Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Coherent Detection of Turbo Coded OFDM Signals

Transmitted through Frequency Selective Rayleigh


Fading Channels

K. Vasudevan
Dept of EE
IIT Kanpur
Email: vasu@iitk.ac.in

Abstract—This work addresses the problem of coherently channel estimation techniques discussed in the literature are
detecting turbo coded orthogonal frequency division multi- done in the frequency domain, using pilot symbols at regular
plexed (OFDM) signals, transmitted through frequency selective intervals in the time/frequency grid [18]–[22]. However in
Rayleigh fading channels. A single transmit and receive antenna this work, maximum likelihood (ML) channel estimation using
is assumed. The channel output is distorted by a carrier frequency the preamble is investigated. This approach does not require
and phase offset, besides additive white Gaussian noise (AWGN).
any knowledge of the channel and noise statistics. Simulation
A new frame structure for OFDM, consisting of a known pream-
ble, cyclic prefix and data is proposed. A filter that is matched to results show that for a sufficiently long preamble, the ML
the preamble is used for start-of-frame (SoF) detection. A two- estimator performs as good as the situation when the channel
step ML detector for frequency-offset estimation is proposed, is known perfectly at the receiver. OFDM is a suggested
which has a much lower complexity compared to the single step modulation technique for digital video broadcasting [23], [24].
ML detector. Turbo decoding and data interleaving is used to It has also been proposed for optical communications [25].
significantly enhance the bit-error-rate (BER) performance of
the coherent receiver. Simulation results show that the BER This paper is organized as follows. Section II describes
performance of the practical coherent receiver is close to the the system model. The receiver algorithms are presented in
ideal coherent receiver, for data length equal to the preamble section III. The bit-error-rate (BER) results from computer
length, and attains a BER of about 4 × 10−5 at an SNR of just simulations are given in section IV. Finally, section V presents
8 dB. It is also shown that the probability of erasure is less than the conclusions and future work.
10−6 for a preamble length of 512 QPSK symbols. The proposed
algorithms are well suited for implementation on a DSP-platform. II. S YSTEM M ODEL

Keywords—OFDM, coherent detection, matched filtering, turbo It is assumed that the data to be transmitted is organized
codes, frequency selective Rayleigh fading. into frames, as depicted in Figure 1. The frame consists of
a known preamble of length 𝐿𝑝 symbols, a cyclic prefix of
length 𝐿𝑐𝑝 , followed by data of length 𝐿𝑑 symbols. Thus, the
I. I NTRODUCTION total length of the frame is 𝐿 = 𝐿𝑝 + 𝐿𝑐𝑝 + 𝐿𝑑 . Let us assume
Future wireless communication standards aim to push the a channel span equal to 𝐿ℎ . The channel span assumed by the
existing data-rates higher. This can only be achieved with the receiver is 𝐿ℎ𝑟 (> 𝐿ℎ ). The length of the cyclic prefix is 𝐿𝑐𝑝 =
help of coherent communications, since they give the lowest 𝐿ℎ𝑟 −1 [4]. Throughout the manuscript, tilde is used to denote
bit-error-rate (BER) performance for a given signal-to-noise complex quantities. However, complex (QPSK) symbols will
ratio (SNR). Conversely, they require the lowest SNR to attain be denoted without a tilde e.g. 𝑆1, 𝑛 . Boldface letters denote
a given BER, resulting in enhanced battery life. There is also vectors or matrices. The channel coefficients ℎ̃𝑘, 𝑛 for the 𝑘 𝑡ℎ
a growing realization that the present-day (3G) mobiles do frame are C N (0, 2𝜎𝑓2 ) and independent over time (𝑛), that
not attain the theoretical performance limits [1]. The recent is:
advances in cooperative wireless communications has yielded 1 [ ]
low-complexity solutions, which are not necessarily power- 𝐸 ℎ̃𝑘, 𝑛 ℎ̃∗𝑘, 𝑛−𝑚 = 𝜎𝑓2 𝛿𝐾 (𝑚) (1)
2
efficient [2]. The “complexity” of coherent reception can be
overcome by use of parallel processing, for which there is a where “*” denotes complex conjugate and 𝛿𝐾 (⋅) is the Kro-
large scope. It is emphasized that this paper presents a proof- necker delta function. This implies a uniform channel power
of-concept, and is hence not constrained by existing standards delay profile. The channel is assumed to be quasi-static, that is
on wireless communications. ℎ̃𝑘, 𝑛 is time-invariant over one frame and varies independently
from frame-to-frame, that is
1 [ ]
The basics of OFDM can be found in [3]–[5]. Timing and
frequency synchronization for OFDM has been investigated 𝐸 ℎ̃𝑘, 𝑛 ℎ̃∗𝑗, 𝑛 = 𝜎𝑓2 𝛿𝐾 (𝑘 − 𝑗) (2)
earlier in [6]–[15]. However the novelty of this work lies 2
in the use of a filter that is matched to the preamble, to where 𝑘 and 𝑗 denote the frame indices. The AWGN noise
acquire timing synchronization [16], [17]. Most flavors of the samples 𝑤 ˜𝑘, 𝑛 for the 𝑘 𝑡ℎ frame at time 𝑛 are C N (0, 2𝜎𝑤 2
).
978-1-4673-6190-3/13/$31.00 ⃝
c 2013 IEEE

Authorized licensed use limited to: QSIO. Downloaded on June 02,2020 at 04:34:34 UTC from IEEE Xplore. Restrictions apply.
𝑠˜1, 𝑛 𝑠˜𝑘, 2, 𝑛 𝑠˜𝑘, 3, 𝑛 III. R ECEIVER
(a)
Preamble Cyclic prefix Data This section presents the key receiver algorithms, namely,
𝐿𝑝 𝐿𝑐𝑝 𝐿𝑑 start of frame (SoF), coarse/fine frequency offset, channel and
noise variance estimation and finally data detection.
𝑠˜𝑘, 𝑛
(b)
𝑆1, 0 A. Start of Frame and Coarse Frequency Offset Estimation

Preamble QPSK
𝑠˜1, 0 Let us assume that for the 𝑘 𝑡ℎ frame, the channel im-
IFFT pulse response is known at the receiver. The channel length
Parallel
symbols
𝑠˜1, 𝐿𝑝 −1 to assumed by the receiver is 𝐿ℎ𝑟 (> 𝐿ℎ ) such that the first 𝐿ℎ
𝑆1, 𝐿𝑝 −1 serial coefficients are identical to the channel coefficients and the
and remaining 𝐿ℎ𝑟 − 𝐿ℎ coefficients are zeros. Define the 𝑚𝑡ℎ
𝑆𝑘, 3, 0 add
cyclic (0 ≤ 𝑚 ≤ 𝐿𝑐𝑝 + 𝐿𝑑 + 𝐿ℎ + 𝐿ℎ𝑟 − 2) received vector as:
𝑠˜𝑘, 3, 0 [ ]
Data QPSK prefix
IFFT r̃𝑘, 𝑚 = 𝑟˜𝑘, 𝑚 . . . 𝑟˜𝑘, 𝑚+𝐿𝑝 −𝐿ℎ𝑟 . (8)
symbols
𝑠˜𝑘, 3, 𝐿𝑑 −1
𝑆𝑘, 3, 𝐿𝑑 −1 The steady-state1 preamble part of the transmitted signal
appearing at the channel output can be represented by a vector:
𝑦˜𝑘, 𝑛 [ ]
To Channel 𝑠˜𝑘, 𝑛 ỹ𝑘, 1 = 𝑦˜𝑘, 𝐿ℎ𝑟 −1 . . . 𝑦˜𝑘, 𝐿𝑝 −1 . (9)
ℎ̃𝑘, 𝑛
receiver The non-coherent maximum likelihood (ML) rule for frame
detection can be stated as [4]: Choose that time as the start
e j(𝜔𝑘 𝑛+𝜃𝑘 )
𝑤
˜𝑘, 𝑛 (AWGN) of frame and that frequency 𝜔 ˆ 𝑘 , which jointly maximize the
conditional pdf:
Fig. 1. (a) The frame structure. (b) System model. 𝑘 denotes the frame index ∫ 2𝜋
and 𝑛 denotes the time index in a given frame. max 𝑝 (r̃𝑘, 𝑚 ∣ỹ𝑘, 1 , 𝜔
ˆ 𝑘 , 𝜃𝑘 ) 𝑝(𝜃𝑘 ) 𝑑𝜃𝑘 . (10)
𝑚, 𝜔
ˆ𝑘 𝜃𝑘 =0

The frequency offset 𝜔𝑘 for the 𝑘 𝑡ℎ frame is uniformly Substituting for the joint conditional pdf and 𝑝(𝜃𝑘 ), and
distributed over [−0.04, 0.04] radian [11]. The phase offset defining
𝜃𝑘 for the 𝑘 𝑡ℎ frame is uniformly distributed over [0, 2𝜋). 𝐿1 = 𝐿𝑝 − 𝐿ℎ𝑟 + 1 (11)
Both 𝜔𝑘 and 𝜃𝑘 are fixed for a frame and vary randomly from
frame-to-frame. yields:
Note that 1 1
max 2 ) 𝐿1
𝐿𝑝 −1 ˆ 𝑘 2𝜋 (2𝜋𝜎𝑤
𝑚, 𝜔
1 ∑ ∫ 2𝜋 ( ∑𝐿 −1 )
𝑠˜1, 𝑛 = 𝑆1, 𝑖 e j 2𝜋𝑛𝑖/𝐿𝑝 − 𝑖=0
1 𝑟˜𝑚+𝑖 − 𝑦˜𝑘, 𝐿 −1+𝑖 e j(ˆ𝜔𝑘 𝑖+𝜃) 2
𝐿𝑝 𝑖=0 exp ℎ𝑟

2𝜎𝑤 2
𝐿𝑑 −1 𝜃=0
1 ∑ × 𝑑𝜃.
𝑠˜𝑘, 3, 𝑛 = 𝑆𝑘, 3, 𝑖 e j 2𝜋𝑛𝑖/𝐿𝑑 . (3) (12)
𝐿𝑑 𝑖=0
where
It is assumed that 𝑆𝑘, 3, 𝑖 ∈ ±1 ± j. Since
[ ] [ ] ˆ 𝑘 (𝐿ℎ𝑟 − 1) + 𝜃𝑘
𝜃=𝜔 (13)
2 2 Δ
𝐸 ∣˜𝑠1, 𝑛 ∣ = 𝐸 ∣˜ 𝑠𝑘, 3, 𝑛 ∣ = 2/𝐿𝑑 = 𝜎𝑠2 (4)
incorporates the phase accumulated by the frequency offset
√ over the first 𝐿ℎ𝑟 − 1 samples, besides the initial phase 𝜃𝑘 .
it follows that 𝑆1, 𝑖 ∈ 𝐿𝑝 /𝐿𝑑 (±1 ± j). In other words, Observe that 𝜃 is also uniformly distributed in [0, 2𝜋).
the average power of the preamble part must be equal to the
average power of the data part. One of the terms in the exponent
𝑡ℎ ∑𝐿1 −1
The received signal for the 𝑘 frame can be written as 𝑖=0 ∣˜𝑟𝑚+𝑖 ∣2
(for 0 ≤ 𝑛 ≤ 𝐿 + 𝐿ℎ − 2): 2𝜎𝑤2
(14)
( )
𝑟˜𝑘, 𝑛 = 𝑠˜𝑘, 𝑛 ★ ℎ̃𝑘, 𝑛 e j(𝜔𝑘 𝑛+𝜃𝑘 ) + 𝑤
˜𝑘, 𝑛 is approximately proportional to the average received signal
power, for large values of 𝐿𝑝 and 𝐿𝑝 ≫ 𝐿ℎ𝑟 , and is hence
= 𝑦˜𝑘, 𝑛 e j(𝜔𝑘 𝑛+𝜃𝑘 ) + 𝑤
˜𝑘, 𝑛 (5) (approximately) independent of 𝑚 and 𝜃. The other exponen-
where “★” denotes convolution and tial term
∑𝐿1 −1
𝑦˜𝑘, 𝑛 = 𝑠˜𝑘, 𝑛 ★ ℎ̃𝑘, 𝑛 . (6) 𝑦𝑘, 𝐿ℎ𝑟 −1+𝑖 ∣2
𝑖=0 ∣˜
2
(15)
2𝜎𝑤
The set of received samples can be denoted by the vector:
1 By steady-state it is meant that all the channel coefficients are involved in
r̃𝑘 = [ 𝑟˜𝑘, 0 ... 𝑟˜𝑘, 𝐿+𝐿ℎ −2 ] . (7) the convolution to generate 𝑦˜𝑘, 𝑛 in (6)

Authorized licensed use limited to: QSIO. Downloaded on June 02,2020 at 04:34:34 UTC from IEEE Xplore. Restrictions apply.
is clearly independent of 𝑚 and 𝜃. Therefore (12) simplifies 1.0e+00
to (ignoring constants):

Probability of frame erasure


1.0e-01
1
max
𝑚, 𝜔
ˆ𝑘 2𝜋 1.0e-02
⎛ {∑ }⎞
∫ ℜ
𝐿1 −1
2˜ ∗
𝑟𝑚+𝑖 𝑦˜𝑘, −j(ˆ
𝜔𝑘 𝑖+𝜃)
𝐿ℎ𝑟 −1+𝑖 e
2𝜋
𝑖=0
exp ⎝ ⎠ 1.0e-03
2
2𝜎𝑤
𝜃=0
1.0e-04
× 𝑑𝜃 (16)
1.0e-05 Lp=64
which simplifies to [4]: Lp=128
( ) Lp=256
𝐴𝑚, 𝜔ˆ 𝑘 1.0e-06
0 1 2 3 4 5 6 7 8
max 𝐼0 2
(17)
𝑚, 𝜔
ˆ𝑘 2𝜎𝑤 SNR per bit (dB)

where 𝐼0 (⋅) is the modified Bessel function of the zeroth-order Fig. 2. Probability of frame erasure as a function of the preamble length 𝐿𝑝 .
and
𝐿 −1
∑1
∗ ˆ𝑘 𝑖
−j 𝜔
𝐴𝑚, 𝜔ˆ 𝑘 = 2˜
𝑟𝑚+𝑖 𝑦˜𝑘, 𝐿ℎ𝑟 −1+𝑖 e . (18) B. Channel Estimation

𝑖=0
This section describes the maximum likelihood (ML)
Noting that 𝐼0 (𝑥) is a monotonic function of 𝑥 and ignoring method of channel estimation. It is assumed that the SoF has
constants, the maximization in (17) simplifies to: been estimated using (22) with outcome 𝑚0 (0 ≤ 𝑚0 ≤
𝐿 −1 𝐿ℎ − 1) and the frequency offset has been perfectly canceled.
∑1
∗ ˆ𝑘 𝑖
−j 𝜔 Define
max 𝑟˜𝑚+𝑖 𝑦˜𝑘, 𝐿ℎ𝑟 −1+𝑖 e . (19)
ˆ𝑘
𝑚, 𝜔
𝑖=0
𝑚1 = 𝑚0 + 𝐿ℎ − 1. (24)
Observe that (19) resembles the operation of demodulation and
matched filtering. The ideal outcome of (19) to estimate the
The steady-state, preamble part of the received signal for
SoF and frequency offset is:
the 𝑘 𝑡ℎ frame can be written as:
𝑚 = 𝐿ℎ𝑟 − 1
r̃𝑘, 𝑚1 = s̃1 h̃𝑘 + w̃𝑘, 𝑚1 (25)
𝜔
ˆ𝑘 = 𝜔𝑘 . (20)
In practice, the receiver has only the estimate of the channel where
(ℎ̂𝑘, 𝑛 ), hence 𝑦˜𝑘, 𝑛 must be replaced by 𝑦ˆ𝑘, 𝑛 , where [ ]𝑇
r̃𝑘, 𝑚1 = 𝑟˜𝑘, 𝑚1 . . . 𝑟˜𝑘, 𝑚1 +𝐿𝑝 −𝐿ℎ𝑟
𝑦ˆ𝑘, 𝑛 = 𝑠˜1, 𝑛 ★ ℎ̂𝑘, 𝑛 (21) [(𝐿𝑝 − 𝐿ℎ𝑟 + 1) × 1] vector
[ ]𝑇
is the preamble convolved with the channel estimate. When w̃𝑘, 𝑚1 = 𝑤˜𝑘, 𝑚1 . . . 𝑤 ˜𝑘, 𝑚1 +𝐿𝑝 −𝐿ℎ𝑟
ℎ̂𝑘, 𝑛 is not available, a heuristic method of frame detection is [(𝐿𝑝 − 𝐿ℎ𝑟 + 1) × 1] vector
proposed as follows: [ ]𝑇
h̃𝑘 = ℎ̃𝑘, 0 . . . ℎ̃𝑘, 𝐿ℎ𝑟 −1

𝐿𝑝 −1 [𝐿 × 1] vector
∑ ⎡ ℎ𝑟 ⎤
max 𝑟˜𝑚+𝑖 𝑠˜∗1, 𝑖 e−j 𝜔ˆ 𝑘 𝑖 (22) 𝑠˜1, 𝐿ℎ𝑟 −1 . . . 𝑠˜1, 0
𝑚, 𝜔
ˆ𝑘
𝑖=0 ⎢ .. .. ⎥
s̃1 = ⎣ . ... . ⎦
where again 𝑠˜1, 𝑖 denotes the preamble as shown in Figure 1. 𝑠˜1, 𝐿𝑝 −1 . . . 𝑠˜1, 𝐿𝑝 −𝐿ℎ𝑟
The ideal outcome of (22) is: [(𝐿𝑝 − 𝐿ℎ𝑟 + 1) × 𝐿ℎ𝑟 ] matrix (26)
0 ≤ 𝑚 ≤ 𝐿ℎ − 1
where again 𝐿ℎ𝑟 (> 𝐿ℎ ) is the channel length assumed by the
ˆ𝑘 =
𝜔 𝜔𝑘 (23) receiver. The statement of the ML channel estimation is as
depending on which channel coefficient has the maximum follows: find ĥ𝑘 (the estimate of h̃𝑘 ) such that:
magnitude. In practical situations, one also needs to look at ( )𝐻 ( )
the ratio of the peak-to-average power of (22) to estimate the r̃𝑘, 𝑚1 − s̃1 ĥ𝑘 r̃𝑘, 𝑚1 − s̃1 ĥ𝑘 (27)
SoF [17]. When 𝑚 lies outside the range in (23), the frame is
declared as erased (lost). The probability of frame erasure as a is minimized. Differentiating with respect to ĥ∗𝑘 and setting
function of the preamble length is shown in Figure 2. Observe the result to zero yields [4], [26]:
that for 𝐿𝑝 = 512, the probability of erasure is less than 10−6 ( )−1 𝐻
and is hence not plotted. ĥ𝑘 = s̃𝐻1 s̃1 s̃1 r̃𝑘, 𝑚1 . (28)
The coarse frequency offset estimate 𝜔 ˆ 𝑘 is obtained by
dividing the interval [−0.04, 0.04] radian into 𝐵1 frequency At this point, it must be mentioned that in the absence of
bins and selecting that bin which maximizes (22). noise, the channel estimate obtained from (28) depends on the

Authorized licensed use limited to: QSIO. Downloaded on June 02,2020 at 04:34:34 UTC from IEEE Xplore. Restrictions apply.
SoF estimate 𝑚0 obtained from (22). When 𝑚0 = 𝐿ℎ − 1, the 1.0e-02
RMS coarse
channel estimate in the absence of noise would be: RMS fine

Freq offset est error (radians)


Max coarse
[ ]𝑇 Max fine
ĥ𝑘 = ℎ̃𝑘, 0 . . . ℎ̃𝑘, 𝐿ℎ −1 0 . . . 0 (29)
When 𝑚0 = 0, the channel estimate (in the absence of noise)
is : 1.0e-03
[ ]𝑇
ĥ𝑘 = 0 . . . 0 ℎ̃𝑘, 0 . . . ℎ̃𝑘, 𝐿ℎ −1 . (30)
Thus
𝐿ℎ𝑟 = 2𝐿ℎ − 1. (31)
1.0e-04
0 1 2 3 4 5 6 7 8
Observe that the channel estimation matrix s̃1 in (26) remains
the same, independent of 𝑚0 . The channel estimate at 10 dB SNR per bit (dB)

Magnitude response of actual channel Fig. 4. RMS and maximum frequency offset estimation error for 𝐿𝑝 = 512.
Amplitude

Amp
Coarse

Magnitude response of estimated channel


Freq (rad)
Time (samp)
Amplitude

Amp
Fine

Subcarriers Freq (rad)


Time (samp)
Fig. 3. Channel estimation at 10 dB SNR per bit after fine frequency offset
compensation. 𝐿𝑝 = 512. Fig. 5. SoF, coarse and fine frequency offset estimate for 𝐿𝑝 = 512, SNR
per bit equal to 0 dB, 𝐵1 = 𝐵2 = 64.
SNR per bit is shown in 3.
The complexity of the two-stage approach is 𝐵1 + 𝐵2 =
C. Fine Frequency Offset Estimation 128 frequency bins. The resolution of the two-stage approach
For the purpose of fine frequency offset estimation, (19) is is 2 × 0.005/𝐵2 = 0.00015625 radian. For obtaining the
used with 𝑦˜𝑘, 𝑛 replaced by 𝑦ˆ𝑘, 𝑛 as given in (21). Moreover, same resolution, the single stage approach will require 2 ×
since the initial estimate of the frequency offset (ˆ 𝜔𝑘 ) is already 0.04/0.00015625 = 512 frequency bins. Therefore, the two-
available, (19) must be modified as follows: stage approach is four times more efficient than the single stage
𝐿 −1 approach.
∑2
∗ −j (ˆ 𝜔𝑘, 𝑓 )𝑖
𝜔𝑘 +ˆ
max 𝑟˜𝑚+𝑖 𝑦ˆ𝑘, 𝑖 e (32) D. Noise Variance Estimation
𝑚, 𝜔ˆ 𝑘, 𝑓
𝑖=0
It is necessary to estimate the noise variance for the purpose
where of turbo decoding [4]. After the channel has been estimated
𝐿2 = 𝐿ℎ𝑟 + 𝐿𝑝 − 1 using (28), the noise variance is estimated as follows:
0 ≤ 𝑚 ≤ 𝐿ℎ𝑟 − 1. (33) 2 1 ( )𝐻 ( )
𝜎
ˆ𝑤 = r̃𝑘, 𝑚1 − s̃1 ĥ𝑘 r̃𝑘, 𝑚1 − s̃1 ĥ𝑘 (34)
2𝐿1
Observe that the span of 𝑦ˆ𝑘, 𝑖 is 𝐿2 . The fine frequency
offset estimate (ˆ 𝜔𝑘, 𝑓 ) is obtained by dividing the interval where s̃1 is defined in (26) and 𝐿1 is defined in (11).

𝜔𝑘 − 0.005, 𝜔 ˆ 𝑘 + 0.005] radian into 𝐵2 frequency bins [17].
The reason for choosing 0.005 radian can be traced to Figure 4. E. Turbo Decoding
Note that the maximum error in the coarse estimate of the fre-
quency offset is approximately 0.004 radian (at 0 dB SNR per The generating matrix for each of the constituent encoders
bit) over 104 frames. Thus the probability that the maximum is given by:
[ ]
error exceeds 0.005 radian is less than 10−4 . Figure 5 gives 1 + 𝐷2
the results for SoF, coarse and fine frequency offset estimation G(𝐷) = 1 . (35)
1 + 𝐷 + 𝐷2
for one particular frame at 0 dB SNR, with 𝐵1 = 𝐵2 = 64.
The advantage of the two-stage approach (coarse and fine) for The overall rate of the encoder is 1/2. After SoF detection,
frequency offset estimation [17] is explained below. frequency offset compensation and channel estimation, the

Authorized licensed use limited to: QSIO. Downloaded on June 02,2020 at 04:34:34 UTC from IEEE Xplore. Restrictions apply.
output of the FFT can be written as (for 0 ≤ 𝑖 ≤ 𝐿𝑑 − 1): coded, “data” denotes 𝐿𝑑1 = 𝐿𝑑 /2, “Pr” denotes practical
˜ 𝑘, 𝑖 = 𝐻
ˆ 𝑘, 𝑖 𝑆𝑘, 3, 𝑖 + 𝑊
˜ 𝑘, 𝑖 . receiver (with acquired synchronization and channel estimates)
𝑅 (36) and “Id” denotes ideal receiver (ideal synchronization and
Note that 𝐻 ˆ 𝑘, 𝑖 and 𝑊˜ 𝑘, 𝑖 in (36) are the 𝐿𝑑 -point DFT of channel estimates).
the estimated channel ĥ𝑘 in (28) and 𝑤 ˜𝑘, 𝑛 in (5) respectively 1.0e+00
and 𝑆𝑘, 3, 𝑖 denotes the data symbols for the 𝑘 𝑡ℎ frame, for
0 ≤ 𝑖 ≤ 𝐿𝑑 − 1.
The variance of 𝑊 ˜ 𝑘, 𝑖 is 𝐿𝑑 𝜎𝑤
2
and that of 𝐻 ˆ 𝑘, 𝑖 (assuming
perfect channel estimates, that is 𝐻 ˆ 𝑘, 𝑖 = 𝐻 ˜ 𝑘, 𝑖 ) is 𝐿ℎ 𝜎 2 . For 1.0e-01
𝑓
the sake of brevity, the BCJR algorithm will not be discussed

BER
here [4]. Observe also that dividing (36) by 𝐻 ˜ 𝑘, 𝑖 results in
˜ ˆ
interference (𝑊𝑘, 𝑖 /𝐻𝑘, 𝑖 ) having a complex ratio distribution 1.0e-02
UC, data=512, Pr
TC, data=512, Pr
UC, data=1024, Pr
[27], [28], which is undesirable. TC, data=1024, Pr
UC, data=4096, Pr
TC, data=4096, Pr
UC, data=4096, Id
F. Data Interleaving TC, data=4096, Id
1.0e-03
Assuming ideal channel estimates, the autocorrelation of 0 1 2 3 4 5 6 7 8

the channel DFT at the receiver is: SNR per bit (dB)

1 [ ˜ ˜∗ ]
𝐿∑ℎ −1

𝐸 𝐻𝑘, 𝑖 𝐻𝑘, 𝑗 = 𝜎𝑓2 e− 2𝜋𝑛(𝑖−𝑗)/𝐿𝑑 . (37) Fig. 6. Simulation results without data interleaving.
2 𝑛=0

It has been found from simulations that the performance of the Observe that for 𝐿𝑑1 = 512, the practical receiver has a
turbo decoder gets adversely affected due to the correlation in performance that is less than 1 dB inferior to the ideal receiver.
𝐻˜ 𝑘, 𝑖 . To overcome this problem, the data is interleaved before However, the throughput of this system is just 50%, since the
the IFFT operation at the transmitter and deinterleaved after data length is equal to the preamble length. Next, for 𝐿𝑑1 =
the FFT operation at the receiver, before turbo decoding [29]. 1024, the practical receiver is about 1 dB inferior to the ideal
This process essentially removes any correlation in 𝐻 ˜ 𝑘, 𝑖 . An receiver and the throughput has improved to 66%. When 𝐿𝑑1 =
alternative to data interleaving would be to use a prediction 4096, the performance of the practical receiver is no better than
˜ 𝑘, 𝑖 [30]. uncoded transmission. This is due to the fact that the residual
filter to exploit the correlation in 𝐻
RMS frequency offset estimation error (fine) in Figure 4 is
about 1.5 × 10−4 radian (at 8 dB SNR per bit), which is about
IV. S IMULATION R ESULTS 19.5% of the subcarrier spacing (2𝜋/𝐿𝑑 = 0.000767 radian),
In this section, simulation results are presented for turbo- resulting in severe intercarrier interference (ICI). Note that the
coded OFDM. Since there are two coded QPSK symbols for frequency offset estimation error depends only on 𝐿𝑝 and the
every uncoded bit, the SNR per bit (over two dimensions) is performance of the ideal receiver is independent of the data
defined as: length 𝐿𝑑1 .
[ 2 ]
ˆ
2𝐸 𝐻𝑘, 𝑖 𝑆𝑘, 3, 𝑖 1.0e+00

SNR per bit = [ ]


ˆ 2
𝐸 𝑊 𝑘, 𝑖
1.0e-01

[ ] [
Bit error rate

]
ˆ 2 2 1.0e-02
2𝐸 𝐻𝑘, 𝑖 𝐸 ∣𝑆𝑘, 3, 𝑖 ∣
= [ ]
ˆ 2 1.0e-03
𝐸 𝑊𝑘, 𝑖 DIV, Data=512, Pr
DIV, Id
No DIV, Id
2 × 2 × 2𝐿ℎ 𝜎𝑓2 1.0e-04 No DIV, Data=512, Pr
= 2
UC, Id
2𝐿𝑑 𝜎𝑤 DIV, Data=1024, Pr
DIV, Data=4096, Pr
2 1.0e-05
4𝐿ℎ 𝜎𝑓 0 1 2 3 4 5 6 7 8
= 2
. (38)
𝐿𝑑 𝜎 𝑤 SNR per bit (dB)

The SNR per bit in decibels is Fig. 7. Simulation results with data interleaving.
( )
4𝐿ℎ 𝜎𝑓2
SNR per bit (dB) = 10 log10 2
. (39) Figure 7 presents the simulation results with data inter-
𝐿𝑑 𝜎 𝑤
leaving, as discussed in Section III-F. Again, the performance
In the simulations, the channel length 𝐿ℎ is equal to 10, hence of the ideal receiver is independent of 𝐿𝑑1 . Observe that the
𝐿ℎ𝑟 = 19. The fade variance 𝜎𝑓2 = 0.5. The simulation results practical receiver exhibits more than two orders of magnitude
are presented in Figure 6, for 𝐿𝑝 = 512 and different values improvement in the BER (compared to the case where there is
of 𝐿𝑑 . The term “UC” denotes uncoded, “TC” denotes turbo no data interleaving), at an SNR of 8 dB and 𝐿𝑑1 = 512. When

Authorized licensed use limited to: QSIO. Downloaded on June 02,2020 at 04:34:34 UTC from IEEE Xplore. Restrictions apply.
𝐿𝑑1 is increased, the performance of the practical receiver [11] H. Minn, V. K. Bhargava, and K. B. Letaief, “A Robust Timing
deteriorates, which is again due to the fact that the residual and Frequency Synchronization for OFDM Systems,” IEEE Trans. on
RMS frequency offset estimation error (fine) is a significant Wireless Commun., vol. 2, no. 4, pp. 822–839, July 2003.
fraction of the subcarrier spacing, leading to severe ICI. [12] Z. Zhang, K. Long, M. Zhao, and Y. Liu, “Joint Frame Synchronization
and Frequency Offset Estimation in OFDM Systems,” IEEE Trans. on
It is worth mentioning at this point that, while the results in Broadcasting, vol. 51, no. 3, pp. 389–394, Sept. 2005.
[29] (Figures 7 and 8) are impressive, the effect of frequency [13] C. Kuo and J.-F. Chang, “Generalized Frequency Offset Estimation in
offset is not considered. OFDM Systems Using Periodic Training Symbol,” in Proc. IEEE Intl.
Conf. on Commun., May 2005, pp. 715–719.
[14] S. Ahn, C. Lee, S. Kim, S. Yoon, and S. Y. Kim, “A Novel Scheme
V. C ONCLUSIONS AND F UTURE W ORK for Frequency Offset Estimation in OFDM Systems,” in 9th Intl. Conf.
on Adv. Commun. Technol., Feb. 2007, pp. 1632–1635.
In this paper, the problem of coherent detection of turbo-
[15] M. Henkel and W. Schroer, “Pilot Based Synchronization Strategy for
coded OFDM signals transmitted over frequency selective a Coherent OFDM Receiver,” in IEEE Wireless Communications and
Rayleigh fading channels, was addressed. Simulation results Networking Conference (WCNC), March 2007, pp. 1984–1988.
show that for small data lengths, the performance of the [16] K. Vasudevan, “Synchronization of Bursty Offset QPSK Signals in the
practical receiver is as good as the ideal receiver. Future work Presence of Frequency Offset and Noise,” in Proc. IEEE TENCON,
could be to improve the BER performance for larger data Hyderabad, India, Nov. 2008.
lengths and increasing the overall code-rate to unity and the [17] ——, “Iterative Detection of Turbo Coded Offset QPSK in the Presence
use of diversity. of Frequency and Clock Offsets and AWGN,” Signal, Image and Video
Processing, Springer, vol. 6, no. 4, pp. 557–567, Nov. 2012.
[18] J.-J. van de Beek, O. Edfors, M. Sandell, S. K. Wilson, and P. O.
ACKNOWLEDGMENT Börjesson, “On Channel Estimation in OFDM Systems,” in Proc. of
the IEEE VTS 45th Vehicular Technology Conf., July 1995, pp. 815–
This work is supported by the India-UK Advanced Tech- 819.
nology Center (IU-ATC) of Excellence in Next Generation [19] O. Edfors, M. Sandell, J.-J. van de Beek, S. K. Wilson, and P. O.
Networks, Systems and Services under grant SR/RCUK- Börjesson, “OFDM Channel Estimation by Singular Value Decompo-
DST/Next Gen(F)/2008, sponsored by DST-EPSRC. sition,” IEEE Trans. on Commun., vol. 46, no. 7, pp. 931–939, July
1998.
This work was presented as an Invited Talk at the In- [20] S. Coleri, M. Ergen, A. Puri, and A. Bahai, “Channel Estimation
ternational Federation of Nonlinear Analysts (IFNA) World Techniques Based on Pilot Arrangement in OFDM Systems,” IEEE
Congress, University of Athens, Greece, 25th June–1st July, Trans. on Broadcasting, vol. 48, no. 3, pp. 223–229, Sept. 2002.
2012. [21] C. Ribeiro and A. Gameiro, “An OFDM Symbol Design for Reduced
Complexity MMSE Channel Estimation,” Journal of Communications,
Academy Publisher, vol. 3, no. 4, pp. 26–33, Sept. 2008.
R EFERENCES
[22] S. Kinjo, “An MMSE Channel Estimation Algorithm Based on the
[1] L. Hanzo, H. Haas, S. Imre, D. O. Brien, M. Rupp, and L. Gyongyosi, Conjugate Gradient Method for OFDM Systems,” in The 23rd In-
“Wireless Myths, Realities, and Futures: From 3G/4G to Optical and ternational Technical Conference on Circuits/Systems, Computers and
Quantum Wireless,” Proc. IEEE, vol. 100, no. Special Centennial issue, Communications (ITC-CSCC), July 2008, pp. 969–972.
pp. 1853–1888, May 2012. [23] H. Sari, G. Karam, and I. Jeanclaude, “Transmission Techniques for
[2] R. Zhang et al., “Advances in Base- and Mobile-Station Aided Cooper- Digital Terrestrial TV Broadcasting,” IEEE Commun. Mag., vol. 33,
ative Wireless Communications,” IEEE Veh. Tech. Mag., vol. 8, no. 1, no. 2, pp. 100–109, Feb. 1995.
pp. 57–69, March 2013. [24] U. Reimers, “Digital Video Broadcasting,” IEEE Commun. Mag.,
[3] J. A. C. Bingham, “Multicarrier Modulation for Data Transmission: An vol. 36, no. 6, pp. 104–110, June 1998.
Idea Whose Time Has Come,” IEEE Commun. Mag., vol. 28, no. 5, pp. [25] H. Takahashi, “Coherent OFDM Transmission with High Spectral
5–14, May 1990. Efficiency,” in 35th European Conference on Optical Communication,
[4] K. Vasudevan, Digital Communications and Signal Processing, Second Sept. 2009, pp. 1–4.
edition (CDROM included). Universities Press (India), Hyderabad, [26] S. Haykin, Adaptive Filter Theory, 3rd ed. Prentice Hall, 1996.
www.universitiespress.com, 2010.
[27] R. J. Baxley, B. T. Walkenhorst, and G. Acosta-Marum, “Complex
[5] L. Hanzo and T. Keller, OFDM and MC-CDMA: A Primer. John Gaussian Ratio Distribution with Applications for Error Rate Calcu-
Wiley, 2006. lation in Fading Channels with Imperfect CSI,” in Proc. IEEE Global
[6] J.-J. van de Beek, M. Sandell, M. Isaksson, and P. O. Börjesson, Telecomm. Conf., Dec. 2010, pp. 1–5.
“Low-Complex Frame Synchronization in OFDM Systems,” in Proc. [28] J. J. Sánchez-Sánchez, U. Fernández-Plazaola, and M. Aguayo-Torres,
of the 4𝑡ℎ IEEE International Conference on Universal Personal “Sum of Ratios of Complex Gaussian RVs and its Application to a
Communications, Nov. 1995, pp. 982–986. Simple OFDM Relay Network,” in Proc. of the IEEE VTS 71st Vehicular
[7] K. L. Baum, “A Synchronous Coherent OFDM Air Interface Concept Technology Conf., May 2010, pp. 1–5.
for High Data Rate Cellular Systems,” in Proc. of the IEEE VTS 48th [29] R. F. H. Fischer, L. H.-J. Lampe, and S. H. Müller-Weinfurtner, “Coded
Vehicular Technology Conf., May 1998, pp. 2222–2226. Modulation for Noncoherent Reception with Application to OFDM,”
[8] M. Julia Fernández-Getino Garcia, O. Edfors, and J. M. Páez-Borrallo, IEEE Trans. on Veh. Technol., vol. 50, no. 4, pp. 910–919, July 2001.
“Frequency Offset Correction for Coherent OFDM in Wireless Sys- [30] K. Vasudevan, K. Giridhar, and B. Ramamurthi, “An Efficient Subop-
tems,” IEEE Trans. on Consumer Electronics, vol. 47, no. 1, pp. 187– timum Detector Based on Linear Prediction in Rayleigh Flat-Fading
193, Feb. 2001. Channels,” Signal Processing Journal, Elsevier Science, vol. 81, no. 4,
[9] D. Landström, S. K. Wilson, J.-J. van de Beek, P. Ödling, and P. O. pp. 819–828, April 2001.
Börjesson, “Symbol Time Offset Estimation in Coherent OFDM Sys-
tems,” IEEE Trans. on Commun., vol. 50, no. 4, pp. 545–549, April
2002.
[10] I. Bradaric and A. P. Petropulu, “Blind Estimation of the Carrier
Frequency Offset in OFDM Systems,” in 4th IEEE Workshop on Signal
Processing Advances in Wireless Communications, June 2003, pp. 590–
594.

Authorized licensed use limited to: QSIO. Downloaded on June 02,2020 at 04:34:34 UTC from IEEE Xplore. Restrictions apply.

You might also like