You are on page 1of 13

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2016.2594176, IEEE
Transactions on Wireless Communications

Self-Coherent OFDM with Undersampling


Down-conversion for Wireless Communications
Qianyu Jin, Student Member, IEEE and Yi Hong, Senior Member, IEEE

AbstractIn this paper, we introduce self-coherent orthogonal Self-heterodyne (self-het) OFDM was proposed by Shoji
frequency-division multiplexing (OFDM), a well-known non- et al. in [4] to cope with high level oscillator instabilities in
coherent technique in optical communications, for wireless radio 60 GHz wireless communications using quadrature amplitude
frequency (RF) communications. Self-coherent OFDM provides modulation (QAM)/OFDM signalling. In a self-het OFDM sys-
complete immunity against phase noise (PN) using a non- tem, the local radio frequency (RF) carrier is transmitted with
coherent receiver and a significantly higher spectral efficiency
than self-heterodyne (self-het) OFDM, which utilizes at most
the information subcarriers. This guarantees that the carrier
50% of the available spectrum for communications. We present phase is perfectly synchronous with the subcarriers and it pro-
the performance analysis of self-coherent OFDM over additive vides complete immunity against PN. Moreover, a square-law
white Gaussian noise (AWGN) and frequency selective fading circuitry (self-mixing) is used, instead of a super-heterodyne
channels, and show by simulations that self-coherent OFDM structure, to down-convert the RF signal. Thus the processes
provides both higher spectral efficiency and better bit error of local carrier generation, carrier frequency correction and
rate (BER) performance than self-het OFDM. Considering that carrier phase recovery can be omitted. This greatly reduces
filter realization in high frequency bands is challenging, we the complexity of the self-het OFDM transceivers compared to
adopt a undersampling down-conversion technique in conjunction conventional OFDMs using super-heterodyne receivers. There
with self-coherent OFDM. We show that with the self-coherent have been a number of research developments on self-het
demodulation, the additional PN introduced by undersampling
down-conversion can be significantly reduced. We compare ana-
OFDM for additive white Gaussian noise (AWGN) and two-
lytically the system performance of self-coherent OFDM using ray channels [5]. Very recently, Fernando et al extended the
undersampling down-conversion with two other conventional self-het OFDM system to frequency selective fading channels
OFDM systems: one with super-heterodyne receiver and the other and applied coding techniques to further improve the system
with undersampling down-conversion. We show theoretically and performance [6], [7]. However, the disadvantage is that self-het
by simulations that both in AWGN and frequency selective OFDM uses at most 50% of the available spectrum.
fading channels, self-coherent OFDM with undersampling down- To improve spectral efficiency, while maintaining the sim-
conversion outperforms the two conventional OFDM systems even plicity of the RF front-end receiver and PN immunity, we
when intercarrier interference (ICI) compensation schemes are consider self-coherent OFDM, a popular technique in optical
applied. communications, originally proposed by Tetsuya Miyazaki
Keywordsnon-coherent, OFDM, self-coherent, phase noise, in [8]. Self-coherent OFDM jointly transmits a carrier and
multipath fading, low-complexity receivers, undersampling down- information subcarriers separated by a guard band, much
conversion. smaller than self-het OFDM, to ensure phase synchronization
between transmitter and receiver. In optical communications,
I. I NTRODUCTION the homodyne detection with a polarization-modulation tech-
Orthogonal frequency division multiplexing (OFDM) is a nique was used to generate a pilot carrier at the transmitter
popular technique that brings the advantages of high spec- and a pilot-carrier combining module at the receiver. Recent
tral efficiency, robustness to intersymbol interference, simple research on self-coherent OFDM for optical communications
channel equalization and efficient implementation using Fast is summarized in [9].
Fourier Transform (FFT). It is known that the intercarrier inter- In this paper, we adapt self-coherent OFDM to wireless
ference (ICI) is a major problem in conventional OFDM due to communications over AWGN and frequency selective fading
Doppler frequency drift, phase offset, local oscillator frequency channels. In particular, at the receiver of self-coherent OFDM,
drift, and sampling clock offset [1]. For very high frequencies, the carrier and information subcarriers are separated using a
used in millimeter-wave and terahertz communications, high low pass filter (LPF) and a high pass filter (HPF), respectively,
level residual phase noise (PN) due to instabilities of oscillators and are further processed by two square-law devices and a
and mixers of the coherent reception can be severe [2]. LPF to down-convert the RF signal. For very high frequency
Hence, new OFDM techniques that are robust to ICI and bands such as millimeter-wave RF bands, the realization of
equipped with low complexity transceiver structures are needed such LPF and HPF can be challenging. We thus introduce
to implement millimeter-wave wireless communications [3]. the undersampling down-conversion technique to self-coherent
OFDM. After undersampling at the receiver, the received
Qianyu Jin and Yi Hong are with Department of Electrical and Computer Systems signal is down-converted to a lower frequency band, where
Engineering (ECSE), Faculty of Engineering, Monash University, Clayton, VIC 3800, filters can be more easily implemented to demodulate the in-
Australia (e-mail: {qianyu.jin, yi.hong}@monash.edu). This work is supported by the
Australian Research Council Discovery Projects with ARC DP130100336 and ARC formation subcarriers. In the undersampling down-conversion,
DP160100528. PN is additionally introduced to the system by the instability

1536-1276 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2016.2594176, IEEE
Transactions on Wireless Communications

of analogue-to-digital converter (ADC). We will further show denotes the real part of a complex number. Let
that self-coherent demodulation can significantly reduce the
impairment of such PN and guarantee the system robustness. N1
The main contributions of our paper are: s(t) , Xk ej2(fc +kf )t

k=Ng
Performance analysis of self-coherent OFDM: we anal-
yse theoretically the bit error rate (BER) performance represent the information-bearing bandpass signal for 0
of self-coherent OFDM over AWGN and frequency t < T , and Gs (f ) be the corresponding power spectral density
selective channels. To the best of our knowledge, self- (PSD).
coherent OFDM has only been studied for optical When x(t) is transmitted over AWGN channels, the received
communications based on simulations and experimen- bandpass signal at the receiver antenna can be represented as
tal tests, yet the performance analysis of self-coherent r(t) = x(t) + n(t), where n(t) N (0, 2 ) is AWG noise
OFDM has never been investigated. process with the power 2 . The spectrum at the receiver is
Self-coherent OFDM with undersampling down- depicted in Fig. 1(b), where a bandpass filter (BPF) with a
conversion: we introduce the undersampling down- bandwidth Bl + Bg + Bs + Bh is used to isolate the channel
conversion technique to self-coherent OFDM and of interest, where Bl and Bh represent its lower and upper
compare analytically the system performance with transition bands, respectively. Subsequently, a LPF with output
self-het OFDM and two conventional OFDMs: one rc (t) is operated to extract the carrier, while a HPF with output
with super-heterodyne receiver and the other with rs (t) is employed to extract the information signal, given by
undersampling down-conversion. We show theoretically
rc (t) = {Aej2fc t } + nc (t) rs (t) = s(t) + ns (t)
and by simulations that self-coherent OFDM with
undersampling down-conversion outperforms self-het where nc (t) and ns (t) denote the noise components at the
OFDM over a wide range of SNRs, and these two output of the LPF and HPF, whose PSDs are indicated by
conventional OFDMs, even when ICI compensation Gnc (f ) and Gns (f ), respectively. The PSDs of rc (t) and
schemes are applied [19], [20], [22]. rs (t) are denoted by Grc (f ) and Grs (f ), respectively (see
Fig. 1(b)).
Finally, the signal y(t) is obtained by two square-law
II. S YSTEM M ODEL AND P ERFORMANCE A NALYSIS OF devices that perform the self-coherent down-conversion as
S ELF -C OHERENT OFDM
1
y(t) = {[rc (t) + rs (t)]2 [rc (t) rs (t))]2 }
A. Self-coherent OFDM over AWGN Channels 4
= rc (t) rs (t) . (2)
A block diagram of a self-coherent OFDM communication
system is shown in Fig. 1(a), where Bg is the frequency gap A LPF then extracts the baseband information subcarriers. This
between the RF carrier and the first OFDM subcarrier, and Bs enables the system to demodulate the received signal without a
is the bandwidth of the effective OFDM subcarriers. Different local carrier generated by an oscillator, which can be unstable
from self-het OFDM we can have Bg Bs . However, Bg at very high frequencies (e.g. 60GHz). Then we obtain the
should be large enough to accommodate the filter transition PSD of y(t) similar to [12], [13],
bands. Let f = 1/T represent the subcarrier spacing, T
the symbol period, Ng = Bg /f be the number of guard Gy (f ) = Grs (f ) Grc (f )
band subcarriers, Ns = Bs /f denote the number of OFDM A2
= Gs (f ) {(f fc ) + (f + fc )}
subcarriers used to encode information. Let N = Ns + Ng be 4
| {z }
the size of Fast Fourier Transform (FFT), fc be the RF carrier {1}
frequency, and Ts be the sampling clock period at the receiver.
A2
At the transmitter, information bits are first mapped into + Gns (f ) {(f fc ) + (f + fc )}
QAM symbols Xk , k = Ng , . . . , N 1 and located over the | 4 {z }
last Ns OFDM subcarriers, while the remaining Ng OFDM {2}
subcarriers are set to zero1 . Through IFFT, parallel-to-serial + Gs (f ) Gnc (f ) + Gns (f ) Gnc (f ) (3)
conversion, addition of cyclic prefix (CP), digital-to-analogue | {z } | {z }
conversion (DAC), and up-conversion, the time-domain OFDM {3} {4}

symbol without CP can be written as [10] where denotes convolution operation. For simplicity, we
assume that Bg = Bl + Bh , where Bl = f Nl , Bh =

N 1 f Nh (LPF and HPF transition bands, respectively), and
x(t) = Aej2fc t + Xk ej2(fc +kf )t (1) Ng = Nl + Nh . We assume that both s(t) and n(t) are zero-

k=Ng mean Gaussian random processes with PSD of S0 and N20 .
Term {1}, produced by mixing the carrier with the information
for 0 t < T , where A is the RF carrier amplitude and () subcarriers, contains the useful information. Term {2} is the
down-converted ns (t) noise obtained by mixing with the RF
1 The carrier can be added after IFFT. carrier. Term {3} is the mixing of carrier noise nc (t) and

1536-1276 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2016.2594176, IEEE
Transactions on Wireless Communications

fc Subcarriers

rc (t )
Bg Bs LPF
| |2 y (t )
N-IFFT Ts N-FFT
OFDM BPF LPF OFDM
TX r (t ) RX
| |2
HPF
rs (t )

(a)

HPF
LPF
BPF
fc Gs ( f )
Grc ( f ) Grs ( f )
Bl Bh
Gnc ( f ) Gns ( f ) f

Bl Bg Bs Bs Bh

(b)

Fig. 1. (a) Self-coherent OFDM system model. (b) Self-coherent OFDM PSD of x(t) with Bg Bs and filter masks. Note: for self-het OFDM, the dashed-line
box in (a) is replaced by a single square law device and in (b) Bg Bs [6].

the OFDM subcarriers. Finally, term {4} comes from the In particular, Bg of self-coherent OFDM is controlled by the
mixing of the two noise terms ns (t) and nc (t). Note that the LPF parameters. If an ideal LPF is used then Bg can be zero
combination of the terms {2}, {3} and {4} represents the total and the spectral efficiency is 100%. If a non-ideal LPF is used,
noise plus interference impairments. Hence, we can define an then Bg 0. For example, in our simulation with a non-ideal
equivalent baseband model in the frequency domain as LPF, we use Ns = 472 and Ng = 40, the spectral efficiency
is 92%.
Yk = AXk + Zk , k = Ng , ..., N 1 (4)
where Yk is the received information symbol of the k-th
OFDM subcarrier and Zk denotes the equivalent noise formed B. Self-Coherent OFDM over Frequency Selective Channels
by the terms {2}, {3}, and {4} in (3). We note that the variance Let h(t) be the frequency selective channel impulse re-
of Zk varies with k, and Zk is correlated with input Xk . Note sponse. Using the same settings as above, the received signal
that X0 = A and Xk = 0, for k = 1, ..., Ng 1. For large r(t) is given by r(t) = h(t)x(t)+n(t), where x(t) is given in
FFT, we can neglect such correlations [6]. (1). The PSD of r(t) is represented by Gr (f ) which is given
Comparing the spectrum of self-coherent and self-het 2
by Gr (f ) = (fc ) + |H(f )| Gs (f ) + Gn (f ), where H(f )
OFDM symbols (see Fig. 1(b)), we note that the major denotes channel response and
difference lies in the guard band Bg : for self-coherent OFDM
Bg Bs , while for self-het OFDM Bg Bs . Specifically, A2 |H(fc )|
2

self-coherent OFDM uses two-square law devices to obtain (fc ) = {(f fc ) + (f + fc )}


4
the down-converted signal y(t) in (2). In contrast, self-het
OFDM uses a self-mixing device to square the received signal where H(fc ) denotes the channel response at carrier frequency.
r(t), i.e., y(t) = |r(t)|2 . This squaring creates a number of Fig. 2 illustrates Gr (f ) over frequency selective channels,
inter-modulation products and requires Bg Bs to isolate where Gx (f ) represents PSD of the transmitted signal. At the
the dominant inter-modulation impairment. Disregarding the receiver, similar to the AWGN case, the nonlinear operation in
CP in both cases, the spectral efficiency of self-het OFDM is (2) is performed after the signal is passed through the BPF, and
Bs /(Bs + Bg ) 50% since Bg Bs , while the spectral effi- then followed by the LPF and HPF. As the transmitted signal
ciency of self-coherent OFDM is much higher since Bg Bs . is propagating over frequency selective channels, the PSD of

1536-1276 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2016.2594176, IEEE
Transactions on Wireless Communications

Gx ( f )
A2 A2 The individual terms in (5) are illustrated in Fig. 3, which
Subcarriers 4 4 Subcarriers result from the same procedure described in (3). Note that the
S0 S0
term {1} that contains the useful information data is not only
f (a) affected by H(f ) but also H(fc ) due to the term (fc ). In



this scenario, the received carrier is faded independently, and
Bl Bg Bs Bh
2 utilized for the down-conversion of the information signal. The
H(f)
equivalent frequency model in this case is given by
Yk = AHc Hk Xk + Zk (6)
f
(b) where () denotes complex conjugation, Hc and Hk are the


Bl Bg Bs Bh
channel responses at the carrier and the k-th OFDM subcarrier
Gr ( f )
respectively, and Zk is the equivalent noise due to the terms
S0 H ( f )
2
1 2 2 1 2 2 S0 H ( f )
2
{2}, {3} and {4} in (5), as discussed in the following section.
A H ( fc ) A H ( fc )
4 4 To provide BER analysis over frequency selective channels,
N0 fc we first analyse the signal-to-interference-noise ratio (SINR)
2 f (c ) of the k-th OFDM subcarrier, denoted by k . Referring to

Bl Bg Bs Bh Fig. 3, the SINR of the k-th subcarrier is a power ratio that


can be computed by integrating all the terms in (5) over the
Fig. 2. Baseband PSDs: (a) PSD Gx (f ) of transmitted signal, (b) |H(f )|2 signal bandwidth, i.e.,
frequency response of the frequency selective channel, (c) PSD Gr (f ) of the
received signal. A2 |Hc |2
4 S0 |Hk |2
k = ( )2 (7)
N0 A2 |Hc |2 N0 N0
{1} Carrier s(t) S0 H ( f )
2
S0 f (k) + + 2Bl
|2 {z } | 4 {z 2}
| {z
2
}
{2} {3} {4}
f
Bl Bh Bs where
k+N
{2} nc (t) s(t) f + Bl 2 N0
m=Ng |Hm |
l 2
Ng k Ng + Nl 1
f -Bl S0 H ( u ) du
2 (k) , k+Nl
m=kNl |Hm |
2
Ng + Nl k Nh + Ns 1
f
Ns
m=kNl |Hm | Nh + Ns k Ng + Ns 1 .
2

2 Bl 2 Bl Bh Bs - 2 Bl
Note that the term {2} is the result of the convolution integral
{3} Carrier ns (t) 1 2
A H ( fc ) N 0
2

8
f +Bl
N0
S0 |H(u)|2 du .
f f Bl 2
Bl Bh Bs Bh From (7), we also note that the square law detection causes
{4} nc (t) ns (t) 1 performance degradation, and the interference-and-noise item
Bl N 02 is non-Gaussian [31]. Furthermore, we define the power ratio
2
f between the RF carrier and information subcarriers as
Bs + 2 Bh - 2 Bl 1 2
2A
2 Bl 2 Bl
, (8)
2S0 Bs
Fig. 3. Self-coherent OFDM baseband components.
and the average signal-to-noise ratio (SNR) as
2S0
y(t) is given by , . (9)
N0
2 According to (7), the SINR of the k-th OFDM subcarrier
Gy (f ) = {Gs (f )|H(f )| } (fc ) depends on the |Hc |2 and |Hk |2 distributions. To simplify
| {z }
{1} notation, we define
2
+ {Gs (f )|H(f )| } Gnc (f ) , |Hc |2 and k , |Hk |2 . (10)
| {z }
{2}
We further assume that and k undergo independent fading
+ Gns (f ) (fc ) + Gns (f ) Gnc (f ). (5)
| {z } | {z } since the carrier and OFDM information subcarriers are sepa-
{3} {4} rated by a guard band. If the channel is Rayleigh fading (i.e.

1536-1276 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2016.2594176, IEEE
Transactions on Wireless Communications

Hc , Hk NC (0, 1)), then both and k follow an exponential


distribution. Substituting (8), (9), and (10) into (7) yields rc (t)
LPF
k ro (t) | |2 y(t)
k = (k) 2Nl
. TSS
Ns ++ Ns
Ts N-FFT
BPF r[n] LPF OFDM
r (t) RX
where the noise power of the discrete equivalent baseband Ideal
model in (6) is given by undersampling | |2
( ) HPF
2 2 2 (k) 2Nl rs (t)
Zk = A ++ .
Ns Ns
If M-QAM information symbols are used to transmit the Fig. 4. Receiver structure of self-coherent OFDM with undersampling down-
data, the average BER of the k-th (k = Ng , . . . , N 1) conversion technique.
OFDM subcarrier can be derived as (see the detailed proof in
Appendix A),
[ ( )]
1 2(M 1) E((k)) 2Nl
BERk U , 0, +
2 3 + 2(M 1) N Ns
( )( s ) BPF BPF
2 1 1 3
1 1 f



log2 (M ) M 2 3 + 2(M 1) - fc f L fc fH

where E((k)) denotes the expected value of (k), which is (a)


a constant depending only on k, and U(a, b, z) is the confluent
hypergeometric function of the second kind and given by [14]

1
U(a, b, z) = ezt ta1 (1 + t)ba1 dt
(a) 0
f



where () is the gamma function. The overall BER of self-
f Bgb fu
coherent OFDM is the average BERk across the OFDM - u
2 2
information subcarriers which is given by
(b)
N 1
1
BER , BERk . (11)
Ns
k=Ng

To improve the overall performance, the subcarrier pairing


technique [7] may be used to exploit the unbalanced qualities f

of the subcarriers. However, the SINR differences between the


f fu
subcarriers of self-coherent OFDM are much less than self- - u
2 2
het OFDM (see Fig. 6). Thus the performance gain of self-
coherent OFDM using pairing may not be as significant as (c)
self-het OFDM.
Fig. 5. (a). self-coherent OFDM RF band signal before undersampling; (b).
self-coherent OFDM signal after undersampling when M is odd; (c). self-
III. D OWN - CONVERSION OF S ELF -C OHERENT OFDM coherent OFDM signal after undersampling when M is even.
U SING U NDERSAMPLING
For millimeter wave (mmWave) communications, self- A. Undersampling Down-conversion Technique
coherent OFDM needs to operate at very high frequencies
such as 60GHz and requires analog LPF/HPFs with narrow We assume that the uniform ideal undersampling down-
transition bands. Unfortunately, such LPF/HPFs at 60GHz conversion technique is used. To avoid aliasing, we first
usually have a transition band over 1GHz [29], [30]. To solve use a BPF to isolate the frequencies in the range [fL , fH ]
this problem, we consider undersampling down-conversion containing the band pass signal (see Fig. 5(a)). Then we
technique in our scheme that can down-convert the OFDM undersample the bandpass signal at rate fu = T1u , where Tu
signal to a lower frequency band, in which LPF/HPFs can be is the undersampling interval. If this is below the Nyquist
realized more easily. Fig. 4 illustrates the receiver structure rate, one copy of the undersampled signal spectrum will
when the ideal undersampling down-conversion technique is appear within [fu /2, fu /2] baseband without any aliasing, as
used after the BPF. illustrated in Fig. 5. In particular, Fig. 5(a) illustrates the PSD

1536-1276 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2016.2594176, IEEE
Transactions on Wireless Communications

Channel fL (GHz) fH (GHz) M fu range(GHz)


1 57.24 59.40 26 4.569 4.579 where Ncp is the length of cyclic prefix and u[i] are mutually
2 59.40 61.56 28 4.397 4.400 independent Gaussian random variables with zero mean and
4.551 4.560 T
3 61.56 63.72 28 variance u2 = 2f 2 . Another important variable that is used
4 63.72 65.88 30 4.392 4.394 cN
as the measure of the variance of timing jitter or PN is the ratio
TABLE I. U NDERSAMPLING FREQUENCY RANGE FOR 60GH Z r3dB = 2f
, which is referred to as the relative PN bandwidth
COMMUNICATIONS IN IEEE 802.11ad [24].
[19].
In the following, we compare theoretically the system per-
formance of different OFDM systems: (i) self-coherent OFDM
|F {y(t)}|2 of the bandpass signal and Figs. 5(b), (c) the PSD with undersampling down-conversion, (ii) conventional OFDM
|DTFT{y(nTu )}|2 of the undersampled signal, where F {} with super-heterodyne receiver, (iii) conventional OFDM with
represents Fourier transform and DTFT() denotes discrete- undersampling down-conversion, and (iv) self-het OFDM2 . We
time Fourier transform, respectively. compare the performance of these four OFDM systems and
The appropriate sampling rate is given by [26] derive the respective SINR expressions.

2fH 2fL fH
fu , 1<M , for some integer M
M M 1 fH fL C. Timing Jitter Cancellation of Self-Coherent OFDM with
where fL and fH are the lower and upper cut-off frequencies Undersampling Down-conversion
of the BPF, and represents the floor function. Specifically, As introduced in Section III.A, when M is chosen as even,
when M is odd as in Fig. 5(b), the carrier is located at a lower the output signals of the LPF and HPF are swapped. Different
frequency than that of the subcarriers. In order to demodulate from standard self-coherent OFDM in Fig. 1(a), the HPF is
the signal without aliasing, a guard band Bgb > Bs is needed used to extract the carrier and the LPF is used to extract the
and this increases the undersampling frequency. When M is subcarriers. In Fig. 4, let ro (t) represent the band-limited signal
even, no guard band is required as shown in Fig. 5(c), while filtered by BPF. After undersampling at rate fu = T1u , the
the output signals of the LPF and HPF are swapped. In our resulting signal r[n] = r0 (nTu )(t nTu ) = x[n] + z[n],
case, we choose the largest even integer M within the bounds where () represents the Dirac function, and x[n] and z[n]
to reduce the undersampling frequency fu as shown in Table I, are the n-th samples of the OFDM symbol and the Gaussian
where fL and fH are given in IEEE 802.11ad [24]. noise, respectively.
In general, due to the instability of the sampling clock, the After undersampling, (1) becomes
undersampling down-conversion technique introduces timing { }
jitter, which results in PN degrading the system performance. x[n] = Aej2fc (nTu +u [n])
However, we will show later that if self-coherent demodulation
is used, the effect of PN can be significantly reduced. N 1
+ Xk ej2(fc +kf )(nTu +u [n]) (12)

B. Timing Jitter Model k=Ng

In a free-running oscillator, PN is usually modeled as a where u [n] denotes the time jitter at the n-th sample of an
Wiener process [16], OFDM symbol, which is due to the clock instability of the
t ADC.
(t) = (u)du Let fc = nfu + fI , where fI , the remainder of fc divided
0
by fu , is the intermediate frequency (IF) of down-conversion,
where (u) is a zero-mean Gaussian random process with then (12) can be rewritten as
variance 2, where is the PN linewidth, i.e., frequency { }
spacing between 3dB points of its Lorentzian PSD. Therefore, x[n] = Aej2fI nTu ej2fc u [n] (13)
(t) is a normal process
[ with] zero mean and variance increas-
ing with time, i.e., E 2 (t) = 2t, where E[] represents N 1
expectation. + Xk ej2(fI +kf )nTu ej2(fc +kf )u [n]

Phase noise (t) and timing jitter (t) are related by (t) = k=Ng
t
2fc (t) [17]. The timing jitter has variance E[2 (t)] = 2f 2
c If the sideband of PN is narrow, i.e. r3dB is small, the HPF
and autocorrelation function [16] can extract the carrier and most of the PN. Then rc (t) and
rs (t) can be written as
E[(t1 ) (t2 )] = min(t1 , t2 ) .
2fc2 { }
bc (t)
rc (t) = Aej2fI t ej2fc u (t) + n
At the receiver, after the N -point FFT, the time jitter affecting
the n-th sample of one OFDM symbol can be modeled as a
2 Self-het OFDM uses a BPF and a self-mixing square-law device for
discrete-time Wiener process [15], i.e.,
the down-conversion, which does not require undersampling. Hence, we

n only compare analytically the performance of self-coherent OFDM with
[n] = u[i], n = 0, . . . , N + Ncp 1 undersampling down-conversion and self-het OFDM without undersampling
i=0 down-conversion.

1536-1276 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2016.2594176, IEEE
Transactions on Wireless Communications


N
1
N 1

rs (t) = Xk ej2(fI +kf )t ej2(fc +kf )u (t) +b


n (t) ek = 1
N e[n]ej2nk/N
n
s N n=0
k=Ng
and Xk is the k-th information subcarrier of one symbol.
where nbc (t) and n
bs (t) denote the I and Q Gaussian noise com-
Similarly, for frequency selective channels, Yk can be written
ponents affected by PN with Gaussian probability distribution
as
functions (PDFs) [28]. We remark that after passing through
LPF/HPF, the signals are converted back from discrete-time to Yk = AHc Xk I0 + + N
(I) ek
k
continuous-time. Here we assume ideal sampling to simplify
our analysis. In practice, real sampling can be used instead where

N 1
and our analysis can be easily extended to this case. AHc Xl Hl I(lk)modN
(I)
k =
Then, the signal y(t) is obtained by two square-law devices l=0,l=k
as in (2), i.e.,
represents the ICI in self-coherent OFDM with undersampling
y(t) = rc (t) rs (t)
down-conversion, where Hc and Hl are the channel responses
N 1 at the carrier and at the l-th OFDM subcarriers. Assuming
= AXk ej2kf t ej2kf u (t) N is large and E[|Hl |2 ] = 1, the variance of ICI can be
approximated by [21, Eq. (5)]
k=Ng

N 1 4
2(I) u T A2 S0
bc (t)
+ n Xk ej2(fI +kf )t ej2(fc +kf )u (t) k 3

k=Ng where u is the two-side linewidth of the Lorentzian PSD of
{ }
the PN term ej2N f u (t) , S0 is the PSD of the signal s(t)
bs (t) Aej2fI t ej2fc u (t) + n
+ n bc (t)b
ns (t) (14)
and = |Hc |2 .
From the above results, we can see that different subcarriers are We can model the noise term N ek in (15) as a Gaussian
impaired by different levels of PN. However, as 2f u (t) is random variable with N (0, N 2
ek ), with
a very small phase perturbation, the differences of PN between ( )
subcarriers can be ignored. We approximate the subcarriers 2 2 2 2 (k) 2Nl
Ne = Zk = A ++
PN by assuming u (t) = 2N f u (t). Since fc kf , k Ns Ns
we ignore kf u (t) in the second term of (14). Accordingly, where (k) is given in (7).
y(t) can be approximated as It was shown in [21] that, when N is large enough, we have

N1 1
N
[ ] [ ] 2T
y(t) Aej2N f u (t) Xk ej2kf t + n e(t) E |Ik |2 = 1 and E |I0 |2 = 1 .
3
k=Ng k=0

where Consequently, the SINR of the k-th subcarrier of self-coherent


OFDM using undersampling down-conversion can be given as

N 1 [ ] [ ] [ ]
e(t) =
n bc (t)
n ej2fc u (t) Xk ej2(fI +kf )t A2 E |Hc |2 E |Xk |2 E |I0 |2
k
(I)
=
k=Ng N
1
{ } N2 + E [|Xl |2 ] E [|Hl |2 ] E [|Ilk |2 ]
e
bs (t) Aej2fI t ej2fc u (t) + n
+ n bc (t) n
bs (t) k
l=0,l=k

is the mixed noise term produced by square-law devices. 1 23 u T


= ( ) (16)
This noise term is a function of the I and Q Gaussian noise (k)
1
Ns + + N2Ns
l
+ 32 u T
components n c (t), n
s (t) and the PN.
After sampling y(t) with the interval Ts and applying a N - where is given in (9).
FFT 3 , the OFDM symbol is converted back from continuous-
time to discrete-time and can be represented as D. Analytical Performance Comparison

N 1 1) Performance Comparison of Self-coherent OFDM and
Yk = AXk I0 + ek
Xl I(lk)modN + N the Conventional OFDM with Super-heterodyne Receiver in
l=0,l=k 60GHz: For conventional OFDM, the received transmitted
where signal is given as [22]
{ 1
}
N 1
1 j2N f u [n] j2nk/N N
L1
Ik = e e (15) r (t) = e
(c) j2fc t
hl Xk ej2kf (tl )
+ n(c) (t)
N n=0 l=0 k=0
(c)
3 Note that in practice, a 2N -real FFT can be used. However, for simplicity where superscript represents the conventional OFDM, the
of notations, we write the equations using N -point complex FFT. coefficients {hl }L1
l=0 represent the equivalent discrete-time

1536-1276 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2016.2594176, IEEE
Transactions on Wireless Communications

channel impulse response of length L, l is the time delay which followed by a BPF to attain the signal located in
of the l-th generic impulse, and n(c) (t) is the AWGN noise. intermediate frequency band. We remark that after passing
At the receiver, the local carrier is impaired by PN due to through BPF, the signals are converted back from discrete-
the instability of the local oscillator. After down-conversion by time to continuous-time. Then, we downconvert the signal to
the local carrier, the signal is given as [22] baseband using a local oscillator at intermediate frequency fI ,
1
with an output denoted by ej2fI (t+I (t)) , where I (t) is the

L1
N
equivalent time jitter of the oscillator at fI with the 3-dB
y (c) (t) = ej2fc c (t) hl e(c) (t)
Xk ej2kf (tl ) + n
linewidth I . The output signal is impaired by two different
l=0 k=0
phase noises: ej2fI I (t) from the local oscillator at frequency
where c (t) is the timing jitter in the local oscillator with a fI and ej2(fc +N f )u (t) from the clock instability of ADC,
carrier frequency such as 60GHz and n e(c) (t) is the AWGN respectively. The output signal can be described as
noise impaired by timing jitter. { }
After N -point FFT, the sampled y (c) (t) becomes a discrete- N1
L1
time signal, i.e., [19] y(t) = e j2fI I (t) j2(fc +N f )u (t)
e hl Xk ej2kf t

l=0 k=0
Yk
(c) (c) e (c)
= Hk Xk J0 + k + N + e(t) .
n
k

where Then, y(t) is resampled at the interval Ts , and fed into FFT,

N 1
(c) resulting in
k = Hl Xl J(lk)modN (17)

N 1
l=0,l=k
Yk
(s)
= Hk Xk J0 + e
Hl Xl Jlk + N
(s)
4 k
is the ICI component with variance = c2
where c 3 c T S0 , l=0,l=k
is the two-side linewidth of Lorentzian PSD of the PN effect (s) e (s)
ej2fc c (t) , and = Hk Xk J0 + k + N k

N 1 (s)
1 j2fc c [n] j2ni/N where k has the same form as (17) and its variance is s2 =
Ji = e e . 4
N n=0 3 s T S0 , where s = u + I .
Similar to (18), the SINR of the conventional OFDM with
Consequently, the SINR of the k-th subcarrier (c) (k) can undersampling down-conversion is
be written as 1 23 s T
(s)
k = 1 . (20)
2S0 (1 23 c T ) 1 2 c T 2
3 s T +
(c)
k = 4 = 2 3 1 . (18)
3 c T S0 + N0 3 c T +
Consequently, we obtain
Comparing (16) and (18), we can see that, when SNR goes (I)
to infinity, the component 23 c T dominates the denominator, k 1 23 u T 23 s T s I
lim = > =1+ . (21)
yielding k
(s) 2
3 u T 1 3 s T
2 u u
(I)
k 1 23 u T 32 c T c As c and I are much bigger than u , the analytical
lim = > . (19)
(c)
k
2
3 u T 1 3 c T
2 u comparison in (19) and (21) demonstrate self-coherent OFDM
has better performance than conventional OFDMs. This is
2) Performance Comparison of Self-coherent OFDM and the because as long as the transition band of the HPF is large
Conventional OFDM with Undersampling Down-conversion: enough to capture most of the PN, the effect of PN can be
The transmitted signal in RF band is directly down-converted estimated accurately and mostly compensated. In self-coherent
to baseband by undersampling, i.e., [23] OFDM, a fraction of subcarriers are utilized as guard band and
{L1 N1 } a carrier is added to the system. To make a fair comparison,

y[n] = hl Xk e j2(fI +kf )nTu j2(fc +kf )u [n])
e we will adopt PN compensation schemes [20], [19], [22] for
the conventional OFDMs in the following section. Based on
l=0 k=0
the simulation results, we demonstrate that the performance
+ e[n].
n
of self-coherent OFDM with undersampling down-conversion
where u [n] is the same timing jitter as in (12) and n e[n] is still performs better than the conventional OFDM systems,
AWGN noise impaired by PN and sampled at the receiver. As even when phase compensation is applied.
2f u [n] is small, we equate all PN in each sampled data 3) Performance Comparison of Self-coherent OFDM with
to 2(fc + N f )u [n], and y[n] becomes Undersampling Down-conversion and Self-het OFDM: Recall-
{ } (h)
ing the SINR at the k-th subcarrier, denoted by k , of self-het
N1
L1 OFDM in [6]:
y[n] e j2(fc +N f )u [n]
hl Xk e j2(fI +kf )nTu
(h)
l=0 k=0 k = e
(k)
+ e[n].
n 1+ f
Ns

1536-1276 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2016.2594176, IEEE
Transactions on Wireless Communications

Parameter Value
Carrier and subcarrier power ratio () 0.5 frequency selective fading channels. In order to guarantee the
Number of OFDM subcarrier (Ns ) 472 same spectral efficiency, we adopt 4-QAM for conventional
Number of bandgap subcarriers(Ng ) 40 OFDM, a mix of 4-QAM and 16-QAM for self-coherent
Number of upper bandgap subcarriers(Nh ) 4
Number of lower bandgap subcarriers(Nl ) 36 OFDM, and 16-QAM for self-het OFDM. Specifically, given
IFFT/FFT size (N ) 512 that the information bits per symbol is fixed (1024 bits),
Channel length 32
Cyclic prefix length 32 + 1
we apply 4-QAM and 16-QAM for 432 and 40 subcarriers,
Channel band allocation 59.40 61.56 GHz respectively. Furthermore, based on Fig. 7, we observe that the
Carrier frequency (fc ) 59.53 GHz interference power over the first and last few subcarriers are
Intermediate frequency (fI ) 4.710 GHz
undersampling frequency (fu ) 4.569 GHz smaller than that over the remaining middle subcarriers. Hence,
3dB linewidth of oscillator at fc (c ) 100 KHz we assign 4-QAM for the first 216 and last 216 subcarriers and
3dB linewidth of oscillator at intermediate frequency fI (I ) 25 KHz 16-QAM to the remaining 40 middle subcarriers. In Fig. 7,
ADC time jitter (u ) variance (rms) 550 ps
Wireless fading channel model AWGN, Rayleigh we observe that in both channels self-coherent OFDM per-
TABLE II. S IMULATION PARAMETERS .
forms better than self-het OFDM, but worse than conventional
OFDM. In the AWGN case, self-coherent OFDM is 3.5dB
better than self-het OFDM, but 4dB worse than conventional
OFDM. In frequency selective channels, self-coherent OFDM
where N fs denotes the number of self-het OFDM information
performs 5.5dB better than self-het OFDM and 6dB worse than
e
subcarriers and (k) is given in (13) of [6]. Recall the SINR at conventional OFDM. We also observe that the analytical BER
the k-th subcarrier of self-coherent OFDM with undersampling in (11) for self-coherent OFDM agrees with the simulation
(I)
down-conversion, k , in (16). When , self-het results over both channels.
(h)
OFDM has k due to its immunity to PN, while Fig. 8 illustrates the effect of the power ratio between
self-coherent OFDM with undersampling down-conversion has the carrier and information subcarriers on BER performance
(I) 1 23 u T over frequency selective fading channels, for Eb /N0 = 10dB,
k approaching a constant 2 (error floor). Similar
3 uT 20dB, 30dB. We further note that, when = 0, this implies
observations can be made for the BER performance of both that there is no carrier in the system, which coincides with a
systems over Rayleigh fading channels (see Figs. 12 and 14). conventional OFDM system, exhibiting the best performance
over the AWGN channel [32]. However, in our system setting,
IV. S IMULATION R ESULTS
we assume that self-coherent OFDM is adopted and a carrier
In this section, we analyze the performance of self-coherent is always inserted. In such a case, according to [32], the
OFDM over AWGN and multipath fading channels in terms optimal choice of is 0.5. This is also demonstrated in Fig. 8,
of where = 0.5 provides the best BER performance. In such a
1) BER performance comparisons among self-coherent case, we compare the effect of on the self-coherent OFDM
OFDM, self-het OFDM [6], and conventional OFDM performance in this figure, and demonstrate that = 0.5
with a super-heterodyne receiver [11]; provides a good trade-off between BER performance and data
2) impact of power ratio between the carrier and sub- transmission power.
carriers on the BER performance; Fig. 9 compares the effects of ideal LPF and HPF band-
3) impact of ideal LPF and HPF transition bands on the widths on the BER performance. In Fig. 9 (a) and (c), we fix
BER performance; Nh = 32 and increase Nl from 32 to 192 with an increment
4) impact of non-ideal LPF on the BER performance; of 32. In Fig. 9 (b) and (d), we fix Nl = 32 and increase
5) BER performance comparisons among self-coherent Nh from 32 to 192 with an increment of 32. Comparing the
OFDM using the undersampling down-conversion tech- effects of ideal LPF and HPF in AWGN channels (see Fig. 9
nique, conventional OFDM with super-heterodyne re- (a) and (b)), we observe that increasing Nl yields 4.5dB loss,
ceiver and conventional OFDM using the undersam- while increasing Nh yields 2.5dB loss at BER of 102 . This
pling down-conversion technique in both AWGN and demonstrates that the BER performance is much more sensitive
frequency selective channels under various conditions. to LPF rather than HPF. For the frequency selective fading case
We conducted all simulations in time-domain following the in Fig. 9 (c) and (d), we have similar observations.
system model in Fig. 1(a) to validate the equivalent frequency- Fig. 10 shows the impact of non-ideal LPF parameters on
domain models in (4) and (6). All simulation parameters are the BER performance, where a Chebyshev II, infinite impulse
given in Table II. Fig. 6 illustrates the PSDs of each signal response (IIR) filter is used. The top three sub-figures in
component in Fig. 3 for the information subcarriers over Fig. 10 illustrate the magnitude responses (dBs) of such filters
AWGN channel at = 12dB. As shown in Fig. 6, we see that with stop-band attenuations Astop of 30dB, 40dB, and 60dB,
the interference resulting from {2} and {3} are much more respectively. Under these three different LPF settings, the
significant than the interference resulting from {4}. The SINR corresponding BER performance of the system are shown in
in each subcarrier is essentially constant, different from self-het the bottom three sub-figures of Fig. 10. We fix the normalized
OFDM [6], where it degrades at higher frequency subcarriers. cutoff frequency fpass = 0.083 and the normalized stop-band
Fig. 7 compares BER performance among self-coherent frequency fstop = 0.25, which correspond to the location
OFDM, self-het OFDM, and conventional OFDM with super- of carrier and the first information subcarrier in Fig. 1(b),
heterodyne receiver without PN presenting in both AWGN and respectively. Then we increase the stop-band attenuation Astop

1536-1276 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2016.2594176, IEEE
Transactions on Wireless Communications

10

0
20 10
Eb Eb Eb
= 10 dB =20 dB =30 dB
N0 N0 N0

40

1
60 10
Power(dB)

BER
80

{1}Carrier s(t)
100 {2}Carrier ns (t) 2
10
{3}s(t) nc (t)
{4}nc (t) ns (t)
120

3
140 10
100 150 200 250 300 0 0.5 1 1.5 2
kth subcarrier

Fig. 6. Simulated PSDs of signal components in information subcarriers of Fig. 8. Effect of power ratio of the carrier and subcarriers on BER
Eb
self-coherent OFDM baseband version. performance when N is 10dB, 20dB and 30dB.
0
0
10
selfhet OFDM(AWGN)
conventional OFDM(AWGN) LPF(AWGN) HPF(AWGN)
0 0
Ana:selfcoherent OFDM(AWGN) 10 10
Sim:selfcoherent OFDM(AWGN)
selfhet OFDM(Rayleigh)
1
10 conventional OFDM(Rayleigh) 4.5 dB 2.5 dB

BER
BER
Ana:selfcoherent OFDM(Rayleigh) 2 2
Sim:selfcoherent OFDM(Rayleigh) 10 10
BER

2
10 4 4
10 10
5 10 15 5 10 15
Eb/N0 Eb/N0
LPF(Rayleigh) HPF(Rayleigh)
1 1
4 dB 3.5 dB 6 dB 5.5 dB 10 10
3
10

6 dB 2 dB
BER

BER
2 2
10 10
4
10
5 10 15 20 25 30 35 40
E /N
b 0
3 3
10 10
15 20 25 30 15 20 25 30
Eb/N0 Eb/N0
Fig. 7. Comparisons of BER performance among self-coherent OFDM, self-
het OFDM, and conventional OFDM in both AWGN and Rayleigh channels. 32 64 96 128 160 192

Fig. 9. Comparisons of BER for various cutoff frequencies of the ideal LPF
from 30dB to 40dB, to 60dB, and obtain the minimum filter and HPF on self-coherent OFDM over AWGN and frequency selective fading
orders 3, 4, 5, respectively. From the figure, we note that when channels.
Astop = 30dB, the BER performance is severely degraded
after Eb /N0 = 20dB. When Astop = 40dB, the BER
performance improves with a 1dB loss at Eb /N0 = 35dB, we adopt various PN compensation techniques including linear
but increasing for higher Eb /N0 . When Astop = 60dB, the minimum mean square error (LMMSE) [20], ICI compensation
BER performance is similar to that of an ideal LPF. The plan (ICI) [19], the subblocks PN compensation (subblocks)
realization of such filters may be very challenging at the mm- [22]. In [20], a classic LMMSE technique is applied to estimate
wave frequencies. This motivated the use of the undersampling ICI by using the effective data obtained by the common PN
technique and here we show the advantages in the simulation (CPE) compensation plan. In [19], the scheme approximates
results. the current PN waveform by using a Fourier series approxi-
In Figs. 11 and 12, we compare the BERs of self- mation to cancel the effects of the ICI beyond common phase
coherent OFDM using undersampling down-conversion tech- error. In [22], the OFDM symbol is partitioned into subblocks
nique, self-het OFDM, and the conventional OFDM using and the estimate of the time-average of PN at each subblocks
super-heterodyne receiver with and without PN over AWGN is used to compensate ICI effect. For the PN parameters of
and frequency selective fading channels. To make a fair the four systems, as shown in Table. II, c corresponds to the
comparison, for the conventional OFDM, in the simulations, PN of the local oscillator at 60GHz, I the local oscillator

1536-1276 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2016.2594176, IEEE
Transactions on Wireless Communications

11

Order = 3 Order = 4 Order = 5


0 0 0
Bg Bs Bg Bs Bg Bs 1
10
20
Astop 20 20
Magnitude(dB)

Astop
40 40 40
2
10
Astop
60 60 60
fstop

BER
3
80 80 80 10

fB
0 0.5 1 0 0.5 1 0 0.5 1
B
conv. w.o. PN
B conv. w. PN
Normal Frequency conv. w. PN+LMMSE
1 1 1
10 10 10 4 conv. w. PN+ICI
10
conv. w. PN+subblocks

2
Real LPF 2 2
selfco w.o. PN
10 10 10 selfco w. PN
selfhet w. PN
BER

4 6 8 10 12 14 16 18
3
10
3
10
3
10
Eb/N0

Ideal LPF
4 4 4 Fig. 11. Comparisons of BER performance among self-het OFDM, self-
10 10 10
10 20 30 40 10 20 30 40 10 20 30 40 coherent OFDM with undersampling down-conversion technique and conven-
tional OFDM with super-heterodyne receiver over AWGN channels.
E /N
b 0

Fig. 10. Impact of non-ideal LPF parameters on the BER performance. We 1


10
fix fpass = 0.083 and fstop = 0.25, and let Astop = 30dB, 40dB, 60dB,
for filter orders of 3, 4, and 5 respectively.

2
10
BER

af intermediate frequency fI , and the time jitter of ADC u


is mapped to PN according to the linear relationship [17]. In conv. w.o. PN
conv. w. PN
Figs. 11 and 12, we observe that when Eb /N0 is moderate, 3
10 conv. w. PN+LMMSE
the effects of PN dominates the interference and noise items, conv. w. PN+ICI
and thus the impairment of noise can be ignored. We note conv. w. PN+subblocks
selfco w.o. PN
that conventional OFDM with compensation plans incurs an selfco w. PN
error floor, while self-coherent OFDM does not. We also 4 selfhet w. PN
10
observe that self-coherent OFDM with undersampling down- 5 10 15 20 25 30 35 40
Eb/N0
conversion outperforms self-het OFDM by 2dB at BER of
104 in the AWGN channel and by 5dB at BER of 103
in the Rayleigh fading channel when Eb /N0 40dB. When Fig. 12. Comparisons of BER performance among self-het OFDM, self-
Eb /N0 > 40dB, it exhibits an error floor, while self-het OFDM coherent OFDM with undersampling down-conversion technique and con-
ventional OFDM with super-heterodyne receiver over frequency-selective
does not due to immunity to PN. channels.

From Figs. 12 and 14, when we compare self-coherent


OFDM in frequency selective channels, we see that the BER V. C ONCLUSIONS
incurs an error floor at very high Eb /N0 (> 40dB). One
major reason of this error floor is that some subcarriers may We have adapted the self-coherent OFDM technique, orig-
experience the joint effect of deep fades and the receiver PN. inally used in optical communications, to wireless communi-
The other reason is evident from (16): we can see that the self- cation systems. We have analysed its performance in terms of
coherent OFDM system experiences a doubly faded channel, BER over AWGN and frequency selective fading channels. We
where one fade is due to multipath and the other is incurred have found that self-coherent OFDM achieves better spectral
by the RF carrier in the link between TX/RX antennas. efficiency and BER performance than self-het OFDM.
Since filter realization can be very challenging in high
In Figs. 13 and 14, we compare the BERs of self-coherent frequency mm-wave RF bands, we have introduced undersam-
OFDM and the convectional OFDM, both of which use under- pling down-conversion to self-coherent OFDM. We have com-
sampling down-conversion, as well as self-het OFDM. Even pared analytically its system performance with self-het OFDM,
when the PN compensation plans [19], [20], [22] are applied and two conventional OFDMs: one with super-heterodyne re-
to the conventional OFDM, we obtain the similar observations ceiver and the other with undersampling down-conversion. We
to that of Figs. 11 and 12. have shown theoretically and by simulation that self-coherent

1536-1276 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2016.2594176, IEEE
Transactions on Wireless Communications

12

gains at the k-th subcarrier and the carrier (i.e., k and ). We


recall that (k) in k (see (7)) involves various subchannel
1
10 coefficients m , where m = k. Hence, here we make an
approximation by replacing (k) with the expected value
E((k)), which is a constant that depends only on k. This
2
10 approximation removes the correlation between the effective
noise in the subcarriers. Moreover, in the Rayleigh channel, k
and follow an exponential distribution. Hence, by averaging
BER

3
10 Pk (e|k , ) over k yields
conv. w.o. PN
conv. w. PN
conv. w. PN+LMMSE Pk (e|) = Pk (e|k , )ek dk
( 0 )
4
10 conv. w. PN+ICI
conv. w. PN+subblocks 1
selfco w.o. PN 1
selfco w. PN
( M )
selfhet w. PN
4 6 8 10 12 14 16 18 3
Eb/N0 1
2(M 1)( + E((k)) 2Nl
Ns + Ns ) + 3
Fig. 13. Comparisons of BER performance among self-het OFDM, self-
coherent OFDM and conventional OFDM both with undersampling down-
Then the average symbol error rate of the k-th OFDM
conversion technique over AWGN channels. subcarrier can be obtained as

Pk (e) = Pk (e|)e d
1
10 0[ ( )]
1 2(M 1) E((k)) 2Nl
U , 0, +
2 3 + 2(M 1) Ns Ns
( )( )
1 1 3
2
10 1 1
M 2 3 + 2(M 1)
BER

where U(a, b, z) is the confluent hypergeometric function of


conv. w.o. PN
3 conv. w. PN
the second kind and has an integral representation given by
10
conv. w. PN+LMMSE [14]
conv. w. PN+ICI

conv. w. PN+subblocks 1
selfco w.o. PN U(a, b, z) = ezt ta1 (1 + t)ba1 dt.
selfco w. PN (a) 0
4 selfhet w. PN
10
5 10 15 20
EbN0
25 30 35 40
where () is the gamma function. For 1 and Gray-
mapping, the average BER of the k-th OFDM subcarrier can
be approximated as
Fig. 14. Comparisons of BER performance among self-het OFDM, self-
coherent OFDM and conventional OFDM both with undersampling down- BERk 2Pk (e)/ log2 (M ) (22)
conversion technique over frequency-selective channels. [ ( )]
1 2(M 1) E((k)) 2Nl
U , 0, +
2 3 + 2(M 1) Ns Ns
OFDM with undersampling down-conversion outperforms self- ( )( )
het OFDM in a wide range of SNRs (up to 40dB). It also 2 1 1 3
1 1 .
offers better performance and higher robustness to PN than log2 (M ) M 2 3 + 2(M 1)
conventional OFDM with ICI compensation.
A PPENDIX A
A NALYTICAL BER OF S ELF - COHERENT OFDM IN R EFERENCES
R AYLEIGH FADING C HANNELS
[1] A.G. Armada and M. Calvo, Phase noise and sub-carrier spacing
Since M -QAM signalling is applied, the symbol error rate effects on the performance of an OFDM communication system, IEEE
of the k-th OFDM subcarrier is given by [28] Communications Letter, vol. 2, no. 1, Jan. 1998, pp. 11-13.
( ) ( )
[2] R.C. Daniels and R.W. Heath, 60 GHz Wireless Communications:
1 3k
Pk (e|k , ) = 2 1 Q Emerging Requirements and Design Recommendations, IEEE Vehicular
M M 1 Technology Magazine, vol. 2, no. 3, pp. 41-50, Sept. 2007.
[3] C. Cao and K.K. O, Millimeter-wave voltage-controlled oscillators
where k is the instantaneous SINR of the k-th OFDM in 0.13-m CMOS technology, IEEE Journal of Solid-State Circuits,
subcarrier in (7), which depends on the equivalent channel vol. 41, no. 6, pp. 1297-1304, June 2006.

1536-1276 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TWC.2016.2594176, IEEE
Transactions on Wireless Communications

13

[4] Y. Shoji, M. Nagatsuka, K. Hamaguchi, and H. Ogawa, 60 GHz band [23] V. Syrjala, and M. Valkama, Jitter mitigation in high-frequency
64 QAM/OFDM terrestrial digital broadcasting signal transmission by bandpass-sampling OFDM radios, Wireless Communications and Net-
using millimeterwave self-heterodyne system, IEEE Transactions on working Conference, pp. 1-6, Budapest, April 2009.
Broadcasting, vol. 47, no. 3, pp. 218-227, Sept. 2001. [24] IEEE Standard for Information technologyTelecommunications and
[5] R.A. Pacheco and D. Hatzinakos, BER analysis of self-heterodyne information exchange between systemsLocal and metropolitan area
OFDM transmission scheme. IEEE Canadian Conference on Electrical networksSpecific requirements-Part 11: Wireless LAN Medium Access
and Computer Engineering, vol. 4, pp. 1953-1956, May 2004. Control (MAC) and Physical Layer (PHY) Specifications Amendment 3:
[6] N. Fernando, Y. Hong, and E. Viterbo, Self-heterodyne OFDM trans- Enhancements for Very High Throughput in the 60 GHz Band, IEEE
mission for frequency selective channels, IEEE Transactions on Com- Std 802.11ad-2012 (Amendment to IEEE Std 802.11-2012, as amended
munications, vol. 61, no. 5, pp. 1936-1946, May 2013. by IEEE Std 802.11ae-2012 and IEEE Std 802.11aa-2012) , vol., no.,
pp.1,628, Dec. 28 2012 doi: 10.1109/IEEESTD.2012.6392842.
[7] N. Fernando, Y. Hong, and E. Viterbo, Subcarrier pairing for self-
heterodyne OFDM, 2013 IEEE International Conference on Commu- [25] A. Stephens, IEEE 802.11 TGn Comparison Criteria, doc.: IEEE
nications (ICC), pp. 3105-3109, Budapest, Hungary, June 2013. 802.11-03/814r31.
[8] T. Miyazaki and F. Kubota, PSK self-homodyne detection using a pilot [26] R.G. Vaughan, N.L. Scott, and D.R. White, The theory of bandpass
carrier for multibit/symbol transmission with inverse-RZ signal, IEEE sampling, IEEE Transactions on Signal Processing, vol. 39, no. 9,
Photonics Technology Letters, June 2005, vol. 17, no. 6, pp. 1334-1336. pp. 1973-1984, Sep. 1991.
[27] K. Nikitopoulos and A. Polydoros, Compensation schemes for phase
[9] S, Adhikari, S.L. Jansen, M. Alfiad, B. Inan, V.A.J.M. Sleiffer, A. Lo-
noise and residual frequency offset in OFDM systems, Global Telecom-
bato, P. Leoni, and W. Rosenkranz, Self-coherent optical OFDM:
munications Conference, San Antonio, TX, Nov. 2001, pp. 331-333.
an interesting alternative to direct or coherent detection, 13th IEEE
International Conference on Transparent Optical Networks (ICTON), [28] D. Tse and P. Viswanath. Fundamentals of wireless communication,
pp. 1-4, Stockholm, Sweden, June 2011. Cambridge university press, 2005.
[10] Q. Jin, Y. Hong and E. Viterbo, Self-coherent OFDM for wireless com- [29] S. Pinel, R. Bairavasubramanian, J. Laskar and J. Papapolymerou,
munications, 2015 IEEE International Conference on Communications Compact planar and vialess composite low-pass filters using folded
(ICC), pp. 4303-4308, London, June 2015. stepped-impedance resonator on liquid-crystal-polymer substrate, IEEE
Transactions on Microwave Theory and Techniques, vol. 53, no. 5, pp.
[11] A.A. Abidi, Direct-conversion radio transceivers for digital commu-
1707-1712, May 2005.
nications, IEEE J. Solid-state circuits, vol. 30, no. 12, pp. 1399-1410,
Dec. 1995. [30] M. Ho, W. Hong, and L. Lin. Low-pass filter of suspended stripline
design with finite transmission zeros for stop-band rejection improve-
[12] Y. Hong, E. Viterbo, and A.J. Lowery, Improving the sensitivity ment, Microwave and Optical Technology Letters, vol. 56, no. 2, pp.
of direct-detection optical OFDM systems by pairing of the optical 297-301, Feb. 2014.
subcarriers, 2011 European Conference and Exposition on Optical
Communications, pp. 1-3, Geneva, Switzerland, Sept. 2011. [31] W.B. Davenport and W.L. Root, An introduction to the theory of random
signals and noise, McGraw-Hill, 1st Edition, 1958.
[13] Y. Hong, A.J. Lowery, and E. Viterbo, Sensitivity improvement and
carrier power reduction in direct-detection optical OFDM systems by [32] J. Hancock and P. Wintz, Signal detection theory, McGraw-Hill, 1966.
subcarrier pairing, Optics express, vol. 20, no. 2, pp. 1635-1648, Jan.
2012.
[14] G. Kristensson, Second Order Differential Equations: Special Functions
and Their Classification, 1st ed. Springer New York, 2010.
[15] S. Wu and Y. Bar-Ness, OFDM systems in the presence of phase noise:
consequences and solutions, IEEE Transactions on Communications,
vol. 52, no. 11, pp. 1988-1996, Nov. 2004.
[16] A. Demir, A. Mehrotra, and J. Roychowdhury. Phase noise in oscil-
lators: a unifying theory and numerical methods for characterization,
IEEE Transactions on Circuits and Systems I: Fundamental Theory and
Applications, vol. 47, no. 5, pp. 655-674, May. 2000.
[17] A. Demir, Computing timing jitter from phase noise spectral for
oscillators and phase-locked loops with white and 1/f noise, IEEE
Transactions on Circuits and Systems I: Regular Papers, vol. 53, no. 9,
pp. 1869-1884, Sept. 2006.
[18] L. Pan, Y. Bar-Ness, and J. Zhu. Effects of phase noise at both
transmitter and receiver on the performance of OFDM systems, IEEE
40th Annual Conference on Information Sciences and Systems, pp. 312-
316, Princeton, New Jersey, March. 2006.
[19] D. Petrovic, W. Rave, and G. Fettweis. Effects of phase noise on
OFDM systems with and without PLL: Characterization and compensa-
tion, IEEE Transactions on Communications, vol. 55, no. 8, pp. 1607-
1616, Aug. 2007.
[20] S. Wu, P. Liu and Y. Bar-Ness. Phase noise estimation and mitigation
for OFDM systems, IEEE Transactions on Wireless Communications,
vol. 5, no. 12, pp. 3616-3625, Dec. 2006.
[21] S. Wu and Y. Bar-Ness. A phase noise suppression algorithm for
OFDM-based WLANs, IEEE Communications Letters, vol. 6, no. 12,
pp. 535-537, Dec. 2002.
[22] M. Lee, K. Yang, and K. Cheun. Iterative receivers based on subblock
processing for phase noise compensation in OFDM systems, IEEE
Transactions on Communications, vol. 59, no. 3, pp. 792-802, Jan. 2011.

1536-1276 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

You might also like