff between two types of errors: type I error wherethe mobile is erroneously accepted and the congestion occursand type II error where the mobile is erroneously rejected whileit can be supported along with other active mobiles. For our pur-pose, type I error should be minimized. In addition, even though

Manuscript received March 31, 1998; revised July 20, 1999.The author is with the School of Electrical Engineering and Com-

puter Science, Hanyang University, Kyunggi 425-791, Korea (e-mail:[email protected]).

Publisher Item Identifier S 0018-9545(00)03694-X.

the mobile would be properly accepted, the admission controlshould guide the evolution of the power level transmitted by thenew mobile. Otherwise, the interference introduced by the newmobile usually results in dropping transmission quality of on-going calls for the time period that is needed by the transmittersto “reconverge” to aneweffective power vector.

A centralized linear/combinatorial programming approachcould provide an optimal admission decision as well as ade-quate power levels letting all mobiles meet their CIR targets[6]–[8]. Its practical implementation, however, is normallyprohibited since it requires to gather and process a large amountof data including all link-gain information in the system. Thus,in designing the proposed call admission algorithm, our mainstress is placed oneasy implementationwith an acceptablelevel of performance.

With the simplest form of admission control,all-admissioncontrol, which admits all mobiles if there are available channels,type II error is minimized, but type I error is maximized. Fromthe user’s perspective, however, type I error is more harmfulthan type II error, since type I error will result in an outage andmay cause the dropping of ongoing calls. Obviously, screeningthe arrived calls and admitting them in a selective way will re-duce type I error though increase type II error. Received CIR, re-ceived signal strength, and present system load are well-knownexamples of the screening criteria. According to the observationin [9], a tight admission policy can improve type I error perfor-mance without compromising too much of type II error. How-ever, the screening methods do not provide a systematic way forsetting the new transmitter power.

Power-controlled admission algorithms have been proposedin [10] and [11]. They admit a new mobile at the moment itarrives in a controlled manner where the CIR of existing callsis adequately protected, and make the final accept/reject deci-sion after numbers of power-updating iterations. That is, the newmobile is allowed to interact with the network before the finaldecision is made, in order to learn how the network respondsto. Since the new mobile is accepted only if an effective powervector is obtained, the algorithms are error free. However, theirslow convergence due to the large number of iterations requiredbefore the final decision often renders them impractical. More-over, the algorithms also need to deliver control data generatedby the new mobile to all the existing mobiles in the system,which requires significant network communications cost. Thus,the performance improvement in theseinteractivemethods is atthe expense of the admission delay and network signaling ef-forts.

It is the objective of this paper to provide efficient interactiveadmission control that may curtail the unfavorable costs. For

0018–9545/00$10.00 © 2000 IEEE

1018 IEEE TRANSACTIONS ON VEHICULAR TECHOLOGY, VOL. 49, NO. 3, MAY 2000

this purpose, two error-free interactive algorithms are presented.One is acoordinatedmethod that still requires base stations toexchange measurement or control data, but greatly improves thecall admission speed compared to the existing methods. Theother is adistributedone where the new mobile adjusts its trans-mitter power with a predefined simple intellectual rule and di-minishes the intensive network communications cost as well asreaches the final decision more quickly. The latter temporarilydrops the transmission quality of the existing links. However,both the number and the magnitude of droppings are below rea-sonable values, respectively.

This paper is organized as follows. In Section II, we intro-duce the power-controlled mobile systems considered in thispaper. In Section III, we define the coordinated method, pro-vide its valuable property, discuss its operational details, andcompare it with the existing methods. The distributed method ispresented in Section IV. In Section V, simulation results are pro-vided. We finish this paper with some concluding discussions inSection VI.

II. THE POWER-CONTROLLED MOBILE SYSTEM

Our admission algorithm is built on the distributed con-strained power control (DCPC) scheme [2] with slightgeneralization which also has been adopted in [10]. Assumingthere are -deployed mobiles using a particular channel, werestrict our attention to the uplink case (from mobile to base).The downlink (from base to mobile) can be modeled in thesame way, though some implementational definitions need tobe modified (for example, see [5] and [8]).

The transmitted power of mobileis which is limited bya maximum power level as

for (1)

Let denote the link gain on the path between transmitterand base. At base that is used by transmitter, the receivedCIR is given by

(2)

where is the receiver noise including spurious and thermalnoise and the denominator is the total interference received andwill be denoted by . The objective of power control is to obtaina nonnegative power vector which satis-fies the maximum power constraint (1) and transmission qualityrequirement

for (3)

where is the required CIR which depends on the traffic typethat mobile is transmitting. We call a transmitter (or mobile)activewhen it keeps the quality target.

Power adjustments in DCPC take the form of

(4)

where the superscript denotes iteration time, which will beoften omitted in the following unless necessary. Each mobilecontrols its transmitter power within the maximum level, basedon the information about its own power level and CIR measure-ment forwarded from the corresponding base. Givenand thenumber of mobiles which simultaneously occupy the samefrequency channel, DCPC eventually achieves a power alloca-tion that lets all mobiles meet the requirements if it is feasible[2], [3], [5].

Note that although the above power control model often hasbeen used in the context of TDMA or FDMA systems, it is par-ticularly appropriate for the uplink of a CDMA system in whichunsynchronized wideband CDMA signals of other mobiles inthe home cell as well as in other cells can be modeled as theinterfering sources [12]. In this case, the link gain in (2)includes the processing gain and would be interpreted as asignal to interference ratio. In the simulation, to evaluate themethods to be proposed, a two-dimensional (2-D) TDMA (orFDMA) system is examined.

III. T HE COORDINATED INTERACTIVE CALL ADMISSION

CONTROL (CICAC)

When a mobile newly arrives, it immediately starts to transmitpower, though it is finally either accepted or rejected after an in-teractiveadmission phase. The transmission quality of new mo-bile is normally below its target during the admission phase, forwhich the mobile does not transmit significant traffic informa-tion such as actual voice and data.

All the existing mobiles are assumed active whena new mobile arrives and let zero be the index that denotes thenew mobile. When mobile 0 begins to transmit power at, eachof the active link suffers additional interference determined by

. In order to overcome the additional interfer-ence, the powers of active transmitters are boosted byat themoment where the new mobile begins to transmit. Moreover,the transmitter power of mobile 0 is restricted by artificial limit

to protect ongoing calls.CICAC Procedure:Given active transmitter powers,

, we have the following.

Step 0) (Initializing the Admission Phase)Set , and compute

(5)

and

(6)

where

(7)

and set

(8)

for (9)

KIM: EFFICIENT INTERACTIVE CALL ADMISSION CONTROL 1019

Step 1) (Testing the Initial Condition)If , then reject mobile 0, stop CICAC, and

continue DCPC only for mobiles .Step 2) (Making an Admission Decision)

With , make an admission de-cision where the outcome is one ofaccept, reject,andnot-determine. If the decision isnot-determine,then go to Step 3); otherwise, terminate the admis-sion phase and continue DCPC with the transmitterscorresponding to the outcome.

Step 3) (Updating Transmitter Powers in the AdmissionPhase)

Update each power with DCPC rule (4) for allmobiles while setting ,and go back to Step 2) with setting .

When a new mobile starts to seek admission, network coordi-nation between bases is required first to computeand thatare used in setting initial powers in Step 0). Each mobile in thesystem then updates its power with ordinary DCPC except thatthe new mobile uses the computed power constraint before thefinal decision is made.

A. Accept/Reject Decision Criteria

If in (5), that is, a certain mobile is already transmittingat its maximum, there is no way to admit the new mobile withoutintervening ongoing calls. Thus, the attempt should be denied toprotect CIR of the active link from dropping below the require-ment in Step 1). Before discussing more decision criteria thatcould be used in Step 2), we need to state a property of CICACprocedure.

Proposition 1: During the admission phase, each of the ac-tive transmitters keeps its transmission quality, that is, for all

for (10)

Proof: Suppose are active transmitterpowers given when a new mobile is introduced. If all the activetransmitters simultaneously boost their powers by , at re-ceiver the interference added by the new mobile

will be allowed if it satisfies

(11)

and then we have

(12)

(13)

(14)

where is referred to as a maximum allowable interference atreceiver . In addition, by (6) and (8)

(15)

Thus

(16)

Therefore

(17)

Suppose that is an iteration index where mobile 0 achievesfor the first time during the procedure. Then

until . If (10) holds at iteration , then, and we hence have

(18)

where is the total interference received by baseat timeincluding the interfering power from mobile 0. This means, withinequality (17), (10) holds for every by mathematicalinduction. Since for all the transmitters includingmobile 0, (10) holds for every analogous to (18). Thiscompletes a proof.

By the above proposition, the CIR corresponding to each ac-tive transmitter does not drop below the threshold at any momentuntil the procedure terminates, which is usually calledsoftad-mission [10]. Moreover, once mobile 0 achieves , it alsokeeps the transmission quality and its power decreases contin-uously. In this case, the tight power constraint, if it is being ap-plied, is obsolete and can be relaxed to an original one. CICACnow shows exactly the same behavior of DCPC. On the otherhand, if (and hence ) for all ,DCPC with the tight constraint converges to a power allocationsuch that for (see [2] and [3] for the conver-gence proof). Accordingly, in Step 2), two criteria are given toterminate the procedure:acceptandrejectcriteria, respectively

Accept Criterion (AC)

If then accept mobile 0

Reject Criterion (RC)

If and for

then reject mobile 0

With AC, the base serving the new mobile determines the accep-tance without network-wide coordinations and it is type I errorfree as discussed above. It may take a long time for RC to beactivated and the decision needs information from other bases.To relieve this burden, one natural choice is limiting the numberof power control iteration in the admission phase as follows:

Iteration-Limited Reject Criterion (IRC)

If the new mobile is not accepted until iterations

where is a design parameter, then reject it

Since too small may result in increasing type II error, addi-tional engineering effort should be devoted to selecting an ad-equate value. Both of RC and IRC may not be free from thetype II error. When for and ,

1020 IEEE TRANSACTIONS ON VEHICULAR TECHOLOGY, VOL. 49, NO. 3, MAY 2000

convergent DCPC achieves a unique minimum power alloca-tion in respect to active transmitters [3], and then CIR of mobile0 cannot be improved any more. Thus, the new mobile may berejected “safely” if

Type II Error -Free Reject Criterion (TRC)

When if for and

then reject mobile 0

However, usually in a heavy traffic network.When capacity becomes a primary issue of admission control,

should be relaxed to to reduce type II error. The fol-lowing describes how Step 2) could be modified in order to pre-vent type II error.

Modified Step 2)

Evoke AC and TRC. If for

and then go to Step 0)

and reinitialize the admission phase

with the following change

set before restarting Step 0)

use instead of in (7)

in (8), set

Otherwise, go to Step 3)

Suppose that the modified CICAC does not stop,(also )determined in (8) by the above new rule increases infinitely,which is impossible since it is bounded above by . Thus,the modified CICAC should be terminated by one of Step 1),AC, and TRC, each of which is an error-free decision.

B. Comparison with the Existing Methods

When a new mobile arrives, in order to protect CIR’s of ac-tive (ongoing) mobiles from dropping below their targets, theirtransmitter powers are set higher than normally required, andthe power evolution of new mobile is limited with a tighterconstraint that is gradually relaxed during the admission phase.Though operational definitions in detail are different, a similaridea is considered in [10] and [11], which also can be imple-mented in an error-free mode.

However, our method has advantageous features that mainlyimprove the call admission speed: simple AC and efficientcomputation with in (7). With AC, we can accept the newmobile before DCPC converges to a power vector that achieves

for , which accelerates the admissionprocess. In computing [see (6) and (7)], if for some, the initial power may be set higher compared with the ex-

isting methods where only term is used. The latterfeature is more useful when another mobile arrives immediatelyafter the previous mobile was rejected in Step 2), since( ) without the interfering transmitter.

CICAC and the existing methods have two common burdensthat should be carried for obtaining soft and error-free admis-sion, but often render the implementation impractical: coordina-tion efforts between bases in computingand in (5) and (6),

respectively, and decision delay for achieving . In thenext section, to get around these problems in implementing in-teractive call admission, we propose distributed admission con-trol which computes and without the coordination and alsoaccelerates the final decision.

IV. THE DISTRIBUTED INTERACTIVE CALL ADMISSION

CONTROL (DICAC)

DICAC enables each of the active mobiles to compute its ownpower scaling-up factor distributedly and the new mobile imme-diately to set the initial power without any network-wide coordi-nation. DICAC, however, would not be soft. Thus, our purposeof this section is to provide apractical call control with an ac-ceptable level of performance in safeguarding ongoing calls.

DICAC Procedure: Given active transmitter powersand a constant , we have the following.

Step 0) (Initializing the Admission Phase)Set , and for each active transmitter

set

for (19)

where

(20)

and for the new mobile, set

(21)

where is a set of neighbor cells that at least oneof mobiles is activated in and the new mobile islistening to control signals from, and

(22)

Step 1) (Testing the Initial Condition)If , then reject mobile 0, stop DICAC, and

continue DCPC only for mobiles .Step 2) (Updating Transmitter Powers in the Admission

Phase)Update each power with DCPC rule (4) for all

mobiles while setting ,where

(23)

Step 3) (Checking Acceptance)If any mobile detects

, then go to Step 4). Otherwise, acceptthe new mobile.

Step 4) (Checking Rejection)If , then set and go to Step

2). If , then set and updateeach power with DCPC; if DCPC converges orthe following condition holds for any mobile

:

and (24)

KIM: EFFICIENT INTERACTIVE CALL ADMISSION CONTROL 1021

where and are small positive numbers, thenreject the new mobile; otherwise, go to Step 3).

In DICAC, in contrast to CICAC, the initial powers can be setwithout the network coordination between bases sinceand

are computed independently like (20) and (23), respectively.Each base station can calculate allowable interferencein (22)only by measuring the corresponding link and broadcasts it on arelevant channel. Note that is calculated with a conservativeview where base expects that the other active mobiles have amargin in their transmitter power enough to boost by. Weassume the new mobile is listening to the control signals from aset of neighbor cells and can obtain for . Then, itcan measure the downlink gains from base . From thereciprocal radio propagation assumption, the new mobile alsoknows the uplink gains . The elements in mayvary depending on mobile locations and propagation environ-ments. If is empty, we assume .

In CICAC, each increment in active transmitter power at theinitial step is determined as the same amount that can be simul-taneously allowed by all the transmitters. Since the quantity iscontinuously changing, depending on traffic load and mobilelocations, the instantaneous computation requires intensive net-work communications. In DICAC, each base accumulates infor-mation about and periodically computes such that

(25)

where is a design parameter. If %, then is chosenas the least possible increment during the considered time pe-riod. is then determined by

(26)

The tight power constraint on a new transmitter is relaxed fol-lowing a negative exponential function in (23). A critical pa-rameter for this function is by which we can adjust the re-laxation speed. Since DICAC begins to consider rejecting themobile when , the smaller is, the faster a final deci-sion can be obtained though ongoing calls tend to be affectedby more interference.

The condition (24) is called aninstability detection condition(IDC) that monitors whether the quality target is not expectedto be achievable at the current state during DCPC procedure[4]. The first inequality is a heuristic condition that detects theconvergence of transmitter power levels. For too large, it failsto make accurate detection of convergence, but for too small oneit may require many iterations to satisfy the inequality. When theconvergence is detected, the second inequality checks whetherit is impossible to achieve the target value from the current CIRlevel.

Since DICAC accepts the mobile only when forall in Step 3), it is type I error free. Ifis set sufficiently small (or without IDC), the reject condition isnot satisfied until DCPC converges, and hence DICAC becomesalso type II error free. If is set too large and is set toosmall, DICAC may terminate very fast with relatively high typeII error. In Section V, we numerically test the performance ofDICAC for various ( , ) combinations.

V. SIMULATION

A. Simulation Environment

Simulation for evaluating the performance of the proposed al-gorithms is done for a 2-D 7 11 Manhattan-like microcellulararray (see Fig. 1 and [10] and [13]). Streets are running betweenthe building blocks of length 100 m in two directions—hori-zontal and vertical. We assume an asymmetric fixed channel as-signment scheme where the cluster size is three. In Fig. 1, thedark crosses are the cochannel cells where the correspondingbase station is at one of the street corners at lamppost level.

The link gain is modeled as a product of two variables,. The variable is the variation in received

signal due to shadow fading and assumed to be independent andlognormally distributed with a mean of 0 dB and a standard de-viation of 4 dB [14]. The variable is the large-scale propa-gation loss, which depends on the transmitter and receiver loca-tion, and on the type of geographical environments. From [15],

for a typical metropolitan environment can be modeled by

(27)

where and are the horizontal and the vertical distances, re-spectively, between the mobile and base station,is the speedof light, is the transmission frequency, and are thestreet widths in the horizontal and vertical directions, respec-tively, and parameters and are propagation con-stants [14]. In our simulation, we use the same parameter valuesconsidered in [10] such as MHz, m,

m, and m.

B. Performance Measures and Evaluation Method

To compare the proposed algorithms, we numerically eval-uate the following performances:

• admission speed: the number of power-updating iterationsmade before the final decision;

• probability of type II error: (the number of erroneous re-jection) (the total number of rejection);

• probability of CIR dropping: (the number of ongoingtransmitters dropping below the CIR requirement duringthe interactive admission phase) (the accumulatednumber of ongoing transmitters involved in the admissionphase);

• average magnitude of CIR dropping: (the sum of(in decibels) for ongoing transmitters andadmission iterations , such that )

( ).Since the proposed methods are all type I error free, we do not

measure type I error performance, and the probability of type IIerror might be traded with the admission speed. The first twoperformances are evaluated for various CICAC and DICAC al-gorithms, respectively. The last two performances are computedonly for DICAC that would not support the soft admission.

1022 IEEE TRANSACTIONS ON VEHICULAR TECHOLOGY, VOL. 49, NO. 3, MAY 2000

Fig. 1. A Manhattan-like microcellular array with cluster size 3. The dark crosses are the cochannel cells, and the white squares are the buildings seen from above[10], [13].

Fig. 2. The sample distribution ofp =p obtained from the simulation.

For the simulation, we assume that the mobile in the centeredcell of our 77-cell plan is always mobile 0 and other mobiles areuniformly distributed in the other cells. Before considering ad-mission of the new mobile, we initialize the system with the statewhere all ongoing transmitters are active and using a stationarypower vector. After the initialization, we consider 100 indepen-dent instances where the mobile seeks admission at a randomlocation in the central cell. We execute independent 1000 runsof initialization. Thus, the total number of simulated instancesis 100 000. When for every , we stopupdating the powers and regard as a stationary powervector. We also assume every mobile requires the same constantCIR, . The maximum transmitter power is set to 1 W foreach mobile, and the receiver noise is taken 10W.

TABLE ISAMPLE MEANS (�) AND STANDARD

DEVIATIONS (�) FOR p =p AT THE INITIALIZATION STATE OF THE

SIMULATION (IN DECIBEL SCALE)

At the initialized states, the relative frequency of (indecibels) is counted and shown in Fig. 2 (also summarized in

KIM: EFFICIENT INTERACTIVE CALL ADMISSION CONTROL 1023

Fig. 3. The accept/reject rate realized from the simulation.

Fig. 4. Comparison of the admission speed between coordinated methods.

Table I). The distribution of looks like a normal distri-bution that is slightly skewed to the left. Fig. 2 shows that themean of gets smaller as the target CIR increases. Foreach instance considered in the simulation, we also record theerror-free accept/reject outcomes. Fig. 3 shows the accept/re-ject rate realized from the simulation. Since the new mobile istrivially accepted when the required quality is below 14 dB, weperform the simulation with the CIR target from 15 to 26 dB forthe convenience.

C. Results for CICAC Performance

Since CICAC withModified Step 2)is both type I and typeII error free, we do not include it in the simulation. In [10], theadmission speed is reported for such an algorithm to require tens

of thousands of power updates to reach the final decision. Wecompare the following CICAC algorithms.

• CICAC-AR: CICAC with AC and RC.• CICAC-AI20: CICAC with AC and IRC for .• CICAC-AI30: CICAC with AC and IRC for .• F-SAS: type I error free and soft call admission control

presented in [10] [it is identical to CICAC only with RC(without AC) if the admission phase starts with an activeand stationary power vector; it tests the acceptance after itfinds a “new” stationary power vector].

The admission speed of the algorithms is shown in Fig. 4.Since CICAC-AI20 and CICAC-AI30 make the final decisionbefore at least 20 and 30 power updates, respectively, theiradmission speeds are bounded above respectively by 20 and30. CICAC-AR saves approximately 300 iterations compared

1024 IEEE TRANSACTIONS ON VEHICULAR TECHOLOGY, VOL. 49, NO. 3, MAY 2000

Fig. 5. Comparison of the probability of type II error between coordinated methods.

Fig. 6. Comparison of the admission speed between DICAC methods without IDC, for variousT ’s; k = 3: (a)A = � and (b)A containing five cells.

with F-SAS, but its admission speed also gets slower if the CIRtarget increases to keep the required transmission quality (whenthe mobile environment goes worse). Though CICAC-AI20and CICAC-AI30 achieve the fast admission decision, theysuffer severe type II error as shown in Fig. 5. For CICAC-ARand F-SAS, the probability of type II error is maintained below0.03 at the expense of slow decision.

D. Results for DICAC Performance

DICAC is evaluated for different parameter settings as thefollowing.

• DICAC, without IDC, for various values of the relaxationparameter in (23) (see Figs. 6, 8, and 9).

• DICAC, without IDC, for various values of the power-uplimit in (20) (see Fig. 7).

• DICAC with IDC and, respectively, with and(see Figs. 10 and 11).

In addition, since the performances of DICAC also depend onthe number of neighboring cells in , we test two cases, respec-tively, where and consists of five cochannel cellsnearest to the centered cell. In Figs. 7, 10, and 11, the dottedlines result from the former case and the solid lines are from

KIM: EFFICIENT INTERACTIVE CALL ADMISSION CONTROL 1025

Fig. 7. Comparison of the admission speed between DICAC methods without IDC, for variousk’s; T = 10.

Fig. 8. Comparison of the probability of CIR dropping between DICAC methods without IDC for ( ; k) pairs where = 18; 20; 26; andk = 0; 3: (a)A = �and (b)A containing five cells.

the latter case. In Figs. 6, 8, and 9, the part (a) of each figureresults from the former case and the part (b) is from the lattercase. When we include two parts in a figure, we use the samescale to clearly contrast them.

To set the value of in (20), we use the sample mean ()and the sample standard deviation () of obtained atthe initialization states in the simulation. Table I summarizesthe results. We parameterize them as

. For an example, if we assume ’s are nor-mally distributed with and , only 1% of is likely tobe less than when is set to three.

Fig. 6 shows the admission speed of DICAC without IDC for. Fig. 6(a) is for the case of and

Fig. 6(b) is for the other case. Since DICAC without IDC is bothtype I and II error free, the admission speed obtained in Fig. 6shows its potential for practical implementation, compared withthe results by CICAC-AR and F-SAS that may make erroneousrejection. If Fig. 6(a) is compared with Fig. 6(b), up to 100 it-erations can be saved at low CIR targets when the neighbor-cellinformation is available. The admission speed is also dependenton , i.e., . The smaller , i.e., the greater , is, the fasterthe speed is as illustrated in Fig. 7.

1026 IEEE TRANSACTIONS ON VEHICULAR TECHOLOGY, VOL. 49, NO. 3, MAY 2000

Fig. 9. Comparison of the average magnitude of CIR dropping between DICAC methods without IDC for ( ; k) pairs where = 18; 20; 26; andk = 0; 3:(a)A = � and (b)A containing five cells.

Fig. 10. Comparison of the admission speed between DICAC methods with IDC for (� , � ) pairs where� = 10 ; 10 ; 10 ; 10 , and� = 10 ; 10 ;k = 3: (a)T = 10 and (b)T = 20.

Though DICAC without IDC is error free, the CIR of on-going links may drop below the target. Figs. 8 and 9, respec-tively, show the probability of the CIR dropping and its averagemagnitude for . Results are given for dif-ferent ( ) pairs where and . For allcases, the probability of dropping is below 0.065% and the av-erage magnitude is less than 0.003 dB. Thus, the performance

degradation of DICAC, compared with CICAC, is fairly accept-able.

IDC accelerates the admission speed, but causes type II error.DICAC with IDC is tested for various ( , ) pairs where

and . Fig. 10 showsthe admission speed. Fig. 10(a) and (b) is obtained from setting

and , respectively. Compared with Fig. 6, almost

KIM: EFFICIENT INTERACTIVE CALL ADMISSION CONTROL 1027

Fig. 11. Comparison of the probability of type II error between DICAC methods with IDC for (� , � ) pairs where� = 10 ; 10 ; 10 ; 10 , and� =

10 ; 10 ; k = 3: (a)T = 10 and (b)T = 20.

a half of the number of iterations is reduced. The smaller,i.e., the tighter detection condition, is, the more iterations arerequired for reaching the final decision. In this case, the tradeoffrelation between the admission speed and the probability of typeII error still holds. Fig. 11 shows that the smalleris, the bettertype II error performance is achieved. If is set as a greatervalue, it also tightens IDC, as a result, reaches the final decisionslower, and attains lower probability of type II error as shownin Figs. 10 and 11.

VI. CONCLUDING DISCUSSIONS

In this paper, we present two types of interactive admissioncontrol and evaluate them in a variety of operational settings. Allthe proposed algorithms are type I error free, which is extremelybeneficial from the user’s perspective. Since the existing inter-active algorithms suffer from the slow admission speed and re-quire costly network-wide coordination, our main objective inthis paper is to find fast and distributed call control. CICAC di-rectly improves the admission speed with a new accept rule,and DICAC enhances the speed further without the intensivenetwork signaling effort while restricting the admission errorsbelow an acceptable level.

The tolerable time, for which the new user can wait until thefinal admission decision is made, is usually limited in practicalmobile systems. If the new user can wait 4 s before transmit-ting actual voice or data information and it takes 100 ms forone power control iteration, the admission phase should be ter-minated within 40 iterations. CICAC-AI20, CICAC-AI30, andDICAC’s with IDC set ( ) for ,and or pass the above requirement. Especially,DICAC with IDC set , restricts

type II errors within 1% of the rejected instances. Thus, theDICAC is most promising for practical implementation.

ACKNOWLEDGMENT

The author would like to thank the anonymous reviewers fortheir valuable comments which improved the presentation ofthis paper.

REFERENCES

[1] J. Zander, “Distributed cochannel interference control in cellular radiosystems,”IEEE Trans. Veh. Technol., vol. 41, no. 3, pp. 305–311, 1992.

[2] S. A. Grandhi, J. Zander, and R. Yates, “Constrained power control,”Wireless Personal Commun., vol. 1, no. 4, 1995.

[3] R. D. Yates, “A framework for uplink power control in cellular radiosystems,”IEEE J. Select. Areas Commun., vol. 13, pp. 1341–1347, Sept.1995.

[4] D. Kim, K.-N. Chang, and S. Kim, “Efficient distributed power controlfor cellular mobile systems,”IEEE Trans. Veh. Technol., vol. 46, pp.313–319, May 1997.

[5] D. Kim, “Downlink power allocation and adjustment for CDMA cellularsystems,”IEEE Commun. Lett., vol. 1, pp. 96–98, July 1997.

[6] M. Andersin, Z. Roseberg, and J. Zander, “Gradual removals in cellularPCS with constrained power control and noise,”Wireless Networks, vol.2, pp. 27–43, 1996.

[7] S. Kim and D. Kim, “Optimum transmitter power control in cellularradio systems,”INFOR, vol. 35, no. 1, pp. 37–47, Feb. 1997.

[8] D. Kim and S. Kim, “Forward link power control in CDMA cellularsystems,” , to be published.

[9] Z. Liu and M. E. Zarki, “SIR-based call admission control forDS-CDMA cellular systems,”IEEE J. Select. Areas Commun., vol. 12,pp. 638–644, May 1994.

[10] M. Andersin, Z. Roseberg, and J. Zander, “Soft and safe admission con-trol in cellular networks,”IEEE/ACM Trans. Networking, vol. 5, pp.255–265, April 1997.

[11] N. D. Bambos, S. C. Chen, and G. J. Pottie, “Channel access algorithmswith active link protection for wireless communication networks withpower control,” , submitted for publication.

[12] C. Y. Huang and R. D. Yates, “Rate of convergence for minimum powerassignment algorithms in cellular radio systems,”Wireless Networks,vol. 4, no. 3, pp. 223–232, April 1998.

1028 IEEE TRANSACTIONS ON VEHICULAR TECHOLOGY, VOL. 49, NO. 3, MAY 2000

[13] M. Gudmundson, “Cell planning in Manhattan environments,” inIEEEProc. Trans. Veh. Technol. Conf., vol. VTC-92, pp. 435–438.

[14] J.-E. Berg, R. Bownds, and F. Lotse, “Path loss and fading models formicrocells at 900 MHz,” inIEEE Proc. Trans. Veh. Technol. Conf., vol.VTC-92, pp. 666–671.

[15] J.-E. Berg, “A simplified street width dependent microcell path lossmodel,” COST 231, TD(94)035, 1994.

Dongwoo Kim (M’95) received the B.A. degree ineconomics from Seoul National University, Seoul,Korea, in 1987 and the M.S. and Ph.D. degrees intelecommunications engineering/operations researchfrom the Korea Advanced Institute of Science andTechnology (KAIST), Daejon, Korea, in 1989 and1994, respectively.

Since 1994, he was with Shinsegi Telecomm, Inc.,Seoul, where he was involved in the design, develop-ment, and testing of CDMA cellular systems for suc-cessful commercial CDMA deployment in Korea. He

is currently with the School of Electrical Engineering and Computer Science,Hanyang University, Kyunggi, Korea. His research interests are in the areas ofefficient power control and advanced mobile system design.