REVISTA FACULTAD DE INGENIERÍA, U.T.A. (CHILE), VOL 11, 2003, pp.

55-60

GENERAL PACKET RADIO SERVICE PHYSICAL-LAYER PERFORMANCE

SIMULATION
Ailton Akira Shinoda1

Received may 23, 2003, accepted december 31, 2003

ABSTRACT

In this paper, we present some results on GPRS physical- layer performance based on a GPRS simulator. The simulator
strictly abides by the GSM standards implementing the detailed transmission chain. We analyze the results of a
comprehensive simulation of the GPRS traffic channels. We evaluate the GPRS performance under various environment
and service rates. The reference channel model used for the simulations is entirely based on the Ricean channel model
and provides a reasonable scenario for radio propagation. The results shown here provide an insight into how to assess
the actual GPRS data channels performance under various service rates.

Keywords: GPRS, GSM, simulation, wireless

INTRODUCTION frame duration of about 4.6 ms. In this paper, we

describe the implementation of a GPRS simulator
Nowadays the current growth rates of data traffic in operating in the FDD/TDD mode. Its is built as add-on
fixed communication environments are extrapolated for modules for SimulinkTM so as to allow for an easy
wireless environment. It is expected that there will be a integration of the GPRS models and the algorithms and
significant demand for wireless data services in the near techniques developed within MatlabTM . The simulator
future. It seems that the success of wireless data at the implements in detail and rigorously all the physical-
moment is just prohibited because certain requirements layer aspects of the GPRS, as specified in GSM
for data rates and cost are not fulfilled. But it is not recommendations Series (e.g., [5-9] ).
visionary to predict that this will change with enhanced
data services e.g. for GSM with HSCSD (High Speed
Circuit Switched Data), GPRS (General Packet Radio In section GPRS System we introduce the GPRS link
Service), EDGE (Enhanced Data Rates for GSM transmission chain. In section Channel we discuss the
Evolution), and UMTS (Universal Mobile Ricean channel model. In section Simulation Setup we
Telecommunication System). discuss the simulation configurations concerning the
simulation setups for each type of environment
GSM already supports data users with circuit switched operation. In section Simulation Results we illustrate
services covering data rates from 1.2 to 64 kbits/s (with some results for the proposed simulation setups and
HSCSD). GPRS takes off to offer improved services for analyze the GPRS performance. In section Conclusion
data users in mobile networks. GPRS is the packet- we draw the final comments and conclusions about the
oriented extension of GSM. It allows data transmission actual status of this work.
up to rates of more than 100 kbps. Since GPRS operates
as a packet-oriented extension, many users can share the
scarce radio resources. In this way a very flexible access GPRS SYSTEM
is possible while idle users can still be online always
and anywhere. A detailed description of GPRS is given This section describes the GPRS as implemented in the
in [1-4]. simulator. The simulation of the GPRS is divided in two
main components: 1) the implementation of the GSM
The access scheme for GPRS is frequency and time standards for the transmitter and 2) the implementation
division system (FDD/TDD), each physical channel of a suitable receiver matched to the transmitter
being characterized by a carrier frequency and a time specifications. The other relevant aspect within the
slot number with channel widths of 200 KHz. Each simulation is the channel model to be used. In this
carrier is divided into frames of 8 time slots, with a paper, the Ricean channel model is included.

1
The author is with the Department of Electrical Engineering, Londrina State University, Londrina, Paraná-Brazil (e-mail:[email protected])
Ailton Akira Shinoda

Fig.1 shows the framework of the implemented GPRS Sequence (40 bits for CS1 and 16 bits for CS2-CS4) and
simulator based on the GSM standards. addition of 4 tail bits (CS1-CS3). In the following is
evaluated the convolutional code (CS1-CS3), which is a
In GPRS radio interface, for each 20 ms the Coding powerful channel coding scheme for wireless mobile
Scheme (CS) gives the information rate from Media communications. The convolutional encoder consists of
Access Control (MAC). The details of the codes are shift registers and modulo-2 adders connected to some
shown in Table I. of the shift registers. The performance of convolutional
code depends on two parameters – coding rate (r) and
Each CS uses, in this order, the following sequence of constraint length (M). The input sequence is fed to the
operations: M-stage shift registers, and output data is calculated
using the contents of M-stage shift registers. The
generator polynomials determine the encoding process
• The information bits are coded with systematic [8]. After this, the puncturing block is applied (CS2 and
block code, building words of information + parity
CS3) taking into account the match rate. Because
bits. convolutional codes perform poorly on bursts errors,
• These information + parity bits are encoded with a interleaving is used in the transmitter to randomize the
convolutional code, building the coded bits. errors so that the convolutional codes can correct them.
The purpose of the interleaving algorithm is to avoid
• Reordering and interleaving the coded bits, and loss of the consecutive information bits. The result of
adding a stealing flag, gives the interleaved bits. the interleaving is a distribution of the reordered 114
bits of a given block, in a diagonal way over
All those operations are made block by block, the size consecutive blocks [8]. The following block, mapping
of which depends on the CS. However, most of the 456 on a burst, adds bits in the block interleaved with values
coded bits, are interleaved and mapped onto bursts in a pre-defined to identify the codification (CS1 -CS4)
very similar way for all of them. Fig. 2 and Fig. 3 are scheme employed. The procedure applied in the
diagrams showing the general structure of the CS mapping was based on [8]. The burst formatting block
scheme. includes 26 training sequence bits, which are defined as
modulating bits with states as given in [1] according to
The first step in the block codification process is the the training sequence code (TSC). Finally, in the
pre-coding USF (Uplink State Flag) [7]. The first 3 bits transmission subsystem, the GSMK block performs the
from CS are coded in the 3 (CS1), 6 (CS2 or CS3) or 12 modulation, taking into account [9].
(CS4) bits. The next step performs the Block Check

CS Rate Mapping
USF BSC TAIL CODING Interleaving Modulation
Source Matching Burst

BER
DPDCH Channel

Mapping
CS USF BSC Tail Rate
Decoding Deinterleaving Burst Demodem
Information Removal Removal Removal Matching
Removal

Fig. 1.- GPRS Simulator Structure

56 Revista Facultad de Ingeniería, Chile, Vol. 11, 2003

 GPRS physical layer performance

Table 1.- GPRS Coding Scheme

Scheme Code Payload BCS Pre-coded Tail bits Coded Puncturing

Data rate
rate USF bits bits
(kbps)
CS1 1/2 181 40 3 4 456 0 9.05
CS2 ≈2/3 268 16 6 4 588 132 13.4
CS3 ≈3/4 312 16 6 4 676 220 15.6
CS4 1 428 16 12 0 456 0 21.4

Radio Block

USF BCS

rate 1/2 convolutional coding

puncturing

456 bits

Fig. 2.- Structure of the CS scheme (CS1-CS3)

Radio Block

USF BCS

block
no coding
code

456 bits

Fig. 3.- Structure of the scheme

Revista Facultad de Ingeniería, Chile, Vol. 11, 2003 57

Ailton Akira Shinoda

The next step in the simulation process is the In the microcell systems, the received signal is
transmission of the processed data through the wireless composed of a strong direct wave and many reflected,
channel. The channel model used is based on a Ricean scattered, or diffracted waves because line of sight
model [10]. (LOS) conditions are dominant in such systems. When
the transmitted signal is given by
The second part of the implemented GPRS simulation
deals with the receiver aspects. At the reception end, the
received signal is processed by the GMSK demodulator. sT (t ) = Re[m(t )exp ( j 2πf c t )] (1)
A de-burst formatting and de-mapping block are then
implemented for extracting the additional bits inserted
in the transmission process. After theses blocks, the a direct wave with amplitude Adir is arriving at angle
received packet is de-interleaved, rate matched, and
of θ dir , and other components are expressed as the
decoded by Viterbi hard-decision decoder [3]
respectively. The next process is the extraction of tail, complex random Gaussian variables, the received signal
BCS, and USF bits. The resulting data sequence is can be expressed as
compared with the original transmitted data sequence to
check for errors. s R ( t) = Re m ( t ) exp ( j2 πf ct ) 

∞ 
+ Re  ∑ Ai m( t) exp{ j2π ( fc + fd cos θi ) t + φi }
CHANNEL
(2)
i =- ∞ 
Radio waves propagate through space as traveling
electromagnetic (EM) waves. The energy of signals
exists in the fo rm of electrical (E) and magnetic (H) = Re {Adir exp ( j2πfd cos θdir t ) + crand ( t)}m ( t ) exp (2πf ct )
fields. Both electrical and magnetic fields vary with
time. The two fields always exist together because a
change in electrical field generates a magnetic field and = Re c ( t ) m ( t ) exp ( j2πf c t )
a change in magnetic field develops an electrical field.
Thus there is a continuous flow of energy from one field
to the other. where crand (t ) is the variation of the nondirect waves
Radio waves arrive at a mobile receiver from different expressed as
directions with different time delays. They combine via
vector addition at the receiver antenna to give a c rand (t ) = crandI + jc randQ (t ) (3)
resultant signal with large or small amplitude depending ∞
upon whether the incoming waves combine to c randI (t ) = ∑ A cos(2πf
i d cosθi t + φi ) (4)
strengthen each other or cancel each other. As a result, a
i =−∞
receiver at one location may experience a signal ∞
strength tens of dB different from a similar receiver c randQ (t ) = ∑ A sin (2πf
i d cosθi t + φi ) (5)
located only a short distance away. As a mobile moves i = −∞
from one location to another, the phase relationships
c(t ) is given by
between the various incoming waves also change. Thus,
there are substantial amplitude and phase fluctuations, and fading variation
and the signal is subjected to fading. It should also be
noted that , whenever relative motion exists, there is a c(t ) = Adir exp ( j 2πf d cosθdir t ) + crand (t ) (6)
Doppler shift in the received signal. In the mobile radio
case, the fading and Doppler shift occur as a result of
the motion of the receiver through a spatially varying In the case of Ricean fading, the power spectrum
()
field. Doppler shift also results from the motion of the
scatterers of the radio waves (e.g., cars, trucks, density of c t is given by
vegetation). The effect of multipath propagation is to
produce a received signal with amplitude that varies
S ( f ) = bdirδ ( f d cosθdir ) +
brand
quite substantially with location. At UHF and higher (7)
 f 
2
frequencies, the motion of the scatterers also causes
fading to occur even if the mobile set or handset is not πf d 1 −  
in motion [2].  fd 

58 Revista Facultad de Ingeniería, Chile, Vol. 11, 2003

 GPRS physical layer performance

The first term in equation (7) represents a direct wave results and this is due to increased influence of the long
component, and the second one represents nondirect fading periods experimented by the mobile in low
wave components. bdir and brand represent the power speed.
of the direct wave and the total power of the nondirect 1
waves, and the power ration of them respectively,
0,1

b A2
K = dir = dir (8) 0,01
brand 2brand
1E-3

BER
is called Ricean factor. Because nondirect waves are
CS1
subject to a complex Gaussian random process, the
()
1E-4 CS2

probability density function of c t is given by CS3

CS4
1E-5

 (c I − Adir )2 + cQ2 
(
p c I , cQ ) =
1
exp 



(9) 1E-6

2πb 02 2 b02
-2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12

  K (dB)

Fig. 4.- BER for user’s low mobility (v= 5 Km/h)

SIMULATION SETUP

This section describes the simulation methodology used 1

to assess GPRS performance given in terms of Bit Error

Rate (BER). During the simulation, data are transmitted 0,1
continuously in a frame-by-frame (only one slot) basis
over the Ricean channel. For the results shown here the 0,01
channels are generated as if an MS is moving within the
cell with velocities of 5 km/h, 50 km/h, and 120 km/h. 1E-3
BER

The simulation employs a carrier frequency of 1800

CS1
MHz. 1E-4 CS2
CS3
CS4
The channel environment is changed through Ricean 1E-5
factor (K) since an additive white Gaussian noise
(higher K) until flat fading (lower K). This variation is
1E-6
made in dB, employing a step of 1 dB. -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12

K (dB)

The selection of data rates to be simulated is chosen to

match the data rates of the GPRS data coding scheme of Fig. 5.- BER for user’s medium mobility (v= 50 Km/h)
9.05 kbps, 13.4 kbps, 15.6 kbps, and 21.4 kbps. No
antenna diversity technique is considered for the Through Fig. 5, we observe that the CS1 coding scheme
simulation. performs better than CS2-CS4. In this scenario, the
observed performance difference is due to the higher
coding gains obtained with the use the convolutional
SIMULATION RESULTS code and no puncturing in the transmission. In spite of
no convolutional code in CS4, it shows a better
This section describes the simulation results obtained performance than CS2-CS3 coding scheme.
for the simulation setups. Fig. 4 to Fig. 6 shows some Analogically this is due to no puncturing process in the
results for BER as a function of K (Ricean factor) for CS4 coding scheme.
user’s typical mobility environment.
Fig. 6 shows the BER performance for high mobility CS
Fig. 4 shows the BER performance for low mobility data rate. Note the improvement for CS1-CS3 scheme
CS1-CS4 data rate. We note that, as expected, the coding taking into account the me dium mobility
higher the direct wave (Ricean factor) the better the environment. In high mobility scenarios the
performance. Besides, CS1-CS4 presents the same convolutional code compensates the puncturing process

Revista Facultad de Ingeniería, Chile, Vol. 11, 2003 59

Ailton Akira Shinoda

for CS2 and CS3 scheme coding. As the CS4 does not Service”, IEEE Communications Magazine, pp.
perform the convolutional code, its performance 94-104, August 1997.
remains the same.
[2] R. H. Clarke, “Statistical Theory of Mobile Radio
1
Reception”, Bell System Technical Journal 47,
July-August 1968, pp.957-1000.
0,1 [3] T. Irnich and P. Stuckmann, “Analytical
performance evaluation of Internet access over
0,01 GPRS and its comparison with simulation results ”,
Personal, Indoor and Mobile Radio
1E-3 Communications, vol. 5 , pp. 15-18, Sept. 2002.
BER

CS1 [4] W. Featherstone and D. Molkdar, “Capacity

1E-4 CS2 benefits of GPRS coding schemes CS-3 and CS-4”,
CS3
CS4 3G Mobile Communication Technologies,
1E-5
pp.287-291, May, 2002.
1E-6 [5] G. Divsalar, M. K Simon, and J. H. Yuen, “Trellis
-2 -1 0 1 2 3 4 5 6 7 8 9 10 11 Coding with Asymmetric Modulation”, IEEE
K (dB)
Transaction Communication, vol. Com 35, no.2,
February 1987.
Fig. 6.- BER for user’s high mobility (v= 120 Km/h)
[6] GSM Specification Series 05.01, V3.3.2, “Physical
Layer on the Radio Path: General Description”,
CONCLUSION 1991,www.3gpp.org/ftp/Specs
[7] GSM Specification Series 05.02, V3.8.0,
The proposed and described implementation of the “Multiplexing and Multiple Access on the Radio
GPRS uplink link has been carried out successfully. For Path”, 1991, www.3gpp.org/ftp/Specs
the implementation, the simulation was implemented
within Mathworks/Simulink using C language to [8] GSM Specification Series 05.03, V3.6. 1, “Channel
optimize the simulation performance. The simulator C Coding”, 1991, www.3gpp.org/ftp/Specs
code was aggregated to Simulink using S-Functions [9] GSM Specification Series 05.04, V3.1.2,
compiled as dlls (Dynamic Link Libraries) called from “Modulation”, 1991, www.3gpp.org/ftp/Specs
with Simulink interface. The Simulink framework
environment provides a quick and simple interface that [10] Hata, M., “Empirical Formula for Propagation Loss
can be used for a reliable insight into the GRPS uplink in Land Mobile Radio Services”, IEEE
behavior. Transactions on Vehicular Technology 29(3),
1980.
The observed differences in the BER performance
according to the MS velocity are mainly dependent
upon the radio propagation environment conditions.

We have shown that the simulator can be efficiently

employed as a test bed for new technologies based on
the GSM specification standards.

Taking into account the time spent in the simulation, the

lower the BER the longest was the simulation time. For
BER in the range of 10-6 ( Pentium 4 processor and 512
MB RAM) the time elapsed was 30 hours.

REFERENCES

[1] G. Brasche and B. Walker, “Concepts, Service, and

Protocols of new GSM Phase 2+ General Radio

60 Revista Facultad de Ingeniería, Chile, Vol. 11, 2003