IEEE COMMUNICATIONS LETTERS, VOL. 11, NO.

12, DECEMBER 2007 913

CDMA with Partitioned Spreading

Christian Schlegel, Senior Member, IEEE

Abstract— An enhancement to code-division multiple access (see e.g. [8]). As a consequence, even though non-linear
(CDMA) is proposed whereby the signature sequences are parti- signal-feedback has been used, the performance of these
tioned into sections, which are interleaved before transmission. A systems show many of the traits of iterative linear receivers,
multistage receiver successively forms soft-bit estimates based on
the received sections, which are used to reduce the interference whose supportable system loads are low. For example, the
in subsequent stages via signal cancellation. This significantly load is limited to 17% of the processing gain for certain
increases the user loads which can be supported as the system forms of parallel cancellation (Jacobi cancellers). While serial
effectively suppresses interference, with a marked threshold effect cancellation can support higher loads [6], the loss in signal-to-
below which practically single user performance is achieved. The noise ratio becomes severe for loads larger than about 50%. In
number of stages required is typically small and independent of
the system size, that is the number of users. Furthermore, it is contrast to these symbol-based detectors, our partition-based
shown that the proposed multistage demodulation is insensitive and interleaved system suffers virtually no SNR loss up to
to variations in the received powers. very high loads in excess of 150%, a regime unreachable by
Index Terms— CDMA, iterative processing, turbo demodula- linear receivers.
tion.
II. S YSTEM M ODEL
I. I NTRODUCTION
In a CDMA system a number of transmitters K access a
ODE-division multiple access (CDMA) is a multiple
C access method using individual spreading sequences to
separate different users that share a common frequency band.
common channel, where the signal of user k is
L−1


CDMA has been adopted for various cellular standards [10], xk (t) = Pk bk,l ak,l (t − lT − τk ) (1)
[11], [14], and remains a key technology in the 21st century. l=0
However, CDMA with conventional correlation reception
and transmission is arranged into frames of L symbols. bk,l
suffers from the so-called near-far problem, where strong
denotes the l-th bit, τk the time delay, and Pk the transmit
users can drown the signal of weaker users by introducing
power of user k. The spreading waveform ak,l (t) during
excessive amounts of noise in their demodulation paths. Tra-
symbol time l is composed of N chips as
ditionally, this is addressed using power-controlled CDMA
[13]. But even with power control, the spectral utilization of N

−1
1
CDMA is quite low, which is due to the fact that the signature ak,l (t) = √ ak,l,n pw (t − nTc ) (2)
sequences of different users are not orthogonal, and multiple n=0
N
access interference limits performance.
where the parameter N is the processing gain, and Tc is
In this paper we motivate an enhancement of CDMA which
the repetition rate of the unit-energy chip waveform pw (t).
is based on partitioning the original signature sequences into
The chips ak,l,n are chosen independently and uniformly
sections, which are transmitted individually and interleaved.
from {−1, 1} as in [13], [11]. This is referred to as random
The receiver collects signal contributions from the different
spreading. The ratio of active users to spreading gain, K/N =
partitions and forms soft-estimates of the transmitted bits
α is called the system load.
which are used in successive receiver stages to improve the
The waveform ak,l (t) is divided into M sections that are
signal-to-noise ratio. A number of such stages allows the re-
interleaved (Figure 1), forming the k-th user signal
ceiver to converge to virtually interference-free signals for all
users, allowing for a significant increase in spectral utilization.
√
L−1

M −1

Systems with and without power control are shown to achieve xk (t) = Pk bk,l ck,l,m (t − lT − τk − π(m, l)T /M ) (3)
almost identical performance, obviating power control. l=0 m=0

Iterative interference cancellation methods based on symbol- where

wise cancellation have first been introduced in [12] for parallel
M

−1
interference cancellation, and were further developed in [3]. 1
ck,l,m (t) = √ ak,l,n+mN/M pw (t − nTc ) (4)
Later it was shown that interference cancellation with linear N
n=0
signal feedback was an instance of iterative matrix inversion
is the m-th section of ak,l (t). We assume that M |N . The in-
Manuscript received March 6, 2007. The associate editor coordinating
the review of this letter and approving it for publication was Prof. Yannis terleaver π(m, l) is a permutation among the integers [1, M L],
Psaromiligkos. This work was supported by iCORE Alberta. i.e., the number of partitions in a frame. It is chosen (pseudo)
C. Schlegel is with the Department of Electrical and Computer Engi- randomly to ensure statistical independence among partitions
neering, University of Alberta, Edmonton, AB, T6G 2V4, Canada (e-mail:
[email protected]). in the limit of large interleavers, analogously to the interleaver
Digital Object Identifier 10.1109/LCOMM.2007.070349. in turbo coding [1].
1089-7798/07$25.00

c 2007 IEEE
914 IEEE COMMUNICATIONS LETTERS, VOL. 11, NO. 12, DECEMBER 2007

 2 
(i)
Spreading Sequence Partitioning signal of user k is given by PNk
E bk
,l − b̃k
,l,m , and
ak,1 ak,2 ak,3 ak,4  2 
ck,1,1 ck,1,2 ck,1,3 ck,2,1 ck,2,2 ck,2,3 ck,3,1 ck,3,2 ck,3,3 ck,4,1 ck,4,2 ck,4,3
2 (i)
σb,k
,i = E bk
,l − b̃k
,l,m is the power of the partially
cancelled interference. Adding up all the contributions we
obtain
K

K


2 2 Pk
K,N →∞ Pk

σk,i = σb,k
,i + σ2 −→ 2
σb,k
,i + σ2
k
=1
N N
k
=1
k
=k
(9)
that is, under some mild conditions on the powers Pk
[8],
Fig. 1. Partitioned and interleaved spreading where M = 3 partitions are in-
terleaved using the permutation π(m, l) = [1, 4, 9, 12, 2, 7, 3, 5, 8, 10, 6, 11].
including the term k
= k has no effect for large K, N , as
long as Pk / Pj → 0. This makes the iteration variance
2
user-index independent, i.e., σk,i = σi2 , ∀k.
III. M ULTI -S TAGE R ECEIVER FOR CDMA WITH M
m
=1 zk
,l,m
is Gaussian by the central limit theorem
PARTITIONED S PREADING m
=m
with mean and variance Pk (M −1)/(M σi2 ), and
We consider an AWGN channel, where the first receiver   
stage is a matched filter. Each partition ck,l,m (t) is fil- (i) M − 1 Pk
M − 1 Pk

tered by the corresponding energy-normalized matched filter b̃k
,l,m = tanh + ξ (10)
√ M σi2 M σi2
M ck,l,m (−t), whose output is given by
 
Pk where ξ ∼ N (0, 1). Now, letting b = MM−1 Pσk2

zk,l,m = bk,l + Ik,l,m + nk,l,n (5) 
i

M 2    2 
(i)
where nk,l,n is a white noise sample with power σ 2 and Ik,l,m E bk
,l − b̃k
,l,m = E 1 − tanh b2 + bξ (11)
is interference from the other users.
The a posteriori probability for bk,l can be computed which has no closed-form solution, but the following bounds
assuming Ik,l,m is Gaussian distributed [13], and the vari- are tight [2]:
  
ance σk2 is that of Ik,l,m + nk,l,n , and equals (K − 1)/N + σ 2 ,  2  1
and κ is a normalizing constant. Different partitions must E 1 − tanh b2 + bξ = Φ(b) ≤ min , πQ(b) (12)
1 + b2
agree, and form the cumulative posterior probabilities ∞  2
where Q(b) = √12π b exp − t2 dt is the error integral.
Pr(bk,l = ∓1|zk,l,1 , · · · , zk,l,M ) =
 
1
 2
M 
M
κ exp − 2 zk,l,m ± Pk /M (6) IV. E QUAL AND U NEQUAL P OWER CDMA
2σk m=1
Using (12) in (9) with Pk = P we find
from which we compute a soft-bit estimate b̃k,l of bk,l which  
2 M −1 P
is used to reduce the interference, and in subsequent decoding σi = αP Φ 2 (13)
stages the input signal for user k is M σi−1

(i+1)



K L−1 M

−1 
(i)
 where α is the system load. If we denote P/σi2 = βi as the
yk (t) = xk (t) + Pk
bk
,l − b̃k
,l,m signal-to-noise ratio of all users at stage i, we equivalently
k
=1 l=0 m=0 obtain a one-dimensional iterative equation for βi , given by
k
=k

   −1
T
ck
,l,m t − lT − τk
− π(m) + n(t) (7) M −1 σ2
M βi = αΦ βi−1 + (14)
  M P
(i)
bk
,l − b̃k
,l,m is the (partially) cancelled transmitted bit of
Figure 2 shows the iteration equation (14) for various α’s
user k
and partition m at time l for stage i of the receiver. 2
= 10dB.
The soft bit used for cancellation is
and an SNR of P/σ  −1 For i = 0 the iterations
⎛ ⎞ start with β1 = α + σ 2 /P , which corresponds to the
 M SNR of a correlation receiver [13]. Even for large α, βi
(i) ⎜ 1 Pk
⎟
b̃k
,l,m = tanh ⎝ 2 zk
,l,m
⎠ (8) approaches the interference-free SNR in a few iterations, and
σk
,i M m
=1 convergence to β∞ near 10 dB is achieved for all α below a
m
=m
certain threshold. This threshold depends on M , but quickly
and σk2
,i
is the residual variance of Ik
,l,m + nk
,l,n at stage saturates to α∞ → 1.675 for M → ∞(≈ 50). This sharp
i, with σk2
,0 = (K − 1)/N + σ 2 = σk2 from (5). The dropoff in post-processing SNR happens where the dynamic
“self-information” term zk
,l,m
=m is excluded in (8). This system equation (14) jumps from having one solution to
is customary and avoids signal correlation [9], [7]. having several. At this point the cancellation system collapses,
We now calculate the variance σk2
,i of the interference and but otherwise it essentially completely suppresses interference,
noise at iteration stage i. Due to the random spreading, the as shown in Figure 3, which shows β∞ as a function of
interference power of a user k
on the post-matched filter α. Simulations for systems of frame length L = 2000 and
SCHLEGEL: CDMA WITH PARTITIONED SPREADING 915

10
The additional processing blocks that need to be implemented
5
8 are i) a cancellation structure, and ii) sufficient storage for
Signal-to-Noise Ratio βi [dB]

0.2

.5 the soft values of the partitions. Signal cancellation depends

α=

0
6 α= 75 on accurate timing, phase, and amplitude information of the
α=0. α=1 different users. These parameters are also required for a
4
conventional receiver. For the subtraction of an interfering
2 signal, it is irrelevant if it is synchronous to the wanted
α=1.5 signal or not, and our receiver is applicable without conceptual
0 changes to asynchronous CDMA. The performance analysis,
which is based on interference power, is also insensitive to
-2 signal asynchronism apart from minor variations due to chip
-4 waveform cross-correlations which can easily be quantified
demodulation iteration i once a particular chip waveform is chosen. Adequate storage
0 1 2 3 4 5 6 7 8 9 10 is easily provide by recent VLSI advances, and is also required
Fig. 2. Post-iteration SNRs as a function of iteration index i for partitioned for the operation of advanced coding in conventional systems.
CDMA with M = 3. Square markers are simulation points of a medium-sized
system.
V. C ONCLUSION
solid: equal power system dashed:
10 Μ=50
A novel demodulation method for CDMA based on parti-
cellular power
distribution tioned spreading and simple multistage receiver processing has
Signal-to-Noise Ratio β∞ [dB]

been proposed, which, with only a small number of processing

8 Μ=3 Μ=4
stages, can achieve significant gains in terms of system loading
Μ=4
Μ=3 and spectral efficiency. The system is inherently near-far
6
resistant, maintaining high loads even with received power
Μ=10 Μ=10 levels with significant spread. An iterative equation based on
4 asymptotic density evolution analysis was given that allows
the study of the post-processing signal-to-noise ratios for
2 all sufficiently large systems. The accuracy of this equation
was verified with sample simulations. The proposed iterative
0 Non-PS Non-PS processing system architecture involves only low-complexity
iterative iterative spinodal
(Jacobi) (with tanh()) threshold
signal processing, given by matched filtering and several
α stages of soft-bit generation and cancellation.
0 0.5 1 1.5 2
Fig. 3. Final post-iteration SNR β∞ as a function of load α for M = 3. The
dashed curves are for unequal power scenarios. Also shown is the performance R EFERENCES
of traditional non-interleaved iterative system w/out the tanh(·) function in [1] C. Berrou and A. Glavieux, “Near optimum error correcting coding and
the feedback loop. decoding: turbo-codes,” IEEE Trans. Commun., vol. 44, pp. 1261–1271,
Oct. 1996.
[2] M. V. Burnashev, C. Schlegel, W. A. Krzymien, and Z. Shi, “Analysis of
K = 27 are also shown. Since (K − 1)/N < α, (14) is a the dynamics of interference cancellation in iterative decoding,” Problemy
(tight) lower bound of the SNR of a finite-sized system. Peredachi Informatsii, vol. 40, no. 4, pp. 297–317, 2004.
The performance curves for a traditional system without [3] D. Divsalar, M. K. Simon, and D. Raphaeli, “Improved parallel inter-
ference cancellation for CDMA,” IEEE Trans. Commun., vol. 46, pp.
partitioning and interleaving shown in Figure 3 illustrate the 258–268, Feb. 1998.
significant gains achievable. [4] A. Grant and P. Alexander, “Random sequence multiuser for synchronous
For unequal received powers, the variance iteration (13) has code-division multiple-access channels,” IEEE Trans. Inf. Theory, pp.
2832–2836, Nov. 1998.
to be modified, and defining βi = E [Pk ] /σi = P /σi we find [5] W. C. Jakes, Microwave Mobile Communications. John Wiley, 1974.
a one-dimensional iterative equation for βi as [6] P. Patel and J. Holtzman, “Analysis of a simple successive interference
K   −1 cancellation scheme in a DS/CDMA system,” IEEE J. Sel. Areas Com-

α Pk M −1 Pk σ2 mun., vol. 12, pp. 796–807, June 1994.
βi = Φ βi−1 + (15) [7] T. J. Richardson and R. L. Urbanke, “The capacity of low-density parity-
K P M P P check codes under message-passing decoding,” IEEE Trans. Inf. Theory,
k=1
vol. 47, no. 2, pp. 599–618, Feb. 2001.
(15) has a similar behavior to (14) for most power distribu- [8] C. Schlegel and A. Grant, Coordinated Multiple User Communications.
tions, and Figure 3 includes the load thresholds for an example Springer Publishers, 2006.
[9] C. Schlegel and L. Perez, Trellis and Turbo Coding. IEEE Press, 2004.
of non-uniform power distribution. For this “cellular” example, [10] TIA/EIA, Mobile Station-Base Station Compatibility Standard for Dual-
the transmitters were randomly and uniformly distributed in a Mode Wideband Spread Spectrum Cellular System, July 1993.
circle of radius R = 1. With constant transmitted power P the [11] TIA, TIA/IS-2000 interim standard, Introduction to cdma2000 Spread
Spectrum Systems, July 2004.
received power equals Pk = P/rγ , where γ is the power loss [12] M. K. Varanasi and B. Aazhang, “Multistage detection in asynchronous
factor, typically 2 – 4 [5]. We used K = 54 with i transmitters code-division multiple-access communication,” IEEE Trans. Commun.,
located at distance i/10 for i = 2 · · · 10, using γ = 2. Figure vol. 38, pp. 509–519, Apr. 1990.
[13] A. J. Viterbi, CDMA: Principles of Spread Spectrum Communication.
3 illustrates that this system is highly near-far resistant. Addison-Wesley, 1995.
The complexity of the proposed system is only marginally [14] R. L. Peterson, R. Ziemer, and D. E. Borth, Introduction to Spread
larger than that of a conventional correlation-based receiver. Spectrum Communications. Prentice-Hall, 1995.