Anda di halaman 1dari 15

Data Throughput in EDGE Networks

Data Throughput in EDGE Networks

Alcatel File Reference Date Edition Page


347232356.doc 3DC 21150 0285 TQZZA 01/2003 03 1
All rights reserved. Passing on and copying of this document, use and communication of its contents not permitted without written authorisation.
Data Throughput in EDGE Networks

CONTENTS

1. SCOPE AND APPLICATION......................................................................................


2. SOME DEFINITIONS..................................................................................................
2.1 Where do we define the Data Throughput ?...............................................................................
2.2 Different Definitions for Data Throughput in the Network Context..............................................
2.3 Data Throughput for Static versus Dynamic Situation..................................................................

3. NETWORK COVERAGE AND THROUGHPUT........................................................


3.1 Some Basics......................................................................................................................................
3.2 GPRS and EGPRS Performance......................................................................................................
3.2.1 Introduction..........................................................................................................................................
3.2.2 Simulation Results...............................................................................................................................
3.2.3 Some Remarks to Spectrum Efficiency...............................................................................................
3.2.4 Influence of Frequency Hopping on Data Throughput.........................................................................
3.2.5 Influence of Real Link Adaptation on Data Throughput.......................................................................
3.3 Link Budgets for GPRS/EGPRS ?..................................................................................................
3.4 How to do a Correct Data Throughput Estimation ?...................................................................
3.4.1 Two Possibilities for Data Throughput Planning.................................................................................
3.4.2 Radio Network Planning Process for GPRS/EGPRS.........................................................................

4. SOME EXEMPLARY RESULTS FOR MACROCELLS............................................


4.1 Dense Urban Scenario....................................................................................................................
4.2 Suburban (Residential) Scenario...................................................................................................
4.3 Conclusion.......................................................................................................................................

5. GENERALIZATION TOWARDS COMPLEX NETWORK STRUCTURES..............

Notice of proprietary information


This document contains proprietary technical information belonging to Alcatel. By accepting this
material, the recipient agrees that this material will not be reproduced or used in whole or part
except as otherwise agreed between Alcatel and the recipient.

Alcatel File Reference Date Edition Page


347232356.doc 3DC 21150 0285 TQZZA 01/2003 03 2
All rights reserved. Passing on and copying of this document, use and communication of its contents not permitted without written authorisation.
Data Throughput in EDGE Networks

1. SCOPE AND APPLICATION


This document provides explanations around the issue of data throughput definitions and results for
EDGE networks.

When talking about EDGE, we always refer to the application of EGPRS (packet data). The circuit-
switched application for EDGE (EDGE Circuit Switched Data, ECSD) is so far not requested from
the market and therefore not covered in this document.

As an introduction to the general subject, it is recommended that the reader goes through the
document Introduction to GPRS and EGPRS available from Alcatel (document number 3DC
21084 0003 TQZZA), before reading this document.

2. SOME DEFINITIONS

2.1 Where do we define the Data Throughput ?

When talking about data throughput in GPRS and EGPRS, we refer to the rate at the RLC/MAC
layer. This is not exactly, what the end users see. However, the relation between the data rate at the
RLC/MAC layer and the end user application is very much dependent on the application (protocol
type, overhead from header, header compression, operating system etc.). This application is outside
of the influence of the mobile network supplier. When comparing the mobile infrastructure, the basis
can only be the RLC/MAC throughput.

2.2 Different Definitions for Data Throughput in the Network Context

Usually, the data throughput specified refers to one PDCH. In GPRS and EGPRS, several PDCH
per user can be combined to get a higher data throughput.

Due to the asymmetric traffic expected for packet data services (much more downlink traffic than
uplink traffic), GPRS and EGPRS networks are planned commonly for the downlink only. The
throughput is therefore specified for the downlink.

There are different ways to define the data throughput in a network:

Mean Data Throughput The area of the network is divided into equally sized pixels. The data
throughput is calculated pixel per pixel considering level (pathloss to
serving cell) and interference. The pixels close to the base station will
have a high data throughput; the pixels at the cell border provide only
a low throughput. The mean data throughput is the average of all
pixels within one cell.

This is the most common definition for data throughput.

x% Data Throughput Again, the data throughput is calculated pixel per pixel. The defined

Alcatel File Reference Date Edition Page


347232356.doc 3DC 21150 0285 TQZZA 01/2003 03 3
All rights reserved. Passing on and copying of this document, use and communication of its contents not permitted without written authorisation.
Data Throughput in EDGE Networks

data throughput is the minimum value reached by at least x % of all


the pixels. Common values for x are 50%, 90% and 95%.

Example: 90% Data throughput of 20 kbit/s. 90% of the cell area


provides a data throughput of at least 20 kbit/s, 10% of the cell area
are below that value.

Guaranteed Data Throughput A guaranteed throughput in the pure sense does not exist. It is similar
to voice: A coverage probability of 100.0% is not possible.

When talking about a guaranteed data throughput, in fact a minimum


throughput with a certain area probability is meant. The definition is
usually the same as above for x% Data Throughput.

The diagram below visualizes the difference between the mean data throughput and, as example,
the 90% data throughput:
kbit/s Assumed cell border
60

Data throughput as function of


50 the distance mobile to BTS

Mean 40
data
through-
put 30
90%
through- 20
put

10

0
0 200 400 600 800 1000 1200 1400 1600
Distance BTS - Mobile [m]

Why is the Mean Data Throughput the most common definition ?

Because this definition describes best the performance of the complete system. GPRS/EGPRS is
not, like voice, limited to one timeslot. Depending on the mobile capability, several timeslots (up to
eight in theory, up to five with real mobiles) can be combined to provide a higher throughput.

A mobile close to the base station receives a higher data throughput. It therefore occupies less
timeslots. The mobile farther away from the base station receives a lower data throughput, but can
get more timeslots to compensate for that.

Alcatel File Reference Date Edition Page


347232356.doc 3DC 21150 0285 TQZZA 01/2003 03 4
All rights reserved. Passing on and copying of this document, use and communication of its contents not permitted without written authorisation.
Data Throughput in EDGE Networks

Or in other words, even a mobile at the cell border benefits from the higher throughput: The mobiles
close to the BTS occupy less time-slots for a given data throughput or completing the data transfer
earlier. The mobile at the cell border (assuming mobiles with multi-slot capability) can compensate
its lower throughput per timeslot by combining timeslots not occupied by the mobile close to the
BTS. This is an important idea on optimizing the spectrum usage and often ignored !

The mean data throughput is therefore the appropriate value to design the network.

2.3 Data Throughput for Static versus Dynamic Situation

The data throughput is generally stated for a static network scenario: The mobile stays at the same
location throughout the data transmission.

In case the mobile would move and cross cell-borders, there would be cell re-selections during the
data transmission. During the cell re-selection, there is a short interruption of the data flow leading
to a lower average throughput. As information on the mobility is hardly available and the modeling
would be quite complex, only the static throughput is used ignoring potential interruptions.

3. NETWORK COVERAGE AND THROUGHPUT

3.1 Some Basics

From the design of GSM networks for voice, we are used to cell ranges, cell areas respectively, in
which, with a certain probability, coverage is provided. Considering a certain, market-defined
coverage class and coverage probability, a cell range or a cell area could be defined, where mobile
telephony service can be offered.

The major tool for calculating the service area is the link budget. Based on the equipment
parameters like transmit power, receiver sensitivity, consideration of losses and gains, a path-loss
can be calculated which, using empirical models like Hata-Okumura, COST 231 etc. results in the
cell range. In fact, also for the classical voice services, interference has to be considered. To
establish a link budget, the addition of a fixed interference degradation margin of 3 dB (1.5 to 2 dB
in case of frequency hopping respectively) was sufficient.

Alcatel File Reference Date Edition Page


347232356.doc 3DC 21150 0285 TQZZA 01/2003 03 5
All rights reserved. Passing on and copying of this document, use and communication of its contents not permitted without written authorisation.
Data Throughput in EDGE Networks

With GPRS and EGPRS, this is different: Service can be offered everywhere within the cell range
as for voice, but due to the link adaptation, the throughput differs significantly from the center of the
cell towards the edge of the cell. The comparison between voice and packet data:

Voice service Packet data service (GPRS/EGPRS)

Service available, or service not available Service available, service quality (data
(based on a certain, specified probability) throughput) dependent on the location

Service Service
Quality Quality

Distance from base station Distance from base station

The depiction is very much simplified

The conclusion: There is no fixed cell range for GPRS/EGPRS. The coverage area depends
directly on the requested data throughput. If the requested data throughput is low, the cell range for
GPRS/EGPRS can be much larger than for voice. If the requested data throughput is high, the cell
range can be lower as for voice.

3.2 GPRS and EGPRS Performance

3.2.1 Introduction

The performance in terms of data throughput per PDCH depends on several parameters:

The carrier-to-interference and noise ratio C/(I+N)

- Carrier -> Level (fieldstrength) from the serving cell at the mobile location

- Interference -> Co- and adjacent channel interference from non-serving cells

- Noise -> assumed to be mainly thermal noise

Use of frequency hopping

Link adaptation

Use of Incremental Redundancy

Some examples for data throughput in macro-cellular networks under different environments are
given in the next chapter.

Alcatel File Reference Date Edition Page


347232356.doc 3DC 21150 0285 TQZZA 01/2003 03 6
All rights reserved. Passing on and copying of this document, use and communication of its contents not permitted without written authorisation.
Data Throughput in EDGE Networks

3.2.2 Simulation Results

The following diagrams show exemplary simulation results for the data throughput in purely macro-
cellular networks dependent of the distance between the mobile and the base station for EGPRS.
The results are shown for four different frequency reuse cluster sizes (1x3, 3x3, 4x3 and 7x3) and a
fixed inter-site distance. A homogenous morpho-structure over the full cell area is assumed.

Dense urban, GSM 900, with incremental redundancy, no frequency hopping, three-sectored

Fading profile TU3


60
Indoor loss 17 dB
Mean Throughput [kbit/s]

50
Body loss 3 dB
40 7x3 BTS antenna height 20 m
4x3
30 Antenna beamwidth 65
3x3
20 1x3 EIRP GMSK 57.7 dBm

EIRP 8PSK 52.9 dBm


10

0
0 500 1000 1500
Distance BTS - Mobile [m]

Diagram 1: EGPRS, mean throughput over distance, GSM 900, dense urban

Rural, GSM 900, with incremental redundancy, no frequency hopping, three-sectored

Fading profile RA100


60
In-car loss 8 dB
Mean Throughput [kbit/s]

50
Body loss 3 dB
40 7x3
BTS antenna height 35 m
4x3
30
3x3 Antenna beamwidth 90
20 1x3
EIRP GMSK 59.3 dBm

10 EIRP 8PSK 54.5 dBm

0
0 2000 4000 6000 8000 10000
Distance BTS - Mobile [m]

Diagram 2: EGPRS, mean throughput over distance, GSM 900, rural

Alcatel File Reference Date Edition Page


347232356.doc 3DC 21150 0285 TQZZA 01/2003 03 7
All rights reserved. Passing on and copying of this document, use and communication of its contents not permitted without written authorisation.
Data Throughput in EDGE Networks

We can clearly see the influence of the interference: The more intensively we are using the
spectrum, the lower the data rates per timeslot get. The difference between a 3x3 and a 4x3 reuse
is rather small, while a 1x3 reuse leads to a severe degradation. Going to a 7x3 reuse increases the
data throughput per timeslot within a certain cell. As the frequencies are less utilized, the total
spectrum efficiency is reduced.

Remark: Important for the interpretation of the above diagrams is to keep in mind that they are
done as function of the distance mobile to BTS for a fixed cell range/site distance. Close
to the base station, the C/I and therefore the data throughput is high (field strength from
serving base station high, distance to neighbor cells high -> interference low). Further
away from the base station, the C/I is low and therefore the data throughput is lower (field
strength from serving base station low, distance to neighbor cells lower -> interference
higher).

An alternative representation would be to show the data throughput as function of the cell
range (mobile would always be located at the cell border for varying cell ranges/site
distances). The resulting diagram would look different. As the C/I is more or less constant
for different cell ranges (reducing the cell range improves the field strength from serving
base station, but also increases the interference level from neighboring base stations), the
resulting graph would show a more constant bit rate for different cell ranges.

You can find more details on how these simulations were done in the document "Introducing 3 rd
Generation Mobile Communication Services on Evolved GSM Platforms" by Dr.-Ing. Michael
Tangemann, Dr.-Ing. Andreas Weber and Dr.-Ing. Dirk Nikolai.

3.2.3 Some Remarks to Spectrum Efficiency

GPRS and EGPRS will provide a much better spectrum efficiency (in kbps/MHz) compared to
voice. The reason is that the design of a voice network is done to ensure a minimum quality level in
the cell. This leads to an "excessive" good signal-to-noise-and-interference-ratio SINR in most of
the cell area. An example for the probability function is given below:

0.18
0.16
0.14
0.12
Probability

0.10
0.08
0.06
0.04
0.02
0.00
-4 0 4 8 12 16 20 24 28 32 36 40
SINR / dB

Diagram 3: Probability for SINR

Alcatel File Reference Date Edition Page


347232356.doc 3DC 21150 0285 TQZZA 01/2003 03 8
All rights reserved. Passing on and copying of this document, use and communication of its contents not permitted without written authorisation.
Data Throughput in EDGE Networks

A SINR-value of around 17 dB has the highest probability, while 9 dB would be sufficient for voice.

This can be utilized by GPRS and EGPRS by transporting more information bits per second
wherever possible: The total spectrum usage per cell is optimized.

As the throughput of GPRS is reaching its saturation point (the point, where an improvement in C/I
does not result in increased data throughput anymore) earlier than EGPRS, the spectrum efficiency
of EGPRS is better. This is visualized in the curves below:

60

50
Throughput [kbit/s]

40
EGPRS
30
GPRS
20

10

0
-10.0 0.0 10.0 20.0 30.0 40.0 50.0
C/I [dB]

Diagram 4: Throughput of GPRS and EGPRS as function of C/I

3.2.4 Influence of Frequency Hopping on Data Throughput

As we have seen above, the data throughput for GPRS and EGPRS depends to a large extend on
the interference experienced within the coverage area. For circuit-switched speech-traffic,
frequency hopping was proven to be a good feature to combat the effects of co-channel and
adjacent channel interference. Frequency hopping is especially useful for slow-moving subscribers.
It is now interesting to know, whether the same positive effect of frequency hopping can be utilized
for GPRS and EGPRS.

The simulations show different effects depending on the modulation and coding scheme and the
fading situation. The conclusion of the simulations:

GPRS step 1 (CS-1 and CS-2): Frequency hopping is recommended

GPRS step 2 (CS-1 to CS-4): Frequency hopping is not recommended

EGPRS (MCS-1 to MCS-9): Frequency hopping is not recommended

3.2.5 Influence of Real Link Adaptation on Data Throughput

The network will adapt the data throughput dynamically. The target is the optimum selection of the
modulation and coding scheme for a given radio condition. The radio condition is defined by the
received level (RXLEV) and the received quality (RXQUAL). For the downlink, this is measured by
the mobile and transmitted in signaling messages to the BSS.

Alcatel File Reference Date Edition Page


347232356.doc 3DC 21150 0285 TQZZA 01/2003 03 9
All rights reserved. Passing on and copying of this document, use and communication of its contents not permitted without written authorisation.
Data Throughput in EDGE Networks

Most simulation results are done for ideal Link Adaptation, i.e. always the most appropriate
modulation and coding scheme is used.

In reality, the results are below this optimum value. This is mainly due to the delay between the
measurements of RXLEV and RXQUAL and the execution of the modulation/coding scheme
change. The delay comes mainly from necessary averaging of the values and the time implied by
the protocol. The real performance depends on lots of parameters including fading profile (mobile
movements), packet length, BSS parameter settings and the link adaptation algorithm.

The data throughput loss under real conditions can be in a range of 7% to around 20%.

3.3 Link Budgets for GPRS/EGPRS ?

Designers used to plan GSM networks for voice are sometimes asking for link budgets for GPRS
and EGPRS. The expected schemes are concentric rings around the base station giving the cell
range for each modulation and coding scheme MCS. The MCS found at the cell border is then
taken as equal to the minimum data throughput.

Now, the explanations given above make something very clear: Link budgets for GPRS / EGPRS
do not make sense. They neither consider the influence of interference, nor the properties of the link
adaptation algorithms, incremental redundancy, frequency hopping etc.

The results from link budgets for GPRS/EGPRS is misleading and should never be used to
estimate the expected data performance.

Below, the results (mean data throughput for EGPRS) of a pure level based approach are
compared to a correct design considering both level and interference (C/I):

Pure level based approach C/I approach


Kbit/s

As we can see, the pure level approach is much too optimistic.

Alcatel File Reference Date Edition Page


347232356.doc 3DC 21150 0285 TQZZA 01/2003 03 10
All rights reserved. Passing on and copying of this document, use and communication of its contents not permitted without written authorisation.
Data Throughput in EDGE Networks

3.4 How to do a Correct Data Throughput Estimation ?

3.4.1 Two Possibilities for Data Throughput Planning

Within Alcatel, there are two possibilities for data throughput planning:

Cell range versus data throughput based on a simulation tool chain. The result is either the data
throughput per PDCH for a given cell range, or the cell range for a given data throughput. This
process complements the link budget used for voice. This approach assumes a homogenous morpho-
structure on flat terrain within the cell area.

Radio Network Planning, providing result plots with a prediction of data throughput per timeslot at
given locations (i.e. pixels on a map). With this approach, real morpho- and topographical
distributions can be considered. This process is explained more in detail in the next sub-chapter:

3.4.2 Radio Network Planning Process for GPRS/EGPRS

The basis for GPRS/EGPRS planning is the radio network planning tool A955. Required are digital
databases for morpho-structure (clutter, land-use) and height (Digital Terrain Model DTM).

Morpho-structure Digital Terrain Model DTM

Urban
Suburban

80 m
Rural

40 m

Alcatel File Reference Date Edition Page


347232356.doc 3DC 21150 0285 TQZZA 01/2003 03 11
All rights reserved. Passing on and copying of this document, use and communication of its contents not permitted without written authorisation.
Data Throughput in EDGE Networks

Some of the planning steps are well known:

Calculation of field strength prediction


per pixel
Some details: GSM 1800
53.4 dBm EIRP 8PSK
(58 dBm EIRP GMSK not shown)
Antenna height over ground 23m

dBm

Best server plot

1000 m

Site Cell 1 Cell 2 Cell 3


Automatic frequency assignment name TRX 1 TRX 2 TRX 3 TRX 1 TRX 2 TRX 3 TRX 1 TRX 2 TRX 3
e 01 582 563 571 561 569 577 576 584 573
e 02 582 570 579 586 568 576 589 574 560
e 03 570 560 566 568 580 588 564 583 586
e 04 575 562 579 560 585 577 583 570 566
e 05 585 561 565 579 582 567 576 589 563
e 06 572 588 577 579 568 586 584 581 564
Average frequency reuse (approximately) 10 except e 07 587 578 573 589 580 567 576 561 585
between the sites e 01, e 05 and e 06, were (by e 08 563 571 588 578 581 585 568 573 561
e 09 585 571 562 577 573 568 583 588 575
intention) a higher interference was created. Three
e 10 563 582 575 565 578 573 570 588 560
TRX per cell are assigned. The ones with the green e 11 583 569 564 588 560 585 566 571 574
background are used for EGPRS. e 12 565 589 561 587 568 572 570 585 574
e 13 589 569 581 562 567 571 587 583 565
e 14 589 571 583 577 569 580 564 586 567
e 15 582 588 565 572 563 584 579 560 586

Alcatel File Reference Date Edition Page


347232356.doc 3DC 21150 0285 TQZZA 01/2003 03 12
All rights reserved. Passing on and copying of this document, use and communication of its contents not permitted without written authorisation.
Data Throughput in EDGE Networks

Interference calculation per pixel


Between the cells e01 and e06, a spot with high
interference was created by intention. This was
done to show the influence of the interference on
the throughput (see the plots below). In a real
system, such spots would be avoided.

C/I

As a result, we have for each pixel the signal level, the interference level, and implicitly the noise
level (just the thermal noise). This is exactly, what we need to calculate the throughput per pixel.
The result for the above example:

Data throughput on RLC/MAC layer.

If you compare with the field strength plot and the


interference plot above, you can see

- the influence of low field strength

- and the influence of high


interference

Kbit/s

Alcatel File Reference Date Edition Page


347232356.doc 3DC 21150 0285 TQZZA 01/2003 03 13
All rights reserved. Passing on and copying of this document, use and communication of its contents not permitted without written authorisation.
Data Throughput in EDGE Networks

4. SOME EXEMPLARY RESULTS FOR MACROCELLS

4.1 Dense Urban Scenario

Parameter: GSM 900 with EGPRS, EIRP GMSK 58.4 dBm, EIRP 8PSK 53.6 dBm, antenna height
25 m, dense urban, slow fading margin 6 dB, indoor+body loss 18 dB, cell range
defined for voice 1.0 km and 0.5 km (for comparison), fading profile TU3, no frequency
hopping, with incremental redundancy, real Link Adaptation (HYSTE_XY).

Definition Frequency reuse 12 No consideration of


interference (level only)

1.0 km 0.5 km 1.0 km 0.5 km

Mean data throughput per PDCH 29.1 kbit/s 30.3 kbit/s 39.9 kbit/s 49.9 kbit/s
for given cell range

90% data throughput per PDCH 9.2 kbit/s 9.5 kbit/s 18.4 kbit/s 45.3 kbit/s
for given cell range

95% data throughput per PDCH 6.7 kbit/s 7.1 kbit/s 14.0 kbit/s 40.5 kbit/s
for given cell range

4.2 Suburban (Residential) Scenario

Parameter: GSM 900 with EGPRS, EIRP GMSK 58.0 dBm, EIRP 8PSK 53.2 dBm, antenna height
30 m, residential, slow fading margin 6 dB, indoor+body loss 15 dB, cell range defined
for voice 3.0 km and 1.5 km (for comparison), fading profile TU50, no frequency
hopping, with incremental redundancy, real Link Adaptation (HYSTE_XY).

Definition Frequency reuse 12 No consideration of


interference (level only)

2.5 km 1.5 km 2.5 km 1.5 km

Mean data throughput per PDCH 28.7 kbit/s 31.1 kbit/s 35.0 kbit/s 45.8 kbit/s
for given cell range

90% data throughput per PDCH 9.4 kbit/s 10.0 kbit/s 13.4 kbit/s 31.2 kbit/s
for given cell range

95% data throughput per PDCH 7.2 kbit/s 7.8 kbit/s 12.2 kbit/s 24.5 kbit/s
for given cell range

Alcatel File Reference Date Edition Page


347232356.doc 3DC 21150 0285 TQZZA 01/2003 03 14
All rights reserved. Passing on and copying of this document, use and communication of its contents not permitted without written authorisation.
Data Throughput in EDGE Networks

4.3 Conclusion

We can clearly see that the level based approach (no consideration of interference) is much
too optimistic.

The stated throughput values vary drastically between the definitions (nearly factor 4.5
between 95% and mean data throughput).

When including the interference into the calculation, the result becomes nearly independent
from the cell range. This is logical: If we reduce the cell range, we do not only increase the
level from the serving cell, but also the interference from the now closer neighbor cells.

This effect is suppressed when ignoring the interference. This clearly shows how inadequate
the pure level (link budget) approach is.

When only looking to the results for 90% or 95% throughput with consideration of the
interference, no significant difference between EGPRS and GPRS data rates can be seen. This
would lead to the (wrong) conclusion, that EGPRS is not better than GPRS. The much better
spectrum efficiency of EGPRS (see chapter 3.2.3) is suppressed using this definition, but
clearly visible using the mean data throughput.

5. GENERALIZATION TOWARDS COMPLEX NETWORK STRUCTURES

Todays GSM network do often contain hierarchical cell structures (concentric cells / IOU,
microcells, picocells for in-building solutions, multiband). GPRS and EGPRS will of course be
applied to such complex structures as well.

As the data throughput depends quite a lot on interference, and interference can be better
contained in micro-and picocellular environments, the expected data throughput will be higher as in
a pure macrocellular network.

It is in the nature of complex network structures to be very individual to a certain location. Different
to pure macrocells, it is therefore not possible to give generalized values for throughputs.

End of Document

Alcatel File Reference Date Edition Page


347232356.doc 3DC 21150 0285 TQZZA 01/2003 03 15
All rights reserved. Passing on and copying of this document, use and communication of its contents not permitted without written authorisation.