Escolar Documentos
Profissional Documentos
Cultura Documentos
Transactions Papers
Abstract—This paper proposes a general framework of Space- latter possesses a number of advantages, compared to the
Time-Frequency Codes (STFCs) for Multi-Band Orthogonal former, in terms of low complexity of receivers to capture
Frequency Division Multiplexing (MB-OFDM) Ultra-Wide Band sufficient multipath energy, and in easier radio frequency
(UWB) communications systems. A great similarity between
the STFC MB-OFDM UWB systems and conventional wire- (RF) design. One of the main candidates for carrier-based
less Complex Orthogonal Space-Time Block Code (CO STBC) UWB communications is Multi-Band Orthogonal Frequency
Multiple-Input Multiple-Output (MIMO) systems is discovered. Division Multiplexing (MB-OFDM), supported by the Wi-
This allows us to quantify the pairwise error probability (PEP) Media Alliance [1]. MB-OFDM UWB is designed for very
of the proposed system and derive the general decoding method high bit rate up to 480 Mbps with low cost and low power
for the implemented STFCs. Based on the theoretical analysis
results of PEP, we can further quantify the diversity order and consumption.
coding gain of MB-OFDM UWB systems, and derive the design On the other hand, one of the emerging techniques to
criteria for STFCs, namely diversity gain criterion and coding resolve the bottleneck of traffic capacity in wireless networks
gain criterion. The maximum achievable diversity order is found
to be the product of the number of transmit antennas, the number is the use of Multiple-Input Multiple-Output (MIMO) sys-
of receive antennas, and the FFT size. We also show that all tems. Communication theory [2], [3], [4] shows that MIMO
STFCs constructed based on the conventional CO STBCs can systems can provide potentially a very high capacity that, in
satisfy the diversity gain criterion. Various baseband simulation many cases, grows approximately linearly with the number
results are shown for the Alamouti code and a code of order 8. of antennas, without paying any additional power. The main
Simulation results indicate the significant improvement achieved
in the proposed STFC MB-OFDM UWB systems, compared to feature of MIMO systems is space-time processing. Space-
the conventional MB-OFDM UWB ones. Time Codes (STCs) are the codes designed for the use in
MIMO systems. In STCs, signals are coded in both temporal
Index Terms—UWB, MB-OFDM, STFC, MIMO, STC, CO
STBC, design criteria, diversity order, coding gain. and spatial domains. Among a variety of STCs, of particular
interest are Complex Orthogonal Space-Time Block Codes
(CO STBCs), which possess a simpler decoding method than
I. I NTRODUCTION other STCs, such as Space-Time Trellis Codes (STTCs) [5].
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:49 from IEEE Xplore. Restrictions apply.
702 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 2, FEBRUARY 2009
Hopping (TFC)
OFDM Modulation
Convolutional
STC MIMO communications, to find the diversity and coding
Bit Interleaver
Constellation
(QPSK/DCM)
Frequency
Scrambler
(Insert Pilots, GI
Puncturer
& Filtering
Mapping
gains of the proposed STFC MB-OFDM UWB system in the
Ecoder
log-normal distribution case [17], without any restriction or
UWB Channel
additional assumption on the time delay and average power
of transmit-receive links. Our analysis is based closely on
OFDM Demodulation
Bit Deinterleaver
Constellation
Demapping
Dehopping
Depuncturer
(QPSK/DCM)
Descrambler
Frequency
(Suffix Processing,
IEEE 802.15.3a UWB channel model. We discover that the
Filtering &
Decoder
Viterbi
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:49 from IEEE Xplore. Restrictions apply.
TRAN and MERTINS: SPACE-TIME-FREQUENCY CODE IMPLEMENTATION IN MB-OFDM UWB COMMUNICATIONS 703
TABLE I
Band Group 1 Band Group 2 Band Group 3 Band Group 4 Band Group 5
N UMBERS OF M ULTIPATHS N p10dB , N p85% , AND N̄p [15].
Band 1 2 3 Band 4 5 6 Band 7 8 9 Band 10 11 12 Band 13 14
CM 1 CM 2 CM 3 CM 4
f N p10dB 12.5 15.3 24.9 41.2
3432 3960 4488 5016 5544 6072 6660 7128 7656 8184 8712 9240 9768 10296 N p85% 20.8 33.9 64.7 123.3
MHz
N̄p 287.9 739.5 1463.7 3905.5
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:49 from IEEE Xplore. Restrictions apply.
704 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 2, FEBRUARY 2009
OFDM Modulation
xOFDM xZP
coefficients hm,n,l of the l-th path, l = 1, . . . , Lm,n , in this
Padding
s channel are modeled as independent log-normally distributed
IFFT
Zero
Tx 1
Convolutional
Constellation
Space-Time
(QPSK/DCM)
RVs. Let Lmax = max{Lm,n }, for m = 1, . . . , M, n =
Frequency
Interleaver
Puncturing
Ecoder &
Encoder
Mapping
. .
. .
S/P
. .
Padding
IFFT
matrix as
Zero
Tx M
and-Add
Overlap-
FFT
Rx 1
Deinterleaver
Space-Time
Constellation
Demapping
(QPSK/DCM)
Frequency
Decoder
. .
Decoder
P/S
. .
& Viterbi
r
rOFDM
and-Add
rZP
Overlap-
At the transmission of the t-th MB-OFDM symbol, the
FFT
Rx N
OFDM Demodulation
received signal at the n-th Rx antenna is calculated as
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:49 from IEEE Xplore. Restrictions apply.
TRAN and MERTINS: SPACE-TIME-FREQUENCY CODE IMPLEMENTATION IN MB-OFDM UWB COMMUNICATIONS 705
After going through the FFT block at the receiver, the may be considered as interferences for the received signals.
received signal becomes Eq. (7) represents the real received signals at the Rx antennas,
while (12) shows the realistic concept used at the STFC
M
F F T {r̄OF DM,t,n } = s̄t,m • h̄m,n + F F T {n̄t,n } decoder to decode the original transmitted signals. Therefore,
m=1 in order to simulate the realistic performance of the proposed
(11) system, the signals received at the Rx antennas should be
calculated from (7) with the linear convolution between the
Denote transmitted MB-OFDM symbols and the fully long multipath
r̄t,n = [rt,n,1 , rt,n,2 , . . . , rt,n,Nf f t ]T = F F T {r̄OF DM,t,n } channels, while decoding algorithm should be carried out
based on (12), i.e. based on the circular convolution.
and On the other hand, for a channel vector h̄m,n =
n̄t,n = [nt,n,1 , nt,n,2 , . . . , nt,n,Nf f t ]T = F F T {n̄t,n }. [hm,n,1 , hm,n,2 , . . . , hm,n,Lm,n ]T , we always have the fol-
lowing property for the Nf f t -point FFT operation
Then (11) can be rewritten as follows
F F T {h̄m,n } = F F T {[hm,n,1, hm,n,2 , . . . , hm,n,Lm,n ]T }
M
r̄t,n = s̄t,m • h̄m,n + n̄t,n . (12) if the length Lm,n of the vector is not smaller than Nf f t .
m=1 This means that, by FFT-ing the received signals with a limited
Recall that s̄t,n is the original QPSK or DCM transmitted FFT size Nf f t , and decoding signals based on (12), the Nf f t -
signal (before IFFT). point FFT operation truncates the impact of a long vector
Denote H = {h̄m,n }M×N to be the matrix whose ele- h̄m,n to the length of Nf f t . Therefore, the higher Nf f t is,
ments are the Nf f t -point FFTs of the respective elements the closer the approximation between the linear convolution
in the channel-coefficient matrix H. Further, denote R = and the circular convolution is, and thus the better the system
{r̄OF DM,t,n }T ×N to be the received signal matrix, R = performance is. However, FFT and IFFT blocks significantly
{r̄t,n }T ×N to be the received signal matrix after FFT, and decide the complexity and the cost of transmitter and receiver.
N = {n̄t,n }T ×N to be a noise matrix. We can rewrite (12) in As a result, there must be a suitable compromise between the
matrix form as follows cost/complexity and the system performance.
R = S ◦ H + N. (13)
IV. PAIRWISE E RROR P ROBABILITY
From (13), we can realize that there exists a similarity
Due to the similarity between the mathematical model of the
between the mathematical model of the STFC MB-OFDM
STFC MB-OFDM system and that of the conventional wire-
UWB system and that of the conventional wireless STC
less STC MIMO system as shown in (13), we can calculate
MIMO system [16], [18], [22]. The only difference between
the PEP of the proposed STFC MB-OFDM system, based on
the two mathematical models is that the matrix elements are
Tarokh’s method mentioned in [5] for conventional wireless
scalar numbers in the conventional STC MIMO system, while
STC MIMO communications, with proper modifications to
they are Nf f t -length vectors in the STFC MB-OFDM UWB
account for the MB-OFDM UWB channels. We consider a
system.
STFC MB-OFDM MIMO system with M Tx, N Rx antennas,
Because the vector elements in S will be transformed
and with the STFC S defined in (3). We denote Es to be
with the IFFT to generate MB-OFDM symbols with Nf f t
the average energy of the signal constellation. The following
subcarriers, we refer to S as a Space-Time-Frequency Code.
analysis is derived without considering a specific modulation
scheme, thus can be applied to different modulation schemes.
B. Realistic Channel Condition We consider the probability that a ML receiver decides
The error performance of the proposed system with realistic erroneously in favor of a signal
UWB channel conditions is inferior, compared to the theoret-
ical performance, due to the following two main reasons. e = [e1,1,1 . . . e1,1,Nf f t , e1,2,1 . . . e1,2,Nf f t , . . . ,
In theory, the length of CP or ZPS must be longer than the eT,M,1 . . . eT,M,Nf f t ] (14)
longest multipath in an OFDM-based system to turn the linear
assuming that
convolution between the transmitted signal and the channel
vector into the circular convolution. However, in practice, the c = [c1,1,1 . . . c1,1,Nf f t , c1,2,1 . . . c1,2,Nf f t , . . . ,
multipath length is very likely to exceed the length of CP
cT,M,1 . . . cT,M,Nf f t ] (15)
or ZPS. This is especially true in MB-OFDM UWB systems
where the average number of multipaths N̄p is usually much was transmitted. Note that each group of Nf f t consecutive
bigger than NZP S = 37 (see Table I). The transition from data inside c in fact form a certain vector s̄ as defined in (3).
(7) to (12) is an approximation, due to the fact that the If the transmission coefficients are known at the receiver,
circular convolution in (9) is approximate, but not exactly the probability of transmitting c and deciding e at the decoder
equal to the first Nf f t samples achieved by the OAAO of is well approximated by the Chernoff bound
the linear convolution x̄ZP,t,m ⊗ h̄m,n in (7). The energy
of multipath components within the ZPS window will be P (c → e|m,n,k , m = 1, . . . , M ; n = 1, . . . , N ;
captured, while the multipath components outside this window k = 1, . . . , Nf f t ) ≤ exp(−d2 (c, e)Es /4N0 ) (16)
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:49 from IEEE Xplore. Restrictions apply.
706 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 2, FEBRUARY 2009
2
where N0 /2 is the noise variance per dimension and Log − N (μZ , σZ ), for instance, using the Fenton-Wilkinson
approximation [23] with the following distribution parameters
N
T N fft
M
2
2
d (c, e) = m,n,k (ct,m,k − et,m,k ) . J 2 2
2
e2μi +σi (eσi − 1)
n=1 t=1 k=1 m=1 σZ = ln 1 + i=1
J 2
(17) ( i=1 eμi +σi /2 )2
J
2 σ2
Setting Ωn,k = (1,n,k , . . . , M,n,k ) for n = 1, . . . , N, k = μZ = ln eμi +σi /2 − Z
1, . . . , Nf f t , we rewrite (17) as i=1
2
N N
fft M M
Note that, because J is a finite number, the central-limit
d2 (c, e) = m,n,k ∗ḿ,n,k theorem cannot be applied in this case. Also, the distribution
n=1 k=1 m=1 ḿ=1 parameters of a log-normally distributed RV x should be
T
understood to be the mean and variance of the normally
(ct,m,k − et,m,k )(ct,ḿ,k − et,ḿ,k )∗ .(18) (Gaussian) distributed RV associated with x.
t=1
Expanding this property for the case
With the square, order-M matrix Ak (c, e) = {Ap,q,k }, where
J
T Y = pi eai Xi (24)
∗
Ap,q,k = (ct,p,k − et,p,k )(ct,q,k − et,q,k ) (19) i=1
t=1
where ai are real or complex numbers, and pi = ±1, based
for p, q = 1, . . . , M , Eq. (18) becomes
on Eqs. (15)–(17) in [23] and with simple manipulations, we
N N fft can deduce the distribution parameters of the approximate log-
2 normally distributed RV Z as follows
d (c, e) = Ωn,k Ak (c, e)ΩH
n,k . (20)
n=1 k=1
J 2 2
2
p2i e2ηi +σi (eσi − 1)
It is clear that Ak (c, e) is Hermitian, i.e. Ak (c, e) = σZ = ln 1 + i=1
J 2
Ak (c, e)H . Hence, there exits a unitary matrix Vk and a real ( i=1 pi eηi +σi /2 )2
diagonal matrix Dk such that Vk Ak VkH = Dk . The diagonal
J
2 σ2
elements of Dk are the eigenvalues λm,k , for m = 1, . . . , M , μZ = ln pi eηi +σi /2 − Z (25)
2
and k = 1, . . . , Nf f t , of Ak (c, e). By construction, the matrix i=1
⎡ ⎤ where ηi = μi + ai .
c1,1,k − e1,1,k . . . c1,M,k − e1,M,k
⎢ c2,1,k − e2,1,k . . . c2,M,k − e2,M,k ⎥ Since the average number of multipaths of UWB channels
Bk (c, e) = ⎢ ⎣
⎥
⎦ N̄p is much higher than Nf f t (see Table I), the elements
... ... ...
cT,1,k − eT,1,k . . . cT,M,k − eT,M,k m,n,k of the channel vector h̄m,n , which is the Nf f t -point
(21) FFT of the channel vector h̄m,n , can be represented as
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:49 from IEEE Xplore. Restrictions apply.
TRAN and MERTINS: SPACE-TIME-FREQUENCY CODE IMPLEMENTATION IN MB-OFDM UWB COMMUNICATIONS 707
parameters of Zm,n,k are derived as follows Denote Km,n,k to be the mean of |βm,n,k |2 . Thus Km,n,k is
Nf f t 2ηm,n,k,l +θ2 2 the second moment of |βm,n,k |, and therefore [17]
2
l=1 e
m,n,l (eθm,n,l − 1)
σZm,n,k = ln 1 + Nf f t 2 Km,n,k =
2
e(2μm,n,k +2σm,n,k ) (32)
( l=1 pm,n,l eηm,n,k,l +θm,n,l /2 )2
Nf f t
2
2
σZ If |βm,n,k | is a log-normally distributed RV, so is |βm,n,k |2 .
μZm,n,k = ln pm,n,l eηm,n,k,l +θm,n,l /2 − m,n,k From (16), (20), and (23), to compute the upper bound on the
2
l=1 average probability of error, we simply average
(28)
N N
fft M
where ηm,n,k,l = Ehm,n,l − j2π(l−1)(k−1)
Nf f t . exp − (Es /4N0 ) λm,k |βm,n,k |2 (33)
From the mutual relation between the mean and variance of n=1 k=1 m=1
a log-normally distributed RV and the normally distributed RV with respect to independent log-normal distribution of
associated with it [17], we can quantify the mean and variance |βm,n,k |2 . We denote Ξm,n,k = |βm,n,k |2 .
of m,n,k , denoted as Em,n,k and 2m,n,k respectively, as It is noted that, if Ξm,n,k is a log-normally distributed RV
follows (i.e. Ξm,n,k = ey where y ∼ N (μ, σ 2 ), and μ, σ 2 are finite
2
μZm,n,k +σZ /2 numbers) and α is a positive coefficient, it is possible to prove
Em,n,k = e m,n,k
2 2
that E{e−αΞm,n,k } ≤ α1 e−E{Ξm,n,k } for a sufficient large α,
2μZm,n,k +σZ
2m,n,k
σZ
= (e m,n,k − 1)e m,n,k (29) i.e. α ≥ α0 (see Appendix A). Therefore, if a sufficiently large
signal-to-noise ratio Es /N0 is considered, we have
As a result, the Nf f t -point FFT of a channel vector h̄m,n
−rN Nf f t
Nf f t r
consisting of Lm,n (Lm,n > Nf f t ) real multipath compo- −N
nents hm,n,l (for l = 1, . . . , Lm,n ), whose magnitudes are P (c → e) ≤ Es /4N0 λm,k
independent log-normally distributed RVs, produces a channel k=1 m=1
N N
vector h̄m,n consisting of Nf f t complex multipath compo- fft r
from the mathematical viewpoint, if the FFT size Nf f t is not where r = min{rk }, and rk , for k = 1, . . . , Nf f t , is the rank
smaller than the average number of multipath components N̄p , of matrix Bk (c, e).
which is usually not the case of MB-OFDM UWB systems, We realize that a diversity order of rN Nf f t and a coding
then the Nf f t -point FFT of h̄m,n produces a vector h̄m,n gain (over MB-OFDM systems without STFCs) of
consisting of N̄p (rather than Nf f t ) independent log-normally
Nf f t r
distributed RVs. 1/rNf f t
λm,k ×
Since Vk is unitary, the rows {v1,k , v2,k , . . . , vM,k } of k=1 m=1
Vk are a complete orthonormal basis of the complex M - −1/rN Nf f t
N Nfft
r
dimensional space C M , and thus, from (22) and the general exp(−Km,n,k ) (35)
property (24) of a sum of independent log-normally dis- n=1 k=1 m=1
tributed RVs, βm,n,k (for m = 1, . . . , M , n = 1, . . . , N and
are achieved. Therefore, the maximum achievable diversity
k = 1, . . . , Nf f t ) can also be reasonably well approximated
order in STFC MB-OFDM UWB systems is the product
to be complex, independent log-normally distributed RVs.
between number of Tx antennas, the number of Rx antennas,
Note that if βm,n,k is a complex log-normally distributed
and the FFT size.
RV, then |βm,n,k | is a real log-normally distributed RV. That
We note that the maximum diversity order is defined here as
is because we can always represent βm,n,k in the form
the maximum diversity order of the outgoing signals from the
βm,n,k = eω = eωR +jωI , where ωR and ωI are the real and
OFDM demodulation block in Fig. 1, which are evaluated by
imaginary parts of the complex, normally distributed RV ω.
(12). The maximum diversity order of the incoming signals
Therefore, |βm,n,k | = eωR is a real, log-normally distributed
received at Rx antennas (before the OFDM demodulation
RV with the associated normally distributed RV ωR .
block, calculated by (7)) may be very large due to the very
Denote the mean and variance of |βm,n,k | as Em,n,k
2 dispersive multipath channel, while the maximum diversity
and γm,n,k respectively, then |βm,n,k | are independent log-
order of the outgoing signals from the OFDM demodulation
normally distributed RVs with the following pdf
block is limited because the FFT size is normally very
1 limited, compared to the full length of multipaths. Thus the
p(|βm,n,k |) = √ exp − (ln(|βm,n,k |)
|βm,n,k |σm,n,k 2π maximum diversity order of the outgoing signals from the
OFDM demodulation block should be considered, rather than
−μm,n,k )2 /(2σm,n,k
2
) (30)
the incoming signals. The former represents the effect of an
where important technical specification of the system, i.e. the FFT
1
2
γm,n,k size of the OFDM demodulation block, while the latter does
μm,n,k = ln(Em,n,k ) − ln 1 + 2 not.
2 Em,n,k
2
It is interesting that the diversity order achieved by our
2
γm,n,k approach agrees with that mentioned in Eq. (23) of the
σm,n,k = ln 1 + 2 . (31)
Em,n,k independent work [12], provided we consider the case where
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:49 from IEEE Xplore. Restrictions apply.
708 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 2, FEBRUARY 2009
and the number of jointly encoded OFDM symbols. Further- h̄∗7 • r̄3 + h̄∗6 • r̄2 − h̄∗3 • r̄7 − h̄∗8 • r̄3 − h̄7 • r̄∗4 +
more, our PEP formulation is carried out directly for the case h̄∗4 • r̄7 + h̄3 • r̄∗8 + h̄4 • r̄∗8 − s̄|.ˆ2+
(−1̄ + 2 8m=1 |h̄m |.ˆ2) • (|s̄|.ˆ2) 2F
of log-normal distribution of UWB channel magnitudes, which
has not been examined in [12]. Our PEP is formulated without
assumptions on the same time delays and the same average s̄4 arg mins̄∈C ND | h̄∗7 • r̄2 − h̄2 • r̄∗7 − h̄∗1 • r̄8 −
power of all transmit-receive links as in [12]. Additionally, h̄2 • r̄8 + h̄3 • r̄5 − h̄7 • r̄∗1 + h̄8 • r̄∗1 − h̄∗5 • r̄3 +
∗ ∗
h̄∗6 • r̄3 + h̄1 • r̄∗7 + h̄∗8 • r̄2 − h̄∗4 • r̄5 + h̄5 • r̄∗4 +
the approach in [12] does not reflect clearly the essence of h̄6 • r̄∗4 − h̄3 • r̄∗6 − h̄4 • r̄∗6 − s̄|.ˆ2+
the effect of the FFT operation to the diversity order. In
(−1̄ + 2 8m=1 |h̄m |.ˆ2) • (|s̄|.ˆ2) 2F
contrast, our approach shows clearly that the FFT size actually
decides the number of independent log-normally distributed
RVs |βm,n,k |, that in turn decides the diversity order. If the
FFT size Nf f t is not smaller than the average number N̄p taken over distinct codewords c and e has to be maxi-
of multipath components, then Nf f t in (34) must be replaced mized.
with N̄p , because there are only N̄p independent RVs among
Nf f t variables |βm,n,k |, for certain numbers m and n, thus the Interestingly, all CO STBCs proposed in the literature for
maximum diversity order is M N N̄p . However, this is usually conventional STBC MIMO systems can satisfy the diversity
not the case for MB-OFDM UWB channels due to the fact gain criterion. This is because, a CO STBC S transmitted
that UWB channels are richly dispersive, while the FFT size is through a wireless system with M Tx antennas during T time
limited. Finally, our approach also clearly shows that, within slots can be defined [18], [24] as a T × M -size matrix whose
a range up to N̄p , the higher the FFT size is, the lower the nonzero entries are the indeterminates s1 , s2 , . . . , sk , √ their
error bound in (34) is. This agrees with the analysis mentioned conjugates s∗1 , s∗2 , . . . , s∗k , or their products with j = −1
previously in Section III-B. over the complex number field C, such that
SH S = D (36)
V. D ESIGN C RITERIA FOR STFC S IN MB-OFDM UWB
Assuming that all the multipath components are known at where D is a diagonal matrix of size M × M with diagonal
the receiver. From (34), we derive the following design criteria entries Di,i , for i = 1, 2, . . . , M , of the form (li,1 |s1 |2 +
of STFCs in MB-OFDM UWB systems for a large SN R to li,2 |s2 |2 + · · · + li,k |sk |2 ). The coefficients li,k are strictly
minimize the upper bound of error probability positive, real numbers. Since the time delay for transmitting
• Diversity Gain Criterion: In order to achieve the maxi- the CO STBC T is at least equal to M , from (36), the rank
mum diversity order of M N Nf f t , the minimum rank of of S is always equal to M . Therefore, if the structures of
matrices Bk (c, e), for k = 1, . . . , Nf f t , over any two conventional CO STBCs are used to create STFCs in MB-
distinct code words c and e must be equal to M . If OFDM UWB systems, the diversity gain criterion is always
the minimum rank is r, then a diversity of rN Nf f t is guaranteed. In other words, from the diversity viewpoint, the
achieved. design criteria are the same for both conventional STBC
• Coding Gain Criterion: Suppose that a diversity of order MIMO system and STFC MB-OFDM UWB system.
rN Nf f t is of our interest. The minimum of the product As an example, the Alamouti code [16] for two Tx antennas
is a STFC providing a full diversity of 2N Nf f t for any
Nf f t r
1/rNf f t two distinct codewords c and e. Our proposed codes in
λm,k ×
k=1 m=1 [18] for 8 Tx antennas are some examples, among various
−1/rN Nf f t other examples of STFCs for 8 Tx antennas, providing a full
N Nfft
r
exp(−Km,n,k ) diversity order of 8N Nf f t , for any distinct codewords c and
n=1 k=1 m=1 e.
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:49 from IEEE Xplore. Restrictions apply.
TRAN and MERTINS: SPACE-TIME-FREQUENCY CODE IMPLEMENTATION IN MB-OFDM UWB COMMUNICATIONS 709
VI. M AXIMUM L IKELIHOOD D ECODING A LGORITHM metrics for data at the k-th sub-carrier, for k = 1, . . . , ND , in
the MB-OFDM symbols s̄1 and s̄2 are
From (13), the ML decoding expression can be derived in
the most general form as follows s1,k = arg min | ∗1,k r1,k + 2,k r∗2,k − s|2 +
s∈C
2
{s̄dec,t,m} = arg min R − S ◦ H 2F . (37)
{s̄t,m } (−1 + |m,k |2 )|s|2 (40)
This decoding metric is too complicated to be performed if m=1
s2,k = arg min | ∗2,k r1,k − 1,k r∗2,k − s|2 +
S has non-orthogonal structures. Fortunately, because S has s∈C
the similar structures as the conventional CO STBCs, the 2
orthogonality of S is preserved for the MB-OFDM symbols (−1 + |m,k |2 )|s|2 .
s̄t,m transmitted inside the code block S. As a result, the m=1
simplicity of decoding CO STBCs in conventional wireless For the DCM scheme, the decoding process is slightly more
STBC MIMO systems is also preserved in STFC MB-OFDM complicated, compared to PSK or QAM modulation schemes.
UWB systems, i.e., each MB-OFDM symbol s̄t,m can be A pair of data points sm,k and sm,k+50 , where sm,k and
sm,k+50 are the data points within the MB-OFDM symbol
decoded separately, rather than jointly. Therefore, decoding s̄m and are separated from each other by 50 tones, must be
metrics for the MB-OFDM symbols s̄t,m can be easily found, simultaneously decoded. For instance, decoding metrics for
based on those decoding metrics for the respective CO STBC s1,k and s1,k+50 , for k = 1, . . . , ND /2, are as follows
of S, with slight modifications. Furthermore, each data point
[s1,k , s1,k+50 ] = arg min
among ND data points (ND = 100 data sub-carriers) within s∈CDCM ,s50 ∈CDCM,50
∗
a MB-OFDM symbol s̄t,m can also be decoded separately, | 1,k r1,k + 2,k r∗2,k − s|2 + (41)
rather than the whole ND data in a MB-OFDM symbol s̄t,m ∗
| 1,k+50 r1,k+50 + 2,k+50 r∗2,k+50 − s50 |2
are decoded simultaneously. Thus the decoding process is
2
2
completely linear, and relatively simple. +(−1 + |m,k |2 )|s|2 + (−1 + |m,k+50 |2 )|s50 |2
According to the WiMedia’s MB-OFDM UWB specifica- m=1 m=1
tions, the convolutional encoder is utilized at the transmitter. where the complex space CDCM denotes all potential possi-
Therefore, the transmitted binary information will be recov- bilities that the symbol s can take, while the complex space
ered at the receiver by using a Viterbi decoder, after all of the CDCM,50 denotes all potential possibilities that the symbol
signals at the FFT points are separately decoded as mentioned s50 can take. The exact values of CDCM and CDCM,50 can be
above. found in [19].
To illustrate the ML decoding algorithm for STFCs imple- Similarly, for the code S8 , we can easily derive the decoding
mented in MB-OFDM UWB systems, we consider here two metrics of the MB-OFDM symbols s̄1 , s̄2 , s̄3 , and s̄4 by slight
CO STBCs: the Alamouti code for two Tx antennas modification from the detailed decoding metrics mentioned in
[18] for this CO STBC in conventional STC MIMO systems,
s̄1 s̄2
S2 = (38) and arrive at the decoding metrics in Table III. Hence, the
−s̄∗2 s̄∗1
data point sm,k at the k-th tone, for k = 1, . . . , ND , can be
and the code proposed in [18] for 8 Tx antennas decoded separately. For instance, the decoding metric for the
⎡ data s1,k is
s̄1 s̄1 −s̄∗2 s̄∗2 −s̄∗3 s̄∗3 −s̄∗4 s̄∗4 ⎤
⎢ −s̄∗1 s̄∗1 s̄2 s̄2 s̄3 s̄3 s̄4 s̄4 ⎥ s1,k = arg min |(∗8,k r7,k + ∗6,k r5,k + ∗7,k r7,k +
⎢ s̄∗2 −s̄∗2 s̄1 s̄1 −s̄4 s̄4 s̄3 −s̄3 ⎥ ⎥
⎢ s∈C
⎢ −s̄2 −s̄2 −s̄∗1 s̄∗1 s̄∗4 s̄∗4 −s̄∗3 −s̄∗3 ⎥ ∗3,k r3,k + ∗1,k r1,k + 6,k r∗6,k + ∗2,k r1,k −
S8 = ⎢ ⎥
⎢ s̄∗3 −s̄∗3 s̄4 −s̄4 s̄1 s̄1 −s̄2 s̄2 ⎥
⎢
⎢ −s̄3 −s̄3 −s̄∗4 −s̄∗4 −s̄∗1 s̄∗1 s̄∗2 s̄∗2 ⎥
⎥ 5,k r∗6,k + 2,k 2r∗2,k − 1,k r∗2,k + ∗5,k r5,k +
⎣ ⎦
s̄∗4 −s̄∗4 −s̄3 s̄3 s̄2 −s̄2 s̄1 s̄1 ∗4,k r3,k − 3,k r∗4,k + 4,k r∗4,k − 7,k r∗8,k +
−s̄4 −s̄4 s̄∗3 s̄∗3 −s̄∗2 −s̄∗2 −s̄∗1 s̄∗1 8
(39) 8,k r∗8,k ) − s|2 + (−1 + 2 |m,k |2 )|s|2 (42)
m=1
Detailed decoding metrics for these codes can be found in
the following sections. A deduction similar to (42) is applied for the data s2,k , s3,k ,
and s4,k in PSK or QAM schemes. Accordingly, a deduction
as in (42) can be applied for S8 with the DCM scheme.
A. MISO system
First, we consider a MISO system using the Alamouti code B. MIMO system
with PSK or QAM modulation schemes, and with only one Next, we consider the MIMO system with N Rx antennas.
Rx antenna. For simplicity, we denote the channel coefficient Linear combinations of the received signals from N Rx an-
vectors between the Tx antennas and the Rx antenna to be tennas are used to decode the transmitted symbols. Therefore,
h̄m , for m = 1, 2. The decoding metrics for the MB-OFDM the decoding metrics for the transmitted symbol vectors s̄1 ,
symbols are presented in Table II. s̄2 , s̄3 , and s̄4 can be deduced simply by replacing h̄m , for
Clearly, from Table II, the data at each sub-carrier can be m = 1, . . . , M , in Tables II and III by the term h̄m,n ; r̄t
N
decoded separately, rather than jointly. Therefore, the decoding by r̄t,n , the first bracket expression . by n=1 . , while
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:49 from IEEE Xplore. Restrictions apply.
710 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 2, FEBRUARY 2009
TABLE IV 0
10
S IMULATION PARAMETERS .
Parameter Value
−1
10
FFT and IFFT size Nf f t = 128
Data rate 320 Mbps −2
Convolutional encoder’s rate 1/2 10
Convolutional encoder’s constraint length K=7
Convolutional decoder Viterbi −3
10
Decoding mode Hard
BER
Number of transmitted
MB-OFDM symbols 1200 −4
10
Modulation QPSK & DCM MBOFDM,SI2O,CM1
IEEE Channel model CM1, 2, 3 & 4 MBOFDM,SI2O,CM2
ND = 100
−5
Number of data subcarriers 10 MBOFDM,SI2O,CM3
Number of pilot subcarriers NP = 12 MBOFDM,SI2O,CM4
Number of guard subcarriers NG = 10 −6
STFC MBOFDM,2I2O,CM1
STFC MBOFDM,2I2O,CM2
Total number of subcarriers used NT = 122 10
STFC MBOFDM,2I2O,CM3
Number of samples in ZPS NZP S = 37 STFC MBOFDM,2I2O,CM4
Total number of samples/symbol NSY M = 165 −7
10
Number of channel realizations 100 −5 0 5 10 15
SNR(dB)
0
10
Fig. 6. Alamouti STFC MB-OFDM UWB performance with QPSK modu-
lation/demodulation, and with 2 Rx antennas.
−1
10
0
10
−2
10
−1
10
−3
10
BER
−2
10
−4
10
MBOFDM,SISO,CM1
MBOFDM,SISO,CM2 −3
10
−5 MBOFDM,SISO,CM3
10
BER
MBOFDM,SISO,CM4
STFC MBOFDM,2ISO,CM1 −4
−6
10
10 STFC MBOFDM,2ISO,CM2
MBOFDM,SISO,CM1
STFC MBOFDM,2ISO,CM3
MBOFDM,SISO,CM2
STFC MBOFDM,2ISO,CM4 −5
−7 10 MBOFDM,SISO,CM3
10 MBOFDM,SISO,CM4
0 5 10 15
SNR (dB) STFC MBOFDM,2ISO,CM1
−6
10 STFC MBOFDM,2ISO,CM2
STFC MBOFDM,2ISO,CM3
Fig. 5. Alamouti STFC MB-OFDM UWB performance with QPSK modu- STFC MBOFDM,2ISO,CM4
lation/demodulation, and with 1 Rx antenna. −7
10
0 5 10 15
SNR (dB)
M
replacing the term |h̄ |. ˆ 2 in the second bracket Fig. 7. Alamouti STFC MB-OFDM UWB performance with DCM modu-
M mN
m=1
expression by the term m=1 n=1 |h̄m,n |. ˆ 2, respectively. lation/demodulation, and with 1 Rx antenna.
Similar to MISO systems, each data point can be separately
decoded.
the Alamouti code and 1/8 for our order-8 code. Thereby,
the average transmitting power from all M Tx antennas at a
VII. S IMULATION R ESULTS certain time is kept the same for both MB-OFDM systems
To examine the performance of the proposed STFC MB- with and without STFCs. Both modulation schemes QPSK
OFDM UWB systems, we used the codes S2 in (38) and S8 and DCM are simulated. The simulation parameters are listed
in (39), and ran several Monte-Carlo simulations, each with in Table IV.
1200 MB-OFDM symbols. As suggested in [15], 100 channel The simulation results show that a significant improvement
realizations of each IEEE 802.15.3a channel models (CM 1, in bit error ratios (BER) can be achieved with the proposed
2, 3 and 4) were considered for the transmission of each MB- STFC MB-OFDM system, compared to the conventional MB-
OFDM symbol. In simulations, SN R is defined to be the OFDM system. Fig. 5 shows that an improvement of at least
signal-to-noise ratio (dB) per sample in a MB-OFDM symbol 5 dB at BER = 10−4 can be achieved in the Alamouti STFC
(consisting of 165 samples), at each Rx antenna. It means MB-OFDM 2ISO system (2 Tx, 1 Rx antennas), compared
that, at a certain Rx antenna, SN R is the subtraction between to the conventional MB-OFDM SISO system (1 Tx, 1 Rx
the total power (dB) of the received signal corresponding to antennas) with a QPSK scheme, while Fig. 6 shows that
the sample of interest and the power of noise (dB) at that this improvement is of at least 2 dB when the number of
Rx antenna. To fairly compare the error performance of MB- receive antennas is doubled. By comparing Figs. 5 and 6,
OFDM systems with and without STFCs, the average power the use of two Rx antennas advances the error performance
of the signal constellation points in the STFC MB-OFDM of MB-OFDM systems without STFCs (and with STFCs
system is scaled down by a factor of 1/M , which is 1/2 for respectively) by approximately 7.5 dB (4.5 dB respectively)
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:49 from IEEE Xplore. Restrictions apply.
TRAN and MERTINS: SPACE-TIME-FREQUENCY CODE IMPLEMENTATION IN MB-OFDM UWB COMMUNICATIONS 711
0 0
10 10
−1 −1
10 10
−2
10 −2
10
−3
10 −3
10
BER
BER
−4
10 −4
MBOFDM,SI2O,CM1 10
MBOFDM,SI2O,CM2 MBOFDM,SI2O,CM1
−5 MBOFDM,SI2O,CM2
10 MBOFDM,SI2O,CM3
−5
MBOFDM,SI2O,CM4 10 MBOFDM,SI2O,CM3
STFC MBOFDM,2I2O,CM1 MBOFDM,SI2O,CM4
−6
10 STFC MBOFDM,2I2O,CM2 STFC MBOFDM, 8I2O,CM1
−6
STFC MBOFDM,2I2O,CM3 10 STFC MBOFDM, 8I2O,CM2
STFC MBOFDM,2I2O,CM4 STFC MBOFDM, 8I2O,CM3
−7
10 STFC MBOFDM, 8I2O,CM4
−5 0 5 10 15 −7
10
SNR(dB) −5 0 5 10 15
SNR(dB)
Fig. 8. Alamouti STFC MB-OFDM UWB performance with DCM modu-
lation/demodulation, and with 2 Rx antennas. Fig. 10. Order-8 STFC MB-OFDM UWB performance with QPSK modu-
lation/demodulation, and with 2 Rx antennas.
0
0 10
10
−1
−1 10
10
−2
−2 10
10
−3
−3 10
10
BER
BER
−4
−4 10
10 MBOFDM,SISO,CM1
MBOFDM,SISO,CM1
MBOFDM,SISO,CM2 MBOFDM,SISO,CM2
−5
−5 MBOFDM,SISO,CM3 10 MBOFDM,SISO,CM3
10 MBOFDM,SISO,CM4
MBOFDM,SISO,CM4
STFC MBOFDM,8ISO,CM1 STFC MBOFDM,8ISO,CM1
−6
−6 STFC MBOFDM,8ISO,CM2 10 STFC MBOFDM,8ISO,CM2
10 STFC MBOFDM,8ISO,CM3
STFC MBOFDM,8ISO,CM2
STFC MBOFDM,8ISO,CM4 STFC MBOFDM,8ISO,CM4
−7
−7 10
10 −5 0 5 10 15
−5 0 5 10 15
SNR (dB)
SNR (dB)
Fig. 11. Order-8 STFC MB-OFDM UWB performance with DCM modu-
Fig. 9. Order-8 STFC MB-OFDM UWB performance with QPSK modula-
lation/demodulation, and with 1 Rx antenna.
tion/demodulation, and with 1 Rx antenna.
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:49 from IEEE Xplore. Restrictions apply.
712 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 8, NO. 2, FEBRUARY 2009
0.35 E{e
−α x
} E{e
−α x
}
(−E{x}) (−E{x})
e /α e /α
0.3 0.25
Values of E{e−α x} and e−E{x}/α
0.2
0.15
0.15
0.1
0.1
0.05
0.05
0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
α α
(a) (b)
1 −E{x}
Fig. 13. E{e−αx } and α
e as functions of α. (a) μ = 0, σ = 0.5. (b) μ = 0.25, σ = 0.75.
0
10
Therefore, (43) becomes
∞
−1
10
αe−αx p(x)dx ≤ e−E{x} . (45)
0
−2
10
Using the property [25]
10
−4
where δ(x) is the Dirac delta distribution, we have
MBOFDM,SI2O,CM1
MBOFDM,SI2O,CM2 ∞ ∞
−αx
lim p(x)dx =
−5
10 MBOFDM,SI2O,CM3 αe δ(x)p(x)dx
MBOFDM,SI2O,CM4 α→∞ 0 0
STFC MBOFDM,8I2O,CM1
10
−6
STFC MBOFDM,8I2O,CM2 = p(0). (46)
STFC MBOFDM,8I2O,CM3
−7
STFC MBOFDM,8I2O,CM4 By definition of the log-normal distribution [17], we have
10
−5 0 5 10 15 p(0) = 0. Clearly, the left-hand side of (45) approaches zero
SNR(dB)
when α goes to infinity, while the right-hand side is a positive
Fig. 12. Order-8 STFC MB-OFDM UWB performance with DCM modu- constant. Therefore, the inequality (45) (and thus (43)) holds
lation/demodulation, and with 2 Rx antennas. when α is large enough, i.e. α ≥ α0 , where α0 is a function
of μ and σ 2 . The inequality (43) has been proved.
analysis is carried out directly for the log-normal distribution Example 1: For illustration, the simulation 13(a) shows the
of UWB multipath amplitudes in a general scenario, without values of E{e−αx } and α1 e−E{x} as functions of α for the
any restriction on the time delay and average power of case μ = 0, σ = 0.5. We realize that E{e−αx } ≤ α1 e−E{x}
transmit-receive links as in [12]. when α ≥ 2.1.
Example 2: We present another example for μ = 0.25, σ =
A PPENDIX A 0.75. Fig. 13(b) shows that E{e−αx } ≤ α1 e−E{x} when α ≥
P ROOF OF I NEQUALITY 3.8.
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:49 from IEEE Xplore. Restrictions apply.
TRAN and MERTINS: SPACE-TIME-FREQUENCY CODE IMPLEMENTATION IN MB-OFDM UWB COMMUNICATIONS 713
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on July 28, 2009 at 06:49 from IEEE Xplore. Restrictions apply.