Académique Documents
Professionnel Documents
Culture Documents
5, MAY 2014
Abstract—The aim of this paper is to study the bit-loading and ICI, such as time domain windowing [7], frequency-
and power allocation problem in the presence of interference domain equalization [8] or ISI and ICI self-cancellation [9].
(Inter-carrier Interference (ICI) and Inter-Symbol Interference
(ISI)) in Orthogonal Frequency Division Multiplexing (OFDM) Both current and next generation of PLC systems employ
systems. ISI and ICI significantly degrade the performance of frequency band from 2 to 30 MHz (and over), e.g., HPAV1,
OFDM systems and make the resource management optimized HPAV2, IEEE P1901, ITU-T G.9963 [2], [10]. In this fre-
without the assumption of interference less efficient. To solve quency band, many radio applications such as amateur radio,
this problem, an initial solution based on the greedy approach
is proposed in this paper. Then, several reduced complexity urgency and military services have already been exploited.
approaches, which yield a little degradation compared to the To avoid interference with those systems, a spectral mask is
initial solution, have been developed. Simulation results presented specified for PLC systems [10]. The IEEE P1901 standard
in the context of Power Line Communication (PLC) show that uses the Windowed-OFDM instead of the conventional OFDM
the performance of proposed algorithms is tight with their technique to adapt to the spectral mask [10]. In [11], the Her-
upper bound. Moreover, these algorithms efficiently improve the
system performance as compared to the constant power water- mitian Symmetry Offset Quadrature Amplitude Modulation
filling allocation algorithm as well as maximum power allocation (HS-OQAM) is proposed for future use to increase the data
algorithm. rate as well as adapt to new spectral masks in Europe.
Index Terms—Bit-loading, power allocation, ICI, ISI, power Regarding resource allocation, the bit-loading can be de-
line communication, windowed-OFDM, greedy algorithm. signed to achieve different objectives in OFDM systems, such
as bit allocation, power allocation, code rate adaptation, etc.
I. INTRODUCTION
Different algorithms have been proposed among which we
can enumerate: adaptive-rate algorithms which maximize the
N the past decades, the use of PLC systems for high
I rate indoor broadband communications has spread rapidly.
Since PLC doesn’t require any new wire installation, it is
capacity under the power and bit-error rate constraints (BER)
[12], [13]; margin-adaptive algorithms which minimize the
consumed power under data rate and BER constraints [14];
economically attractive for indoor local area network (LAN). BER minimization under data rate and power constraints [15].
It can also be complementary with wireless LAN [1], [2]. If the ISI and ICI are present, the system can be modeled as a
PLC systems exploit the OFDM technique to combat the Gaussian interference channel [16]. In this case, by using joint
effect of multi-path channels with severe frequency selectivity coding and decoding, the system capacity is maximized by
[3], [4]. The conventional OFDM divides the entire bandwidth using the water-filling algorithm derived in [16]. In the absence
into many orthogonal subcarriers and data are transmitted in of joint coding and decoding, the interference is considered
parallel over these subcarriers. Therefore, it can support high as noise when solving the resource allocation problem [17],
transmission data rate and achieves high spectral efficiency. [18], [19], [20], [21]. In [18], the achievable throughput of
Unfortunately, many phenomena such as frequency offset PLC systems is maximized by combining the bit-loading
between transmitter and receiver, insufficient guard interval or algorithm and adaptive cyclic prefix. The bit-loading algorithm
Doppler frequency shift, etc. induce inter-carrier interference used in [18] relies on two simplifying assumptions. First,
(ICI) as well as inter-symbol interference (ISI) that signifi- the conventional OFDM is taken into account to calculate
cantly degrade the performance of OFDM systems [5], [6] ISI and ICI in PLC systems while, as mentioned above,
and make the resource management more complicated. Many PLC systems exploit the Windowed-OFDM instead of the
techniques have been developed to combat the effects of ISI conventional OFDM. The second simplification is enabled by
on/off power loading and integer bit number constraint. The
Manuscript received August 21, 2013; revised December 20, 2013. The
editor coordinating the review of this paper and approving it for publication latter leads to bit discretization that causes non optimal power
was A. Tonello. use. Actually, the discretized bit-loading can be achieved with
This work is financially supported by PRACOM. lower power consumption than with the approach in [18] and
T. Nhan Vo, K. Amis, and T. Chonavel are with the Signal and Com-
munication Department of Télécom Bretagne–Institut Télécom, CNRS Lab- the residual power could be exploited to increase data rate
STICC (UNR 6285), Brest 29200, France (e-mail: {thanh.vo, karine.amis, or/and transmission quality. CP design for maximizing the
thierry.chonavel}@telecom-bretagne.eu). achievable throughput for Windowed-OFDM systems is con-
P. Siohan is with the Orange Labs, Rennes 35510, France (e-mail:
pierre.siohan@orange.com). sidered in [22]. Unfortunately, the power allocation strategy
Digital Object Identifier 10.1109/TCOMM.2014.031614.130660 in [22] is also non optimal due to the bit discretization. In
0090-6778/14$31.00 ⃝
c 2014 IEEE
VO et al.: ACHIEVABLE THROUGHPUT OPTIMIZATION IN OFDM SYSTEMS IN THE PRESENCE OF INTERFERENCE AND ITS APPLICATION . . . 1705
[18], [22], it is shown that the choice of a cyclic prefix (CP) II. S YSTEM M ODEL
length equal to the channel impulse response length makes
Let us consider a Windowed-OFDM system with L subcar-
PLC systems less efficient in terms of achievable throughput.
riers used out of M , which are activated under a given spectral
Shorter CP evidently results in ISI and ICI, but the gain
mask constraint. The demodulated sample on the m0 -th used
offered by shortened CP may exceed the losses caused by
subcarrier and n0 -th OFDM symbol is given by
interference. Another approach to CP length adaptation relies
on the statistical channel state information as derived in [23], y(m0 , n0 ) = α(m0 , n0 )cm0 ,n0 + ICI(m0 , n0 )
[24]. This approach causes a capacity loss when compared to
+ISI(m0 , n0 ) + b(m0 , n0 ), (1)
the bit and power allocation with the instantaneous channel
state information (CSI). In our paper, we assume that perfect where α(m0 , n0 ), cm0 ,n0 , ISI(m0 , n0 ), ICI(m0 , n0 ),
instantaneous CSI is available and we use it to optimize the b(m0 , n0 ) denote the channel multiplicative factor, the
resource allocation. Moreover, our interest is to optimize the symbol of interest, the ISI and ICI coefficients and the
throughput for fixed GI length and we take into account complex circularly Gaussian noise sample at the m0 -th used
possible interference in the bit/power allocation. subcarrier and n0 -th OFDM symbol.
We assume that the channel is block-based time invariant.
Recently, several approaches to search for the upper-bound Without loss of generality and for the sake of simplicity, in a
of the achievable throughput have been studied in [19] but no block of many OFDM symbols, Eq. (1) can be re-written as
practical solution of the achievable throughput maximization
problem with low-complexity has been given. A solution y(m0 ) = α(m0 )cm0 + ICI(m0 ) + ISI(m0 ) + b(m0 ) (2)
based on the greedy approach has been developed in [25]
for the multi-carrier interference channel in multi-users ADSL Since the number of subcarriers used in practical PLC systems
systems. However, it only takes into account the ICI caused is quite large, we assume that the interference on a given
by the asynchronous cross-talk effect. Moreover, due to the subcarrier is normally distributed (following the central limit
difficulty of exact additional power computation, the algorithm theorem) [3], [26]. Different normality tests for the interfer-
in [25] exploits the gradient information to approximate the ence have been introduced in [3]. These tests have confirmed
power adjustment when modifying the number of bits on the validity of Gaussian distribution of the interference in
subcarriers. The algorithm detailed in [25] approximates the practical PLC systems. Then, the signal to interference plus
matrix inversion via power series expansion to reduce the noise ratio (SINR) and the theoretical capacity on the m0 -th
complexity. This method works only if the assumption of con- used subcarrier are given as follows:
vergence of the power series expansion is valid. In this paper, |α(m0 )|2 P (m0 )
we focus on single-user Windowed-OFDM-based systems in SIN R(m0 ) = (3)
PICI (m0 ) + PISI (m0 ) + σb2 (m0 )
the presence of ICI as well as ISI. Our proposed algorithms ( )
are also based on the greedy principle, which is detailed in the SIN R(m0 )
C(m0 ) = log2 1 + (4)
section III. However, a judicious iterative procedure for accu- Γ
rate matrix inversion calculation and an iterative bit-loading
procedure exploiting the incremental power needed to transmit where P (m0 ) is the power allocated to m0 -th used subcarrier
an additional bit are utilized instead of the approximations and Γ is the SNR gap that models the practical modulation
proposed in [25]. and coding scheme for a targeted symbol-error rate (SER):
[ ( )]2
1 SER
The main contribution of this paper is to use the greedy Γ= Q−1 (5)
principle to solve the achievable throughput maximization 3 4
problem in PLC systems in the presence of interference and where Q−1 (x) is the inverse tail probability of the standard
under power and bit-error rate constraints. For this purpose, the normalization distribution [27].
ISI and ICI due to the presence of insufficient cyclic prefix in Let us denote Ause the set of used subcarriers, then
PLC systems have been analyzed. Relying on the ISI and ICI #(Ause ) = L, where # denotes the cardinality of Ause . On a
analysis and the greedy principle, we propose an initial greedy given subcarrier, the interference power generally depends on
algorithm to calculate the achievable throughput maximization the power allocated to other used subcarriers (see Appendix)
in the presence of ISI and ICI. Then, several approaches are and can be written as
proposed in order to reduce the complexity with a negligible
degradation compared to the initial greedy solution. ∑
L
PI (m0 ) = PICI (m0 ) + PISI (m0 ) = W (m0 , m)P (m)
m=1
The rest of the paper is organized as follows. In Section II, (6)
the OFDM model in the presence of ISI and ICI is described.
In Section III, the greedy principle and several approaches to where W (m0 , m) is the interference contribution of the m-th
reduce the complexity are presented. In Section IV, the achiev- used subcarrier on the m0 -th used subcarrier. In other words,
able throughput maximization problem is formulated and an W (m0 , m) is the interference contribution of the m-th entry
initial solution, which exploits the greedy principle is given. of Ause on the m0 -th entry of Ause .
A reduced complexity algorithm is also proposed. Numerical Let W denote the interference matrix with entry (a,b)
results are reported in Section V. Finally, conclusions and equal to W (a, b) for a, b ∈ {1, ..., L}. Let P denote the
perspectives are drawn in Section VI. allocated power vector and N the noise power vector. The total
1706 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 62, NO. 5, MAY 2014
interference power and the SINR on the m0 -th used subcarrier number of bits allocated to the subcarrier cannot be increased.
are given by In this case, to avoid meaningless loops, the cost function
F(m0 , k) is set to infinity and this subcarrier must not be
PI (m0 ) = [WP](m0 ) (7)
considered for additional bits allocation in following iterations.
|α(m0 )|2 P (m0 ) The procedure stops if there is no subcarrier left to allocate
SIN R(m0 ) = (8)
PI (m0 ) + N (m0 ) additional bits.
already demonstrated in [25] where it is shown that if B1 ≤ B2 search. This algorithm becomes intractable as the number of
component-wise, then the corresponding power vectors satisfy subcarriers becomes significant.
P1 ≤ P2 component-wise. It means that when the number of 2) Efficient computation of the cost function: The com-
bits on a subcarrier is increased, the resulting power levels on plexity of the standard greedy algorithm is mainly due to
all subcarriers are higher than or equal to the current ones. matrix inversions in the cost function calculation. To reduce
1708 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 62, NO. 5, MAY 2014
= (I − Λk+1 W)−1
(m) (m)
≈ (I − Λk W)−1 ∆Λk+1 (WPk + N)
(m) (m)
Mk+1 ∆Pk+1
= (I − Λk W − ∆λk+1 em eTm W)−1
(m) (m)
≈ Mk ∆λk+1 (WPk + N) (28)
| {z }
Wm
The required power per additional bit on m-th subcarrier is
(M−1 −1
(m)
= k − em ∆λk+1 Wm ) (m)
Mk em Wm Mk S(∆Pk+1 )
= Mk + ∆Pbit (m, k + 1) = (m)
(m) − Wm Mk em
1 ∆bk+1
∆λk+1
(m)
(m) S(Mk (:, m))∆λk+1 [WPk + N](m)
= Mk + ∆Mk+1 (23) ≈ (m)
(29)
(m) (m) (m) ∆bk+1
Pk+1 = Mk+1 Λk+1 N (24)
where Mk (:, m) is the m-th column vector of matrix Mk . Note
Note that M0 = I because B = 0 or equivalently Λ0 = 0 at the
that the matrix inversion in the calculation of ∆Pbit (m, k + 1)
first iteration. At any iteration, we can derive the power vector
(m) has already been calculated to derive Pk :
Pk+1 when the number of bits on m-th subcarrier is increased.
We choose the m0 -th subcarrier to increase its number of bits Pk = (I − Λk W)−1 Λk N = Mk Λk N (30)
as in Eq. (17). Finally, matrix M, the power vector and the
bit-allocation corresponding matrix are updated to Then, at every iteration, the number of matrix inversions to
calculate is only one. Thus the complexity reduction gain at
(m ) (m ) (m )
Mk+1 , Pk+1 , Λk+1 = Mk+10 , Pk+10 , Λk+10 (25) iteration (k + 1) is of about Lk+1
on as compared to the standard
(m) greedy algorithm.
Due to the simple expression of em , the complexity of ∆Mk+1 2) Reduction of the number of iterations: The choice of
calculation is only O(2L2 ). Thus, the total complexity at null initial bit-allocation vector B = 0 or equivalently P = 0
iteration (k + 1) of this algorithm is O(4L2 ∗ Lk+1 on ), i.e. and Λ = 0 involved by the greedy approach makes the number
(m)
O(3L2 ∗ Lk+1
on ) to calculate M k+1 , O(L 2
∗ Lk+1
on ) to calculate of iterations high. We can enhance the convergence rate by
(m)
Pk+1 for all the values of m. Finally, the complexity is judiciously initializing vector B. An initial bit-loading vector
reduced by a factor of about L/4 thanks to the proposed that can be utilized is the bit-loading solution obtained by the
implementation of matrix inversion. water-filling followed by bit discretization. In other words, the
In the following, the greedy algorithm which uses this power allocation obtained by the Constant Power Water-Filling
computation is called the proposed greedy algorithm and (CPWF) [17] is used to calculate the continuous bit-loading.
referred to as GR to distinguish from the standard greedy From the CPWF solution we perform the discrete bit-loading
algorithm (Standard GR), which exploits the matrix inversion and we determine the corresponding power allocation applying
to calculate the cost function. (15). It is exploited to judiciously initialize bit-loading and
power allocation vectors.
C. Reduced complexity approaches
Theorem 1. The power allocation corresponding to the bit-
In this section, we try to reduce the total complexity of loading initialized by CPWF algorithm followed by bit dis-
the greedy algorithms by relying on the reduced complexity cretization always satisfies the power constraints.
approaches proposed in Section III-B. Proof: Let CCP W F and PCP W F be the continuous bit-
1) Cost function approximation: to further reduce total loading and power allocation vectors obtained by the CPWF
complexity, an approximation of the incremental power needed algorithm; B0 and P0 be the discrete bit-loading vector and its
to transmit an additional bit on m-th subcarrier is proposed. corresponding power allocation vector after the discretization
By using Eq. (15), we obtain: and from the monotonicity of the right-hand side of (15) (see
{
Pk − Λk WPk = Λk N VI.A), we obtain:
(m) (m) (m) (m) (26)
Pk+1 − Λk+1 WPk+1 = Λk+1 N 0 ≤ B0 ≤ CCP W F ⇒ 0 ≤ P0 ≤ PCP W F (31)
VO et al.: ACHIEVABLE THROUGHPUT OPTIMIZATION IN OFDM SYSTEMS IN THE PRESENCE OF INTERFERENCE AND ITS APPLICATION . . . 1709
TABLE I
C OMPLEXITY PER ITERATION , NUMBER OF ITERATIONS AND EQUIVALENT 4
TOTAL NUMBER OF MATRIX INVERSIONS .
h(t) × 103
Complexity Number of Number
Algorithm per iteration iterations of matrix
inversions 0
62 14
60
58
12
Upper bound
56 MPA (continuous)
CPWF (continuous)
10 CPWF (discretized)
54 13.54
0 0.1 0.2 0.3 0.4 13.52 Greedy
P0 (Normalized) 13.5 Greedy + Init
13.48
RCA (K=1)
13.46
Fig. 2. An illustration of non-convexity of the problem (33). MPA (discretized)
8
0 20 40 60 80 100
The non-convexity of the problem (33) has already been Max Total Power (Normalized)
mentioned in [17]. To support this result, it is quite easy to find
directions in the constrained power area where non-concavity Fig. 3. Achievable throughput of different algorithms.
arises. For instance, letting:
{ 100
P0 m ∈ [401, 420]
P (m) = (34)
13.5 2,000
Achieved Throughput (Mbps)
200
Total used Power (Normailzed)
Fig. 6. Total consumed power comparison. Fig. 8. Proposed algorithms complexity comparison.
TABLE II
RUN - TIME PER ITERATION , NUMBER OF ITERATIONS AND TOTAL
135
Achieved Throughput (Mbps)
90
400
200
and Ptotal = 500 (normalized). In this figure, the RCA 200 400 600 800
algorithms improve the achievable throughput by about 10%
Max Total Power (Normalized)
when compared to the CPWF + discretization. Moreover, we
can observe that the GI value that maximizes the throughput Fig. 11. RCA vs CPWF in Class 9 channel.
coincides with the one that maximizes the capacity. This
observation can be used to optimize jointly the GI value and
bit/power allocation. In other words, firstly, we could use the greedy algorithms as well as a reduced complexity algorithm.
CPWF algorithm to calculate the capacity for every value of Simulation results have clearly shown that the reduced com-
GI specified in the IEEE P1901 standard. We would choose plexity algorithm is efficient, i.e. the achievable throughput
the GI value corresponding to the maximum capacity. Then, loss is negligible and the complexity is significantly reduced
with this GI value, we use the RCA algorithm to determine the as compared to the greedy solutions. It is also shown that
bit-loading and power allocation that maximize the achievable the proposed algorithm is tight with its upper bound of
throughput. performance and outperforms the MPA and CPWF algorithm.
The cumulative distribution function (CDF) of allocated We have also proposed a joint optimization of the GI value and
number of bits and of allocated power for 100 channel bit/power allocation based on the proposed algorithm. Finally,
realizations are plotted in Figures 13 and 14. We can clearly as the reduced complexity algorithm is based on the greedy
see that transmitting on Class 5 channel requires less power principle, it can be also exploited for the margin-adaptive
than on Class 2 channel (mean value: 0.13 vs 0.2) while the problem with some modifications in the initialization of bit
throughput obtained in Class 5 channel is higher than that of allocation vector. Therefore, it is a good candidate to solve
Class 2 channel (mean value: 8.8 vs 6.2 bit). resource management problems for single-user Windowed-
Finally, to adapt the proposed algorithm to the periodically- OFDM systems in the presence of significant interference.
time varying PLC channels [44], [45], we can re-use the
channel adaptation principle described in [44]. In [44], it is
assumed that the channel periodically switches among several
A PPENDIX
values. We can use this block-based invariance to apply
our proposed algorithm for each channel state. If the frame I NTERFERENCE C ALCULATION IN PLC S YSTEMS
error rate is high, the channel estimation and the bit/power
allocation must be re-operated. In this section, we derive ISI and ICI formula in PLC
systems. In other words, we carry out the value of W (a, b),
VI. C ONCLUSION ∀a, b ∈ [1, L]. In this appendix, subcarrier m denotes the m-th
In this paper, new robust algorithms for the problem of available subcarrier. Under a given spectral mask constraint,
achievable throughput maximization in PLC systems in pres- only a part of available subcarriers is used. According to [11],
ence of interference have been devised. We have developed the in absence of noise, the demodulated sample on the m0 -th
VO et al.: ACHIEVABLE THROUGHPUT OPTIMIZATION IN OFDM SYSTEMS IN THE PRESENCE OF INTERFERENCE AND ITS APPLICATION . . . 1713
200 1
Achieved Throughput (Mbps) CPWF (continuous)
RCA (K=5)
180 RCA (K=20)
0.8
CPWF (discretized)
160 0.6
CDF
140
0.4
120
0.2
Class 2
100 Class 5
0
10 20 30 40 50 0 0.2 0.4 0.6 0.8 1
GI length (µs) Allocated power
Fig. 12. Throughput evolution with different GI values. Fig. 14. CDF of the allocated power.
1
Class 2 RI GI T0
Class 5
0.8
0.6
CDF
T0
0.4 g(t-nT)
GI g(t-nT- )
f(t-n0T)
0.2 n0T (n0+1)T
Ig (m − m0 , n, n0 , τl )
subcarrier and n0 -th OFDM symbol is ∫ GI+T0
1
−1 −1
= δn,n0 ej2π(m−m0 )F0 (u+n0 T ) du
∑
M ∑
+∞ ∑
P T0 GI
y(m0 , n0 ) = cm,n hl e−j2πmF0 τl = δn,n0 δm,m0 (36)
m=0 n=−∞ l=0
∫ +∞ where we used that T0 = F0−1 and δu,v = 1 if u = v and null
1
× g(t − nT − τl ) f (t − n0 T )e j2π(m−m0 )F0 t
dt (35) anywhere else.
T0 −∞
| {z }
Ig (m−m0 ,n,n0 ,τl )
• GI − RI < τl < T: Figure 16 illustrates the relative po-
where sition between both filter responses. By observing Figure
• cm,n : Complex QAM, 8-PSK or BPSK symbol located 16, we deduce that the term Ig (m − m0 , n, n0 , τl ) is not
at the m-th subcarrier of the n-th OFDM symbol; null if and only if n = n0 or n = n0 − 1.
• F0 : frequency separation between adjacent carriers; If n = n0 :
• T0 = F10 : FFT window period;
• RI: Roll-off interval of the filter at transmitter (utilized in Ig (m − m0 , n, n0 , τl )
∫ T
Windowed-OFDM), in the conventional OFDM RI = 0; 1
• GI: Guard interval (GI > RI); = g(u − τl ) f (u)ej2π(m−m0 )F0 (u+n0 T ) du
T0 GI
• T = T0 + GI: OFDM symbol period; ∫ T
1
• g(t), f (t): filter responses at transmitter (length of T + = e|j2π(m−m } T0 GI g(u − τl )e
0 )F0 n0 T j2π(m−m0 )F0 u
{z du
RI) and∑ receiver (length of T0 ); C(m−m0 ,n0 )
P −1
• h(t) = l=0 hl δ(t − τl ): Multi-path channel impulse
∫ T
1
response; = C(m − m0 , n0 )[ ej2π(m−m0 )F0 u du
T0 GI
Firstly, the calculation of Ig (m − m0 , n, n0 , τl ) is given ∫ τl +RI
according to the value of τl . There are two cases: 1
+ (g(u − τl ) − 1) ej2π(m−m0 )F0 u du]
• τl < GI − RI: The filters at the transmitter g(t−nT ) and T0 GI
| {z }
at the receiver f (t−n0 T ) as well as their relative position −G(m−m0 ,τl )
are illustrated in Figure 15. In this case, we obtain = C(m − m0 , n0 )δm,m0 − Ag (m − m0 , n0 , τl ) (37)
1714 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 62, NO. 5, MAY 2014
∑
P −1
where Hg (m0 , m) = | hl e−j2πmF0 τl G(m − m0 , τl )|.
l=IN
Because P (m) = 0, ∀m ∈ / Ause , then the interference
Fig. 16. Relative position of g(t − nT − τl ) and f (t − n0 T ) in the case power on subcarrier m0 ∈ Ause
can be re-written as
of τl > GI − RI. ∑
PI (m0 ) = W (m0 , m)P (m) (41)
m∈Ause
where Ag (m − m0 , n0 , τl ) = C(m − m0 , n0 )G(m − m0 , τl ). Finally, the value of W (a, b), ∀a, b ∈ [1, L] is defined as
W (a, b) = W (m0 , m) where m0 = Ause (a), m = Ause (b)
If n = n0 − 1:With the same calculation and since and W (m0 , m) is calculated as in Eq. (40).
g(u + T ) = 1 - g(u), ∀u ∈ [0, RI] we obtain
R EFERENCES
Ig (m − m0 , n, n0 , τl )
∫ τl +RI [1] H. Ferreira, L. Lampe, J. Newbury, and T. Swart, Power Line Commu-
1 nication. Wiley, 2010.
= C(m − m0 , n0 ) g(u + T − τl )ej2π(m−m0 )F0 u du [2] H. P. Alliance, HomePlug AV Baseline Specification, 2007.
T0 GI
∫ τl +RI [3] A. Cortés, J. Cañete, L. Dı́ez, and M. Torres, “On PLC channel models:
1 ( ) an OFDM-based comparison,” in Proc. 2013 IEEE ISPLC, pp. 333–338.
= C(m − m0 , n0 ) 1 − g(u − τl ) ej2π(m−m0 )F0 u du
T0 GI [4] H. Latchman, K. Afkhamie, S. Katar, B. Mashburn, R. Newman, and
= Ag (m − m0 , n0 , τl ) (38) L. Yonge, “High-speed multimedia home networking over powerline,”
NCTA Technical Papers, pp. 9–22, 2005.
[5] M. Russell and L. Stuber, “Bounds on the interchannel interference of
OFDM in time-varying impairement,” IEEE Trans. Commun., vol. 49,
∑P −1 no. 3, pp. 401–404, 2001.
The channel impulse response h(t) = l=0 hl δ(t−τl ) where
[6] Y. Lang, C. Shixing, and W. Hang, “Effects of cyclic prefix on OFDM
τl < τl+1 and τIN −1 ≤ GI − RI < τIN can be rewritten as systems over time-varying channels,” in Proc. 2005 IEEE PIMRC, vol. 2,
pp. 750–753.
∑
IN −1 ∑
P −1 [7] L. Wan, C. Huang, and C. Shen, “An ISI suppression technique using
h(t) = hl δ(t − τl ) + hl δ(t − τl ) (39) time-domain windowing for OFDM systems,” in Proc. 2006 IEEE VTC,
vol. 5, pp. 2518–2522.
l=0 l=IN [8] J. Martinez, E. Naya, T. Entrambasaguas, and M. Abril, “Low complex-
ity synchronization and frequency equalization for OFDM systems,” in
Replacing (39) into (35) and taking into account (36), (37) Proc. 2007 ICECS, pp. 987–990.
and (38), the analytic formula of y(m0 , n0 ) becomes [9] S. Chen and C. Zhu, “ISI and ICI analysis and mitigation for OFDM
systems with insufficient cyclic prefix in time-varying channels,” IEEE
−1 −1 Trans. Consum. Electron., vol. 50, no. 1, pp. 78–83, 2004.
∑
M ∑
+∞ ∑
IN
[10] I. C. Society, IEEE Standard for Broadband over Power Line Networks:
cm,n hl e−j2πmF0 τl δn,n0 δm,m0 Medium Access Control and Physical Layer Specication, 2010.
m=0 n=−∞ l=0 [11] P. Achaichia, M. L. Bot, and P. Siohan, “OFDM/OQAM: a solution to
−1 ∑ −1 efficiently increase the capacity of future PLC networks,” IEEE Trans.
∑
M +∞ ∑
P
Power Delivery, vol. 26, no. 4, pp. 2443–2455, 2011.
+ cm,n hl e−j2πmF0 τl {δn,n0 [δm,m0 [12] L. Dinh, Q. Yin, K. Deng, and Y. Meng, “A practical computationally
m=0 n=−∞ l=IN efficient power and bit allocation for wireless OFDM systems,” in Proc.
−Ag (m − m0 , n, τl )] + δn,n0 −1 Ag (m − m0 , n0 , τl )} 2006 IEEE ICC, vol. 12, pp. 5628–5633.
( −1
) [13] E. Guerrini, G. D. Amico, P. Bisaglia, and L. Guerrini, “Bit-loading
∑
P algorithms and SNR estimate for HomePlug AV,” in Proc. 2007 IEEE
−j2πmF0 τl
= cm0 ,n0 H(m0 F0 ) − hl e Ag (0, n0 , τl ) ISPLC, pp. 77–82.
[14] C. Y. Wong, R. Cheng, K. Letaief, and R. Murch, “Multicarrier OFDM
l=IN
| {z } with adaptive subcarrier, bit, and power allocation,” IEEE J. Sel. Areas
α(m0 ) Commun., vol. 17, no. 10, pp. 1747–1758, 1999.
[15] L. Goldfeld, V. Lyandres, and D. Wulicj, “Minimum BER power loading
∑
M −1 ∑
P −1
for OFDM in fading channel,” IEEE Trans. Commun., vol. 50, no. 11,
− cm,n0 hl e−j2πmF0 τl Ag (m − m0 , n0 , τl ) pp. 1729–1733, 2002.
m=0,m̸=m0 l=IN [16] W. Yu, W. Rheem, S. Boyd, and J. M. Cioffi, “Iterative water-filling for
| {z } gaussian vector multiple access channel,” IEEE Trans. Inf. Theory, vol.
ICI(m0 ,n0 ) 50, no. 1, pp. 145–152, 2004.
−1 −1
[17] S. D’Alessandro, A. M. Tonello, and L. Lampe, “On power allocation in
∑
M ∑
P
adaptive cyclic prefix OFDM,” in Proc. 2010 IEEE ISPLC, pp. 183–188.
+ cm,n0 −1 hl e−j2πmF0 τl Ag (m − m0 , n0 , τl ) [18] M. Tonello, S. Alessandro, and L. Lampe, “Cyclic prefix design and
m=0 l=IN allocation in bit-loaded OFDM over PLC Channels,” IEEE Trans.
| {z } Commun., vol. 58, no. 11, pp. 3265–3276, 2010.
ISI(m0 ,n0 ) [19] M. Wolkerstorfer, P. Tsiaflakis, M. Moonen, and D. Statovci, “Joint
power-loading and cyclic prefix length optimization for OFDM-based
In practical PLC systems, cm is i.i.d ∀m ∈ [0, M/2 − 1] communication,” 2013 IEEE ICASSP.
[20] W. Yu, G. Ginis, and J. M. Cioffi, “Distributed multiuser power control
and cm = c∗M −m , c0 = cM/2 = 0. Furthermore, we assume for digital subscriber lines,” IEEE J. Sel. Areas Commun., vol. 20, no.
that channel is block-based invariant. Then the values of 5, p. 1105-1115, 2002.
VO et al.: ACHIEVABLE THROUGHPUT OPTIMIZATION IN OFDM SYSTEMS IN THE PRESENCE OF INTERFERENCE AND ITS APPLICATION . . . 1715
[21] S. T. Chung, S. J. Kim, J. Lee, and J. M. Cioffi, “A game-theoretic ap- [43] H. Steendam and M. Moeneclaey, “Analysis and optimization of the
proach to power allocation in frequency-selective Gaussian interference performance of OFDM on frequency-selective time-selective fading
channels,” 2003 IEEE Int. Sym. on Inf. Theory. channels,” IEEE Trans. Commun., vol. 47, no. 12, pp. 1811–1819, 1999.
[22] D. Alessandro, M. Tonello, and L. Lampe, “Adaptive pulse-shaped [44] K.-H. Kim, H.-B. Lee, Y.-H. Kim, and S.-C. Kim, “Channel adaptation
OFDM with application to in-home power line communication,” for time-varying powerline channel and noise synchronized with AC
Telecommun. Syst., vol. 51, no. 1, pp. 3–13, 2012. cycle,” in Proc. 2009 IEEE ISPLC, pp. 250–254.
[23] Y. Huang and B. Rao, “On using a priori channel statistics for cyclic [45] M. Picorone, R. Amado, and V. Ribeiro, “Linear and periodically time-
prefix optimization in OFDM,” in Proc. 2011 IEEE Wireless Commu- varying PLC channels estimation in the presence of impulsive noise,”
nications and Networking Conference, pp. 1460–1465. in Proc. 2010 IEEE ISPLC, pp. 255–260.
[24] ——, “Awareness of channel statistics for slow cyclic prefix adaptation
in an OFDMA system,” IEEE Wireless Commun. Lett., vol. 1, no. 4, Thanh Nhan Vo received his telecommunication
pp. 332–335, 2012. engineering diploma at Telecom Bretagne, France in
[25] M. K. Chan and W. Yu, “Multiuser spectrum optimization for discrete 2012. He is currently a PhD student at the Signal and
multitone systems with asynchronous crosstalk,” IEEE Trans. Signal Communication Department of Telecom Bretagne,
Process., vol. 55, vol. 11, pp. 5425–5435, 2007. France. His research activity focuses on power-line
communication, digital signal processing, resource
[26] J. Seoane, S. Wilson, and S. Gelfand, “Analysis of intertone and
allocation algorithms, MIMO techniques.
interblock interference in OFDM when the length of the cyclic prefix
is shorter than the length of the impulse response of the channel,” in
Proc. 1995 IEEE GLOBECOM, pp. 2069–2074.
[27] J. M. Cioffi, A Multicarrier Prime.. Available: http://www.stanford.edu/
group/cioffi/publications.html
[28] T. H. Cormen, C. E. Leiserson, R. L. Rivest, and C. Stein, Introduction Karine Amis [M’00] received her Dipl.-Ing. degree
to Algorithms, 3rd ed. MIT Press, 2009. from the Ecole Nationale Superieure des Telecom-
munications de Bretagne (ENST Bretagne), France
[29] J. M. Cioffi, Advanced Digital Communication. Available: http://www.
in 1998 and her Ph.D. degree from the University
stanford.edu/class/ee379c/readerfiles/chap4.pdf
of Rennes 1, France in 2001. Since 2001, she is
[30] W. Al-Hanafy and S. Weiss, “A new low-cost discrete bit loading using
an associate professor at Telecom Bretagne, France.
greedy power allocation,” 2009 Mosharaka International Conference on
Her current research interests focus on communica-
Communications, Computers and Applications.
tion and coding theory with application to MIMO
[31] J.-H. Wen, C.-H. Chiang, T.-J. Hsu, and H.-L. Hung, “Resource manage- systems, interference channels, cooperative commu-
ment techniques for OFDM systems with the presence of ICI,” Wireless nications and more recently power line communica-
Personal Commun., vol. 65, no. 3, pp. 515–535, 2011. tions.
[32] E. Guerrini, L. Guerrieri, D. Veronesi, P. Bisaglia, and R. Cappelletti,
“LLR-based bit-loading algorithm and its applications to HomePlug Thierry Chonavel received the Ph.D. thesis in
AV over OPERA power-line channels with impulsive noise,” IEEE J. signal processing from ENST, Paris, France, in
Commun., vol. 4, no. 7, pp. 454–462, 2009. 1992. His thesis was about trigonometric moment
[33] R. Cendrillon, W. Yu, M. Moonen, J. Verlinden, and T. Bostoen, problems and array processing. He joined Institut
“Optimal multiuser spectrum management for digital subscriber lines,” Telecom in 1993, where he is a Professor at Telecom
IEEE Trans. Commun., vol. 54, no. 5, pp. 922–933, 2007. Bretagne, Brest, France. His researches are about
[34] W. Yu and R. Lui, “Dual methods for nonconvex spectrum optimization signal processing and communications with appli-
os multi-carriers systems,” IEEE Trans. Commun., vol. 54, no. 6, pp. cations ranging from speech processing to radar,
1310–1322, 2007. array processing, underwater acoustics, waveforms
[35] R. Cendrillon and M. Moonen, “Iterative spectrum balancing for digital design or powerline communications (PLC). In the
subscriber lines,” in Proc. 2005 IEEE ICC, vol. 3, pp. 1937–1941. field of powerline communications, he worked with
[36] Y. Xu, T. Le-Ngoc, and S. Panigrahi, “Global concave minimization for Orange Labs on studies related to energy consumption of PLC plugs, MIMO
optimal spectrum balancing in multiuser DSL networks,” IEEE Trans. modelling of inhome PLC channels, and power allocation in MIMO-OFDM
Signal Process., vol. 56, no. 7, pp. 2875–2885, 2008. PLC systems.
[37] L. Qian, Y. Zhang, and J. Huang, “Mapel: achieving global optimality
for non-convex wireless power control problem,” IEEE Trans. Wireless Dr. Pierre Siohan (SM’1999) received the Ph.D.
Commun., vol. 8, no. 3, pp. 1526–1563, 2009. degree from the Ecole Nationale Superieure des
[38] N. Papandreou and T. Antonakopoulos, “A new computationally efficient Telecommunications (ENST), Paris, France, in 1989.
discrete bit-loading algorithm for DMT applications,” IEEE Trans. In 1977 he joined the Centre Commun d’Etudes
Commun., vol. 53, no. 5, pp. 785–789, 2006. de Telediffusion et Telecommunications (CCETT),
[39] C. M. Akujuobi, J. Shen, and M. N. O. Sadiku, “A new parallel greedy Rennes, where his activities were first concerned
bit-loading algorithm with fairness for multiple users in a DMT system,” with the communication theory and its application
IEEE Trans. Commun., vol. 54, pp. 1374–1380, 2006. to the design of broadcasting systems. Afterwards,
[40] M. Tonello, F. Versolatto, B. Bejar, and S. Zazo, “A fitting algorithm he was in charge of the CCETT Mathematical and
for random modeling the PLC channel,” IEEE Trans. Power Delivery, Signal Processing Group. From September 2001
vol. 27, no. 3, pp. 1477–1484, 2012. to September 2003, he took a two-year sabbatical
[41] Seventh Framework Programme: Theme 3 ICT-213311 OMEGA, Deliv- leave, being research director with the Institut National de Recherche en
erable D3.2, “PLC Channel Characterization and Modelling,” Informa- Informatique et Automatique (INRIA), Rennes. Now, he is Expert in Access
tion and Communication Technologies (ICT), Tech. Rep., 2011, v1.2. Networks at Orange. His current research interests are in signal processing for
[42] R. Hashmat, P. Pagani, and T. Chonavel, “Analysis and modeling of communication, especially for radio and power line communication systems.
background noise for inhome MIMO-PLC channels,” in Proc. 2012 Dr. Siohan has authored 31 papers in refereed international journals, 30 patents
IEEE ISPLC, pp. 316–321. and 90 papers in international conferences.