Documente Academic
Documente Profesional
Documente Cultură
11b
Christof Jonietz† , Wolfgang H. Gerstacker† , and Robert Schober‡
†
Chair of Mobile Communications, University of Erlangen-Nürnberg, Cauerstrasse 7, D-91058 Erlangen, Germany
Email: {jonietz, gersta}@LNT.de
‡
Department of Electrical and Computer Engineering, University of British Columbia, Vancouver, Canada
Email: rschober@ece.ubc.ca
Abstract
In this paper, complementary code keying (CCK) transmission over frequency–selective fading channels is
investigated. A vector signal model is introduced since CCK can be interpreted as a block code. Based on pairwise
error probabilities we derive a performance approximation for optimum detection obtained for given channel
statistics, and compare the theoretical results with simulations. Because optimum maximum–likelihood sequence
estimation (MLSE) is not practical for large channel delay spreads, minimum mean–squared error decision–feedback
equalization (MMSE–DFE) is employed. Two different DFE approaches are considered here: (a) the conventional
scalar DFE and (b) a block–based DFE, which is well suited for the underlying vector signal model. For various
common indoor channel profiles, numerical results are given for MLSE, scalar DFE and block DFE in terms of bit
error rates (BERs) and packet error rates (PERs).
Note, that ϕ0 is contained in all chip phases and simply where hT (t) and hR (t) denote the transmit and re-
rotates the whole code word c. ϕ1 is contained in all ceiver input filter impulse responses, respectively, and
n(t)
a)
dF [·] CCK cF [·] r[µ]
hT (t) hC (t) hR (t)
encoder t=µT
n[µ]
b)
dF [·] CCK cF [·] r[µ]
h[l]
encoder
n[k]
c)
d[k] CCK c[k] r[k] ML ĉ[k] CCK d̂[k]
H s [m] demapper
encoder detector
Fig. 1. Block diagrams of a) the continuous–time, b) the chip–based discrete-time, and c) the symbol–based discrete–time equivalent baseband
transmission model.
The PEP Pe (α, β) is the probability that a code word where γ0 has to fulfill γ0 < ℜ{s1 }. γ0 = 10−3 is
c = cα is transmitted and the maximum–likelihood typically a good choice. The relevant pairs of (α, β)
receiver decides in favour of ĉ = cβ (cα , cβ ∈ C, α 6= are then determined by comparing the corresponding
β). According to [12] values of Φ∆(α,β) (γ0 ).
2 2
Pe (α, β) = Pr{||r − C α β
m h|| > ||r − C m h|| } 4.3 Decision–Feedback Equalization
= Pr{ ∆(α, β) < 0 }, (21) Because the complexity of MLSE seems to be too high
where Cα and C βm
are defined similarly to Eq. (17). for a practical implementation, suboptimum receiver
m
Thus, the PEP can be reduced to the probability that concepts for CCK modulation are required. In [14],
a random variable ∆(α, β), which depends on the combined equalization and decoding of CCK signals
mean and the autocorrelation matrix of h (see [12] for based on the Fano sequential decoding algorithm has
details), with pdf p∆(α,β) (x) takes on a negative value been proposed, and it has been shown that such a
[13] sequential decoder performs close to the optimum
receiver for certain scenarios. However, the compu-
Z0 tational load of a sequential decoder is a random
Pe (α, β) = p∆(α,β)(x) dx. (22) variable which might be a drawback if regularity is
x=−∞ required in an implementation. Therefore, alternative
In order to evaluate the PEP, the Laplace transform low–complexity concepts are of interest.
Φ∆(α,β) (s) of the pdf p∆(α,β) may be used, which is In the following, combined equalization and decod-
given in [12] in closed form. Then, the PEP for an ing of CCK signals based on feedback of previous
arbitrary code word pair can be determined as decisions is investigated. Because block–coded signals
are transmitted, decision feedback has to be done at
Z ∞
γ+j
block level, i.e., block decision–feedback equalization
1 ds
P (α, β) = Φ∆(α,β) (s) (23) (DFE) [15], [16] is applied. For the simplest version of
2πj s
γ−j ∞ a block DFE, the decision rule is given by [15]
n
for 0 < γ < ℜ{s1 }, where s1 denotes that pole of
ĉ[k] = argmin r[k] − H s [0] c̃[k]
Φ∆(α,β) (s) which has minimum positive real part (ℜ{·} e
c[k]
is the real part operator). We evaluate Eq. (23) numer- )
Xq̃h 2
ically based on Gauss–Chebyshev quadrature rules as
− H s [m] ĉ[k − m] ,
described in [12]. m=1
(28)
4.2.2 Approximation of the Bit Error Rate where ĉ[k−1], . . . , ĉ[k− q̃h ] denote previously decided
code words.
Based on the PEP, the Union Bound on the average bit The performance of a block DFE can be improved
error rate Pb is given by if an FIR matrix feedforward (FF) filter with impulse
|C| |C| response F [m] is introduced in order to concentrate
1 X X n(α, β)
Pb ≤ · Pe (α, β), (24) the energy of the overall response in the first matrix
|C| α=1 8
β=1 tap, cf. Fig. 3. In the feedback (FB) section, the
β 6= α
postcursor ISI corresponding to the overall forward
where n(α, β) denotes the number of bit errors if cα impulse response is subtracted using a FB filter with
is transmitted and cβ is detected. coefficient matrices B[m]. The FF and the FB filter
In order to avoid the computation of the PEP for all may be jointly optimized according to the minimum
possible pairs of (α, β), only those (dominant) pairs mean–squared error (MMSE) criterion, cf. [17]. For
of (α, β) are considered for evaluation of Eq. (24), for this, the autocorrelation matrix of the code words is
which required, which is given by a scaled 8 × 8 identity
matrix because the chips of a code word c are mutually
Pe (α, β) ≥ ξ · Pe,max (25) uncorrelated.
is true, where For sufficiently high signal–to–noise ratios (SNRs)
n o and FIR filter orders, the MMSE solution tends to the
Pe,max = max Pe (α, β) (26) zero–forcing (ZF) solution. The transfer function (z–
(α,β)
(α6=β) transform of impulse response) of the ZF–DFE FF filter
n[k]
B[m]
reads [18] Here, Hmin (z) and H(z) denote the z–transforms of
hmin [·] and h[·], respectively.
F (z) = H −1 −1
s,min [0] H s,min (z) H s (z). (29) Hence, for high SNRs the application of an MMSE
Here, H s,min (z) is the minimum–phase equivalent of matrix FF filter at symbol level is equivalent to op-
the transfer matrix H s (z), with timum FF filtering at chip level. On the other hand,
for low–to–moderate SNRs the MMSE matrix FF fil-
HH ∗ H ∗
s (1/z ) H s (z) = H s,min (1/z ) H s,min (z) (30) ter yields a somewhat better performance than scalar
MMSE FF processing, cf. Section 5. FF filtering at
((·)H : Hermitian transposition), and det(H s,min (z))
chip level and subsequent collection of chips in words
(det(·): determinant of a matrix) has roots only inside
can be also described by Fig. 3, where the FF filter
the unit circle. Furthermore, similar to the scalar case,
matrices F [m] can be obtained similarly to H s,min [m]
H s,min [·] is characterized by a better energy concen-
(Eqs. (32), (33)) from the scalar FF filter coefficients
tration in the front part than H s [·] [18],
f [0], . . . , f [qf ] (qf : FF filter order). Note that in this
k
X k
X case the additional 8 × 8 whitening matrix W (dashed
kH s,min [κ]k2F ≥ kH s [κ]k2F , 0 ≤ k ≤ q̃h part in Fig. 3) is not necessary because the components
κ=0 κ=0 of the error vector of the DFE are uncorrelated. On the
(31)
other hand, if the optimum matrix DFE filters according
(k·kF : Frobenius norm of a matrix).
to [17] are chosen, the chips of the error vector are
Because of the special structure of H s (z)
correlated, and spatial whitening is necessary if the
(cf. Eqs. (9), (10)), it turns out that the coefficient
Euclidean metric is applied for forming code word
matrices of the minimum–phase equivalent can be
decisions3 . For high SNRs, W = H s,min [0] results.
determined (up to an arbitrary unitary matrix factor)
from an 8 (q̃h + 1) × 8 matrix H ′s,min as follows:
5 Numerical Results
H s,min [0]
H s,min [1]
For three test channels, the derived performance ap-
H ′s,min = .. , (32)
. proximation for MLSE is compared with simula-
H s,min [q̃h ] tions. We consider the AWGN channel, a test channel
(ProB ) with fixed channel impulse response h[·] =
where, similar to Eq. (9), H ′s,min is defined as the {0.407, 0.815, 0.407} [2], and a stochastic test channel
Toeplitz matrix (3TAP ) with 3 taps. The amplitudes of the 3TAP
H ′s,min = [h′0,min . . . h′7,min ] (33) channel are Rayleigh distributed with equal variances.
The taps of the ProB and 3TAP channels are Tc –spaced
which is composed of column vectors contain- h′ν,min (Tc = 90.9 ns). For these channels, memories of qh = 2
ing the scalar minimum–phase equivalent hmin [·] of (corresponding to 16 trellis states and 16 branches per
the scalar channel impulse response h[·], hyper–branch) result, respectively.
h Fig. 4 shows the BER and packet error rate (PER)
h′ν,min = 0Tν hmin [0] versus Eb /N0 (Eb : received energy per bit), respec-
iT tively. One packet consists of K = 1000 code words. A
. . . hmin [qh ] 0T8 (q̃h +1)−(qh +1)−ν
. good agreement of the simulated BER with the derived
(34) performance approximation for MLSE can be observed
It can be easily checked that the resulting impulse at moderate Eb /N0 already. For high Eb /N0 both
response fulfills Eqs. (30) and (31). For transforming curves are identical. Hence, the derived performance
H s (z) into H s,min (z) via FF filtering, block pro- approximation is an efficient tool to estimate the BER
cessing is not necessary in this special case. Instead, of MLSE.
filtering the received sequence first at chip level with 3 The reason for this is that a standard (optimum) block DFE results
an allpass filter F (z) = Hmin (z)/H(z) and then in a reference tap close to the identity matrix. Therefore, correlations
collecting the filtered chip sequence in words suffices. of the chips of the error vector cannot be avoided.
0 0
10 10
−1
10
−2
10
−1
10
−3
10
BER →
PER →
−4
10
−2
10
−5
10
−6
10
MLSE, AWGN MLSE AWGN
MLSE, 3TAP MLSE 3TAP
−7
MLSE, ProB −3
MLSE ProB
10 10
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
10⋅log10(Eb/N0) [dB] → 10⋅log10(Eb/N0) [dB] →
a) b)
Fig. 4. a) BER and b) PER versus Eb /N0 for MLSE (solid lines) and the derived performance approximation for MLSE (dashed lines). ’◦’:
AWGN channel, ’+’: 3TAP channel, ’△’: ProB channel.
0 0
10 10
−1
10
−2
10
−1
10
−3
10
BER →
PER →
−4
10
−2
10
−5
10
Fig. 5. a) BER and b) PER versus Eb /N0 for DFE (solid lines) with scalar and matrix FF filter and the derived performance approximation
for MLSE (dashed lines). ’◦’: Office A channel and scalar FF filter, ’+’: Office A channel and matrix FF filter, ’△’: Residential C channel and
scalar FF filter, ’×’: Residential C channel and matrix FF filter.
TABLE III
(JTC) to characterize indoor radio environments [19].
JTC C HANNEL PROFILES : O FFICE A AND R ESIDENTIAL C
Three channel profiles (A, B, C) for each of three
different propagation environments (office, residential,
Channel Channel
Office A Residential C and commercial area) are standardized. The JTC pro-
files Office A and Residential C are depicted in Table
Excess Rel. Excess Rel.
Tap Delay Att. Delay Att. III (delay and relative attenuation (Rel. Att.) of each
(ns) (dB) (ns) (dB) path). Regarding the large excess delays of the JTC
1 0 0 0 -4.6 channels and multipath spacings unequal to multiples
2 50 -3.6 50 0 of Tc , simulations for MLSE are not feasible. However,
3 100 -7.2 150 -4.3
in addition to DFE with both FF filtering approaches,
4 225 -6.5
5 400 -3.0 the proposed performance approximation for MLSE is
6 525 -15.2 also shown in Fig. 5.
7 750 -21.7 For low–to–moderate Eb /N0 the matrix FF filter at
symbol level yields a slightly better performance than
scalar MMSE FF filtering. As expected, at high Eb /N0
The performance of DFE with scalar and matrix both DFE schemes achieve the same performance, since
FF filter, respectively, is compared in Fig. 5. Here, the MMSE matrix FF filter is equivalent to optimum FF
the comparison is based on multipath Rayleigh fading filtering at chip level in this case, cf. Section 4.3.
models developed by the Joint Technical Committee
6 Conclusions [17] N. Al-Dhahir and A.H. Sayed. The Finite-Length Multi-Input
Multi-Output MMSE-DFE. IEEE Trans. Signal Processing,
48:2921–2936, October 2000.
In this paper, the performance of various receiver [18] W.H. Gerstacker and D.P. Taylor. On Prefiltering for Reduced–
concepts for complementary code keying (CCK) trans- State Equalization of MIMO Channels. Accepted for ITG
mission over ISI channels has been investigated. Inter- Conference on Source and Channel Coding, Erlangen, 2004.
[19] Joint Technical Comittee on Wireless Access. Final Report on
preting CCK as a block code, a performance approxi- RF Channel Characterization, Sept. 1994.
mation for optimum MLSE is derived and validated by
simulations. Furthermore, MMSE–DFE with optimum
scalar and matrix feedforward filters, respectively, is
proposed. Our results show, that the performance for
matrix filtering is slightly better than that for scalar
filtering at low–to–moderate SNRs, whereas both ap-
proaches are equivalent at high SNRs.
References
[1] IEEE Standard 802.11b, Part 11: Wireless LAN Medium Access
Control (MAC) and Physical Layer (PHY) Specifications –
Higher Speed Physical Layer Extension in the 2.4 GHz Band,
1999.
[2] J.G. Proakis. Digital Communications. Mc Graw Hill, 2000.
Boston.
[3] M.A. Webster et al. Rake Receiver with Embedded Decision
Feedback Equalizer, May 2001. United States Patent, No.: US
6,233,273 B1.
[4] M.V. Clark, K.K. Leung, B. McNair and Z. Kostic. Outdoor
IEEE 802.11 Cellular Networks: Radio Link Performance. In
IEEE International Conference on Communications, pages 512
–516, April 2002. New York.
[5] H. Luo and R.-W. Liu. Apply Autocorrelation Matching Method
to Outdoor Wireless LAN on Co–Channel Interference Suppres-
sion and Channel Equalization. In Wireless Communications
and Networking Conference, pages 459–464, March 2002.
[6] M.J.E. Golay. Complementary Series. IRE Trans. Inform.
Theory, pages pp. 82–87, April 1961.
[7] C.C. Tseng and C.I. Liu. Complementary sets of sequences.
IEEE Trans. Inform. Theory, vol. IT-8(no. 5):pp. 644–651, Sept.
1972.
[8] R. Sivaswamy. Multiphase complementary codes. IEEE Trans.
Inform. Theory, vol. IT-24(no. 5):pp. 546–552, Sept. 1978.
[9] R.L. Frank. Polyphase complementary codes. IEEE Trans.
Inform. Theory, vol. IT-26(no. 6):pp. 641–647, Nov. 1980.
[10] K. Halford, S. Halford, M. Webster and C. Andren. Comple-
mentary code keying for RAKE-based indoor wireless commu-
nication. In IEEE International Symposium on Circuits and
Systems, pages 427–430, May 1999.
[11] Y. Kakura. A Path Restricted Sequence Estimator using Code-
word Characteristics. In 11th IEEE International Symposium
on Personal, Indoor and Mobile Radio Communications, pages
1305–1309, Sept. 2000. London.
[12] R. Schober, W.H. Gerstacker and L. Lampe. Performance
Analysis and Design of STBC’s for Frequency-Selective Fad-
ing Channels. To appear in IEEE Transactions on Wireless
Communications, 2003.
[13] E. Biglieri, G. Caire, G. Taricco and J. Ventura-Travest. Com-
puting Error Probabilities over Fading Channels: a Unified
Approach. European Trans. on Telecom., 9(1):15–25, Jan.-Feb.
1998.
[14] C. Heegard, S. Coffey, S. Gummadi, E.J. Rossin, M.B. Shoe-
make and M. Wilhoyte. Combined Equalization and Decoding
for IEEE 802.11b Devices. IEEE Selected Areas on Com.,
21(2):125–138, Feb. 2003.
[15] D. Williamson, R.A. Kennedy and G.W. Pulford. Block decision
feedback equalization. IEEE Trans. on Com., 40(2):255–264,
Feb. 1992.
[16] A. Stamoulis, G. Giannakis and A. Scaglione. Block FIR
Decision-Feedback Equalizers for Filterbank Precoded Trans-
missions with Blind Channel Estimation Capabilities. IEEE
Trans. on Com., 49(1):69–83, Jan. 2001.