7225059 Admission Control in Frequency Hopping GSM Systems

download 7225059 Admission Control in Frequency Hopping GSM Systems

of 5

Transcript of 7225059 Admission Control in Frequency Hopping GSM Systems

  • 8/14/2019 7225059 Admission Control in Frequency Hopping GSM Systems

    1/5

    Admission Control in Frequency Hopping GSM SystemsPer Beming and Magnus FrodighEricsson Radio Systems AB

    S-164 80 StockholmSweden

    Abstract: One way to improve the ability to handleoffered traffic variations in a cellular system is to use softcapacity, i.e., to have an Admission Control. algorithm thataccepts new calls as long as the quality of the alreadyongoing calls are preserved. The increased flexibility inthe use of the radio resources will increase the capacity inthe system.

    In GSM, the operators continuously tightens their cellplans. Suggestions of going to a cluster size of 3 with thehelp of soft capacity exists. This paper investigates theperformance of an Admission Control algorithm based onthe number of active users in each cell. The purpose of theAdmission Control algorithm is to block some calls inorder to preserve the quality in the system. By means ofsimulations it is shown that a traffic load based AdmissionControl algorithm works well in a system with cluster sizeof 3. The results show that the method preserves the qual-ity both when the traffic is uniformly and non-uniformlydistributed. The system is using frequency hopping, dis-continuous transmission and quality based power control.

    I. IntroductionThe fast growth of the cellular market necessitates high

    capacity cellular systems. One way to increase the capacity isto use a tighter frequency reuse plan.

    It has been shown, [11, that a GSM operator having 5 MHzavailable bandwidth can gain 190% in capacity with a fre-quency plan using a cluster size of 3 compared to a frequencyplan using a cluster size of 12 if random frequency hopping,quality based power control and discontinuous transmissionare used. However, going from a cluster size of 12 to 3implies that one cannot occupy every available channel inevery cell without introducing a severe C/I situation. Thus, amechanism that guarantees the quality in the system by limit-ing the utilization of the total number of available channelsmust be deployed. A simple solution is to limit the utilizationof the channels by not installing more transceivers than possi-ble from a quality point of view. However, due to the localvariations in the offered traffic (hot spots or just the variationin time), the blocking probability with such a solution will beunnecessary high. Hence, a solution that limits the number ofactive calls based on other criteria than the number of trans-ceivers is preferable. An Admission Control method isneeded.

    In [2-41 a new user is blocked if there is no available link

    giving the new call an estimated C/I over some threshold. Themethods consider power capabilities and estimated path gainof the new call plus measured interference on the free chan-nels when deciding whether the new call shall be blocked ornot.

    In [5-61, the method of preserving the quality has anotherfocus, namely the handover failure rate. The methods con-sider the number of active calls in the cell and neighbouringregion. The method in [5] blocks the new call if there aremore active calls in a cluster than a threshold (there are stillchannels left, but they are reserved for handover calls). Themethod in [6]blocks a new call if the estimated probability ofhandover failure becomes too high. The probability is esti-mated from the number of active calls in the cell and itsneighbour cells.

    The purpose of this study is to evaluate one method ofAdmission Control called Traffic Load Admission (TLA).TLA aims to preserve the speech quality in terms of C/I as in[2-41. TLA, however, considers the already active users,which will be disturbed by a new call, when deciding to admita call or not. The blocking probability and the characteristicsof C/I are derived for two different traffic environments. Theresults are compared to results obtained with hard blockingdue to limited hardware capacity (less transceivers than fre-quencies per cell) and with a traditional hard blocking system(a system that is not using any admission control at all). Thestudy is mainly intended for GSM, but when applicable themethod may also be used for other systems.

    Chapter I1 defines the quality and performance criteriawhile Chapter III presents the TLA algorithm. In Chapter IVand V the performed simulations and the obtained results arepresented and finally in Chapter VI there are some conclu-sions.

    11. Definition of Quality and Performance MeasuresAdmission Control shall guarantee that the active users in a

    system have an acceptable quality. The quality is in thispaper defined by the following criterion:

    90 percentile of C/I 2 X dB, i.e. 90 percent of theusers shall have a C/I greater than or equal to X dB.

    The probability of blocking a new call, P,, is used as a per-formance measure. P, shall be compared when the criterionis fulfilled and the comparison shall be done while the meth-ods have approximately the same quality defined by the cri-terion above.

    0-7803-3659-3/97 $ 1 0.00 0 1997 IEEE 1282

  • 8/14/2019 7225059 Admission Control in Frequency Hopping GSM Systems

    2/5

    111. Traffic Load AdmissionA. Intelference Areas

    The basic idea in this method is to divide the cell plan intointelference areas, i.e. each cell together with the co-channelcells that generates the dominating interference (e.g. the sixnearest co-channel cells) build an interference area.

    Consider for example the system using a cluster size of 3 inFigure 1. Here, cell number 22 build together with cell num-ber IO, 11, 21, 23, 33 and 34 one interference area if wedefine an interference area as the cell itself and its six nearestco-channel neighbours. Denote this set of cells by A22. Nowconsider the interference area corresponding to cell number34, A34.This interference area consists of the cells 2 2 , 2 3 , 3 3 ,3 4 , 3 5 , 4 2 and 43 . Note that cell number 22 is a part of AS,+ Itis easy to see that cell number 22 is a part of the interferenceareas A I , , A l l , A,,, A,,, A23, A3 3 and A34. Call this set ofinterference areas where cell 22 is included by S22. Generaliz-ing this, there is a set of interference areas coupled to eachcell, i.e., Si is the set of interference areas including cell i.B. Utilization Factor

    Denote the number of available channels in cell i by ci andthe number of active users in cell i by ui. he utilization factorin interference area Ai, enoted by F A , , is then:

    C u.i(3.1)

    j e A,For example, if a system using a cluster size of 3 has 12

    carriers then each cell will be assigned 4 carriers. Each carrierhave 8 time slots, thus ci = 8 * 4 = 32. If the interference areasare defined as above then the denominator in Equation (3.1)will be 32 * 7 = 224.C. The Algorithm

    Consider a new call in cell i. The call is admitted ifF A j ' t h r e s ho l d ) VA j Si (3.2)

    Here, the threshold Fthreshold determines the trade-offbetween capacity and quality in the system.

    The algorithm in words: check the utilization factor in eachof the interference areas where cell i is included. If the utiliza-tion factor in any of the interference areas in Si xceeds thethreshold, FrhreshoLd, block the call else admit the Call.

    IV. SimulationsThe simulator tool that is used has a birth-death process

    with an arrival traffic according to a Poisson process andexponential distributed call duration.

    Figure 1. Interference areas n a system using a cluster size of 3. A cell,together with its six nearest co-channel ce lls, build an nterference area. Cell

    number 22, together with cell number 10,1 , 21 ,23 ,33 and 34, build aninterference area. Note that cell number 22 belongs to seven different inter-

    ference areas together with the shaded cells.A. System Parameters

    In the simulations frequency hopping, discontinuous rans-mission and quality based power control were used. Therewas no mobility. The simulator uses a cell plan with a wraparound technique to avoid border effects. Further, the systemwas interference limited.

    A 5 MHz operator was assumed, i.e., 24 available frequen-cies. 12 frequencies were used as BCCH carriers and not con-sidered in the simulations. Hence, 12 frequencies remains.These 12 frequencies were divided into three groups. Since 8time slots was simulated, ci = 8 * 4 = 32. All system parame-ters used in the simulations are summarized in Table I.B. TrafJic Environment

    In this study, the Traffic Load Admission algorithm wassimulated for two different traffic environments: Uniform andHot spot. Uniform was realized according to the following:

    The mobiles were distributed with equal probabilityover the coverage area of the system.

    Hot spot was realized according to the following:50% of the mobiles were distributed with equalprobability over the coverage area of the system.The remaining 50% of the mobile were distributedaccording to a two dimensional Gaussian distribu-tion. The standard deviations in the two dimensionsof the Gaussian distribution were both set to 1500meters.

    The cells that covers at least 90% of the traffic generated

    1283

  • 8/14/2019 7225059 Admission Control in Frequency Hopping GSM Systems

    3/5

    Propagation modelLognormal fading standard devia-

    Okumura-Hata (35 og d)

    No of frequenciesCluster size

    Time slots used

    I I 75 Io of cells3 sector

    Cell radiiAntennas

    IRandom Frequency Hopping I On a) UniformDT X factor

    Quality based Power Control

    0 km/hCall duration

    VelocityIoICCH frequencyI

    TABLE I System parameters used in the simulations.with the two dimensional Gaussian distribution are hereafterdenoted as Hot spot cells. With this definition there was 15Hot spot cells. The cells that are not Hot spot cells are hereaf-ter denoted as Uniform cells.

    Realizations of the traffic environments with 2000 mobilesare shown in Figure 2. The shaded cells in the Hot spot caseare the Hot spot cells.C. Measure of Offered Trufic

    The offered traffic in Erlangkell was measured as the aver-age traffic in Erlang over whole the coverage area of the sys-tem divided by the ni :ber of cells in the system. Hence, Xaverage Erlangkell in the Hot spot case implies X/2ErlangAJniform cell and X/2+75/15*X/2 = 3X Erlang/Hotspot cell.D. hreshold Setting

    The threshold was set to 60% (Fthreshold :=0.6)which withour assumptions and system parameters resulted in a 90 per-centile of C/I above 9 dB which in this paper is assumed togive an acceptable quality. An adjustment of the thresholdwould result in another 90 percentile of C/I.E. Reference Systems

    To be able to compare the obtained results, simulationswere also done for a system using a cluster size of 3 with 60%load/cell hard blocking (i.e. 32*0.60 = 19 channelskell) and asystem using a cluster size of 3 with 100% loadkell tradi-tional hard blocking (i.e. 32 channels/cell).

    b) Hot SpotFigure 2. Traffic distributions: a) Uniform. b) Hot Spot.

    V. ResultsA . Uniform Trufic Ca se

    In Figures 3-4, the blocking probabilities and 90 percentileC/I levels for TLA, 60% loadkell and 100% loadkell areshown for the Uniform traffic case.

    Here it is shown that for high loads (offered traffic > 20Erlangkell) the performance of TLA and 60% load/cell areapproximately the same. For interesting loads (offered trafficthat generates a blocking around 2%), however, TLA outper-forms 60%loadkell. 100% loadkell outperforms TLA for allloads (at least for blocking probabilities above 0.1%).

    The quality decreases with offered traffic for all methods.In the 100% loadkell case, the assumed 9 dB quality thresh-old is broken. TLA and 60% loadkell, however, preserve the90 percentiles of C/I around 9 dB even at high loads. Notethat there is no significant difference between TLA and 60%loadkell concerning the quality.

    The offered traffic can be increased from 11.9 Erlangkellto 15.7 Erlangkell and still maintain a 2% blocking withacceptable quality when TLA is used compared to 60%loadkell. This is an increase with 32%.

    1284

  • 8/14/2019 7225059 Admission Control in Frequency Hopping GSM Systems

    4/5

    Blocking probabilityvs offered raffic. Uniform1on

    10.'ff-DD2mx-85 10

    13

    g12--

    I II , , , I5 10 15 20 25 301 -3 Offered traffic (Erlang/cell)

    -

    Figure 3. Block ing probability for uniform traffic distribution

    m510-

    z 9 -ea

    8 -7 -

    6

    90 percentileof C/I vs. offered traffic - Downlink, Uniform15 1 I

    - - - - - - --1_ _ -\- 0% load/cell- - 100% oadlcell - _ - -

    90 percentile of C/I vs offered traffic- Uplink, Uniform:

    Figure 4. 90 percentiles of CII for Uniform traffic distribution

    B. Ho t Spo t TrafJic CaseIn Figures 5 - 6, the blocking probabilities and 90 percen-

    tile C/I levels for TLA, 60% loadcell and 100% load/cell areshown in the Hot spot traffic case. The results comprise thetotal system regardless of cell type.

    Here it is shown that for high loads (offered traffic > 20average Erlangkell), the performance for TLA and 60%loadkell are approximately the same. For the interestingloads (offered traffics that generates a blocking around 2%),the TLA and the 100% oadkell have almost identical perfor-mance.

    The quality decreases with average offered traffic for allmethods. In the 100% oadcell case, the assumed 9 dB qual-ity threshold is broken. TLA and 60% loadkell, however, pre-serve the 90 percentiles of C/I around 9 dB even at high loads.

    With preserved quality, the average offered traffic can beincreased from 1.62 Erlangkell to 2.41 Erlangkell with TLAcompared to 60% load/cell at 2% blocking. This is anincrease with 49%.C. Ho t Spo t TrafJic Case - Hot Spot and Uniform Cells

    For the Hot spot case, the blocking probability and 90 per-centile of C/I levels are shown in Figures 7 - 8 for Uniformand Hot spot cells.

    Here it is shown that the blocking probability in the Uni-form cells are considerable lower than in the Hot spot cells. Infact, comparing Figure 7 with Figure 5 shows that the block-ing in the Hot spot cells dominates the total blocking proba-bility.

    The quality decreases with the average offered traffic but itis preserved for both Uniform and Hot spot cells in the down-link. In the uplink, the quality falls below the 9 dB qualitythreshold for the Hot spot cells. However, antenna diversitywill reduce the necessary threshold in the uplink. Thus, it isno problem to have a worse uplink than downlink.

    VI. ConclusionsThe Traffic Load Admission (TLA) algorithm is shown to

    preserve the quality in the system. Further, TLA is shown towork well in an non uniform traffic environment. However, asystem which does not have an Admission Control algorithmis shown to violate the 90 percentile quality criterion for highloads regardless of traffic environment.

    The TLA algorithm is shown to have better performance interms of blocking compared to a hard blocking method wherethe number of installed transceivers limits the served traffic(TLA can handle 32% more traffic than 60% loadkell). Thegain with the TLA algorithm increases when the traffic is notuniformly distributed (from 32% to 49%, even more if onlyHot spot cells are considered). Further, TLA is shown to havealmost optimum performance for a non uniformly distributedtraffic scenario if the blocking probability is kept at a reason-able level (around 2%).

    1285

  • 8/14/2019 7225059 Admission Control in Frequency Hopping GSM Systems

    5/5

    Blocking probabilityvs offered traffic, Hot spot1 o

    10>1-2gea

    5 10 15 20 25 301o - ~ Offered traffic (average Erlanglcell)

    Figure 5. Blocking probability for Hot spot traffic distribution.

    90 percentile ofCll vs. offered traffic - Downlink,Hot spot

    8 9 1 - 0% loadlcell- - 100% loadlcell

    10 15 20 25 30Offered traffic (average Erlanglcell)90 percentile of CII vs offered traffic- Uplink, tiot spot

    60% load/cell- - TLA- - 100% loadlcell

    5 10 15 20 25 30Offered raffic average Erlanglcell)

    Figure 6. 90 percentiles of C/I for Hot spot traffic distribution

    BlockinaDrObabllltvVS. offered trafflc- TLA. Hot SDOI

    p1 0 - 2

    t 1,5 i o 15 20 25 30Offered traffic (average Eriang/ceil)

    Figure 7. Blocking probability fo r Uniform and Hot spot cells, Hot spottraffic distribution.

    90 percentile of CA YS. offered traffic - TLA, Hot spot15

    14

    13

    g 1 2I$ 1 1m-5 108 9

    6 5 10 15 20 25 30Offered traffic (aver age Erlang/ceii)Figure 8. 90 percentiles of C/I for Uniform and Hot spot cells, Hot spot

    traffic distribution.VII. References

    [I]H Olofsson, J Naslund, B Ritzen & J Skold, Interference Diversity asMeans for Increased Capacity in GSM, In Proceedings of the 1stEuropean Personal and Mobile Communications Conference, 1995, pp.97-102.

    [2]Chen Nee Chuah, Roy D. Yates and David J . Goodman, IntegratedDynamic Radio Resource M anagement, In proceedings of the VTC 95,pp. 584-588.

    [3]Y Argyropoulos, S Jordan and S P R Kumar, Dynamic ChannelAllocation Performance under uneven Traffic Distribution Conditions,In proceedings of the ICC 95, pp. 1855-1859.

    [4]M Andersin, M Frodigh and K-E Sunell, Distributed Radio ResourceAllocation in Highway Microcellular Systems, In proceedings of th e5th Winlab workshop, April 1995, pp. 77-85.[5]M Naghshineh and A S Acampora, Design and Control of Micro-

    Cellular Networks with QOS Provisioning for Real-Time Traffic, Inproceedings of the ICUPC 9 4, pp.376-381.

    [6]M Naghshineh and M Schwartz, Distributed Call Admission inMobilelWireless Networks, In proceeding of the PIMRC 95, pp. 289-293.

    1286