A Comparison of CDMA and TDMASystems
`Bjorn Gudmundson, Johan Skéld and Jon K. Ugland
`Ericsson Radio Systems
`§-164 80 Stockholm, Sweden
`
`Abstract
`
`In this reports two candidates for high capacity cellular
`systems are simulated and analysed, one CDMA and one
`TDMAsystem. Simulations of the CDMA example in-
`dicate a high sensitivity to variations in certain system
`parameters.
`The TDMA example is a GSM system using ran-
`dom frequency hopping and operating without frequency
`planning. The simulations show that the TDMA system
`has at least the same capacity as the CDMAcandidate.
`Soft capacity, efficient use of voice activity and diversity
`are features available in both systems.
`
`1
`
`Introduction
`
`CDMAhas been introduced as a candidate for high ca-
`pacity cellular systems [1, 3]. Another atractive alter-
`native is to refine todays TDMA systems already in
`operation into high capacity cellular TDMA or hybrid
`TDMA/CDMAsystems for the future.
`Introduction of dynamic channel allocation has been
`suggested as an evolution of TDMA[4, 5]. Anotheralter-
`nativeis the frequency hopping TDMAsystem presented
`below. It can be seen as a TOMA/CDMAhybrid.
`In this report, the issues of system capacity, flexibility
`and operational features are investigated for the CDMA
`system example and a comparison is made with the fre-
`quency hopping TDMAsystem.
`
`2 The CDMAsystem
`A description of the CDMA system is given in [3]. A
`direct sequence spreading technique is used, spreading
`the R = 8 kbps user data over the bandwidth W =
`1.25 MHstrough a low-rate convolutional encoder and a
`Walsh-Hadamard transform.
`Scrambling the data with PN-sequences provides
`decorrelation of the different users. The result is that
`the interference from all co-existing channels will add to
`a noise-like interference which is efficiently supressed in
`the despreading process.
`The wide channel bandwidth allows a high resolution
`when extracing multipaths. Thus, a high degree of path
`diversity is availiable. For complexity reasons however
`only a few paths will be resolved. The remaining paths
`have to be supressed.
`
`The analysed system is specified with a re-use fac-
`tor of 1, ie.
`the same frequency can be re-used in all
`base stations (BSs). This configuration is possible by
`the inherent interference diversity. Since the co-channel
`interference will be the average interference from several
`mobile stations (MSs), the worst case interferer will no
`longer determine the re-use factor as it does in conven-
`tional TDMA- and FDMAschemes.
`For a CDMA system to work, the users sharing the
`same carrier must be received with equal power levels.
`Otherwise some users will jam the others. This makes
`heavy demands on the up-link design where the power
`control must have a wide dynamic range to compensate
`for the near/far effect. The power control must also be
`very fast since it has to compensate the multipath fad-
`ing for slowly moving MSs. For fast moving MSsthis is
`not necessary since the interleaving and coding probably
`provides sufficient quality in this case.
`In [3] BS controls the MS power trough a closed loop.
`Every millisecond a commandis transmitted to the MS
`demandingeither an increase or a decrease by 0.5-1.0 dB
`of its power level. The algorithm is claimed able to track
`Rayleigh fading for vehicle speeds up to 25-100 mph.
`Power control in the down-link is not as critical as in
`the up-link. Here all signals propagate along the same
`path, naturally giving balanced signal levels. It may how-
`ever be used to increase capacity.
`A study of the performance and capacity of the
`CDMAsystem is madein [1] and [2]. In both papers ana-
`lytical expressions are used for the capacity calculations.
`Here, Monte-Carlo simulations are performed, allowing
`the introduction of hand-off margin, power controllimi-
`tations etc. The evaluationis limited to the up-link since
`this is the mostcritical connection in the system.
`A system comprising one local BS and three rings of
`interfering BSs is simulated. A hexagonal omni-cell pat-
`tern is assumed, and the MSs are randomly spread with
`uniform distribution in a circle having an equivalent area
`of the 37 hexagons. For each MS the power attenuation
`to each of the BSs are calculated. A distance dependent
`path-loss r~* is used where @ is the propagation expo-
`nent. The shadowing effects are modelled by indepen-
`dent, log-normally distributed random variables having
`a standard deviation of 8 dB. Hand-off is made to the BS
`resulting in the least attenuation.
`In each simulation the C/I-level for the MSs con-
`
`
`
`732
`
`0-7803-0673-2/92 $3.00 1992 IEEE
`
`arn
`
`APPLE1018
`Page 1 of 4
`
`APPLE 1018
`Page 1 of 4
`
`

`

`
`
`OutageprobabilityPo
`
`101 35
`
`10
`
`is
`
`20
`
`25
`
`E [users/cell/MHz}
`
`
`
`CumulativeDistribution
`
`101f- 102-1
`
`Cf [4B]
`
`Figure 1: CDMA outage probability vs spectrum effi-
`ciency. Right: Ideal. Middle: canging channel activity to
`9 = 50%. Left: adding handoff margin Apo = 6 dB.
`
`Figure 2: CDMA C/J-distributions in Rayleigh fading
`environment with max MS powerlimits »y = 5, 10 and 20
`dB.
`
`path loss exponents mainly between a = 3 and 4, but also
`values as low as 2 and as high as 5 occured. Changing
`the path loss exponent from 4 to 3 in the simulations
`gives a capacity reduction of 20%.
`In deep fading dips the MSs must dramatically in-
`crease their output power levels in order to compensate
`the multipath fading. If a MS is located close to a neigh-
`bouring BS this may cause severe interference to the MSs
`at that BS. In the simulations a Rayleigh distributed fad-
`ing is assumed. It is modeled by scaling the signals with
`independently Nakagami-2 distributed random variables
`corresponding to ideal, two-antenna diversity. All MSs
`will perfectly handle the fading, i.e. move at speeds less
`than 25 mph. However, parts of the deepest fading dips
`will remain due to an upper limiting of the MS output
`power. C/J-distributions for the upper limits v = 5,10
`and 20 dB are given in Figure 2. For the most realistic
`power limit, v = 10 dB, the C/I performanceis reduced
`by 2 dB compared to a non-fading environment. Trans-
`lated to capacity this is a reduction of 45%, implying
`that power control of the Rayleigh fading may cause se-
`vere problems to the system.
`Perfect power control has been assumed in the capac-
`ity calculations in [1]. Here long-term power controler-
`
`nected to the local BS is calculated. Here C is the re-
`ceived power from the desired MS and I is the sum of
`undesired powers from all MSs sharing the same band-
`width. C'/I-values for MSs not connected to the local BS
`are discarded due to boundary effects. Repeated simula-
`tions give a C/I-distribution.
`The capacity is constrained by the minimum accept-
`able transmission quality BER < 10-3 after decoding.
`In [1] this is claimed to be reached at E,/No > 7 dB. As-
`suming neglegible background noise this is directly trans-
`lateable to the minimum C/J-level
`
`_ R-(Es/No)
`(1)
`=e
`which for the vocoder bitrate R = 8 kbps is y = —14.9
`dB. The probability of having a C/I-level below 7 is
`called the outage probability Po and is found from the
`C/TI-distribution. Note that outage implies that ali MSs
`connecetd a the BS have unacceptable transmission qual-
`ity.
`
`Simulated values of Po vs different values of spec-
`trum efficiency E (users/cell/MHs) are given in Figure1.
`The rightmost curve yields the parameter values used in
`[1]. For Po = 1% the capacity is found to be FE = 29
`users/cell/MHz which gives 36 users/cell according with
`the result in [1]. However, this capacity figure is achiev-
`able only under idealized conditions. Choosing slightly
`more pessimistic parameter values reduce performance
`significantly.
`The voice activity factor of 9 = 37.5% claimedin [1]
`seems very optimistic.
`In [2] the value # = 60% was
`used instead. Here 9 = 50% has been used giving the
`expected capacity reduction of 30% shown in Figure 1.
`The choice of propagation parametera is known to
`depend heavily on geography. Own measurements gave
`
`Table 1: Relative capacity degradation for variations in
`CDMAsimulation parameters
`Parameter:
`
`Voice activity #
`
`Multipath fading
`HO-margin Ayo
`Power control cerr
`
`Path-loss « Capaci
`
`Test
`value
`
`change
`
`a P
`
`age 2 of 4
`
`733
`
` Page 2 of 4
`
`

`

`rors are included by scaling the MS transmit-powers with
`an independent,sero mean, log-normally distributed ran-
`dom variable, having standard deviation o.,. Simula-
`tions show that the degradation is large. For the mod-
`erate error value err = 1 dB the capacity is reduced by
`35%, demonstrating an extremely high accuracy needed
`in the power control to maintain operation.
`To prohibit too frequent hand-offs, a hysteresis has
`commonly been used in cellular systems. This has been
`obtained by prohibiting the hand-off until a BS occures
`that gives more than AyodBreduction in path loss. Soft
`hand-off is an alternative to using hand-off margins, how-
`ever, this has not been a prefered solution since connect-
`ing the MSs to more BSs simultaneously results is a dra-
`matic increase in network complexity. The leftmost curve
`in Figure 1 shows Po vs E for the CDMA system using
`hard hand-off with Azo = 6dB. A capacity reduction of
`40% is observed compared to the case of ideal hand-off
`implying that soft hand-off is mandatory in this system.
`Table 1 is a summary of the sensitivity to parameter
`variations. All capacity changes cannot be added, but
`the total picture is that the optimistic capacity claims
`in (1] are unrealistic. Accounting for the reduced voice
`activity of 9 = 50% and including the effect of multipath
`fading gives the capacity F = 12 users/cell/MHs. This
`can be compared to the AMPS analog 30 kHs system
`which with a 21 re-use frequency plan has the capacity
`1.6 users/cell/MHsz. Thus, a realistic CDMA capacity
`gain would probably be somewhatless than 10 compared
`to AMPS.
`
`3 Frequency Hopping TDMA
`Comparisons between CDMA and TDMA systems have
`been made by several authors (2, 4, 6].
`that
`[2] suggest
`Results reported by Baier et.al.
`TDMA and CDMA system capacity is similar. They
`proposed a TDMA system using slow frequency hopping
`to provide interferer diversity [6]. The cluster size K was
`varied continously to simulate a continous set of spec-
`trum efficiency values E. The interference level in [2] was
`calculated as the average interference over all frequencies
`used for hopping,i.e.
`T= external = 3 Vk
`
`1
`
`i
`
`(2)
`
`is the number of frequencies used for fre-
`where Ny;
`quency hopping and J; is the interference contribution
`from the ith external user. The system parameters used
`were taken from the GSM-system with a half-rate speech
`codec, see Table 2.
`Here we will simulate TDMA in the same way but
`with a fixed cluster sise K = 1, i.e. all frequencies are re-
`used in all sites and no frequency planning is required.
`By varying the subscriber load, different values of F can
`be simulated. Random frequency hopping is performed
`over N;=8frequencies. The frequency hopping patterns
`
`
`
`
`
`15
`
`2
`
`25
`
`35
`30
`B [users/cell/MHz]
`
`“0
`
`45
`
`50
`
`Figure 3: TDMA and CDMAoutage probability vs. spec-
`trum efficiency.
`
`within the cell are strictly orthogonal, while the patterns
`are uncorrelated between the cells. Power control is used
`to achieve equal received powers at the site.
`Each BS serves on the average N, users per TDMA
`time slot, where N, < Ny. N, determines the spectrum
`efficiency E in users/cell/MHs:
`
`N,M
`(3)
`E= x.W
`where M is the number of TDMAtimeslots and W is
`the bandwidth, see Table 2. The maximum available
`capacity of a BS is F = 80 users/cell/MHs.
`The simulation model is in al] other aspects equiva-
`lent to the one used for the CDMA simulations, including
`the ideal power control used. Up-link C/I-distributions
`are calculated and used to derive the outage probability.
`Theresult is plotted in Figure 3 together with the middle
`CDMAcurve from Figure 1. The TDMA system capac-
`ity is 19 users/cell/MHz at 1% outage probability, while
`the corresponding CDMA figure is 21 users/cell/MHz.
`This shows that TDMA using 8 frequencies for interferer
`diversity has the same capacity as the CDMA system
`in [1]. The conclusion is confirmed in [2].
`The GSM recommendation specifies a TDMA system
`with options for random frequency hopping and adaptive
`
`
`
`TDMA multiplex factor
`No. of frequencies
`
`User bit rate
`Bandwidth
`
`C/I limit [2]
`
`Channel activity
`
`HO-margin
`
`734
`
`|
`
`i
`
`i
`
`Page 3 of 4
`
` Page 3 of 4
`
`

`

`powercontrol. This indicates that one-cell frequency re-
`use could be possible in GSM. Further simulations of
`the GSM air interface operating in frequency hopping
`conditions will be needed to verify this.
`
`cro cells. This is a disadvantage in CDMA if mobile
`assisited hand-off is used, since transmission is continous
`and no timeslots are available for measurement such as
`in TDMA.
`
`4 System Comparison
`
`The two system proposals presented above have approx-
`imately equal spectrum efficiency under the assumptions
`made. Below is a comparison of other system aspects
`that should be considered.
`A TDMAsystem with random frequency hopping and
`a re-use factor of one gives interferer diversity much in
`the same way as CDMA does. The interference power
`will be the average from manyinterferers instead of a few
`possibly large ones. Interferer diversity makes it possible
`to exploit channel activity to get a capacity increase.
`Frequency diversity in the CDMA system requires a
`multi-tap Rake receiver to exploit the wide bandwidth.
`In the TDMA system, the random frequency hopping
`inherently provides frequency diversity. Other diversity
`schemes such as antenna diversity, transmitter diversity
`and path diversity can be implemented in a TDMAsys-
`tem as well as in CDMA. The path diversity in TDMA
`can be provided by an equaliser.
`In the CDMAsystem, a tight up-link power control
`is essential for operation. The TDMA system uses power
`control to reduce power comsumption in the mobiles and
`to gain capacity, but it is not as critical as in CDMA.
`A CDMAsystem does not have a hard capacity limit.
`Thisis also true for the TDMA-proposal sinceit only uses
`10-30% of the frequencies available at full capacity. Fig-
`ure 3 shows that moresoft capacity is available in TDMA
`since the outage probability increases more slowly.
`The soft capacity also makes TDMAveryflexible to
`unequal cell loading. A number of cells forming a line
`can be loaded up to 200% of the normal capacity if the
`surrounding cells have a 65% load. The high flexibil-
`ity is due to the internal interference always being zero.
`In CDMA,the interference is mostly internal making it
`harder to increase capacity at peaks, since the cell will
`jam itself. A line of cells in the CDMA system can be
`loaded to 120% of nominal capacity, making it necessary
`to decrease the surroundingcells to 30%.
`In the simulations presented in Section 3, 8 frequen-
`cies are used for frequency hopping.
`If the number
`is increased and averaging of interference can be per-
`formed, the system capacity will increase above that of
`the CDMAreference system. A TDMA system with a
`specific Ny will always have a higher spectrum efficiency
`than a similar CDMA system with the same interference
`diversity factor (processing gain). The reason is that the
`users within each cell are orthogonal i TDMA [6].
`In a cellular system including both micro and macro-
`cells, MSs in the macro cells will use higher power lev-
`els and create high interference in the micro cells. The
`solution is to use different frequencies in macro and mi-
`
`5 Conclusion
`
`This report points out some strong and weak points in
`CDMA bysimulation of a system example. Theresults
`indicate that performance can be severely degraded when
`assuming non-ideal conditions. A comparison is made
`with a frequency hopping TDMAsystem.
`Simulations show that the capacity of the TDMA sys-
`tem is equal to or even better than in the CDMA example
`in ideal conditions.
`It also has most of the advantages
`that CDMA has,concerning e.g. frequency diversity and
`use of voice activity. The soft capacity properties of the
`TDMAsystem is superior to CDMA, making operation
`in inhomogeneouscell loads moreefficient.
`The conclusion is that when comparing CDMA and
`TDMAas multiple access schemes, capacity is not the
`only issue. The main pointis how to exploit the capacity
`potential and achieve the system features wanted.
`The overall picture is that a very careful system de-
`sign is necessary to utilize the full potential of CDMA as
`well as TDMAsystems. Further simulations are needed
`to verify the properties of both candidates.
`
`References
`
`[2=
`
`Jacobs, R. Padovani,
`I. M.
`{1] K. S. Gilhousen,
`A. J. Viterbi, L .A. Weaver, Jr., C. E. Wheatly,
`“On the Capacity of a Cellular CDMA System”.
`IEEE Transactions on Vehicular Technology, Vol.
`40, No.2, May 1991, pp. 303-312.
`A. Baier, W. Koch, “Potential and Limitations of
`CDMAfor 3rd Generation Mobile Radio Systems”,
`MRC Mobile Radio Conference, Nice, France,
`November 1991.
`[3] A. Salmasi, K. S. Gilhousen, “On the System Design
`Aspects of Code Division Multiple Access (CDMA)
`Applied to Digital Cellular and Personal Communi-
`cations Networks”, Proceedings of IEEE Vehicular
`Technology Conference, VTC-91, St. Loius, USA,
`May 1991, pp. 57-63.
`[4] B. Gudmundsson, “A Comparison of CDMA and
`TDMASystems”, COST 231 Panel Meeting on Ra-
`dio Subsystem Aspects of CDMA for Cellular Sys-
`tems, Berne, Switzerland, October 1991.
`K. Raith, J. Uddenfeldt, “Capacity of Digital Cel-
`lular TDMA Systems”, JEFE Transactions on Ve-
`hicular Technology, Vol. VT-40, No. 2, May 1991,
`pp. 323-332.
`M. Mouly, “The Compared Capacity of CDMA and
`TDMA systems”, COST 281 Panel Meeting on Ra-
`dio Subsystem Aspects of CDMA for Cellular Sys-
`tems, Florence, Italy, January 1991.
`
`{5=
`
`[6—
`
`735
`
`+=
`
`Page 4 of 4
`
` Page 4 of 4
`
`