A Resource Allocation Algorithm using Frequency Borrowing in Hierarchical CDMA Cellular Systems

Transcription

1 A Resource Allocation Algorithm using Frequency Borrowing in Hierarchical CDMA Cellular Systems Young-uk Chung and Dong-Ho Cho Division o Electrical Engineering Deartment o Electrical Engineering and Comuter Science Korea Advanced Institute o Science and Technology (KAIST) Guseong-dong Yuseong-gu Daejeon, Reublic o Korea Abstract The great increase o the number o users requires eicient management o the sectrum. In hierarchical cellular systems based on CDMA, FBRA algorithm is roosed to allocate requency sectrum more eiciently. This algorithm lends requency sectrum o macrocell to microcells when mass hot sot occurred. We analyze the intererence to signal ratio and reverse link caacity in case o using the roosed algorithm. Based on the result, Algorithm eroance is evaluated in view o blocking, the hando reused robability, and the sectrum eiciency. Numerical result shows that the roosed algorithm can increase the caacity and solve a mass hot sot roblem. with a ew isolated hot sot cells. However, when hot sots occur in several adjacent cells consecutively, such as rush-hour, these schemes do not work well. We roose a new resource allocation algorithm to relieve mass hot sot traic and to increase the system caacity in a CDMA based hierarchical cellular structure. The roosed algorithm lends sectrum o macrocell to its embedded microcells when mass hot sot traic occurs. Comarison o analysis results shows that the roosed schemes can increase the system caacity and can solve mass hot sot traic roblem. I. INTRODUCTION There has been remarkable growth in the number o mobile users in recent years. Moreover, uture mobile communication systems will rovide multimedia services needing more channel resources er user. The great increase o the number o users causes a shortage o caacity in the current system using macrocells with a ew kilometers diameter. The best way to suort increasing numbers o users is the allocation o additional sectrum, which is a limited resource. Eicient management o the sectrum is, consequently, an imortant issue. The IMT-2000 system uses a microcell with diameter o a ew hundred meters. Using such a microcell, we can increase the system caacity. However, in a microcell system, the number o handos increases greatly and the traic load distribution is nonunio. The hierarchical cellular structure, which consists o microcell clusters suerimosed on a macrocell, can solve this hando roblem. In this hierarchical structure, high seed mobile teinals are serviced in the macrocell while low seed mobile teinals are serviced in the microcells to minimize the number o handos. But, the uneven traic load roblem which raises hot sot can not be erectly solved by hierarchical cell structure. Hot sot cells aear mainly in urban environments severe traic congestion at seciic times (rush-hour) is common. Hot sots occur eriodically in a mass o micro/icocells. During rushhour, mobiles move very slowly, new call attemts increase greatly, and hando attemts rarely occur. In hierarchical cellular structures, hot sot traic loads can be alleviated by shiting overlow traic between lower and higher layers[1]. This aroach is an eective solution or a system II. DESCRIPTION OF FBRA ALGORITHM The FBRA (Frequency Borrowing Resource Allocation) algorithm is roosed to relieve mass hot sot traic and to increase the system caacity in a CDMA based hierarchical cellular structure. To avoid intererence between layers, each layer has a dierent sectrum. It is assumed that macrocell has at least two FAs(Frequency Assignments). Allocation o a mobile user to a macro/microcell is deteined by the mobile seed. This hierarchical system is usually located in downtown region that rush-hour outbreaks eriodically. The system status is divided into rush-hour and noal cases. In the noal case, many mobiles are serviced in macrocells because traic is not heavy and there are many ast mobiles. During rush-hour, almost all mobiles are suorted in microcells because traic is heavy and the seed o mobiles is slow. Almost all microcells become hot sot cells in this circumstance. The Frequency Borrowing Resource Allocation(FBRA) algorithm can alleviate this consecutive hot sot roblem. I the same number o FAs is rovided to each macrocell and embedded microcells, resectively, microcells can suort more users than overlaid macrocell because the number o microcell is much more than that o macrocell in the same area. From this oint, we roose the FBRA algorithm that FAs o macrocell are borrowed to microcells. This algorithm is shown in Fig. 1. The FAs o a macrocell are divided into dedicated FAs and common FAs. A dedicated FA is used only in a macrocell. Common FAs can be borrowed to microcells. When a call is originated, a channel o dedicated FAs is assigned. Ater all channels o dedicated FAs are used, channels o common FAs /02/$ IEEE 371

2 Fig. 3. System Model ater FA borrowing Fig. 1. Flow Chart o FBRA Scheme though this algorithm requires additional CEs, this algorithm can use FAs more eiciently. Because a FA is a limited resource and a CE is a relatively unlimited resource, the FBRA algorithm has merit in this alication. (a)noal Case (b) Rush-hour Case Fig. 2. FA Allocation in Each Case are assigned. When rush- hour traic conditions occur, common FAs are borrowed to microcells according to the traic load(fig. 2). I there are mobiles serviced by the FA that is to be borrowed to microcell, the mobiles are handoed to a dedicated FA o the macrocell. During rush-hour, all common FAs have ew mobiles to service because there are ew mobiles to be serviced in macrocells during rush-hour. Consequently, the caacity o each microcell increases greatly and the hot sot roblem can be solved. Ater rush-hour case, the borrowed FAs are returned to the macrocell. The beginning and ending o rush-hour can be detected by historical inoation or E b N 0 inoation o macrocell and embedded microcell. This algorithm has an extra beneit such as the caacity increase o neighboring macrocells. The requency borrowing to microcell reduces the intererence to neighboring macrocells. So, this algorithm rovides additional resources not only to hot sot microcells, but also to neighboring macrocells. To imlement this algorithm in an ordinary cellular system, each microcell must have additional Channel Elements(CE) corresonding to the common FAs o the macrocell. Even III. REVERSE LINK CAPACITY ANALYSIS The roosed algorithm lends sectrum o macrocell to its embedded microcells when mass hot sot traic occurs. This algorithm can increase the system caacity and can solve mass hot sot traic roblem. But, this borrowing can cause additional intererence change to neighboring macrocells. In CDMA system, reverse link caacity is imortant in view o system eroance[2]-[4]. In this section, we analyze the reverse link caacity o microcells and neighboring macrocells when rush-hour occurs and FA o macrocell is borrowed. The system model when common FA o macrocell is borrowed to embedded microcell is shown in Fig 3. We assume that seven microcells are embedded in a macrocell. Each microcell and macrocell in this igure use the same FA and this causes intererence each other esecially in reverse link. We assume this system eros erect ower control. The intererence roduced by a mobile user which is located at a distance r m rom BS(Base Station) o its locating cell and at a distance rom BS o interest cell is calculated as[5] οm I(;r m) ο 0 r 4 m S 4 (ο 0οm) 1: (1) ο 0 is a shadowing loss between a mobile user and BS o interest cell and ο m is a shadowing loss between a mobile user and BS o current cell. ο 0 and ο m are Gaussian random variables with zero mean and standard deviation, resectively. Each ο 0 and ο m are indeendent, and the dierence is also 372

3 Gaussian random variable with zero mean and 2 2 standard deviation. In this aer, we use the model o 8. Let the number o mobile users in a center microcell(cell 0 ) be N cμ, the number o mobile users in a irst circle microcell(cell 1 cell 6 )ben 1μ, the number o mobile users in a irst circle macrocell(cell 7 cell 12 )ben 1m, and the number o mobile users in a second circle macrocell(cell 13 cell 24 ) be N 2m. We ignore the intererence rom macrocells aterward third circle. We assume that macrocell aterward second circle has the same number o mobile users N 2m. I i;j ( ;r m ) is the intererence rom user j in cell i. Then, the intererence-tosignal ratio in center microcell(cell 0 )is shown as I cμ N 6 1μ i1 24 N 2m i13 χ i;j ρ χ i;j I i;j ( ;r m) 12 N 1m i7 χ i;j I i;j ( ;r m) χ i;j I i;j ( ;r m) (2) 1 with robability 0 with robability 1 : Here, is the voice activity actor and is assumed to be 38. The irst te o (2) is intererence rom irst circle microcells and the second te is intererence rom irst circle macrocells. The third te is intererence rom second circle macrocells. Similarly, the intererence-to-signal ratio in a irst circle microcell(cell 6 )isgivenby Ncμ I 1μ N 12 1m i7 χ 0;j I 0;j ( ;r m) χ i;j I i;j ( ;r m) 5 N 1μ i1 N 24 2m i13 χ i;j I i;j ( ;r m) χ i;j I i;j ( ;r m) (3) the irst te is intererence rom center microcell, the second te is intererence rom other irst circle microcells, the third te is intererence rom irst circle macrocells, and the ourth te is intererence rom second circle macrocells. And the intererence-to-signal ratio in a irst circle macrocell(cell 7 ) is described as I 1m Ncμ 12 i8 χ 0;j I 0;j ( ;r m) N 1m 6 i1 χ i;j I i;j ( ;r m) i1316;2227;4142 N 2m N 1μ χ i;j I i;j ( ;r m) χ i;j I i;j ( ;r m) : (4) I we noalize the macrocell radius to unity, the traic densities o a center microcell ρ cμ, a irst circle microcell ρ 1μ,a irst circle macrocell ρ 1m,and a second circle macrocell ρ 2m are shown as below, resectively. 14N cμ 14N 1μ ρ cμ ; ρ 1μ ; 2N 1m 2N ρ 1m ; ρ 2m 2m : (5) Let the area o a center microcell be A cμ, the total area o irst circle microcells be A 1μ 6 A cμ, the total area o irst circle macrocells be A 1m ß 6 7 A cμ, and the total area o second circle macrocells be A 2m 2 A 1m. Then, the average and variance o intererence-to-signal ratio o center microcell are given by ZZ 4 E(I cμ) ZZ ZZ 4 ρ 1μ da 1μ 4 ρ 1m da 1m ρ 2m da 2m (6) ZZ Var(I cμ) ρ 1μ da 1μ h ex ln h ln g ex Q(x) 1 ZZ 8 2 ZZ ß R 1 x e y 2 2 dy : h h 2in 2in 1 Q 1 Q log log 2 ρ 1m da 1m 2 ρ 2m da 2m 2 2 ln 2 2 ln Similarly, the average and variance o intererence-to-signal ratio o a irst circle microcell are given by ZZ 4 E(I 1μ) ρ cμ da cμ ZZ 4 ρ 1μ d(5a 1μ6) ZZ 4 ρ 1m da 1m ZZ 4 ρ 2m da 2m (7) ZZ Var(I 1μ ) ρ cμ da cμ 5 io io 373

4 ZZ ZZ 8 2 ZZ 8 2 ρ 1μ d(5a 1μ 6) 2 ρ 1m da 1m 2 ρ 2m da 2m : And the average and variance o intererence-to-signal ratio o a irst circle macrocell are described as Pr(BER>0.001) center microcell 1st circle microcell 1st circle macrocell macrocell only system E(I 1m ) ZZ 4 ZZ ZZ ZZ 4 ρ cμ da cμ 4 ρ 1μ da 1m 4 ρ 1m d(5a 1m 6) ρ 2m da 2m (8) ZZ Var(I 1μ ) ρ cμ da cμ ZZ 8 2 ZZ 8 2 ZZ 8 2 The equation (6)-(8) are calculated as 2 ρ 1μ da 1m 2 ρ 1m d(5a 1m 6) 2 ρ 2m da 2m : E(I cμ) 0:164N 1μ 0:152N 1m 0:068N 2m Var(I cμ) 0:060N 1μ 0:055N 1m 0:016N 2m E(I 1μ ) 0:021N cμ 0:081N1μ 0:146N 1m 0:073N 2m Var(I 1μ ) 0:008N cμ 0:028N1μ 0:054N 1m 0:018N 2m (9) E(I 1m ) 0:001N cμ 0:017N1μ 0:068N 1m 0:128N 2m Var(I 1m ) 0:0001N cμ 0:004N1μ 0:023N 1m 0:040N 2m : From (9), we can calculate the reverse link caacity. Let the reverse link caacity be N, background noise be, total sread bandwidth be W, and inoation bit rate be R. From revious works, adequate eroance(ber < 3 ) can be obtained on the reverse link when E b N 0 5(7dB). So, the minimum requirement to achieve this level o eroance is E b N 0 5. I lower bound o robability to achieve this eroance is 0.99, the reverse link caacity or multile cell CDMA is given by P N1 k0 N 1 k k (1 ) N1k Q Pr(BER > 3 ) 0:01 () ike(is) Var(IS) Reverse Link Caacity/cell (user/cell) Fig. 4. i Reverse Link Caacity/cell WR E b N 0 S : We assume that the average and variance o intererenceto-signal ratio o a second circle macrocell are E(IS) 0:247N 2m and Var(IS) 0:078N 2m [5]. When W 1:25MHz, R 8kbs, and i 30, N 2m is calculated as 36. From (9) and (), the reverse link caacity o center microcell, irst circle microcell, irst circle macrocell are 25 users/cell, 31 users/cell, and 39 users/cell, resectively. The reverse link caacities o each cell are shown in Fig. 4. In the igure, the reverse link caacity o center microcell is calculated in case o N 1μ 31, N 1m 39, and N 2m 36. Similarly, the caacity o a irst circle microcell is calculated in case o N cμ 25, and N 1m 39. Also, the caacity o a irst circle macrocell is N cμ 25, and N 1μ 31. As the reverse link caacity o macrocell system is 36 users/cell, this result shows that the FBRA algorithm rovides additional resources not only to hot sot microcells, but also to neighboring macrocells. IV. PERFORMANCE ANALYSIS AND RESULTS We deine a system with a hierarchical cell structure that has one macrocell and microcells and assume that the microcell has one FA and the overlayed macrocell has one dedicated FA and N common FAs. We assume that all common FAs are borrowed to microcell during rush-hour. From the result o Section III, the initial number o channels o a microcell is 36, and the added number o channels by one FA borrowing is 30. Using the birth-death rocess, we analyze the eroance o the roosed algorithm. The analysis is eroed using the same method worked by Su et al.[6]. We calculate the numerical examles in case that the FBRA algorithm is used or not, and comare the results. We assume that call duration time is exonentially distributed with mean 0 sec and the ratio o hando region on a cell is 0.3. It is assumed that two channels and our queues are reserved or hando calls. Cell dwell time is assumed to be exonentially distributed with mean 500 sec, because the seed o user is very slow during rush-hour. We analyzed only the eroance o 374

5 1 Blocking Probabilities Hando Reused Probabilities Fig Fig. 5. New Call Arrival Rate without FBRA (N0) with FBRA (N1) with FBRA (N2) Blocking Probability vs. New Call Arrival Rate without FBRA (N0) with FBRA (N1) with FBRA (N2) New Call Arrival Rate Hando Reused Probability vs. New Call Arrival Rate microcell during rush-hour, because the roosed algorithm is used in rush-hour case. During rush-hour, almost all users are serviced in microcell and ew users are serviced in macrocell. So, it is exected that no new call blocking and hando call droing is occurred in macrocell during rush-hour. Numerical results when N 1 are shown in Fig. 5 and Fig. 6. The blocking robability and hando reused robability or the FBRA algorithm are signiicantly lower. From these results, we can see that extra resources are obtained and hot sot traic is relieved by the FBRA algorithm. We also calculated numerical examles when the number o common FAs N is changed. I the number o common FAs is increased, the macrocell can lend more FAs to microcells and this rovides more resources to microcell. To evaluate the sectrum eiciency o the FBRA algorithm, we calculated the number o FAs needed to guarantee 2% blocking robability. The required FA amount is deteined or the worst case o call arrivals in the system. The macrocell worst case occurs when all mobiles are serviced in the macrocell. The microcell worst case is rush-hour case because almost all mobiles are serviced in microcells during rush-hour. Let the number o dedicated FA in macrocell be F d. We assume that microcell has one FA. Let the intererence limitation o CE er one FA be C l, and the amount o required CE be C r. The new call arrival rate or the worst case in each macro- Required Number o FA to guarantee 2% blocking robability without FBRA with FBRA Oered Traic(Erlang) Fig. 7. Required Number o FA to Guarantee 2% Blocking Probability vs. Oered Traic er cell cell and microcell is assumed to be the same. Then, the total amount o required FA in a hierarchical system A is calculated as A ( d Cr C e d Cr l d Cr C l e: C e l F d :, without FBRA, with FBRA (11) We assume F d 1 and C l 20. The FBRA algorithm requires ewer FAs to guarantee the same QoS(as shown in Fig. 7). Thereore, the FBRA algorithm uses sectrum(fa) more eiciently. V. CONCLUSIONS A resource allocation algorithm to relieve a mass hot sot traic was roosed. The roosed FBRA algorithm considers CDMA based hierarchical system in which the FA o a macrocell is borrowed to underlayed microcells during rush-hour. By borrowing o FAs, the FBRA algorithm can solve the mass hot sot roblem that is common in rush-hour. But, this borrowing causes intererence change to neighboring macrocells. Numerical results show that FBRA algorithm rovides additional resources not only to hot sot microcells, but also to neighboring macrocells. REFERENCES [1] M. Lohi, D. Weerakoon, A. H. Aghvami, Trends in Multi-Layer Cellular System Design and Handover Design, IEEE WCNC 99, New Orleans, USA, , 1999 [2] J. Shaira, Microcell engineering in CDMA cellular networks, IEEE Transactions on Vehicular Technology, vol.43, no.4, , Nov 1994 [3] P. Whiting, Caacity Bounds or a Hierarchical CDMA Cellular Network, IEEE PIMRC 96, vol.2, , 1996 [4] J. -S. Wu, J. -K. Chung, and Y. -C. Yang, Co-channel intererence and caacity or two-tier CDMA cellular systems, IEEE PIMRC 96, vol.1,.88-92, 1996 [5] K. S. Gilhousen, I. M. Jacobs, R. Padovani, A. J. Viterbi, L. A. Weaver, Jr., and C. E. Wheatley III, On the Caacity o a Cellular CDMA System, IEEE Transactions on Vehicular Technology, vol.40, No.2, , May 1991 [6] S. L. Su, J. Y. Chen and J. H. Huang, Peroance Analysis o Sot Hando in CDMA Cellular Networks, IEEE JSAC, vol. 14, No. 9, , Dec