. 20
( 132 .)


UDP/IP. TCP/IP is usually used for web browsing, ¬le transfers and email; it is a reliable
connection-oriented system that acknowledges all data transferred. When TCP sends a
segment of data it starts a timer, known as the retransmission timer. If the data arrives
and is acknowledged successfully before the timer expires then the next segment is sent.
However, if the acknowledgement is not received for any reason the timer may expire.
This is referred to as a timeout. This timeout is usually dynamic and related to the round
trip time. Once a timeout has occurred the connection may close, all data transferred
so far may be lost and the transfer will have to start again from the very ¬rst byte. By
enabling the sharing of time slots, the bit rate may reduce for each GPRS user, and the
TCP ¬‚ow control will slow down the data transfer, but the connection will not timeout.
When a user tries to make a voice call in the busy hour, they may under some cir-
cumstances get a network busy tone. Usually a customer will immediately try one or two
more times to connect. After this, in many cases, the user will leave a long time (2 or
3 minutes) before trying again. When they try this time they may be successful. This
situation arises when the number of users simultaneously attempting to make a call in
the same cell is greater than the operator has planned for. To a certain extent, network
planning is based on educated guesswork, where the planners use tools, tables and past
experience to predict the numbers of users in a given geographical area who may wish to
access the network simultaneously. However, extenuating circumstances cannot always

be planned. Consider, for example, the end of a football match or concert, where a large
number of users in the same cell will attempt to make a simultaneous call.
If a connected user had ¬nished their call almost immediately after the ¬rst user had
given up temporarily, the time slot would have been vacant for 2 or 3 minutes. There is no
way that the user trying to connect would know that another user had just ¬nished a call
and thus the time slot resources in GSM are not always utilized to maximum ef¬ciency.
To dimension the air interface within GSM, Erlang tables are used. Since GPRS also
shares the GSM air interface then this can be taken into account when dimensioning for
GPRS. Consider the following example.

If there are 520 calls an hour and the average call duration is 100 seconds how many
traf¬c channels are required?
No. of Erlangs = Calls per hour — Average call time/(60 min — 60 s)
= 520 — 100/3600 ≈ 15 Erlangs
In a GSM cell it is not expected that all calls connect satisfactorily “ this is known
as grade of service (GoS). Commonly GoS may be 2%, which means that on average
two out of every one hundred calls will be blocked and get the network busy tone. By
consulting the Erlang tables, it can be seen that to dimension a cell for the above number
of GSM calls, 22 TCHs are required (Table 4.1). Adding one TCH for BCCH and one
TCH for the stand alone dedicated control channel (SDCCH)/8 means three TRX units
(one TRX = eight channels) are required.
To prevent a high level of call blocking at the busy hour we needed 22 traf¬c channels
to supply an average of 15 calls at any one time. Note that at any one time more than 15
channels may be used by GSM subscribers but on average 15 subscribers will connect sat-
isfactorily. Since only an average of 15 channels are used out of 22, this leaves 7 channels
vacant. They cannot be used for GSM because of the call blocking. However, they can be
used for GPRS since the network can notice that there is a free slot and allocate extra tem-
porary block ¬‚ows (TBFs) to GPRS subscribers. Using CS-2 (13.4 kbps, see Table 4.2)
this provides over 90 kbps of bandwidth without increasing the number of channels.

4.5.2 Air interface coding schemes
Unlike voice, data is very intolerant of errors and generally must arrive error free. This
does present some problems, as the air interface introduces a signi¬cant number of errors.

Table 4.1 Example of Erlang table
Grade of service
No. of
channels 2% 3% 5%
20 13.2 14.0 15.2
21 14.0 14.9 16.2
22 14.9 15.8 17.1
23 15.8 16.7 18.1

Table 4.2 GPRS data rates
Coding scheme Bit rate Raw bit
(kbps) rate (kbps)
CS-1 9.05 8
CS-2 13.4 12
CS-3 15.6 14.4
CS-4 21.4 20

To protect the data, it is necessary to transmit some codes with it which allow either error
checking and/or error correction. Four new coding schemes have been speci¬ed for the
GPRS air interface, known as CS“1-CS“4 and the data rates for these are highlighted
in Table 4.2. The column indicating the bit rate includes the radio link control/medium
access control (RLC/MAC) headers whereas the raw bit rate column is for user data.
CS-1 offers the lowest data rate but also offers the most protection on the data with
both error detection and error correction. CS-4 offers a much higher data rate but with
very little error checking and no error correction. It may initially appear that CS-4 should
be used extensively; however, as mentioned, the air interface is notorious at losing data
due to interference. If CS-4 is used, there is no error correction and if errored the frame
will need to be retransmitted. This introduces transfer delays and also reduces the actual
throughput. A transfer using CS-1 will be susceptible to the same interference. However,
since it incorporates error correction, frames transmitted between the mobile device and
the network can be corrected at the base station without the need for a retransmission.
The offerings from CS-2 and CS-3 are somewhere in between. Figure 4.6, illustrates the
effectiveness of each of the coding schemes in a noisy environment and in a cell with
little interference. C/I is the carrier to interference ratio, which for this purpose can be
regarded as similar to a signal-to-noise ratio.
Changes in the coding scheme used may take place during the call. This is done
dynamically by the network and is dependent on the current properties of the connection
such as number of errors, retransmissions etc., and is transparent to the user. It should be
noted that different subscribers in the same cell might be using different coding schemes
at the same time.
In practice data rates depend not only on the mobile network but also on the capability
of the handset. Currently these are supporting maximum data rates of around 50 kbps.
Higher data rates are possible using advanced modulation techniques, such as eight-
phase shift keying (8PSK), which require new TRXs in the base station (BTS) and higher-
speed links within the BSS due to the higher data rates possible. This system will be used
in enhanced data for global evolution (EDGE) and is known as enhanced GPRS (EGPRS).
EDGE also works with high-speed circuit switched data (HSCSD); when used with this
system it is known as ECSD.

4.5.3 Classes of devices
GPRS mobile devices can be classi¬ed into three general categories, as follows:

Throughput (kbps)
Average Cell
Noise Range




C/I Ratio
good quality cell
noisy cell
more retransmissions

Figure 4.6 GPRS coding scheme throughput

• Type A can connect to both the GSM and GPRS cores, i.e. the MSC and SGSN, simul-
taneously. For example, this would allow a user to talk on the phone while downloading
an email.
• Type B can also connect to both the GSM and GPRS cores, but the connection can
only use one side of the network at any given time. For example, a subscriber who
is currently downloading data would be noti¬ed of an incoming phone call and must
decide whether to accept the call or not. This will put the data transfer on hold.
• Type C typically will be a data card for a PC allowing it to send and receive data
across the GPRS network. This type of device cannot register with both the GSM and
GPRS core networks at the same time.

Due to implementation complexities, currently for 2G systems only types B and C exist
and this is likely to be the case for some time.
The device class de¬nes the maximum data rate at which a GPRS device can send
or receive. Table 4.3, shows the class number de¬ned for GPRS devices, identifying the
time slot combinations allowed. For example, a class 6 device allows the use of three

Table 4.3 GPRS device classes
Class Downlink Uplink Max. slots
1 1 1 2
2 2 1 3
3 2 2 3
4 3 1 4
5 2 2 4
6 3 2 4
7 3 3 5
8 4 1 5
9 3 2 5
10 4 2 5
11 4 3 5
12 4 4 5
13 3 3 Unlimited
14 4 4 Unlimited
15 5 5 Unlimited
16 6 6 Unlimited
17 7 7 Unlimited
18 8 8 Unlimited
19 6 2 Unlimited
20 6 3 Unlimited
21 6 4 Unlimited
22 6 4 Unlimited
23 6 6 Unlimited
24 8 2 Unlimited
25 8 3 Unlimited
26 8 4 Unlimited
27 8 4 Unlimited
28 8 6 Unlimited
29 8 8 Unlimited

time slots in the downlink and two in the uplink. The maximum is four time slots, so a
user that takes up three down link time slots can only use one in the uplink.
There are a number of problems with designing devices that can utilize more time slots
for data transmission. In GSM, transmission and reception are offset by three time slots
to prevent the mobile device from having to transmit and receive at the same time. If
GPRS time slots overlap, more complex transceiver circuitry is required. Also, the more
time slots used, the more power used since the device is transmitting for longer, and this
creates problems for battery life and device cooling. Therefore, GPRS devices are unlikely
to support the higher class numbers in the near future. It should be noted that this is the
capability of the GPRS terminal. However, the maximum number of time slots available
is also highly dependent on the network equipment capability and the operator policy.

4.5.4 Advantages of GPRS over the air
Consider a GSM phone call. The user makes a call setup by dialling the telephone number
and is then connected to the call recipient. While the call is connected, the user is charged

for that call and the charging is based on the call duration. Many studies have demonstrated
that the average user will only talk for approximately 30“40% of the call duration and
thus, on average, a telephone call is 60“70% inef¬cient. In effect, phone subscribers pay
for silence.
If data traf¬c is now examined, for a GSM connection, this problem becomes consid-
erably more pronounced. A typical data transfer is web page access. A user will connect
to a website, retrieve its contents and then proceed to read the page. The download may
take of the order of 3“4 seconds. However, the user is then idle on the network for a
much longer period, while reading. During this long idle period, the user is also being
billed, since the charging mechanism is based on duration, and the user is still occupying
a ¬xed resource in the system. Data traf¬c is generally of this nature, where bursts of
data transfer are interspersed among long periods of inactivity.
In this scenario, it makes more sense to bill based on volume rather than duration as is
done for voice calls. In this manner, a subscriber will only be billed for actual throughput
and not for the periods of network inactivity.
The introduction of the packet switched network approach alleviates this problem by
allowing resources to be shared among users, and records kept of how many packets were
transferred rather than how long they took.
Another important aspect of GPRS is its always connected nature. A mobile device
does not take up any of the scarce resources over the air but when a user wishes to transmit
data it appears that there is a dedicated channel. In reality the network simply remembers
information about the user and thus can give a quick connection. In GSM every time a user
wishes to make a call they must dial a number and wait a few seconds for the connection.
Unlike GSM, a GPRS user can be ˜always connected™ to the network, where the user has
a logical connection in the form of an IP address allocation (referred to as a packet data
protocol, or PDP, context). However, only when the user sends or receives data will they
actually use any resources, and consequently be subject to charging. A GPRS user will
merely request some data resource, for example, by entering a web location. The network
will then service that request in the available time slot space, thus requiring limited setup
time. In this way, GPRS separates the connection from the actual resource usage.

Similar to any communications protocol, GPRS consists of a layered stack, with different
layers providing key functions and communicating with other layers through primitives.
This section discusses the GPRS protocol stack and provides an explanation of the key
functions of each layer.
Figure 4.7 illustrates the control plane protocol stack for GPRS as it is used within
the GSM network across the BSS towards the SGSN. It can be seen that there is little
difference between it and the user plane protocol stack, shown in Figure 4.8. The user
plane uses an additional layer above the logical link control (LLC) between the mobile
device and the SGSN referred to as the subnetwork-dependent convergence protocol
(SNDCP). Each of these protocols will be described in detail in the following sections.
Unlike Figure 4.7, Figure 4.8 continues the protocol stack across to the GGSN; both the


. 20
( 132 .)