7, JULY 2009
Transactions Papers
Abstract—Direct-sequence code-division multiple-access (DS- There are many advanced downlink receiver approaches,
CDMA) cellular systems, such as Wideband CDMA (WCDMA), providing different combinations of performance and com-
are limited in performance by interference. Linear equalization, plexity. Here we consider a relatively simple class of single-
such as generalized Rake (G-Rake) receivers and transversal
chip equalizers, can theoretically provide significant gains in user linear equalizers that employ coherent reception1. A
performance by suppressing interference. In this paper, an representative receiver in this class is the generalized Rake
adaptive G-Rake receiver is developed, employing practical (G-Rake) receiver [3], a type of symbol-level equalizer. It is
algorithms for finger placement and weight computation. Finger based on a maximum-likelihood (ML) formulation, though a
placement is determined by selecting delays from a candidate minimum mean-square error (MMSE) formulation can also
set using a “maximal-weight” criterion. Weight computation
includes estimation of an impairment covariance matrix using be used [4]-[6]. Another member of this class is the MMSE
a parametric approach. While described in the context of the transversal chip equalizer [4], [5], [7]-[9], which has been
downlink, these algorithms can be used in the uplink as well. shown to be equivalent to the G-Rake receiver (for the same
Both single-antenna and dual-antenna receivers are considered. tap and finger placements) [10]. Such receivers were originally
Link and system performance are evaluated for the downlink, developed for the uplink [11], [12] and have a similar structure
showing significant gains in high-rate coverage.
to noncoherent phased-array processing [13].
Index Terms—Coherent demodulation, channel estimation, The G-Rake receiver has the same structure as a Rake re-
rake receivers.
ceiver [14], in that “fingers” (correlators) are used to despread
the received signal at different delays and the despread values
I. I NTRODUCTION are combined to form symbol estimates. With a coherent Rake
receiver, the goal is to collect signal energy by placing fingers
T HIRD generation cellular systems using direct-sequence
code-division multiple-access (DS-CDMA), such as
Wideband CDMA (WCDMA) and CDMA2000, are currently
on the path delays of the dispersive channel and combining the
despread values coherently using weights that correspond to
channel coefficient estimates. With a G-Rake receiver, finger
being deployed worldwide. On the downlink, performance of
placement and weight computation are used to both collect
these systems is limited by various forms of interference.
signal energy and suppress interference.
Within a cell, symbols are transmitted orthogonally, but the
One contribution of this paper is the synthesis of practical
dispersive radio channel destroys this orthogonality. Between
algorithms for finger placement and weight computation,
cells, transmissions are not orthogonal. Performance can be
building on approaches taken from work on G-Rake receivers,
improved by employing advanced receiver structures that
chip equalizers, and linear equalizers. The goal is to achieve
suppress interference, such as linear equalization. The focus
most of the potential performance gains with reasonable com-
of this paper is on developing practical, adaptive algorithms
plexity2 . Finger placement is based on selecting delays from
for advanced receivers and on evaluating link and system
a candidate set using a “maximal-weight” criterion. Weights
performance of such receivers.
are computed from estimates of the channel response and an
Manuscript received June 5, 2006; revised December 13, 2007; accepted impairment covariance matrix. A key aspect of these algo-
February 7, 2009. The associate editor coordinating the review of this paper rithms is that the weights are computed in a way that simplifies
and approving it for publication was M. Sawahashi.
T. L. Fulghum, D. A. Cairns, Y.-P. E. Wang, and G. E. Bottom-
finger placement. While the focus is on the downlink, these
ley are with Ericsson, Inc., Research Triangle Park, NC, USA (e-mail:
1 The receivers are single-user in the sense that only spreading codes
tracy.fulghum@ericsson.com).
C. Cozzo is with Enerdyne Technologies - Viasat, Inc., El Cajon, CA, USA. of desired and pilot symbols are needed, and only for the purpose of
This work was presented in part at the IEEE Digital Signal Processing despreading. Other, more advanced linear receivers, require knowledge of
Workshop, Taos Ski Valley, NM, Aug. 2004, and at the IEEE Vehicular the spreading codes of other symbols to determine combining weights [1] or
Technology Conference, Melbourne, Australia, May 2006. chip equalization filter coefficients [2].
Digital Object Identifier 10.1109/TWC.2009.060330 2 A detailed comparison of approaches is beyond the scope of this paper.
1536-1276/09$25.00
c 2009 IEEE
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 07:57 from IEEE Xplore. Restrictions apply.
FULGHUM et al.: ADAPTIVE GENERALIZED RAKE RECEPTION IN DS-CDMA SYSTEMS 3465
algorithms can be used in the uplink as well. where n(t) represents thermal noise across the receiver anten-
A second contribution of this paper is the evaluation of link nas and is assumed to be a white, Gaussian, and independent
and system performance for the practical G-Rake receiver. across receive antennas.
These results complement existing link results for ideal G- The receiver is a standard Rake receiver in structure. Com-
Rake receivers and MMSE chip equalizers (e.g., [3], [5], bining fingers are used to despread the ath receive antenna
[9]) and system results for ideal G-Rake receivers [15]–[17]. signal to produce a despread value according to
The existing results show that linear equalization can provide ∞ N −1 ∗
1
significant gains. For example, with an ideal single-antenna ya (d) = √ ra (t + d) ck,j (m)p(t − mTc ) dt,
receiver, link gains of 1 - 3.5 dB [3] and system capacity N −∞ m=0
gains of 30% [15], [16] have been shown. (4)
We examine link and system performance of an adaptive where d is a delay specific to the finger.
G-Rake receiver in the High-Speed Downlink Packet Access We will assume, without loss of generality, that the desired
(HSDPA) system. HSDPA is an evolution of WCDMA to symbol is s0 (0). The despread value can be written as ya (d) =
support high-speed packet data. Both single-antenna and dual- ha (d)s0 (0) + ua (d), which describes the despread value as a
antenna reception are considered. Significant gains in system net response times the desired symbol, lumping all undesired
performance are shown. components (interference and noise) into ua (d). From (6) of
The paper is organized as follows. A system model is [15], the net response of the transmitter is defined as
given in Section II. In Section III, the adaptive receiver
L−1
is described, including algorithms for finger placement and ha (d) = Es ga ()Rp (d − τ ()), (5)
weight computation. Performance is evaluated in Section IV, =0
and Section V concludes the paper.
where Rp (·) is the autocorrelation of the chip pulse shape
and Es is the transmit symbol energy of the spreading code
II. S YSTEM MODEL of interest. The output despread values from the combining
Our complex baseband system model extends the model in fingers from all antennas can be collected into a J-length
[15] to multiple receive antennas and is general to both uplink vector, y, where J is the total number of combining fingers.
and downlink. Some of the model development is repeated This gives us a convenient vector notation, y = hs0 (0) + u,
here for convenience. Only a single transmitter is considered, where h and u contain the corresponding values of ha (d) and
though extension to multiple transmitters is straightforward. ua (d).
The radio channel is characterized by discrete, Rayleigh fading The despread values are combined with weight vector w
signal paths. We assume a common path profile (average to form the decision statistic for the desired symbol, z =
strength and path delays) to all antennas, but assume the wH y. For the Rake receiver, the combining weight vector w
realization of the fading coefficients is independent across is simply the vector of medium coefficients (w = g, where
antennas and paths. The impulse response between the trans- g is a stacked vector of all L vectors g()). For the G-Rake
mitter and Na receive antennas can be written as a vector receiver, w is given by
function w = R−1
L−1 u h, (6)
g(τ ) = g()δ(τ − τ ()), (1) where Ru = E{uuH }, with expectation being conditioned
=0 on the fading channel coefficients.
where g() is an Na -length medium coefficient vector at the From (3) and typical independence assumptions, Ru is
th of L resolvable paths for the transmitter, and τ () is the theoretically given by
corresponding path delay. Ru = Ec Ri (g) + N0 Rn , (7)
Combining (1) and (2) of [15], the signal from the trans-
mitter can be written as a sum of K individually spread signal where these terms represent own-cell interference (or self-
components, interference in the uplink) and noise, which models thermal
noise and other-cell interference. (see (12) in [15]). Matrix
K−1 ∞
N
−1
Ri (g) is the normalized interference covariance, and Ec is
x(t) = Ec sk (j) ck,j (m)p(t − mTc ), (2)
the total received energy per chip of the transmitted signal.
k=0 j=−∞ m=0
The normalized noise covariance is represented as Rn , and
where index j represents the symbol period, index k represents N0 is the one-sided noise power spectral density. An element
the spreading code of K possible codes in a multicode system, in Rn corresponding to fingers f1 and f2 is given by
index m represents the chip period, Ec is the total energy per
Rp (d1 − d2 ) a1 = a2
chip, sk (j) is the modulated symbol, ck,j (m) is the spreading rn (f1 , f2 ) = (8)
0 a1 = a2 ,
sequence, and p(t) is the chip pulse shape.
The vector of baseband signals received across the receiver where the delay and antenna settings for finger f1 are given
antennas is by d1 and a1 , respectively (similarly for finger f2 ).
L−1 Similar to (13) in [15], elements in Ri (g) are given by [Eq.
r(t) = g()x(t − τ ()) + n(t), (3) 9] where Tc is the chip duration. The term 1 − δ(m) indicates
=0 that for the desired transmission at chip lag m = 0, there is
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 07:57 from IEEE Xplore. Restrictions apply.
3466 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 7, JULY 2009
L−1
L−1 ∞
ri (f1 , f2 ) = ga1 (1 )ga∗2 (2 ) (1 − δ(m))
1 =0 2 =0 m=−∞
× Rp (d1 − mTc − τ (1 )) Rp (d2 − mTc − τ (2 )), (9)
antenna
signal(s) combining a maximal-weight criterion. Specifically, combining weights
combiner
fingers
corresponding to the candidate delay set are computed3, and
combining delays the delays with the largest weight magnitudes are selected.
weights
path Then, “combining fingers” despread traffic symbols that are
delays
path finger
placement
weight coherently combined to form soft symbol estimates.
searcher computation
weights The channel probing approach is motivated by the fact that
measurement
delays
candidate delays it is difficult to determine instantaneously where to place a
measurement measurements
limited number of fingers. However, it is easier to determine
fingers potentially good finger locations on average (the candidate
set). The mirroring approach is based on identifying delays
Fig. 1. Adaptive receiver block diagram. corresponding to the inverse channel response, as suggested
in [3], [20]. The approach is somewhat ad hoc, as it considers
pairs of paths and their associated inverse response. Inverting
no interference due to spreading code orthogonality. Note that the channel results in own-cell interference suppression, due
the infinite summation in m can also be expressed in closed to the orthogonality property of the spreading codes. The
form [18]. maximal-weight criterion is used due to its simplicity.
The SINR of symbol estimate z, important for transmission The notion of selecting delays from a candidate set can
rate adaptation, is given by be found in traditional equalization [21]-[23]. In a deci-
({wH h})2 sion feedback equalizer (DFE), feedforward tap locations
SINR = , (10) are selected from an evenly-spaced grid using the maximal-
w H Ru w +
weight criterion [21], [22], thresholding the weight magnitudes
where {·} denotes the real part of a complex number. The
[23], or evaluating an approximate expression for the signal-
interfering effect of the quadrature part of the symbol, denoted
to-interference-plus-noise ratio (SINR) [22]. For DS-CDMA
by , can be neglected.
equalization, both SINR [24] and mean-square error (MSE)
[25] have been used as criteria for combining finger delay
III. A DAPTIVE R ECEIVER S TRUCTURE selection in a recursive sequential search approach. In [26], an
The main components of the receiver are finger placement incremental SINR criterion is used in an iterative procedure.
and weight computation as shown in Figure 1. The purpose The procedure terminates when either the incremental SINR
of finger placement is to generate a set of delays for fingers falls below a threshold or the number of available fingers is
that will be combined (combining delays) to produce soft reached.
symbol values. To determine the combining delay set, the The mirroring approach for determining the candidate set
finger placement algorithm generates a superset of delays is related to existing approaches for finger placement in G-
called candidate delays. This set includes the set of mea- Rake receivers and chip equalizers. It can be viewed as an
surement delays (used to estimate receiver parameters), which extension of the approach in [27], [28], which places only two
incorporates the set of path delays (i.e. L resolvable paths of extra fingers or taps for each pair of path delays. The approach
(1)). Weight computation provides set(s) of coefficients used in [29], motivated by locating interfering symbol images, only
by the combiner to generate soft symbol values for forward places one extra finger per pair. In [26], the approach in [27] is
error correction decoding. An overview of finger placement used to obtain an initial candidate set, and additional candidate
and weight computation is given below, along with detailed delays are added as finger delays are selected.
design equations. An analytical approach to finger placement is given in
[25]. It is based on approximating the convolutional inte-
A. Finger Placement Overview gral for continuous-time linear equalization. Sampling times
For finger placement, a “channel probing” approach is used, are determined from a density function that is optimized.
in which the combining delay set is chosen from a set of Interestingly, the optimized density function depends on the
candidate delays. The candidate delay set is determined from magnitude-square of the continuous-time equalizer weighting
path delays of the channel response, which change slowly function, indicating that more samples should be taken where
over time and can be determined using a form of “channel the weighting function is large. The maximal-weight approach
sounding” [19]. The candidate delay set is determined using can be interpreted as a coarse approximation, as selecting
an ad hoc “mirroring” approach that considers pairs of paths maximal-weight fingers from the candidate set provides a
and determines a number of candidate delays corresponding simple way of sampling more where the weighting function
to an inverse channel response. The selection process is 3 Note that it is possible to compute weights for arbitrary delays due to the
performed at a relatively high rate (the fading rate), using parametric approach to covariance estimation.
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 07:57 from IEEE Xplore. Restrictions apply.
FULGHUM et al.: ADAPTIVE GENERALIZED RAKE RECEPTION IN DS-CDMA SYSTEMS 3467
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 07:57 from IEEE Xplore. Restrictions apply.
3468 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 7, JULY 2009
0.5
[32], [33]. The channel response is estimated from a common
4/4 pilot channel using the subset of measurement fingers placed
8/4 on the path delays. The impairment covariance matrix is es-
8/6 timated using a parametric approach, in which the covariance
12/12 matrix is estimated based on a model of the impairment. With
Rake the parametric approach, there are fewer quantities to estimate,
0
−6 −4 −2 0 2 4 6 improving estimation accuracy. Also, combining weights can
SINR (dB) be computed at arbitrary delays, allowing the maximal-weight
finger selection approach to be performed without the need of
Fig. 2. CDF of G-Rake SINR, two equal strength tap channel, various
combinations of candidate delays/combining fingers.
placing measurement fingers on all candidate delays.
The impairment covariance is estimated using the channel
response estimate and estimates of total base station signal
1 power and noise power. These powers, generally unknown at
the receiver, are obtained by fitting a measured impairment
covariance matrix to the modeled one. The measured impair-
ment covariance matrix is typically too noisy to use directly,
and smoothing the measured value over an extended period
prob. of select
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 07:57 from IEEE Xplore. Restrictions apply.
FULGHUM et al.: ADAPTIVE GENERALIZED RAKE RECEPTION IN DS-CDMA SYSTEMS 3469
impairment
^
Ru weight h̃a = A2 g̃a , (15)
covariance solver
estimation where A2 relates the L medium coefficients to the J elements
weights
of the net response vector.
2) Covariance estimation: To estimate Ru , we employ
candidate
delays a parametric approach similar to [35] that estimates model
parameters from pilot symbols. From (7), Ru can be estimated
Fig. 4. Adaptive receiver weight computation. as
R̂u = αRi (g̃) + βRn , (16)
channel estimates [42]. Finally, it is possible to replace the where theoretically β = N0 and α = Ec /Ep , with
impairment covariance estimate with an estimate of the data Ep representing the energy per pilot symbol. The fitting
correlation, either before despreading [5], [43]-[45] or after parameters α and β need to be estimated. The elements of
despreading [43]. Rn are given in (8) and the elements of Ri (g̃) are given in
(9).
D. Weight Computation Details The fitting parameters are obtained as follows. Measure-
Based on (6), the practical combining weights, ŵ, are ment fingers are used to obtain a short-term measurement
(p)
obtained by solving of Ru given by [Eq. 17] where yp (m) stacks ya (m) from
different receive antennas and h̃ is the net channel response
R̂u ŵ = h̃ , (11)
vector obtained by stacking h̃a from all antennas. The fitting
where h̃ is a scaled estimate of h and R̂u is an estimate of parameters are then obtained using a least-squares fit between
Ru . The vector h̃ is a scaled estimate because it is based on the measurement and the model, i.e.,
pilot symbols which may differ in symbol energy from traffic
Ru ≈ αRi (g̃) + βRn , (18)
symbols. Similar to [35], [36], [46], (11) is solved using the
Gauss-Seidel method [47]. where ≈ is used to signify equivalence in a least-squares
The weight computation process is illustrated in Figure 4. sense. The least-squares fit is formulated by stacking the upper
Measurements are used for channel estimation and impairment triangular elements of the constituent matrices. Complex-
covariance estimation. The weights are computed for the set of valued equations are treated as two real equations, as the
candidate delays for the purpose of finger selection, and then fitting parameters are purely real. As the fitting parameters
recomputed for the set of combining delays for the purpose of are positive by definition, negative values are clipped to zero.
combining. Recomputing involves taking a subset of the rows
and columns of R̂u and h̃. E. Example
1) Channel estimation: Correlation-based estimation of the To better understand the impact of estimation error, consider
net channel response at all candidate delays may be too noisy, the following dual-antenna example. The channel consists of
as the channel is estimated in locations where the signal energy 4 chip-spaced paths at relative powers of 0, -3, -6, and -
is low or absent. For this reason, we first “measure” the net 9 dB (exponential profile)7 , and each path is block faded.
channel response at the path delays using pilot symbols, then SINR distributions (over the fading) are shown in Figure 5 for
compute the vector of medium coefficients at the path delays, practical Rake and G-Rake weights. Results with ideal weights
and finally estimate the net channel response for all fingers, are also provided for Rake (“Rake w id”) and G-Rake (“G-
similar to [48]. Rake ref” with unlimited fingers8 and “G-Rake w id” with a
For the ath receive antenna, the net channel response vector limit of B=18 candidate delays (9 per antenna), select J=10).
at the path delays is estimated as Semi-ideal results for G-Rake (“G-Rake w semi”) correspond
1
Ns −1 to using ideal values for the fitting parameters α and β but
h̃a = s∗ (j)y (p)
a (j) , (12) practical channel estimates. For the practical G-Rake results
Ns j=0
(“G-Rake w pract”), measurement fingers were placed on
where j is a time index, s(j) is a known pilot symbol, Ns B = 12 (6 per antenna) of the candidate delays: delays 0
(p)
is the number of pilot symbols, and ya (j) is the vector of through 3 (the path delays in chip periods) as well as -1 and
despread pilot symbol values during symbol period j. -2.
From (5), we can write the relation of the net response 7 This can be recognized as the multipath profile of the 3GPP Case 3 channel
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 07:57 from IEEE Xplore. Restrictions apply.
3470 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 7, JULY 2009
N
s −1
1
Ru = (s∗ (m)y(p) (m) − h̃)(s∗ (m)y (p) (m) − h̃)H , (17)
Ns − 1 m=0
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 07:57 from IEEE Xplore. Restrictions apply.
FULGHUM et al.: ADAPTIVE GENERALIZED RAKE RECEPTION IN DS-CDMA SYSTEMS 3471
0 0
10 10
−1
10
BLER
BLER
−1
10
−2
10
Ideal G−Rake1 Practical G−Rake2
Practical G−Rake1 Practical G−Rake1
Ideal Rake1 Practical Rake2
−2 Practical Rake1 −3 Practical Rake1
10 10
−9 −8 −7 −6 −5 −4 −3 −2 1.5 2 2.5 3 3.5 4
Ec/Ior (dB) Rate (Mb/s)
Fig. 6. Link results, BLER vs. Ec /Ior , exponential channel, ideal and Fig. 8. Link results, BLER vs. data rate, exponential channel, one and two
practical receivers, one receive antenna, Ior /Ioc = 10 dB. receive antennas, practical receivers, Ior /Ioc = 10 dB, Ec /Ior = −3 dB.
0 TABLE I
10 M APPING BETWEEN INFORMATION RATE AND SYMBOL SINR. ( COMPILED
Ideal G−Rake2
BASED ON AWGN SIMULATIONS USING 12 CHANNELIZATION CODES )
Practical G−Rake2
Ideal Rake2 symbol SINR (dB) range bits per symbol
Practical Rake2 (−∞, −11.5) 0.0626
[−11.5, −10.5) 0.0758
[−10.5, −9.5) 0.0990
[−9.5, −8.5) 0.1253
BLER
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 07:57 from IEEE Xplore. Restrictions apply.
3472 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 7, JULY 2009
12 15
G−Rake2 G−Rake2
G−Rake1 G−Rake1
10
Rake2 Rake2
Rake1 Rake1
8 10
4 5
0 0
−5 0 5 10 15 0 0.5 1 1.5
Ior/Ioc (dB) distance (km)
Fig. 9. Link results, median data rate vs. Ior /Ioc , exponential channel, one Fig. 10. System results, median data rate vs. distance, exponential channel,
and two receive antennas, practical receivers. one and two receive antennas, practical receivers.
B. System Performance tions as specified in [50]. Like the case of exponential profile
channel, G-Rake fingers are assigned as described in Section
The performance of the practical receivers is evaluated in a
III-B. For all cases B = 12. For single antenna simulations,
macro-cell environment, where the site deployment consists of
(P = 16, J = 6) for Pedestrian A, and (P = 27, J = 18) for
a uniform hexagonal pattern containing 19 sites, each equipped
Pedestrian B and Vehicular A. For two antenna simulations,
with three sectorization antennas. This results in a total of 57
(P = 20, J = 9) for Pedestrian A, (P = 30, J = 27) for
cells. The sector antenna is normalized to have a constant 0
Pedestrian B, and (P = 32, J = 27) for Vehicular A. We
dBi gain in the 120 degree sector illumination directions and
see that G-Rake2 achieves a median data rate higher than
infinite attenuation in other directions. The site-to-site distance
10 Mb/s at a distance of 0.5 km from the base station, even
is 3 km, and the users are uniformly distributed. This cell
in heavily dispersive channels. Also, within this distance, G-
layout is wrapped around to achieve continuity at the edges.
Rake1 outperforms Rake2. At the edge of the cell (1 km),
The total transmission power per cell is 20 Watts (W) for
G-Rake2 achieves median data rates higher than 5 Mb/s in
all the base stations in the system, including 4.4 W allocated
heavily dispersive channels.
to downlink overhead channels (e.g. common pilot channel,
synchronization channel, and signaling channels). Power and
code allocation to a scheduled data user are the same as those V. C ONCLUSION
used to generate the results in Figure 9. Path loss, shadowing
loss, and receiver noise figure are based on the recommended Practical algorithms have been provided for both finger
parameters for the Vehicular Environment given in [49] (see placement and weight computation in a G-Rake receiver. Fin-
also Table I in [15]). While [49] specifies a receiver noise gers are placed by selecting delays from a candidate set using
figure of 5 dB, the requirement in [31, §25.101] allows the a maximal-weight criterion. Weights are determined using
noise figure to be much higher. We assume a 9 dB receiver channel estimates and a parametric estimate of an impairment
noise figure. Other-cell interference is modeled as white noise covariance. The parametric approach allows weights to be
in both weight formulation and SINR evaluation. determined for the candidate delay set without having to place
Figure 10 shows the median achievable data rate for the measurement fingers at all delays. While the algorithms are
practical receivers in the exponential profile channel as a func- described in the context of a G-Rake receiver and the down-
tion of distance from the serving base station. As expected, link, they can be used in other linear equalization contexts (a
data rate goes down as the user moves farther away from the chip equalizer, uplink). Link and system performance results
base station. Both G-Rake2 and G-Rake1 effectively reduce show that significant improvements in coverage for high-rate
own-cell interference and thus reach higher data rates than services in HSDPA can be achieved with practical receivers.
Rake1 and Rake2 for users close to the base station. At the
cell border, performance is more noise-limited, so the G-Rake ACKNOWLEDGMENT
receivers perform only slightly better than the Rake receivers
and the dual-antenna receivers provide array gain. The authors gratefully acknowledge A. S. Khayrallah and H.
Results in other channel conditions are summarized in Table Björkegren for guiding and supporting this work; J.-F. Cheng
II. Specifically, results are produced for Pedestrian A (mildly for providing the data in Table I; and K. Hooli for help in
dispersive), Pedestrian B (heavily dispersive), and Vehicular identifying and obtaining references. The authors would also
A (heavily dispersive) channels [31, §25.101]. Path delays of like to thank H. Koorapaty, R. Ramésh, and the anonymous
these channels are quantized to half-chip grids in our simula- reviewers for helpful comments.
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 07:57 from IEEE Xplore. Restrictions apply.
FULGHUM et al.: ADAPTIVE GENERALIZED RAKE RECEPTION IN DS-CDMA SYSTEMS 3473
TABLE II
M EDIAN ACHIEVABLE DATA RATES IN VARIOUS DISPERSIVE CHANNELS (75% POWER ALLOCATION AND 15 CHANNELIZATION CODES , QPSK AND
16-QAM)
receiver PA PB VA
0.5 km 1 km 0.5 km 1 km 0.5 km 1 km
Rake1 6.66 Mbps 3.75 Mbps 3.17 Mbps 2.18 Mbps 3.17 Mbps 2.56 Mbps
G-Rake1 11.77 Mbps 3.75 Mbps 6.08 Mbps 3.17 Mbps 6.66 Mbps 3.75 Mbps
Rake2 10.87 Mbps 5.50 Mbps 4.92 Mbps 3.75 Mbps 4.92 Mbps 3.75 Mbps
G-Rake2 14.40 Mbps 6.66 Mbps 10.87 Mbps 5.50 Mbps 14.40 Mbps 6.08 Mbps
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 07:57 from IEEE Xplore. Restrictions apply.
3474 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 7, JULY 2009
[39] K. Li and H. Liu, “A new blind receiver for downlink DS-CDMA Doug Cairns received a B.S. in electrical engineer-
communications," IEEE Commun. Lett., vol. 3, pp. 193-195, July 1999. ing from the Milwaukee School of Engineering in
[40] I. N. Psaromiligkos and S. N. Batalama, “Recursive short-data-record Milwaukee, Wisconsin in 1989 and the M.S and
estimation of AV and MMSE/MVDR linear filters for DS-CDMA antenna Ph.D degrees, both in electrical engineering, from
array systems," IEEE Trans. Commun., vol. 52, pp. 136-148, Jan. 2004. Duke University in Durham, North Carolina in 1991
[41] I. S. Reed, J. D. Mallett, and L. E. Brennan, “Rapid convergence rate in and 1995 respectively.
adaptive arrays," IEEE Trans. Aerospace Electronic Systems, vol. AES- He worked for IBM from 1991 to 1998 and
10, pp. 853-863, Nov. 1974. then joined Ericsson Research in Research Triangle
[42] J. Vidal, M. Cabrera, A. Agustin de Dios, and M. A. Lagunas, “En- Park, North Carolina where he is currently em-
hanced MDIR receiver for space-time dispersive channels," IEEE Trans. ployed. His work focuses on wireless communica-
Wireless Commun., vol. 2, pp. 99-107, Jan. 2003. tions, particularly equalization, channel estimation,
[43] T. Hasegawa and M. Shimizu, “A chip correlation MMSE receiver and transceiver design.
with multipath interference correlative timing for DS-CDMA systems,"
Carmela Cozzo received the Laurea degree in
in Proc. IEEE Veh. Technol. Conf. (VTC), Stockholm, May 2005, pp.
Electrical Engineering from the University of Rome,
1740-1744.
La Sapienza, Italy, and the Ph.D. in Electrical
[44] M. Lenardi and D. T. M. Slock, “A RAKE receiver with intracell
Engineering from North Carolina State University,
interference cancellation for a DS-CDMA synchronous downlink with
Raleigh, NC, in 2001. From 2001 to Nov. 2008 she
orthogonal codes," in Proc. IEEE Veh. Technol. Conf. (VTC), Tokyo,
was with Ericsson Research, NC, USA. Her work
Japan, May 2000, pp. 430-434.
focused on wireless receiver algorithms, interference
[45] E. Tiirola and J. Ylitalo, “Performance of smart antenna receivers
suppression for CDMA systems and HSPA physical
in WCDMA uplink with spatially coloured interference," in Proc. IST
layer. Since Nov. 2008 she has been a Senior Mem-
Mobile Commun. Summit, Sitges, Barcelona, Spain, Sept. 2001.
ber of Technical Staff at Enerdyne Technologies -
[46] L. Mailaender, “Low-complexity implementation of CDMA downlink
ViaSat, CA, USA. Her work focuses on wireless
equalization," in Proc. IEE Intl. Conf. 3G Mobile Commun. Technol.,
communications for unmanned aerial systems. Her research interests include
Mar. 2001, pp. 396-400.
signal processing for wireless communications, interference cancellation and
[47] G. H. Golub and C. Van Loan, Matrix Computations. Baltimore, MD:
multi-antenna systems. She has authored 19 issued or pending patents in the
The Johns Hopkins University Press, 1996.
field of wireless communications. Dr. Cozzo is an Associate Editor for the
[48] B. C. Ng, M. Cedervall, and A. Paulraj, “A structured channel estimator
IEEE T RANSACTIONS ON V EHICULAR T ECHNOLOGY, and a member of the
for maximum-likelihood sequence detection," IEEE Commun. Lett., vol.
IEEE Vehicular Technology Society BoG serving as VTS Liaison to the IEEE
1, pp. 52-55, Mar. 1997.
Women in Engineering Committee.
[49] “Selection procedures for the choice of radio transmission technologies
of the UMTS (UMTS 30.03)," ETSI Tech. Rep. 101 112, version 3.1.0,
Nov. 1997. Y.-P. Eric Wang (S’91 - M’96) received the B.S. de-
gree in electrical engineering from National Taiwan
[50] “Modified channel models and channelisation codes for LMMSE sim-
University in 1988, and the M.S. and Ph.D. degrees,
ulations," 3rd Generation Partnership Project (3GPP); Technical Speci-
both in electrical engineering, from the University
fication Group Radio Access Networks; Working Group 4, R4-050112,
of Michigan, Ann Arbor, in 1991 and 1995, respec-
Feb. 2005.
tively. He has been a member of Ericsson Research
in Research Triangle Park, North Carolina, USA,
Tracy L. Fulghum (S’87 - M’88 - SM’05) received since 1995. His work focuses on wireless com-
the B.S. and Ph. D. degrees in electrical engineer- munications, including both mobile satellite com-
ing from North Carolina State University in 1987 munication systems and terrestrial cellular systems.
and 1997, respectively, and received the M.S. in His research interests include coding, modulation,
electrical engineering from University of Florida in synchronization, MIMO, and interference cancellation and suppression. He
1993. holds over 40 U.S. patents in the area of wireless communications.
From 1988 to 1994, he was with Motorola, Inc.,
working in land mobile radio research and develop- Gregory E. Bottomley (S’81 - M’85 - SM’99 -
ment. Since 1997, he has been with Ericsson Inc., in F’07) received the B. S. and M. S. degrees from
Research Triangle Park, NC. His research interests Virginia Polytechnic Institute and State University,
are mobile radio systems and signal processing Blacksburg, in 1983 and 1985, respectively, and the
for mobile communications, including equalization, array processing, and Ph.D. degree from North Carolina State University,
interference cancellation, having contributed to 19 patents in these areas. Raleigh, in 1989, all in electrical engineering.
Dr. Fulghum is a member of the IEEE Communications, Signal Processing, From 1985 to 1987 he was with AT&T Bell
and Vehicular Technology Societies, and is Secretary and Board Member for Laboratories, Whippany, NJ, working in the area
the VTS. He is a registered Professional Engineer in the state of Florida. of sonar signal processing. In 1990, he was a Vis-
iting Lecturer at North Carolina State University,
Raleigh. Since 1991, he has been with Ericsson Inc.,
Research Triangle Park, NC, where he is currently a member of Ericsson
Research. He has contributed to over 80 patents in the area of wireless
communications. His research interests are in baseband signal processing,
including channel equalization and interference suppression.
Dr. Bottomley is a member of Sigma Xi. He served as an Associate
Editor and then as the Editor for the IEEE T RANSACTIONS ON V EHICULAR
T ECHNOLOGY. He also served as Technical Program Committee Co-chair for
the IEEE Vehicular Technology Conference Fall 2007.
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 3, 2009 at 07:57 from IEEE Xplore. Restrictions apply.