Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
I. INTRODUCTION
Manuscript received June 4, 2001; revised April 22, 2002 and November 4,
2002; accepted November 7, 2002. The editor coordinating the review of this
paper and approving it for publication is W.-Y. Kuo.
The author is with Ericsson Inc., Research Triangle Park, NC 27709 USA
(greg.bottomley@ericsson.com).
Digital Object Identifier 10.1109/TWC.2003.817434
PROJECTION
891
892
is obtained
(8)
where
(5)
where
(6)
can be first filtered by a filter
Thus, the received signal
matched to the chip pulse shape, yielding the received samples
. The detected bit value is given by
(7)
denotes taking the real part (
will denote
where
taking the imaginary part). The sign operation can be omitted to
obtain a soft bit value.
III. ML SOLUTION
An ML receiver is developed for the received sequence
, assuming knowledge of the channel
coefficients and , the desired signals spreading sequence
, and the interfering signals complex scrambling
. For this formulation, it is assumed that
,
sequence
so that there are no other in-cell (orthogonal) user signals. This
assumption is relaxed in the next section.
(9)
is a sequence of independent circularly symmetric
and
Gaussian random variables. Both real and imaginary compo.
nents have power
Observe that the interfering signal term consists of an unand a known complex component
known real component
. Similarly, the desired signal consists of a real compo. These components are
nent and a complex component
shown graphically in Fig. 2, which illustrates the complex plane
consisting of the real axis (in-phase, or component) and the
component). Intuitively, as
imaginary axis (quadrature, or
shown in the figure, the desired signal component can be projected to be orthogonal to the interference component, eliminating the interference.
The interference component is modeled as noncircular complex Gaussian noise that is independent from one sample to
is large, so that
the next. Such a model is reasonable when
is approximately Gaussian by the central limit theorem.
As thermal noise and interference are assumed to be independent on each sample, log-likelihood ratio (LLR) values from
different chip periods can simply be added to obtain a LLR for
bit .
As used in [3], the components of complex receive value
, denoted
and
, are used to form a real vector
PROJECTION
893
(10)
(21)
where
(11)
where
(12)
(22)
(13)
and
The LLR for
.
given the vector
is given by
(14)
where the overall noise correlation matrix is given by
(15)
denotes expected value. Ignoring constant scaling
and
is symmetric, an
factors1 and using the property that
equivalent LLR is given by
(16)
where
(17)
is a vector of combining weights.
1Omitting common scaling factors has no impact on bit detection, and it
has no impact on decoder performance using soft information for certain decoders, such as a conventional ML sequence estimation decoder for convolutional codes.
(25)
(26)
(27)
(23)
894
Fig. 3.
I /Q projection receiver.
(28)
(29)
The ML solution can be related to the conventional solution by
substituting (25)(27) into (28), giving
(30)
where
(5) and
in
(31)
is a second, despread value that depends on both the desired
signals spreading sequence and on the interfering signals
scrambling sequence. Thus, the decision variable of the ML
solution can be formed using conventional despreading and a
second despreading operation, as illustrated in Fig. 3. The two
despread values are weighted, depending on the thermal noise
and interfering signal powers.
With the decision variable in (28), different chip values are
weighted differently. As a result, when other orthogonal users
, loss of orthogonality to these users ocare present
curs. Thus, one form of interference is suppressed but another
form is introduced.
The ML solution for this case is difficult to obtain, as the
orthogonality property is more apparent after despreading. Instead, a ZF equalization approach is used to preserve orthogonality between in-cell signals, at the expense of a reduction in
To preserve orthogonality between in-cell users, a chipdependent real scaling factor is introduced into the ZF term,
which is determined by analysis of the decision variable when
orthogonal users are present. This scaling factor is shown
to depend on the scrambling sequences of the desired and
interfering signals, as well as the relative angle between the
channel coefficients of the desired and interfering signals.
Because one of the two terms that forms the decision variable
has been changed, the relative weighting of the two terms
needs to be reoptimized. This relative weighting is optimized
to maximize the overall signal-to-interference-plus-noise ratio
(SINR) of the decision variable. Finally, an adjustment factor
is introduced to optimize the soft information.
is introSpecifically, a chip-dependent scaling factor
term), as the conventional term
duced into the ZF term (
preserves orthogonality. Second, the conventional and
ZF components are weighted by real weights
and , which
are optimized for performance. Third, an overall adjustment
factor is introduced, which scales the decision variable properly for soft information generation. The resulting receiver, referred to as / projection, is given by
(32)
where
(33)
(34)
PROJECTION
895
A. Scaling Factor
that preserves orthogonality,
To determine the value for
the ZF term from (32) is analyzed. For this analysis, it is convenient to express the ZF term as
(35)
This term is made to be zero when the received signal is an ar,
bitrary, orthogonal user. From (8), this implies
where
(36)
Substituting (36) for
(37)
where
(38)
(39)
denotes the angle or phase in the complex plane
Note that
and
correspond to orof complex number . As
thogonal Walsh codes, the sum in (37) is zero when the term in
brackets is constant. Setting the term in brackets to unity gives
(40)
For a fixed , observe that
only two possible values, as
, as a function of
is either
or
, takes on
.
C. Adjustment Factor
As the resulting solution can no longer be interpreted in an
ML sense, a scaling factor is needed so that the soft decision
variable has properties similar to an LLR. Specifically, the soft
decision scalar variable should have the form
(44)
where
has variance SINR. From the expression for SINR
given by (70) of the Appendix, it is straightforward to show that
this is achieved by setting
(45)
(43)
where
is given in (60) of the Appendix. Observe that the
multiplies the noise power in (43), indicating how
term
equalization enhances the thermal noise.
896
V. PERFORMANCE
Performance is evaluated for the correlation receiver and the
/ projection receiver under both static (nonfading) and fading
channel conditions. For both receivers, the channel responses
and are assumed to be known. In practice, they would be
estimated using pilot channels.2
Analytical results are obtained using the SINR expression in
for the correlation receiver
(70) of the Appendix, where
[see (43)] for / projection. Modem bit-error-rate
and
(BER) results are obtained using the standard relationship between signal-to-noise ratio and BER for BPSK modulation [29]
(46)
Semianalytical results are obtained for fading channels by averaging (46) for 10 000 fading realizations using Monte Carlo
simulation.
Link simulation is used to verify the analytical results and to
provide frame-erasure rate (FER) results based on one of the
coding options used in TIA/EIA-95-B [1]. For each frame, a
block of 184 information bits plus eight tail bits are encoded
convolutional code with constraint length nine,
using a rate
producing a 384-bit frame. Pseudorandom puncturing of 1 out
of 12 bits is performed for power control signaling purposes, so
. Thus, the energy-per-inforthat the effective code rate is
. Each coded
mation-bit ( in [30]) for user is
symbol (modem bit) is spread using a spreading factor of 64.
At the receiver, ideal error detection is assumed, so that a frame
is erased if any of the information bits is in error. In practice,
there is a cyclic redundancy code which is used to detect errors.
As TIA/EIA-95-B has a 1% FER operating point, results that
include this operating point are provided.
For both static and fading channel cases, 10 000 frames of
384 bits each were generated. For the fading channel cases, these
10 000 frames were generated using 200 independent realizations of 50 frames each, assuming a Doppler spread of 7 Hz
and the classic Jakes spectrum. The TIA/EIA-95-B chip rate of
1.2288 Mchips/s was assumed.
Unless otherwise indicated, two base-station signals are simand
ulated assuming 33 equal-power user signals (
).3 The desired base-station signal includes the user
of interest. The interfering base-station signal was simulated
using a sufficiently delayed version of the desired base stations
signal. The ratio of average powers of the two base stations is de(
using
noted as the carrier-to-interference ratio
the notation in [30]). In general, typical values for
depend
on whether a system employs soft handoff and how soft handoff
is implemented. To provide general results, a wide range of
values are examined.
2The traditional receiver would estimate only c, whereas, the proposed approach would require estimation of d as well. The accuracy of these coefficient
estimates would depend on the pilot channel strengths as well as how fast the
channel was changing (Doppler spread).
3Downlink power control was not simulated. However, experience indicates
that with power control, performance is more often interference limited, so that
gains from interference suppression increase.
= 0 dB.
A. Static Channel
For the static channel case, we first examine the SINR gain of
/ projection relative to conventional correlation using the analytical expressions. From these expressions, the SINR gain depends on the relative angle between the desired and interfering
and the interference power
relative to the noise
signals
power
. Analytical results are given in Fig. 5, where only
results for the first octant are shown, due to octant symmetry.
Observe that the gain is maximized when the desired and interfering signals are 45 apart, which provides maximum separation between the two signals. Also, the gain increases the larger
the interference power is, relative to the noise power.
In Fig. 6, modem BER results for correlation and / pro. For these
jection receivers are shown as a function of
results, the two base stations have equal power (
dB),
which is a worst case scenario. Both analytical (markers) and
link simulation (lines) results are shown.
For / projection, results for a fixed best case relative angle
of 45 are given, indicating the best possible performance.
In addition, modem BER averaged over uniformly distributed
relative angles is also given. As the correlation receiver performance is independent of relative angle, only the random angle
result is shown.
Observe that the conventional correlation receiver has a high
error floor, due to interference from the interfering base station.
The / projection receiver is able to suppress that interference, providing significantly improved performance. Also observe that there is good agreement between the analytical and
simulation results.
=N
PROJECTION
= 0 dB.
=N = 21 dB,
897
=N = 21 dB, one or
Performance was also evaluated for the case of one base station and a two-ray chip-spaced channel. This is equivalent to
transmit delay diversity in a nondispersive channel, in which a
second antenna is used to transmit the base-station signal delayed by one chip period. At the receiver, there are two copies
of the base-station signal, fading independently in the presence
is either split evenly between the
of thermal noise. Energy
two rays (the ratio of the second ray power to the first ray power
is 0 dB) or unevenly (the ratio of ray powers is 6 dB).
The correlation receiver was replaced by the traditional, coherent RAKE receiver [29], in which the despread value for each
signal image is weighted by the conjugate of the corresponding
channel coefficient, then added to form the decision variable.
is replaced by
, where subscripts
Thus, in (7),
zero and one denote different signal images. For / projection,
the approach was applied separately to each signal image, and
the results were then added.
Performance results for both RAKE and / projection are
given in Fig. 10. FER is plotted as a function of the fraction of
the total base-station power given to the desired user (related to
in [30]). Note that in this case, there are still 33 users,
but the desired users power differs from the power level used
for
by the other 32 users. To maintain a fixed noise level,
the other 32 users is fixed at 21 dB.
Observe the gains in performance using the / projection
approach. At 1% FER, the gain of / projection over RAKE
reception for the cases of even and uneven rays is 2.7 and 2.4 dB,
respectively.
898
APPENDIX
In this Appendix, the SINR of / projection with equalization is determined as a function of a parameter . By differentiating with respect to and setting the result to zero, the optimal
value for is determined. It is straightforward to show that the
second derivative is negative, so that the result maximizes SINR.
The decision variable is given in (32), and the overall positive scaling factor may be omitted without loss in generality.
Substituting (33), (34), and (40) into (32), the decision variable
can be expressed as
(47)
Fig. 10. FER versus fraction of base-station power given to the desired signal,
two-ray, chip-spaced Rayleigh-fading channel (second ray with relative power
of 0 dB or 6 dB), E=N = 21 dB.
The fact that the two signal images employ the same scrambling sequence is not essential to either RAKE or / projection
performance. Thus, if the two signal images employ different
scrambling sequences, which corresponds to a soft handoff situation, the performance gains would be the same.
VI. CONCLUSION
In the downlink of a DS-CDMA system, interference can be
suppressed in the complex / plane using partial knowledge
of the spreading codes that form the interference. Specifically,
knowledge of a common complex scrambling mask can be used
to project the desired signal away from interfering signals in the
complex plane. To preserve orthogonality with own base-station
signals, ZF equalization can be used, at the expense of decreased
interference suppression.
When there are multiple interfering signals, the / projection approach can be applied to the strongest interfering signal,
treating the rest as noise. Though not explored here, this approach can be refined by selecting which signal to treat as interference based on output SINR, which depends on both signal
strength and phase, relative to the desired signal.
While developed for chip-aligned interference from another
base station, the technique can be applied to misaligned interference, treating it as multiple interfering signals, based on the
ringing of the pulse shape. It can also be applied to the multipath
dispersion, transmit delay diversity, and soft handoff scenarios,
in which the / projection approach is applied to each signal
image separately, treating the other images as interference.
For future research, it would be interesting to formulate the
ML solution for the case of chip-asynchronous interference.
This is a challenging problem, as this introduces interference
correlation between successive chips. It would also be interesting to formulate the ML solution for the case of dispersive
signals.
where
(48)
Substituting the model (8) into (47), the power of the signal,
noise, and interference components of can be determined to
give an overall SINR for of the form
(49)
Each term in (49) is determined separately.
1) Signal Term: Substituting
and (48) into (47), the signal component of is
shown in (50), at the bottom of the page. Substituting the fact
in (50) gives
that
(51)
and
are defined in (38) and (39), respecwhere
is the reciprocal of
tively. Observe, from (40), that
, so that (51) further simplifies to
(52)
Thus, the signal power in
is given by
(53)
and
(54)
(50)
PROJECTION
into
899
where
(64)
(55)
(65)
(66)
where
(67)
(56)
where is the power in the real or imaginary part of
.
For further simplification, we assume that is nonzero and
use the fact that
(68)
(57)
and
. Substituting (48) into (57) and using steps
similar to the signal power analysis gives
(58)
is
, there are two possible values for
.
Since
is
for
of the chips and
for the other
Assuming
chips, which is reasonable for large , (56) using (58) gives
(59)
(70)
The optimal setting for , denoted , is obtained by differentiating the SINR expression in (70) with respect to and setting
it equal to zero, i.e.,
where
(60)
(71)
(72)
The resulting SINR can be obtained by substituting (72) in
(70). SINR for the conventional correlation receiver is obtained
.
by evaluating (70) using
ACKNOWLEDGMENT
The author would like to thank R. Ramsh and E. Sourour
for the help in developing the link simulation tool used, as
well as H. Arslan and H. Koorapathy for providing key references. The author also gratefully acknowledges H. Arslan,
A. S. Khayrallah, Y.-P. E. Wang, K. Zangi, the anonymous
reviewers, and the editor of this paper for providing helpful
comments and suggestions.
900
REFERENCES
[1] Mobile StationBase Station Compatibility Standard for Wideband
Spread Spectrum Cellular Systems, Telecommun. Ind. Assn. Standard
TIA/EIA-95-B, 1999.
[2] Physical Layer Standard for cdma2000 Spread Spectrum Systems Addendum 1, Telecommun. Ind. Assn. Standard TIA/EIA/IS-2000.2-A-1,
2000.
[3] B. Picinbono, On circularity, IEEE Trans. Signal Processing, vol. 42,
pp. 34733482, Dec. 1994.
[4] F. D. Neeser and J. L. Massey, Proper complex random processes with
applications to information theory, IEEE Trans. Inform. Theory, vol.
39, pp. 12931302, July 1993.
[5] B. Picinbono, Second-order complex random vectors and normal distributions, IEEE Trans. Signal Processing, vol. 44, pp. 26372640, Oct.
1996.
[6] W. M. Brown and R. B. Crane, Conjugate linear filtering, IEEE Trans.
Inform. Theory, vol. IT-15, pp. 462465, July 1969.
[7] W. A. Gardner, Cyclic Wiener filtering: Theory and method, IEEE
Trans. Commun., vol. 41, pp. 151163, Jan. 1993.
[8] B. Picinbono and P. Chevalier, Widely linear estimation with complex
data, IEEE Trans. Signal Processing, vol. 43, pp. 20302033, Aug.
1995.
1 narrow-band
[9] D. B. Williams and D. H. Johnson, On resolving 2
signals with an M sensor uniform linear array, IEEE Trans. Signal Processing, vol. 40, pp. 707711, Mar. 1992.
[10] P. Chevalier, Optimal array processing for nonstationary signals, in
Proc. IEEE Int. Conf. Acoustics, Speech, and Signal Processing, Atlanta,
GA, May 710, 1996, pp. 28682871.
[11] S. V. Schell, A separability theorem for 2M conjugate-symmetric signals impinging on an M-element sensor array, IEEE Trans. Signal Processing, vol. 45, pp. 789792, Mar. 1997.
[12] A. Ferrol and P. Chevalier, On the behavior of current second and
higher order blind source separation methods for cyclostationary
sources, IEEE Trans. Signal Processing, vol. 48, pp. 17121725, June
2000.
[13] J. Galy and C. Adnet, Blind separation of noncircular sources, in Proc.
10th IEEE Workshop Statistical Signal and Array Processing, Pocono
Manor, PA, Aug. 1416, 2000, pp. 315318.
, Canonical correlation analysis: A blind source separation using
[14]
noncircularity, in Proc. 2000 IEEE Int. Workshop Neural Networks
for Signal Processing, Sydney, NSW, Australia, Dec. 1113, 2000, pp.
465473.
[15] P. Chevalier and B. Picinbono, Complex linear-quadratic systems for
detection and array processing, IEEE Trans. Signal Processing, vol.
44, pp. 26312634, Oct. 1996.
[16] D. T. M. Slock and H. Trigui, An interference cancelling multichannel
matched filter, in Proc. IEEE Global Telecommunication Conf., Communication Theory Mini-Conf., London, U.K., Nov. 1822, 1996, pp.
214218.
[17] H. Trigui and D. T. M. Slock, Cochannel interference cancellation
within the current GSM standard, in Proc. IEEE 1998 Int. Conf.
Universal Personal Communications, Florence, Italy, Oct. 59, 1998,
pp. 511515.
[18] G. Gelli, L. Paura, and A. M. Tulino, Cyclostationarity-based filtering for narrowband interference suppression in direct-sequence
spread-spectrum systems, IEEE J. Select. Areas Commun., vol. 16, pp.
17471755, Dec. 1998.
M0