AssTRACT Tis arte ponies a dese dsc umber of pushed papers ithe irl source sa hanael allocation schemes. The authors provide a survey fa ‘of ine, dyunic and nbd location schemes snd compare thet ead of sof ‘compli an purormance. We ao ivespne thse chanaa locaton schemes based om other flo such isirbuted centralized eonvol and adapabliy to wae conditions Moreover we provide a detaled discussion 00 see partitoning schemes, he ffs of handofl and priiiztion schemes inl, we deus othr important sues in resourealoation sich a ve el frequency plnsing and owe Channel Assignment Schemes for Cellular Mobile Telecommunication Systems: A Comprehensive Survey |. KATZELA AND M, NAGHSHINEH echnological alvances and rapid development of handheld wireless terminals have facilitated the rapid srowth of witeless communications and mobile com puting. Taking ergonomic and economic factors ino account, and considering the new trend in the telecoramunications industry to provide ubiquitous information aeves, the popula tion of mobile users will continue to grow at a tremendous rate. Another important developing phenomenon is the shift ff many applications to multimedia patforms in order to pee- sent information more effective “The tremendous growth of the wireless/mobile user popu- lation, coupled with the bandwidth requirements of multime- sia applications, requires efficient reuse of the scarce radio specitim allocaied to wirelessimobile communications, Ef lent use of radio spectrum i also important fram a cost-of- service point of view, where the number of base stations required to service a given geographical area is an important factor. A reduction inthe numberof base stations, and hence in the cost of sexvice, can be achieved by more efficient reuse ‘tthe radia spectrum. The basie prohibiting factor ia radio spectrum reuse is interference caused by the environment oF ‘ther mobiles, Interference ean be reduced by deploying eli ‘ent radio subsystems and by making use of channel assign tment techniques Tn the radio and transmission subsystems, techniques such as deployment of time and space diversi stems, use of lw noise filters and efficient equalizers, and deployment of eff cient modiation schemes canbe used to suppres interference land to extract the desired signal, However, co-channel inter Terence caused by frequency reuse isthe most restraining fac- tor on the averall system capacity in the wireless networks, tnd the min idea behind channel assignment algorithms i 0 make use of radio propagation path los [1,2] characteristics fn order ta minimize the carrer-to-interference ratio (CIR) ‘and hence increase the radio spectrum reuse efficioney. “The focus ofthis article isto provide an overview of differ. cent channel assignment algorithms and compare them in fe1ms of performance, flexibility, and complenity. We fest start by ging an overview of the channel assignment problem ina cellular environment and disouss the general idea behind sajor chaane allocation schemes, Then we proceed to discuss ‘ferent channel allocation schemes within each category. Channel Allocation Schemes What Js Channel Allocation? A given radio spectrum (or bandwidth) can be divided into a Set of disioint oF noninterfeing radio channels, All such chan nels ean be used simultaneously while maintaining an accep able received radio signal. In order to divide a given radio Spectrum into such ehanels many techniques sich a8 tre {quency division (FD), time division (TD), or code division (CD) can be used. In FD, the spectrum is divided into disjoint Trequency bands, whereas in TD the channel separation is achieved by dividing the usage of the channel into disjoint time periods calcd time slots. In CD, the channel separation isachieved by using different modulation codes. Furthermore, more elaborate techniques can be designed to divide a radio spectrum into 2 set of eisjoit channels based on combining the above techniques, For example, a combination of TD and FD can be used by dividing each frequency band of an FD Schieme into time slots. The major driving factor in determin. ing the number of channels with certain quality that ean he used for a givon wireless spectrum is the Tevel of received sige al quality that ean be achloved in each channel Let $;(K) be denoted as the set (7) of witeiess terminals that communicate with each other using the same channel k By taking advantage of physical characteristics of the radio environment, the same channel k can be reused simultaneous Iy by another set if the members of sets and j sre spaced sufficiently apart: ALI such sets which use the stme channel ae referred to a8 co-channel sts or simply co-eharels. The ‘minimum distance at which co-ehannels can be reused with "tn prucice, cach channel ean gencrte some interne the aac hanes Hover the eft of sch merece canbe rec ode quate adsacen chan eparion. 107.991696805 00.6 196 IEEE TEE Personal Communiatons * June 1995acceptable interference is called the “eovchannel reuse dane" “rhs posse because due 0 propagation path os inthe radio Environment. the average power 55 depending onthe piel Shvronment any the erage {Panamiter power: For example, foram indoor emianment wih te average power at dis tance Ais ahout percent of the imerage poner resiea a dance ahs ajnting the ease ter poe level anor the dance Seinson cochanmsn s cannelcan Deve tye number of sh tet the CIR in eck crchonel is above the required vale Ci ere he caer () represents he eceed sgn powctin «channel andthe nerterence (represent he Sim ot eve sol powers all coche. "Asan example consider Fg where a wicess tation labled in at stance om anster station beled rng areowand radio channel, We eet tate aso hannel wed by to commanieate ro asthe rterence Shane In thre we havea shown ve ober sons Inbled 1,2, which ue the sme channel the see BBenotng the tansmited pow of station by Pa the i tance of tation frm Ryd he average CIR tthe ree. Sie ation en crn" Leds +No here represents the environmental noise To achieve a Cerin ee of CIR at the eernce ato ere et tsivean be wad For xnmple the Utance ete tations P32 Snine the ocala th eterence station Cane ietesed to rede the co-channel nrfrence lve Many channel locaton schemes are hase om thi ia of physical sparaton Another solution to rece the CR at 8 {Moreau the teeing power uns fom fe as anor to tees the desired sana power Ith ea behing power control scheme These two methods presen the underiing concep for channel "Signment sipritins in cellar stems Ech of hee so Fahmy uses afferent method t achieve a CIRyi st etch ‘oe trminsl by soaring co-channss anor aston th trannies Different Channel Allocation Schemes Channel allocation schemes can be divided into a number of sifferent categories depending on the comparison basis. For ‘example, when channel assignment algorithms are compared based on the manner in which co-channels ate separated they an be divided into fixed channel allocation (FCA), dynamic hannel allocation (DCA), and hybrid channel allocation (HCA), ‘In FCA schemes, the area is partitioned into 2 number of cells, and a number of channels are assigned to each cell according to some reuse pattern, depending on the desired Signal quality. FCA sehemes are very simple, however, they do ‘not adapt to changing traffic conditions and user distribution In order to overcome these deficiencies of FCA schemes, Vas oh i: Figure 1 imerfeence. ow Interfering DCA strategies have been intro- duced, Tn DCA, all channels are placed in a pool and are assigned to new calls as needed such that the CiRjun criterion i satisfied, At the cost'of higher complexity, DCA, Schemes provide Mexbilty and traf fic adaptability. However, DCA strategies are less efficient than FCA under high load conditions. ‘To overcome this drawback, HCA techniques were designed bj com> bining FCA and DCA schemes. Channel assignment schemes can be implemented in many dif- ferent ways, For example, a chan rel can be assigned to's radio cell based on the coverage area of the radio cell and its adjacent cells such thatthe CLR is main {ained with high probability ia a radio cells. Channels could be also assigned hy taking the local CIR measurements of the rmobile's and base station's receiver into account. That is, instead of allocating a channel blindly to # cell based on worst-case conditions (sach as leting e0-channels be located at the closest boundary), a channel can be allocated to a ‘mobile based on its local CIR measurements [3,4 ‘Channel assignment schemes can be implemented in cen- tralized or distibuted fashion. In the centralized schemes the channel is assigned by a central controller, whereas in dis- tributed schemes a channel i selected either by the local base Station of the cell from which the calli initiated or selected autonomously by the mobile, In a system with cell-based con- trol, cack base station Keeps information about the eurrent le channels sity, Here the chanel valaity information is updated by exchange of status information between base stations Filly in autonomously oreazed die tributed schemes, the mobile chooses ehanacl hase on its Toca CIR measurements without the ivoement of a sna Call assignment entity. Obviously this scheme has much lower complesiy a the cow of loner efficent impor tant co note that channel assignment ased on local asi tment canbe doe for both FCA snd DCA schemes Fixed Channel Allocation the FCA stratay a sc of nomial chanel is permancet Jicsiossctto scarce ne Hoe det tte celationship is assumed between cach channel and cach clin acondanee to corchannel ese consans (5-12 “Th ttl nombor of salle channel inte sytem Ci livid into ses, and the minimum number of chanel Sets N required t serve the entire coverage area is related tothe reube distance follows [6 12} N = (1/3)0?, for hexagonal cells e Here a in defincd ax DIR, where Ry the rads ofthe ccll and Dis the pyscal distance between the two cell sn ters [5.4 can asume only the integer values 347,95 00, teneraly presented bythe series + J)2 wlth and) boing integers [87], Figures 24 and 2b give the allocation of Channel ses to eels for = 3 (0 = 3)andN = 7 (@ = 44), respect. Tn the simple BCA strategy, the same numberof nominal chanoels is allocated to cach el This uniform chann! ste Uti is efficient if he traffic dstbution af the sytem ts ke uniform In that cae, the oneal averape Hocking prob IEEE Personal Communsstion * June 196 "Figure 2.0) 8 bility ofthe mobile system is the same as the call blocking probability in a cel, Because trafic in cellular systems can be Fonuniform with temporal and spatial uctwations, uniform allocation of channels to cells may result in high blocking in some cells, while others might have a sizeable number of Spare channels. This could result in poor channel utilization, Wis therefore appropriate to tailor the number of channels in ‘cell to match the load init by nonuniform channel allocation (13, 14] oF susie borowing (15, 16) in nonuniform channe! allocation the number of nominal chanaeis allocated 10 each cell depends on the expected traf profile in that cll. Tavs heavily loaded cels are assigned Shannels than ightly loaded ones, In 13} an algorithm, nonuniform compact patter allocation, is proposed Tot allocating channels to cells aecording (9 the trafic distribution in each of them. The proposed technique attempts to allocate | channels to cells in such a way that the average blocking probability inthe entire system is minimized, Let there be N eelis And M channels in the system, The alloce- tion of a channel to the set of co-channel cols forms a pattern which i refereed to as the allocation pattem (13) In adlition, the compact allocation pattern ofa channel is defined as the pattern with minimum aver tage distance between cols, Given the traffic loads in each of the cells andl the powsible ‘compact pattern alloations forthe MY ehan- nels, the nonuniform compact pate alloca tion algarithm attempts to find the compatible compact patterns that minimize the average blocking probability inthe Entire system as nominal channels are assigaed one at time ‘A similar technique for noamniform channel allocation is also employed in the algorths proposed in (14. Simulation results in [13] show thatthe blocking probabil ty using nonuniform compact pattern allocation is always lower than the blocking probability of uniform channel alloca tion. Ie is interesting to note that the reduction of blocking probability almost uniformly 4 percent for the range of traf fie showa in {13} Also for the stme blocking probability the system can carry, on the average, 10 percent (maximum 22 Percent) more traffic with the use of the nonuniform pattern allocation [13] In the static borrowing schemes proposed in (15, 16) unused eiannels from lightly loaded cells are reassigned to heavily loaded ones at distances = the minimum reuse dlistance 6. Although in static borrowing schemes channels are permanently assigned to cells, the number of nominal thaninels assigned ip each cell may be reassigned periodical- ly according to spatial inequities in the load. This can be 2 Call ava aes of 20-200 als’ foreach el aiigure 3. Chanel ick done in a scheduled or predictive manner, with changes in traffic known in advance or based on messurements, respec tively Channel Borrowing Schemes In sehannel borrowing scheme, an acceptor cll hat has used all ts nominal channels can bortow free channels from is eghboring cells (donors) to accommodate new el, A cha tel can be borrowed by 4 eel the borrowed channel does tot interfere wth existing cals, When channel is borrowed, Several other cells are prohibited from sing i. Thief called channel oct. The number of sch els depends on the cal layout andthe ype of inital allocation of channel 0 eels For example, ora hexagonal planar layout with reuse is fance of one cll (a = 3) borrowed channel stocked ip {ree additonal neighboring cll a is shown in Fig 3, while tora one-dimensional lyout or hexagonal planar gid layout swith two-cell reuse distance, it fs Tocked in two additional nelghboring cel in contest o sate borroving, channel borrowing state- ses deal with short-term allocation of borrowed channel fells once a call is completed, the borrowed channel is returned tots ominal ell. The proposed channel borrowing Shiemesdifler in the way a free channel is selected from & donor eel 0 be Borromed by an aceptor cel "The channel borrowing scheoes can be ded into simple an tsb, In ipl channel borrowing schemes, ay nominal ‘hate ina cellcan be borrowed by 2 ci boring eel or temporary ese. In hybrid Chantel borrowing strategies, the se of hannelsasigned to each cel is divided {nto eo subsets (standard or local chan- ck) and (nonstandard or borrowable Channels). Subset fs for use only i the sominally asigned cell, while subset Bis Hllowed to be lent to neighboring cell. ‘Table summarizes the channel borrowing schemes proposed inthe literature. Inthe beat ewo aubbections we discuss the simple fd hybrid borrowing scenes ip deta Simple Channel Borrowing Schemes ~ ta the simple borrowing (SB) strategy [15-20], ‘nominal channel set fs assigned to a cell, as in the FCA tase, After all nominal channels are used, an available chan= hel from a neighboring cell is borrowed. Ta be available for borrowing, the channel must not interfere with existing cll, Althougit channel borrowing can reduce eal blocking i ean cause interference in the donor cells from which the channel Is borrowed and prevent future call in these cells from being, completed (21) ‘As shown in [20], the SB strategy gives lower blocking probability than static FCA under light and moderate trafic, But static FCA performs better in heavy traffic conditions, ‘This is due tothe fact that in Tight and moderate tafe conde tions, borrowing of channels provides a means to serve the ‘uetiations of offered traffic, and as long as the attic inten sity slow the number of donor cells is smal. In heavy trafic, the channel borrawing may proliferate to such an extent, die to channel locking, that the channel usage efficiency deops Arastically, causing an increase in blocking probability and a ‘deerease in channel utization [23 Because the set of borrowable channels ina cell may con- tain more than one candidate channel, the way & channel is selected from the set plays an important role in the perfor. mance of s channel borrowing scheme. The objective of all the schemes is 10 reduce the number of locked channels, 2 IEEE Personal Communations June 1996caused by channel borrowing. The difference between them is the specific algorithm used for sclecting one of the candidate channels for bor towing. Alogg these fines, several variations of | the SB strategy have been proposed where chat nels are borrowed from nonadjacent cells [13, 15, 16,17, 19, 20), Ia the following, we discuss briefly cath of the proposed schemes, Sorrow from the Richest (SBR) ~ In this | scheme, channels that are candidates for borrow ing are available channels nominally assigned to ‘one of the adjacent cells of the acceptor cel [15]. {Ef more than one adjacent cell as channels avail- able for borrowing, a channel is borrowed from the cell with the greatest number of channels available for borrowing, As discussed earlier, channel bortow- ing can cause channel locking, The SBR scheme does not take channel locking into account when choosing a candidate eh rel for borrowing. aT. Basic Algorithm (BA) — This is an improved version of the SBR strategy which takes channel locking into account when selecting a candidate channel for borrowing [15, 16]. This Scheme tries to minimize the future call blocking probability in the col that is most affected by the chanael borrowing. As in the SBR case, channels that are candidates for borrowing are available channels nominally assigned t0 one of the adja Cent cells ofthe acceptor cell. The algorithm chooses the ean- ddidate channel that maximizes the number of available ‘nominal channels in the worst-case nominal eelP ia distance & to the acceptor call Basic Algorithm with Reassignment (BAR) ~ This scheme provides forthe transfer of a cal from a borrowed channel 10 a nominal channel whenever a nominal channel becomes available. The choice of the particular borrowed channel to be freed is again made in a manner that minimizes the maximum probability of future call blocking inthe cell most affected by the borrowing, as inthe BA scheme [15 Borrow First Available (BFA) ~ Instead of trying to optimize when borrowing, this algorithm selects the first eandidate channel it finds [15]. Here, the philosophy of the nominal channel assignment is also different. Instead of assigning, channels directly to cells, the channels are divided into sets, and then each set is assigned to cells at reuse distance 0: ‘These seis are numbered in sequence. When setting up a call, channel sets are searched in « preseribed sequence to find a cundidate channel Performance Comparison ~ A general conclusion reached by most studies on the performance comparison of the previous Schemes is that adapting a simple test for borrowing (e.8.. | Snle chanel borrowing yond channal bowrowing V- Channel booming schemes. Simple bettowing (58) Borrow from the richest (SBR) L Basie algorithm (6a) Basie algorithm with reassignment (BAR) | Borrow fst avaliable (FA) Serco oro ems EH ‘Borrowing with channel ordering (BCO) | Coe ee paca somes Se me tn booing nd oan | sce Sm et ogo wih ee ae borrowing the first available channel that satisfies the or can- straint) yields performance results quite comparable 10 $s tems which perform an exhaustive and complex search method 1 find a candidate channel [13, 15-17], SBR, BA, and BFA, were evaluated by simulation in [15] using a two-dimensional hhexagonal col avout with 360 service channels, The offered mad was adjusted for an average blocking of 0.02. The results show tht all tree schemes exhibit nearly the same average blocking probability versus oad with about 25 percent increase in offered load to achieve an average blocking of 0.02. The [BFA has an advantage over the other two in that its comput ing effort and complexity are signiticanty less. Here the com- plexity of each algorithm is determined based on the average fhumber of channel tests per call while searching fr a candi- date channel to borrow. In [15], simulation results showed a large variation in the complexity of these algorithms depend- ing on network load. For example, fr a 20 percent increase in the traffic, SBR requires 50 percent, and the BA 100 percent, ‘more channel tests compared to BEA. A summary ofthe com. parison results between the BFA, SBR, BA, and BAR Schemes is given in Table 2. Hybrid Channel Borrowing Schemes ~ In the following we ‘will describe diferent hybrid channel borrowing sciemes. ‘Simple Hybrid Channel Borrowing Strategy (SHCB) ~ In the SHC strategy {5, 13, 17] the set of changes assigned to cach cell is divided into two subsets, (standard) and B (borrow able) channels. Subset 4 is nominally assigned in each cell, While subset B is allowed to be lent to neighboring cells, The Fatio [4:18] is determined a priori, depending on an estima- tion of the traffic conditions, and can be adapted dynamically 1 scheduled or predictive manner [17} Borrowing with Channel Ordering (BCO) ~ The BCO, intro- ‘duced in [20] and analyzed in [13, 17), outperforms SHCB by ‘dynamically varying the local to borrowable channel ratio ‘according 10 changing traffic conditions [17, 2]. In the BCO. strategy, all nominal channels are ‘ordered sue that the first channel has the highest priority for being assigned to the next local cal | the last channels given the high- Borrow from the richest (SBR) Moderate | Moderate | Few j Ga neaeeny lea eran [basic algritho (6A) High Moderate | Alot of CO tty called CO i algorithm wit easignmant (BAR) | High Modeate | Alot Brow fst ovalble (FA) tow ‘ow | Very few ee , '§ Table 2. Comparison between BEA, SBR, BA, and BAR. IEEE Personal Communications * June 19% ‘nominal ase ares noma ell| ponoc ca lular handoff, that is, immediate reallo cation of a released high-rank channel tora call existing ina lower-anke channel in order to minimize the channel locking elect. Borrowing with Directional Channel Locking (BDCL) ~ In the BCO strategy, ‘a channel is suitable for borrowing only fit is simultancously free in three sarby co-channel cells. This tequiee iment is too stringent and decreases the rumber of channels available for bor wing. In the BDCL. strategy, the chan- nel locking ia the co-channel cells Is restricted to those directions affected by the borrowing, Thus, the number of thannels available for borrowing is greater than that in the BBCO strategy. To determine in which case a “lacked” channel ean he borrowed, “lock directions” are specified for each locked channel. The scheme also incorporates reallocation of calls from borrowed to nominal channels and between bor- owed channels in order 0 minimize the channel borrowing ff future calls, especially the multiple-channel borrowing ‘observed during heavy traffic, Performance Comparison — s shown by simulation io [13] BDCL ghes the lowest blocking probabiliy, followed by BCO and FCA, for both uniform and nonuntlorm traffic. The reduction of the blocking probability for BDCL and BCO, ‘over FCA for the system in [13] 6 almost uniformly 0.04 and (0.08, respectively for the range of traffic loa tested, Note that the nonuniform pattern allocation FCA scheme, siscussed in the previous section, ean be also applied in the case of the hybrid channel borrowing strategies. With the use ‘of nonuniorm patter allocation the relative performance of the BDCL, BCO. and uniform FCA schemes remain the same as before, but the trafie-carrying capacity ofa system can be Increased by about 10 percent. This advantage isin addition to thore guined from the chanel harrowing strategies [13]. A summary of the comparison results between the BCO, BDX and FCA schemes is given in Table 3. Sharing with Bias (SHB) — In [23] SHB was proposed: a Scheme of chanel borrowing with coordinated sectoring. The SHB strategy s similar to the join biased queue role [24], which isa simple but effective way t balance the load of servers in the presence ‘of unbalanced traffic. Fach cel in the system is divided in three see ,Z, 98 shown in Fig. & initiated in one of these sectors can borrow channels trom the evo adjacent cells neighboring it (donor cells), In addition, the channels in donor eels ate divided in two subsets, and B, as in the SHCB case, Channels from set A can only be used inside the Toffc cried eapaciy Channel utilization donor eell, while channels in set ‘can be loaned to an acceptor cell. Compal omploaly Crane erereapaet |e oe in U5] consis of 9 hag onal els where each cll saat’ 10 ‘harels and fila oying fom 20-200eallh, onc, Figure 4 Sharing wih bias Table 4. Comparison beoween FCA, CARB, and ‘An example is shown io Fig. 4, A call | initiated in sector X of cell number 3 ‘ean only borrow a channel from set of the cells numbered Land 2. Channel Assignment with Borrowing and Reassignment (CARB) ~ The CARB scheme proposed in [16] is sta {istially optimum in a eertain min-max sense. Here channels are borrowed on the basis of causing the least harm to seighboring cells in terms of future eal blocking probability. Likewise, reassign sient of borrowed channels done in way to cause maximum relief to neighboring cells Ordered Channel Assignment Scheme with Rearrangement (ODCA) - The ODCA scheme, proposed in [25], combines the merits of CARB and BCO with improvements to yield higher performance, In ODCA, when call requests service, the base station of the cell checks to see if there are ay nom: inal channels available. If there are channels available, the user willbe assigned one on an ordered basis, as in BCO. Here all channels are numbered in predetermined order according to the same criterion as in the CARB scheme, and the lowes ‘numbered availale idle channel is always selcted. If all rom: inal channels are busy, the cell may borrow a nonstandard charnel from a neighboring cell, Once t nonstandard channel is asigned, the availability lists of all affected cells where the assigned channel ean cause interference ate updated, When= ever a channel is no longer required, the avallability lists of the affected cells are updated accordingly. Whenever ‘dard channel is available, the channel re fs initiated to ensure efficient utilization, If there is & nonstaa- ard channel in use in the cel the ell served by that channel [s switched to the newly Irced standard chan, the necessary availabilty lists are also updated. If no nonstandard channels fre used in the cel call served by a standard channel with lower priority than the newly freed one is switehed to the newly freed channel [25]. Performance Comparison ~ The performance of ODCA was studied in [25] fora highway mierocellular environment with nonuniform tele-traffic load. Performance comparison with the FCA and CARB shows signifi- cant improvement. The ODCA ‘Scheme exhibits better channel uti lization compared to the CARB and FCA; the ODCA scheme | Fea. sco, soc, = forms beter than CARD and FCA Cae at blocking probabilities below 0.1 For example, ata blocking proba bility 0f 0.05 ODCA is capable of supporting 4 percent more traffic than CARB and 35 percent mare lealie than FCA [25}, However, the ODCA scheme incur a higher com- putational overhead in asigning and FA CARS. ODCA Feassigning channels, anid more fe- es ‘quent switching of channels due 40 ‘OBER Cafe, FER the reassignment propagation effect. > “The performance Comparison ress between ODCA, CARB, and FCA schemes are summarized in Table 4, Finally, a summary of the com parison between FCA schemes is piven in Table 5 >| “ IEEE Personal Communisatios + June 1996Dynamic Channel simiefca tow vow | exter tan gai nd yd ertowing in heavy wate Allocation = : \oshectecn vemporat boron Low moderate | Moderate | Batter than FEA id gpa variations of eat, Simple channel borewing| Modeatenigh|High | eter han FCA and stat borowing fic incellvlar systems, FCA inlightand moderate vate seems af notable fo attain High fybrd channel borawing| Moderate | Moderate | Better than FCA ight and this, DCA schemes have been stud- oe jied during the past 20 years. In con- borrowing in heavy loads : trast 10 FCA, there is no fixed relationship between channels and cls in DCA. All channels are kept ina central pool and are assigned dynamically to radio cells as new calls arrive in the system (18, 26}. After a cal is completed, its channel is returned to the ‘entra poo! Th DCA, a channel is eligible for use in any eell provided that signal interference constrains ate satislied. Because, in general, more than one channel might be available in the cen- {tal poo! to he assigned toa cell that requires a channel, some Strategy must be applied to select the assigned channel [10] The main idea ofall DCA schemes isto evaluate the cost of using each candidate channel, and select the one with the ‘minimum cost provided that certain interference constraints fre satisfied, The selection of the cost funetion is what differ: entiatos DCA schemes (10) The selected cost function might depend on the future blocking probability inthe vicinity of the eel, the usage fre~ ‘quency of the candidate chanoel, the reuse distance, channel ‘sccupaney distribution under current traffic conditions, radio ‘channel measurements of individual mobile users or the aver- ‘age blocking probability of the system [22]. “Although many claims have been made about the relative performance of eich DCA scheme to one or more alternative Schemes, the trade-off and the range of achievable capacity ‘tins are still unclear, and questions remain unanswered: How ‘does each dynamic scheme produce its gain? What are the basic trade-offs? Why do some schemes work only under cer- tain traffic patterns? Can different schemes be combined? ‘What i the Value of addtional status information of the near- Table 5. Comparionbiween id anne alocaionschemes by culls? What is the best posible use ofthe bandwidth [18]? Based on information used for channel assignment, DCA. strategies could be classified either as eallby-cll DCA oF adap- tive DCA schemes (27 In the cal-by-call DCA, the channel ‘assignment is based only on current channel usage conditions in the service area, while in adaptive DCA the channel assignment is adaptively carried out using information on the previous as well a present channel usage conditions (27,28) Finally, DCA schemes can be also divided into centralized and distributed schemes with respect to the type of control they ‘employ, Table 6 gives a list of the proposed DCA schemes. Centralized DCA Schemes In centralized DCA schemes, a channel from the central pool is assigned to a call for temporary use by a centralized Co troller. The difference between these schemes isthe specifi cost function used for selecting one ofthe candidate channels for assignment. First Available (FA) ~ The simplest of the DCA schemes is the EA strategy. In FA the first available ekannel within the reuse distance encountered during a channel search is assigned to the call. The FA strategy minimizes the system computa- tional time; and, as shown by simolation in [10] fora linear cellular mabyle system, provides an inerease of 20 percent in the total handled traffic compared to FCA for low and mod- erate traffic load. Locally Optimized Dynamic Assign- ment (LODA) — In the LODA Strategy [13, 17] the selected cost entrabend OCA Fist vale) function based on the future lock | (ocaly optima dynamic assignment (LDA) ing probably inthe vicinity of the | SSectonvthmanmum usageontnerevse og (ONG) | cel in which acl fina. eon square 0) Nesrest neighbor (NN) Channel Reuse Optimization Nearest neighbor #4 (NN +1) Schemes ~The objective of sny ive mobile system isto maximize the Disrbuted DR ocaly packing dtbuted DCA P-DDEAY efficiency of the sytem, Maximum Locally packing datbutedD tffcieney is equivalent to maxim | aes Utilization of every channel ia the [Gi mensurement OCA schemes usin ‘Dynamic chanael selection (OCS) Chane! segregation ‘win Minimum interference (Ma) ic——_—-} ‘One Dimension Systems Table 6, Dynamic channel location schemes. IEEE Personal Communications * June 1996 ‘Sequential channel search (8S) Random minimums interference (0) Random minimum interference wth reassignment (RMI) Sequential minimum interference SM ) sstem. It is obvious that the short- er the channel reuse distance, the treater the channel reuse over the twhole service area. The cost func- Hons selected in the following schemes attempt to maximize the | cele a the astem by optim ing the rouse of a channel in the sgstem area Selection with Maximum Usage ‘on the Reuse Ring (RING) ~ In the 15RING strategy [10], a candidate channel is selected which isin use in the most cells inthe co-channel -—=—— set. I more than one channel has | Fo%eed termination this maximum usage, an arbitrary _rate selection among such channels is made to serve the call. It none is available, the selection is made based on the FA scheme. Cored wate ‘Mean Square (MSQ), Nearest Neighbor KN) Nearest Neighbor plus One (NN + 1) — The MSQ Scheme selects the available anne) that minimizes the mean square ofthe distance among the cells using the same channel. The NN strategy selects the available ‘channel occupied in the nearest cel in distance > @, while the NN-+ 1 scheone selects an eligible channel occupied in the near est cell within distance 2 6 + 1 or distance @ if aa available ‘channel 8 not found in distance @ +1 (10) Performance Comparison ~ Computer simulations of FCA, MSO, NN, and NN + T strategies show that under light trafic conditions, NN exhibits the lowest blocking rate, followed by MSQ, FA, and NN 1 [27], Aso, the NN + 1 srategy, when applied to 4 mierocellular system, leads to lower forced call termination and chanel changing because the mobile unit is mote likely to keep the same channel When it moves to sn Adjacent cll [29]. In addition, simulation results of FA, RING, and NN (10, 30] show that for both one- and two-dimensionsl mobile s- tems, all of the above schemes operate at very low blocking rates until the offered traffic reaches some critical value. A small increase in the offered traffic above ths value prolaces ‘considerable inerease in the blocking probability of new calls and results in very little increase in the trafic carried by the system; the load at which blocking begins to aecur in one dimensional systems [30] is somesthat greater than that in {vo -dimensional systems [10], Finally, the simulation results in [30] show that strategies like RING and NN, which use ‘hannel reuse optimization approach, are able to carry 5 per cent more traffic ata given blocking rate of 3 percent com pared to a channel assignment strategy like FA, which docs ‘ot employ any channel reuse optimization. A surat’ of the performance comparison of the channel reuse optimization Selyemes is given in Table 7 1-Clique ~ Al four previous schemes employ local channel reuse optimization schemes. A global channel reuse optimiza tion approach is used in the I-elique strategy. The I-clique scheme uses a set of graphs, one for each channel, expressing the non-co-channel interference structure over the whole ser Vice area for that channel, In each graph a vertex represents 4 cell, and cells without co-channel interference are connevted ‘with edges. Thus, each graph reflects the results ofa possible channel assignment. A’channel is assigned from the several possiblities such that a8 many vertices as possible still remain available aler the assignment. This scheme shows a low prob ability of blocking, but when there are a lot of cells the required computational time makes quick channel selection shihicul 26), Schemes with Channel Rearrangement ~ Compased to FCA, Schemes, DCA schemes da not earry as much traffic at high blocking sates because they ate not able to maximize channel reuse as they serve the randomly offered eall attempts. In ‘order to improve the performance of DCA schemes it large 6 Blocking probability | NN, NSQ, FA, NN A _ NN TLV, SQ, FA sul NN. NN 7, RING, SQ, A su Wable 7. Channel reuse optimization schemes. then techniques have been sees 5 [8 10,31), The baste goal of hate! reasdignment ts swtch calle already in press, whenever posible toate channels these falls ave using to other chanrels, with the objecive of keeping the Sistance between cells wing the . i Same channel simltancousl os tuininum, Thus, channel ree more consents ad more ta fctan be cared porchanrel at ven bodang rte Distributed DCA Schemes “Microcellula systems have shown great potential for eapacity| improvement ia high-density personal communication net ‘works [2 32, 33.34] However, propagation characteristics wil be less predictable and network control requirements more Intense than in the present systems, Several simulation and analysis results have shown that centralized DCA schemes cat Produce neac-optimum channel allocation, but at the expense ‘of high centralization overhead [28, 35-38]. Distributed Schemes are therefore more attractive for implementation in ricrocelilar systems, due to the simplicity of the assignment slgrithm in each base station “The proposed distbuted DCA setemes us ether local infor ration about the cursent available channels inthe cells vei 1y (cel-based) [39-42] or signal strength measurements [43-15] In cell-based schemes a channel is allocated toa call by the ‘base station at which the cal is initiated, The difference with the centralized approach is that each base station Keops infor. ‘mation about the current available channels in its vieinity, The ‘channel pattern information is updated by exchanging status information between base stations. The cell-based scheme provides neat-optimum channel allocation atthe expense of excessive exchange of status information between hase st tions, especially under heavy traffic ads Particularly appealing are the DCA interference adapta: sion sehtemes that rely on signal strength measurements [43] In these schemes a base station uses only local information, Without the need to communicate with any other Buse sation in the network. Thus, the sytem i self-organizing, and chan nels can be placed or added everywhere, as needed, 10 increase capatity or t0 improve radio coverage in distributed fashion. These schemes allow fast real-time processing and maximal channel packing® atthe expense of increased co- chann! interference probability with respect to oagoing calls in adjacent cells, which may lead to undesirable effects such as interruption, deadlock, and instability (Cell-Based Distributed DCA Schemes Local Packing Dynamic Distributed Channel Assignment (UP-DDCA) ~In'the LP-DDCA scheme proposed in [39}, each base station assigns channels to calls using the augment fed channel occupancy (ACO) mattis, which contains neces- sary and sulfieient local information for the base station to make « channel assignment decision, Let M be the total num ber of available channels inthe system and &; the aumber of neighboring cells to cell within the co-channel interference sistance. The ACO matt, as shown in Table 8, has M+ 1 $ Charl packing efoto he area wher a chanel cannot be reed and how lose hse vee are packed. IEEE Personal Commericaions * Sune 1996columns and kj + 1 rows. The first The simulation results of mod Bt columns correspond to te Bcd LP-DDCA {a0 show tat thane The atom indesies x | ‘nhon the co-cell chanel spar the channel occupaney in cell’ iq {oe [TE | ton is tess than four, whichis the andthe remaining ky ows indie So casein most rea systems, the cate the channel occupancy pat- i impact of the additional con- tern nthe nelebonhood of as G2) stein on the compleiy ofthe Chinined from neighboring bse Channel selecaon procedere fs Stations. The last column of the EE xi 2282) insignificant. Also, the fact that ‘ matrix corresponds o the number fof current available channels for each of the k; + I co-channel cells. Thus, an emply column indicates an idle channel whieh fan he assigned to cel. When a call requests service from fell ts base sation uses the ACO matrix and assigns the first channel with an empty columa, The content of the ACO {able i updated by collecting channel oseupancy information from interfering cells. Whenever change of channel occ: panes happens in one cell, the base station ofthe eel informs the base stations ofall the interfering cells regarding the ‘change in order 10 update the information in the local ACO Adjacent Channel Interference Constraint — In ad ‘constraining co-channel interference, the design of w fexillar system must also include measures to limit adjacent channel interference (ACI). Channel impairments such as ‘rosstal, premature bandofis, and dropped calls may result from ACI, leading to degradation of quality of service. ‘Although channel filters in both the base station and the mobile unit receivers significantly attenuate signal from adj cent channels, severe interference may occur in circumstances ‘were the received signal level of an adjacent channel greatly fexceed that ofthe desired channel, This situation arses often {in mobile cellular environments due to the distance differ- fences between the mabile units and the base stations. To reduce ACI, typical cellular stems employing FCA avoid the ‘se of adjacent channels in the same base stato. ‘Al the DCA schemes discussed so fat assign ehannels 10 calls based on the constraint imposed only by co-channel Interference, overlooking ACI. Any of the previous described DCA schemes could he modified so that ey assign channels to cals respecting both the minimum co-channel interference and ACT constraints atthe expense of 2 reduction in the total fatried wate, LP-DDCA with ACI Constraint ~ In [10], a modified version of the LP-DDCA scheme was proposed that incorporates the ‘ACL constraint. ‘The variation of LP-DDCA imposes addtional conditions fon the channel selection fram the ACO matrix [40], Ifthe Fequited channel separation between channels to avoid ACL Inferference is Nag, the Nag ~ U columns to the left and right of that channel should have empty entries in te frst row of the ACO matrix. When a call requests service fom eel is ‘nase station seurches inthe fist row of the ACO matrix for @ {rOup OF 2Nyq ~ | consecutive empty entries where the center colunin ot the group is empty. If successful, it assigns the channel; otherwise, the base station seurches for 2Nyg)~ 1 ‘consecutive empty entries in the First row, where the center columns has ooly one mark. Ita channel is found. it checks 10 See whether the cell that uses the channels additonal chan nels availabe, In that cas lt sends a meseage tothe corre- sponding col, and the base station ofthat cell switehes the call sing the cane in relation to a new one. Thus the hase Station of cell rean use the chanel. Otherwise the call is blocked TERE Personal Communications» Jane 1998 Table 8. 4CO matic ar base station i ‘modified LP-DDCA is ACT interference is prima to its ability to provide flexible reuse packing of channels by allowing up to one local reas- sSgnment wo ascomimodate anew: cll ‘Moving Direction (MD) ~ ‘The MD strategy was proposed in [41, 42} for one-dimensional mierocelfular systems. In these systems, ford call termination and channel changing occur frequently necause oftheir small cell size [12]. The MD str ety uses information on moving directions ofthe mobile units to decrease both the forced call termination blocking proba bility and the channel changing. An available channel is, selected among those assigned to mobile units that are ese Whore inthe service area and moving inthe same dizection as the mobile in question, ‘The search for such a channel starts from the nearest noninterfering col fo the one where the new call was initiated, and stops atthe cell that is @ reuse dis tances away, where ais 4 parameter 'A channel assignment example given in Fig. $ where be. and eare the available channels, and DR is the minis tease distance. For this example the parameter asset to one. The new call attempt is assigned channel b because the mobile requesting the chanel is moving in the same direction ts the mobile in cell number 5, ‘The sets of mobiles moving in the same direction and assigned the stme channel are thus formed. Thus, when a tmobile of set etsses a cell boundary, is ikely that a same. Sct of mobiles has already crossed ott of its cll tothe next Gell: In this manner, a mobile can use the same channel after hhandott with higher probability. Tis lowers the probability of ‘oth changing channels and forced cll termination. The st egy is efficient in systems where mobiles move at nearly the Same speed through the cell aid along a toad of & highway and for one-dimensional microcellular systems “The simulation results in [2] for a one-dimensional stem show that the MD strategy provides lower probability of Forced call termination compared to the NN, NN + 1, and FCA strategies Although the MD scheme has altrative fe tures, itis not obvious how it could be expanded to a wo- tion scheme uses channels efficiently and decreases the num. ber of inracel handofts that i the reassignment of channels 1 avoid interference. I also decreases the Toad of the switeh- ing system as well as quality degradation during a handott period [44] Simulation results show thac interference due t0 farrier sense etror is reduced by 1/10-1/100 with channel seg” rezation [43) Also, the blocking probabil is greatly reduced compated to FCA and DCA schemes, Speed of convergence to the optimum global channel allocation is an important issue in implementing channel segregation. Based on the al- gis in [44}, channel segregation quickly reaches some subopt ‘nal allocation, but convergence to the optimum global allocation takes prohibitively large amount of time Because there are many local optimum allocations. The discussion in [45] shows that channel segregation can be successfully applied to a TD multiple access/FD multiple access (TDMA/EDMA) of multicarrier TDMA system. AS slscussed in [15], the difference in the performance of the FDMA and TDMA’ FOMA systems using channel segres tion is sll, and one-carrier TDMA and FDMA have, in principle, similar performance. The advantages of channel eg gasion ate summarized in Table 10 One-Dimensional Cellular Systems ‘Allthe DCA schemes deseribed inthis section are applic ie for one-dimensional cellular mobile stems: One-ahmen Sonal structares ea be ented ncasex sucha streets with tal buildings shielding interference on ether side (30 Minimum Interference (MD) ~The MI scheme is well known and among the simplest Tor one-dimensional celluar systems Itis incorporated in the Enhanced Cordless Telephone (CT- 2) and DECT systems [50], We present here the MI and its modifications. Than MI scheme, a mobile signals its need fora channel to IEEE Personal Communications June 196its nearest base station. The hase station then measures the interfer= locking probability Foteed cal termination | how one could implement these schemes in u two-dimensional sys fem because an order of sevice is ing signal power on all channels, not already assigned to other MO.NN + L.FCA.HIN| difficult to recognize in a two- [ channel changing mobiles The mobile assigned the Ee gage channel withthe minimum interfer- sdimensional system. A summary of the. performance comparison (eo fence. The order ia which mobiles B Table 9, Comparison benveon MD.N NNN + between the centralized, cell-based, reassigned chamnin fects the ™ 90 nd meesurement based dstbted Shion of channel ese. Taking "4 FCA DeA'shemen ese n Table int sonsiderston he ore OFS secgesconne itis of [Raa Comparison Between andom minimum inerfrence | Moree FCA and DCA (RMI): In this scheme, the mobiles are served according 10 the MI scheme in a random fotder or, equivalently, in the frder in whieh calls arrive in the system, + Random minimum interference with feussignment (RMIR): Ia RMIR, mobiles are firs served according to the RMI scheme. Each mobile then reassigned ‘channel by its base station according to the MI scheme. ‘Those mobiles denied service by the initial RMI scheme also try to obtain a channel again. The order in which mobiles re reassigned is random. The number of times this procedure is carried ut isthe numberof reassignment, R {so Sequential minimurn interference (SMI): kn the SMI scheme, miohiles are assigned channels according to the MI scheme ina sequential order. The sequence followed is such that Any mobile is served only atte all the mobiles that are thea of it have had a ehunce to be served. This procedure ‘would require some coordination between base stations because ofthe sequential order of service. MINMAX — Another scheme applicable for one-dimensional ‘cellular systems is the MINMAX strategy. In this scheme mobile i assigned a channel that maximizes the minimum of the CIRS of all mobiles being served by the system at that time. A mobile i served only alter all mobiles tothe left of i have had a chance to be served. This sequential (let to right) ‘onder of servi is chosen because it appears to be the best ‘way for reusing the channel [Si]. The mobile immediately 0 the right of given set of mobiles with channels assigned is the one that will cause the most interference at the base sta tion servicing the given set of mobiles, and is also the one ‘which has the most interference from that st of mobiles, Performance Comparison ~ in [50], RMI, RMIR, and SMU are compared for 8 one-dimensional mierocellular system. Also, their performance was compared to the MINMAX Scheme, which gives an upper bound on the performance of distibuied chasinel assignment schemes for one-dimensional ‘stems, The system performance is defined as the probability ‘of eall blocking a function of load, The simulation results in [50] show sha the call blocking probability decreases for FCA, RMI, RMIR, SMI, and MINMAX schemes in that order, RMI exhibits approximately 30 percent improvement in the blocking probability compated (@ FCA, RMIR gives an add tional 8 percent improvement over RM, and SMI gives an additional 2 percent over RMIR. ‘One would expect that the relative behavior of RMI, MIR, SMI and MINMAX schemes would not change very ‘much ina two-dimensional system: however, itis not obvious IEE Perso Commenications une 196 | Dectease the numberof inacel andots Decrease load to switching system Reduce interference de to carter sense ear Reduce blocking probabilty | Quietly reaches a sab-opsimal allocation Table 10. Advantages of channel segregation. general here isa trade-off Josisccnsuts ose implementation complesiy of the change allocation sgortins, tnd pecan uiztion ete, Sinution |, 9:10) and sna {08} rests show thas under low Ualficintenaty, DCA strategies Performs better, However, FCA {hemes become superior st igh fered aft, epi inthe ene of unio at athe ase of nonuniform tite and light o moderate loads its tellved thatthe DCA scheme wil perform beter do the fat that under ow rac tte, DCA sues chanel ore tffceni than FCA. In the FCA case channels ae press: Sno to lao three oscasons whe, do faction ina, calls ae locked, even thogh there ate channel Ballable in ajacent cel, In adtion, a base fact of ees Phone traffic ensincrin stata sever with eaacty Cs FRore efficient tans numberof small ones wit he me foal ggoqas expat. Tha forthe se Svetage oc: Ing patty autem with hth capacity as hehe se: Tih [52) FCA scomes baba ls numberof sal groupe tscvers wile DCA prodes aay of making thes sl froups of svets behave ikea laner ser "he ination of requests for service fom cel cll it a random procs hersiore, when dynamism {iteent chanel are signed to serve cle random too Because of ths randomness ound that els whieh have tonuved te sane chanel or se me, om verge ape {reacr distance apart than the moimun eave dtc Com SSqucn yam sigement schemes are not ways uc cede in easing the chanel the maxima possible mumber aims, On the other han, n PCA apelie chanel ca be Sscigned to celle that are the minimum distance part auch itn itsference occurs, The assignment is done sch & tray thatthe maionum reurabity of channel is always Sled, That wy the PCA exhib superior performance Sompared to DCA unde es lod conto "imation results 9, 13-35] roe withthe above and stow tht in the ese of DCA seems the tem snot ever ip sensitive tote and spatial changes tn offered ta, gi ings to almont stable prtrmanc in each cel In aio, inthe DCA the grade of service within an inerfsrence group ‘teal depends om the average ondng within that group nis spa dribaton >, 15,53, On te omer hin the tise of FCA the sce deviation, measure of he gra of ‘Since fncantio from one ft another, fry meh Somened by ine and spatial li changes In geneal, for ihe same blocking rate DCA has lower ten sal ermiaton rate than FCA. In FCA cal st be handed of ito another channel a every handoff esate he Same channel isnot availabe In adjcont celle In DCA thepee ernie New optimem [channe allocation s si ft M Table 11. Comparison berween DCA seems. same channel can be assigned in the nev cell if co-channel interference does n01 occur. In microcelluar systems, mobiles ross cell boundaries frequently and the trafic of each cell, varies drastically. Thus. a large amount of channel assignment Control is required, which resus in frequent invocation of network control functions. Application of DCA schemes in these systems willbe advantageous in solving the above prob lems due to flexibility in channel assignment. As shown by simulation in [54], the traffic pexformance of FCA deterio Fates when cells are smal, while DCA provides much steadier performance. If we also add the geographical load variations the gain of DCA ever FCA will be drastically increased, System Complesity Comparison ~ ln FCA, the asim on- tote made independent in each ely tcecint sachet hanct arog tose led Yo thst el advance DCA the knuwtedge of ecaped chants in oer cells vel the clin quston necessary. The amount o onal ferent in euch DCA strategy. Ifthe DCA requires ot of procesing and complete knewledge of the sae ofthe ene Sistem, the eal cup dely woud be pny tng wiht Highapeed computing and ignaling. As discussed in {39 the implementation compesity of the DCA i higher than FCA. The pips implementation of DCA requte ret deal of processing power to determine optima allocations and a heavy sigan laud On the er Hand FCA regs com pict and tot ene frequency planing efor To sl up & Sse, whieh aot the ee for he DCA sehemes (1) Regarding pe of contol FCA he sutble fora eeneal- ined contol sytem, while BGA We applcable toa desentral- ine contol seen A cSnraied emia scheme creates & hige contol volume fn a mirocelluar tem, which cn ead to bottleneck: One solution to divide the comrol are nto several subarn of sultabe size To eapure all of te abate Tadeo summary ofthe performance comparison of FCA nd DCA Schemes even in Table 1 Comparison Models ~ Due to the complexity of the problem, most of the performance comparison studies between FCA, and DCA strategies are based on simulation models [18]. principal problem with sination comparison isthe lack of ‘common context and scenarios within each strategy. Thus, ‘more unified realistic quantitative studies are necessary. Sim lations to compare the performance must be done under com mon conditions such as cel stuctare, number of channels land traffic intensity in each cell. In addition, simulation with lime-varying traffic is necessary lor more realistic sconaios Minimum communication set organized Increases system capacity, Fast cl time processing [iensie communication |Inceased co-channel Increased iterption, ey ie The problem of performance analysis of cellular mobile systems using dynamic channel allocation hasbeen discussed in several papers [5, 56,57). Io [38] an improved sin lation model suitable for furure mobile systems was proposed which can be used for the teletraffic cab ulations and dimensioning of the system, and to describe the radio coverage of the system with an appropriate level of detail, The sala diference between that mode ‘and ones used in other papers is that it allows overlapping cell areas, some practical aspects, such as fading handoffs and adjacent chan nel interference, are ignored, the channel assignment problem is essentially a queuing optimization problem [21] Along these lines, Kelly [59,60] studied analy tally the benefits of maximum packing over FCA, providing & capacity upper bound for some dynamie schemes. ‘The analysis ‘in [61] finds 9 bound of the blocking probelilities fora similar system. In [62], 2 “Shannon type bound” for single service class was derived. However, all ofthese studies ignote hand offs entirely. In [63], dynamsie and fixed allocation using the notion of stochastic dominance, which incorporates handoffs, Was studied. Furthermore, the conditions in which dynamic scliemes, for the case of uniform traffic and well defined cells, perform better were derived [18] In (18), 4 comparison is made between the maximum pack ing allocation, fixed allocation, and optinal contol policies.” ree the system model is a specific example of multiple-ser- 1; multple-resource system similar to that deseribed in [64]. “The cellular system is modeled as multidimensional time: reversible Markov chain in which states are the number of calls in progress in each cell, The strength ofthe model i that both basic frequency reuse constraints and any additional DCA, constraints can be incorporated in the same model; therefore, competing strategies can be compared equally and the differ: ences between them easily understood. The pi ‘ess of the model is tat i ignores handoffs, whichis (o achieve a tractable form for the stationary distibution and ‘optimal control. In addition, computational considerations limit the size of the state space for which the optimal policies under specific traffic Toad can be caleated [18 The analysis in [18] showed that for a symmetric cellular system (same size of eels, uniformly distributed traffic load), the total system throughput for the FCA, maximum packing, and optimal policies are increasing and concave with the increase in system capacity the same behavior is observed in| the case of an inerease in cell load. At low loads, the total throughput under maximum packing is higher than under fixed allocation, while at high loads the total throughput Under maximum packing is lower than under fixed allocation, Therefore, there exists a unique crossover paint of the 10 throughputs versus load carves However, af low loads both policies achieve throughput close to the offered load, but maximum packing obtains & Tower probability of blocking, At high loads both strategies ub optimum chanel ‘loeston imple assignment algorithm le of local Information| ‘ith other base stations ficiency adie coverage dative to trate changes Tnterteresce ‘deadick probability ane Instat © Ths provides an upper bound on the performan forevery DCA poi hs povides an xtet upper bound on the maximum achievable os pu of the stom anges bight on how inrema perfomance i gba. » IEEE Personal Communications *hine 19ieve a throughput close to the capacity ofthe cellular system, but FA obtains lower probability of blocking because it more often avoids states in which the instants ‘neous throughput is suboptimal. "Ata moderate load itis natural toask whether it might be valwable to combine these two strategies by reserving some of the channels for Performs better under hea trafic {Low flexibility in channel assignment ‘Maximum channel reusability | Sensitwe to time and spatial changes | Not stable grade of service per call | inan interference cll group High forced cal temination probabiity terorns better under ighvnoerate atc Flexible location of channels Not always massmur channel reusability Insensitive to time and time spatial changes Stable grade of sevice per clin in an interference ell group Low to maderateforead cll termination | probabiity Suitable fr large cel ensronment “Suitable in mler-callule environment ton Heil igh fest each cell and sharing the remain- [i der among the calls Indeed, as will be discussed in the next section, a lot of policies have been proposed along these lines. In [18] 4 policy was considered that at low loads resembles maximum packing, at high loads FCA. Hybrid Channel se ee Allocation 5 Table 12. Comparison between FCA and DCA. pybrid channel assignment Sehiemes are a mixture of the FCA and DCA techniques. In HCA, the total number ‘of channels available for service is divided into fired and ‘dynamic sets. The fixed set contains a number of nominal channels that are assigned to cell a in the FCA schemes and, in all eases, are to be preferred for use in their respestive cells, ‘The second set of channels is shared by al users in the system 1 increase fexbiiy. When a call requires service from i cell and all of its nominal channels are busy, a channel from the ‘dynamic set is asigned to the call. The channel assigament procedure from the dynamic set follows any of the DCA Strategies described in the previous setion. For example, inthe stucles presented in [5, 63], the FA and RING strategies are used, respectively, for DCA. Variations of the main HCA, schemes include FICA with channel reordering (65] and HCA Schemes where calls that cannot find un available channel are ‘queued instead of blocked {6 The call blocking probability or an HCA selieme i defined asthe probability that a call arr Jing to a oell finds hoth the fixed and dynamic channels busy Performance evaluation results of uifferent HCA schemes hhave been presented in[S 6,8, 6]. In [S] a study is done for an HCA scheme with Erlang-b service discipline for uniform Size and shape cells where trafic uniformly distributed over the whole jstem. The measre of interest isthe probability of| blocking as the load increases for different ratios of fixed 10 ddynamie cells. As shown in [5], for a system with fixed 10 ‘dynamic channel ratio 3:1, the HHCA gives a better grade of ‘Service than FCA Tor load increases up to 50 percent. Beyond this load HCA has been found to perform better in all eases Studied in [5] A similar pattern of behavior is obtained from the analysis in [6] where the HCA scheme employed uses the FA DCA scheme and Erlang-e service discipline (calls that cannot find an available channel are queued instead of blocked). In addition, the HCA scheme with Erlangee sevice discipline [6] as lower probability of blocking than the HCA, Scheme with Erlang-b service discipline [5]. Tas phenomenon 's expected because in the former ease calls are allowed to be (queued until they ean be served, This inp assumption; there possi hat some dbname ‘channel are, but thecal cannot wse thm case the itererence ‘canara ar old. ‘The ratio of fixed to dynamic channels is a significant parameter which defines the performance ofthe system. It ‘would be interesting o find the optimum ratio in order to achieve better system performance. In general the ratio of fixed to dynamic channels is funetion ofthe trafic load and would vary over time according to offered load distribution ‘Simulation results in [5,6] showed that systems with the most dynamie channels give the lowest probability of queuing for load increase up to 15 percent over the baste load. For load increase of 13-32 petcent, systems with the medium dynamic channels give the best performance. From load of 32-40 percent, systems with low dynamic channels give the best performance. Finally for loads of over 40 percent sys ems with no dynamie channels give the best performance. ‘The general nature of the results presented in [5,6] is very reasonable. As discussed earlier, BCA performs best at low load offerings. When the load i néreased Substantially, the fined allocation performs best because of its optimal reuse of the ‘channel. HCA at load olferings close to the base load behaves as ifthe load offered tothe dynamic channels is Low. Ths because the traffic offered is shared, though not equally, hetween the fived snd dynamic channels thereore, there i n01 much blocking at Jow-pereentage load increases. However, as the fad increases more than a eriain pereentageabore the base load, schemes with aot of dynamic channels begia to block cals with substantial probability. This phenomenon is again ebar- ‘acteristic of the DCA seheme. Io the ease of nonamiform traf ie distribution, a similar performance trend is expected when HICA is used, Tt i believed that the HICA scheme would show its superior performance with nonuniform tealie because it includes dynamic channels which could move around to serve the random Muctuation in the offered trafic [5,6 Studies in (5,6, 8] have provided some simulation results for HICA schemes, Because simulation to study Uke bebavior ‘of a large system is time-consuming and costly, an analytical Imethod would be appealing. Unfortunately, an exact analyt al solution forthe blocking probability inthe HICA system i ‘ot feasible, and one must use approximations. In [66], two sifferent approximating models Were presented. In the fist ‘model the trafic offered inthe dynamte channels is modeled as an interrupted poison process, while the second modeled IEEE Personal Communications «tune 1996 athe system is modeled as « GUMim(m) queuing model. The blocking probability versus the arrival rate for bath models present the same pattern of behavior asthe simulation results of 5,6). Finally, HCA schemes have variants which add channel reordering. that i, switching channels asigned to some of the calls in progress to maintain a nearly optimum separation between coverage areas by simultaneously using the sa channel in order to reduce ineffcieney a high load. As inthe hybrid borrowing strategy, channel seordering is dane when nominal (Hed) channels become vacant, Namely, a nomine! channel is assigned instead of the dynamic channel, which Fequites channel handoffs hetween occupied channels 1 real ize an optimal allocation, This improves perlormance greatly by producing a significant increase in chanel occupancy. but a huge amount of computing is required for channel re arrangement in large ssstem. For example, inthe system analyzed in [65], which has a uniform distribution of fixed scons and-ff prioritizing schemes provide improved performance at the expense of a reduction in the total admitted traffic and an inerease in the blocking proba- bility of new calls. channels and was operated with uniform spatial distribution of offered traffic, the channel vccupancy was increased by twosthirds ever & pute FCA system at the blocking rate of one percent. Ths cortesponds to a channel savings of 40 percent Tor the same carried traffie at one percent blocking by the hybrid systems that were studied Flexible Channel Allocation the flexible chanel allocation (FICA) schemes, the tof [ieee ait inied ane ei Exch cll angned ast of tned channels tt typically sulfces under 4 ight tat lod The flexible channels 76 ssigded to those ces whose chaanels have become nade {lute ender inerng tats loads, The asignment ofthese Emergency channels amon the cells done neither sched tied or predictive manner [67] Inthe ineratue proposed FHA techniques fer according tothe meat which ana The basis on which adtonalchanne ae 3 Inthe predstve strategy, the tac intensity of, equi: lent the bhockingprobatis constantly mesure st every Cells so tate relocation of he feuble channels cane Carved tay point in ime [22]. Fed and fete channels are determined and assigned (of leased f(r Irom) each Gell aeording tothe change in tafieInkensty of Mocking probably measured in each coll The numberof yam Ehannels required ina cel is determined according tothe increase in measured trafic intensity, The scared Fexble ‘Thnels can be used n'a manner knelt the fined chan nels ina cell slong as the cell possesses the channels, As ing accel ho several free fed channels Hex chan nel ave aesgned (ot if the wale inteny is below acertan Threshold fo Tr the lexible channels are ssiened on schedule basi is asumed that toe variation af tae, such as the movement tt tafe peaks in tine a space, are exinated » prior The hangs inasignment of fexble channel then nade at he redetermined peaks of rai change (2) Flexible asigoment strategies se Centralized contol and 2 require the centeal controller to have up-to-date information| fsbout the trafic pattern in ils tea in onder to manage the fssignment of the flexible channels [22]. In addition, the Scheduled flexible assignment is not adaptive to unexpected changes of tatfic. However, as presented in [67], the flenble allocation schemes sufficientiy reduce the processing load of the system controller as compared to the DCA scheme. Fixed and Dynamic Channel Allocation "CA and DCA which tres to realize the lower of each techniguc’s blacking rate sepending on traffic intensity Tn low teat intensity the DCA scheme is used; in heavy teal fie situations the FCA strategy is used. The transition from ‘ane strategy tothe other should be done gradually because 3 ‘sudden transition will cause Tot of Blocking. In [42], the authors developed an optimization ‘model involving a single channel, a donor group, find an acceptor group of cell. An explicit form lais derived for the value of the load below ‘dynamic assignment of the channel from the {donor group to the acceptor group to minimized the overall blocking probability. This study ana- Istically validates the belie that a strategy for DCA should he sensitive tothe load of the sys- tem, and yields an important insight in that DCA should be disallowed in certain situations even if channels are free. The fixed and dynamie strategies allow assignment of channels in a dynamic fashion only ita minimum number of channels are fice. This numther depends on the value of the measured load. As the load inereases, the minimum number of channels, decreases, and eventually, under heavy loads, the seheme Starts to resemble the fied allocation scheme [22} Handling Handoffs ASS Fe and dynamic channel sssignment isa combination of tions did not take into account the effect of handofts in the performance of the system. “Handoff” is defined as ‘the change of radio channel used by a wireless terminal. The new radio chaanel can be with the same base station (itracell, hhando) or with a now base station (Intercell andof. Tn general, the handoff event is caused by the radio fink degradation ot initiated by the system that reacranges Fadio channels in order to avoid congestion. Our focus in this sec- tion son the first kind of hand, where the eause of hand fff is poor radio quality due to a change in the environment fr the mavernent of the wireless terminal, For example, the mobile subscriber might cross cell boundaries and move to an adjacent cell while the calli in process. In tis ease, the call ‘must be handed off to the neighboring cell in order to provide Uunintersupted service to the mobile subseriber. If adjacent cells do not have enough channels to suppoxt the handoff, the call is forced to be blacked. In systems where the cel sie is rela- tively smal (so-called mierocellulae systems), the handott proce dare hasan important effec on the performance of the system. Here, an important issue isto limit the probability of forced call termination, because from the point of view of a mobile user forced termination of an ongoing call is Less desirable than blocking a new call. Therefore, the system must reduce the chances of unsuccessful handofis by reserving some chan- nels explicitly for handoff calls, For example, handofl prio tizing schemes are channel assignment strategies that allocate ‘channels to handoff sequests more readily than new calls IEE Personal Communications » June 1996meneame fe guiewing of hand-off requests, with or Handoff prioritizing schemes provide improved performance atthe expense of redve- tion in the total admitted traffic and an increase in the blocking probability of new calls. Recently, a number of wireless call admission control Schemes have been proposed and studied which can be used to limit the handoff blocking praba- bility to-2 predefined level [68, 69]. Moreover, in [14, 29, 70-73] different prioritizing schemes ‘were presented, ‘The simplest way of giving priority to handoff ealls is to reserve some channls for handoff calls explicitly in each cell, In the literature, tis scheme is referred to as the cutoff prin ‘scheme (CPS) (14,70, 71} 0 the guard channel scheme [72,73 Other prioritizing schemes allow either the handoff 10 be {queued [71,72] or new calls to be qucued [73] unl new channels fre bined inthe vel, Several variations of the hase eu pri orly scheme, with queuing of handoff requests or of new call Fequests, have also been discussed in th literature [71— "The guard channel concept can be used in FCA or DCA, schemes. Here guard channels are not assigned to cells per~ manently; instead, the system can keep a collection of chan- nels to be used only for handoff requests, or have a number of flexible channels with associated probabilities of being allocat- ced Tor hand requests Guard Channels Schemes ‘The guard channel concept was introduced inthe mid80s for mobile systems [70, 72, 74, 75]; however, policies based on uard channels, have long been used in felecommunication systems [26,77]? The guard channel approach offers a generic ‘means of improsing the probebiity of succesful hands Py sim= ply reserving a number of channels exclusively for handatts in each cell, The remaining channels can be shared equally between handolfs and new call. "The penalty sa eduction inthe total cui traffic de tothe fact that Tewer channels ae prant- eto new calls. This disadvantage may be bypassed by allow ing the queuing of new call, Intuitively, we ean say that the latter methox! s feasible because new cals are less sensitive to delay than handott calls [22]. Another shoricoming of the ‘employment of guard channels, especialy with FCA schemes, i the risk of insutficient spectrum utilization. Careful estimation ‘of channel occupancy time distributions and knowledge of the traffic pattern are essential in order to minimize this risk by determining the optimum number of guard channels 22] Handoff Queuing Schemes ‘The queuing of handoff requests, with or without employing {guard channels, is another prioritizing scheme whieh reduces the probability of forced termination of handott calls atthe expense of increased cll Backing probability and a decrease {nthe ratio of carried to admitted traffic (71, 72}, The reason is that inthis scheme no new call is granted a channel before the handoff requests in the queue are served. The scheme is briefly deseribed as follows. When the power level received by the base station in the current cell reaches a certain threst old, namely the handoff threshold, thecal is queued for se. Vice from a neighboring cel. The call remains queued until rither an available channel in the new cell is found or the power by the ase station in the eurrent eell drops below a ccond theeshold called the receiver threshold!" Tt the cal 1 Rajored as uk reservation shames” 10 The hand shold i st at the po where the power eve ty he base tation na neigiboring cll has eared 0 eco he per reed ‘yh curent base station TEBE Personal Communizations «June 1996 without the employment of guard channels, is another prior- itizing scheme which reduces the probability of forced termi- nation of hand-off calls at the expense of an increased call blocking probability and a decrease in the ratio of carried to admitted traffic. reaches the receiver threshold and a new channel has not heen found, then the call is terminated. Queuing handoff requests is made possible by the enistence ofthe time interval that the mobile station (MS) spends between these two Uresholds. This interval defines the maximum allowable wait- ing time in the quewe [14, 29) Based on the traffic pattern ‘and the expected number of handoff requests, the maximum size of the handoff queue could be determined Tn the handoff queuing scheme, the probability of forced termination is decreased. However, a handoff call may stil be «dropped because the handoff requests can only wait until the receiver threshold is reached; in the case of high demand for handotis, handoff ealls will be denied queuing due tothe lim- ited size of the handoff queue, The baste queuing discipline in ‘queuing handoff requests s first-in first-out (FIFO) [22, 7] ‘One of the goals of current research isto improve the perfor” mance of the handoff queuing scheme by modifying the queu- ing discipline . In [71] a nonpreemptive priority queuing discipline based on a mobile’s subscriber measurement was used for queuing handoffs. A handoif request is ranked ‘according to how close the mobile stands to, and possibly how fast it is approaching, the receiver level. Because the radio measurements are already made, there is no additional com- plexity in the employment ofthis scheme, The simulation and analysis results in [71] clearly indicate that the proposed Scheme offers a better performance in terms of quality of ser- Vice and spectrum efliiency New Call Queuing Schemes “The delay insensitivity of new calls makes it more feasible to queue new call attempts instead of handott attempts In (31), {Tmethod was proposed involving the introduetion of guard channels and the queuing of new calls. The performance anal- Jsis in [73] showed that the blocking of handoff calls decreas- @ much faster than the queuing probability of new calls Increases: th rest agrees with the analysis in (72) tn add tion, the analysis in (31] shows thatthe method not only min izes blocking of handoft calls, but also increases total Carried trafic. This is de tothe fat thatthe decrease in the blocking probability of handoff alls results in an inerease of {otal caried afi and hecause the new cll are allowed to be queued, they wil ultimately receive service, Thus, the total aficcatied by the sytem is increased. The gai i total el ried tral between a system wih guard channels and queuing ff now calls done without queuing Is sstants: about 24 Erfangs for a system with 44 channels and 38 Erlangs of ofeced atic [73 System Dimensioning Procedures for Priortized Channel Assignment In systems with prioritized channel assignment, one important issue is to decide the minimum number of guard channels required in each coll so that a desired evel of quality of ser ‘ce (i terms of a limit on forced termination probability) for Tre reaver esol the poi a which the rected power from the ‘ure bae ation iat th minmars acceeable ee (22). 2handoff cals is met. Traffic models and performance measures of typical handott Priority schemes are discussed in (14, 7, 72, 73], FCA with priority is simulated, and a method for selecting the number of reserved channels suggested, in [75] However, this scheme fails to guarantee a preseribed level af quality of service {in terms of eal acceptance probability) for new call attempis. Here, the overall blocking probability is used as the pet- formance measure, and die 10 the eo putational intensity of simulation aad its Tong runtime, t may net be used adap. rly to deal with changes in traffic parameters such as arrival rates andior holding times of calls) In [14], dimensioning procedures for prioritized channel assigament were considered. Moreover, under the eutolt pr Drty discipline, the prioritized channel assignment procedure for single- and multcell systems were formulated 28 nonlinear diserete capacity allocation problems. Exact incremental ag rithms which efficiently solve the proposed problems are derived based on the properties of the Blocking probabilities ff new and handoff calls. As shown fom analysis in (71), for any ratio of puard to regular channels in a cel, the probability bf blocking handoff cals is less than the probability of block. ing new cals. Also, the probability of blocking handoff calls, ‘decreases whenever an additional channel is assigned to the ‘ell Finally, the probability of blocking new cell atiempts is “ecreased sf one or more channels are assigned as ordinary ‘hannel(s) to the cell and increases if one oF more channels fate assigned as guard ehnnel(s) in the cel, In the remainder of this section we briefly describe three. Aiferent dimensioning procedures (algorithms SP1, SP2, and MP) proposed in [14]- Given the numberof available channe's together withthe arrival rates and the required blocking prob blities for both new and handoff calls in each cell, SPl gen- trates an optimal channel assignment whieh ensures pricrity of handoff ealls. Given the arrival rates of he required block ing probabilities for new ane hand calls, SP2 finds the min ‘mum number of regular and guard channels required in each ‘cell Finally, algorithm MP extents algorithm SP! to a multi ‘ell system and provides the prioritized channel assignment for all cells in the stem. Algorithm SPI ~ Given the numberof available channel ina cel the artival rate of no ells, and hand ells an iit for block: ing probability of new calls algorithm SPI generates an optimal ‘channel assignment between regular and guard channels which ensures print of handoff calls and guarantees the desired Blok probably ofthe new cll [14]. The arith i simple, Fist, the number of guard channels ise o 2t0, and the smallest nun ber of ordinary channels (using the Erlang B formula) that guarantee the blocking probability Tor the view calls found ‘Then the number of guard channels is incremented one at & as the blocking probability for new calls isnot vio he total umber of ordinary and word channels is less than the total umber of channels allowed to the cel Algorithm SP2 ~ tn cells with few call handoff attempts, only ‘8 small number of guard channels would sufficiently reduce the chances of unsuecessful handoffs Ia order to avoid giving excessive priority to handoffs in these cells, a desired blecking probability of handoff calls cam be prescribed in addition to the blocking probability of new cals. Given the aval rates of both types of traffic and the distinct blocking probabilities of sow and handoff calls, algorithm SP2 finds the minimum Figure 6 Concenii sub-cel numberof channels required in each cell in order to limit both probabilities of blocking to a guaranteed level. The pro: cedure of SP2 is as follows, First, he ‘umber of ordinary channels i found so thatthe blocking probability of the new calls are met. Then, guard channels are added one at atime as Tong asthe block ing probability of new cals isnot viola fed and the total number of channels is less than the maximum available number fof channels inthe cell If there ae still channels available and the blocking probability of new eas is violated, the number of ordinary channels is increased by one and the pro- cedure of adding new guard channels is repeated Algorithm MP ~ The previous two algorithms axe applicable ina single cell system, Algorithm MP extents algorithm SPI in 4 multicell ystom and provides the prioritized channel assign: ‘nent forall cells inthe system, The model could be extended jn a mulicell environment where the weighted average of the blocking probability of handoff calls is used as the perfor: ance measure for the entire system. In FCA, the total num ber of available channels in the system is divided into disjoint ses Fach channel set is then assigned to cells in the noninter: fering cell cluster, and clusters are deployed ina regular pat- tem to provide continuous service across the service region By applying the MP algorithm 1o exch cluster inthe system, te procedure can be implemented adaptively so that the total umber of channels in the cluster is allocated to cells accord- ng to the traffic Ructuation, Given the arrival rates of handott and new calls in each cell ofthe cluster and the desired prob- abilities of blocking of new cals in each cel in the cluster, algorithm MP finds the best allocation of regular and guard shhannels io each cel of the cluster so that a weighted average of the blocking probabilities of handoff calls is minimized Geuals of the procedure are given in [13]. “Algorithm SPI could he incorporated into a fixed alloca- tion procedure very well. Given the set of nominal channels allocated to each cell by an FCA scheme, it determines the 5number of guard channels in each cell. The algorithm can be fxecuted in each cell site separately. ‘Algoritim SP2 can be applied to various assignment schemes. For example, it can be incorporated in the FCA, Scheme deseribed in [67] in both the scheduled and predictive cases If algorithm SP2 is applied to this scheme, not oly the total number of channels hat aso the ratio between the ordinary and guard channels in each cell can be determined. The third Scheme could be applied to both the fixed and flexible assign- sent schemes. Given the number of avilable channels i the cluster it determines the number of ordinary and guard chaa= nels for each cell in the cluster. The tied algorithm for the luster may, because of interference fsutes, force a nonopti- sal assignment in other clusters, but this problem is common anyway to systems that employ the fixed allocation scheme. ‘All three algorithms can solve problems of practical size efciently; therefore, they can be incorporated into an adap tive assignment scheme where new assignment of channels sust be provided immediately whenever attval rates of calls ‘0 both spe of trate vary with time, Reuse Partitioning What Is Reuse Partitioning? Reuse partitioning (RUP) isan effective concept to get high spectrum eiteney in celular systems, In RUB, as shown i [EEE Pecsonal Communication + Sune 1996Cn ee Fixed ceuse partitioning Fig. 6, each cell inthe system is divided into to ‘or more cocentricsubeells (ones). Because the | Jnner 2ones are loser to the base station located | atthe center of the cell the power level required to achieve a desiced CIR inthe inner zones can be much lower compared to the outer zones. Thus, the channel reuse distance (ic. the dis tance between cells using the same channel) ean be smaller for the inner zones than for the outer ‘ones, resulting in higher spectrum efficiency Reuse partitioning schemes could be divided into fixed [87, 57,38, 79] and adaptive (41, 80-85}, and are sum- ‘marized in Table 13, We diseuss these schemes inthe follow ing subsections. Fixed Reuse Partitioning ‘Simple Reuse Partitioning ~ Sipe RUP was introduced in 75. In this scheme, available charmels ate split among several nea el plas with diferent reuse distances. The undenying Principle behind RUP [78,79] isto eeducesignalto-interer= Ence ratio (SIR) for those units that already have more than Adequate transmission quality while offering greater protec tion to those units that require it. The goals to preuce an ‘overall SIR astribution that satisfies system quality objectives While beinging about a general increase in system capacity For the same SIR objective, reuse panitioing has the poten= tial to obtain a significant inerease in sytem capacity when compated toa nystem that uses only single reuse factor [78 Simple RUP can be implemented by dviding the spectrum allocation into two [37,787] or more [57] groups of mutual iy exclusive channels. Chane! assignment within the ith group is then determined by the reuse factor N; for Mobile units with the best assigned lo the group of channels with the smallest Value factor value, while those with the poorest received signal {quality wll be assigned to the group of channels with the largest reuse factor value As the received signal quality for a mobile unit changes, i can be handed off to @ channel that belongs toa different reuse group on the same zone atthe same cell, 0 & channel that belongs to the same or toa differ= tent grOUp on anather zone atthe same cell, or Wo 8 channel helonging tothe same or 8 different group at another eel ‘Typically, the mobile unis closer to 4 eel ste will be served Iychannels font a group having a smal value of 7S) There are two main design issues related to the simple UP concept. The first issue fs the capacity allocation prob. Jem, which i to decide how many channels should be assigned to each zone. The second iss fs the actual assignment of channels to calls. In [57] the performance limits of the RUP concept were explored, and methods for allocating capacity to the different cell zones as well as optimum fea-time channel assignment schemes have been presented [57,86 ‘Simple Sorting Channel Assignment Algorithm — In (57, 86}, a generalized RUP method called the "simple sorting chan fel assignment algorithm” is presented. Here, each cells tivided nto a number of cocentrie ones ad signed nm berof channels, asin simple RUP. For each mobile in the call the bave station measures the level of SIR. and places the measurements ina descending order. Then it assigns chan- nels tothe set of at most AT mobiles with the largest values of SIR, where Af is the number of available channels in the entire cell, The mobile in the set with the smallest value of SIR is assigned a channel from the outer cell zone. The Assignment of mobile channels according to ascending values OF SIR continues until all channels from the outer zone are ‘sed, The base sation continues co assign channel in the next | Simple euse partivoning | Sieple sorting channel assignment algorthom {B Table 13. Reuse partitioning zone, and so on, unt ll mobiles in the set have ben assigned channels [85 As shown in [86] the simple sorting channel algorithm achieves almost optimum performance. It also allows L4-3 times more traffic than the FCA seheme [86]. An important remaining isue is that the sorting scheme only determines Which cell plan each mobile should use: it does not assign "actual channels, which must be done with some care, In adai- tion, ial cells using a certain channel group started the chan nel assignment by using the fiest channel in the group, we ‘would get an exceptionally high interference level on that par ticular channel. A'random selection procedure would be one \way to sole this problem ($6) Performance Comparison ~ The simple RUP schemes pro: posed in [57, 78, 79, 86] are improved versions of the FCA Scheme, Therefore, they suffer from the drawbacks of the FCA schemes, suchas the difficulty in handling time-variant traffic [12] In ation, the employment of microcels ina ys tem results in ineressing complexity of propagation patteras land further complicates the reuse pattern design process, ‘When the RUP concept is applied 1o a microcellular system, the planning or channel assignment hecomes dificult because the distribution of channels among zones should be frequently changed to mateh the changes in raffic. In addition, the eapacty allocated to different cell zones is based on an est ‘mation of co-channel interference, which is harder task in 2 ‘iicrocell environment due to complicated deformed cell shapes. Therefore, an autonomous or self-organized method for channel sssignment is desired (4). Adaptive Channel Allocation Reuse Partitioning Schemes Several researchers have investigated adaptive channel alloca tion (ACA) RUP schemes inn tempt to avoid the draw backs of the fixed RUP sehemes (80-85). With ACA RUP, tany channel in the sjsiem ean be used by any base station, as long asthe required CIR i matntained. It should be noted that reducing the CIR margin in each channel leads to an improvement inthe traffic handling capacity. Based on this fact, a number of approuches such as exible reuse schemes [si] and selrorganiring schemes (88,80, 82,84, 8, 7, 88] have heen proposed. ln [80. sutonomous RUP (ARP) was proposed, which assigns to all the fist channel found to Exceed a CIR threshold in an ordered sequential channe) ‘arch foreach el. The ARP technique was further improved in another scheme called flee reuse, i which the channel sth the minimum CIR margin is asigned [S!]-Anather Scheme based on the ARP concept, called the distributed con- {fol channel allocstion (DCCA) scheme, was proposed ia [87-89], In [84] allchannel concentric allocation (ACCA), Which isan improved distibuted version of the RUP scheme, vias proposed, Another scheme, selforganized RUP (SORP), Shih is based om signal power measurements a each station, ‘was proposed in [83] In [38, 82] the channel assignment IEEE Personal Communications © June 196 2eiar stem im Figure 7 rnin ofthe chanel concent llocaton. under the RUP concept was formulated as an optimization problem that maximizes the number of served calls, In the fl fowing, we provide a detailed description and discussion of the above-mentioned RUP schemes. Autonomous Reuse Partitioning ~ The first ACA RUP ‘scheme -~ ARP — was discussed in [80] It is based on the UP concept and real-time CIR measurements, In this tech- hhique, all the channels are viewed in the same order by all, hase stations, and the fist channel which satisfies the thresh ‘ld condition i allocated (o the mobile attempring the call ‘Thus, each channel is reused at'a minimum distance with respect to the strength of the received desired signal. ARP ‘easily achieves “reuse partitioning” in which channels bigher in the order are used at shorter distance by mobile stations from which stronger sigaal levels are received atthe base st tion, The resulting pattern is similar to that of the simple RUP [78]. In ARP base stations conduct their allocations independent of one another, and no cooperative contra is necessary. Performance of the ARP scheme has been evaluated in [80] by means of simulations. As compared to simple FCA, ARP Goubles the traffic handling capacity of the system and Adecreases the co-channel interference ty 1/4. ARP improves. Ute traffic handling atthe cost of the SIR margin in each ‘channel. This ereates problems to fast-moving mobile stations Such as car-mounted units, which suffer from rapid fluctua” tions in signal level. If power contro i employed, an addtion- al percent improventen in the capacity is observed Flexible Reuse ~ The ARP was further improved in another ACA RUP scheme, flexible reuse (FRU) [Si]. In the FRU Scheme, whenever a call requests service, the channel with the smallest CIR’ margin among those available is selected. IF there is no available channel, the eal is blocked. Simulations [81] showed that FRU can effectively improve system pacity, especially for users with portable wits. More specif cally, a capacity gain of 2.3-2.7 of FRU over FCA was dbserved. However the FRU strategy requires a large number of CIR measurements, which makes it virtually impractical for high-density miercelluar systems Sell-Organized Reuse Partitioning Scheme ~ Io [85] another SORP Scheme was proposed, In this method, each base sta tion has a table in Which average power measurements for teach channel in its cell aad the surrounding celis are stored When a call arrives, the base station measures the received power of the calling mohile station (in ordee to define at Which subcell the mobile station is located) and selects a Channel, which shows the average power closest to the mea ‘ured power. The channel is used iF available; otherwise, the Second closest candidate is tried, The content of the table for the chosen channel is updated with the average value oft measured power and the power of the mobile stations using the same channel. The power level of the other mobile sti tions is Broadeast by thet hase station. As a consequence of this procedure, in each base station channels that correspond to the same power are grouped autonomously fr sel-orga nized pantitioning In [85], « performance comparison is made between SORP, conventional ARP, and random DCA schemes. The Simulation analysis showed that SORP and ARP show almost the same performance, which is far superior to random DCA. Moreover, SORP can reduce the aceurrence of intracell handott and can reach a desired channel quickly, while achieving high traffic capacity. The essential difference hetwecn ARP and SORP is that ARP always senses the chan nels in the same order until one is available, while SRP. learns which channel is proper forthe calling mobile, so it an Find a desired channel more quickly [85) All-Channel Concentric Allocation ~ In [84] 2 dynamic chan ‘pel assignment algorithm called “all-chapnel concentric allocs- tion” (ACCA) was proposed, which is an extension of the RUP concept. Here, the RUP coneept was extended 28 fol: lows. All radio channels of a system are allocated nominally i the same manner for each cell as in Fig. 7. Each cell citid- fe into W conveatrie regions; cach region has its own charinel lllocation. Here, each channel is assigned a mobile belonging to the concentric region in which that channel i allocated, and has a specific desired signal level corresponding to the channel location. Therefore, each channel has its oWn reuse distance determined from the desired signal level. Thus, ACCA accomplishes effective channel allocation in 4 glabal sense, though it isa self-organizing distributed control algo rithm. Computer simulations showed that the system capacity ata blocking rate of 3 percent is improved by a factor of 2.5 compared to the FCA" Ht, in addition, a transmitter power ‘conttol i implemented on top of ACCA, the system accom plishes a capacity 3. times greater than PCA. Distributed Control Channel Allocation (DCCA) ~ The recently proposed DCCA [87-89] i 4 dynamic channe! alloca: tion scheme based on the ARP concept. In this scheme all cells are identical, and channels are viewed in the same order, Starting with channel number one, by all the base stations in the network, The decision to allocate a channel is made lacal- ly based on CIR measurements, The architecture ofa cell ia BCA is shown in Fig. 8. It consists of an onnnidisectional ‘central station connected to six symmetrically oriented subst tions. The substations are simple transceivers, and can be sivitched on and off under the control of the main station When the tafe density ofthe cell low all the substations are off and the only operating station fs the main station, a the center of the eel covering the entre cell area. Gradualy, as eal traffic increases, forced cal Blocking will occur due 0 an unacceptable level of co-channel interference or the Uinavalabilty of resources. In this cas, the main base station ‘switches on the nearest substation to the mobile unit demand ing acess. This in effect relocates the main base station closer to the mobile requesting service; therefore, CIR measure ‘ments wil now be higher, thus improving the probability of Finding an acceptable channel. If the trafic is teduced, the main station soches off a rumber of substations, The systems therefore automatically adapts itself to time-variant cal afc lfensity. As a result, an improvement in both system efficiency tnd wrative capacity can be achieved. As discussed in [89], the DCCA system results in lower probability of forced termina tion of cals. Computer simalation showed a drastic reduction in the number of handoffs and almost S0 percent less forced TEFE Personal Communications Fun 196Sie the Se resented | eg tions (38, 81, 82], frequent rearrange ment of channels (38, 82], and/or ‘cooperative contro! among base sta- tions in order o maintait an optimal allocation of channels to different cell zones, The proposed scheme in [38] is | 4 CiR-adaptive but complicated method which showed a potential for producing excellent efficiency. It requis chanel reassignment every 5 S Tor optimal performance and some data communication between base stations. Finally, in [83], the possibilty of using Hopficid’s neural network to solve the optimal chanel assign mien problem under the RUP concept was investigated. ‘Although the idea is appealing, it not practical for present systems, In Table 14, summary ofthe important characters- tics of channel allocation schemes based on reuse partitioning 's provided. Other Schemes Overlapping Cells Beneen the extreme schemes based on fixed allocation, there are many possible alternatives, hybrid sehemes and schemes Such as dicted retry (DR) and dirscted handott (DH), which take advantage of the fact dat some percentage of the mobile stations may’ he able to obtain sufficient signal quality from two oF more cells. With DR, if call Finds js frsattempt cell thas no free channels, it can then try Tora free channel in ny bother eel that can provide suicient signal quality, The DH Scheme takes this idea further, in that when 2 cell has all or almost all of its channels in use it may, using DH, diret some of the cals currently in progress in its domain to attempt handoff to an adjacent cel. The motivation here is to attempt 10 redistribute calls in heavily loaded cells to lighter loaded cells [53] Both ihe above schemes are expected to improve system performance. This improvement depends on the percentage of| Falls that could communicate with to or more cells simulta heously or equivalently to the percentage of overlapping, hhetween adjacent cells. This percentage has been reported be as high as 30-45 percent (83) ia which the performance of both the above sehemes was compared with the MP dynamic scheme which provides an upper bound in the performance of DCA schemes. The conclusions reached by simulations in [53} were that both schemies improve the efficiency ofthe system, For the DR scheme an inerease in the overlapping between cells leads to an increase in the grade of service provided by the system. In audition, the DE scheme has very good sensi tivity propertios with respect to variation inthe spatial traffic profile of the system Selective Handover ~ Another scheme, selective handover for traffic balance (SHOT), is based on the concept of FCA and overlapping cells proposed in [90]. Ifthe teaflie of a cell, Increases temporarily such thatthe resource utilization rate exceeds & threshold, SHOT hands off some eall tothe appro- priate adjacent cell. Whenever a call reaches the overlapping ftea ica he served by the hase station of either of the ver lapping cells. Therefore, in the case of a temporary trafic Figure & CCA cet mere inerease calls ean be distributed to adjacent cells which share the overlap. ping area, The wider the cell overlap~ ping, the more trafic performance is fest to ieprone ‘Simulation Fesults in [90 show that SHOT improves traffic performance ‘under the condition of uniformly dis- iriboted traffic, and enhances the fre quency utilization in the time domain through the handott of mobiles inthe overlapped areas of the cell. This | method i superior to DCA because it Utilizes the conventional intercell handoff function, and no new func tions are necessary [90}. The perfor- mance improvement achieved by SHOT depends greatly on the algoritim used for selecting & ‘mobile station for handoff from a heavily loaded cell to new Selected cell. Inthe following, we discuss theee algorithms for hhandoff selection proposed in 190), SHOT? ~ In SHOT! the algorithm seleess the mobile station ‘with the minimum reception level — the mobile further vy from the hase station. Though it provides very simple seletion| control which measures onl the reception level of the mebile at the base station, the selected mobile station does not neces sarily have the required reception level atthe new cel SHOT2 — 1m the second algorithm, SHOT2, all mobile sta tions in the original eell measure the reception level ofthe adjacent cells which have one or more ide channels. SHO Selects the mobile with the maximum recep Although the control in SHOT? is more complicate it pro- vides better signal quality. Both SHOT1 and SHOT? do not take into account o-channt interference SHOTS ~ In SHOTS, all mobile stations in the original cell ‘reastre the reception evel fom the adjacent base stations ‘that have at least one idle channel. The mobile sation and the ase station that have the highest reception level are selected and called the “fest priority par,” Similarly, the "sec- Oona!” and “third priority” pairs are formed. Bach selected base Station makes its mobile station measure the interference of the candidate handover channel. The same is applicable for the second and third par. The pair with the least interference 's then selected ‘As shown in [90], the improvement in traffic handling depends on the required SIR value for channel interference. For the ftst two methods, asthe required SIR increases the frequency utilization gain degrades, Although SHOTS isa lit- te complex, it provides a performance improvement of about Si percent. The above results are for uniform traffic: for ‘nonuniform traffic conditions, al SHOT algorithms are expected to perform more effectively. In Table 15 a simple ‘comparison Between the three SHOT algorithms is provided In edition, Table 16 provides a summary of the advantages ‘of the overlapping cll schemes Overlaying Macrocellular Scheme In microcetular sytem, requent handoffs are very commen, ‘A channel asignment scheme diferent from the schemes ds cussed thus far isthe overlay scheme, Here, a luster of Inicrocll are grouped together and covered by a macrocell [91]. In overlay schemes, the total wireless resource is divided between the macrocel and all the microcell nits domain, In as of oongestion if ere are not enough mierocll channels BEE Personal Communications une 19 afor handoff calls, then macrocell channels can be used, Because the macrocell base station covers a much larger area than a microcell, ts cransmitied power is higher than thet of microcell base stations Inthe pas, different channel assign ment schemes for overlay cellular systems based on FCA and DCA schemes have been studied. In [91], 2 microcelular clus ter having contiguous highway mierocels, each with its own hase station, is considered. Overlaying the micracelluar clus: ter is a macrocell whose base station also fulfils the ole of tbe mobile switching center (MSC) of the mierocellular cluster. ‘The macrocell base station has X channels at its disposal ‘composed of X; for new eas generated in the macrocell, >for bandos from other macrocels in the mactocell cluster, and 2X; for handoffs from the microcelalar system, A mobile sta tion that is Blocked duting a Aandoltaftempe due to insuliient thannels a a mierocellvlar base station requests a channel from its MSC. Ifthe macrocell has a free chasnel, it assigns the channel to the mobile station, Later, if an appropriate channel becomes available in a microcell the macrocell channel is released and the calls handed off to the microcell chanos ‘As shown with simulations in [91], withthe use of the above reassignment scheme, the probability of terminating calls feduced atthe expense of an increased number of handofts. Frequerey Planning In the previous sostons we diacensed a namber of dierent Camel assignment technics and evalstd ther performance seth respect oer performance eter. Al thee tshrghes Ea coo Ss assume that a number of channels Cis available to the systom and try to find the best way of assigning these channels 10 falls so thatthe utilization ficiency of the system is increased. ‘Another important question related tothe efficiency of the system isthe following: Given the traffic profile fora system. and.a predefined blocking probability, what isthe minimum ‘numberof channels required to accommodate the trafic? In [35] the MP concept is proposed, which finds the mini ‘mum. number of channels requited to handle a given number bf calls, based on cell compatiblity information. Along the Same lines, the study in [48] evaluates the minimum number ff channels assigned to mobiles under given operating cond tions such that gwen interference conditions are satisfied. The ‘operation conditions refer to the knowledge or lack of knot edge ofthe location ofthe mobiles, The interference condi tions refer to the acceptable level of interference so that two mobiles will be assigned the same channel. In [48] the mini ‘mum number of required channels is evaluated by eonstruct, ing matrix, defined asthe compatibtiy mati, of dimension NXN (N: gumber of mobiles inthe system). Fach mobile i ‘evaluated with cach other mobile to see if they can se the Same chasnel. A graph i then composed, where each mobile corresponds to each Yertex and an edge connects two vertices it and only i the two mobiles are incompatible (i.e., cannot use the Same channel simultaneous). set of graph-coloring llgorthms could then be employed to find the minimicm num ‘per of colors to cor the vertices inthe composed graph such that no two vertices interconnected by an edge are the samme ‘olor. Thus, the nomer of colors is equal to the number of required channels. This problem is equiva Poca fase arioning | Opiinim pertervance TDraubacs af FX feat to finding the minimum nm eee Cartes more vafethan FCA —_ifcalty m handing time | er of cliques that cover all the : vorlon tafe Xertces in the, complementary Alford crete protein to —_[Difietimpementaton | graph. Because the coloring prob- those ues tat neste most | inmicvcear system | Kin fs NP-complete [92], heuristics imum imosctonceltcke—|pitcut chanel ate used. The Reus used i [8] is the algorithm proposed in [93]. Re achitectre 2signment planing | Higher vate henaing [GR margin eich Cot “capacity than FOA Rapid fuctuaion in signal ive Which gives an upper bound on the ‘minimum number of required col- ‘rs. The results in [48] shoved that the MP scheme ean reduce the ‘number of required channels almost by a factor of for interference dis ‘ance 2.0) compared t9 PCA schemes. sere een one FA Sets organizes Tiree ima oe | © | Reset theintcet Ronco | Fees nt Requires great deal of Impact or miro-ctlr | 1 Power Control s discussed above, the pur- pose ofall erie! assignment gorithm is to assign radio ‘channels to wireless usets such that ‘certain evel of CTR is maintained ‘every wireless terminal. One can also use power control schemes to achieve the CIR level. Power eon- trol schemes play an important role in spectrum and resource allocation in cellular networks. The idea ‘behind power control schemes is ‘pase oa the fact thatthe CIR at _ | Seter vate adap ec enmaminer feral Seewonnata tn: Gein” cee | neta Le eeanimesaaienl anime | Salat geen <1 co-channel interferers. Thus, By W Table 14. Comparison of rese patinoning schemes. Increasing the transmitted power of IEEE Personal Communications * June 1996,[Do not take into account co-channel interference Simple selection contol ‘Maderate selection contol Better Communication quality [bo not ake int account ‘co-channel interference the desired signal and/or decreasing the power SHOT level of interfering signals the CIR level ean he accommodated, However, this approach is based)” 3H0T2 on posing requirements because an increase in the power level of the desired signal level corre= toa certain witeles station also resus SHOTS spond ian increase in the interference power level corresponding (o a different wireless station using the same channel, The purpose of different power control schemes is simply to find a trade~ bff between the change of power level in oppos- ing directions. In a way, power control schemes tty to reduce the overall CIR in the system by measuring the received power and inereasing (oF decreasing) the transmitted power in order (0 iximize the minimum CIR tn a given channel jocation of the system. This ean resll ina dr matic increase of overall system capacity mes Sted in ferme ofthe number of mobiles that can be supported. Power control can be done in cither senralized or distributed fashion, Central- ized power control schemes require a central controller that has complete knowledge of all radio links and their power fevels in the system [94, 95]. In the distributed approach [96, 97], ‘each wireless terminal adjusts is transmitter’s power level based on local measurements, Gen erally, distributed schemes for power control converse rapidly toa Stable state ifthe system can accommodate all existing Hinks. Otherwise, some of these algorithms can result in fact ‘ations of the power level and converge to 4 minimum CIR, level, which is unsatisfactory. In [98] a set of link admission control algorithms have been introduced, the purpose of ‘which isto avoid such unstable or undesirable conditions in the distributed power control algorithms. [shor Conclusions growing interest in the area of wireless ions in recent years, the wireless resource allocation problem has received tremendous atten- tion. Asa resulta vast amount of research has heen done 10 extend the earlier work as well as to introduce new tech- niques. Most of the recent work bas been in the area of dis- tributed, adaptive, measurement-based, power-coatrol-based, priority-based, and overlay channel allocation schemes. In| Addition, avast amount of results have been published which provide an insight into the performance, complexity, and sta- bility of different channel allocation algorithms. In this article, we have provided an extensive survey ofthe resource alloca- tion problem in wireless networks and presented a detailed and comparative discussion of the major channel allocation Schemes. Wilh recent trends in the ateas of microcellular net- ‘works and wireless access broadband networks where milti- ‘media applications will be extended to end vsers over wireles Tinks, we are faced with new, interesting and important chal lenges to the wireless resource allocation problem. These challenges have arisen as result of emerging and new tech: nologies, a result of recent advances in the design of low- power handheld wireless terminals, the design of advanced radio modems and antennas, and, finaly, recent developments inthe area of spread-spectrum systems. These emerging new reas will intsoduce a new set of constraints in the resource fund channel allocation problems. The solution ofthese prob- lems will ply an important role in providing ubiquitous sooess to multimedia applications in personal communication net- works, Directed handoft Directed retry Improves watfic handling ‘opacity {Complex selection control 1 “Table 15. Compan mer ahe SHOT agritorc Has very good sensitivity properties with respect to ‘aration in spatial traffic profile of the system asthe capabiity to offer a large increase in system performance significant name of all cn heat two.or mare cells smultaneatsly Improves system performance Improves tafe handling capadty Enhances frequency utizaten Utiizes the intercell hand procedure [The more cel overlapping, the more traffic improves, ‘Table 16. Comparison of overlapping cel schemes References LW. lates, Merowave Mabie Commanizatins IEE Ress. [2]. Yl he Cllr Senemes for Spewal Eton,” EE ean ‘on Voi ech sl VIG 987 9p. 188 3, {a1 "Zander, sasymptone Bounds an the Ferformance of» Clas of yma se Asignment rhs FEE a 1. 193, (41 PEt chang, "Pextrmance Ise and Algorithms for Oynaic Cham fn Besant IE BAC: wl TY, 183,86 (5173. ta an nL aargans Hj Chanel Asignmen eee in Large Seale Cally Structures MOble Communtetion sytams EE Trans on Commun. vl COM 26,1378, 943238. (611 Sin and NGoarginas "A Srsiton Sty ofa bid Chanel Assign ‘nant chen for Cele tan Mie Ras Star wth Erara So tle” IEE Tas, on Commun. at. COM9, 196. Pp 83-47 (7) W°& ¥. en Mobile Ctr Commonention Ste, 1989, (010. Cox and 0.0. Rau teasing Channa Ocupane Large Sale Nisbet Stems Dynan Chas hansnmen EEE Tron o0 eter Tecnology va VIS, 197) pp ast (91 "x'Comparson of Some Non insor® Sparta Demand Profiles on ‘Mobis oa Stem Pertormance," EE Trans on Commun, vl CON [101 € cor and, ©, Reus, “Oymamie Chanel Assignment Two Dimenson Largest Mab an Stes,” Bol Ss Teh J ol 3 1972 pp et 28 117 Sh "ate Capay of Thee Types of Coma Use Ral Com Imancatn Systems 12 Tans on Commun. Tech vl COMB, Wo Ural ake rs lrowave Mobile Communications New York: Wes, [nal hang and 1:5, Yum, “the Non Uniorm Compact Pater Aloe: tis Aigthry for Cllr Mable Systems IEEE ans on Veni Teoh 128 pp 387-3, [04] 5H Oh eta SPrortned channel Aesignment in Cllr Radio twos” ie annon Commu. ,1099 pp 935-03 15) Anderson. “A smutation Sd of Sole Byam Chanel Asin petals an apc oe forme Se (06) 8 get and M. Petey” “tatty Optimary Dynami Server ‘Asn in Syste wh nang Sets IE Trans on VehCe ep vel vt, 1893 op 200-8 L171 thang, Comparaans of Channa Assignment Satis i Cellar site Wgphane Sytem” BEE Transat Vertu Toh a 138 1989. pp. 211-13 [1815 Jodan and: Khan, “Optimal Dyramic Aloation in Cllr Sy ‘esate for putin, 1993. 09} Sawada at, "Tachnigus fr ieeaing Frequency Spec Ut Vston® tech, Rep hoe 100, 1968 [ao] singh 5 inaub, and ©, Gupta, “A Now Frequency channel ‘assignment algerthm in igh Capoety Mobile Communications 5s IEEE Personal Communications» lune 196 2»vet em ney 4 [22] 5. Tekinay and 8. Jabbari, “Handover and Channel Assignment in iterate te a 25].5. 5: Kuek, “Ordered Oynaenic Channel Assignrnent Scheme with Ress wht anon or em span age eee ree ai on to wl a Ptr nna nr GLOBECOM ‘82, pp. 1414-18. ei ao Marte ies skies a teat eg eee ia enema ook ahi ge eee ai gh ieualhcaae reread tee eter tae at re Ce nanos Ul" Fnn a Aa Chal enn “A Bie anes (451, Akatwa and HW. Andoh, "Channel Segregation: self Organized Sen rn i re EM an aa ei iet Co eS (511, 8. Part and D, Sparreboom, "Mathematical Models forthe Analysis wait ech i (Bie Sere eg tat eo 151, eet Are meet of an Maton Trens. an Vehicular Tech,, vol 42, 1993, pp. 444-5, solutes patra A of ct aie ‘sr find) eth, "apacy ean a Chan ne sr fade Spare ig tare trig Pe eee tal lonen hap sd ter tea of Sh oscly Mint ee i: Cen Sp HEECE 958, off tly “ang meting het Sted Net ents ged stn 1 Tu soft Pity aang ne ete tei ptinzaton Shad (oF HBR ana a tend ote Acca au Communication, Control, and Signal Processing, 1990, . (eaLEN Sorrel er Orc Cae Ilona adr: Tt Yen ch a (1p A Rama chung and Mei, Dyan ned (Sater loti maa csi Comaiesns wrt HEE a Go Cina Sl 158 tn 0 Red meg wt cng ae Se soa a ee alte ho orm ca sete aemuniton san ith Wd Chl Ange ene Me ons a Vaca rly na vo (oi a nd mam, & tae ee hal Att pneumonia rs one os (con dears Nghia, “onl nd Gali of Serie rronsionnain highspeed Msc eto Wad Ps Cae sat ae Uc aging dM. Seba “Dud Cal mien Con islet ete rae Para ndon be ase (ro ata "ale eng en Ptr ely on eit her et rien ne nde ho Nan ae thang a tra aera uc atc fe Ctl ai tt rabbi hance nts pp Dana 2183 trace “ucens Boing Ste to Ay hl teams nd ‘ar ind He? rt 8 Conc ap vat charg iat se Bang Watt 3 Mt ont et Rio” BES Tak ono a 3k "Se op 36 (9516 lint, anda a a Tel Mating cl crit ont nd hae Rh aan ene onl’ Sasa Shae "Oe ian nd romano ct ‘ty ews in tlcing yeni Sel Bone cnt ape Ege tong Cr pens Hon Dt ines, “ai gin ny a ae Metin” Po (ol at Aue Fron Cl Ste 181] 5, Onoe and 5, Yasuda, “Flexible Re-Use for Dynamic Channel Assign rete tab ana ioe REECE tah TC weal a hc le alt ns angel Ctr ee Vat WEE Vida 50508 cea Sepa “aon spt Ar Mob Conmi Shon ei oq acah Realnea te et tel tte ill. “Aran Cnc lo loon elu Spe aE, pa vest Fulawa and Raa St pad ea rong» bine an ign ted eke Ste Me Sid tela coll aed eve Pion Clara Radi tat Conmancans ster 5m anwar and A AgoRearaeetbaed Col IEEE Pensonal Commicaion + June 1995,channel location (OCCA) unde ifm and Non-Untorm ei bration nbs ao Corsten EK 998 {sein wadal and gam, “DCoRA Ostbuted Conta Channel ican Sefer Mireceluar Communastion Netrre” Pre ZOE Fr Coton Pr an "#915 convo chanel ilceaonYOCEN far Mircea as Sem, aaa 22 90} ai “sine Handover fr Tate Slance in Mabe Commurica "re ro teSupecommnCe 99 vat 982 pp BATT tong, Si eal em Hades Pmance oh (021 47 Garey and 0s sean, Comput nt natty: & Gide ‘othe Tea af Wampler NewYork Wf eoran 1975 (021 Jon and 0. Watt, robs Buds sa Het go hms for olving Lage Random Gash Tach, Reps SU, Dal, 04] Grandi ot f,"Cetrlzed Romer Contin Cau Rai Sy tens TEE Trane. an Vea few 80.50% 1988 95} dander, “berormance of Gum transmits Fowet Contain tities Sr eo bay hw sna ‘sens IE Tran on Yoh ena 982 (971, "a honehronoes Dante Algorithm fr Power Con inet taper Ph AD. rd Gere {2a} amor Schon and ote Rad ik Adition goths oc wel Wtons wth Re Conte an Ate Lk Quay Pe tecton® oe Ioan 35 Boston, MN Ap 1985 Biogrophies these tr ez he pom ee egnesing Kom te Wetee etl Ces atbere Ceats i 550 tS or estes cos Cnneriy i n8 a 90, rapecve Caren ee thong he Pb dei ines uk ong ASR ty Se 8 sc ben aoa eth ‘ey Sh i's morro he national Techical Chamba of Greece. Het ‘Siveae asset hnapenent Serr e098 a [Sis ef pues reese! ean uous Nacionnce ier mano he Thomas ‘wort nthe Wiles and Mabie Networks group. Ne jan sh 008 From 19a 991, he worked one vara research nd devant [oj desing wt eng and any of oa ea rotor cs ‘on poco and fe pct sutrevoadbad nes See 29) thee been wordigintneaver Of wueles en mode APM, vreens ina abe ary, Rew ors 0 andi enc, Sand the fE Tech Conti on Compe Cooma > we 2 he ices comatee on Pesralcomanean Beso odio 12 Penoal Commune mega, Ne ah ay er he teacher» course on werelesmobiecommurscatgne and networking TEEE Personal Commusications * June 06