B Christer V Johansson
Ericsson Radio Systems AB
S-16480 Stockholm
Sweden
Abstract. An analytical formula similar to the Qualcomm describes the simulation set up for the CDMA test system,
capacity formula for speech is theoretically derived for cellular including some radio network algorithms used. Section IV
packet data with ARQ. The value of the target E,/N, that presents system simulation results for the CDMA test system,
gives the highest system capacity can from the formula together and these results are in Section V compared to the theory.
with the E,/N, to BLER (Block Error Rate) mapping be Section VI gives some resent results for the alpha-group con-
found. (We define a block to be the smallest unit that is re-trans-
cept (WCDMA) in ETSI. Finally, Section VI1 discusses the
missible on the air-interface.) The system capacity can be found
except for a multiplicative constant which depends on the reuse results found in this paper.
and on the path loss function.
II. The Capacity Formula
System simulations aiming for confirming the theory has been A. Derivation
performed for a downlink 32 kbps data channel in a CDMA test
system with non-orthogonal DL. Results for the alpha-group Excluding thermal noise and interference from other cells,
concept (WCDMA) in ETSI are also shown. the number of users Nc!rcuft,uL that simultaneously can
communicate in the up link In a cellular CDMA telephone
The optimal value of the target E,/N, found by simulations system is given by (see [ 11)
is very close to the one given by the theory. Furthermore, the
ratio of the capacity in the two cases investigated (with and with- -
out fast PC) is very close to what was predicted by the theory. ‘circuit, U L
The result in this paper indicate that using the theory makes it Here W is the band width, R is the information bit rate,
possible to optimize coding, interleaving and block size with and E,/N, the information bit energy to the noise ratio that
respect to system capacity by only performing link simulations. is required for the link to operate satisfactorily.
Hence, system simulations are probably not needed for this opti- We will in the following derive a similar formula usable for
mization procedure. a cellular ARQ packet data system. The arguments we will
use are valid for down-link as well as up-link, and holds for
I. Introduction CDMA system as well as TDMA systems. However, it is nec-
In this paper we discuss how to set the E b / N , target in a essary to have a power control algorithm that gives all users
CDMA packet data system: If the target is set too low, we will the same E,/N,.
suffer from many retransmissions. The result is a capacity Let a cellular packet data system have Npacket users per
decrease. However, the number of mobiles that a system can cell, all transmitting with the information bit rate R . Let
serve simultaneously is inversely proportional to the E,/No TP,,,,, (“throughput cell”) be the number of information bits
target, so also a too high .!?,/No target will result in a capac- that per second are successfully transmitted in one cell.
ity decrease. This paper shows how one can find the optimal Assuming that all users operate at the same E,/N, value
E,/No target (with respect to system capacity) from the yields
.??,/No to BLER mapping alone. From this mapping also an
estimate of the capacity can be found. Hence, when different TP,,,, = Npucker. R . [ 1 - B L E R ( E b / N o ) ], (2.2)
coding schemes (e.g. coding rate, block length and interleav-
where BLER is the block error rate. A block is defined to
ing depth) and power control strategies (using fast PC or not)
be the smallest unit that is re-transmissible on the air-inter-
are compared, the best scheme (with respect to system capac-
face.
ity) can be found from the E,/N, to BLER mapping alone.
The interference I in a cell is proportional to the number of
Section 11 derives the capacity formula. Section I11 briefly
users in a cell times the carrier energy C, so
. .. . .
- -w. 1 -BLER(Eb/No)
k
lo+
3 4 5 6 7 8 9 10 11
Comparing (2.1) with (2.4) shows that the main difference Eb/NO [dBl
is the factor [ 1 - B L E R ( E b / N o ) ] coming from the retrans- Figure 1. The Block Error Rate (BLER) as a function of E/N0 for the 32
kbps data channel in the CDMA test system. The channel has 4 equally
missions. Also note that in (2.1) we must derive the E b / N , strong Rayleigh fading taps, and the mobile speed is 6 d s .
that gives acceptable speech quality. In (2.4) the quality is Perfect first PC
taken into account by the factor [ 1 - B L E R ( E b / N o ) ] . _ _ _ _ - - _No
_ -fast PC
We now define the capacity function as
1on
fcap
We write (2.4) as
i
0 199SiEEE
0-7803-4320-4/98/$5.00 1879 VTC '98-
sponding to BLER = 20%.
~
more than can be compensated for by the lower number of New users were put in a queue. We admitted users from
retransmissions. Queueing the queue one by one as long as the BS befim admitting
Finally we compare the maximum of f c U p for the two strategy 1 the user was capable of giving the admitted users the
target E@@ (A new user that was put in an empty
cases. This shows that using PC according to this theory
queue and immediately admitted was not considered
should give delayed in the queue.)
Information 32kbps
Eb/No,rarget, namely E b / N o , t , ,,
= 5 dB, 6 dB and 7 dB.
In this particular simulation the ofiered traffic was 18 Erlang
data rate
(corresponding to 115 kbps/MHz/cell) and fast PC was used.
Block length 10 ms = 320 bit = 40 byte Each simulation started with an empty system, and run for 1.2
seconds (i.e. 120 time steps). We see that the system was sta-
Packet length 10 blocks = 100 ms = 3.2 kbit = 400 byte
ble only for E,/N,,,,,, = 6 dB. Both 5 dB and 7 dB
Poisson distributed. Each user was born, transmitted resulted in growing queues.
process one packet containing 10 blocks, and died. 2500
Simulation
time
1.2 seconds simulated. Statistics collected between 0.2
and 1.2 seconds. I 0 0.5
time [SI
1 0 05
time [SI
1 0
Figure 3. The number of mobiles (admitted +in the queue) in the system as
05
time [SI
1
/irl/i!r
The large queueing time for 5 dB (lower left figure)
depends on that the system accepts too many mobiles at the Eb/N I target [dB1 Eb/N I target [dB1
same time, resulting in many retransmissions. These retrans-
missions in turn result in low throughput, and therefore also Figure 5. The delay (in blocks) as a function of E@o,,urRrt for the case that
long queueing time. We conclude that 5 dB is a too smail tar- fast PC is used. The left figure. is for using queueing strategy 1 , the right for
get value. using queueing strategy 2. Stable cases (no growing queues) are marked with
3 0
4000 0 0 r 1 rings and connected by solid lines. Unstable cases (growing queues) are
marked with crosses. (Note that the finite value of the delay in the unstable
case is solely due to the finite simulation time).
The different curves are for 16, 17, 18 and 19 Erlang offered traffic.
2000
5 dB 6 dB 7 dB N o Fast PC N o Fast PC
2000 2000 2000 Queueing strategy 1 Queueing strategy 2
O0 Ll 10 20 ou
ou 0 10 20 0 10 20
I Fast PC,
Queueing 1 6 - 6.5 1 6 I 4.4
[Erlang per cell]
Spectrum efficiency
[kbps/cell/MHz]
122
I
strategy I
0.240
f f"Plmax
Queueing 6 4.4
1
I
strategy 2 f I
c t l p lmux
k = 1.97
spectrum efficiency
TABLE 111 The DL capacity (offered traffic per cell),L.a,,,mu and k. Wis
No Fast PC, 5 MHz. Remember that the system simulation was performed with non-
Queueing 6-7 20 - 36 orthogonal DL.
Figure 8. The percent satisfied users as a function of for the 2 VIII. Acknowledgments
Mbps packet data service in [3]. The left figure shows the UL and the right
shows the DL. Note that the maximum is derived close to where the capacity Maria Gustafsson at Ericsson Radio Systems AB is very
function fcUp in Figure 7 has its maximum. much acknowledged for providing the link simulation results.
IX. References
Figure 8 shows the percent satisfied users (a user is satis-
fied if the mean session transmission speed is over 200 kbps, [I]Gilhousen, K., Jacobs, J, Padovani, R., Viterbi, A., Weaver, A. Jr. and
see [3]) as a function of the target E b / N o for the 2 Mbps Wheatley, C. 111: On the Cupucify of a Cellular CDMA System. IEEE
Transactions on Vehicular Technology, Vol. 40, No. 2, p 303-312, May
packet data service. For the UL (left figure) the offered load 1991
was 290 kbps/MHz/cell. For the DL (right figure) the offered
load was 490 kbps/MHz/cell. (The DL BLER mapping is [2]Maria Gustafsson, Ericsson Radio Systems AB. Personal communication.
almost identical to the UL BLER mapping, and this particular [3]UMTS 30.03, version 3.0.0: Selection procedures for the choice of radio
DL simulation was performed using the same mapping as in transmission technologies of the Universal Mobile Telecommunications
the UL.) We see that in both cases around 99% of the users Systems (UMTS).
are satisfied (ETSI, [3], only require 98%). [4]Tdoc SMG2 UMTS 11 1197: Concept Group Alphu- Wideband Direct-
The most interesting observation for this paper is derived Sequence CDMA (WCDMA). EVALUATION DOCUMENT (2.0).
by comparing the best Eb/NO,target found in the system
simulations to the theoretically best E b / N O , t a r g e t . For the
UL they were the same, and for the DL the best value found
by simulations was only 0.5 dB larger than the theoretical
value.
VII. Conclusions
We have derived a capacity formula for a CDMA packet
data ARQ system. The formula gives the capacity as a func-
tion of the E b / N o to BLER mapping (found in link simula-
tions), the used ,!?,/No, the used band width W and a
constant k that basically depends on the reuse and on the path
loss function. The formula can be used for finding the optimal
value of Eb/NO,rarger.If k is known, the system capacity in
terms of information bits per time and cell can be found. Even
if k is not known, the formula makes it possible to optimize
the coding, interleaving and block size with respect to system
capacity by only performing link simulations. The system
capacity gain derived by using fast PC compared to only
using slow PC (not following the Rayleigh fading) can also be
derived by only performing link simulations. System simula-
tions are not necessary for these purposes.
System simulations aiming for confirming the theory has
been performed. Comparing the theory and simulations