You are on page 1of 31

Influence of GPRS on capacity and coverage

planning

Michal Szczesniak,
Polkomtel S.A., RF HQ
michal.szczesniak@polkomtel.com.pl.
Outline

 overview of GPRS radio interface


 GPRS radio network parameters (Nokia case)
 simulation model
 QoS descriptors and objectives
 simulation results
 coverage vs. capacity issues
 conclusions

GPRS Conference, May 2000,


Cannes
Overview of GPRS radio access, phase 1
GPRS phase 1 features:
 multislot connections - typically 3DL/1UL, max.6 TSL/direction
supported by network
 independent uplink/downlink allocations
 dynamic channel (timeslot) sharing - max. 9 DL/7 UL simultaneous
connections on one TSL.
 dynamic channel allocation as the main medium access mode -
access determined by USF flag sent in previous block (except for
uplink PACCH blocks where RRBP flag is used)
 only coding schemes CS1 & CS2 are supported
 no GPRS dedicated common signalling channels

GPRS Conference, May 2000,


Cannes
Radio interface protocol layers
Network layer

SNDCP layer (header 3-4 bytes)

LLC layer (header 2-4b., FCS=3b.)

RLC/MAC layer(header 3-4b)

payload Convolutional coding

header
Physical layer
FCS or BCS
User data frame overhead: 5-8 b for LLC&SNDCP + 3-4 b for each RLC block
tail GPRS Conference, May 2000,
Cannes
Radio interface capacity
LLC user
RLC raw user max. data
coding block data rate rate
scheme length [kbps] [kbps]

CS1 181 9.05 7.85

CS2 268 13.4 12.20

CS3 312 15.6 14.40

CS4 428 21.4 20.20

 The LLC throughput is further decreased through higher layers


overheads (depending on the user packet length)
GPRS Conference, May 2000,
Cannes
TBFs 0 1 2 3 4 5 6 7

7
BCCH
6

Speech call
5

4 TBF1
3 TBF2
2
TBF3
1

 TBF (Temporary Block Flow) is released if the are no more packets to be sent e.g. MS is not using its uplink TSL allocations or NW has nothing to send
 establishment of the TBF requires immediate assignment procedure
 load of TBFs per TSL is balanced across TSLs

GPRS Conference, May 2000,


Cannes
Channel allocation methods

TRX1 GPRS enabled

TRX2 GPRS disabled

BCCH

TCH for CS calls (CS only)


Additional GPRS capacity (CS by default, GPRS possible)
Default GPRS Capacity (GPRS by default, CS when needed)

Dedicated GPRS Capacity (GPRS only)

• more intracell handovers due to territory changes!

GPRS Conference, May 2000,


Cannes
GPRS Radio Network Parameters
 X1 - number of TBFs/TSL triggering the upgrade procedure, e.g. X1=1.5
 X2 - requested number of TBFs/TSL. e.g. X2=1
 X3 - number of TBFs/TSL triggering the downgrade procedure, e.g. X3=0.5
 dedicated and default GPRS capacity
 territory upgrade guard time, e.g. 5s. Downgrade is done when needed
 preferredBCCHfreqGPRS, recommended YES
 ch = max power control change with respect to +39dBm e.g. 33dB,  = slope
e.g. 0.8 but closed loop control preferred (e.g.  = 0 )
 Tavg_w (idle mode), Tavg_t(active mode), Navg_i(interference), UPDATE_C,
PC_MEAS_CHAN
 link adaptation thresholds and risk levels

NOTE: In Nokia system parameters X1, X2 and X3 are preset in the network and
can not be changed by the user

GPRS Conference, May 2000,


Cannes
Simulation model
 GPRS simulator has been developed in-house featuring:
• hundreds of GPRS and normal users simulated in one cell during
the period of several hours
• ARQ selective repetition algorithm implemented
• modelling of different traffic source types (poisson, bursty etc.)
• dynamic channel allocation strategies based on Nokia
descriptions
• calculation of important performance indicators like user
throughput, 95-percentile blocking etc.
 discrete simulation is the only practical way of observing
GPRS performance and network behaviour
GPRS Conference, May 2000,
Cannes
QoS objectives
 mean or 95-percentile delay below threshold as defined in e.g.
ETSI GSM 2.60 or
 mean user throughput are the main KPIs
 network throughput is not a good measure as it rises monotonically
with the offered traffic increase ( provided the system is stable)
which is in contradiction to quality perception
 reliability level (BER, RBER, BLER) should be dependent on
network level & application level protocols - high reliability should be
assumed at the start
 GPRS blocking should be no problem provided that upper layers
protocols can cope with it (implementation dependent) - blocking has
effect on delay which is very difficult to estimate!
GPRS Conference, May 2000,
Cannes
QoS objectives continued
95- Mean user Mean user
Mean delay percentile throughput Mean delay throughput
objective [s] delay for [bps] for objective [s] [bps] for
for packets packets < packets for packets packets
<128 bytes 128 bytes <128 bytes <1024 bytes <1024 bytes

Delay class 1 0.5 1.5 2048 2 4096

Delay class 2 5 25 205 15 546

Delay class 3 50 250 20 75 109


 Based on ETSI GSM 2.60
 classes 2, 3 and best effort are likely to be disappointing for customers and should be avoided through call admission
control mechanisms ( vendor dependent )

GPRS Conference, May 2000,


Cannes
Source characterisation
 GPRS traffic sources are not poissonian but bursty in nature
 fixed network most common packet size is between 256 and 512 bytes -
mobile networks are expected to have shorter packets due to specialised
protocols (e.g. WAP)
 packet length distribution is heavy-tailed (very long packets occur with
nonnegligible probability) -> traffic measurements will have high variation
 source that is forced to release a TBF performs an immediate retry (this in
not inline with traditional traffic models)
 source must not send a new frame before the previous one has been
transmitted ( additional packet delay in the source )
 in simulation: on average 256 bytes packets, 10 packets/burst, 600ms
interval between packets in a burst

GPRS Conference, May 2000,


Cannes
Sources of delay in GPRS
 One MS can be sending only one LLC frame at the time - all
subsequent frames are buffered in the source
 16 kbit/s A-bis I/F incurs 20ms delay for each CS1/CS2 packet
(PCU frame)
 frame retransmission, especially in multislot transmissions.
Frame retransmission can be due to transmit errors, receive
errors or sender stall condition
 internal source/destination delays (implementation dependent)
 dynamic channel sharing
 radio I/F frame structure is not synchronised with data sources

GPRS Conference, May 2000,


Cannes
LLC vs packet throughput
Let t(i) be the i-th packet RLC transfer time, w(i) the i-th packet total time in
the system ( waiting in the LLC queue + RLC transfer time) and l(i) be the i-
th packet length.

average _ LLC _ throughput   l (i ) /  t (i )


i i

average _ packet _ throughput   l (i ) /  w(i )


i i

GPRS Conference, May 2000,


Cannes
Influence of LLC layer queuing
Conditions: cells are nominally loaded with CS traffic at 2% GOS, dynamic channel sharing, CS-1,
3 TSL connections

2 TRX cell 4 TRX cell

 LLC throughput is 1-3 times higher than the packet throughput


 in most cases user throughput is equivalent to LLC throughput

GPRS Conference, May 2000,


Cannes
Ack/Nack sending rate (polling interval)
 GPRS uses the selective ARQ procedure
 Ack/Nack sending rate should be coupled with current BLER, multislot
allocation and CS in order to maximise throughput (bearing in mind
that the max. nr. of outstanding blocks is 64)
 Ack/Nack sending rate on downlink should be a trade-off between
downlink signalling load and uplink delay
 Ack/Nack sending rate on uplink should be fast enough to make the
downlink transmission fluent
 double retransmission or double acknowledgement can be used in
high error situation to minimise the delays
 Note that RTD for acknowledgement is at least 100ms (comparable to
sending time of 5 RLC blocks)

GPRS Conference, May 2000,


Cannes
Ack/Nack sending rate cont.
Conditions: 30 additional GPRS TSLs, bursty traffic of 3.2 Erl, nominal CS load = 21.9 Erl, BLER=0%, CS-1

1 TSL connections, delay 3 TSL connections


increases by almost 2 seconds more blocking due to no channel
with longer ACK intervals reorganisation at new call attempt,
more delay increase than in 1 TSL
case due to frequent restransmissions
GPRS Conference, May 2000,
Cannes
How much can we load a GSM cell with GPRS
traffic?
 Conditions: 14 voice/packet TSLs, constant voice traffic = 8.2 Erl, dynamic
TSL sharing, max. 3TSL connections, 256 mean packet length
 Objective: delay according to GSM 2.60, class 1 (95% short packets with
delay < 3s), no voice quality degradation

Only 0.8 Erl


capacity
with delay class 1
(1 GPRS Erl = 9050
bits/s in RLC layer)
GPRS Conference, May 2000,
Cannes
How much can we load a GSM cell with GPRS
traffic? - cont.
 Conditions: 30 voice/packet TSLs, constant voice traffic = 21.9 Erl, dynamic TSL
sharing, max. 3TSL connections

1.3 Erl capacity


with delay class 1
 Conclusion: Capacity for nominally loaded 4 TRX cell is about 50% higher than for nominally loaded 2 TRX
cell because of differences in mean available GPRS capacity = TCH_nr - nominal_CS_load -free_channels

GPRS Conference, May 2000,


Cannes
How much can we load a GSM cell with GPRS
traffic? - cont.

2 dedicated no dedicated

 Cell with 2 GPRS dedicated channels has less throughput with higher traffic because more packets
are accepted than in no dedicated channels case ( no call admission control function is apparent)

GPRS Conference, May 2000,


Cannes
How much can we load a GSM cell with GPRS
traffic? - bursty vs poisson traffic

bursty poisson

 System has slightly better throughput for poisson traffic and much smaller delay
 usage of CS traffic smoothing features can improve GPRS throughput

GPRS Conference, May 2000,


Cannes
Smoothing of CS traffic
GPRS territory

CS traffic envelope

time time
 CS traffic is smoothed using traffic reason handovers to surrounding cells ( e.g. Nokia AMH feature )
 GPRS territory has more constant capacity -> less GPRS signalling, less downgrades, less delay variation, acts like a
dedicated resource

GPRS Conference, May 2000,


Cannes
Performance of GPRS on 8 dedicated channels
Conditions: 100 GPRS users, 256-bytes packets, CS-1, BLER=0%

1 TSL connections

3 TSL connections

• Note that the throughput of 3-TSL connection is less than 3


times better than of 1-TSL connections due to processing delays
and sender stall condition
GPRS Conference, May 2000,
Cannes
Mobility management in GPRS
 No handovers - cell reselections only based on C1, C31 and C32
 3 possible states : GMM-IDLE, GMM-STANDBY and GMM-READY.
 MS has to perform Cell Updates when in Ready State. Paging in
ready state is done only in 1 cell. Mobile leaves the Ready state if
there are no packets to transmit within 32s ( default ).
 Location area subdivided into Routing Areas (RA). MS has to perform
RA Updates
 Periodic Location Timer T3312
 Network Controlled cell reselection is optional and not supported in
phase 1

GPRS Conference, May 2000,


Cannes
Performance of signalling channels
 GPRS phase 1 reuses standard GSM common signalling channels (BCCH,
RACH, PCH, AGCH)
 GPRS packets are shorter than speech -> we have more attempts per
generated Erlang (15-30 times depending on the source type) -> more
signalling
 30 Erl of speech traffic uses around 5% of usable RACH capacity. 1 Erl of
packet traffic uses also 5%!
 In 50000 subs LA around 20% of PCH/AGCH capacity is used. The same 20%
of PCH/AGCH can be consumed by paging for 5000 GPRS users assuming 10
packet transmissions per hour. AGCH load for 1000 users in a cell consume
also around 20% of PCH/AGCH usable capacity (with 9 slots per multiframe)
 1 Erl of GPRS traffic may generate 0.2-0.3 Erl on 16 kbps A-bis signalling
channel -> more TRXSIG channels needed

GPRS Conference, May 2000,


Cannes
Performance of signalling channels -
RACH calculations
GSM typical:
RACHs/51 multiframe 51.00
RACH capacity 780 065.01 msg/h
Max. RACH usage 0.20
Max. RACH throughput ( not couting retransmissions
156 013.00) usable channel bandwidth [msg/h]
Ave call length 50.00 s
Traffic/user 0.02 Erl
MCR 1.20 multiple channel request ratio
LU/user/h 2.00 location update rate
Traffic 10.00 20.00 30.00 40.00 50.00 Erl
Users 666.67 1 333.33 2 000.00 2 666.67 3 333.33
CSA 2 464.00 4 928.00 7 392.00 9 856.00 12 320.00
RACH usage by speech [%] 1.58 3.16 4.74 6.32 7.90

GPRS estimated:
Most frequent packet length 256.00 bytes
Channel troughput ( user rate ) 5 000.00 bit/s
GPRS traffic (DL+UL) 1.00 2.00 5.00 10.00 Erl
GPRS users in the cell 22.00 45.00 112.00 225.00
GPRS attempts (PSA) 8 789.06 17 578.13 43 945.31 87 890.63
RACH usage by GPRS [%] 5.63 11.27 28.17 56.34

PS/CS ratio[%] 35.67


PSA/CSA ratio[%] 35.67
GPRS Conference, May 2000,
Cannes
GSM typical:
PCHs/51 multiframe 9 3
Number of multiframes between paging 6
PCH capacity per paging group 2549.232 msg/h
Max. PCH utilisation with acceptable delay
in one paging group 0.5
Max. PCH throughput per group 1274.616 msg/h
Average number of pages per message 2
Total PCH/AGCH throughput 68829.27 msg/h
Ave call length 50 s
Traffic/user 0.015 Erl 750
Numer of users per paging area 50000
Number of cells per paging area
MT/all ratio
37.5
0.5
@20Erl/cell traffic
Performance of signalling
MT CSA
PCH usage by speech [%]
27000
19.61
channels - PCH/AGCH
AGCH usage by speech [%] 2.09
calculations
GPRS estimated:
Most frequent packet length 256 bytes
Number of GMM-READY states (sessions)
per hour 3 10
Channel troughput ( user rate ) 5000 bit/s
Traffic volume/user 800000 bit/h 0.044444 Erl
Number of users per routing area 100 500 1000 100 500 1000
Total GPRS traffic in routing area 4.44 22.22 44.44 4.44 22.22 44.44
Users 100 500 1000 100 500 1000
MT/all calls ratio 0.5 0.5
GPRS paging messages 150 750 1500 500 2500 5000
PCH usage by GPRS [%] 0.11 0.54 1.09 0.36 1.82 3.63
Number of high traffic GPRS cells per
routing area 37.50 37.50
AGCH messages 1041.67 5208.33 10416.67 1041.67 5208.33 10416.67
AGCH usage by GPRS [%] 1.51 7.57 15.13 1.51 7.57 15.13
Actual usage of PCH/AGCH channel [%] 1.73 8.66 17.31 2.24 11.20 22.40
GPRS Conference, May 2000,
Cannes
Spectrum efficiency
 Spectrum efficiency is defined as:

spectrum _ efficiency 
partial _ reuse _ factor
 coding _ scheme _ throughput
reuse _ factor
 Best available coding scheme is automatically selected by the system based on RXQUAL measurements -> coding scheme throughput is the throughput of the
best coding scheme under a given C/I determined by the reuse factor
 optimum BLER (block error rate) is <10% (double retransmission < 1%)

GPRS Conference, May 2000,


Cannes
Spectrum efficiency vs. frequency reuse

K[reuse factor] 1 3 9 12 15
C/I[>90%] 7 8 9 12 15
PRF[partial reuse] 0.14 0.38 1.00 1.00 1.00
EFF/TRX 1.71 1.50 1.33 1.00 0.80
CS1[kbit/s] 7 8 8.5 8.75 9
CS2[kbit/s] 7 9 10 12.5 13
CS3[kbit/s] 6 7 8.5 13 15
CS4[kbit/s] 2.5 3 4 9 13
BESTCS[kbit/s] 7 9 10 13 15
GPRSEFF 12.0 13.5 13.3 13.0 12.0
Calculations made for 12 available channels
GPRS Conference, May 2000,
Cannes
Main concerns
 Restrictions in mobility management -> applies particularly to
networks where quality driven handovers are in place ->
solution - put GPRS into clean BCCH layer because IUO
mechanisms don’t work
 no quality management - all GPRS calls have equal chances
in the channel reservation algorithm - it is very easy to kill
the signalling with some packet sending schemes unless the
billing scheme restricts that
 unknown terminal behaviour with regard to upper layer
management - assumptions for one service may not be valid
for another
GPRS Conference, May 2000,
Cannes
Conclusions
 existing capacity reserves resulting from non-ideal trunking
efficiency in cells bring little capacity for GPRS purposes
( less than advertised by vendors ) unless some CS traffic
smoothing is implemented
 coverage should be no concern with CS1
 restrictions in mobility management mean that high quality
frequency plans have to be used for GPRS traffic purposes
 quality of service is not observed which means that in
conjunction with high traffic variations no committed
information rate can be guaranteed to end users

GPRS Conference, May 2000,


Cannes

You might also like