04683394 - Department of Electronics

Two Enhanced Decision Feedback Equalizers for
10Gb/s Optical Communications
Matthew Hagman, Student Member, IEEE
Tadeusz Kwasniewski, Senior Member, IEEE
Department of Electronics
Carleton University
Ottawa, Canada
mhagman@doe.carleton.ca
Department of Electronics
Carleton University
Ottawa, Canada
usually lost in intensity modulated direct detection (IM/DD)
systems. Coherent detection techniques are much more
expensive due to the specialized optical components required.
In shorter metro-reach links (40km to 120km) it is still
advantageous to use EDC even with the square-law detection.
In these applications, EDC is effective in extending the reach
from 60km to 100km by partially compensating for ISI due to
chromatic dispersion. The most common EDC approach for
IM/DD systems is FFE/DFE. It has been shown that a 7-tap,
T/2 spaced FFE filter with a 1 tap DFE section can extend the
reach of a 10Gb/s link to 100km [1], [3] and [4]. Non-linear
equalizers such as MLSE have demonstrated reach extensions
up to 200km. These equalizers are particularly robust.
However, they come with the price of high power
consumption. To the author’s knowledge, the most recent
10Gb/s MLSE-EDC solution was implemented in 90nm
CMOS, utilizing a 4-state Viterbi algorithm and it consumes
4.5W. For this reason, the discussion here is restricted to
FFE/DFE-based EDC.
Abstract—In this work, two unique decision feedback equalizers
(DFE) for use in 10Gb/s optical communications are presented.
These equalizers are effective at cancelling post-cursor ISI as well
as pre-cursor ISI without the use of a feed-forward equalizer
(FFE). The removal of the FFE equalizer is desirable as it is very
expensive from a chip real-estate perspective. The synthetic
transmission lines used to achieve the analog delay in FFE filters
also suffer from performance issues such as limited bandwidth,
impedance mismatches, and nonlinearities which degrade the
efficacy of the filter. The proposed filter structures will be
evaluated via numerical simulation, and a comparison with
standard FFE/DFE techniques will be made.
Keywords-Electronic Dispersion Compensation, Decision
Feedback Equalizer, IIR, Equalization, Chromatic Dispersion.
I.
INTRODUCTION
With the recent trends in commoditization of 10Gb/s
optical components and the continuing improvements in
CMOS and SiGe processes, it has now become attractive to
implement Electronic Dispersion Compensation (EDC)
techniques to enhance the reach of 10Gb/s optical links, or to
replace legacy 2.5Gb/s links. There are two fundamental
challenges with EDC. The first challenge is the processing
speed of the electronics to deal with the high frequency
distortions which occur in the optical channel. The second
more fundamental challenge is the non-linear square-law
detection that is used to convert the intensity modulated optical
signal to an electrical signal [2]. Chromatic dispersion is in
fact a linear distortion that can be quiet effectively “un-done”
in the optical domain with the use of dispersion shifted fiber, or
with optical FIR filters [1]. These optical techniques are bulky,
expensive, and inherently non-adaptive.
It should be
mentioned however, that optical methods of chromatic
dispersion compensation do offer the advantage of multichannel compensation which is the reason that dispersion
compensation modules (DCFs) are still used in many legacy
long haul systems.
One of the challenges of FFE/DFE based EDC is the feedforward continuous time filter section. This filter is typically
designed using a delay line such as an LC-ladder or synthetic
transmission line. A series of variable gain amplifiers usually
serve as the filter taps. Process variation alone can add up to
20% uncertainty in the achieved delay from this filter
architecture. In the work by S. Reynolds et al., [5] it has been
shown that a 7-tap FFE filter can take 5mm of chip real estate.
Experimental evidence also shows that the bandwidth limit
imposed by the FFE section, as well as non-linearity in the tap
scaling amplifiers reduces the effectiveness of this filter. Since
the DFE section is effective at cancelling only post cursor ISI,
the FFE section is generally used with DFE to cancel at least
the linear portion of ISI. In further sections of this paper,
modifications to the DFE filter structure will allow for some
pre-cursor ISI and thus making it possible to remove the FFE
section.
II.
At the present time, EDC has been successfully used in two
different systems configurations. In long haul systems, it is
common to use coherent detection, optical QPSK, or predistortion techniques [2]. These techniques when combined
with EDC are particularly powerful as the electronic
compensation is able to recover the phase information which is
978-1-4244-2921-9/08/$25.00 ©2008 IEEE
SIMULATION MODEL
In order to simulate an optical links, it is essential to model
the effects of chromatic dispersion. Although chromatic
dispersion combined with square-law detection is a difficult
impairment to mitigate, it is actually quite easy to simulate
accurately. Using MATLAB™ a PRBS 211-1 NRZ signal is
125
Authorized licensed use limited to: Carleton University. Downloaded on June 2, 2009 at 11:10 from IEEE Xplore. Restrictions apply.
transmitted. The electro-absorption (EA) transmitter (most
common in metro links) is modeled with a 30ps rise and fall
time, an extinction ratio of 10dB, and optical chirp
characteristic of slightly positive (3GHz peak-to-peak) transient
chirp. These values were selected by consultation of the
datasheets of two commercially available 10Gb/s EA
transmitter modules. The chromatic dispersion is applied
directly to the transmitted signal using both the intensity and
frequency modulation characteristics with an over-sampling of
40 samples per bit (much more than required). The transmitter
output is shown in node “A” in Figure 1.
III.
A. Digital 1-bit Predictive DFE
The design of this filter structure attempts to exploit some
of the features of ISI resulting from chromatic dispersion. It
can be shown from equations (1), (2) and (3) that the ISI
resulting from chromatic dispersion has temporal symmetry.
The primary problem with any attempt to remove the FFE
section is the fact that the feed-forward equalizer is the only
element which treats pre-cursor ISI. The proposed modified
DFE shown in Figure 3 uses a second order DFE circuit
followed by a 100ps delay. The un-delayed DFE circuit
serves as the 1-bit predictor to cancel the ISI due to the
“future” bit.
100km SMF-28 Fibre
PRBS
1kΩ
Laser
Driver
A
B
EDC-EQ
C
THE PROPOSED 1-BIT PREDICTIVE DFE FILTERS
D
Figure 1: Simulation Diagram
The mathematical models for chromatic dispersion were
found in the derivation shown in [6]. The simulated eyediagrams from 0km to 130km show good agreement with
measured results. The signal shown at node “B” in Figure 1 is
the unfiltered optical signal after 80km of single-mode optical
fiber. The signal at node “C” is the point where amplifier
noise is added for bit error ratio (BER) analyses. Node “D”
shows the received signal where the pre-amplifier has applied
a bandwidth limit of 7.5GHz.
Figure 2: Diagram of the Proposed Digital Realization
One of the unique features of this filter is the fact that only
two coefficients are needed. Using temporal symmetry, it is
obvious that the ISI from the (k+1)th bit has the same influence
on the kth bit as the (k-1)th bit. The argument of temporal
symmetry however, does not take into account the impulse
response of the 10Gb/s optical receiver. As mentioned earlier,
the receiver consisting of a photo-detector and transimpedance amplifier functions as the matched filter and is
frequently modeled as a 4th order Bessel-Thompson filter.
This type of low pass filter does not have a temporally
symmetrical impulse response. However, since the length of
the impulse response of the optical receiver is short compared
to the effects of ISI due to chromatic distortion. In practice,
there may also be small differences in the d1 and d2
coefficients as only one of the DFE experiences the bandwidth
limitations and nonlinearities of the analog delay element. For
this work, these effects have been neglected.
The trans-impedance amplifier in optical communication
links serves as the matched filter and pre-amplifier. Here, the
trans-impedance and photo-detector are modeled by a 4th order
Bessel Thompson filter and a gain factor of 1kΩ. This is
typical of optical receivers used in the field. It is important to
mention here that when using this type of EDC, it is essential
to have a linear optical receiver with a suitable automatic gain
control to keep the output voltage amplitude at a constant.
This type of trans-impedance amplifier is readily available in
the field.
For noise analyses, the noise generated in the transimpedance amplifier is treated as the dominant noise source.
This assumption is only valid for un-amplified links and metro
links are generally un-amplified. In order to simulate an
amplified link, it is necessary to add Gaussian noise before the
detector. This adds the complexity of having to treat nonlinear noise which has a Χ-squared distribution due to the
squaring function of the optical detector.
B. Analog 1-bit Predictive DFE
The following DFE filter was designed to address one of
the weaknesses of the filter described in the previous section.
Although it will be shown later that the digital predictive DFE
filter offers an improvement in system BER, there still exists
an issue with error propagation due to the possibility of
126
Authorized licensed use limited to: Carleton University. Downloaded on June 2, 2009 at 11:10 from IEEE Xplore. Restrictions apply.
two proposed predictive DFE filters. The BER simulations
assume that the primary noise mechanism is due to the input
referred noise of the optical receiver. This assumption is a safe
one considered that the optical links discussed here are
unamplified. Figure 8 shows the simulated BER performance
of all of the filters discussed in this paper. The optical receiver
is assumed to have a back to back sensitivity of -19dBm with
an input optical extinction ratio of 10dB. This is common to
most PiN based 10Gb/s receivers in the field.
incorrect decisions being made at the slicer circuits (in this
case, the combined slicer and D flip-flops).
This circuit replaces the un-delayed DFE section with an
analog filter section. This is used as the predictor and can be
described as an IIR filter with transfer function:
H (z ) =
1 − c ff z −1
1 − c fb z −1
(1)
Where the “c” coefficients are shown in Figure 3 along with
the block diagram.
Unfiltered Receiver Output 2200ps/nm
Receiver Electrical Output (V)
0.15
0.1
0.05
0
-0.05
-0.1
Figure 3: Block diagram of the proposed analog predictive DFE filter
-1
For stability, “cfb” must be constrained to the range of
-1<cfb<1. Simulations show that at 130km of single mode
fiber, cfb takes a typical value of -0.3. This value rapidly
approaches 0 for shorter fiber reaches and thus, stability is not a
concern. The interesting feature of this proposed filter is the
fact that the IIR filter section can be shown to approximate the
FFE filter in the popular FFE/DFE architectures used in 10Gb/s
optical links where “cff” offers 100ps of pre-cursor ISI
reduction, and the feed-back coefficient “cfb” cancels postcursor ISI. From the simulations, there was a high degree of
similarity in the impulse response of the optimized symbolspaced FFE filters, and the proposed IIR structure. This was
observed after the FFE filter taps were set via an LMS
algorithm with 75km, 100km, and 130km of chromatic
dispersion.
IV.
-0.5
0
Time (s)
0.5
1
-10
x 10
Figure 4: Eye-diagram of the unfiltered receiver output (130km)
Output Eyediagram with Standard 2nd Order DFE at 130km
0.2
0.15
Amplitude (V)
0.1
RESULTS
0.05
0
-0.05
The unfiltered (no EDC) electrical output of the receiver is
shown in Figure 4. This was simulated assuming a EA-type
transmitter with virtually zero chirp and 30ps rise/fall time and
130km of SMF-28 fiber. The modeled receiver has a linear
response with a trans-impedance gain of 1kΩ and a 4th order
roll-off with a bandwidth of 7.5GHz. The output of a 2nd order
DFE at 130km with no FFE section is given in Figure 5. The
asymmetry is due to the lack of pre-cursor ISI reduction.
Finally, the digital 1-bit predictive DFE gives an electrical
output waveform as shown in Figure 6. The symmetry is
restored, and there is 1dB additional eye opening. In order to
estimate the dispersion penalty, the eye opening was measured
for various lengths of optical fiber and the results were plotted
in Figure 7. Here, the 1-bit predictive DFE offers
approximately 20km (3dB power penalty) additional reach over
the standard DFE. At fiber lengths below 80km, there is little
difference between the standard DFE and the predictive DFE.
Given that error propagation is a known issue in DFE filters,
some bit error rate simulations were carried out to compare the
-0.1
-0.15
-1.5
-1
-0.5
0
Time (s)
0.5
1
1.5
-10
x 10
Figure 5: Eye-diagram of the received signal filtered by a 2nd order DFE
(130km)
127
Authorized licensed use limited to: Carleton University. Downloaded on June 2, 2009 at 11:10 from IEEE Xplore. Restrictions apply.
BER Performance Comparison of DFE Filters
-1
10
0.2
10
0.15
10
0.1
10
-2
-3
-4
BER
Amplitude (V)
Output Eyediagram with Enhanced DFE at 130km
0.25
0.05
-5
10
-6
10
0
-7
10
-0.05
No EDC
1-bit Pedictive DFE (digital)
2nd order DFE
-8
10
-0.1
-1.5
-1
-0.5
0
Time (s)
0.5
1
7-tap FFE, 2-tap DFE
1-bit Predictive DFE (analog)
1.5
-9
-10
10
x 10
Figure 6: Output signal from the proposed 1-bit predictive DFE (130km)
-21
-20.5
-20
-19.5
-19
-18.5
-18
-17.5
-17
Input Power to a 10Gb/s PiN-Based Receiver (dBm)
V.
Dispersion Power Penalty (dB)
0
-4
-6
No EDC
Standard 2nd Order DFE
1-bit Predictive DFE (digital realization)
0
20
40
60
80
100
Reach (km of SMF-28 Fibre)
120
CONCLUSION
Two modified DFE filter structures have been presented
and their system performance has been analyzed using
numerical simulation. The 1-bit predictive DFE offers
comparable performance to the FFE/DFE structure commonly
used in today’s optical links with a significantly shorter
physical length and fewer implementation complexities. When
completed, a circuit implementation in 0.13µm CMOS will be
added to this research. The present design takes up a space of
less than 1mm by 1.5mm offering a greater than 3-fold
reduction in chip area [5]. The power consumption determined
by simulation is 75mW and this may be reduced with further
optimization. This is four times less power than FFE structures
offering similar reach enhancement.
-2
-10
-16
Figure 8: Simulated bit error rate performance at 130km of a 2nd order DFE and
both realizations of the 1-bit predictive DFE
Simulated Dispersion Penalty of the Predictive DFE and Standard DFE
2
-8
-16.5
140
Figure 7: Dispersion penalty of a 2nd order DFE and the 1-bit predictive DFE
From the BER measurements, there is marked performance
improvement in the analog realization of the 1-bit predictive
DFE filter. As mentioned earlier, this is due to lower error
propagation from incorrect decisions made at the slicer in the
DFE. In these numerical results, the 7-tap FFE, 2-tap DFE
structure outperforms both predictive DFE designs. This was
expected given that the FFE structure allows for more precursor ISI reduction. The IIR-like response of the proposed
predictive design still offers comparable performance with a
significantly shorter physical length.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
Nielson, Torben, Chandrasekhar, S., “OFC 2004 Workshop on Optical
and Electronic Mitigation of Impairments,” IEEE Journal of Lightwave
Technology, Vol. 23, No. 1, January, 2005.
C. Laperle, B. Villeneuve, Z. Zhang, D. McGhan, H. Sun, M.
O’Sullivan, “Wavelength Division Multiplexing (WDM) and
Polarization Mode Dispersion (PMD) Performance of a Coherant 40Gb/s
Dual-Polarization Quadrature Phase Shift Keying (DP-QPSK)
Transceiver,” Optical Society of America, 2007.
P. M. Watts, V. Mikhailov, S. Savory, M. Glick, P. Bayvel, R. I. Killey,
“Electronic signal processing techniques for compensation of chromatic
dispersion,” Invited paper, workshop WMJ on Electronic Equalization
for Multigigabit Communications, at IEEE MTT-S International
Microwave Symposium (IMS 2006), San Francisco, CA, 11-16 June
2006.
F. F. Dai, S. Wei, R. Jaeger, “Integrated blind electronic equalizer for
fibre dispersion compensation,” IEEE International Symposium on
Circuits and Systems. ISBN: 0-7803-8834-8, May 2005.
S. Reynolds, P. Pepeljugoski, J. Schaub, J. Tierno, D. Beisser, “A 7-Tap
Transverse Analog-FIR Filter in 0.13µm CMOS for Equalization of
10Gb/s Fibre-Optic Data Systems,” IEEE International Solid-State
Circuits Conference, Session 18, High-Speed Interconnects and Building
Blocks, Paper 18.2, 2005.
Van Schyndel, Andre. “Post Detection Chromatic Dispersion
Compensation.” US Patent #US 6,775,631 B2. Assignee: Nortel
Networks Limited. Aug 10, 2004.
128
Authorized licensed use limited to: Carleton University. Downloaded on June 2, 2009 at 11:10 from IEEE Xplore. Restrictions apply.