674 IEEE PHOTONICS TECHNOLOGY LETTERS, VOL. 16, NO. 2, FEBRUARY 2004
Coherent Detection Method Using DSP for
Demodulation of Signal and Subsequent Equalization
of Propagation Impairments
Michael G. Taylor, Member, IEEE
Abstract—A new approach to coherent detection is demon-
strated which achieves the same high sensitivity as homodyne
detection but without the need to phase lock the local oscillator
laser. In addition, 1470 ps/nm of chromatic dispersion is compen-
sated with zero net penalty by electronic domain equalization, a
result which has not been achieved before because zero-penalty
equalization is not possible after direct detection. The method pro-
poses the use of high-speed digital signal processing technology,
and the experimental results are obtained using burst-mode
sampling followed by offline signal processing.
Index Terms—Chromatic dispersion (CD) compensation,
coherent detection, optical transmission, phase diverse detection.
I. INTRODUCTION
C
OHERENT detection has many advantages over direct
detection, which is why it was the subject of much
research activity ten years ago. It is sensitive to the phase as
well as the amplitude of the optical wave, and coherent detection
offers an inherent ultranarrow optical filtering capability useful
for dense wavelength-division multiplexing [1]. The mode
of coherent detection that gives the best sensitivity [lowest
bit-error rate at a given optical signal-to-noise ratio (OSNR)] is
homodyne detection, but this mode requires the use of special
narrow linewidth lasers that are phase locked, which makes it
expensive to implement. This letter proposes a new method of
coherent detection based on phase diverse detection and using
new high-speed digital signal processing (DSP) technology.
It possesses all the advantages of homodyne detection but
without the need to phase lock lasers, and so will be much more
cost effective. The reported experimental results show that the
method can detect a 10-Gb/s signal with very good sensitivity.
DSP technology operating at 10 Gb/s has very recently been
applied to soft decision forward-error correction [2], and as
a means to overcome distortions from chromatic dispersion
(CD) and polarization-mode dispersion in single-mode fiber
(SMF) and modal dispersion in multimode fiber [3]. To date,
all proposals for impairment compensation have been to use
DSP in conjunction with direct detection. The detected signal
is digitized by an analog-to-digital (A/D) converter operating at
the bit rate, and then passed to a digital integrated circuit which
applies a mathematical algorithm operating on near-neighbor
samples to reverse any distortion that the signal experienced
Manuscript received May 6, 2003; revised September 11, 2003.
The author is with the Optical Networks Group, Department of Electronic and
Electrical Engineering, University College London, London WC1E 7JE, U.K.
(e-mail: mtaylor@unodos.net).
Digital Object Identifier 10.1109/LPT.2003.823106
earlier in the optical domain. Kanter et al. [3] have demonstrated
compensation for the CD of 75 km of SMF at 10 Gb/s, but the
penalty was reduced rather than eliminated. This will always
be the case with CD compensation following direct detection,
because the phase information is discarded upon detection and
it is not possible to restore the signal to its original state. In
this letter, penalty-free compensation of CD is demonstrated
for the first time by equalization after coherent detection.
II. T
HEORY
Coherent detection involves beating an incoming signal with
light from a local oscillator (LO) laser [1]. With phase diverse
detection both the signal and LO are split in two paths and then
combined with one another before being detected by two pho-
todetectors [4], as shown in the diagram of the experimental
arrangement in Fig. 1. The LO receives an extra phase shift
of
in one arm. If the signal is written in complex form
as
, and the LO as , then the
powers in the two arms are
(1a)
(1b)
Either one of these detected powers alone cannot be used to re-
cover the information on the signal in the case where the differ-
ence frequency is less than the bit rate, because the information
vanishes during the null of the difference frequency envelope.
The information can be recovered from the two powers com-
bined, as follows:
(2)
where
, etc., and given that ( ) and the
phase of
are first determined during a lock acquisition
phase and then continuously tracked. Equation (2) is derived
from (1) assuming that the signal-squared term can be neglected,
which is reasonable because the power of the LO was kept about
20 dB higher than the signal during the experiment. Note that
(2) is different from the way information is usually obtained
using phase diverse detection [4]. Previous implementations
have been equivalent to squaring and adding
and , which
discards the absolute phase of the signal, whereas (2) generates
a complex (phasor) quantity. After digitizing the electrical
1041-1135/04$20.00 © 2004 IEEE
TAYLOR: COHERENT DETECTION METHOD USING DSP FOR DEMODULATION OF SIGNAL 675
Fig. 1. Arrangement of experiment to demonstrate sampled coherent detection.
signals from the two photodetectors, (2) can be implemented
in a DSP circuit. The LO and signal lasers do not have to
be phase locked, and the DSP is doing the last stage of
synchronous demodulation via a digital phase-locked loop. A
similar operation in the field of digital radio is referred to
as quadrature sampling. The result is that a continuous set
of samples of the signal’s complex electric field is available
within the DSP. The transmitted information can be extracted
directly, or signal processing can be applied first to compensate
for impairments such as CD.
For any coherent detection method to be feasible, it is impor-
tant to have a means of tracking the incoming signal state of
polarization (SOP). Although not demonstrated here, this can
be done by adding polarization diversity to the phase diverse ar-
rangement [4]. This would require four photodetectors instead
of two, but in fact, it is possible to perform polarization demul-
tiplexing using the polarization diverse configuration, so the in-
formation carrying capacity of the signal is doubled also.
III. E
XPERIMENT
To demonstrate the basic features of quadrature-sampled
detection, a binary phase-shift keying (BPSK) transmitter
was used, comprising a LiNbO
phase modulator driven via
a broad-band RF amplifier with a
pseudorandom
sequence at 10.66 Gb/s. A controlled amount of optical noise
was added to the signal through the two erbium-doped fiber
amplifiers shown in Fig. 1. The phase diverse configuration was
made up of individual fiber pigtailed components. It was found
that the phase difference between the arms drifted over time,
and to obtain a phase difference of
several measurements
were taken for each data point; in every case at least one was
found subsequently to have close to
phase difference. The
signal and LO lasers were external cavity lasers. The frequency
difference between them drifted and was about 100 MHz. The
SOPs of the signal and LO were approximately matched by
manual manipulation of the fibers. The outputs of the pho-
todetectors were recorded by a Tektronix TDS6604 real-time
sampling oscilloscope, having 6-GHz front end bandwidth.
This instrument was able to download continuous 4-
s blocks
from its two inputs in parallel at 20 GSa/s, and then the data
was transferred to a pesonal computer for offline processing.
The oscilloscope + postprocessing was used because, although
inexpensive when produced in volume, the cost of a dedicated
DSP for this purpose would be prohibitive.
Fig. 2. Measured
Q
factor versus OSNR.
Two sets of data of factor versus OSNR were collected, for
a noise-loaded back-to-back configuration and for transmission
over 89-km standard SMF with noise loading. The results are
shown in Fig. 2. Each data point comes from a pair of traces
4
s in length. The measured data was processed as follows.
1) Each trace was convolved with a nine-point impulse re-
sponse vector to reverse the nonflat frequency response of
the detector + scope front end. The two impulse response
vectors were determined by a simple adaptive process to
give the best
factor on one of the noise-loaded mea-
surements, and then the same vectors were applied to all
measurements, with and without fiber.
2) The transmit clock (10.66 GHz) and beat envelope (about
100 MHz) were recovered for the two channels.
3) The 20-GSa/s data was retimed using an interpolation
technique to exactly two samples per bit, with alternate
samples lying at the bit center. The same data would have
been obtained if the A/D converters were clocked syn-
chronously with the signal clock (and 2
the frequency).
For the quadrature sampling process, only one sample per
bit (at the bit center) is needed, and for CD compensation,
two samples per bit are required.
676 IEEE PHOTONICS TECHNOLOGY LETTERS, VOL. 16, NO. 2, FEBRUARY 2004
4) The two channels were combined according to (2), to give
the complex envelope of the signal electric field.
5) For the measurements over fiber, an equalization filter
was applied. The complex signal was convolved with the
impulse response (inverse Fourier transform) of the CD
transfer function
, truncated to seven
points. The value
s was used, equiv-
alent to 89 km of standard SMF.
6) The
factor was calculated using the decision threshold
technique on the alternate samples at the bit centers.
The signal processing operations above are all of the kind that
can be executed by a high-speed DSP. For Step 5, the seven-
element vector was found to give the same results within 0.2 dB
as a much longer vector (i.e., no truncation). Also, it was found
that the simple convolution used (a fully nonrecursive digital
filter) was as good as any mixed recursive-nonrecursive filter
with the same number of taps.
The calculated
versus OSNR curves are shown in Fig. 2.
The points for the back-to-back case indicate a very good sen-
sitivity, requiring an OSNR of 10.5 dB for
, which is
only 2.7 dB from the theoretical minimum. This is believed to
be the best ever reported back-to-back sensitivity at 10 Gb/s,
and is 4–5 dB better than a typical intensity modulation-direc-
tion result. The theoretical curve in Fig. 2 is for BPSK obtained
from the equations in [5], adapted to refer to OSNR instead of
received power.
Also in Fig. 2 are the curves for transmission over 89-km
fiber, both without and with postdetection equalization.
The equalization operation restores the sensitivity to the
back-to-back values. The penalty due to CD is reduced from
5 to 0 dB at
. As mentioned previously, the elimination
of the CD penalty by postdetection signal processing can be
achieved with coherent detection, but not direct detection.
IV. D
ISCUSSION AND
CONCLUSION
What has been shown in this letter is that the incorporation of
new high-speed DSP technology means that coherent detection
can be implemented with the high sensitivity associated with
homodyne detection but without phase locking the LO laser.
The detection of a BPSK encoded signal was demonstrated in a
burst-mode experiment with offline processing of the data. Sub-
sequent processing of the phasor representation of the signal
is possible by the digital signal processor to compensate for
propagation impairments, and this processing stage is equiv-
alent to inserting an optical component having characteristics
defined by the signal processing algorithm. A consequence of
this optical domain-equivalent nature is that propagation impair-
ments can be equalized completely. Penalty-free compensation
of 1470 ps/nm of CD was shown, and a similar zero-penalty
result can be expected for compensation of other impairments.
The method also opens up the possibility of employing phase
and polarization-encoded modulation formats cost effectively,
which are in general more tolerant to fiber propagation effects
and have higher spectral efficiency than the conventional for-
mats used with direct detection.
A
CKNOWLEDGMENT
The author thanks Tektronix for the loan of the oscilloscope
and H. Yaffe and F. Moody for the loan of key items of
equipment.
R
EFERENCES
[1] G. P. Agrawal, Fiber-Optic Communication Systems, 2nd ed. New
York: Wiley, 1997.
[2] T. Mizuochi, K. Ouchi, T. Kobayashi, Y. Miyata, K. Kuno, H. Tagami, K.
Kubo, H. Yoshida, M. Akita, and K. Motoshima, “Experimental demon-
stration of net coding gain of 10.1 dB using 12.4 Gb/s block turbo code
with 3-bit soft decision,” presented at the Optical Fiber Communications
Conf. (OFC 2003), Atlanta, GA, 2003, Paper PD21.
[3] G. Kanter, P. Capofreddi, S. Behtash, and A. Gandhi, “Electronic equal-
ization for extending the reach of electro-absorption modulator based
transponders,” presented at the Optical Fiber Communications Conf.
(OFC 2003), Atlanta, GA, 2003, Paper ThG6.
[4] L. G. Kazovsky, “Phase- and polarization-diversity coherent optical
techniques,” J. Lightwave Technol., vol. 7, pp. 279–292, July 1989.
[5] Y. Yamamoto, “Receiver performance evaluation of various digital op-
tical modulation-demodulation systems in the 0.5–10
m wavelength
region,” IEEE J. Quantum Electron., vol. QE-16, pp. 1251–1259, Nov.
1980.
