Radio network planning and optimisation for umts 2nd edition phần 10 ppsx

The converse is also possible: that is, capacity that is by default allocatedfor CS use but allowed for EDGE and GPRS use when the packet traffic load is suchthat extra capacity would be d

The conclusion drawn from the results is that the autotuning of cell-based downlinklink maxima and load targets improves significantly the system performance asmeasured with throughput particularly in comparison with cautious or incorrectparameter settings Therefore, the feature is a promising candidate for implementationinto the NMS.

9.3.6 Capacity Optimisation and Traffic Balancing

In the following sections some mechanisms to share resources between circuit switchedand packet switched traffic or between cells is discussed A similar logic can be applied

to GPRS as demonstrated in Chapter 10 The traffic control mechanism betweensystems, inter-system handover, is introduced in Chapter 4

9.3.6.1 Autotuning of P-CPICH Power

The primary objective of the methods presented in this section is to minimise the usage

of power resources for the P-CPICH, while ensuring good enough P-CPICH coverage.This is even more important if the power levels of all other common channels are setwith respect to P-CPICH power – i.e., higher amounts of power resources can be savedand more traffic served The original work is presented in [36]

Method for Autotuning

P-CPICH defines the power of the P-CPICH in the cell Increasing or decreasing thepilot power makes the cell larger or smaller Thus, the tuning of pilot powers can beapplied to balance cell load among neighbouring cells and, additionally, to providesufficient signal reception for the terminals The common pilot coverage issues arediscussed in Section 9.3.3.3

In the rule-based method of [36] the pilot power of a cell was increased or decreased

by 0.5 dB if the cell load was significantly lower or higher than the neighbour cell load

as indicated by statistics in Section 9.3.3, Equation (9.14) If the load was not

Table 9.8 Micro 46-cell scenario: results for circuit switched speech and packet switched traffic

——————————————————————————————————————————

PtxTarget PtxTarget PtxTarget PtxTarget

33 dBm, 33 dBm, 35.5 dBm, 35.5 dBm,fixed offset DL link fixed offset DL link5.5 dB maxima 5.5 dB maxima

significantly unbalanced among the cells, but the pilot signal reception was significantlylower or higher than the target, the pilot power was increased or decreased by 0.5 dB,respectively Pilot power was limited between 3% and 15% of the maximum BS power.Pilot power control actions are presented in Table 9.9.

In Table 9.9 t (load balance) and c (coverage balance) are calculated as follows Thetest statistic of the difference between own-cell and neighbour cell loads was obtainedusing:

The terminals in the sector reported the received Ec=I0of the pilot For each reported

Ec=I0 cell-specific counter Necio was incremented If Ec=I0 exceeded 18 dB, counter

Nover was also incremented The counters were reset at the point of pilot poweradjustment as shown in Table 9.9

The test statistic of the difference between cell coverage and target coverage, C, wascalculated according to:

c0¼ Nffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiover Necio C

Necio C  ð1  CÞ

The target coverage was set to C¼ 0.98 As above, statistic c0was quantified to three

levels of coverage balance:

Table 9.9 Pilot power control actions

Load balance, t Coverage balance, c Pilot power change and counter reset

1 1 Increase pilot and reset all counters

1 0 Increase pilot and reset load counters

1 1 Increase pilot and reset load counters

0 1 Increase pilot and reset coverage counters

0 1 Decrease pilot and reset coverage counters

1 1 Decrease pilot and reset load counters

1 0 Decrease pilot and reset load counters

1 1 Decrease pilot and reset all counters

Results from Simulations

A full set of results for the optimisation of pilot power using a rule-based method ispresented in [36] The mixed macro-cell and micro-cell scenario depicted in Section 9.3.2was used in the simulations Table 9.10 shows the improvement of downlink packetdata performance measures obtained with the autotuning method Table 9.11 showsthat the average downlink total transmission powers (PtxTotal) increased slightly whenrule-based optimisation was applied The initial P-CPICH power values were 5% of thetotal maximum transmission power in each cell – i.e., 200 mW for micro-cells and 1 Wfor macro-cells Increased total BS powers explain the improved performance of rule-based optimisation This can be taken as an indication that the load was more evenlydistributed As the target pilot coverage was 98%, the results in Table 9.11 show thatthe coverage deteriorated with autotuning Coverage could be improved by increasingits weight in the cost function and by adjusting the rule priorities

The results corroborated that the fixed setting of the pilot power –i.e., by default to5% of the maximum BS power – is a warranted choice Coverage was sufficient and thepacket data performance was close to that obtained with the pilot control Thus, pilotpower control may only benefit performance in congested cells However, loadbalancing, which was clearly attained, can benefit single cells whose performance isnot reflected in the total network performance but which subjectively can be highlysignificant The total BS powers show the effect of load balancing obtained with thepilot power control Mean macro-cell total power moved closer to the target of 10 Wand the standard deviation of the power decreased The decrease of standard deviationwas shown in micro-cells as well

Table 9.10 Improvement of packet data performance withpilot power optimisation compared with initial pilot powersetting

Rule-based method[%]

Active session throughput 21

95th percentile of packet delay 5

Table 9.11 Performance results of pilot power optimisation

No optimisation Rule-based methodMacro PtxTotal (std) [W] 9.0 (2.0) 9.4 (1.3)

Micro PtxTotal (std) [W] 1.9 (0.3) 1.9 (0.2)

Macro pilot power (std) [W] 1 (0) 1.6 (0.9)

Micro pilot power (std) [W] 0.2 (0) 0.24 (0.15)

These results suggest that, first, the balancing of load among cells and aiming to achieve

a specific coverage level is feasible using the simple heuristic rules that control pilotpower Second, the pilot power control method improves the air interface performance.Finally, the method is a valid means for improving network operability by itsautomation

In [19] another optimisation technique for adjusting pilot powers in CDMA systems

is presented The idea in this technique is to reduce the unused pilot signals seen bymobiles, thus reducing the number of pilot powers within a certain margin relative tothe strongest pilot The results in [19] showed that the lowering of pilot power pollutiongives some improvement in downlink link coverage and capacity in addition toreduction in deployment efforts spent in optimising pilot powers

9.3.6.2 Autotuning of Dedicated Capacity for Non-Real Time Services or for

High-speed Downlink Packet Access

The basic RRM without the dedicated capacity for NRT services allows only onethreshold for any traffic when performing admission control for the new enteringRAB, when modifying an existing RAB or when performing packet scheduling Thismeans that RT and NRT traffic will use the same entry criteria and in case the cell isfully loaded with RT traffic there will be no room for NRT traffic at all With thededicated NRT traffic capacity feature the operator can guarantee at least somecapacity for NRT traffic as well

The dedicated NRT traffic capacity feature provides uplink and downlink targetpower thresholds for RT and NRT traffic separately This feature improves the QoS,because it provides a possibility to guarantee some capacity for NRT traffic on a cell-by-cell basis The capacity reservation for NRT traffic requires support in NEs in terms

of algorithms and configuration parameters to do the resource reservations in practice

In Figure 9.37 the idea underlying dedicated NRT capacity reservation is presented

In phases A, B and C both RT and NRT traffic are getting the needed capacity, andthere are no traffic restrictions In phase D NRT traffic experiences capacity shortage

Planned target load

Maximum guaranteed NRT traffic capacity

A B C D E F

RT Traffic NRT Traffic

Figure 9.37 Conceptual presentation of the operation of dedicated non-real time traffic capacityreservation

and new RT RAB setups are rejected until NRT traffic gets the capacity it requires.

At point E RT traffic experiences blocking NRT traffic is allocated the capacity that isleft over from the RT traffic At F both traffic types experience blocking, new RT RABsetups are rejected and NRT traffic is given the maximum guaranteed capacity.Optimisation of dedicated NRT traffic capacity would take care that the thresholdcontrolling the size of the dedicated territory would be adaptive Some of the resourcesavailable in the uplink and downlink can be dedicated to NRT traffic During heavyload a tradeoff between RT traffic blocking and NRT traffic queuing can be performed.One possible method is to attach costs to blocked and queued bearers Autotuning can

be done so that the dedicated NRT traffic capacity is increased if the cost of queuedbearers is significantly higher than the cost of blocked bearers, and correspondinglydecreased if the cost of queued bearers is significantly lower than the cost of blockedbearers

Similarly, HSDPA functionality shares the physical and logical resources in terms ofpower and codes with Dedicated Channels (DCHs) Should the HSDPA performance

be degraded due to lack of power resources or codes, a similar method to the one aboveshould reallocate physical and logical resources optimising the performance of HSDPAchannels and of DCHs

9.3.6.3 Intra-frequency Traffic Balancing Using Cell Individual Offsets

As discussed in Chapter 4 handovers within the UTRA-FDD system can be classified asintra-frequency handovers and inter-frequency handovers In intra-frequency softhandover, an MS is allowed to connect simultaneously to several BSs, which areadded or removed from the terminal’s active set by applying relative handoverthresholds The most important ones are the addition threshold, the addition timer,the dropping threshold and the dropping timer In principle, if a received P-CPICH

Ec=I0 from a new BS is within a window defined by the addition threshold relative tothe best serving BS’s Ec=I0for a time period longer than the addition timer, it is addedinto the user’s active set When the P-CPICH Ec=I0from a BS in the active set is lowerthan the P-CPICH Ec=I0 of the best serving BS by a margin defined by the droppingthreshold that BS is removed from the active set Typically, the measurement quantity

is P-CPICH Ec=I0 but it can also be path loss

Further, a Cell Individual Offset (CIO) value can be used to make one neighbouringcell more attractive than another This is demonstrated with Figure 9.38 In order tomake the handover to a cell with P-CPICH 3 happen earlier, an offset is applied tomanipulate the terminal’s decision The offset raises the P-CPICH 3 curve

The terminal measures the Ec=I0levels of the pilot signals of neighbouring cells Theterminal initiates changing of the active set by sending a measurement report and anASU request to the RNC The reporting conditions have the following general form:CPICHðmonitoredÞ þ AdjsEcNoOffsetðbest; monitoredÞ

> ReportingCriterionðCPICHðbestÞÞ ð9:27Þwhere CPICHðmonitoredÞ and CPICHðbestÞ are the measurement results (Ec=I0)

of the monitored cell and the best active set cell, respectively; and

AdjsEcIoOffsetðbest; monitoredÞ is the cell individual offset added to the measured Ec=I0

of the monitored cell It is specific to the primary cell in the active set – i.e., there is anoffset for each neighbour of a cell

The neighbour set for a specific combination of cells in the active set is a union orintersection of the neighbour sets of the individual active set cells formed using aparticular method The maximum size of the combined neighbour set is 32 cells.When an ASU is made, the terminal gets signalled the new neighbour set and CIO-related information

The idea of congestion relief is to utilise these CIOs to force traffic from a highlycongested cell to neighbouring cells that are less loaded (see Figure 9.39) Such asituation applies, for example, in the business areas of city centres A few highlyloaded cells might serve certain office buildings, and the surrounded cells are lowloaded during business hours Thus these surrounding cells can be taken to serve thebusiness complex with the proposed method The offsets between two cells A and B –i.e., AdjsEcIoOffsetðA; BÞ and AdjsEcIoOffsetðB; AÞ are adjusted if the ratios of blockedcalls differ significantly between the cells Blocking can be measured and evaluated inseveral ways – for instance, as:

soft blocking due to insufficient power resources (downlink total transmission powerexceeding its target level);

Figure 9.38 Cell individual offset A positive offset is applied to P-CPICH 3 before eventevaluation in the terminal

Figure 9.39 Conceptual presentation of the congestion relief logic After optimisation theaverage behaviour is the same, but the blocking performance is improved for the highly congestedcell

hard blocking due to insufficient hardware or logical (codes) resources; or

hard and soft blocking combined, soft handover overhead-related information – ifabnormal, increased blocking may be due to a poorly set AdditionWindow which

is indicated to the user The user then may wish to perform AdditionWindowoptimisation first

The proposed control method gives the best gain if insufficient hardware resourcescaused the blocking but the method is able to balance the load from soft-blocked cells

to other cells as well The combination of hard and soft blocking as the blockingmeasure is the best solution

The algorithm collects blocking statistics during specified hours and/or loadconditions and, as soon as significant differences between the blocking of a cell pair

is detected, handover event-triggering parameters between the cells are adjusted inorder to balance the load with handover actions With conservative setting of controlmethod parameters, the method reacts slowly to differing blocking ratios If the averageblocking ratio in a cell pair is 2%, the number of samples required for detecting ablocking ratio difference with sufficient statistical accuracy is some hundreds in bothcells

When a blocking ratio difference bigger than a certain threshold is detected, theparameters for event-triggered measurement reporting are changed slightly;the change in the CIO values is a function of the difference between the blockingratio of the cells in question

The control method inherently assumes that the downlink is the limiting directionwith respect to power resources If the load is high in the uplink, the control actions ofthe method can cause terminals to run out of power Thus, control actions should bemade cautiously with the operator monitoring the cells with high uplink load.Load-balancing Process

CIOs are a tool to move the cell border Thus, adjusting the offsets can reduce traffic incongested cells and increase traffic in low loaded cells Traffic in congested cell A ismoved to a neighbouring less loaded cell B by decreasing AdjsEcIoOffsetðB; AÞ andincreasing AdjsEcIoOffsetðA; BÞ Decreasing AdjsEcIoOffsetðB; AÞ inhibits softhandovers from cell B to cell A – i.e., cell A is more difficultly added to andmore easily dropped from the active set when the user is close to cell B IncreasingAdjsEcIoOffsetðA; BÞ makes users close to cell A favour cell B in the soft handover,which moves traffic from other neighbours of cell A to cell B as well The algorithm mayhave the following internal parameters that the user can adjust:

Normal – the level of blocking ratio that can still be considered normal Its valuemust not be zero

Step – the adjustment step of the CIOs in decibels Its range begins with 0.1 dB; it can

be a function of the differences in cell-blocking ratios

Max – the maximum absolute value of the CIOs

Threshold – the threshold for indicating a significant difference in the blocking ratios.The parameter determines the sensitivity required to make an offset adjustment Itsconservative values lie between 2% to 3% in absolute terms However, setting itcloser to zero can increase adaptation speed without significant adverse effects

Control is performed for the selected group of cells periodically The actual value ofthe period is not crucial and it can be as often as is practically possible; a change ismade when a significant difference in the blocking rates is detected The controlalgorithm is described using the steps in the following list, but, before optimisation,the planned CIOs of a selected group of cells, C, are stored in the reference configura-tion management database:

1 Iterate Steps 2 to 10 for all cell pairs (c1; c2) in the cell group selection Note: Do notrepeat steps for a cell pair (c2; c1) if (c1; c2) is already processed

2 Obtain the measured blocking ratios for cells (c1; c2) from the performancemanagement database and KPI calculation engine

3 Obtain the current CIO values from the configuration management database

4 Check for situations when blocking ratios are not greater than Normal If theblocking statistics show that cell blocking is at an allowed level and the difference

in blocking ratios between cells is less than a pre-defined threshold then move theoffsets towards their initial planned values in the reference database (provided thatvalues differ) and continue with Step 10 Otherwise, continue with the next step

5 Retrieve the KPI for the average blocking ratio

6 Compute the deviation, D, using the blocking ratios (B1; B2 for cells 1 and 2,respectively; N1; N2 being the number of samples, Nlimit being, e.g., 5) This isobtained using:

D¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiB1 B2



Bð1  BÞ

1

N1þ 1

N2



where Bis the average blocking However, D is zero if:

min½ B; ð1  BÞ  N1< Nlimit or min½ B; ð1  BÞ  N2< Nlimit ð9:29Þ

7 Compute the change in the CIOs:

DOffset ¼ StepStep if D < Thresholdif D> Threshold

(

ð9:30Þ

in which Step is the CIO adaptation step

8 Compute the new CIOs with range checking

9 If both computed new CIOs of cells c1; c2 differ from their current setting, changethe CIOs and provision the change to network

10 If there are unprocessed cell pairs, take the next one and continue with Step 2

11 Reset the congestion measurement counters and KPIs of the cells whose CIOs werechanged

It is the operators’ choice what level of congestion to allow and tolerate even after thecongestion relief algorithm The proposed method is not an answer to all blocking-related problems, but it can solve certain traffic hotspot-type situations

Another identified application area for CIO optimisation is for areas along highways.The soft handovers of a mobile user can be controlled by prioritising the adjacencydefinitions using offset values Cells intended to cover the highway are higher prioritised

in handover evaluation and thus unnecessary ASUs (cell addition and immediatedeletion again), involving cells aside the highway not intended for highway usage butwhich can be locally received, can be avoided This would reduce signalling load andwould be especially beneficial for fast-moving mobiles.

In this chapter advanced analysis methods for cellular networks were introduced NMSlevel intelligence is needed in order to cope with the challenges arising from theincreased amount of traffic and new mobile services Further, some WCDMA-specific automation examples were presented The presented methods bring first ofall operational efficiency, owing to the high level of process, analysis and decisionlogic automation Second, with automation network performance is improved andnetwork resources are used more efficiently

References

[1] ETSI, TS 100.908, v8.10.0, GSM Technical Specification 05.02: Digital Cellular munications System (Phase 2þ); Multiplexing and Multiple Access on the Radio Path,2001

Telecom-[2] Vehvila¨inen, P (2004) Data mining for managing intrinsic quality of service in digitalmobile telecommunications networks Thesis (Doc.Tech.), Tampere University ofTechnology

[3] Fayyad, U., Piatetsky-Shapiro, G and Smyth, P., From data mining to knowledgediscovery: An overview In: U Fayyad, G Piatetsky-Shapiro, P Smyth and R.Uthurusamy (eds), Advances in Knowledge Discovery and Data Mining, pp 1–34, MITPress, 1996

[4] Laiho, J., Raivio, K., Lehtima¨ki, P., Ha¨to¨nen, K and Simula, O., Advanced AnalysisMethods for 3G Cellular Networks, Report A65, Publications in Computer and Informa-tion Science, Helsinki University of Technology, 2002 Modified version resubmitted toIEEE Transactions on Wireless Communicationsend 4/2002

[5] Han, J and Kamber, M., Data Mining: Concepts and Techniques, Morgan Kaufmann,2001

[6] Ho¨glund, A.J., Hatonen, K and Sorvari, A.S., A computer host-based user anomalydetection system using the self-organising map Proc IEEE-INNS-ENNS InternationalJoint Conf on Neural Networks (IJCNN 2000), Vol 5, pp 411–416, 2000

[7] Pyle, D., Data Preparation for Data Mining, Morgan Kaufmann, 1999

[8] Breiman, L., Friedman, J., Olshen, R and Stone, C., Classification and Regression Trees,Chapman & Hall/CRC Press, 1984

[9] Vesanto, J and Alhoniemi, E., Clustering of the self-organising map, IEEE Transactions

[12] Kohonen, T., Self-organising Maps, Springer-Verlag, 1995.

[13] Kohonen, T., Analysis of processes and large data sets by a self-organising method, Proc

of 2nd International Conf on Intelligent Processing and Manufacturing of Materials, 1999,vol 1, pp 27–36

[14] Kohonen, T., Oja, E., Simula, O., Visa, A and Kangas, J., Engineering applications of theself-organising map, Proceedings of the IEEE, 84(10), October 1996, pp 1358–1384.[15] Kohonen, T., New developments and applications of self-organising maps, Proc of Inter-national Workshop on Neural Networks for Identification, Control, Robotics, and Signal/Image Processing, 1996, pp 164–172

[16] Ahola, J Alhoniemi, E and Simula, O., Monitoring industrial processes using the organising map, Proc of IEEE Midnight Sun Workshop on Soft Computing Methods inIndustrial Applications, 1999, pp 22–27

self-[17] Ha¨ma¨la¨inen, S., Holma, H and Sipila¨, K., Advanced WCDMA radio network simulator,Proc of PIMRC 1999, Aalborg, Denmark, October 1997, pp 509–604

[18] Raivio, K., Simula, O and Laiho J., Analysis of mobile radio access network using the organising map, Proc of IEEE International Conf on Data Mining, San Jose, California,November/December 2001, pp 457–464

self-[19] Vesanto, J., Himberg, J., Alhoniemi, E and Parhankangas, J., SOM Toolbox for Matlab 5,Report A57, Helsinki University of Technology, 2000

[20] Vehvila¨inen, P, Ha¨to¨nen, K and Kumpulainen, P., Data mining in quality analysis ofdigital mobile telecommunications network, Proc of XVII IMEKO World Congress,Dubrovnik, Croatia, June 22–27, 2003, pp 684–688

[21] Suutarinen, J., Performance measurements of GSM base station system Thesis (Lic.Tech.),Tampere University of Technology, 1994

[22] Ha¨to¨nen, K., Kumpulainen, P and Vehvila¨inen, P., Pre- and post-processing for mobilenetwork performance data, In: R Tuokko (ed.), Automaatio03, Seminaaripa¨iva¨t [Automa-tion Makes it Work]: Automaation sovellukset ja ka¨ytto¨kokemukset, September 9–11, 2003,

pp 311–316, Finnish Society of Automation

[23] Ha¨to¨nen, K., Laine, S and Simila¨, T., Using the LogSig-function to integrate expertknowledge to Self-Organising Map (SOM) based analysis, IEEE International Workshop

on Soft Computing in Industrial Applications, Birmingham University, New York, June 23–

25, 2003, pp 145–150

[24] Johnson, R.A and Wichern, D.W., Applied Multivariate Statistical Analysis (4th edn),Prentice Hall, 1998

[25] Everitt, B.S., Cluster Analysis, Edward Arnold, 1993

[26] Davies, D.L and Bouldin, D.W., A cluster separation measure, IEEE Transactions onPattern Analysis and Machine Intelligence, 1(2), pp 224–227, April 1979

[27] Rousseeuw, P.J., Silhouettes: A graphical aid to the interpretation and validation of clusteranalysis, Journal of Computational and Applied Mathematics, 20, November 1987, 53–65.[28] McGill, R., Tukey, J.W and Larsen, W.A., Variations of boxplots, The AmericanStatistician, 32, pp 12–16, 1978

[29] Ha¨ma¨la¨inen, S., Slavina, P., Hartmann, M., Lappetelainen, A., Holma, H and Salonaho,O., A novel interface between link and system level simulations, Proc ACTS Summit 1997,Aalborg, Denmark, October 1997, pp 599–604

[30] Ho¨glund, A and Valkealahti, K Automated optimisation of key WCDMA parameters,Journal of Wireless Communications and Mobile Computing, in press

[31] Olofsson, H., Magnusson, S and Almgren, M., A concept for dynamic neighbor cell listplanning in a cellular system, Proc 7th IEEE International Symposium on Personal, Indoorand Mobile Radio Communications (PIMRC’96), pp 138–142

[32] Love, R.T., Beshir, K.A,., Schaeffer, D and Nikides, R.S, A pilot optimisation techniquefor CDMA cellular systems IEEE VTS 50th Vehicular Technology Conf., VTC 1999, Fall

1999, Vol 4, pp 2238–2242

[33] 3GPP, TS 25.133, v3.50, Requirements for Support of Radio Resource Management, 2001.[34] Valkealahti, K., Ho¨glund A and Novosad, T., UMTS radio network multi-parametercontrol, Proc IEEE PIMRC 2003, pp 616–621

[35] Valkealahti, K., Ho¨glund, A., Parkkinen, J and Flanagan, A., WCDMA common pilotpower control with cost function minimisation, Proc IEEE 56th VTC Fall 2002, Vol 4,

[39] Zhu, H., Buot, T., Nagaike, R and Schreuder, H., Load balancing in WCDMA systems

by adjusting pilot power, Proc 5th International Symposium on Wireless PersonalMultimedia Communications 2002, Vol 3, pp 936–940

[40] Lee, W.C.Y and Lee, D.J.Y, Optimise CDMA system capacity with location, Proc 54thIEEE VTC, Atlantic City, NJ, October 2001, Vol 2, pp 1015–1019

[41] Flanagan, A and Novosad, T., Automatic selection of AdditionWindow in a WCDMAradio network based on cost function minimisation, Proc IEEE ISSSTA, September 2002,Prague, Vol 3, pp 672–676

[42] Ho¨glund, A., Po¨llo¨nen, J., Valkealahti, K and Laiho, J., Quality-based auto-tuning of celluplink load level targets in WCDMA, Proc IEEE VTC, Spring 2003, Vol 4, pp 2847–2851

[43] Ho¨glund, A and Valkealahti, K., Quality-based tuning of cell downlink load target andlink power maxima in WCDMA, Proc IEEE 56th VTC, Fall 2002, Vol 4, pp 2248–2252

Other 3G Radio

Access Technologies

Jussi Reunanen, Simon Browne, Pauliina Era¨tuuli,

Ann-Louise Johansson, Martin Kristensson, Jaana Laiho,

Mats Larsson, Toma´sˇ Novosad and Jussi Sipola

This chapter deals with two technologies that are different from UMTS TerrestrialRadio Access Frequency Division Duplex (UTRA FDD) Section 10.1 discussesthe General Packet Radio Service (GPRS) used in Global System for Mobilecommunications (GSM) technology GPRS brings variable-rate packet data trafficinto the air interface of what was originally a Circuit Switched (CS) and singledata rate service-oriented technology Thus, this technique is now paving the way inpacket data communications towards Third Generation (3G) and Universal MobileTelecommunications System (UMTS) From the planning point of view, GPRS andits modifications are affected by the variability in user data rate in a somewhat similarway to Wideband Code Division Multiple Access (WCDMA)

The second new technology, covered in Section 10.2, is the Time Division Duplex(TDD) mode of WCDMA (UTRA TDD), a potentially interesting technology for highdata rate indoor users Unlike FDD, TDD does not need a paired spectrum, but itpresents several potential coexistence problems with FDD due to mutual interference,

as discussed later in this chapter

This part of the chapter deals with issues relating to the planning of GPRS andEnhanced GPRS (EGPRS) services on the GSM network Data rate variability hasintroduced another variable into network planning This also affects air interface andtransmission issues

Radio Network Planning and Optimisation for UMTS Second Edition

Edited by J Laiho, A Wacker and T Novosad # 2006 John Wiley & Sons, Ltd

10.1.1 Introduction

Packet data services were introduced in GSM with the GPRS By May 2001 themajority of the existing GSM networks supported or were about to support theGPRS service GPRS will enable packet data rates of up to 20 kbps per timeslot overthe existing GSM network; four possible coding schemes are implemented

Due to the ever-increasing pressure to boost throughput and data service speed, anenhancement to GPRS – namely, EGPRS, which is part of EDGE (Enhanced Data forGlobal Evolution) – was developed and standardised during 1999 EGPRS will enhancedata throughput up to around 60 kbps per timeslot

EDGE is a common convergence of two standards, from the US tions Industry Association (TIA) and the European Telecommunications StandardsInstitute (ETSI) (lately ETSI specification work was transferred to 3GPP) Standards(Table 10.1)

Telecommunica-EDGE services will be carried over GSM/GPRS networks, utilising their existingcontrol channels and traffic channels Interim Standard 136 (IS-136 or TDMA in theUS) is based on a different approach, providing only data services in a relatively smallfreed dedicated spectrum

In this chapter some general ideas about how to plan and optimise GPRS and EDGE(over the GSM/GPRS) networks are discussed It should be noted that GPRS refers tofeatures (or issues) applicable only to GPRS, and EDGE refers to features (or issues)applicable only to EDGE The EDGE issues discussed in this chapter are onlyconsidered for the EDGE system over GSM system

10.1.2 Modulation and Coding Schemes

GPRS uses Gaussian Minimum Shift Keying (GMSK) modulation Four codingschemes are defined: CS-1 to CS-4 CS-1 offers the highest level of error protection,while CS-4 offers no error protection of user content, as shown in Table 10.2 [6] In

Table 10.1 Standard convergence

Specification status GSM (ETSI SMG2) TIA/EIA 136 (TIA TR 45.3)(standardisation group)

circuit-switched infrastructure circuit-switched infrastructuremax 14.4 kbps max 9.6 kbpsEnd 1998þ GPRS 200 kHz, GMSK 30 kHz, 8-PSK

GPRS infrastructure GPRS infrastructuremax 171.2 kbps max 52.2 kbps

GPRS infrastructuremax 473.6 kbps

consequence, the user data rate increases with higher coding schemes, at the expense of

an increasing signal-to-interference level requirement

For EDGE, both GMSK and 8-Phase Shift Keying (8-PSK) are defined asmodulation schemes, and for both of these there are several different code rates: seeTable 10.3 [6] EDGE offers user bit rates between 8.8 kbps and 59.2 kbps perradio timeslot The use of 8-PSK allows for a trebling of the air interface bit ratewhen compared with GMSK, albeit with increased signal-to-interference ratiorequirements [7]

10.1.2.1 Protocol Stack

The GPRS or EDGE air interface consists of a layered protocol structure that providescontrol procedures, such as error correction and retransmission, to user data Theprotocols that should be considered when analysing the air interface performance areshown in Figure 10.1, and are used between the Mobile Station (MS) and the BaseStation Controller/Serving GPRS Support Node (BSC/SGSN)

Table 10.2 GPRS coding schemes

Scheme Code rate Radio blocka Coded bits Punctured bits Data rate

a Excludes uplink state flag and binary coded signal bits.

Table 10.3 EDGE modulation and coding schemes

Scheme Code rate Header Modulation RLC blocks Raw data Raw data

code rate per radio within one rate

block radio block

A brief explanation of the function of the protocols is given below:

Sub-Network Dependent Convergence Protocol (SNDCP) Maps the network-levelPacket Data Units (N-PDUs) to the underlying Logical Link Control (LLC) layer.Also provides optional compression functionality, of both the Transmission ControlProtocol/Internet Protocol (TCP/IP) header and the data content

Logical Link Control (LLC) layer Provides a reliable ciphered link between theSGSN and the MS This protocol is independent of underlying radio interfaceprotocols The layer can be operated in both acknowledged and unacknowledgedmodes, and this is one of the parameters defined by the reliability class field present in

a Packet Data Protocol (PDP) context Quality of Service (QoS) profile [8]

Radio Link Control (RLC) layer A key layer in the air interface, this providesreliable transmission of data using optional Automatic Repeat reQuest (ARQ) \barfunctionality In addition, segmentation/desegmentation of data from/to the LLClayer is performed The RLC layer can be operated in both acknowledged and

\bar unacknowledged modes, and, as with LLC, this is defined by the reliabilityclass in the QoS

Medium Access Control (MAC) layer This layer controls MS access to the commonair interface and provides scheduling of the associated signalling

GSM Radio Frequency (RF) The GSM Time Division Multiple Access (TDMA)physical interface Bit interleaving, modulation/demodulation and power control areexamples of functionality within this layer

Figure 10.2 shows the typical data block format within the MS–GPRS protocolstack Application level data may split into multiple TCP and IP blocks, depending

on data volume and TCP packet size Typically, each TCP/IP packet then maps one toone through the SNDCP and LLC layers before being split into a number of RLCblocks These have a header added in the MAC layer before being sent on the airinterface in four bursts, which are sent over four consecutive TDMA frames, with anaverage duration of 20 ms

L1bis

NW sr BSSGP

SNDCP LLC

BSSGP

NW sr L1bis BSS

Figure 10.1 SGSN–MS protocol stack

Performance over the GPRS network is very dependent on the interaction betweenthe different layers in the protocol stack, and these interactions must be understood ifthe network is to be optimised.

In particular, the TCP layer is found to interact strongly with RLC [3]: reliability atthe RLC level can have a serious impact on TCP throughput and this implies the use ofRLC acknowledged mode where TCP is used In addition, high RLC block error ratescan lead to significant delay at LLC, which, in turn, can cause TCP congestionavoidance algorithms to trigger The use of User Datagram Protocol (UDP) ratherthan TCP as a transport protocol allows for a more transparent mode of service,and, if limited packet loss can be tolerated, permits the RLC and LLC layers to beoperated in unacknowledged mode

10.1.3 EDGE Radio Link Performance

In this section, EDGE radio link performance is analysed using link-level simulations[1] The input to the simulation is a characterisation of the radio channel, includingsignal-to-noise or Carrier-to-Interference (C/I) ratio, delay profile and fading charac-teristics The most important quantity to be analysed is throughput, but also delay isdiscussed

10.1.3.1 Simulation Assumptions

The simulations in this section are made assuming one MS and one timeslot Thechannel is a typical urban channel with the MS moving at 3 km/h, in the 900 MHzband Ideal frequency hopping is used These simulations study the performance in aninterference limited network, and C/I is the ratio of the corresponding mean powers.One continuous interferer is modelled using the same modulation and channel statistics

as the wanted signal Simulation length is 40000 bursts

SN-DATA PDUs LLC

The RLC protocol is modelled with the following parameters: transmitter window

128 RLC blocks, polling every 320 ms, roundtrip delay 220 ms Temporary Block Flow(TBF) establishments and releases are not modelled, but a continuous data stream isassumed

10.1.3.2 Performance without Enhancements

The purpose of this analysis is to show the basic link level performance of EDGE,without any link level improvements Later, the effect of two such improvements(incremental redundancy and link adaptation) will be analysed

Block Error Rate (BLER)

The RLC layer produces RLC blocks, which are mapped to radio blocks at the physicallayer Each radio block consists of four normal bursts, and the average duration of aradio block is 20 ms A radio block contains one or two RLC blocks depending on thecoding scheme (see Table 10.3)

BLER is the percentage of erroneously received RLC blocks The ARQ protocolcauses a retransmission for each erroneously received block Therefore, BLER is alsothe ratio of the number of retransmissions to the number of all transmissions.Figure 10.3(a) shows the BLER of the 8-PSK coding schemes As can be seen, ahigher Modulation and Coding Scheme (MCS) always has a higher BLER

MCS-5 MCS-6 MCS-7 MCS-8 MCS-9

can have multiple users and/or a user may use multiple timeslots Therefore, thethroughput experienced by the user is not necessarily the same as the throughput pertimeslot.

If the BLER is known, throughput can be calculated as tp¼ tppeak ð1  BLERÞ,where tppeakis the maximum throughput of the MCS The throughput curves in Figure10.3(b) show how the optimal MCS (that giving the best throughput) depends on theC/I value

10.1.3.3 Incremental Redundancy (IR)

IR combines channel coding and ARQ protocol It is based on soft combining differenttransmissions of the same block at the receiver, thereby increasing the probability ofcorrect reception of retransmissions Original transmissions are not affected andtherefore IR has no effect unless ARQ is used

The EDGE standard supports IR operation by specifying separately coded headers

to identify blocks before channel decoding, different puncturing schemes for eachcoding scheme and a requirement for IR combining capability at the MS receiver

Operation

Figure 10.4 shows the operation of the transmitter and the receiver when IR is used Atthe transmitter, the payload is encoded using convolution coding at the rate 1/3 Thecodeword is punctured by removing two of the three bits The remaining bits aretransmitted with an effective code rate of 1 Retransmissions are performed in thesame way, except that different sets of bits are punctured This means changing thepuncturing scheme For each transmission, the effective code rate is 1, since the number

of bits transmitted equals the number of bits in the payload In EGPRS, codingschemes MCS-4 and MCS-9 have this property For other MCSs, fewer bits arepunctured but the same principle still applies

payload codeword 1st transmission

correct word

decoding

decoding

2 2 2 2

1 1 1 1

3 3 3 3

At the receiver, the first transmission is received normally by performing puncturing and decoding In this example, the decoding fails, so a new transmission

de-is performed However, the received bits are now combined with those of the firsttransmission, yielding an effective code rate of 1/2 This combination is then fed tothe channel decoder The increased redundancy increases the probability of correctreception If, however, the second transmission still fails, there will be a thirdtransmission, which will be combined with the first two transmissions

Benefits of Incremental Redundancy

The gain of IR is significant when observing the throughput of an individual codingscheme Figure 10.5(a) shows the throughput with and without IR The gain is veryhigh when throughput is low relative to the peak throughput For example, at 20 kbpsthe gain is 9 dB, but at 50 kbps it is less than 1 dB The reason is that IR only improvesretransmissions, and at a lower throughput there are more of them

The gain at low throughputs means that a high MCS can achieve the samethroughput as a low coding scheme even at a low C/I This is seen in Figure 10.5(b),which illustrates all the 8-PSK coding schemes of EDGE This also makes MCSselection much easier, which is also a big benefit of IR

Effect of Finite Receiver Memory

The receiver needs to store the soft values of the incorrectly received blocks until theblocks are correctly received Because of the nature of the selective ARQ and depending

on the polling frequency and roundtrip delay, tens of blocks can be incomplete at agiven time instant Therefore, the memory consumption of the receiver increases alongwith the number of required retransmissions

MCS-5, IR MCS-6, IR MCS-7, IR MCS-8, IR MCS-9, IR

a) MCS-9   b) Different MCSs with IR  !""  !  

Figure 10.5 Performance of incremental redundancy

Figure 10.5(a) also shows the performance with finite receiver memory In the curvelabelled ‘20 ksv’, the receiver is assumed to be able to store 20480 soft values (softdecisions at the output of the equaliser) For low C/I, the high number of retransmis-sions means that not all blocks can be stored in memory, therefore the throughput isdecreased For high C/I, ideal IR performance is achieved Finite receiver memory isone of the reasons for using a lower MCS when C/I is low.

10.1.3.4 Link Adaptation (LA)

The task of LA is to select the best performing coding scheme for each channelcondition The need for link adaptation is evident if incremental redundancy is notused (see Figure 10.3) If IR is used, a high-coding scheme such as MCS-9 could always

be used, except that the high number of retransmissions causes a long delay and memory consumption

high-Ideal Link Adaptation

Ideal LA is a conceptual algorithm that selects the coding scheme with best throughputfor each C/I value Link simulations do not model shadowing or path loss, but theireffect is assumed to be included in the C/I value Therefore, ideal LA will adapt toshadowing and path loss Fast fading is not included in the C/I value but it is reallysimulated, so ideal LA does not adapt to fast fading In terms of throughput plots, thethroughput of ideal LA is the envelope of all the coding schemes

Sometimes it is even possible to achieve a better performance than with ideal LA, iffast fading can be taken into account, or if IR is utilised efficiently by using a highercoding scheme for retransmissions than for original transmissions The performance of

a real LA algorithm can be evaluated by comparing it with ideal LA

Bit Error Probability (BEP) Measurements

The EDGE radio standard defines BEP measurement for the MS BEP is defined as theprobability of a bit error during a burst in a given channel instance BEP measurementincludes both the mean and variance of burstwise BEP Variance is calculated duringfour bursts of a radio block, and therefore reflects any fast changes in the channel –e.g., due to frequency hopping or high mobile speeds

Simulation Results with Incremental Redundancy and Link Adaptation

Figure 10.6(a) shows the throughput of EDGE with both link-level enhancements – IRand LA A realistic (20480 soft values) receiver memory is assumed It can be seen that

LA is able to select the optimal MCS is most cases

Delay Analysis

Delay of an LLC frame is measured from the radio transmission of the first bits of theframe to the time when the block is completely and correctly received In this analysis,LLC frames are 8192 bits The delay experienced by the user will additionally includeany queuing in various buffers as well as delays caused by the core network or higherlayers, but these are not taken into account here

Figure 10.6(b) shows the delays of individual coding schemes and LA The maincriterion for LA is throughput and not delay, but the delay achieved with LA is stillrather low, usually less than 0.5 s.

10.1.4 GPRS Radio Link Performance

In this section, the performance of the GPRS radio link is analysed using link-levelsimulations Input to the simulation is a characterisation of the radio channel, includingsignal-to-noise or C/I ratio, delay profile and fading characteristics The mostimportant quantity to be analysed is throughput

The actual performance of each of the four GPRS coding schemes is dependent uponthe channel C/I The results shown in Figure 10.7 are based on a frequency-hoppingenvironment and the typical urban 3 km/h channel type (TU3) In a non-hopping casethe crossing points of the coding scheme are different

Ideal LA ensures that the coding scheme changes from one coding scheme to another

as the C/I increases or decreases to maximise the user data rate In practice the BLER of

a TBF is obtained by recording the number of RLC blocks that require retransmission.This is used as the basis to make the LA decision to change from one coding scheme toanother

The maximum throughput per timeslot depends on the number of coding schemesimplemented If only CS-1 and CS-2 have been implemented, it is clear that CS-2provides a better throughput in most cases The C/I has to be lower than 6.5 dBbefore CS-1 will provide the highest data rate For a non-hopping cell this crossing

MCS-1 MCS-2 MCS-3 MCS-4 MCS-5 MCS-6 MCS-7 MCS-8 MCS-9 LA

  #$

Figure 10.6 Performance of bit error probability-based link adaptation in typical urban channel

3 with intelligent frequency hopping, incremental redundancy memory size 20 ksv

point is even lower, hence CS-2 will provide a better throughput in all cases when aconnection can be maintained.

10.1.5 Coverage

In GPRS and especially in EDGE the coverage area shrinks when the throughput pertimeslot increases as the required Es=N0 (Eb=N0 for GPRS) increases when highercoding schemes are required Coverage can be increased by using link enhancementfeatures Cell ranges without and with enhancements for different coding schemes (CS-1

to CS-4 for GPRS; MCS-1 to MCS-9 for EDGE) are discussed in this section.The C/I – i.e., the quality of coverage – gives another dimension to the achievablethroughput (this was discussed in Section 10.1.3)

10.1.5.1 Input Parameters

The coverage discussion presented in this section is based on link simulations (Section10.1.3.1) and the simple coverage assumptions listed below:

outdoor urban environment considered;

frequency band used: 900 MHz;

slow fading margin: 7.36 dB;

Base Transceiver Station (BTS) output power: 43 dBm (41 dBm with 8-PSK); MS noise figure: 10 dB;

Antenna gain (MS, BTS): 2 dBi, 18 dBi (þdiversity gain: 3 dB for BTS receiving end); propagation model: COST231 Okumura–Hata with 4 dB area type correction factor; BTS/MS antenna height: 25 m/1.5 m;

Required Eb=N0 (GMSK) and Es=N0 (8-PSK) are simulated by using a simulatorwith a TU3 non-frequency hopping multi-path profile

Using the above parameters, the normal (GMSK) speech cell range is calculated as3.9 km

0 5 10 15 20 25

10.1.5.2 Cell Range without Enhancements

In Figure 10.8 cell range vs throughput per timeslot is presented Graph (a) is forGMSK modulation and graph (b) for 8-PSK modulation Both graphs are withoutany improvement effects

As can be seen from Figure 10.8, cell range decreases rapidly with increasing datarate, especially when 8-PSK modulation is used If the network was designed for anoutdoor speech service with a 7.36 dB slow fading margin, the achieved maximumthroughput per timeslot from the network (only the coverage dimension included) isabout 11 kbps for GMSK and about 8 kbps for 8-PSK When comparing the GMSKand 8-PSK cell ranges for certain throughputs per timeslot, it should be rememberedthat there is 2 dB backoff in transmitted power when 8-PSK is used in order to securetransmitter linearity When comparing CS-1 to CS-4 cell ranges with MCS-1 to MCS-4ranges, refer to Tables 10.2 and 10.3 to see the difference in coding rates

10.1.5.3 Incremental Redundancy

IR increases the achieved throughput as explained in Section 10.1.3.3 In Figure 10.9the effect of IR on the throughput vs cell range curves is shown Input parameters arethe same as mentioned in Section 10.1.5.1 It should be noted that IR is only applicablefor EDGE

Figure 10.9 shows that IR enhances throughput per timeslot for ranges above about

1 km Comparing the range with the speech cell range of 3.9 km, note that theachievable throughput per timeslot is about 12 kbps with GMSK and about 11 kbpswith 8-PSK The increase after introducing IR was only 1 kbps per timeslot for GMSKmodulation but 3 kbps per timeslot for 8-PSK

0 10 20 30 40 50 60

MSC5 MSC6 MSC7 MCS8 MCS9

Figure 10.8 Cell range for different coding schemes

10.1.5.4 Downlink Diversity

One way to enhance coverage is to introduce downlink diversity methods – e.g., byusing two transmitters with a certain symbol delay between the two transmissions.Downlink diversity will add 3 dB to received power, and in addition the symboldelay will add or subtract an extra1 dB to þ1 dB (depending on the environment)

to the link budget Figure 10.10 represents throughput per cell range with downlink

0 10 20 30 40 50 60

MSC5 MSC6 MSC7 MCS8 MCS9 MSC5 - IR MSC6 - IR MSC7 - IR MCS8 - IR MCS9 - IR

0 10 20 30 40 50 60

7.0 6.0 5.0 4.0 3.0 2.0 1.0

MSC5 MSC6 MSC7 MCS8 MCS9 MSC5 - IR MSC6 - IR MSC7 - IR MCS8 - IR MCS9 - IR

Figure 10.10 Cell range for different coding schemes with downlink diversity scheme

diversity The Eb=N0 and Es=N0 values are taken from simulations provided by thesimulator presented in Section 10.1.3.1.

As can be seen from Figure 10.10, downlink diversity increases throughput pertimeslot for large ranges For the 3.9 km cell range the achievable throughput pertimeslot for 8-PSK is about 26 kbps with downlink diversity compared with 11 kbpswith only IR For GMSK the corresponding throughputs per timeslot are 16.5 kbpsand 13.5 kbps By introducing IR and downlink diversity, throughput can be trebled forthe speech cell range from the original without any enhancements

10.1.6 Capacity Planning

Once link-level results are available, it is possible to estimate the air interface ance and perform network dimensioning based on a given network topology Thiswould typically involve the use of a planning tool, and this section considers some ofthe issues that would need to be considered when developing such a tool, or theassociated processes Both traffic and signalling capacity are considered

perform-10.1.6.1 Traffic

GPRS and EDGE traffic and GSM CS traffic use a common air interface resource Thechallenge in dimensioning a network for capacity is therefore in dividing this capacitywhile offering a satisfactory grade of service to both user types

Figure 10.11 illustrates one way that traffic resource within a cell (here TRX 2) can bedivided into CS, GPRS or EDGE territories TRX 1 carries all the signalling channels

In [6] it is stated that that the capacity available to GPRS and EDGE may be eitherfixed or follow the capacity on demand principle The two cases are now considered

Fixed GPRS Capacity

It is possible to assign fixed (or dedicated ) GPRS capacity, where one or more timeslotsare allocated on a permanent basis to GPRS These timeslots are always configuredfor GPRS and cannot be used by CS traffic This ensures that GPRS capacity isalways available in a cell The drawback with this approach is that, for a given cell

CCCHTRX 1

TRX 2

PCCCH TCH TCH TCH TCH TCH

TCH PDTCH PDTCH PDTCH PDTCH TCH

SDCCH

PDTCH PDTCH

Fixed (E)GPRS Dynamic (E)GPRS

Figure 10.11 Illustration of cell resource

configuration, blocking levels for CS traffic will increase since the number of availablechannels is reduced.

The decision on whether to assign fixed GPRS or EDGE territory is a tradeoffbetween providing a minimum level of GPRS and EDGE service and increasing theblocking for CS services This decision needs to take into account network operatorpriorities, network performance and predicted GPRS and EDGE usage levels

CS services The converse is also possible: that is, capacity that is by default allocatedfor CS use but allowed for EDGE and GPRS use when the packet traffic load is suchthat extra capacity would be desirable and the CS load is low Although at first thoughtthere may appear to be little difference between these two cases, there can be implica-tions for EDGE and GPRS, in terms of the air interface performance due to thealgorithms used to trigger the capacity switch and to EDGE and GPRS capacity atthe Packet Control Unit (PCU) in the BSC Some implementations use both types ofdynamic capacity simultaneously

Depending on the CS load level and the traffic profiles for CS and packet traffic,dynamic capacity will typically form a substantial part of the EDGE or GPRS capableresource, particularly while EDGE traffic levels are at low/medium levels Dedicatingsignificant resource to EDGE or GPRS is typically costly to network operators, andtherefore the use of dynamic capacity in the main is generally preferred initially Astraffic levels grow, however, the balance is likely to change with increasing use ofdedicated resources The trend of increasingly higher dedicated resources is clear incase of frequent multi-slot devices like EDGE capable laptop cards, etc

Available EDGE or GPRS Resources within the Circuit Switched Design

Since EDGE and GPRS typically make use of the resource not carrying CS traffic, thefirst step in packet capacity planning is determination of the CS load The design for CScapacity typically involves application of the Erlang B formula:

level, there is typically spare instantaneous capacity that can be utilised for packet datatransmission As long as the packet traffic can be temporarily interrupted toaccommodate peaks in CS traffic, then no degradation in CS services will result.Let us illustrate this with an example: a cell offering a CS load of 14 Erlangs with 21traffic channels will, on average, have 7.2 spare circuits These could carry packet data

on an on-demand basis, relinquishing the channels for CS traffic when required In thisway, the blocking probability of the CS facility is not degraded, even though the trafficchannels are subject to higher overall utilisation

Table 10.4 shows the mean number of timeslots (TCHs) available for EDGE andGPRS for different numbers of TRXs per cell and for CS blocking probabilities of 1%and 2% Due to the trunking efficiency for CS traffic, the capacity available to EDGEand GPRS does not increase linearly with increasing cell configurations, as can be seen.Having determined the capacity available to EDGE and GPRS, it is necessary to beable to dimension that capacity in order to provide a satisfactory user grade of service.Since it is generally accepted that packet data traffic is very different from CS traffic,being rather bursty in nature and typically represented by a succession of shorttransactions, the use of the Erlang B equation is not suitable The actual design musttake into account the traffic profile, in particular the applications being used

Estimation of the actual average maximum supportable cell throughput for EDGE

or GPRS is given by simply multiplying the average capacity available by the data rateper timeslot

Taking the previous example of 7.1 timeslots available, and assuming the use ofEDGE MCS-7 at a 10% BLER, the average available cell throughput is given by:

Throughput¼ EDGEtimeslots Meandatarate=TSL

¼ 7:1  44:8  ð1  0:1Þ ¼ 286:2 kbps ð10:1ÞThe dimensioning process is a way of producing, from the available cell throughput,the throughput that can be achieved while preserving a satisfactory user grade ofservice As such, network level simulations are required for the anticipated trafficprofile and these should form the basis of the planning and optimisation processesand tools

Table 10.4 Mean number of GPRS timeslots available

(TCH)/cell @ 1% @ 2% (E)GPRS available (E)GPRS available

blocking blocking TCH (CS load @ 1% TCH (CS load @ 2%

Figures 10.12 and 10.13 show example output from such simulations [2], giving netthroughput and LLC frame delay vs offered EGPRS load Simulations are based onthe link-level simulation results presented in Section 10.1.3 A network of 75 cells is thebasis for the work, and the performance includes the effects of LA and IR (see Sections10.1.3.3 and 10.1.3.4) Four reuse schemes were simulated: 1/3, 2/6, 3/9 and 4/12.Packet Switched (PS) downlink traffic was simulated, with MS multi-slot capability

of three timeslots

The key to the capacity-dimensioning task is in determination of the upper loadthreshold that will support a satisfactory user grade of service This threshold maybedetermined from network-level simulation results, such as the above, after designcriteria are defined This typically involves the required net throughput or maximumallowable mean LLC delay For example, using the graphs above, and a re-use of 4/12,

Figure 10.12 Net EGPRS throughput vs offered load (IRþ LA)

Figure 10.13 Logical link control frame delay vs offered EDGE load (IRþ LA)

the maximum offered load for the given network should not exceed 200 kbps/km2/MHz

if a mean net throughput of 50 kbps is to be achieved Conversion between the unitsused in the offered load results may be required – e.g., to cell offered load in kbps.The actual design target will depend heavily on the traffic mix and the sensitivity ofuser applications to delay and/or compromised throughputs In addition to theexponentially increasing packet delays at high load, some interaction with TCP(where used) may be anticipated and this will compound the issue [3] Themechanism to monitor network performance is clearly very important when trafficlevels are increasing, and the necessary mechanisms should be available to upgradenetwork capacity as and when required

Increasing EDGE and GPRS Capacity

As shown, a typical network will, at most times, have some limited available capacityfor EDGE and GPRS traffic To increase the available capacity, further resources will

be required in the form of traffic channels These can be obtained by dedicatingresources to EDGE and GPRS (at the expense of CS blocking), by upgrading thecell with additional TRX capacity, or by the use of network traffic managementfeatures Another way is to apply the Adaptive Multi Rate (AMR) codec withsubstantial half-rate usage for speech traffic The feature gains additional capacityfor speech due to the fact that half-rate usage accommodates two speech users in onetimeslot AMR requires AMR capable mobiles and, thus, the effect depends uponAMR mobile penetration Half-rate usage can only be applied in good radioconditions If one considers a spectrum limited scenario AMR mode brings substantialspectral capacity gain [22]

Dedicating EDGE or GPRS Resources

If a cell is found to be performing within the CS-blocking criterion, however, packetperformance is compromised, then dedication of resources to packet traffic could beconsidered The impact of such changes on the CS service should be evaluated, asblocking levels can change substantially Figure 10.14 shows the effect of dedicatingdifferent numbers of timeslots on blocking, for different cell configurations and a CSoffered load resulting in a 2% blocking level (in the absence of a dedicated EDGE orGPRS resource)

It can be seen, for example, that dedicating two timeslots in a four TRX cell results in

an increase in blocking from 2% to almost 4% for the same CS load The increase inblocking in smaller configurations is clearly even greater Therefore, it may be the casethat dedicating resources should be combined with traffic management features thatreduce end-user blocking

Additional TRX Capacity

By monitoring the grade of service offered to both the CS and packet services, it can bedetermined whether greater capacity is required in a cell If dedicating resources wouldresult in unsatisfactory CS performance then extra TCH resources may be required, bymeans of cell capacity expansion Assuming the CS load remains constant when a TRX

is added, the increase in capacity for packet traffic will be significant For example,based on Table 10.4, in a five TRX cell EDGE or GPRS available resources will

increase from 9.2 timeslots to 17.2 timeslots with the expansion to a six TRX cell, theadditional TRX offers an additional eight timeslots EDGE capacity expansion iscrucial in case of wide usage of multi-slot EDGE laptop cards The end user cannotuse these devices efficiently in case of capacity deficiencies.

Traffic Management

As an alternative, in some cases, to capacity expansion, traffic management featuresmay be used in a network to give capacity gain Typically, these offer a means todistribute CS traffic between cells where coverage overlaps exist The net effect ofthis is that peaks in the CS load are smoothed in each cell and therefore an increase

in average load can result

An example of such functionality is the BSC-controlled Traffic Reason Handover(TRHO) feature This is a load based algorithm that changes the power budgetthreshold for one or more outgoing adjacencies for a given source cell, depending onits load If the load exceeds a given level, and the load in a neighbouring cell is belowanother given level, then the power budget threshold for this adjacency is reduced, with,

in effect, the source cell shrinking in coverage area Load level discrepancies aretherefore balanced between the two cells; overall blocking levels can therefore bereduced and/or greater utilisation of the cell resource achieved (hence the term

‘capacity gain’)

The same principles can be used to attempt to improve packet resource availability,

by the use of appropriate thresholds

Figure 10.15 shows the TRHO concept with associated parameters In this example,

if the load level on the source cell exceeds 75% while that on the neighbour is less than50%, the normal power budget handover margin is changed from þ6 dB to 10 dB.Note that the received level from the target cell must exceed80 dBm for the handover

to be allowed

0 2 4 6 8 10

In this way, each cell attempts to retain the CS load within 75% of the cell TrafficChannel (TCH) resource, thus allowing 25% of resource to be available to packettraffic.

10.1.6.2 Signalling Capacity

The signalling associated with EDGE or GPRS data transfer can utilise either existingCommon Control Channel (CCCH) resources or be carried on the dedicated packetsignalling channel (PCCCH) if this channel is deployed

The deployment of the PCCCH, typically in a separate timeslot, allows the signallingfor the packet traffic to be removed from the CCCH This is clearly of benefit to anetwork operator whose CCCH (usually the downlink PCH/AGCH) capacity islimited A typical configuration for the logical signalling channels is shown in Figure10.16

The physical channel supporting the Packet Broadcast Control Channel (PBCCH)/PCCCH also supports the use of the Packet Data Traffic Channel (PDTCH) –i.e., traffic carrying capability The actual capacity offered by the PCCCH depends

on the configuration of this channel: it is possible to set a usage split by means of anumber of parameters and, in addition, where resource logical channels are allocateddynamically based on demand, the actual capacity will depend on the respectivepriorities The mapping of downlink logical channels is done according to thefollowing rules [20]:

PAGCH Resource assignment

Broadcast DL

UL DL

Figure 10.15 Illustration of BSC controlled traffic reason handover

The PBCCH is mapped onto the 52-multiframe The parameter BS_PBCCH_BLKSspecifies the number of radio blocks allocated to the PBCCH.

Radio blocks which are not available for paging, defined by the operator parameterBS_PAG_BLKS_RES, are allocated These blocks can be used for the PacketAccess Grant Channel (PAGCH), PDTCH and Packet Associate Control Channel(PACCH) and can therefore carry assignment messages, packet data and TBF-associated signalling

The remainder of the radio blocks in the 52-multiframe can be used for the PacketPaging Channel (PPCH), PAGCH, PDTCH and PACCH These blocks can carrypaging messages in addition to the messages specified in the previous point

The 52-multiframe divides into 12 blocks (each of four frames, and labelled B0–B11)assigned to the above logical channels and four idle frames

Figure 10.17 shows an example of downlink PBCCH/PCCCH mapping, whereBS_PBCCH_BLKS¼ 3 and BS_PAG_BLKS_RES ¼ 4

or PACCH

PBCCH PAGCH, PDTCH

or PACCH

PPCH, PAGCH, PDTCH

or PACCH

PBCCH PAGCH, PDTCH

or PACCH

PPCH, PAGCH, PDTCH

or PACCH

PAGCH, PDTCH

or PACCH

PPCH, PAGCH, PDTCH

or PACCH

PPCH, PAGCH, PDTCH

or PACCH B0 B1 B2 B3 B4 B5 B6 B7 B8 B9 B10 B11

Figure 10.17 An example of downlink PBCCH/PCCCH mapping onto the 52-multiframe

For the uplink case, the only signalling resource is the Physical Random AccessChannel (PRACH) Radio blocks can carry PDTCH traffic in addition to this By use

of the parameter BS_PRACH_BLKS, it is possible to define those blocks that can onlycarry PRACH The remaining blocks can carry PRACH or PDTCH, subject todemand and priority Figure 10.18 shows an example, where BS_PRACH_BLKS¼ 5

PRACH

(fixed) PRACH (fixed) PRACH (fixed) PRACH (fixed) PRACH (fixed)

B0 B1 B2 B3 B4 B5 B6 B7 B8 B9 B10 B11

Figure 10.18 An example of fixed PRACH mapping onto the 52-multiframe

The priority for each type of message on a given block determines the instantaneousconfiguration, and this is typically scheduled in the PCU

The downside to PCCCH deployment is the removal of one timeslot that can carry

CS traffic, so this must be taken into account However, the higher signalling capacityand packet specific idle mode cell reselection criteria have to be taken into account

If the PBCCH/PCCCH physical timeslot is not deployed then the existing CCCHsignalling channels are used These are Random Access Channel (RACH), AccessGrant Channel (AGCH) and Paging Channel (PCH) Typically the AGCH/PCHcommon resource is that which has the greatest capacity constraint, and thereforethe additional load generated by EDGE and GPRS should be taken into account

Signalling Traffic Estimation

In the initial deployments of GPRS it was found that packet data flow required ratherheavy usage of signalling channels This can be understood by consideration of Figure10.19, which illustrates a typical message flow associated with a TCP session

The figure represents the transfer of a small data volume (1.5 kB), and illustrates thelevel of handshaking performed A total of ten blocks of data are transferred, with atleast nine associated TBFs The use of the CCCH for the establishment of each TBFtherefore required extensive use of RACH and AGCH capacity Consequently anumber of change requests were submitted to ETSI with the aim of reducing the use

of the CCCH and, instead, using the PACCH for TBF establishment where possible,based on the fact that the time between subsequent data flows was typically small (up toseveral hundred milliseconds) In addition to the reduction in required signallingcapacity, the changes also led to significantly reduced data transfer times

Initial observations based on specific TBF establishment and release procedures haveshown a significant reduction in usage of the CCCH, by a factor of at least 75%, for atransfer such as the above Because the actual capacity requirement is so criticallydependent on radio resource algorithms, which are somewhat vendor-specific (subject

to the constraints of the standards) and may change in the short term, it is notconsidered appropriate to give absolute values that may be anticipated Rather, it isassumed that monitoring of signalling resource usage will be performed on a networkand on capacity dimensioning, based on the results of this together with the data trafficgrowth trend

The actual capacity in terms of the absolute number of signalling messages conveyed

is unlikely to differ significantly between the use of the CCCH and PCCCH, andtherefore the load seen on the CCCH can be used as the basis for the load to beexpected on the PCCCH, if this is to be deployed at a later stage

Data ACK Data ACK Data Control

Control

Figure 10.19 TCP message flow

Table 10. 4 shows the mean number of timeslots (TCHs) available for EDGE andGPRS for different numbers of TRXs per cell and for CS blocking probabilities of 1 %and 2% Due to the trunking efficiency for. .. level simulations are required for the anticipated trafficprofile and these should form the basis of the planning and optimisation processesand tools

Table 10. 4 Mean number of GPRS timeslots