V N U Journal o f S cien ce , M ath em a tics - P hysics 24 (2008) 171-178
E stim ation on efficient o f distribute channel allocation schem es for cellular netw ork
Do Huu T r i '’*, Vu Duy Loi^ Bui Thien Minh^
^Ministry o f Information and Communication (MIC), Ỉ8 Nguyen Du Street, Ha Not
‘Center o f Information Technology Office o f Central Commitee o f Communist Party ofVietnaniy Ỉ Hung Vuong Street, Ha Noi
^ Viet Nam Post and Telecommunication Group (VNPT), Ỉ Dao Duy Anh, Dong Da Street, Ha Noi
Received 20 March 2008; received in revised form 20 May 2008
Abstract. The efficient management and sharing of the spectrum among the users is an important issue, fre q u e n c y channels h av e to be reu sed as m uch as po ssib le in o rd e r to su p p o rt the m any thousands o f simultaneous call in any typical of mobile system. In cellular architecture, cell is serviced by a base station located at its center. A number of cells are linked to a mobile switching ccnter which also acts as a gateway of ihe cellular network to the existing wired networks liked PSTN, ISDN or LAN-WAN based networks. A base station communicates with the mobile station through wireless links, and with the MSC’s through wire-line links. This paper deals with some estimations on efficient of disừibute channel allocation scheme for cellular networks and then LBSB scheme is selected to improve it’s efficient, that is, the first, channel boưowing is not only neighbours but in compact pattern and the second, when a cell become hot, it repeat to boưov/
channel until reaching average degree of coldness of networks, that mean X=C.(dc^^^-dc).
1. Introduction
Efficient o f distribute channel allocation scheme is estim ated by: new call blocking probability, drop call, num ber o f hand off, delay o f channel assignm ent and traffic. In selecting a channel assignm ent m ethod, the objective is to gain a high degree o f spectrum utilisation for a given a quality o f service with the least possible num ber o f database lookups and sim plest possible algorithms at the base station or M SC . In this paper, we present distribute channel allocation schemes in way systematic, estim ation advantage and w eak point o f them.
2. Channel b orrow ing and locking schemes (FCA)
The cellu lar netw orks w ere im plem ented using static frequency channel assignment: after careful frequency planning, channels are assigned to cell sites and these sets aren ’t changed except for a new long-term reconfiguration, this is fixed channel assignm ent, FCA.
Corresponding author. Tcl.: (84) 0912285227 E-mail: [email protected]
171
172 D .H . Tri et aỉ. / V N V J o u r n a l o f Science, M a th em a tics - P h ysics 2 4 (2008) Ỉ 7 Ỉ - Ỉ 78
T he first p roposed schem e was Sim ple B orrow ing (SB) schem e [1] In this scheme, when an incom m ing call req u est aư iv es in the cell and there are no m ore available nom inal channels, the base station can b o ư o w a channel from a neighbouring cell to serve the call request, provided this frequency does no t interfere with the existing calls. M SC supervises the borrow ing procedure follow ing an algorithm that favours channels o ff cells with less traffic demand. The cells (within a distance o f one or tw o cell units aw ay from the boưow er cell) that have a nom inal channel o f the same frequency as a borrow ed channel will not be able to use it because o f the co-channel interference.
Therefore, the M SC “locks” the frequency channel in those cells. T he M SC keeps a record of free, serving, boưovved and locked channels. The SB scheme gets a low er call blocking probability than FC A under light and m oderate traffic conditions o f the expense o f additional storage requirem ent at the M SC and the need for database lookups. In heavy traffic conditions, the channel utilisation efficiency in SB is very m uch degraded because the locked channels reduce the available capacity.
2 . 1. H ybrid C hannel Borrow ing Schemes
The m ain problem w ith the SB schem e is the absence o f control in the num ber o f channels that can be lent by a cell. In the Sim ple H ybrid Channel B orrow ing schem e (SH C B ) [2], the set o f channels assigned to a cell is divided into two groups, A and B. G roup A channels are local channels that can only be used to sever call requests inside the cell. N eighbouring cells can borrow channels o f group B w hich are “ borrow able” channels. The ratio A:B is determ ined a priority, the optim um ratio depends on the percentage increase in the ừaffic density.
The B orrow ing w ith Channel Odering (BCO) schem e (BCO ) [3] also divides the nominals channels into tw o groups, but the local to boưow able ratio varies dynam ically according to the cuưenl traffic conditions. T he channels o f the cell are ordred such that the first channel has the highest priority to be assigned to the next local call, and the last channel is given highest priority to be borrowed by neighbouring cells. I f the base station perform s this functionality, then the M SC needs to be informed about the resulting assignm ent. The M SC uses an adaptive algorithm to calculate and update each channel probability o f being b oưow ed, base on fraffic conditions. If the channel frequency is free in the three nearest co-channel cells, only then the channel is suitable for borrowing.
B o ư o w in g w ith D irection Channel Locking (DBCL) [4] is sim ilar to B CO w ith channel reassignm ent. H ow ever, D BCL uses an efficient w ay to lock channels. W hen a channel IS locked, it is locked only in the directions that w ould cause co-channel interference. C ells located tow ard the free directions can borrow or lock the channel.
SHCB schem e perform s better than FCA w ith light and m oderate ừaffíc. U nder heavy traffic load, the point w here SHBC outperform s FCA will be dependent on the A:B ratio. BCO and DBCL schem es ou tp erfo m ied FC A in all kind o f traffic conditions. D B CL outperform s BCO and also a dynam ic channel allocation schem e called Locally Optim ised Dynamic A ssignm ent Sừategy (LODA).
2.2. D istributed C hannel Borrowing Schemes
T hese schem es are D istributed Load B alancing w ith Selective Boưovving D -LBSB [5]. BCO and D B C L use centralised control inside the M SC. The M SC has to keep a record o f free, serving, borrow ed and locked channels and to label them w ith updated priority. The need for a continuous up- to-date global know ledge o f the entire m obile netw ork can lead to a slow response and a neavy signalling load. T o alleviate this problem , several authors have proposed m odifications to make the schem es m ore distributed.
D .IỈ. Tri et CIỈ. / V N U J o u r n a l o f Science, M ath em a tics - P hysics 24 (2008) 171-178 173
The D -LBSB schem e m igrates channels from a cell w ith available channels (called “cold cell” ) to an overload cells (called “hot cell”). Together with the borrow ing channel algorithm a ch an n el assig n m e n t stra te g y IS used.
Initially, c channels are allocated to each cell in the network. The classification o f a cell as hot or cold depends on its degree o f coldness dc. If dc <h (a determ ined threshold), then the cell IS hot, otherw ise it is cold. The determ ination o f h depends on the average call arrival and term ination rates o f the entire cellular netw ork, c and also the probability o f channel bo ư o w in g rates from the other cells. Typical values o f h are 0,2 or 0,25. The mobile user is clasified as new , departing or others according to the rules m figure 1.
The base station periodically m onitors the quality o f the received signal strength (RSS) from each user through special control channels. If RSS o f the user is less than a certain threshold, the user is w ithin one o f the shaded peripheral regions in the boundary o f a cell.
A cold cell cannot borrow channels and a hot cell canot lend channels. Three param eters determine the suitablility o f a cell to be a lender, L: degree o f coldness dci,, nearness D e l and hot cell blockabde H i l l . N earness is the cell distance between the borrow er cell B and the lender cell L. The hot cell blockade is the num ber o f hot co-channel cells o f the lender that are not co-channel cells o f the borrower.
@ R S S from A b ■elow thre^old Rt
r
Sian miier to coiuư / ĩime muts
>’
I f tim e r = Ĩ a n d R.SS conT uiues c le c re .iiu ii
Classify user as reset tưiìer.
“dq^nrtme” no changes
R iTtdius o f rhe cell I- widtli o f ứie sliacled reeioa RSS recen'ed Signal Sttenffth Rr RSS threshold
If RSS from B is sreater than RSS from A
and Classify user as ‘ new" iu B RSS &0I11 A o\-er Rt
Ỷ y
Starl tuner to count T tứne luut^ Reset tmwrs.
Classify user as “others" in A
Clas^if^^ user as '‘others’' m B
Fig 1. Classification of Mobile users.
The best lender is the cold cell in the com pact pattern that m axim ise the value o f this function:
F =__
D
Rcp 1
174 D .H . Tri e t a l / VN Ư J o u r n a l o f Science, M a th em a tics - P h ysics 24 (2008) Ỉ 7 Ị - Ỉ 7 8
Rep is the radius o f the com pact pattern in term o f cell distance, 1< D gi <Rcp. 0< Hfii < 6 for hexagonal cellu lar geom etry. Rep and 7 are used for norm alisation. The value o f the function IS p ro p o rtio n al to the d eg re e o f c o ld n e ss o f the c e ll and in v e rse ly pro p o rtion al to Db i and Hịịị . A n o th e r criterion is the lender L should not be hot after lending a channel.
T he base station stores the num ber o f departing users heading tow ard the ith cell in the i(]i elem ent o f a ư a y N u m D e p a rt. The objective o f this inform ation is to possibly b o ư o w a channel from the neighbouring cell i and assign it to departing user heading tow ards that cell.
A hot cell needs to borrow channels until it reach the average degree o f coldness dCavr^ W hen a c e ll b e c o m e s h o t, the n u m b er o f a v a ila b le ch an n els is h X c , the n u m b er o f ch an n els to be b o rro w e d (X) to reach dCavr can be canculated by solving the equation:
, _ h C + X
^ (2)
A s seen in figure 1, it can be assum ed r«R and the shaded portion o f the cell has an area approxim ately given by the product o f the perim eter (p) o f the cell tim es r. Supposing K is the average density o f m obile users m aking calls in a cell, it easy to derive that the num ber o f departing u sers is X X r. I f the u ser o f b o rro w e d channel is co n fin ed to d ep artin g u se rs o n ly, then K X p X r >
X and r is given by:
K x p (3)
Each cell keeps as local param eters its N C C (Set o f non-co-channel cells, considering its com pact pattern), c c (set o f co-channel cells), dc, //Ạrcc(set o f hot non-co-channel cells), //c c ( s e t o f hot co-channel cells).
Algorithm :
1. B sends m essages to the cold neighbouring cells L for w hich N u m D e p a rt[L ] is greater than 0. The m essage contains N C C , Hịmcc, -0^1=1 and requests the cell to com putes Fgi. Each L cell com putes Hbl, Fbl and send back to B.
2. B order the L cells by decreasing values o f Fbl and selects the cell with highest Fai. The selected lender com putes the set o f its co-channels, which are non-co-channel w ith B by com paring N C C w ith its ow n c c .
3. C hannels start to be borrow ed from the lender cell until the num ber o f borrow ed channels is equal to N u m D e p a rt[L ] or the basic criteria is violated, the lender becom e nearly hot, after lending the channel, the len d er cell insừxicts its co-channel cells w hich are non-co-channel cells o f B, to lock that frequency channel. The same procedure is excuted for the other cells in the listed order until the num ber X o f b o rro w ed channels is reached and the algorithm is term inated, or list o f cells IS exhausted, w hereupon it gives step 4.
4. B sends m essages requesting Fbl to the L ’ cells. The cold L ’ cells will answer.
5. B selects L ’ w ith highest Fbl. L ’ com putes the set o f its co-channels w hich are non-co- channel w ith B. Step 4, 5 are repeated until X reached.
Channel reassignment strategy:
T he set o f available channels can be divided into local and borrow ed channels. Hot cells have both, cold cells only local channels. The channel demand is classified into four classes:
C lass 1 requests have highest priority to receive a channel, these are han d o ff requests. It tries to m inim ise the p robability o f disrupting ongoing calls.
C lass 2 req u ests are the channel requests by originating calls.
D.H. Tri e t al. / VNU J o u rn a l o f Science, M a th em a tics - P hysics 24 (2008) 1 7 1 - / 7 8 175
C lass 3 requests are requests for channel re-assignm ents, they are requested by a cell site function that m onitors state o f the channels. The re-assignm ents are divided in two types: The rcassignm ents o f type 1 are for reassigning a departing user using a local channel to borrow ed channel, if the borrow ed channel IS not used to satify class 1 and 2 demands. R equests o f class 3 are for reassignm ent o f type 1.
C lass 4 request are re-assignm ents requests o f type 2.
A lthough the authors o f the D-LBSB scheme claim the channel assignm ent strategy prioritises h an d off requests are classified differently, the execution o f the channel assignm ent algorithm does not prioritise h a n d o ff requests. Each request IS treated independently and discarded if blocked. Therefore h an d off requests and incom ing call requests have same priority, because they are freated equally inside the algorithm
In D -LBSB, the borrow ing algorithm is not executed every time a call or h a n d o ff request IS m ade and there are not m ore available channels to accom m odate the request as is done in BCO and DBCL schem es. It is trigerred before the nom inal channels are all used, once h is reached. M oreover, it does not get only one channel, but a certain num ber o f channel X, the actual num ber depending on the average traffic load o f the w hole network. D-LBSB does not perform as well as BCO and DBCL, but it IS le ss c o m p le x and it p ro v e s to be m uch faster as the lo ad o f tra ffic in c re a se s co m p are d w ith its centralised version.
Simulation results:
C om m unication netw ork com m ercial sim ulator (O PN ET) IS used. The results show m the figures were obtained w ith 90% confidence interval and w ithin 5% o f the sample m ean. The sim ulated cellular system contains N =100 cells,
c=40
channels per com pact. Tw o m terger values X and y (l< x,y< 10) are used to describe the location o f the cells. The shift param eter s„ Sj w ere set to be 3 and 2 respectively, call arrival at each cell was assumed The hold time o f a call w as assum ed to be distributed based on an exponential distribution with a mean \l\i o f 500s, h=0,25. In order to determ ine the position o f user in a cell, a cell is m odeled as a circle w ith a grid o f size 100x360. A user locationIS determ ined by using a radial method, a pair o f (y,0) to indicate position o f a user, w here Y denotes the distance o f the use from the center o f the cell and 0 denotes the angle from a com m on reference line. It is assum ed that a user can m oves to a position over 100 from the center, a h a n d o ff occurs. The allocation o f the new user is given at random.
F ig 2. S im ulation result.
176 D .H . Tri et a l / V N V J o u rn a l o f Science, M a th em a tics - P hysics 2 4 (2008) Ỉ 7 Ỉ - Ỉ 7 8
The cell m odel used in this paper is expected to be more efficient than the other one in the literature since w hen a h an d o ff occurs the new possition in the new cell is d e a rly determ ined. The tim e period, a, used for determ ining a departing user was set to 10 tim es units. If a user keeps departing for at least 5 tim es in shadow region for the last period a, it is classified as a departing user.
O therw ise, it is a local user.
Figure 2 show the call blocking probablity o f the D-LBSB algorithm for various values o f'C , the num ber o f channels initially allocated to each cell under the fixed assignm ent schemes. The load level used here denotes the ratio o f the call arrival rate at one channel in a cell to the call service rate a channel. It w as assum ed that an arriving call in a cell is served im m ediately if there are any channels available for use in a cell. From figure 2 show that the value
c
has sữong effects on the call blocking probablity.3. D C A sch em es
D ynam ic channel assignm ent scheme (DCA) is to assign a channel to a call request with m inim um cost, bu t respecting the signal interference constraints. The cost is evaluated by a cost function. The cost function can be form ulated taking into account the future blocking probability in the vicinity o f the cell, the usage frequency o f the candidate channel, the reuse distance, channel occupancy distribution under c u ư en t traffic conditions, radio channel m easurem ents o f individual m obile users, average blocking probability o f the system. The differentiation factor in DCA schem es is form ulation o f the co st function.
3.1. Centralised D CA schemes
Locally optim ised dynam ic assignm ent scheme (LODA) [4] is a D C A schcm c whose cost function is based on the future blocking probability in the vicinity o f the cell. Sim ulation results com paring LO D A w ith D B CL show that DBCL has better perform ance.
Several D C A schem es form ulate a cost function that m axim ises the channel efficiency by optim ising the reuse channel distance packing. These schemes perform well in light and moderate traffic conditions bu t they are not able to m axim ise the channel reuse in heavy traffic load because the best candidates m ost probably are already serving call requests.
3.2. D istributed D C A schemes
The distributed D CA schem es are norm ally cell-based schem es or signal sừength m easurem ent based schem es. In cell based schemes, the channel assignm ent is perform ed by the base station. The base station (not the M SC as in centralised schem es) is responsible for keeping inform ation about c u ư e n t available channel in the vicinity. The local packing dynam ic disfributed channel assignm ent schem e (LP-D D C A ) [6] uses an augm ented channel occupancy (ACO) matrix for channel assiejim ent. The A CO m atrix contains all the local and vicinity inform ation needed for the selection o f a channel. The base station keeps the ACO m aừ ix updated. The LP-D D C A w ith adjacent channel interference constraint takes into account this kind o f channel interference when selecting a channel from A C O m a tìx . Both schem es provide near optim um channel assignm ent, but cause excessive exchange o f status infom iation betw een cells.
A nother signal strength m easuarem ent based scheme is the Channel Segregation scheme (CS) [7]. This schem e is a self-organised DCA. Each base station scans channels w hen selecting an
D J l Tri e! a!. / V N U J o u r n a l o f Science. M ath em a tics - P hysics 24 (2008) 171-178 177
available channel w ith acceptable signal interference. Each base station will attribute to cach channel a probability o f channel selection P(i). The channel “ selectability” order is perform ed independently by each base station and is review ed through learning methods. For each call request, base station selects channel with highest P(i). Then base station need to checks if the use o f that channel IS possible by m easuring its p o w er level. I f the pow er level is good enough then this channel is considered idle, allocated to serve the call request and its “selectability” increased. If not the channel is busy and P(i) is decreased. If all channels are busy the call is blocked. The
cs
schem e is autonom ous and adaptive to change in traffic load. Sim ulation results show that blocking probability is greatly reduced com pared to FC A and D CA schem ses and quickly reach a sub-optim um channel allocation,cs
uses the channels e ffic ie n tly and re d u c e s the n eed fo r chan nel reallo catio n due to in terferen ce,cs
IS a g o o d so lu tio n fo r TDM A/FDM A.4. Flexible channel assignm ent schemes FICA
The c e ll sites have a sufficient num ber o f pre-assigned nom inal channel to accom m odate light traffic. The rem aining channels are kept in a central pool and assigned to cell sites in need. The dynamic assignm ent can have a scheduled or predictive approach. In the scheduled approach, the assignment o f channels is m ade at determ ined peaks o f traffic. In the predictive approach, the traffic intensity IS m easured constantly at all cells and the M SC can reallocate the channel at any time. Ratio o f fixed and dynam ic channels is a significant param eter that defines the perform ance o f the system.
For heav>' traffic loads FCA gives better blocking probability than FICA, FC A w ill make better use o f the minimum reuse distance than FICA.
5. Conclusion
FCA is too lim iting for m obile netw orks and several strategies have been proposed to m axim ise frequency channel allocation and m inim ise call blocking probability. D CA schem es perform better under low tTaffic intensity. M odified FCA schemes have superior perfonTiance in high traffic loads.
DCA schemes use channels m ore efficiently and for the same blocking rate have a low er forced call tcrminaticn than FC A schem es. In D CA, there is not pre-assignm ent o f frequency channels to the cells o f the cellular netw ork. All frequency channels are kept in a central pool. W hen there is a channel request in one base station, the M SC chooses the appropriate frequency channel that gives m axim um channel efficiency taking into account all the signal interference consfraints. The chanels are assigned for the duration o f a call, after the call has finished, the channel is returned to the central pool or reallocated to a m obile user inside the same cell site that was controlling the channel before.
However, the near optim um channel allocation is the at the expense o f high overheads through its use o f ;en tralised allocation schemes. This overhead mean that such schem es a re n ’t practicable for large netvorks. D C A schem es w ith lim ited inter-cell com m unication suffer less overhead, but need to sub-optiư.um allocations. Such schem es are being proposed for m icrocellular system s as this cell structure illow s inter-cell inform ation sharing by interference m easurem ent and passive non-intrusive monitoring at each base station. For m acrocellular systems, FC A w ith channel borrow ing offers good results a n l less com putational com plexity than DCA. H owever, FCA variant schem es w ith best result use centTclised confrol inside M SC, although they are less com plex than DCA there is still a need to maintain ip -to -d ated global know ledge o f entire m obile netw ork, leading to a slow response and a heavy signalling load.
178 D .H . Tri et aỉ. / V N U J o u rn a l o f Science, M a th em a tics - P h ysics 2 4 (2008) 7 7 / - / 78
N ext section, w e propose m ethod to determ ine num ber o f channels need to borrow when a cell becom e hot in D -LB SB schem e and the channel priority boưow ing. In D -LBSB schem e, the channel is bo ư o w ed from neigbough cells only, but for us, when uses move to any direction, it m ay not stop m a neigbough cells, it stop in any cell in com pact pattern. So, w hen a cell becom e hot, algorithm IS propose that allow hot cells borrow frequecy channel in any cell in com pact pattern. In this method, num ber o f channel need to b o ư o w is X^C-idc^'^-h), in fact, w hen a cell becom e hot, it should be to b o ư o w channel until it reach average degree o f coldness o f netw orks, X=C.(dc^'^-dc). In next paper, we are going to present our algorithm and result o f simulation. .
References
[1] I. Katzcla, M. Naghshineh, Channel Asignment Schcmes for Cellular Mobile Telecommunication Systems: A Comprehensive Survey, ỈEEE Personal Communications Magazine Vol 3, No 2 (1996) 10.
[2] T J. Kahwa, N.D. Georganas, A Hybrid Channel Assigment Scheme in Large-Scale, Cellular-Structurc Mobile Communication Systems. IEEE Trans on Communications Vol. 26, No 4 (1978) 432.
[3] Said M. Elnoubi, Rajcndra Singh and Somcshwar c . Gupta, A new frequency channel assignment algorithm in high capacity mobile communication systems. ỈEEE Trans on Vehicular Technology Vo!. VT-31, No 3 (1982) 125.
[4] Ming Zhang, Tak-Shing p. Yum. Comparisions of Channel Assignment Strategies in Cellular Mobile Telephone Systems. IEEE Trans on Vehicular Technology Vol. 38, No 2 (1991) 387.
[5] Sajal K.Das, Sanjoy K.Sen, Rajeev Jayaram and Prathima Agrawa!. A Distributed Load Balanciníĩ Algorithm for the Hot Cell Problem in Cellular Mobile Networks. Proceedings of the 1997 6’^ ĨEEE International Symposium on ỉỉiglì Performance Distributed Computing (\ 9^1) 254.
[6] Ĩ. Chih-Lin, Pi-Hui Chao, Local Packing - Distributed Dynamic Channel Allocation at Cellular Base Station.
Proceedings o f IEEE Global Telecommunications Conference (GLOBECOM'93), part /. Vol. 1, New York, NY, USA (1 9 9 3 )2 9 1
[7] Y. Furuya, Y. Akaiwa, Channel Segregation: a Distributed Channel Allocation Schemc for Mobile Communication Systems, ỈEĨCE Trans Vol 74 (1991) 949.
