3946 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO.

9, SEPTEMBER 2016

Orthogonal Chirp Division Multiplexing

Xing Ouyang and Jian Zhao

Abstract— Chirp waveform plays a significant role in radar

and communication systems for its ability of pulse compression
and spread spectrum. This paper presents a principle of multi-
plexing a bank of orthogonal chirps, termed orthogonal chirp-
division multiplexing (OCDM) for high-speed communication.
As Fourier transform is the kernel of orthogonal frequency-
division multiplexing (OFDM), which achieves the maximum
spectral efficiency (SE) of frequency-division multiplexing,
Fresnel transform underlies the proposed OCDM system, which
achieves the maximum SE of chirp spread spectrum. By using
discrete Fresnel transform, digital implementation of OCDM is
introduced. According to the properties of Fresnel transform, the
transmission of OCDM signal in linear time-invariant channel is
studied. Efficient digital signal processing is proposed for channel
dispersion compensation. The implementation of the OCDM
system is discussed with the emphasis on its compatibility to
the OFDM system; it is shown that it can be easily integrated
into the existing OFDM systems. Finally, the simulations are
Fig. 1. Illustration of chirps: (a) Near-field Kirchhoff-Fresnel diffraction of a
provided to validate the feasibility of the proposed OCDM. It is
circular aperture, and (b) its diffraction pattern on the second plate; (c) linear
shown that the OCDM system can efficiently exploit multipath chirp waveform which is ‘up-chirp’ from 0 to 1 second and is ‘down-chirp’
diversity and thus outperforms the OFDM, and that it is more from 1 to 2 second, and (d) its spectrogram.
resilient against the interference due to insufficient guard interval
than single-carrier frequency-domain equalization.
The chirp signals use a wideband spectrum and are thus
Index Terms— Fresnel transform, discrete Fresnel transform, resilient to detrimental effects, such as channel noise, mul-
chirp spread spectrum, orthogonal chirp division multiplexing. tipath fading and Doppler effects within the wireless mobile
channel.
I. I NTRODUCTION
For the purpose of communication, chirped waveforms

C HIRP relates to a signal whose phase evolves along

with time in a certain manner, and it is found in many
applications. For instance, the spatial frequency of near-field
are of spread-spectrum technique, i.e., chirp spread spec-
trum (CSS), providing secure and robust communication for
applications, such as, military, underwater and aerospace sce-
Fresnel diffraction pattern increases linearly along with the narios [14]–[16]. In the recent ultra-wideband (UWB) standard
distance to the origin of the screen, in Fig. 1 (a) and (b) [1]–[4]. by the IEEE 802.15.4a group [17], CSS is adopted to meet
In the radar systems, a chirp is a frequency modulated signal the requirement of FCC on the radiation power spectral mask
sweeping over a wideband spectrum with constant ampli- for the unlicensed UWB systems. Meanwhile, it achieves
tude. By correlating the echoes bouncing from target, pulse what an UWB is supposed to do, ranging, measuring and
compression is achieved as if ‘short’ pulses are emitted and communicating [18]–[20].
received. The position of the target can be resolved from Conventionally, a chirp signal is usually generated by analog
the temporally delayed pulses [5]–[7]. In spread-spectrum devices in filtering approach by surface acoustic wave (SAW)
systems, binary information is encoded by modulating the fre- devices [21]–[25] or in frequency-modulation approach by
quency of a carrier, sweeping the frequency linearly from low voltage controlled oscillator, by taking the advantage of the
to high as ‘up-chirp’ for bit ‘1’ or from high to low as ‘down- mature CMOS technology [26]–[28]. In the chirp spread-
chirp’ for bit ‘0’, as shown in Fig. 1 (c) and (d) [8]–[13]. spectrum (CSS) or frequency-modulated systems, a broad
Manuscript received March 28, 2016; revised July 20, 2016; accepted spectrum is occupied for modulating information. Spectral
July 21, 2016. Date of publication July 27, 2016; date of current version efficiency is sacrificed for high processing gain, the ability of
September 14, 2016. This work was supported by the Science Foundation multipath resolution, and other merits of the chirp signal. In a
Ireland under Grant 11/SIRG/I2124 and 15/CDA/3652. The associate editor
coordinating the review of this paper and approving it for publication given period T and bandwidth B, there is only one modulated
was R. Dinis. chirp, as shown in Fig. 2 (a). If there are more than one chirp
X. Ouyang is with the Tyndall National Institute, Cork T12R5CP, Ireland, signals in the same period and bandwidth, interference occurs.
and also with the School of Electrical and Electronic Engineering, University
College Cork, Cork T12 YN60, Ireland (e-mail: [email protected]). Thus, chirp is attractive for low data rate applications in which
J. Zhao is with the Tyndall National Institute, Cork T12R5CP, Ireland (e- reliability is in priority.
mail: [email protected]). To increase the data rate of a UWB system, multi-code
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. UWB is proposed by dividing the entire bandwidth into a bank
Digital Object Identifier 10.1109/TCOMM.2016.2594792 of spectrally separated chirps, see Fig. 2 (a) [18]. The chirps
0090-6778 © 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.

Authorized licensed use limited to: University of Surrey. Downloaded on March 29,2023 at 08:22:38 UTC from IEEE Xplore. Restrictions apply.
OUYANG AND ZHAO: ORTHOGONAL CHIRP DIVISION MULTIPLEXING 3947

exhibit better tolerance against the interference due to insuffi-

cient length of guard interval (GI) than the single-carrier (SC)
FDE system.
This paper is organized as follows. In Section II, the Fresnel
transform and its discrete form, DFnT are introduced for
supporting the OCDM. In Section III, the proposed OCDM
system is presented. In Section IV, the mathematical model
of OCDM system under LTI channel is studied. In addition,
the proposed efficient equalization algorithm is presented. Dis-
cussions about the implementation of the OCDM system and
its compatibility to OFDM system are provided in Section V.
In Section VI, simulations are performed to evaluate the
Fig. 2. Illustration of (a) the multi-code chirp waveform and (b) the digital
implemented OCDM signal in the temporal-frequency-chirp dimension.
performance of the OCDM system under wireless multipath
fading channels, and Section VII finally concludes this paper.
In this paper, we consider the system in baseband rep-
are modulated in binary codes, such as, the Walsh-Hadamard resentation [Chapter 2, 29]. Matrices are upright bold in
code. The receiver retrieves the information from the chirps uppercase while vectors are italic bold in lowercase. Some
of different subchannels over the entire bandwidth. basic mathematical notations are given here.
Heuristically, in this paper we come to think if there is a way √
j −1
of synthesizing a bank of chirps in the same period and band- ∗ linear convolution operator
width without any interference, and if both their amplitude and
 circular convolution operator
phase can be used for information modulation. If yes, can we
H (m, n) (m, n)-th entry of matrix H
implement such system by using the advanced digital signal
h(n) n-th entry of vector h
processing (DSP) technology in the digital domain rather than
(·)∗ complex conjugate operator
the analog domain?
(·)T matrix transpose operator
In this paper, we introduce the answer, i.e., orthogonal chirp
(·) H complex conjugate transpose operator
division multiplexing (OCDM), which is actually the principle
δ(t)/δ(n) Dirac delta/Kronecker delta function
of orthogonally multiplexing a number of chirped waveforms
of the same bandwidth, towards the maximum communication
rate. As shown in Fig. 2 (b), in the OCDM, a bunch of II. F RESNEL T RANSFORM
chirp waveforms overlap temporally and spectrally. The phase In this section, we will introduce the Fresnel transform and
and/or the amplitude of the chirps can be modulated in, for DFnT as the bases for the OCDM system. The linear (circular)
example, phase shift keying (PSK), pulse amplitude modu- convolution property of the Fresnel transform (DFnT) is also
lation (PAM), or quadrature amplitude modulation (QAM). studied for studying the OCDM signal under LTI channels.
In this paper, it is proved that the modulated chirps are
orthogonal along the chirp dimension, without interference to A. Fresnel Transform
each other for transmission. Fresnel transform is an integral transformation originating
We show that the fundamental underlying the OCDM from classical optics [1]–[4]. It is the formula that mathemat-
system is Fresnel transform, just as the Fourier trans- ically describes the near-field optical diffraction. In Fig. 1 (a),
form in orthogonal frequency division multiplexing (OFDM). when a monochromatic plain wave with wavelength λ encoun-
In this paper, digital implementation of the OCDM system ters a slit (grating) whose scale is comparable in size to λ, the
using discrete Fresnel transform (DFnT) is presented. Specif- resulting diffraction pattern on a plate at distance z is given by
ically, the inverse DFnT (IDFnT) generates the OCDM signal π 
e− j 4 π
s (t) e j a (τ −t) dt, (1)
2
at the transmitter, and the DFnT recovers the OCDM signal ŝ (τ ) = F a {s (t)} (τ ) = √
at the receiver. a
We then analyze the transmission of the OCDM signal under where F a {}(·) denotes the Fresnel transform of parameter
linear time-invariant (LTI) or quasi-static channels which can a = λz being the normalized Talbot distance, and s(t) is the
generalize most of practical communication channels, such as, complex transmittance of the grating. The kernel of Fresnel
coaxial cables, optical fibers and mobile radio channels etc. transform is
Based on the convolution property of the Fresnel transform, π π 2
it is shown that the channel distortion can be compensated by ϕa (t) = e− j 4 e j a t . (2)
using either time-domain or frequency-domain equalizer. In 1994, Gori visited the Fresnel transform in his chapter
Moreover, based on the eigen-decomposition for DFnT, an “Why Fresnel is so little known?”. The important but almost
efficient channel equalization algorithm is further proposed for ‘neglected’ property [Theory (1), 30], the Fresnel transform
the OCDM based on single-tap frequency domain equaliza- of a linear convolution is
tion (FDE), which is more efficient than the aforementioned
two equalization schemes. Simulation results show that the r̂ (τ ) = F a {h (t) ∗ s (t)} (τ )
OCDM outperforms the conventional OFDM, and it also = ĥ (τ ) ∗ s (τ ) = h (τ ) ∗ ŝ (τ ) . (3)

Authorized licensed use limited to: University of Surrey. Downloaded on March 29,2023 at 08:22:38 UTC from IEEE Xplore. Restrictions apply.
3948 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 9, SEPTEMBER 2016

where ŝ and ĥ are the DFnTs of s and h, respectively. It can

be observed that the DFnT of a circular convolution in Eq. (7)
is the discrete analogy of Eq. (3).
The DFnT is a trigonometric transform of chirped phases
and has an intimate relation to the chirp Z -transform (CZT)
[40]–[42] and the Zadoff-Chu sequence [43], which are well
known in the signal processing societies. It is shown that the
CZT is a special case of DFnT, and that the root Zadoff-Chu
sequence is the eigen-values of DFnT [39].
Fig. 3. Illustration of Talbot effects; the DFnT matrix of size N represents III. O RTHOGONAL C HIRP D IVISION M ULTIPLEXING
the optical field of the light spots at the fraction of Talbot distance Z T /N .
In this section, following a brief review of the CSS systems,
we introduce that the Fresnel transform provides a theo-
Eq. (3) states that the Fresnel transform of a convolution retical framework of OCDM. Digital implementation of the
equals either one convolving with the Fresnel transform of OCDM system based on DFnT is also presented.
the other. It is different from the convolution theorem of the A. Chirp Spread Spectrum
Fourier transform which states that the Fourier transform of a
Most applications consider the frequency modulated (chirp)
convolution equals the product of the Fourier transforms.
signal whose frequency evolves linearly or equivalently whose
phase changes quadratically over time,
B. Discrete Fresnel Transform 
παt 2 +ϕ0
ψ (t) = e j , (8)
The discrete form of Fresnel transform, DFnT relates to
the Talbot effect which is the periodic grating of Fresnel where α is the chirp rate and ϕ0 is an initial phase.
diffraction [1]–[4], [31]–[35], see Fig. 3. The DFnT matrix Its instantaneous frequency is
gives the field coefficients of the Talbot image or called self- 1 d

image at the fraction of Talbot distance, z = Z T /N, where f (t) = παt 2 + ϕ0 = αt. (9)
2π dt
Z T = d 2 /λ. (4) If the chirp is temporally limited in some period T , the
bandwidth of the chirp B is determined by its chirp rate α
is the Talbot distance and d is the distance of repeated gratings. and periodT , and B × T indicates the processing gain of a
In previous works, the DFnT is formulated for describing chirp signal.
the coefficients of the Talbot image. Degeneracy exists: the In the CSS system, the time-bandwidth product BT  1,
size of DFnT matrices is N/2 if N ≡ 0 and 2 (mod 4), while and thus B  Rs , where Rs = 1/T is the symbol rate.
the size is N if N ≡ 1 and 3 (mod 4) [31], [33], [36]–[38]. The It means that the larger the processing gain is, the less the
degeneracy hinders its applications as a general mathematical spectral efficiency becomes. Moreover, in the conventional
tool. CSS system, the chirp signal is modulated by analog devices.
In the recent work, the DFnT is derived in [39] without The advanced modulation formats, such as QAM, cannot be
such degeneracy. The (m, n)-th entry of the N by N DFnT implemented to enhance spectral efficiency. In the following
matrix  is defined as subsection, we present the principle of OCDM to maximize
⎧ π
1 −j π ⎨ e N 
j (m−n)2 the spectral efficiency of CSS.
N ≡ 0 (mod2)
 (m, n) = √ e 4 × 2
π
N ⎩ e j N m+ 2 −n
1
N ≡ 1 (mod2).
B. Principle of Orthogonal Chirp Division Multiplexing
To introduce the Fresnel transform in the optics for OCDM,
(5)
some constraints are raised. Firstly, the chirped waveforms for
The DFnT matrix is unitary, and its other important properties, modulation is time-limited. Secondly, the spatial Talbot effect
such as its eigen-decomposition, can be found in [39]. It should in Eq. (1) is adapted into the temporal counterpart for OCDM.
be noted that the DFnT of even and odd N is slightly different. Based on Eq. (4), we define the temporal Talbot distance Z T as
The DFnT possesses the circular convolution property T2
which says that the DFnT of the circular convolution of two ZT = (10)
λ
sequences equals either one convolving with the DFnT of
where T is the period of the chirped waveforms. In Eq. (4), the
the other. Given two length-N vectors h and s and two N
temporal length T replaces the periodic distance d. Supposing
by N circulant matrices H and S whose first columns are
that there are N chirped waveforms, one can obtain the chirps
h = [h(0), h(1), . . . , h(N − 1)]T and s = [s(0), s(1), . . . ,
by substituting the fraction of the Talbot distance z = Z T /N
s(N − 1)]T , the circular convolution r (n) = h(n)  s(n) in
into Eq. (2), and thus one can obtain the temporal Talbot
matrix form is
period a = T 2 /N. Substituting the variable a in Eq. (2) by
r = Hs = Sh. (6) a = T 2 /N, the ‘root’ chirp is

The DFnT of the circular convolution is ψ0 (t) =
T (t) ϕ ∗ (t) T 2
a a= N
j π4 − j π T 2 t
N 2
r̂ = r = Hŝ = Sĥ. (7) =e e , 0≤t <T (11)

Authorized licensed use limited to: University of Surrey. Downloaded on March 29,2023 at 08:22:38 UTC from IEEE Xplore. Restrictions apply.
OUYANG AND ZHAO: ORTHOGONAL CHIRP DIVISION MULTIPLEXING 3949

Fig. 5. Schematic diagram of the OCDM transceiver (a) multiplexing and

(b) demultiplexing a bank of N modulated orthogonal chirp waveforms.

Fig. 4. Illustration of a family of 16 orthogonal waveforms in OFDM and

OCDM. The in-phase (solid) and the quadrature (dots) components of the
linear (a) exponential waveforms in OFDM, and (b) chirped waveforms in
OCDM.

where

1 0≤t <T

T (t) = (12)
0 otherwise,
is the rectangular function.
It can be seen from Eq. (11) that the chirp rate α is
equivalent to the reciprocal of the temporal Talbot period a,
1 N
α = = 2. (13)
a T Fig. 6. Illustrations of the spectra of (a) analog and (b) digital implementa-
One can get a set of N chirps by using the root chirp in tions for OCDM. The lines denote the instantaneous frequency of each chirp.
Eq. (11), and the k-th, k = 0, 1, . . . , N − 1, chirp is
  According to Eq. (15), x(m) can be extracted by the matched
∗ T
ψk (t) =
T (t) ϕ T 2 t − k filter to the m-th chirp as shown in Fig. 5 (b), as
N N  T
 2
j π4 − j π T 2 t −k N
N T x  (m) = s (t) ψm∗ (t) dt
=e e , 0 ≤ t < T. (14) 0
It can be readily proved that the chirp waveforms ψk (t) in 
N−1

Eq. (14) are mutually orthogonal, = x (k) δ (m − k) = x (m) . (17)

 k=0
ψm∗ (t) ψk (t)dt In Fig. 6, the spectra of the chirps in the OCDM is
 T  2  2 illustrated. The analog OCDM signal occupies a bandwidth
j π N t −m N
T
− j π N2 t −k N
T
from −B to B, where B = N/T . In the following subsec-
= e T2 e T = δ (m − k) . (15)
0 tion, we will present the digital implementation of OCDM
Eq. (14) formulates a set of N orthogonal chirped wave- signal based on DFnT, and it will be shown that, according
forms. To show an illustrative comparison between the OFDM to the sampling theory, the digital OCDM generates the
and OCDM, Fig. 4 provides the linear exponential waveforms OCDM signal of wrapped spectra, rather than the interpolated
in OFDM that are mutually orthogonal in frequency [44]–[47], OCDM signal in Eq. (16). The spectrum of the digital OCDM
and the quadratic exponential waveforms in OCDM that are signal can be well constrained within the bandwidth from
mutually orthogonal in the dimension of chirp. −0.5B to 0.5B. In addition, the out-of-band aliasing signal in
In the OCDM system, the amplitude and phase of each chirp the chirp based systems, such as CSS, can be well controlled
can be used for modulation. Thus, PAM, PSK, and QAM can by digital filters.
be employed. Depending on the modulation formats, symbols
are chosen from a codebook χ to encode the information bits. C. DFnT Based Digital Implementation for OCDM
Similar to the OFDM symbols which are transmitted block Eqs. (14)-(17) show the concept of OCDM, which can be
by block, the modulated chirps are also transmitted in block. theoretically realized by using analogue electronic or photonic
In an OCDM block, the k-th symbol modulating the k-th chirp devices, such as, surface acoustic wave devices or fiber Bragg
is x(k) ∈ χ. The synthesized and modulated chirp signal in gratings. Here, we present the efficient digital implementation
Fig. 5 (a), is of OCDM by using DFnT. Rather than the interpolated analog

N−1 signal in Eq. (16), the discrete-time OCDM signal is equivalent
s (t) = x (k) ψk (t) 0 ≤ t < T. (16) to the continuous-time OCDM signal in Eq. (16) with wrapped
k=0 spectrum, which will be discussed in Section III-D later.

Authorized licensed use limited to: University of Surrey. Downloaded on March 29,2023 at 08:22:38 UTC from IEEE Xplore. Restrictions apply.
3950 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 9, SEPTEMBER 2016

It should be noted that there are two forms of DFnT matrix

in Eq. (5). Thus, the discrete OCDM signal is,

N−1  
T
s (n) = s (t)|t =n T = x (k) ψk n
N N
k=0

π 
N−1
π
x (k) e− j N (n−k) ,
2
= ej 4 (18)
k=0
for even N, or

N−1  
  T T
s (n) = s (t)| = x (k) ψk n +
t = n+ 12 NT
N 2N
k=0
 2
π 
N−1 π
−j N n−k+ 12
= ej 4 x (k) e , (19)
k=0
for odd N. Inspecting Eq. (18) and (19), one can compare them
with the definition of DFnT in Eq. (5), and find that they are
exactly the IDFnT. Thus, the synthesis of a bank of discretized
modulated chirp waveforms can be realized by the IDFnT. Fig. 7. Schematic diagram of the proposed digital implementation of OCDM.
To express Eq. (18) and (19) in a concise matrix form, we Insets: Illustrations of (a) time-domain and (b) frequency domain equalizer.
stack the symbols in the vector form as x = [x(0), x(1), . . . ,
the difference, Fig. 6 provides the spectrograms of the
x(N −1)]T , and thus the discrete time-domain OCDM signal is
(a) continuous-time and the (b) discrete-time OCDM signals.
s =  H x. (20) It can be observed that in Fig. 6 (a), although each chip signal
ψk (t) has a bandwidth of B, the continuous-time OCDM
Since the DFnT matrix is unitary, at the receiver, the trans- signal in Eq. (16) occupies a bandwidth from −B to B.
mitted symbols can be recovered by performing the inverse It seems that one cannot meet the Nyquist rate, unless a
operation, i.e., DFnT. The recovered symbols are thus sampling rate no less than 2N/T is chosen to generate the
x = s = x. (21) OCDM signal in Eq. (16).
Fortunately, as shown in Fig. 6 (a), each chirp signal
ψk (t) still has a spectral bandwidth of B. In principle, the
D. Differences Between the Continuous and Discrete OCDM
spectrum of a discrete-time signal produces its continuous-
There are differences between the continuous- and discrete- time counterpart with cyclically extended spectrum. In the
time OCDM signals. Firstly, although the sampled chirp in OCDM system, if the sampling rate is exactly at B = N/T Hz,
Eqs. (18) or (19) is from ψk (t) in Eq. (14) by uniform referring to Eq. (18), the spectrum of the discrete-time OCDM
sampling, the interpolated continuous-time chirp from Eq. (18) signal has a period of B, as shown in Fig. 6 (b). All the
or (19) is the periodic extension of ψk (t). Specifically, we have spectra of ψk (t) are folded into the baseband from −B/2 and
the root chirp B/2. As a result, the discrete-time OCDM signal generates
⎧   the OCDM signal of folded spectrum, completely preserving
⎪ T T
⎪ ϕ ∗
⎨ T 2 /N (t) t ∈ − , the information.
ψ0 (t) =  2 2 
⎪
⎪ ∗ T T IV. OCDM S IGNAL U NDER LTI S YSTEMS
⎩ ϕT 2 /N (t − cT ) t ∈ − + cT, + cT ,
2 2 In this section, we formulate the mathematical model of the
(22) OCDM signal under LTI channel in matrix form. The channel
is static or quasi-static, which means that the channel response
as the periodic extension of ψ0 (t) in Eq. (11). The interpolated
remains constant within one OCDM block, and might change
chirped waveform becomes
  in the next block. It is assumed that the channel information
T is available at the receiver via a certain channel estimation
ψk (t) =
T (t) ψ0 t − k , (23)
N method with perfect synchronization. To avoid inter-symbol
Moreover, using the periodic property, we can easily prove interference (ISI), GI is inserted between adjacent blocks, just
that the following relation in Eq. (18) holds as the OFDM system does. Based on the analysis, we will
show that the GI can be filled with either zeros, i.e., zero-

N−1   N−1  
T  T padded prefix (ZP), or the replica of a portion of the end of
s (n) = x (k) ψk n = x (k) ψk n , (24) the signal, i.e., cyclic prefix (CP).
N N
k=0 k=0
for even N, or with a sampling offset 2T /N for odd N. A. Signal Transmission Under LTI Gaussian Channels
Secondly, the spectrum of the discrete-time OCDM signal is In Fig. 7, the signal transmission is illustrated. Suppose that
the periodic extension of the continuous-time one. To illustrate the maximum channel delay is smaller than the length of GI.

Authorized licensed use limited to: University of Surrey. Downloaded on March 29,2023 at 08:22:38 UTC from IEEE Xplore. Restrictions apply.
OUYANG AND ZHAO: ORTHOGONAL CHIRP DIVISION MULTIPLEXING 3951

equalization is employed, and the IDFnT at the receiver is

avoided based on its eigen-decomposition property.
At the receiver, the sampled signal is given in Eq. (25).
Instead of DFnT in Fig. 7, it is firstly transformed into the
frequency domain by DFT as
y = Fr = FH H x + Fn, (28)
where F is the normalized DFT matrix. Using the identity
I = F H F, Eq. (28) can be further given by
y = FHF H F H F H Fx + w
=  H Fx + w, (29)
Fig. 8. Proposed single-tap FDE for the OCDM system.
where  = FHF is the channel frequency response (CFR)
H

matrix and  H = F H F H is a coefficient matrix.

If CP is used, the received signal is According to the eigen-decomposition property of a circu-
lant matrix with respect to DFT, the CFR matrix  is diagonal.
r = Hs + n = H H x + n (25) The k-th diagonal entry of  is the CFR at the k-th frequency
where H is the channel impulse response (CIR) matrix and bin, or equivalently is the k-th eigenvalue to the k-th column
n is the AWGN vector. The CIR matrix H is circulant, and vector of F H .
its first column is h = [h(0), h(1), . . . , h(L − 1), 0, . . . , 0]T , Recalling Eq. (5), the DFnT matrix is also circulant. Thus,
where h(l), l = 0, . . . L − 1, are the CIR taps and L is the the matrix  is also a diagonal matrix whose diagonal entries
maximum delay spread. are the eigenvalues of  with respect to FH . In [39], it
On the other hand, if ZP is used, one can also arrive is derived that the eigenvalue of  is the root Zadoff-Chu
at Eq. (25) by overlap-and-add operation, [Chapter 12, 48]. sequence, given by
 π 2
Therefore, Eq. (25) can be a general model for the OCDM e− j N k N ≡ 0 (mod2)
signal transmitting under LTI channels based on both  (k) = π (30)
e− j N k(k−1) N ≡ 1 (mod2).
CP and ZP.
Before we recover the symbols, we could first compensate Based on the commutative law of the product of two diagonal
the channel H in Eq. (25), and then perform DFnT to recover matrices, Eq. (29) can be further given by
the transmitted symbols in x. Alternatively, we first perform y =  H Fx + w. (31)
DFnT on the received signal as
Before compensating the CFR , we first cancel out the phase
r = r = Hs + n. (26) induced by . The equalized signal is given by
Based on and the convolution property of DFnT in Eq. (7), y = Gy = GFx + Gw, (32)
Eq. (26) is further given by
where G is a diagonal matrix with its k-th diagonal entry G(k)

r = H x + nH to be the coefficients of the single-tap equalizer. For example,
if zero-forcing (ZF) criterion is adopted, G(k) is
= Hx + n. (27)
G Z F (k) = −1 (k) , (33)
Therefore, the chirp waveforms  H vanish and are transparent
to the channel H as if symbols are transmitted directly and if minimum mean square error (MMSE) is adopted,
without modulating the chirps after the DFnT at the receiver. G(k) is
In addition, since  is unitary, the noise n is still AWGN. ∗ (k)
From Eq. (27), the CIR H can be compensated by using the G M M S E (k) = , (34)
| (k)|2 + ρ −1
multi-tap time-domain equalization (TDE) in Fig. 7 (a) or by
where ρ is the signal-to-noise ratio (SNR). Finally, the signal
using the more efficient single-tap FDE in Fig. 7 (b) to recover
is transformed by IDFT to recover the transmitted information.
the symbols x. More sophisticated nonlinear equalizers or
If the ZF equalizer is employed, the recovered signal is
decoding algorithms, e.g., decision feedback equalizer (DFE)
and maximal likelihood (ML) detection, can be more powerful xZ F = x + F H −1 w, (35)
for the improvement of performance [49], [50]. In this paper, and, on the other hand if MMSE is used, the signal is
we will show the block iterative DFE in Section VI.  
 H

xM M S E = F H H F x
B. Proposed Equalization Algorithm for OCDM   + ρ −1 I
In this subsection, we propose an efficient algorithm based H
+ FH w. (36)
on the FDE for compensating the channel dispersion imposed  H  + ρ −1 I
on the OCDM signal. The diagram of the proposed algorithm In practice, the MMSE equalizer can efficiently balance the
is illustrated in Fig. 8. In the proposed scheme, single-tap noise enhancement and channel compensation.

Authorized licensed use limited to: University of Surrey. Downloaded on March 29,2023 at 08:22:38 UTC from IEEE Xplore. Restrictions apply.
3952 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 9, SEPTEMBER 2016

Fig. 9. Schematic diagram of OFDM system (excluding the dash components) and the diagram of OCDM system based on the OFDM system (including
the dash components). (a) Transmitter. (b) Receiver #1. (c) Receiver #2

V. D ISCUSSIONS Consequently, the DFnT can be implemented by FFT in three

In this section, we will study the properties of the Fresnel steps:
and Fourier transforms. Based on these properties, implemen- 1) multiplying the chirp phase 1 ,
tation differences and compatibility of the OCDM and the 2) performing the DFT by FFT algorithm, and finally
OFDM are discussed. We also compare the implementation 3) multiplying the other chirp phase 2 ,
complexity of the proposed OCDM system to the OFDM, the where 1 and 2 are diagonal matrices whose m-th diagonal
DFT-Precoded (DFT-P-) OFDM, and the single-carrier (SC) entries are 1 (m) and 2 (m), respectively.
system with FDE.

A. Relation Between the Fourier and Fresnel Transforms B. Compatibility to the OFDM
In the continuous case, the kernel of the most ‘well-known’ In the previous subsection, Fresnel transform mathemati-
Fourier transform is cally formulates the OCDM, just as the Fourier transform
in OFDM. Inspecting Eq. (1) and (5), the Fresnel transform
ω ( f, t) = e− j 2π f t , (37) and DFnT are trigonometric transforms of quadratic phases.
and the expanded kernel of Fresnel transform in Eq. (1) is Both the Fresnel and Fourier transforms are linear canonical
π π
 transform [51]–[54]. The DFnT can be divided into three-step
ϕa ( f, t) = e− j 4 e j a f 2 −2 f t +t 2
. (38) process by using DFT. One would expect that the OCDM can
be integrated into the existing OFDM system easily without
In the discrete case, they are
significant modification.
1 2π
Fig. 9 provides the block diagram of the conventional
W (m, n) = √ e j N mn , (39)
N OFDM system (excluding the components in the dashed-line
and boxes). At the transmitter, the IDFT multiplexes symbols onto
1 π
parallel subchannels, and at the receiver DFT performs inverse
 (m, n) = √ e− j 4 operation to recover the symbols. After the DFT at the receiver
N
 π
 2 side, single-tap equalizer compensates each subchannel.
e j N m −2mn+n
2
N ≡ 0 (mod2) According to the relation between the DFT and the DFnT
× π π

e j 4N e j N m +m−2mn+n −n
2 2
N ≡ 1 (mod2), in Eq. (37)-(42), the OCDM can be integrated into the OFDM
(40) system with the additional operations in the dashed-line boxes.
At the transmitter, the three-step operation involving IDFT acts
respectively. The Fresnel transform in Eq. (38) or DFnT as the IDFnT. At the receiver end, there are two architectures.
in Eq. (40) consists of the Fourier transform or DFT with One is based on the receiver in Fig. 7, and the other is based
additional quadratic phases. In the discrete form, the additional on the proposed equalization algorithm in Section IV, refer to
quadratic phases are Fig. 8. In receiver #1, it should be noted that, as discussed
 π 2
in Section III-C, the equalizer could be either multi-tap TDE
− j π4 ej N m  N ≡ 0 (mod2)
1 (m) = e × π π
in Fig. 7 (a) or FDE in Fig. 7 (b). If FDE is adopted, in
e j 4N e j N m +m
2
N ≡ 1 (mod2), addition to single-tap equalizer, there is a DFT and an IDFT
(41) operation. The receiver #2 is the same as that in Fig. 8, which
and is the conventional FDE structure in the OFDM and SC-FDE
 π 2 systems. It means that the channel estimation and equalization
e jN n N ≡ 0 (mod2) etc., schemes designed for OFDM can be easily adapted into
2 (n) = π
n 2 −n
(42)
ej N N ≡ 1 (mod2), the OCDM system.

Authorized licensed use limited to: University of Surrey. Downloaded on March 29,2023 at 08:22:38 UTC from IEEE Xplore. Restrictions apply.
OUYANG AND ZHAO: ORTHOGONAL CHIRP DIVISION MULTIPLEXING 3953

In terms of the signal structure, both the OFDM and OCDM TABLE I
systems transmit the modulated waveforms in blocks. Between A DDITIONAL A RITHMETIC C OMPLEXITY OF THE OCDM
S YSTEM C OMPARED TO THE OFDM S YSTEM
the blocks, GI is used to avoid ISI. As shown in Section III-C,
in the OCDM system both CP and ZP can be used for
filling the GI. The structure of OCDM signal is the same
as that of OFDM signal. As a result, the OCDM signal
is also compatible to the OFDM signal. The design of
OCDM system can be well integrated into the OFDM system.
As a result, the generation and recovery of OCDM signal can
be realized using the existing OFDM system with additional
operations involving only phase rotation and one more IDFT
at the receiver.
However, there is difference between the spectra of OFDM
and OCDM systems. In the OFDM system, the aliasing signal
can be separated by setting subcarriers at the edge of the
spectrum to be nulls, and filters are then applied to reject The additional arithmetical complexity of the OCDM sys-
the aliasing signal. In the OCDM system, as discussed in tem compared to the OFDM system is provided in TABLE I.
Section III-D, the spectrum of the OCDM signal is directly It can be seen that, if the receiver #2 is adopted, the complexity
generated by DFnT, ranging from −B/2 to B/2. Thus, before of the OCDM system is slightly increased to that of the OFDM
filtering the aliasing signal, up-sampling is applied by a pulse- with (2 + 0.5 log2 N) multiplications per symbol.
shaping filter to fit the OCDM spectrum into the OFDM In TABLE I, the additional complexity of the DFT-P-OFDM
spectrum. compared to the conventional OFDM is provided. This scheme
requires a length-N DFT precoder at the transmitter and
C. Arithmetic Complexity of OCDM
another length-N IDFT reversing the operation at the receiver.
In Fig. 9, the similarities and differences between the The total additional complexity, to that of the OFDM is log2 N.
OFDM and OCDM systems are illustrated. In the OCDM In SC-FDE system, instead of DFT operation, pulse-shaping
system, there are two receiver schemes. For both schemes, the filter is used at the transmitter. Though the pulse-shaping filter
transmitter is the same, and thus the complexity of OCDM can be implemented by FFT, we assume that the time-domain
system depends on which receiver scheme is adopted. convolution operation is adopted with an oversampling ratio
In this subsection, we compare the arithmetic complexity of of 1. If so, the complexity is M per symbol, where M is the
the OCDM system to the OFDM system in terms of complex length of the pulse shaping filter. The filter length M is usually
multiplication operations. In addition, the DFT-P-OFDM that large enough to control the spectral shape, and it is typically
is used in the 3GPP LTE standard and SC-FDE is also in the order of several tens in practical implementation.
discussed. It should be noted that there are other compulsory
modules in a communication system, such as synchronizations VI. S IMULATION
and channel estimation. Additional consideration is required
because their complexities vary depending on the algorithms A. Peak-to-Average Power Ratio Performance
adopted. In this paper, we will not provide these details for In this subsection, simulations are carried out to investigate
brevity. the PAPR characteristics of the proposed OCDM system. For
At the transmitter, there are two additional phase rotations, comparision, we also consider the conventional OFDM, as
which require 2 additional complex multiplications per symbol well as the DFT-P-OFDM for its low PAPR.
compared to the transmitter of the OFDM system. In the OCDM system, the number of chirps are 256 and
If the receiver #1 is adopted, there are also two phase 1024 and the chirps are modulated in 16-QAM. Similarly,
rotation operations. In Fig. 7, either (a) TDE or (b) FDE can there are 256 and 1024 subcarriers in the conventional OFDM
be adopted. The complexity of TDE depends on the number of and DFT- OFDM systems. For all three systems, the baseband
taps of the transverse filter. If the number of taps is L, which signal is upsampled by a factor of 4 to emulates their actual
is larger than the CIR taps, the complexity of the TDE is L per PAPR. The PAPR is evaluated by complementary cumulative
symbol. On the other hand, in the FDE, besides the single-tap distribution function (CCDF), which is defined as the proba-
equalizer, it needs two more DFT operations, see Fig. 7 (b). bility of the PAPR of a signal exceeding a threshold PAPR0 .
Therefore, the FDE scheme requires additional log2 N complex In Fig. 10, the CCDF of the PAPR in the three systems are
multiplications per symbol. In the applications such as mobile provided. It is shown that the DFT-P-OFDM has the best PAPR
communication whose channel delay spread is relatively large, performance. For example, the probability of PAPR exceeding
FDE is more preferable than TDE in terms of computation 8.5 dB is less than 10−2 . The OFDM and OCDM systems have
complexity. the same PAPR performance. The PAPR0 for the probabilities
In the receiver scheme #2, besides the phase cancella- of 10−2 increases to 11 dB. Higher PAPR may impose more
tion and the single-tap equalizer, an additional IDFT are strigent requirements on the device linearity. Note that various
required. Thus, compared to OFDM, the additional complexity PAPR reduction methods have been proposed in the literature
is 0.5logN. and can be readily applied to the OCDM systems.

Authorized licensed use limited to: University of Surrey. Downloaded on March 29,2023 at 08:22:38 UTC from IEEE Xplore. Restrictions apply.
3954 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 9, SEPTEMBER 2016

Fig. 10. The PAPR characteristics of the OFDM, DFT-P-OFDM, and OCDM Fig. 11. BER performances of the OCDM systems with both ZF and MMSE
signals with 256 and 1024 subcarriers/chirps modulated in 16-QAM. equalizers and the OFDM system under the 10-ray multipath Rayleigh fading
channel.

B. Performance With Linear Equalizers

To study the performance of the proposed OCDM system,
simulations are performed in wireless channels. The bandwidth
is 10 MHz and 1024 chirps are modulated in M-ary QAM with
M from 4 to 16, and 64. The guard interval is filled with CP,
whose length, unless otherwise stated, is chosen larger than
the maximum excess delay of the channel to avoid ISI.
For comparison, both the OFDM and DFT-P-OFDM are
also simulated. Their bandwidth is also 10-MHz which is
divided into 1024 subcarriers modulated in QAM. CP is also
used to fill the guard interval. In the OFDM system, only ZF
equalizer is employed in the OFDM system for simplicity, as
both the ZF and MMSE equalizers achieve the same bit-error
rate (BER) performance in the OFDM system.
The multipath Rayleigh fading channel with 10 rays of equal Fig. 12. BER performances of the OCDM systems and the DFT Precoded
gain is considered. The maximum excess delay of the channel OFDM system under the 10-ray multipath Rayleigh fading channel.
is 5.4 μs, and it is assumed that the channel is quasi-static,
i.e. the channel remains static in one OCDM/OFDM block and For example, in the 64-QAM case, the OCDM system with
may vary for the next. Thermal noise, modeled as AWGN, is MMSE equalizer starts to outperform OFDM system from
added at the receiver. The BER is evaluated by signal-to-noise SNR = 20 dB. Nonetheless, the OCDM system of MMSE
ratio (SNR) which is given in E b /N0 , the ratio of bit energy E b equalizer still gets much better performance than OFDM for
to the noise power density N0 . SNR > 20 dB.
In Fig. 11, the BER performances of the OFDM and OCDM In Fig. 12, the BER performances of the DFT-P-OFDM
systems are provided. For the OCDM system of ZF equalizer, and OCDM systems are compared when the length of GI
it requires higher SNR to achieve the same BER compared to is sufficiently large (greater than the maximal delay of the
the OFDM, especially in low SNR region. The correspond- channel). As the DFT-P-OFDM system is also able to spread
ing BER curves approach those of the OFDM as the SNR the symbol over the entire bandwidth, the DFT-P-OFDM gets
increases. This degradation is from the noise enhancement of the same performance as OCDM does.
the ZF equalizer, which becomes smaller as the SNR increases.
The OCDM with MMSE equalizer outperforms that with C. Insufficient Guard Interval
the ZF equalizer as MMSE equalizer is able to balance the OCDM exhibits more resilience again the interference due
channel compensation and noise enhancement. The multipath to insufficient GI than the DFT-precoded OFDM. In Fig. 13,
diversity contributes to its superior performance over that of we investigate the performance of the three systems in terms
the OFDM. Although MMSE equalizer is able to effectively of various cyclic prefix lengths under the EVA channel model.
suppress noise, it should be noted that, the performance of The power delay profile is provided in TABLE II. The lengths
the OCDM system with MMSE is slightly degraded in the of CP are chosen to be 0.8, 1.6, and 3.2 μs to accommodate the
low SNR region. The degradation is more pronounced as maximum delay of ∼2.5 μs. In the case of 4-QAM, the OCDM
the modulation level increases from 4 and 16 to 64-QAM gets negligible performance degradation for different lengths
because high level modulation formats are sensitive to noise. of CP, while the DFT-P-OFDM has an error floor at the BER

Authorized licensed use limited to: University of Surrey. Downloaded on March 29,2023 at 08:22:38 UTC from IEEE Xplore. Restrictions apply.
OUYANG AND ZHAO: ORTHOGONAL CHIRP DIVISION MULTIPLEXING 3955

Fig. 13. BER performance of the OFDM, DFT-precoded OFDM, and OCDM systems under the LTE extended vehicle A channel model with various guard
interval length. (a) 4-QAM, (b) 16-QAM, and (c) 64-QAM.

TABLE II
P OWER D ELAY P ROFILE OF THE EVA M ODEL

of 10−4 as the GI reduces to 0.8 μs. For higher modulation

level, all the systems get certain degradations. Nonetheless,
the OCDM shows higher tolerance to the interference and
outperforms the other two systems. For example, in the case
of 16-QAM, if the GI is fixed at 1.6 μs, OCDM and the
DFT-P-OFDM have the error floor at BER = 2 × 10−5 and
1 × 10−4 , respectively.
The benefit of OCDM compared to DFT-P-OFDM can be Fig. 14. The BER performance of SC-FDE and OCDM systems with iterative
explained as follows. The DFT-P-OFDM in essence is a block- block decision feedback equalization with (a) 0.8-μs and (b) 3.2-μs GI.
transmitted single-carrier system, where symbols at the edge
of a block are susceptible to the interference from adjacent
blocks if the GI is not sufficient. The corrupted edge symbols (b) 32-μs GI. As the DFT-P-OFDM in essence is a single-
impose irreducible BER. In contrast, OCDM can eliminate carrier system, it can be observed in Fig. 14 (a) that, if the
this effect by spreading the ISI over the entire OCDM symbol GI is insufficient, the SC-FDE gets similar performance as the
period more evenly. This effect contributes to the improved DFT-P-OFDM, and it gets worse performance than the OCDM
performance over the DFT-P-OFDM system. system.
By applying the IB-DFE technique, we can observe that the
BER performance of both the OCDM and SC-FDE improves
D. Decision Feedback Equalization as the number of iterations increases. The improvement
Nonlinear equalization is able to improve the performance becomes negligible as the number of iterations is greater
of the DFT-P-OFDM and SC-FDE significantly, approaching than 2. For the SC-FDE system in Fig. 14 (a), the error
the matched-filter bound. In this subsection, we investigate floor increases from 1 × 10−4 to 1 × 10−5 by applying
the performance of the iterative block DFE (IB-DFE) [50], IB-FDE with 3 iterations. On the other hand, for the OCDM
[55]–[57] for the proposed OCDM system without channel system, by applying IB-FDE with 3 iterations, the performance
coding. In the IB-DFE algorithm, soft-decoding is adopted for degradation caused by insufficient GI becomes negligible.
each iteration [55]. In addition, we focus on the comparison of In Fig. 14 (b), as the GI is large enough, both the SC-FDE
the SC-FDE with the proposed OCDM system, and the EVA and OCDM systems achieves the same performance. One can
channel is still adopted. In both systems, 4-QAM is adopted also expect that, the DFT-P-OFDM system performs the same
for modulation. as the SC-FDE, under the same conditions.
In Fig. 14, the BER performances of the SC-FDE system In Fig. 14, the OFDM is not considered since its perfor-
and the OCDM system are provided with (a) 8-μs and mance is very poor without channel code. In the following

Authorized licensed use limited to: University of Surrey. Downloaded on March 29,2023 at 08:22:38 UTC from IEEE Xplore. Restrictions apply.
3956 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 64, NO. 9, SEPTEMBER 2016

feasibility of the proposed OCDM system. The results show

that the OCDM outperforms the conventional OFDM system,
and that it also exhibits more resilience against the interfer-
ence due to insufficient GI compared to DFT-P-OFDM and
SC-FDE. Consequently, the proposed OCDM system can be
an attractive alternative solution for high-speed communication
systems.

ACKNOWLEDGMENT
The authors would like to thank Prof. Paul Townsend,
Dr. Fatima Gunning, and Prof. Yong Liang Guan for their
comments and help during the preparation of this paper.

R EFERENCES
Fig. 15. The BER performance of the OFDM, SC-FDE and OCDM using [1] H. F. Talbot, “Facts relating to optical science. No. IV,” Philos. Mag.
forward error coding with 3.2-μs GI and code rates of (a) 2/3 and (b) 3/4. Ser. 3, vol. 9, no. 56, pp. 401–407, Dec. 1836.
[2] L. Rayleigh, “On copying diffraction-gratings, and on some phenomena
subsection, we will investigate the performance of OFDM, SC- connected therewith,” Philos. Mag. Ser. 5, vol. 11, no. 67, pp. 196–205,
Mar. 1881.
FDE, and OCDM systems with forward error coding (FEC). [3] J. T. Winthrop and C. R. Worthington, “Theory of Fresnel images.
I. Plane periodic objects in monochromatic light,” J. Opt. Soc. Amer.,
vol. 55, no. 4, pp. 373–381, 1965.
E. Channel Coding Performance [4] J. Wen, Y. Zhang, and M. Xiao, “The Talbot effect: Recent advances
In this subsection, low-density parity-check (LDPC) code in classical optics, nonlinear optics, and quantum optics,” Adv. Opt.
Photon., vol. 5, no. 1, pp. 83–130, Mar. 2013.
is employed to investigate the performance of OFDM, [5] J. R. Klauder, A. C. Price, S. Darlington, and W. J. Albersheim, “The
SC-FDE, and OCDM systems. Block bit-interleaver is theory and design of chirp radars,” Bell System Tech. J., The, vol. 39,
adopted, and the encoded bits are mapped into 4-QAM. EVA no. 4, pp. 745–808, Jul. 1960.
[6] R. J. Dengler et al., “600 GHz imaging radar with 2 cm range
channel is adopted, and the GI is chosen to be 3.2 μs to resolution,” in Proc. IEEE MTT-S Int. Microw. Symp. (IMS), Honolulu,
completely avoid the inter-symbol interference (ISI). HI, USA, Jun. 2007, pp. 1371–1374.
Fig. 15 presents the BER performances of the three systems [7] K. B. Cooper, R. J. Dengler, N. Llombart, B. Thomas, G. Chattopadhyay,
and P. H. Siegel, “THz imaging radar for standoff personnel screening,”
with code rates = (a) 2/3 and (b) 3/4. In both cases, OCDM IEEE Trans. Terahertz Sci. Technol., vol. 1, no. 1, pp. 169–182,
and SC-FDE have the same performance because the GI is Sep. 2011.
long enough to avoid ISI. Compared to the un-coded case, the [8] S. Darlington, “Demodulation of wideband, low-power FM signals,” Bell
coded OFDM gets significant improvement. In Fig. 15 (a), Syst. Tech. J., vol. 43, no. 1, pp. 339–374, 1964.
[9] W. Hirt and S. Pasupathy, “Continuous Phase Chirp (CPC) Signals for
with a code rate 2/3, the coded OFDM system has sim- binary data communication—Part I: Coherent detection,” IEEE Trans.
ilar performance to the OCDM system without IB-DFE. Commun., vol. COM-29, no. 6, pp. 836–847, Jun. 1981.
If IB-DFE of 3 iterations is adopted, both the coded SC-FDE [10] W. Hirt and S. Pasupathy, “Continuous phase chirp (CPC) signals
for binary data communication—Part II: Noncoherent detection,” IEEE
and OCDM systems get about 0.6 dB improvement than the Trans. Commun., vol. COM-29, no. 6, pp. 848–858, Jun. 1981.
coded OFDM system at BER = 1 × 10−5 . In Fig. 15 (b), if [11] F. Russo, “Demodulator structures for pulse-frequency-modulated sig-
nals,” IEEE Trans. Aerosp. Electron. Syst., vol. AES-12, no. 2,
the code rate increases to 3/4, with IB-DFE, both the coded pp. 127–130, Mar. 1976.
SC-FDE and OCDM systems gets 1.5 dB improvement than [12] R. L. Brewster and W. W. S. Jibrail, “Detection of FSK and DPSK data
the coded OFDM. In addition, the coded OFDM gets slight signals by pulse compression,” IEE Proc. F Commun., Radar Signal
degradation to the other two systems without IB-DFE. Process., vol. 129, no. 4, pp. 273–280, Aug. 1982.
[13] J. Pinkney, R. Behin, A. Sesay, and S. Nichols, “High-speed DQPSK
chirp spread spectrum system for indoor wireless applications,” Electron.
VII. C ONCLUSION Lett., vol. 34, no. 20, pp. 1910–1911, Oct. 1998.
[14] A. Kadri, R. K. Rao, and J. Jiang, “Low-power chirp spread spectrum
In this paper, we have proposed the OCDM system, which is signals for wireless communication within nuclear power plants,” Nucl.
in essence the principle of multiplexing a bank of orthogonal Technol., vol. 166, no. 2, pp. 156–169, May 2009.
[15] C. He, M. Ran, Q. Meng, and J. Huang, “Underwater acoustic com-
chirped waveforms for high-speed communication. It achieves munications using M-ary chirp-DPSK modulation,” in Proc. IEEE
the maximum SE for CSS systems just as the OFDM does for 10th Int. Conf. Signal Process. (ICSP), Beijing, China, Oct. 2010,
FDM systems. Implementation of the proposed system is based pp. 1544–1547.
[16] M. Palmese et al., “Experimental validation of a chirpbased underwater
on the Fresnel transform, and its convolution theorem gives the acoustic communication method,” in Proc. Meetings Acoust., Paris,
analytical approach to model the transmission of OCDM signal France, Jun./Jul. 2008, p. 030002, doi: 10.1121/1.2988054.
under LTI channels. In the OCDM, both the time-domain and [17] Wireless Medium Access Control (MAC) and Physical Layer (PHY)
Specifications for Low-Rate Wireless Personal Area Networks (WPANs),
frequency-domain equalizers can be applied for channel com- IEEE Standard 802.15.4a-2007, 2007, pp. 1–203.
pensation. By the eigen-decomposition of the DFnT matrices, [18] H. Liu, “Multicode ultra-wideband scheme using chirp waveforms,”
a simpler and more efficient single-tap equalization algorithm IEEE J. Sel. Areas Commun., vol. 24, no. 4, pp. 885–891, Apr. 2006.
is further proposed for the OCDM system. Simulations under [19] I. Dotlic and R. Kohno, “Low complexity chirp pulsed ultra-wideband
system with near-optimum multipath performance,” IEEE Trans. Wire-
wireless multipath channel are carried out to validate the less Commun., vol. 10, no. 1, pp. 208–218, Jan. 2011.

Authorized licensed use limited to: University of Surrey. Downloaded on March 29,2023 at 08:22:38 UTC from IEEE Xplore. Restrictions apply.
OUYANG AND ZHAO: ORTHOGONAL CHIRP DIVISION MULTIPLEXING 3957

[20] L. Yang and G. B. Giannakis, “Ultra-wideband communications: An [47] A. J. Lowery and L. B. Du, “Optical orthogonal division multiplexing
idea whose time has come,” IEEE Signal Process. Mag., vol. 21, no. 6, for long haul optical communications: A review of the first five years,”
pp. 26–54, Nov. 2004. Opt. Fiber Technol., vol. 17, no. 5, pp. 421–438, Oct. 2011.
[21] A. Springer et al., “A robust ultra-broad-band wireless communication [48] A. Goldsmith, Wireless Communications, 1st ed. Cambridge, U.K.:
system using SAW chirped delay lines,” IEEE Trans. Microw. Theory Cambridge Univ. Press, 2005, p. 674.
Techn., vol. 46, no. 12, pp. 2213–2219, Dec. 1998. [49] B. Ng, C.-T. Lam, and D. Falconer, “Turbo frequency domain equal-
[22] D. P. Morgan, “A history of surface acoustic wave devices,” Int. J. High ization for single-carrier broadband wireless systems,” IEEE Trans.
Speed Electron. Syst., vol. 10, no. 3, pp. 553–602, 2000. Wireless Commun., vol. 6, no. 2, pp. 759–767, Feb. 2007.
[23] D. R. Arsenault and P. Das, “SAW Fresnel transform devices and their [50] N. Benvenuto, R. Dinis, D. Falconer, and S. Tomasin, “Single carrier
applications,” in Proc. Ultrason. Symp., New York, NY, USA, 1977, modulation with nonlinear frequency domain equalization: An idea
pp. 969–973. whose time has come—Again,” Proc. IEEE, vol. 98, no. 1, pp. 69–96,
[24] V. P. Plessky and L. M. Reindl, “Review on SAW RFID tags,” IEEE Jan. 2010.
Trans. Ultrason., Ferroelectr., Freq. Control, vol. 57, no. 3, pp. 654–668, [51] M. Moshinsk and C. Quesne, “Linear canonical transformations and their
Mar. 2010. unitary representations,” J. Math. Phys., vol. 12, no. 8, pp. 1772–1780,
[25] R. Brocato, J. Skinner, G. Wouters, J. Wendt, E. Heller, and J. Blaich, 1971.
“Ultra-wideband SAW correlator,” IEEE Trans. Ultrason., Ferroelectr., [52] S.-C. Pei and J.-J. Ding, “Eigenfunctions of linear canonical transform,”
Freq. Control, vol. 53, no. 9, pp. 1554–1556, Sep. 2006. IEEE Trans. Signal Process., vol. 50, no. 1, pp. 11–26, Jan. 2002.
[53] L. M. Bernardo, “ABCD matrix formalism of fractional Fourier optics,”
[26] A. Georgiadis and M. Detratti, “A linear, low-power, wideband CMOS
Opt. Eng., vol. 35, no. 3, pp. 732–740, Mar. 1996.
VCO for FM-UWB applications,” Microw. Opt. Technol. Lett., vol. 50,
[54] E. Sejdić, I. Djurović, and L. Stanković, “Fractional Fourier transform
no. 7, pp. 1955–1958, Jul. 2008.
as a signal processing tool: An overview of recent developments,” Signal
[27] J. Ryckaert, G. Van der Plas, V. De Heyn, C. Desset, B. Van Poucke,
Process., vol. 91, no. 6, pp. 1351–1369, Jun. 2011.
and J. Craninckx, “A 0.65-to-1.4 nJ/burst 3-to-10 GHz UWB all-digital
[55] N. Benvenuto and S. Tomasin, “Iterative design and detection of a
TX in 90 nm CMOS for IEEE 802.15.4a,” IEEE J. Solid-State Circuits,
DFE in the frequency domain,” IEEE Trans. Commun., vol. 53, no. 11,
vol. 42, no. 12, pp. 2860–2869, Dec. 2007.
pp. 1867–1875, Nov. 2005.
[28] S. Solda, M. Caruso, A. Bevilacqua, A. Gerosa, D. Vogrig, and [56] F. Silva, R. Dinis, N. Souto, and P. Montezuma, “Approaching the
A. Neviani, “A 5 Mb/s UWB-IR transceiver front-end for wireless sensor matched filter bound with block transmission techniques,” Trans. Emerg.
networks in 0.13 μm CMOS,” IEEE J. Solid-State Circuits, vol. 46, Telecommun. Technol., vol. 23, no. 1, pp. 76–85, Jan. 2012.
no. 7, pp. 1636–1647, Jul. 2011. [57] A. Silva, J. Assunção, R. Dinis, and A. Gameiro, “Performance evalu-
[29] J. G. Proakis and M. Salehi, Digital Communications, 5th ed. ation of IB-DFE-based strategies for SC-FDMA systems,” EURASIP J.
New York, NY, USA: McGraw-Hill, 2007. Wireless Commun. Netw., vol. 2013, p. 292, 2013, doi: 10.1186/1687-
[30] F. Gori, “Why is the Fresnel transform so little known?” in Current 1499-2013-292.
Trends in Optics, Lasers and Optical Engineering, J. C. Dainty, Ed.
New York, NY, USA: Academic, 1994, pp. 139–148.
[31] P. Szwaykowski and V. Arrizon, “Talbot array illuminator with multilevel
phase gratings,” Appl. Opt., vol. 32, no. 7, pp. 1109–1114, Mar. 1993.
[32] A. Papoulis, “Pulse compression, fiber communications, and diffraction: Xing Ouyang received the M.S. degree in informa-
A unified approach,” J. Opt. Soc. Amer. A, vol. 11, no. 1, pp. 3–13, tion science and engineering from Dalian Polytech-
Jan. 1994. nic University, China, in 2013. He is currently with
[33] V. Arrizón and J. Ojeda-Castañeda, “Fresnel diffraction of substructured the Tyndall National Institute, University College
gratings: Matrix description,” Opt. Lett., vol. 20, no. 2, pp. 118–120, Cork, Ireland. His current research topics focus on
Jan. 1995. the multicarrier transmission techniques and digital
[34] R. Iwanow et al., “Discrete Talbot effect in waveguide arrays,” Phys. signal processing in both the wireless and optical
Rev. Lett., vol. 95, no. 5, p. 053902, Jul. 2005. communication systems, as well as the transmitter
[35] K. B. Wolf and G. Krötzsch, “Geometry and dynamics in the Fresnel and receiver diversity techniques for fading commu-
transforms of discrete systems,” J. Opt. Soc. Amer. A, vol. 24, no. 9, nication channels.
pp. 2568–2577, Sep. 2007.
[36] V. Arrizón and J. Ojeda-Castañeda, “Multilevel phase gratings for array
illuminators,” Appl. Opt., vol. 33, no. 25, pp. 5925–5931, Sep. 1994.
[37] V. Arrizón, J. G. Ibarra, and J. Ojeda-Castañeda, “Matrix formulation of
the Fresnel transform of complex transmittance gratings,” J. Opt. Soc. Jian Zhao received the B.Eng. degree from the
Amer. A, vol. 13, no. 12, pp. 2414–2422, Dec. 1996. University of Science and Technology of China,
[38] C. Zhou and L. Liu, “Simple equations for the calculation of a multilevel Hefei, China, in 2002, and the M.Phil. and Ph.D.
phase grating for Talbot array illumination,” Opt. Commun., vol. 115, degrees from the Chinese University of Hong Kong,
nos. 1–2, pp. 40–44, Mar. 1995. in 2004 and 2007, respectively. He is currently
[39] X. Ouyang, C. Antony, F. Gunning, H. Zhang, and Y. L. Guan. (2015). with the Tyndall National Institute and University
“Discrete Fresnel transform and its circular convolution.” [Online]. College Cork. Since 2009, he has captured over
Available: http://arxiv.org/abs/1510.00574 A
C2.4 million funding as the principal investigator
[40] L. Rabiner, R. Schafer, and C. Rader, “The chirp z-transform algorithm,” from the Science Foundation Ireland (SFI), EU FP7,
IEEE Trans. Audio Electroacoust., vol. 17, no. 2, pp. 86–92, Jun. 1969. Industry, Enterprise Ireland, and the Irish Research
[41] L. Bluestein, “A linear filtering approach to the computation of discrete Council. He is also the Co-PI of two Hong Kong
Fourier transform,” IEEE Trans. Audio Electroacoust., vol. 18, no. 4, Research Grants Council projects. He was a recipient of the SFI Career
pp. 451–455, Dec. 1970. Development Award 2016, the Starting Investigator Research Award 2012,
[42] L. Rabiner, “The chirp z-transform algorithm—A lesson in serendipity,” and the Technology Innovation and Development Award 2013. The team he
IEEE Signal Process. Mag., vol. 21, no. 2, pp. 118–119, Mar. 2004. led was the EMEA finalist in Alcatel-Lucent Innovation Competition 2012.
[43] R. Frank, S. Zadoff, and R. Heimiller, “Phase shift pulse codes with He holds two patents. He has authored over 110 technical papers (over
good periodic correlation properties,” IRE Trans. Inf. Theory, vol. 8, 75 senior-author papers and 16 invited papers) in peer-reviewed international
no. 6, pp. 381–382, 1962. journals and conference proceedings. His current research interests include
[44] R. W. Chang, “Synthesis of band-limited orthogonal signals for multi- optical OFDM, advanced modulation and detection schemes, digital signal
channel data transmission,” Bell System Tech. J., The, vol. 45, no. 10, processing for optical communications, and fiber nonlinearity mitigation.
pp. 1775–1796, Dec. 1966. He served over ten conferences as a TPC Member/Co-Chair, and is a Reviewer
[45] S. Weinstein and P. Ebert, “Data transmission by frequency-division mul- for Funding for Scientific Research Belgium, Hong Kong Research Grants
tiplexing using the discrete Fourier transform,” IEEE Trans. Commun. Council, and various journals, including Optics Express, Optics Letters, the
Technol., vol. COM-19, no. 5, pp. 628–634, Oct. 1971. Journal of Lightwave Technology, Photonics Technology Letters, the Journal
[46] J. Armstrong, “OFDM for optical communications,” J. Lightw. Technol., of Optical Communications and Networking, and the IEEE T RANSACTIONS
vol. 27, no. 3, pp. 189–204, Feb. 1, 2009. ON V EHICULAR T ECHNOLOGY .

Authorized licensed use limited to: University of Surrey. Downloaded on March 29,2023 at 08:22:38 UTC from IEEE Xplore. Restrictions apply.