The Performance/Bandwidth Trade- Off in 1G and 2G Cellular Networks

The Performance/Bandwidth Trade-
Off in 1G and 2G Cellular Networks
The AMPS/ETACS analog cellular radio networks introduced in the 1980s used very
well established baseband and RF processing techniques. The analog voice stream was
captured using the variable voltage produced by the microphone, companded and
pre-emphasized, and then FM modulated onto a 25 kHz (ETACS) or 30 kHz (AMPS)
radio channel.
The old 1200-bit rate FFSK signaling used in trunked radio systems in the 1970s was
replaced with 8 kbps PSK (TACS) or 10 kbps PSK for AMPS. (As a reminder, AMPS
stands for Advanced Mobile Phone System, TACS for Total Access Communications
System, and E-TACS for Extended TACS—33 MHz rather than 25 MHz allocation.)
At the same time (the Scandinavians would claim earlier), a similar system was
deployed in the Nordic countries known as Nordic Mobile Telephone System (NMT).
This was a narrowband 121⁄2 kHz FM system at 450 MHz.
All three first-generation cellular systems supported automatic handover as a handset
moved from base station to base station in a wide area network. The handsets could
be instructed to change RF channel and to increase or decrease RF power to compensate
for the near/far effect (whether the handset was close or far away from the base
station). We have been using the past tense, but in practice, AMPS phones are still in
use, as well as some, though now few, NMT phones.
Power control and handover decisions were taken at the MSC on the basis of channel
measurements. AMPS/ETACS both used supervisory audio tones. These were
three tones at 5970, 6000, and 6030 Hz (above the audio passband). One of the three
tones would be superimposed on top of the modulated voice carrier. The tone effectively
distinguished which base station was being seen by the mobile. The mobile then
retransmitted the same SAT tone back to the base station. The base station measured
the signal-to-noise ratio of the SAT tone and either power-controlled the handset or
instructed the handset to move to another RF channel or another base station. Instructions
were sent to the mobile by blanking out the audio path and sending a burst of 8-
kbps PSK signaling.
This still is a very simple and robust system for managing handsets in a mobile
environment. However, as network density increased, RF planning became quite complicated
(833 channels to manage in AMPS, 1321 channels to manage in ETACS), and
there was insufficient distance between the SAT tones to differentiate lots of different
base stations being placed relatively close to one another. There were only three SAT
tones, so it was very easy for a handset to see the same SAT tone from more than one
base station.

Given that the SAT tones were the basis of power control and handover decisions,
the network effectively became capacity-limited in terms of its signaling bandwidth.
The TDMA networks (GSM and IS136 TDMA) address this limitation by increasing
signaling bandwidth. This has a cost (bandwidth overhead) but delivers tighter power
and handover control.
For example: In GSM, 61 percent of the channel bandwidth is used for channel
coding and signaling, as follows:
Speech codec 13.0 kbps 39%
Codec error protection 9.8 kbps 29%
SACCH 0.95 kbps 2%
Guard time/ramp time/synchronization 10.1 kbps 30%
TOTAL 33.85 kbps 100%
The SACCH (slow associated control channel) is used every thirteenth frame to provide
the basis for a measurement report. This is sent to the BTS and then on to the BSC
to provide the information needed for power control and handover. Even so, this is
quite a relaxed control loop with a response time of typically 500 ms (twice a second),
compared to 1500 times a second in W-CDMA (IMT2000DS) and 800 times a second in
CDMA2000.
The gain at system level in GSM over and above analog cellular is therefore a product
of a number of factors: 1. There is some source coding gain in the voice codec. 2.
There is some coherence bandwidth gain by virtue of using a 200 kHz RF channel
rather than a 25 kHz channel. 3. There is some channel coding gain by virtue of the
block coding and convolutional coding (achieved at a very high price with a coding
overhead of nearly 10 kbps). and 4. There is a gain in terms of better power control and
handover.
In analog TACS or AMPS, neighboring base stations measure the signal transmission
from a handset and transfer the measurement information to the local switch for
processing to make decisions on power control and handover. The information is then
downloaded to the handset via the host base station. In an analog network being used
close to capacity, this can result in a high signaling load on the links between the base
stations and switches and a high processing load on the switch.
In GSM, the handset uses the six spare time slots in a frame to measure the received
signal strength on a broadcast control channel (BCCH) from its own and five surrounding
base stations. The handset then preprocesses the measurements by averaging
them over a SACCH block and making a measurement report. The report is then
retransmitted to the BTS using an idle SACCH frame. The handset needs to identify cochannel
interference and therefore has to synchronize and demodulate data on the
BCCH to extract the base station identity code, which is then included in the measurement
report. The handset performs base station identification during the idle SACCH.
The measurement report includes an estimate of the bit error rate of the traffic channels
using information from the training sequence/channel equalizer. The combined
information provides the basis for an assessment of link quality degradation due to cochannel
and time dispersion and allows the network to make reasonably accurate
power control and handover decisions.

Given the preceding information, various simulations were done in the late 1980s to
show how capacity could be improved by implementing GSM. The results of base simulations
were widely published in the early 1990s. Table 11.9 suggests that additional
capacity could be delivered by increasing the reuse ratio (how aggressively frequencies
were reused within the network) from 7 to 4 (the same frequency could be reused every
fourth cell). The capacity gain could then be expressed in Erlangs/sq km.
In practice, this all depended on what carrier-to-interference ratio was needed in
order to deliver good consistent-quality voice. The design criteria for analog cellular
was that a C/I of 18 dB was needed to deliver acceptable speed quality. The simulations
suggested GSM without frequency hopping would need 11 dB, which would
reduce to 9 dB when frequency hopping was used. In practice, these capacity gains initially
proved rather illusory partly because, although the analog cellular networks
were supposed to be working at an 18 dB C/I, they were often working (really quite
adequately) at C/Is close to 5—that is, there was a substantial gap between theory and
reality.
The same reality gap happened with coverage predictions. The link budget calculations
for GSM were really rather overoptimistic, particularly because the handsets and
base station hardly met the basic conformance specification.
Through the 1990s, the sensitivity of handsets improved, over and above the conformance
specification, typically by 1 dB per year. Similarly, base station sensitivity
increased by about 3 or 4 dB. This effectively delivered coverage gain. Capacity gain
was achieved by optimizing power control and handover so that dropped call performance
could be kept within acceptable limits even for relatively fast mobility users in
relatively dense networks. Capacity gain was also achieved by allocating 75 MHz of
additional bandwidth at 1800 MHz. This meant that GSM 900 and 1800 MHz together
had 195 + 375 × 200 kHz RF channels available between 4 network operators, 570 RF
channels each with 8 time slots = 4640 channels! GSM networks have really never been
capacity-limited. The capacity just happens sometimes to be in the wrong place. Cellular
networks in general tend to be power-limited rather than bandwidth-limited.

So it was power, or specifically coverage, rather than capacity that created a problem
for GSM 1800 operators. As frequency increases, propagation loss increases. It also gets
harder to predict signal strength. This is because as frequency increases, there is more
refraction loss—radio waves losing energy as they are reflected from buildings or
building edges. GSM 1800 operators needed to take into account at least an extra 6 dB
of free space loss over and above the 900 MHz operators and an additional 1 to 2 dB for
additional (hard to predict) losses. This effectively meant a network density four to five
times greater than the GSM 900 operators needed to deliver equivalent coverage. The
good news was that the higher frequency allowed more compact base station antennas,
which could also potentially provide higher gain. The higher frequency also
allowed more aggressive frequency reuse; though since capacity was not a problem,
this was really not a useful benefit.
It gradually dawned on network operators that they were not actually short of spectrum
and that actually there was a bit of a spectral glut. Adding 60 + 60 MHz of
IMT2000 spectrum to the pot just increased the oversupply. Bandwidth effectively
became a liability rather than an asset (and remains so today).
This has at last shifted attention quite rightly away from capacity as the main design
objective. The focus today is on how to use the limited amount of RF power we have
available on the downlink and uplink to give acceptable channel quality to deliver an
acceptably consistent rich media user experience. 261