Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

DS/CDMA SUCCESSIVE INTERFERENCE CANCELLATION

Jack M. Holtzman
WINLAB, Rutgers University
P.O. Box 909
Piscataway, NJ 08855-0909, USA
holtzmana winlab3.rutgers.edu

ABSTRACT upwards from the conventional detector. That is, we


consider what is believed to be the simplest augmentation
ConventionalDS/CDMA detectors operate. by enhancing a to the conventional detector, a form of successive
desired user while suppressing other users. considered as interference cancellation. We will try to provide a
interference (multiple access interference. MAI) or noise. perspective on such an augmentation.
A different viewpoint is to consider other users not as
noise but to jointly detect all users' signals (multiuser The earliest paper on DSCDMA multiuser detection is
detection). This has significant potential of increasing probably [2]. Another early paper, 131, specifically
capacity and near/far resistance. Optimal multiuser discusses multiuser detection in terms of cancellation.
detection is, however, too complex to implement, thus Key theoretical papers (with earlier references) are [4], [5].
motivating the search for suboptimal algorithms. Our There have been a number of other papcrs on interference
objective is to underline the need for simplicity and to cancellation, e.g. [6-17] (not an exhaustive list).
discuss what is a relatively simple form of multiuser
detection, successive interference cancellation. The The paper is organized as follows. Section I1 places the
cancellation scheme uses only components already present candidate for multiuser detection at the base station in a
in a conventionaldetector. cellular system and Section Ill places bounds on the
improvements to be expected. Section IV motivates the
need for multiuser detection. Section V discusses issues
L UVTRoDUCT ION associated with the successive interference cancellation
scheme. Concluding remarks are given in Section VI.
ConventionalDS/CDMA detectors operate by enhancing a
desired user while suppressing other users, considered as
& MULTIUSER DETECTION IN CELLULAR
interference (mulriple access interference. MAI) or noise.
A different viewpoint is to consider other users not as SYSTEMS
noise but to jointly detect all users' signals (multiuser
detection). This has significant potential of increasing In a cellular system, a number of mobiles communicate
capacity and nearlfar resistance. Since another paper at with one base station (BS). Each mobile is concerned
this conference will extensively discuss multiuser only with its own signal while the BS must detect all the
detection [I]. we will only give enough background on signals. Thus, the mobile has the information only about
multiuser detection here that is needed for our objective. its own chip sequence while the base station "knows" all
Our objective is to underline the need for simplicity and to the chip sequences. For this reason, as well as less
discuss what is a relatively simple form of multiuser complexity being tolerated at the mobile (where size and
detection, successive interference cancellation. weight are critical), multiuser detection is currently being
envisioned mainly for the BS. or in the reverse link
As mentioned, multiuser detection has significant (mobile to BS). It is important to realize, however, that
potential for performance improvement. It is, however, the BS maintains information only on those mobiles in
much too complex to implement. Much effort is its own cell and there are mobiles outside the cell causing
currently in progress on developing suboptimal detectors interference. This plays a role in the limitations on
which are not too complex to implement. The key improvements to be expected in a multiuser detection
question is: Is there a suboptimal version that is system. to be discussed next.

(i) Cost effective to build in a practical system and The next section on limitations on improvement from
which multiuser detection underlines [he importance of
(ii) Still retains enough advantage over the maintaining simplicity.
conventional detector?

Towards answering this question, the orientation of the m.UMITA TlONS TO IMPROVEMENTS
paper is as follows. One cap look at the optimal
multiuser detector and successively make simplifications Before we discuss multiuser improvcments to the
until implementabilityis attained. Instead of working our conventional DS/CDMA detector, it is important to define
way downwards from the optimal. we discuss working factors that limit such improvement. One factor is

69
intercell interference in a system that cancels only the performancelcost uadeoff ;idvantage to multiuser
inuacell interference I . For intercell interference which is detection.
a fraction f of the inuacell interference, the bound on
capacity increase (with all of the intracell interference The bottom line is that there are significant advantages to
canceled) is (1 + f )/ f. For f = 0.55, this factor is 2.8 multiuser detection which are. however, bounded and a
simple implementation is needed.
[IS].
Another factor is the fraction, f,,of energy captured by a
1y, OPTIMAL AND SUBOPTIMAL MULTIUSER
Rake receiver. ?hat is. a Rake receiver with L branches or DETECTION FOR A SIMF'LIFIED DSKDMA
"lingers" will uy to capture the energy in the L strongest
SYSTEM
multipath rays but there will be additional received energy
in additional rays. For the conventional detector, this is
self-interference. Ref. 19 gives examples of the fraction We will explain the basic problem with a very simplified
of captured energy. The fraction of captured energy is a DSlCDMA system. There are a number of
function of chip duration times the number of Rake simplifications which will bc relaxed in the rest of the
branches divided by the delay spread. article. In facc each relaxation of simplification will
represent another factor to consider for the multiuser
detection system. Suppose there are K users sharing the
So, combining the two effects (measured by f and fc),
same bandwidth and each user's baseband signal is
the total interference before cancellation is (1 + f ) l
(neglecting the smaller self-interference due to uncaptured i ~ k ( t ) = A k b , , ~ k ( l - ; Ti )7,S f < ( i + 1 ) 7
multipath power of the desired user). Cancellation
removes at most J c I so the bound on improvement is (IV - 1)
(I+f)l(l-f,+f). Depending on the system
panmeters, f, can vary from near 1 to substantially less where T is the bit interval and Ak, 6*, a k ( l ) are the
(see [19]). For f, = 1, the above bound on capacity amplitude, bit sequence, and spreading chip sequence (or
signature waveform, or code), respectively, of user k. The
improvement of 2.8 remains. For f,=0.5, the bound is
b, are t l and the objective is to detect those polarities,
reduced to 1.5. Note that a small f, also presents a which contain the transmitted information. The received
problem to the conventional detector. signal (at baseband) is the sum of all the users' signals
plus noise:
It should also be recognized that interference cancellation K
is used not only to increase capacity but also to alleviate r(t)= Cuk(t)+z(l)
the nearlfar problem, and the preceding hound does not k=l
account for that benefit. This actually translates into a (IV - 2)
capacity benefit which is, however, more difficult to
quantify than by the above simple SignaVinterference where z ( f ) is the additive noise. Note the simplifications
argument. An interference cancellation scheme could which need to be relaxed:
recapture part of the reduction due to received power
variability by reducing the variability (or it could be used (i) Synchronous reception: all of the users' bits are
to relax the requirements on power control). aligned in time.
(ii) Coherent reception: all of the relative phases are
To put these constraints on improvements into further zero (coherent reception only requires knowledge
perspective, we are assuming here that multiuser detection of the phases, not that they are Uie same).
is a candidate primarily for the reverse link for reasons (iii) The amplitudes are constant.
given in Section 11. Since the reverse link is usually (iv) No multipath.
more limiting than the forward link [20], increasing the
reverse link capacity will improve the overall system The conventional DSlCDMA receiver detects the bit from
capacity. B u t increasing it beyond the forward link user j by correlating the received signal with the chip
capacity will not further increase the overall system sequence of user j . Since the system is synchronous, we
capacity. Thus, can isolate each bit interval. Over the interval [0,T I , the
decision statistic is:
(i) The potential capacity improvements are not
enormous (order of magnitude) but certainly non-
trivial and worth pursuing.
(ii) Enormous capacity improvements only on the
reverse link (the candidate for multiuser detection)
would only be partly used anyway in determining
overall system capacity.
(iii) Hence, the cost of doing multiuser detection
must be as low as possible so that there is a

70
when: it is assumed that the integral of a:(r) is unity. over all components of the bit vector, (IV-7) searches ovcr
one component at a time. Decisions for the other bits arc
The decision on the sign of bo, is given by the sign of used from a previous stage of processing. This leads to
two issues for multistage detectors:
y o j . Note that yoj consists of three terms. The first is
the desired information which gives the sign of the (i) How to choose the initial stage.
information bit boj which is exactly what is sought. The (ii) How to choose the subsequent stages of
second term is the result of the multiple access processing.
interference, and the last is due to the noise. The second
term typically dominates the noise so that one would lie Different alternatives will lead to different suboptimal
to remove its influence. Its influence is felt through the algorithms discussed. One alternative is to use the
cross-correlations between the chip sequences. Using conventional detector and another is the decorreloror
knowledge of the chip sequences(available at the BS). one derecfor which we now motivate. Inspection of (IV-5)
could cancel the effect of one user upon another. This is. immediately suggests a method to solve for b . whose
in fact, the intuitive motivation for interference components boj contain the bit information sought. If 2
cancellation schemes. It is important to also note that the was identically zero; we have a linear systcm of equations.
factors A,bok are also needed. These can be obrained by y = RAb, the solution of which can be obtained by
either multiplying separate estimates of A, and bo, or by inverting R (assuming i t is invertible) sincc A is
an estimate of the product itself. This choice will be diagonal. With a non-zero noise vector ?. inverting R is
further discussed below. still an effective procedure (and actually optimal in cercain
reslricted C i " a n C e S ) . This results in
For the simplified problem discussed above, it may be
shown that the maximum likelihood decision for the R-ly = Ab+ R-'?
vector of bits, b , is given by
(IV - 8)
max [Zy'b-b'RAb]}
y=arg{ b~{-I,+l)'
where it is seen that the information vector, b , is
recovered in the first term but conkminated by a new
noise term (which may be enhanced).

where Consideration of alternative initial and subsequent stages


y = RAb+i of processing is given in [22].

(N- 5 ) Y. SUCCESSWE INTERFERENCE


CANCELLATION
where 2 is the noise term after correlation, A is a diagonal
matrix of amplitudes and R is the matrix of
crosscorrelations. We have seen there can be a number of altematives for
suboptimum multiuser detection. They have different
comparative performance and complexity. What they all
have displayed is improved performance and increased
complexity compared to the conventional detector. We
shall discuss now what is probably the least complex of
the schemes but which still retains superiority over thc
(N- 6) conventional detector.
The exponential complexity motivates the need for
suboptimal approaches. Multistage approaches arc As mentioned, most of the schemes could be explained in
common. One, given in [21], uses (IV-7) instead of (IV- terms of an initial stage of processing followed by one or
4).
more stages of processing of a different type. The initial
stage is typically a conventional detector or a
decorrelating-type detector. The simpler initial stage is
the conventional detector. A scheme for subsequent stages
max [2yrb-brRAb] based on simplicity is to subtract off the conuibutions of
btEi+l.-I)
the multiple access interference. with the order of Uie
b, = b ( m ) , V f t l
subtractions being given by the relative suength of the
users. Contributing to its simplicity is the flexibility to
(N- 7) limit the number of cancellations.
In (IV-7). an estimate at stage ( m + l ) is made using It is most important to cancel the strongest signal before
estimates at stage m . Comparing the two equations, one detection of the other signals because it has the most
sees that while (IV-4) requires the simultaneous search

71
negative effect. Also. the best estimate of signal strength coherent system, it is not, but the information needed for
is from the strongest signal for the same reawn that the the cancellation can be obtained from I and Q channels.
best bit decision is made on that signal: tbe strongest
signal has tbe minimum MAI since the strongest Si@ is The successive cancellationsare carried out as follows:
excluded from its own MAI. Thii is the twofold rationale
for doing successive cancellation in order of signal (i) Recognize the strongest user (one with
strength: maximum correlation value).
(ii) Decode the strongest user.
(i) Canceling the strongest signal has the most (iii) EstimaCe Le amplitude of the decoded user from
benefit the output of the correlator.
(ii) Canceling the strongest signal is the most (iv) Regenerate strwgest user's signal using its chip
reliable cancellation. sequence and lhe estimate of its amplitude.
(v) Cancel the strongest user.
Ranking the Signal Strengths (vi) Repeat (until all users are decoded, or until a
permissible number of cancellations are
Before the successive cancellation can proceed. it is achieved).
necessary to rank the signal strengths. They could be
obtained by separate channel eslimates or they can be The received signal is:
obtained directly from the outputs of the conventional
detector. The sign of the output of the correlator for each K
user provides a bit decision for that user. The amplitude of r(t)= ~ A , b , ( f - ? , ) a , ( f - ? , ) ~ o s ( t v , l + ~ ~ ) + : ( J )
,=I
that correlator output also provides an estimate of the
received signal strength (from the squareof tbe amplitude)
of that user. There is noise associated with this estimate (V - 1)
but we shall find its accuracy sufficient. The channel Where
estimate method corresponds to using separate estimates
of A, and bo, while using the correlator output r(t) - received signal
corresponds to using an estimate directly of the product K - Total number of active users
A, bo, as discussed in Section 111. For specificity in the A, - Amplitude of kIh user
rest of the article. we shall assume that the rankings are
based on the correlator outputs.' This is also consistent 9 ( r ) - bit sequence of krh user
with hying to maintain simplicity -- no components are a, ( t ) - spreading chip sequence of k" user
needed beyond what is already provided for the
conventional detector. n(t) - Additive White Gaussian Noise (two sided
power spectral density = No I 2 )
Successive Cancellation with a Coherent
RPSK System
N = TI T, where, T - bit period and T, - chip period
We shall fmt describe the method for a coherent BPSK
system to fix ideas. These results will be unrealistic for
the following reasons: ek
r, and are the time delay and phase of the kIh user,
which are assumed to be h o w n , i.e., tracked accurately.
(i) A non-coherent system would be used on the (Both these assumptions will be relaxed.)
reverse link because of the lack of a pilot signal
used for a coherent signal [201. The bits and chips are rectangular. Their values are all
(ii) Fading and multipath an?not included. i.i.d. random values with probability 0.5 of f l . The 7,
and are i.i.d. uniform random variables in [O.T] and
Each of lhese factors can be introduced after the basic idea [0,2 X I , respectively, for the asynchronous case.
is illustrated. One important point can be made at this
point regarding non-coherence. To do the cancellation, It will be assumed that the users are labeled from 1 to K
phase information as well as amplitude information (to in accordance with their estimaled signal strengths.
detect relative strengths) is needed. In a coherent system,
the phase informalion is directly available. In a non- At the output of the low pass filter (LPF) of the I-
Channel, we get:

*Refs.[l4] - [17] provide analytic results for using the


correlator outputs for amplitude estimates. which form the
basis of thc present paper. Ref. 7 uses the despread signal lo
select the strongest and Ref. 12 also uses the despread signals
for amplitude estimates. Channel estimates are suggested in
[lo], [ 111 which uses a novel method of cancellation in the
spectral domain. (\' - 2)

12
where n,(r) is the in-phase component of the low pass The Gaussian approximation used in the above was found
to be sufficiently accurate by comparison with
filtered Gaussian noise n ( r ) . Similarly d Q ( r ) is simulations for the simple case considered. Additional
obtained. The first decision variable (correlation value) in analysis is needed in other cases, as will be seen.
the Ichannel is given by:

4'- 1
= - j d ' ( l ) U l ( f - T,)COS($,)df
An illustrative result is given in Figure 1. The example
is for equal received powers (ideal power control) which
TT we shall see is a pessimistic case for successive
cancellation. Figure 1 shows the BER averaged over all
(V - 3 ) the users calculated using the Gaussian approximationand
Similarly, f,Qis obtained. comparison with the conventional detector. Substantial
improvement is shown even with equal powers which
The decision on the bit is then made by using the decision does not take full advantage of a scheme based on ranked
variable power.

i,=i,'+ip Another example is given in Figure 2. This is a case of


one weak user dominated by seven olher equally strong
users. We see the performance of the weakest user is
(V - 4) better than the performance of 6 equal users in the
The corrclation value is then used for cancellation: conventional system. Also seen is the effect of the
additive noise, reflected in Eb / N o . .
d: ( I ) = d' ( I ) - &a, ( I - TI) COS($I )
Comparison with Using Channel Estimates
(V - 5 )
As mentioned, in a conventional detector, the output of
Generalizing for the j'* cancellation, we get: the correlator is used for bit detection -- the sign of the
correlator output is taken as the bit decision. We are also
d j( I ) = d;-l ( I ) - i j U j ( I - ?j )cos($j) using the correlator output to obtain an estimate of the
signal amplitude. and are thus making extra use of a
(V - 6 ) quantity already available. It is of interest U) compare the
accuracy of this method of estimation with using separate
where ijis the correlation (decision variable) after the channel estimates. In the above, we assumed that the
amplitude estimate was obtained from only one bit. We
( j - 1)" cancellation.
shall here assume that n successive bits are used for
estimation using the correlator outputs. For the case
As the users are canceled, the multiple access interference studied in [14], it was found that there is a rough
decreases but the noise due to imperfect cancellation in the equivalence in estimation accuracy if the independent
previous stages increases. After j cancellations, the power estimate's accuracy in estimating each bit energy is
signal to noise ratio conditioned on the ordered amplitudes around 1 dB if n = 10. An example of averaging is given
A, (for the asynchronous case) is found in [I41 to be in Figure 3. The averaging improves the amplitude
given by: estimation by a factor of about 1/ A.

Alternate Ways of Implementing

There are two ways of implementing this type of


interference cancellation. One, which has been assumed in
(v - 7) explaining the method. involves identifying the strongest
user and feeding back that user's chip sequence to the
where qi represents noise accumulated from previous (baseband) received signal and then the conventional
cancellation stages. detector is used on all the users who have not been
The bit error probability at the j + 1" cancellation stage, selected as strongest. An alternate method is to feed back
the cross-correlations between the strongest user and each
conditioned on the ordered set of amplitudes A,, is given remaining user and subtracting that off the correlation
by: output of each of those users. It is easily verified that both
are equivalent as far as the equation leading to BER
Pj'l = P{ij+l < Olbj+l =+I} calculation. The latter method is more in line with the
other methods which use these cross-correlations. The
=a(&) implementation implications are currently being
compared.
(v - 8)

73
Synchronous + Asynchronous The pdfs of the ordered Ak (where AIis the strongest
and Ak is the weakest) is denoted by J A , ( x ) and is
In synchronous reception, all the bits are aligned when
received. In the reverse link, in which multiuser detection obtained as follows:
is a candidate, the different users' bit sequences would be
asynchronous.** This presents a problem, the solution of
which provides an added benefit. For synchronous
systems, it is non-ambiguous as to what is meant by
"strongest signal" as all of the users are aligned. For (v - 10)
asynchronous systems, bits from one user overlap
partially with bits of other users. To take into account all The error probability expression after the j" cancellation
of the interactions between all of the bits actually would
require an infinite memory, or a truncation to reasonably is then unconditioned using the pdf of the j + strongest
capture the interactions. A simple heuristic scheme is to amplitude as follows:
group n bits of each user into a cancellation frame, where
[he maximum time between the first bit start and last bit
end is (n + 1) bit durations. After the entire frame is
received, the correlations of the n bits of each user are
averaged and the ranking of the users is obtained from (V - 11)
rhcse averaged correlations. Then the n chip sequences of
the strongest user are canceled. This introduces a delay and where E,,, [ t ~ , , ~ ]represents the equivalent noise.
some small end effects of pans of a chip sequence not
being canceled. The benefit is from the averaging of the The average probability of error is then obtained as the
correlations which, as pointed out before. improves the average of the BER resulting from all stages of
accuracy of the ranking and of the cancellation. cancellation.

Coherent + Noncoherent An interesting question is how the method performs


compared to ideal power control where all the signals are
As previously mentioned, multiuser detection is received with equal energies. There are really two
appropriate on the reverse link (mobile to base station). questions:
On the reverse link, there is no pilot signal for coherent
detection [20]. So non-coherent detection is used, e.g, M- (i) How does the cancellation scheme's performance
ary onhogonal modulation [20]. The basic issue is that in compare under Rayleigh fading or ideal power
order to cancel chip sequences in the coherent syslem, the Control?
phase as well as amplitude is used. Coherent detection (ii) How does the i m p r o v e n i e n l over the
assumes knowledge of the phase but non-coherent conventional detector compare under Rayleigh
detection is based on the lack of this information. The fading or ideal power control?
cancellation can still be accomplished by canceling the
inphase and quadrature components. The M-ary The answers are displayed in Figure 4. The answer to (ii)
modulation case has been studied for single path Rayleigh is that the improvement is better for the Rayleigh fading
fading [17] and for multipath resolution and combining case because the ranking takes advantage of the unequal
[15]. In both those cases, the Gaussian approximation is powers. Nevertheless, in answer to (i). the scheme still
not adequate and additional analysis was required. works beuer under ideal power control because that is an
easier environment. Thus, the scheme does not have
Rayleigh Fading complete nearlfar resistance but does allow for relaxation
of power control requirements.
When all of the users' signals are received with
independent Rayleigh distributions, this method of Multipath
successive cancellation can be readily analyzed by
considering the order statistics of the received signals. Each user's signal is typically received via several
The amplitudes are assumed to be Rayleigh distributed multipaths. If tbe chip duration is smaller than the time
with unit mean quare value, i.e. its pdf is given by, between the multipaths, the individual multipaths can be
resolved and diversity reception can improve performance.
To d o cancellation with multipath reception is
f , , ( x ) = 22z-x= conceptually simple but increases the complexity.
(v-9) Cancellation is done with all of the multipaths being
tracked. The delails of multipalh reception and M-ary
modulation are given in [IS].

.
I
If all of the users are sufficiently near and the delays small
enough. synchronous operation may be tried.

74
Limiting the Number of Cancellations chip duration ( e = 0 is zero tracking error ). The
interference cancellation scheme retains superiority over
The successive cancellation must operate fast enough to the conventional detector. Similar types of results were
keep up with the bit rate and not introduce intolerable found in [24] from simulation of the scheme of [IO] and
delay. For this reason, it will presumably be necessary to [Ill. While it is premature to draw any general
limit the number of cancellations. The ability to limit conclusions about robustness at this point, these results
the number of cancellations is consistent with the are promising.
objective of controlling complexity by choosing an
appropriateperformandcomplexity aadeoff.
y1 CONCLUDING REMARKS
Parallel vs. Successive Cancellation [ 161
The theoretical foundations of multiuser detections are
Successive cancellation works by successively subtracting fairly well understood now. There is enough potential
off the strongest remaining signal. An alternative (the advantage to investigate the feasibility of practical
parallel method) is to simultaneously subtract off all of implementations.
the users' signals from all of the others. It is found that While the potential advantages are significant, they arc
when all of the users are received with equal strength, the bopnded as discussed in Section 111 so that a relatively
parallel method outperforms the successive scheme simple implementation is dictated. We have discussed
(Figure 5 ) . When the received signals are of distinctly what may be the simplest version, which still retains
different strengths (the more important case), the significant performance advantage over the conventional
successive method is superior in performance (Figure 6). detector. More work is needed on questions of robustness
The important thing to note is that in both cases, both and actual implementability in a real system including
outperform the conventional detector. Another point is questions of complexity and dclay.
that the successive method yields to an accurate analysis
while the parallel method (at least, using the correlator BEFERENCES
outputs to estimate signal strengths) appears to be more
difficult to analyze. The availability of the analytical [l] S . Verdu, "Optimum Multiuser Detectors for
results facilitates parametric studies. CDMA Systems," ISSSTA '94, Oulu, Finland,
July 4-6, 1994.
Sensitivities and Robustness
[2] K. S . Schneider. "Optimum Detection of Code
Almost all of the discussion and analyses of multiuser Division Signals," IEEE Trans. on Aerospoce
detection have assumed a number of idealizations. For nnd Electronic Sysreins. Vol. AES-IS. No. 1, p.
example, in canceling out a chip sequence, it has been 181-185, Jan. 1979.
assumed that there is perfect synchronizationof each user.
Clearly, if there are tracking errors, thc chip sequence [3] R. Kohno, M. Hatori, and H. Imai,
being canceled will be offset and doing an imperfect "Cancellation Techniques of CO-Channel
cancellation. The pertinent question is whether the Interference in Asynchronous Spread Spectrum
tracking error tolerable for the conventional detector is Multiple Access Systems," EIectronics and
tolerable for the cancellation, or how much tighter it must Communicarions, Vol. %-A, No. 5 , pp. 20-29,
be for the interference cancellation. For the conventional 1983.
detector, the synchronization must be within a fraction of
a chip duration. This is analyzed in [23]. For numerical [4] S . Verdu. "Minimum Probability of Error for
results presented here, the interf.erence cancellationscheme Asynchronous Gaussian Multiple Access
was subjected to pessimistic conditions for the interference Channels," IEEE Trans. on Info. Theory, Vol.
canceller: IT-32, No. 1, Jan. 1986.
(i) Did not use averaging of the correlator outputs [5] S. Verdu, "Optimum multiuser asymptotic
for amplitude estimates which significantly efficiency," IEEE Trans. on Coirvrntnicarions.
improves cancellation performance. vol. 34, no.9, pp.890-897, Sept. '86.
(ii) Assumed equal received powers (perfect power
control). The improvement over the [6] A.J. Viterbi, "Very Low Rate Convolutional
conventional detector is much greater in the more Codes for Maximum Theoretical Performance of
realistic case of unequal received powers. Spread-Spectrum Multiple Access Channels,"
IEEE Journal on Selecred Areas in
The processing gain of spread specmm is set to N = 31 Cointnunications, Vol. 8, no. 4, pp. 641-649,
and total numbers of users are ranged from 10 to 24. May 1990.
Figure 7 shows a result for z = 0 (this is optimistic for
the interference canceller). There are 3 curves each for the S . Kubota, S. Kato. and K. Feher, "Inter-channel
interference cancellation scheme and the conventional
p]
interference cancellation technique for CDMA
detector. Each curve represents different standard deviation mobile/personal communications base stations."
e's of tracking error normalized with respect lo ratio of

75
in proceedings. International Symposium on Communications. pp. 150-161. March 1994,
Spread Spearum Techniques and Applications Zurich Swimrland.
(ISSSTA), Yokohama, Japan, pp. 91-94. Dec
92. [18] A.J. Viterbi. "The Orthogonal-Random
Waveform Dichotomy for Digital Mobile
Y.C. Yoon. R. Kohno. and H. Imai, Personal Communications," IEEE Personal
"Combination of an adaptive array antenna and a Communications, pp. 18-24, First Quarter 1994.
canceUerofin~~nce.fordina-Jequenoe spread-
spectqm multiple accesssystem." IEEE Journal [19] L.F. Chang."Dispersive Fading Effects in
on Selened Areas in Comm.. vo1.8. no. 4, May CDMA Radio Systems," Proc. of ICUPC '92,
'90. Dallas, pp. 185-189.9/29-10/2/92.

A. Kajiwara and M. Nakagawa. " Spread [a] K.S. Gilhouscn. IM. Jacobs, R. Padovani. A.J.
spectrum block demodulator with high capacity Viterbi. L.A. Weaver, and C.W. 111, "On the
crossconelation canceller," in proceedings, capacity of a cellular CDMA system," IEEE
Globecom, 1991. Transactions on Vehicular Technology. vol. 40,
no. 2, pp. 303-31 1, May '91.
P. Dent, B. Gudmundson. and M. Ewerbring.
"CDMA-IC: a novel code division multiple [21] M. Varanasi and B. Aazhang, "Multistage
access schemebased on interference cancellation," detection in asynchronous codedivision multiple
in proceedings, PIMRC (Bosron, MA), pp. access communications," IEEE Transucrions on
4.1.1-4.1.5, Oct. 92. Communications. vo1.38, pp.509-519, April 90.

P. Teder. G. Larsson, B. Gudmundson. and M. [22] T.R. Giallorenzi and S.G. Wilson, "Decision
Ewerbring, " CDMA with Interference Feedback Multiuser Receivers for Asynchronous
Cancellation: a technique for high capacity CDMA Syslems." Proceedings Globecom '93,
wireless systems:, in proceedings. IEEE pp. 1677-1682.
international Conf. on CO" (ICC, Geneva),
1993. [23] F.C. Cheng and J.M. Holtzman. "Effect of
Tracking Error on DSlCDMA Successive
M. Kawabe. T. Kato. T. Sato, A. Kawahashi. InterferenceCancellation," To appear.
and A. Kukasawa, "Advanced CDMA scheme for
PCS based on interference cancellation," in [NI L. Levi, F. Muratore and G.
record, ICUPC. pp.IO00-1003. 1993. Romano,"Simulation Results for a CDMA
l n t e r f m c e cancellation Technique in a Rayleigh
Y.C. Yoon, R. Kohno, and H. Imai,"Cascaded Fading Channel." Proceedings of 1994
cochannel interference canceling and diversity International Zurich Seminar on Digital
combining for spread s p e c " multi-access over Cowmunicarions. pp. 162-171, March 1994,
multipath fading channels," IEICE Transactions Zurich S w i t " d .
on Comm.. no.2. pp. 163-168. Feb '93.

P. Patel and J. Holtzman, "Analysis of a simple


successive interference cancellation scheme in
DSlCDMA system using correlations," in
proceedings, Globecom (Houston, Texas), 1993.
P. Patel and J. Hollrman, "Analysis of a simple
successive interference cancellation scheme in
DSKDMA system," to appear in JSAC -
Special issue on CDMA, First Quarter 1994.

P. Patel and J. Holtzman, "Performance


comparison of a DSKDMA system using a
successive interference cancellation (IC) scheme
and a parallel IC scheme under fading," ICC '94
(New Orleans, LA) May 1994.

P.R. Patel and J.M. Holtzman. "Analysis of


Successive Interference Cancellation in M-ary
Orthogonal DS/CDMA System with Single Path
Rayleigh Fading," Proceedings of 1994
International Zurich Seminar on Digital

16
I

........ ., . ~i.
.. . . .

IO'

IOJ

NUMBEROFACnVEUSEES
FIGURE 1 - BER PERFORMANCEOF INTERFERENCECANCELLER AND mCURE4 .AVERAGE BER (BITERROR R A n ) UNDER IDEAL
CONVENTlONALDEIECIION POWER CONIROL AND UNDER RAYLUCII FADING

0.1

b
oai
1

3
ij 0.WI
2
0.Iu)I

Icas

FIGURE2 .BER PEWORMANCEOFCANNLATION UNDER FIGURE5 -PARALLU.VS.SUCCESSNE~CB~nONS


UNEQUAL mwm (IDEAL POWER C O W O L )

I ....... ........... :-<..::.:


. .__
-
. -;-
_---I-

. . . . . . .. .:-.
____---.-.
I I I --

I 0.01
Y

en
:om1 om1

I I I I
Wllmbcrofkd 25
FIGWE~-AVERACEBW(VS~OFACINEUSERS .m
NO AVERAGING. 6 B r AVERAGING AND IO-BTT AVERAGING FIGURE 6 - PARALLEL VS. SUCCFSSIVE CANCELlATlONS
(W\YL€JGH FADING)
78

You might also like