1414

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 48, NO. 5, SEPTEMBER 1999

Analog-to-Digital Transition: Selecting the

Optimal Cellular Radio Mix
Mathilde Benveniste, Member, IEEE

AbstractThe capacity gains from the transition from analogto-digital time-division multiple-access (TDMA) air-interface
technology depend on the number of base-station radios
(transceivers) changed from analog to digital. It is possible to
have no capacity gains, and even losses, if too many radios are
changed over. This paper addresses the problem of nding the
optimal radio mix. An analytical model is developed to compute
the blocking rate experienced by subscribers in a cell, given the
cells offered load, the mix of analog and dual-mode subscriber
units, and the analog-to-digital split of the radio frequency (RF)
channels available to the cell.
Index Terms Blocking probability, capacity maximization,
cellular telephony, digital radios, wireless teletrafc model.

I. INTRODUCTION

HE INTRODUCTION of digital cellular telephony brings

with it the potential for a signicant capacity increase
as digital (North American) time-division multiple-access
(TDMA) technology can accommodate three conversations in
the bandwidth used for one conversation by analog technology.
The full capacity increase will not be realized, however, unless
all subscribers own phones with digital TDMA capability.
Until then, it is expected that analog and digital technology
will coexist, achieving less than the ultimate capacity gains.
The gains realized will depend on the number of base-station
transceivers, referred to simply as radios in this paper, that
are changed from analog-to-digital. It is possible to have
no capacity gains, and even losses, if too many radios are
changed over.
The question that must be answered for the transition phase
is: how many of the radios in a cell must be changed over
from analog-to-digital, given the proportion of subscriber units
with dual-mode capabilitythe ability to interface with both
analog and digital radios. In determining the radio mix, one
might wish to maximize the call handling capacity of a cell,
or minimize the blocking experienced by the subscribers. To
address these questions, we develop an analytical model to
compute the blocking rate experienced by subscribers in a
cell, given the cells offered load, the mix of analog and
dual-mode subscriber units, and the analogdigital split of
the radio frequency (RF) channels available to the cell. The
models results enable us to select the channel split that gives
the lowest blocking rate and to compile charts that show the
Manuscript received June 22, 1997; revised August 5, 1998. Patent application allowed.
The author is with AT&T LaboratoriesResearch, Florham Park, NJ 07932
USA (e-mail: benveniste@att.com).
Publisher Item Identier S 0018-9545(99)07366-1.

Fig. 1. Equivalent random systems.

relationship between the proportion of dual-mode subscriber

units, the number of digital radios in a cell, and the maximum
offered load that can be served at a specied blocking rate.
II. QUEUEING MODEL
To compute the blocking experienced in a cell containing
both analog and digital radios, we represent the two types
of radios as a queueing system consisting of two server
groups. Two independent streams of calls are directed to the
two server groups. They consist of the calls initiated by the
analog subscriber units and the calls initiated by the dualmode subscriber units. The two streams are assumed to be
Poisson distributed (call durations are exponentially distributed
and are independent of each other and of the rate of call
generation). The load for each stream will depend on the total
load offered in the cell and the proportion of analog and dualmode subscriber units. Assuming that new and handoff calls
are treated with equal priority, each stream of calls may contain
both new calls and handoff calls.
Fig. 1 illustrates the relationship between the trafc streams
and server groups. The digital server group receives calls from
dual-mode subscriber units only. If any of these calls nd
all digital radios busy, they are directed to the analog server
group, which, in addition to the overow from the digital
server group, receives also the calls from the analog subscriber
units.

00189545/99$10.00 1999 IEEE

BENVENISTE: ANALOG-TO-DIGITAL TRANSITION: SELECTING THE OPTIMAL CELLULAR RADIO MIX

We wish to compute the blocking probability

for the
for the dual-mode subscriber units, and
for the
system,
be the dual-mode offered load;
analog subscriber units. Let
the analog offered load;
the number of digital servers;
the number of analog servers. The number of digital
and
or analog servers is equal to the number of digital or analog
radios, respectively, times the number of calls each radio can
serve simultaneously. The number of digital servers per radio
is the number of time slots per frame. In the case of North
American TDMA, this number is three. We can compute the
above blocking probabilities using the following relationships.
is equal to the probability
that a call reaching the
analog server group will nd all analog radios busy. That is,
(1)
is equal to the probability that a dual-mode subscriber unit
call will nd both the digital and the analog server groups
busy. Hence
(2)
is the probability that all digital radios are busy.
where
is equal to the ratio of
The system blocking probability
the combined blocked load over the combined offered load.
That is,
(3)
is the dual-mode load blocked at the digital server
where
group.
To compute the desired probabilities we need to know
and
the blocking probability at the analog and digital
is computed
server groups, respectively. The probability
where
through the use of the Erlang loss formula
is the number of servers and the offered load. Given
radios in the digital group and an offered load
the blocking
is
probability

1415

receiving a single equivalent random load. The size of

the equivalent random primary server group and the equivalent
random offered load are the two parameters used to make the
two systems equivalent. The two systems will be equivalent if
the rst two moments of the input to the analog server group
are the same. That is, the mean and the variance of the
total load offered to the analog server group must be the same
in the original and in the equivalent random system. As a
result, the blocking probability at the analog server group will
be the same for the two equivalent systems. We will evaluate
using the equivalent random system.
First, we must nd the values for and that will render
the two queueing systems equivalent. To this end, we write
expressions for and in the two systems and equate them.
In the original system the input to the analog server group
consists of two independent streams, the overow from the
digital server group and the analog trafc stream. Due to
this independence, we can compute its mean and variance as
follows:
(5)
(6)
and
are the mean and variance of the digital
where
overow. Their values are given by
(7)
(8)
is the dual-mode offered load, and
is the number
where
is the analog offered load. As a Poisson
of digital radios [2].
trafc stream, the variance of the analog offered load will be
equal to its mean.
In the equivalent random system the mean and variance of
the overow are given by the following:

(4)
(9)
is more complex.
The computation of the probability
Even though the original trafc streams are Poisson, the input
to the analog server group is not, because it contains the
overow from the digital server group, which (overow) is
not Poisson. Hence, an equivalent approximate method was
employed to calculate the blocking probability at the analog
server group.
A. The Equivalent Random Method
Our analysis is based on an approximation technique known
in teletrafc engineering as the equivalent random method [1].
We represent the queueing system at hand as an equivalent
simple overow system with one primary and one overow
server group, as illustrated in Fig. 1. The analog server group
is the overow group. The two inputs going into the analog
server group are combined and represented as the overow
from a single equivalent random primary server group,

(10)
By rst computing and with (5)(8), one can then use (9)
and (10) to solve for the equivalent random group size and
the equivalent random load .
one can calculate the blocking probability
Given and
at the analog server group. In the equivalent random
is the conditional probability that a call blocked
system,
at the primary equivalent random server group will nd all
servers in the overow group busy. The blocking probability at
The overall blocking
the primary server group will be
where
is the number of
probability will be
analog radios. Using the denition of conditional probability,
we have
(11)

1416

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 48, NO. 5, SEPTEMBER 1999

TABLE I
BLOCKING PROBABILITIES FOR PROPORTION OF DMMs = 0

TABLE III
BLOCKING PROBABILITIES FOR PROPORTION OF DMMs = 0:2

TABLE II
BLOCKING PROBABILITIES FOR PROPORTION OF DMMs = 0:1

TABLE IV
BLOCKING PROBABILITIES FOR PROPORTION OF DMMs = 0:3

Step 3: Compute using (12).

Step 4: Find by inverting (10)

B. Approximate Solution for the Parameters

of the Equivalent Random System
The solution of (9) and (10) for and was achieved by
using Rapps method [3]. Rapp observed that a solution to
(9) and (10) can be obtained by combining with (10) the
following:
(12)
where the peakedness factor

is dened as

This observation leads to the following solution procedure.

C. Algorithm for

and

Step 1: Compute and

Step 2: Compute .

Step 5: Find an integer value for by truncating the value

computed in Step 4 to its integral part.
Step 6: Compute a new value for from (10), which is
now solved for in terms of the integer and

using (5)(8).

Having obtained values for and

we can compute
using (11). With
and
[see (4)] known, one can nd
and
using (1)(3), respectively.
the probabilities
The accuracy of the above described approximation is tested
by simulations whose results are reported in Appendix I.

BENVENISTE: ANALOG-TO-DIGITAL TRANSITION: SELECTING THE OPTIMAL CELLULAR RADIO MIX

TABLE V
BLOCKING PROBABILITIES FOR PROPORTION OF DMMs = 0:4

1417

TABLE VI
BLOCKING PROBABILITIES FOR PROPORTION OF DMMs = 0:5

III. RESULTS
The formulas presented in the preceding section were applied to a sample problem. We considered a cell assigned a set
of 19 RF channels, receiving an offered load of 14.3 erlangs.
Assuming different values for the proportion of dual-mode
subscriber units, we computed the blocking probability experienced by the entire subscriber population and by the dual-mode
and analog subscriber units separately, for a different split of
the available channels between analog and digital.
The results appear in Tables IVI and in Fig. 2. In each of
the tables, column 1 gives the number of digital radios, which
ranged from 0 to 19. Column 2 is the blocking probability
for all subscriber units, column 3 is the blocking experienced
by the dual mode subscriber units, and column 4 is the
blocking experienced by the analog subscriber units. We
observe that dual-mode subscriber blocking is always less
than analog subscriber blocking, as we expected from (1) and
(2).1 The overall blocking probability experienced by the total
subscriber unit population lies between the dual-mode and
analog subscriber blocking probabilities.
In the presence of some dual-mode subscriber units, the
system blocking probability will decrease at rst as channels
switch over from analog-to-digital radios due to the capacity
increase this conversion implies. This blocking decrease is the
result of a combination of two effects: a monotonic decrease
in dual-mode blocking, and an initial decrease in analog
blocking followed by a subsequent increase as the analogto-digital conversion persists. Consequently, the decrease in
overall blocking stops when the higher analog blocking more
than compensates for any decrease in digital blocking.
Table VII shows the number of digital radios that would
minimize blocking for different values of the proportion of
dual-mode subscriber units. As the proportion of dual subscriber units increases, the number of digital radios that
1 The analog subscriber blocking probability is equal to the analog server
group blocking probability. The dual-mode blocking probability is equal to
the analog server group blocking probability times a number less than one.
Hence, the dual-mode blocking would be lower than that experienced by the
analog subscriber units.

Fig. 2. Plot of system blocking probability versus number of digital radios.

TABLE VII
OPTIMAL RATIO MIX

minimizes system blocking probability increases, and the

blocking probability decreases.
When handoff is possible between cells, the offered load in
each cell is a combination of new calls and handoff calls. The
ratio of the two types of load, dual-mode and analog, will be
equal to the ratio of the two types of subscriber units because,
with TDMA, a dual-mode call that has been served by an
analog radio can be served by a digital radio upon handoff.

1418

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 48, NO. 5, SEPTEMBER 1999

Fig. 3. Plot of capacity versus the number of digital radios for different DMM proportions in a cell with 19 channels.

Fig. 4. Plot of capacity versus the number of digital radios for different DMM proportions in a cell with 38 channels.

Otherwise, if calls served by analog radios could be handed off

to other analog radios only, the ratio of the two types of load
would not equal the ratio of the two types of subscriber units.
The offered load ratio would be biased in favor of analog, with
the degree of bias depending on the handoff rate.
A. Capacity Reference Charts
The results of the preceding section illustrate the capability
of the derived model to compute blocking probabilities for a
variable mix of analog and digital radios, given the offered
load and the proportion of dual-mode subscriber units. In
designing a cellular system, one would be interested in the
maximum offered load that could be handled with a specied
blocking probability. This load is referred to as the load
capacity. For this reason, we have compiled charts showing the relationship between load capacity, the number of
digital radios, and the proportion of dual mode subscriber
units, assuming that the desired blocking probability is 2%.

We considered two cases, which differed by the number of

channels available to the cell. In the rst case there was a set
of 19 channels available, and in the second 38 channels.
The capacity charts were generated by the sequential application of the model. Blocking probabilities were computed
for various offered loads. The capacity was found through
interpolation, by equating the blocking probability to 0.02.
The charts appear in Figs. 3 and 4. Fig. 3 gives the capacity
for a cell with 19 channels and in Fig. 4, for a cell with
38 channels. Examining the two charts we observe that, for
a given proportion of dual-mode subscriber units, capacity
increases with the number of digital radios up to a certain
maximum value, after which it decreases. This is due to the
increased blocking experienced by analog subscriber units
when too many analog radios are replaced by digital.
The maximum attainable capacity will increase with the
proportion of dual-mode subscriber units. To achieve the
maximum capacity gain in the absence of digital capability,

BENVENISTE: ANALOG-TO-DIGITAL TRANSITION: SELECTING THE OPTIMAL CELLULAR RADIO MIX

1419

Fig. 5. System blocking versus number of digital radios.

one would have to triple the number of channels available to

the cell. For a lower proportion of dual-mode subscriber units,
the capacity gain would be lower, but signicant. For instance,
a cell with 19 channels and 30% dual-mode subscriber units
would need only two digital radios in order to enjoy a capacity
of 14.8 erlangs, which represents a 21% increase relative to
the all-analog case. For the same dual-mode subscriber unit
proportion, a cell with 38 channels would require ve digital
radios to achieve a capacity of 35.6 erlangs, which represents
a 24% increase.
B. Illustrative Example
The TDMA/AMPS provisioning problem illustrates the usefulness of the derived model. Many cellular engineers determine how to allocate the available channels between analog
and digital radios by ignoring the analog capability of the dualmode phone. That is, the required number of digital radios
is computed by nding the number of channels needed in a
simple server group offered a load equal to the dual-mode
load. This approximation overprovisions in favor of digital
radios and at the expense of analog, thus underestimating the
overall blocking rate. More importantly, it provides no means
of estimating the load capacity of a given conguration.
For example, consider a cell with 19 channels. The results
of our analysis in Fig. 3 show that the load capacity for a target
system blocking rate of 2% varies with the proportion of dualmode phones. According to our model, a load of 18 erlangs
would experience 1.9% blocking rate for a 50-50 split between
dual-mode and analog subscriber units. When only 20% of the
subscriber units are dual mode, however, the lowest achievable
system blocking rate is 8.6% for the same load, by using two
digital radios. If instead of our model, the above approximation
was used, whereby only the dual-mode load is considered,
erroneous conclusions would be drawn. By this approximation,
an 18-erlang offered load would be deemed sustainable at 2%
blocking when 20% of the subscriber units are dual mode,

with three digital radios since, one would reason, 3.6 erlangs
offered to nine servers result in 1% blocking rate. According
to our model, however, the system blocking rate with three
digital radios would be 10%. When only 20% of the subscriber
units are dual mode, our model shows that the maximum load
capacity for a 2% blocking rate is 13.7 erlangs, achieved with
either one or two digital radios.
IV. CLOSING REMARKS
Analytical approximations have been derived to give the
blocking probability experienced by subscribers in a cell
equipped with both digital and analog radios. We made the
following observations concerning the behavior of blocking
probabilities.
The blocking experienced by dual-mode subscriber units
decreases as the number of digital radios increases. A limited
conversion of analog-to-digital radios causes a decrease in
analog blocking as well, but eventually analog blocking will
increase. Hence, the installation of too many digital radios
could lead to capacity loss. Dual mode subscriber units experience always a lower blocking than analog subscriber units.
The overall blocking decreases with an increasing presence of
dual-mode phones and the right number of digital radios at
the cell site.
To facilitate growth planning of cellular systems, we have
compiled capacity charts for two channel allocations: 19 and
38 channels in a cell. These charts show the maximum capacity
attainable for a given proportion of dual-mode subscriber
units as a function of the number of digital radios in a
cell. The impact of TDMA technology on capacity depends
on the proportion of dual mode phones in the subscriber
population. We observe that, while the attainment of the
maximum capacity increase (which is equivalent to a threefold spectrum increase under analog technology) requires a
full changeover from analog-to-digital technology, a limited
change-over could generate noticeable capacity gains. For

1420

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 48, NO. 5, SEPTEMBER 1999

instance, a cell with 19 channels requires only two digital

radios to achieve a 21% capacity increase, assuming a 30%
proportion of dual-mode subscriber units.
APPENDIX I
The approximate method due to Rapp [3] was employed
to solve the system of equations generated by the Equivalent
Random Method [1] described in Section II-A, which is the
approximate method used to solve the queuing problem at
hand. The accuracy of these approximations was tested by
comparison to Monte Carlo simulation results and was found
to be acceptable. Fig. 5 shows a plot of the system blocking
probability as a function of the number of digital radios in the
cell, calculated by both the analytical approximation model
and a simulation for the case considered in Table VI. That is,
we considered an offered load of 14.3 erlangs in a cell with 19
RF channels and assumed that the offered load was split 50-50
between analog and dual-mode subscriber units. 100 000 calls
were simulated.

REFERENCES
[1] R. I. Wilkinson, Theories for toll trafc engineering in the U.S.A.,
Bell Syst. Tech. J, vol. 35, no. 2, pp. 421514, 1956.
[2] R. B. Cooper, Introduction to Queueing Theory, 2nd ed. Amsterdam,
The Netherlands: Elsevier, 1981, p. 140.
[3] Y. Rapp, Planning of junction network in a multiexchange area,
Ericson Technics, 20, no. 1, pp. 77130, 1964.

Mathilde Benveniste (M96) received the B.A. degree from the University of Pennsylvania, Philadelphia, and the M.S.E. and Ph.D. degrees from Johns
Hopkins University, Baltimore, MD.
She has been working in the area of telecommunications planning and performance evaluation
since she joined AT&T Bell Laboratories (AT&T
Laboratories) in 1983. She has worked on wireless
communications for the past ten years.
Dr. Benveniste is a member of the Institute for
Operations Research and the Management Sciences.