Professional Documents
Culture Documents
net/publication/3433265
SLM and PTS peak-power reduction of OFDM signals without side information
CITATIONS READS
196 203
2 authors:
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Chintha Tellambura on 20 March 2014.
Abstract—Selected mapping (SLM) and partial transmit se- may be lost and an irrevocable decoding error can occur. Extra
quence (PTS) are well-known techniques for peak-power reduc- protection bits may need to be sent and the total redundancy can
tion in orthogonal frequency-division multiplexing (OFDM). We thus exceed log2 (U ) bits.
derive a simplified maximum likelihood (ML) decoder for SLM
and PTS that operates without side information. This decoder Consequently, recent research has modified the original
exploits the fact that the modulation symbols belong to a given SLM concept to make it operate without side information.
constellation and that the multiple signals generated by the PTS or Breiling et al. propose a scrambling scheme that does not use
SLM processes are widely different in a Hamming distance sense. explicit side information [9]. A multimode coding technique
Pairwise error probability (PEP) analysis suggests how SLM and based on guided scrambling is proposed in [10]. Modified
PTS vectors should be chosen. The decoder performs well over
additive white Gaussian noise (AWGN) channels, fading channels, repeat accumulate codes are used to generate statistically inde-
and amplifier nonlinearities. pendent symbols in [11]. Although the techniques in [9]–[11]
do not use explicit side information, a small amount of redun-
Index Terms—Orthogonal frequency-division multiplexing
(OFDM), partial transmit sequences, peak-to-average power dancy is introduced at the transmitter. Standard arrays of linear
ratio, selected mapping. block codes are proposed in [12], and this technique sacrifices
the error-correcting ability of the channel code for recovering
side information at the receiver.
I. I NTRODUCTION
In this letter, we develop a new decoder for SLM and PTS
The subcarrier vector in (1), c ∈ C, c = (c0 , c1 , . . . , cN −1 ) is a This minimization can be performed as follows. The minimum
vector of N constellation symbols from a constellation Q. The distance H ⊗ ĉ to r ⊗ P∗0 is determined, where P∗0 is the conju-
size of Q is q. For COFDM, C is an optional channel code such gate of P0 . This can be done by the Viterbi algorithm for coded
as a convolutional code or a trellis code. A cyclic prefix (CP) systems or by searching all q N data sequences for uncoded
is added to the signal s(t) in order to avoid the intersymbol q-ary modulation. This process is repeated for P1 , P2 , . . . ,
interference (ISI) that occurs in multipath channels. Since the PU −1 . The global minimum distance solution yields the best
CP does not impact the PAR issue, we ignore it. The PAR of estimates for c and ũ. The overall complexity is U times that of
OFDM is given by COFDM without SLM.
Consider the quaternary phase-shift keying (QPSK) constel-
max |s(t)|2 lation given by
ξ= (2)
E |s(t)|2 jπm
QQPSK = e 2 , m = 0, 1, . . . , 3 . (5)
where E{.} denotes expectation. PAR does not greatly depend
For the uncoded case, there are U 4N , |.|2 operations to solve
on the constellation Q. The theoretical maximum of the PAR
(4). This is of very high complexity and can be performed only
for N number of subcarriers is 10 log(N ) dB.
for small N . Thus, a suboptimal decoding metric with reduced
complexity is derived next.
III. S ELECTED M APPING W ITHOUT S IDE I NFORMATION Let rn be detected into the nearest constellation point ĉn ,
û
by comparing rn with Ĥn ĉn ejφn , where Ĥn is the estimated
The SLM encoder uses U vectors (SLM vectors) given by channel response. That is, a hard decision is made for each sub-
u u u
carrier. This whole process is repeated for 0 ≤ û ≤ U − 1. The
Pu = ejφ0 , ejφ1 , . . . , ejφN −1 minimum Euclidean distance solution yields the data sequence.
The simplified new decision metric can thus be written as
where φun ∈ (0, 2π] and u ∈ {0, 1, . . . , U − 1}. Let a ⊗ b rep-
N −1 2
resent the vector product of a and b. For a given input frame
min rn e−jφn − Ĥn ĉn .
û
DSLM = min
c, the lowest PAR sequence c ⊗ Pũ , ũ ∈ {0, 1, . . . , U − 1}, is Pû ,û∈{0,1,...,U −1} ĉn ∈Q
n=0
selected for transmission. In the subsequent development, the (6)
value of the optimal transmit sequence number ũ is “NOT”
transmitted to the receiver. Fig. 1 depicts the proposed decoder structure. For COFDM with
We derive the new decoder using the following properties. a given trellis structure, the metric (6) can be computed using
1) cn ’s are restricted to a given signal constellation Q. 2) The set the Viterbi algorithm. The branch metric of the trellis may be
of Pu ’s is fixed and known a priori. 3) c ⊗ Pu and c ⊗ Pv are given by
sufficiently different for u = v. The necessary condition for this 2
u
method to work is cn ejφn ∈ Q for all n and u. The set of Pu
B = rn e−jφn − Ĥn ĉn .
û
(7)
can be chosen readily to ensure this. We expect the performance
of the decoder to be very good, as the Hamming distance Here, ĉn is a branch label of the trellis. Initial results of this
between any Pu and Pv is very large. For simplicity, let us receiver have been presented in [13].
assume a distortionless and noiseless channel. Now, the receiver SLM vectors for the proposed system are constructed by
gets r = c ⊗ Pũ and computes r ⊗ P∗u for u = 0, . . . , U − 1. selecting each φun randomly between 0 and 2π, subject to the
Note that r ⊗ P∗u will not be a vector of symbols from the u
constraint that ejφn is not a member of Q.
constellation Q unless u = ũ. This observation allows us to
dramatically reduce the complexity of the ML decoder.
Consider the received signal rn after the FFT demodulation IV. P ARTIAL T RANSMIT S EQUENCES
at the receiver The input data vector c is partitioned into disjoint subblocks,
jφũ
as {cm |m = 0, 1, . . . , M − 1}, and these are combined to min-
rn = Hn cn e n + nn (3) imize the PAR. Phase angles of subcarriers in each subblock are
changed, and several iterations are performed before obtaining
where Hn is the frequency response of the fading channel the optimally combined sequence. The inverse discrete Fourier
at the nth subcarrier and nn is a complex additive white transform (IDFT) of the cm are called PTSs. Let yk,m for
Gaussian noise (AWGN) sample. The signal-to-noise ratio k = 0, 1, . . . , LN − 1, m = 0, 1, . . . , M − 1, be the LN point
(SNR) is defined as γs = E{|Hn cn |2 }/E{|ηn |2 } where E{} is IDFT of the cm appropriately zero-padded
the statistical expectation operator. Let r = [r0 , r1 , . . . , rN −1 ]
and H = [H0 , H1 , . . . , HN −1 ]. Without side information (not j2πnk
knowing ũ) the optimal ML decoder uses the decision metric yk,m = cn e LN (8)
n∈Im
N −1 2
−jφûn
D= min rn e − Hn ĉn . (4) where Im ∈ {0, 1, . . . , N − 1} is the subset of indices belong-
[ĉ0 ,ĉ1 ,...,ĉN −1 ]∈C
Pû ,û∈{0,1,...,U −1} n=0 ing to the mth subblock and L is the oversampling factor.
2008 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 4, NO. 5, SEPTEMBER 2005
Using the linearity property of the IDFT, the time domain TABLE I
SYSTEM AND CHANNEL PARAMETERS USED FOR SIMULATIONS
samples can be represented as
M
sk = bm yk,m , k = 0, 1, . . . , LN − 1 (9)
m=1
2
M
ξopt. = min max bm yk,m . (10)
b1 ,...,bM 0≤k≤LN
m=1
Fig. 2. System block diagram with channel interleaver and proposed decoder.
TABLE II
POWER DELAY PROFILE OF THE FREQUENCY-SELECTIVE
FADING CHANNEL
less than N . Since we choose φun uniform in (0, 2π), the prob-
ability that φũn = φûn for ũ = û is zero. Note that if φũn was
chosen from a finite constellation such as QPSK, the achievable
diversity order would be less than N . If SLM is “NOT” used at
the transmitter, the numbers of n for which cn = c̃n relate to
the dmin (minimum Hamming distance) of the channel code (if
used). If no channel code is used, dmin = 1. The asymptotic
error rate should then decrease as γs−dmin . We conclude that the
use of long random phase sequence in SLM provides a diversity
order of N . Since N is large, this ensures that our proposed
decoders can achieve nearly ideal performance levels. Equation
(14) also suggests the following decision rules for SLM code Fig. 4. BER of the several TC-OFDM systems with (a) uncoded OFDM
book generation. with pilot-assisted CE, (b) TC-OFDM with pilot-assisted CE, (c) TC-OFDM
with perfect CSI, and (d) TC-OFDM in an AWGN (no fading). Normalized
1) For maximum diversity order, make sure that φkn = φln Doppler frequency is 0.011.
for all l = k.
2) Choosing φkn from a uniform distribution ensures maxi- for trellis-coded orthogonal frequency-division multiplexing
mum diversity benefits. (TC-OFDM) in frequency-selective fading channels. The
frequency-selective channel model is given by h(t) =
L
k=1 hk δ(t − kT ) with E[|hk | ] decaying exponentially with
2
VI. S IMULATION R ESULTS
k (Table II). A normalized Doppler frequency of 0.011 is
Proposed decoders for both SLM and PTS are evaluated over assumed in the simulations. We use a nonsystematic, eight-state
AWGN and fading channels and in the presence of a nonlinear encoder with eight-phase-shift keying (8-PSK) mapping, which
amplifier using Monte Carlo simulations. System and channel has the same spectral efficiency as uncoded QPSK-OFDM.
parameters are shown in Table I. The system block diagram for Each OFDM symbol consists of 64 pilot symbols for CE.
both uncoded and TCM-coded OFDM is shown in Fig. 2. The Fig. 4 shows that the proposed decoder has identical perfor-
interleaver and deinterleaver pair is used for the simulation on mance to a decoder with PSI in all the channels considered
frequency-selective fading channels to randomize the bit errors (a–d). Fig. 4 also shows the effect of imperfect channel state
across subcarriers providing the frequency diversity. information (CSI). The proposed decoder performs well even
Fig. 3 compares the performance of the proposed decoder with imperfect CSI and recovers COFDM completely in both
(6) with an ideal SLM decoder using perfect side information AWGN and fading channels even in the presence of a nonlinear
(PSI-SLM) in AWGN. The two systems show virtually identical amplifier.
performance in AWGN. Comparative performance in AWGN Fig. 5 compares the performance of the proposed decoder
for a soft limiter (nonlinearity) with backoff values of 2, 5, and (12) with an ideal PTS receiver that has perfect side information
7 dB in Fig. 3 shows that both the decoders perform nearly (PSI-PTS) in AWGN and Rayleigh fading. As expected, the
identically. proposed decoder almost perfectly recovers the received data
Given the excellent decoder performance over AWGN in AWGN. It performs nearly as good as the ideal receiver
and nonlinear channels, we next investigate its performance in Rayleigh fading. The slight performance degradation of the
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 4, NO. 5, SEPTEMBER 2005 2011
Note that this analysis is valid only in the high SNR regime.
The interested reader is referred to [18] and [19]. We use this
approach here to show that a side benefit of the SLM expansion
process is to generate an equivalent system with a diversity
order reaching N . We derive the PEP as a complex contour
integral with distinct single-order poles. However, more insight
can be gained by considering the high SNR regime where the
multiple poles coalesce into an N th-order pole. The resulting
PEP expression is presented in Section V.
Let Ĥn = Hn + n where n is a complex Gaussian noise
affecting the CE process. We assume the estimation error n to
be independent of the channel additive noise and the channel.
That is, E[nn ∗m ] = 0 and E[Hn ∗m ] = 0. While this assump-
tion may not be realistic in some cases, it greatly facilitates the
subsequent analysis. Without loss of generality, we assume
E |Hn |2 = 1, n = 0, 1, . . . , N − 1. (15)
Fig. 5. Comparison of BER performance of the proposed decoder for We define the CE SNR as
PTS-OFDM in AWGN and fading with perfect CSI. Normalized Doppler
frequency is 0.011. E |Ĥn |2
γe = for n = 0, 1, . . . , N − 1. (16)
proposed decoder in the fading channel may be overcome by a E {|n |2 }
suitable channel coding scheme, as in the case of SLM.
The performance of the proposed decoder with higher order This means E{|Ĥn |2 } = γe /(γe − 1) and E[|n |2 ] = 1/(γe −
constellations (8-PSK and 16-QAM) was also simulated. In the 1). We also have E{Hn∗ Ĥn } = E{Hn∗ (Hn + n )} = 1. We
high SNR region, the proposed decoder performs fairly close to consider MPSK only for brevity.
the ideal case. From (3), we find E{|nn |2 } = 1/γs and E{|rn |2 } = 1 +
(1/γs ). The PEP is given by
N −1
VII. C ONCLUSION
ũ 2
P [(c, Pũ ) → (ĉ, Pû )] = Pr rn − Ĥn cn ejφn
Selected mapping (SLM) and partial transmit sequence n=0
(PTS) detection without using side information for coded or- −1
N
thogonal frequency-division multiplexing (COFDM) and or- û 2
> rn − Ĥn ĉn ejφn . (17)
thogonal frequency-division multiplexing (OFDM) has been n=0
developed. These decoders recover received COFDM in addi-
tive white Gaussian noise (AWGN), fading, and the presence Let us consider a term
of nonlinear amplifiers. Analytical results show that the ML
û 2 ũ 2
decoder for SLM has an N th-order diversity effect in fading. Dn = rn − Ĥn ĉn ejφn − rn − Ĥn cn ejφn
As N is typically large, excellent performance can be obtained
even though a suboptimal decoder (6) is used. The proposed = |Xn |2 − |Yn |2 . (18)
SLM and PTS systems neither lose throughput due to side This is a special case of the quadratic form, which has been
information nor degrade bit error rate (BER) due to errors in analyzed by Proakis [20]. The complex Gaussian random
side information. However, a reduction in throughput occurs variable Xn and Yn are zero mean and correlated. We need
due to the pilot tones used for channel estimation (CE). Some their variances and the covariance for deriving the moment
increase in the receiver complexity is the price paid for these generating function (mgf) of Dn
benefits.
1
µxx = E |Xn |2
2
A PPENDIX
PEP A NALYSIS 1 û 2
= E |rn |2 + Ĥn ĉn ejφn
2
PEP analysis is a widely used approach for designing coding
schemes such as trellis codes and space–time codes. This − 2Re rn∗ Ĥn ĉn ejφn
û
approach begins with the union bound for the BER. The union
bound is a weighted sum of all the PEPs for a given code. As
1 1 γe
the SNR goes to infinity, the union bound is dominated by a = 1+ +
2 γs (γe − 1)
handful of PEPs, which can be expressed in terms of distance
properties of the code. In general, if the largest PEP decays as
∗ n −φn )
j (φû ũ
− 2Re cn ĉn e . (19)
γs−d , the coded system is said to have a diversity order of d.
2012 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 4, NO. 5, SEPTEMBER 2005
2
1 jφũ We assume that all v1n = v1m for m = n and v2n = v2m for
µyy = E −n cn e n + nn m = n. From (17), we find
2
1 1 1
= + . P [(c, Pũ ) → (ĉ, Pû )] = Pr(D < 0). (25)
2 (γe − 1) γs
Using a complex variable theory, this can be written as
The covariance is given by
1
P [(c, Pũ ) → (ĉ, Pû )] = −Res
ΦD (s)
at poles in LHP
= E rn − Ĥn ĉn ejφn ∗n c∗n ejφn + n∗n
û ũ
µxy s
2
(26)
1 1 1
ĉn c∗n ej (φn −φn ) .
û ũ
= + (20)
2 γs (γe − 1) where LHP stands for left half plane. Under the assumption in
(24), there are only first-order poles at s = −v2n . The exact
In Proakis [20, eq. B-6], we use α1 = α2 = 0, A = 1, B = −1, average PEP is given by
and C = 0 and get
N −1
v1k v2k
µxx − µyy P [(c, Pũ ) → (ĉ, Pû )] =
ωn = (21) v2k (v1k + v2k )
4 (µxx µyy − |µxy |2 ) k=0
N −1
v1n v2n
v1n 1 × . (27)
= ωn2 + ± ωn . (22) (v1n + v2k )(v2n − v2k )
v2n 4 (µxx µyy − |µxy |2 ) n=0
n=k
Note that for high SNR and high CE SNR (γe ), we have µxx ≥ However, this exact average PEP does not immediately give
µyy and µxy 0. In this case any insight. It is suitable for numerical computation only. We
therefore device an approximation valid at high SNRs. From
1 γs (γe − 1) (22), for high SNR, we find that
ωn = .
4µyy 2(γs + γe − 1)
v1n 0 and
The mgf of Dn is given by γs (γe − 1)
v2n = ∆,
(γs + γe − 1)
v1n v2n
ΦDn (s) = E{esDn } = . (23) n = 0, 1, . . . , N − 1. (28)
(v1n − s)(v2n + s)
Using (28) in the denominator of (24) and using (26), we have
To facilitate the analysis, we assume that the availability of
perfect interleaving, making Dn , n = 0, 1, . . . , N − 1 indepen-
P [(c, Pũ ) → (ĉ, Pû )]
dent. Consider a block interleaver of size Nd × Ns . As a result
of interleaving, the effective Doppler is increased by a factor of G
Res at s = −∆ (29)
Nd (the interleaving depth) and this reduces temporal correla- (−s)N +1 (s + ∆)N
tion. Perfect interleaving implies Nd being infinity. However, in
practical cases, fairly small values of Nd are sufficient [21]. where G is defined by (30), shown at the bottom of the page.
N −1
1
G=
n=0
4 (µxx µyy − |µxy |2 )
N −1
1
= 2 (30)
ũ
n=0 1 + 1 jφũn − ĉn ejφûn 2 1 1
− γ1s + 1 ∗ j(φû
n −φn )
γs (γe −1) + cn e γs + (γe −1) γe ĉn cn e
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 4, NO. 5, SEPTEMBER 2005 2013
The N th-order residue is evaluated as [5] J. Tellado, “Peak to average power reduction for multicarrier modulation,”
Ph.D. dissertation, Dept. Elect. Eng., Stanford Univ., Stanford, CA, Sep.
1999.
1 [6] W. Henkel and V. Zrno, “PAR reduction revisited: An extension to
Res − at s = −∆ Tellado’s method,” in 6th Int. OFDM Workshop, Hamburg, Germany,
sN +1 (s + ∆)N 2001, pp. 31-1–31-6.
(−1)N +1 dN −1 1 [7] A. D. S. Jayalath and C. Tellambura, “Use of data permutation to reduce
= the peak-to-average power ratio of an OFDM signal,” Wireless Commun.
(N − 1)! ds N −1 (−s) N +1 Mob. Comput. J., vol. 2, no. 2, pp. 187–203, Mar. 2002.
s=−∆
[8] S. H. Müller and J. B. Huber, “A comparison of peak power re-
1 (N + 1)(N + 2) . . . (2N − 1) duction schemes for OFDM,” in IEEE Global Telecommunications
= (GLOBECOM), Phoenix, AZ, 1997, vol. 1, pp. 1–5.
(N − 1)! (−∆)2N [9] M. Breiling, S. Müller-Weinfurtner, and J. Huber, “SLM peak–power
reduction without explicit side information,” IEEE Commun. Lett., vol. 5,
2N − 1 1 no. 6, pp. 239–241, 2001.
= . (31)
N (∆2 )N [10] Y. Xin and I. J. Fair, “Interpretation of coding for peak-to-average power
ratio reduction in OFDM as constrained sequence coding and the ap-
plication of guided scrambling,” in IEEE Int. Symp. Advances Wireless
Combining (29) and (31), we finally get Communications, Victoria, BC, Canada, 2002, pp. 121–122.
[11] N. Carson and T. A. Gulliver, “PAPR reduction of OFDM using selected
mapping, modified RA codes and clipping,” in IEEE Vehicular Technol-
2N ogy Conf., Vancouver, BC, Canada, 2002, pp. 1070–1073.
2N − 1 γs + γe − 1 [12] K. Yang and S. I. Chang, “Peak-to-average power control in OFDM
P [(c, Pũ ) → (ĉ, Pû )] G .
N γs (γe − 1) using standard arrays of linear block codes,” IEEE Commun. Lett., vol. 7,
no. 4, pp. 174–176, Apr. 2003.
(32) [13] A. D. S. Jayalath and C. Tellambura, “Blind SLM receiver for PAR
reduced OFDM,” in IEEE Vehicular Technology Conf., Vancouver, BC,
Upon combining (30) and (32), we get (13). Canada, 2002, pp. 219–222.
[14] S. H. Müller and J. B. Huber, “A comparison of peak power re-
duction schemes for OFDM,” in IEEE Global Telecommunications
(GLOBECOM), Phoenix, AZ, Nov. 1997, vol. 1, pp. 1–5.
ACKNOWLEDGMENT [15] A. D. S. Jayalath and C. Tellambura, “A blind detection algorithm for
PTS,” in IEEE Int. Symp. Advances Wireless Communications, Victoria,
The authors would like to thank the anonymous reviewers BC, Canada, 2002, pp. 125–126.
[16] ——, “An adaptive PTS approach for the reduction of peak-to-average
for their critical comments, which greatly improved this paper. power ratio of an OFDM signal,” IEE Electron. Lett., vol. 36, no. 14,
pp. 1161–1163, Jul. 2000.
[17] H. Ochiai, M. P. C. Fossorier, and H. Imai, “Performance analysis of
R EFERENCES deliberately clipped OFDM signals,” IEEE Trans. Commun., vol. 50,
no. 1, pp. 89–102, Jan. 2002.
[1] R. Nee and R. Prasad, OFDM for Wireless Multimedia Communications. [18] J. Cavers and P. Ho, “Analysis of the error performance of trellis-
Boston, MA: Artech House, Mar. 2000. coded modulations in Rayleigh-fading channels,” IEEE Trans. Commun.,
[2] E. A. Jones, T. Wilkinson, and S. Barton, “Block coding scheme for vol. 40, no. 1, pp. 74–83, Jan. 1992.
reduction of peak-to-mean envelope power ratio of multicarrier trans- [19] C. Tellambura, “Evaluation of the exact union bound for trellis-coded
mission schemes,” IEE Electron. Lett., vol. 30, no. 25, pp. 2098–2099, modulations over fading channels,” IEEE Trans. Commun., vol. 44,
Dec. 1994. no. 12, pp. 1693–1699, Dec. 1996.
[3] H. Ochiai and H. Imai, “Performance of the deliberate clipping with adap- [20] J. G. Proakis, Digital Communications, McGraw-Hill series in electrical
tive symbol selection for strictly band-limited OFDM systems,” IEEE J. engineering. Communications and signal processing, 4th ed. New York:
Sel. Areas Commun., vol. 18, no. 11, pp. 2270–2277, Nov. 2000. McGraw-Hill, 2001.
[4] K. G. Paterson and V. Tarokh, “On the existence and construction of good [21] C. Tellambura and V. K. Bhargava, “Error performance of MPSK trellis-
codes with low peak-to-average power ratio,” IEEE Trans. Inf. Theory, coded modulation over nonindependent Rician fading channels,” IEEE
vol. 46, no. 6, pp. 1974–1987, Sep. 2000. Trans. Veh. Technol., vol. 47, no. 1, pp. 152–162, Feb. 1998.