a 10 Gbps Burst-Mode Clock and Data Recovery
Circuit with Continuous Clock Output
Runxiang Yu, Roberto Proietti, Shuang Yin,
and S.J. B Yoo
Junya Kurumida
National Institute of Advanced Industrial Science and
Technology
Ibaraki, Japan
kurumida@aist.go.jp
Department of Electrical and Computer Engineering
University of California at Davis, Davis, USA
sbyoo@ucdaivs.edu
Abstract—This paper presents a 10-Gbps burst-mode clock and
data recovery (BM-CDR) circuit based on the phase picking
method. Experiment demonstrates the proposed BM-CDR circuit
is able to align the burst-mode incoming data to the local clock
with a maximum phase misalignment of ±π/8 within 20 ns.
Keywords-Burst mode; clock
synchronization; multiphase clock
I.
and
data
recovery;
bit
extracted data clock with four phase-delayed copies of the local
clock at the same rate as the data. Two important
characteristics of the BM-CDR are its phase acquisition time
and accuracy of the phase alignment. Here, a phase
synchronization time of 20 ns and phase alignment accuracy of
±π/8 are achieved which are sufficient for 10 gigabit Ethernet
PON (10G-EPON) application according to IEEE802.3 av
10G-EPON standards and may be also useful for future optical
packet switching network.
INTRODUCTION
The rapid growing of the last mile solution such as passive
optical networks (PONs) activates research on burst mode data
operation where the optical line terminal (OLT) must deal with
asynchronous packets with different amplitude and phase,
necessitating clock and data recovery (CDR) circuit with fast
clock recovery and data retiming. In future optical packet
switching (OPS) network, burst mode operation is essential to
conduct burst packet aggregation at the OPS edge router [1]. In
both cases, a burst-mode receiver (BMRx) plays a key role
which is responsible for amplitude and phase recovery at the
beginning of every packet. A generic schematic of a BMRx is
shown in Fig. 1. At the front end of the BMRx is a BM limiting
amplifier (BM-LA) with automatic gain control (AGC) and
automatic threshold control (ATC) which is responsible for
amplitude recovery. Fast clock and data recovery (CDR)
together with phase acquisition is then performed by a BMCDR. This paper focuses on the BM-CDR aspect of the
BMRx. Among the existing solutions, [2] incorporates gated
voltage controlled oscillators (GVCO) to acquire rapid phase
locking but results in higher phase noise as it does not filter out
the input jitter. More seriously, the gating behavior would
cause momentary fluctuation on the recovered clock,
potentially incurring undesired jitter and inter-symbol
interference (ISI). BM-CDR based on injection locking
technique suffers from issues such as limited locking range,
process, temperature and supply (PVT) variation and weak
injection signals [3]. Oversampling in time requires a clock
frequency higher than the bit rate while oversample in space
only using multiphase clock with a frequency equal to the bit
rate [4]. Also it is desirable for a BM-CDR circuit to align the
burst data to a local clock so it provides a continuous clock
output for existing PON Large scale integration (LSI) and
RocketIO transceiver on field programmable gate array
(FPGA) for data processing. Here the proposed BM-CDR
technique is based on phase picking method by mixing the
This work was supported in part by DoD contract #H88230-08-C-0202
Electrical
path
BMRx
BM-LA
PD/TIA
AGC
This work
BM-CDR
ATC
Figure 1. Generic schematic of a burst mode receiver. PD: photo-detector;
TIA: transimpedance ampfilier; BM-LA: burst mode limiting amplifier;
AGC:automatic gain control; ATC:automatic threshold control.
II.
Proposed 10GHz BMCDR Circuit
AMP
1:4
AMP
Fanout
AMP
Buffer
AMP
LO
O/E
CIRCUIT DESCRIPTION AND IMPLEMENTATION
1:2
Fanout
Buffer
Clock Extraction
NRZ
BPF LA
to RZ
Delay
Line
1:4
Fanout
Buffer
0º
90º
180º
Comparator
Comparator
Comparator
270º
Comparator
1:4
Fanout
Buffer
0º
90º
180º
FPGA
4:1
selector
270º
Figure 2. Schamtic of the proposed 10Gbps BM-CDR. O/E: optical to
electrical converter; LA: limiting amplifier; AMP: amplifier; BPF:bandpass
filer
The proposed BM-CDR is implemented for operation at 10
Gbps for non-return-to-zero (NRZ) data. The main building
blocks include an Atlys Spartan-3 FPGA development board
from Digilent and a custom designed printed circuit board
(PCB) with individual chips from Hittite Microwave built in as
shown in Fig. 2. All the major components are listed in Table.
I. The incoming optical data is O/E converted then buffered.
Four time-delayed and phase-delayed copies are prepared for
latency. Here four comparators with 700-ps propagation delay
are employed which are equivalent to a 2-bit ADC. The phase
alignment accuracy of this configuration can be expressed as
s = ±π / 2n
(1)
when n is the number of mixers employed in this case.
1:4 Fanout Output (V)
the 4:1 selector. After NRZ-to-RZ conversion, data clock is
extracted by a high Q band-pass filter (BPF) with a Q of 200.
While it is possible to insert the system clock as a subcarrier
inside the transmitted signal [5], it complicats the hardware
required for the transmitter. The extracted clock is then
amplified and phase delayed before entering the RF mixers to
frequency mix with the four copies of the local clock for phase
detection. The threshold of the four comparators is set to 48
mV which is chosen according to the maximum mixer output
voltage in the worst case as described in Fig. 3 and Fig. 5. By
matching the mixer outputs to a look-up table, the FPGA
selects one of the four parallel burst mode packets at the
selector inputs which has the largest phase margin as the
optimum recovered data and feed them to the next stage (either
a PON LSI or RocketIO Transceiver on a FPGA) for further
data processing. A D flip-flop (DFF) could be easily added at
the BM-CDR output for data retiming.
0
Port 1
Port 2
-0.2
Port 3
-0.4
Port 4
-0.6
150
100
Time (ps)
50
0
200
Figure 4. Measured phase delayed copies of the Data before the 1:4 selector
TABLE I.
Model
Number
LIST OF COMPONENTS
Description
Function
Vendor name
HMC720LC3C
14Gbps 1:2 Fanout fuffer
Hittite
HMC940LC4B
13Gbps 1:4 Fanout fuffer
Hittite
HMC958LC4B
14Gbps 4:1 Selector
Hittite
HMC721LC3C
14Gbps XOR/XNOR gate
Hittite
HMC914LP4E
12.5Gbps Limiting amplifier
Hittite
HMC788LP2E
DC-10 GHz Gain block
Hittite
Hittite
National
semiconductor
TLD
7-10GHz double balanced mixer
LMH7323
Quad 700ps high speed
comparator
Atlys
Spartan-6 LX45 FPGA Board
Digilent
Bandpass filter
10GHz band-pass filter Q=200
RLC
Electronics
0
-0.5
-180 -90
0
90
Phase delay (Degree)
(b)
Mixer output (a.u)
Mixer output (a.u)
0.5
III.
Signal 1545 nm
HMC171C8
(a)
The multiphase clock/data generation is achieved by the 1:4
fanout buffer with the electrical co-planer waveguide with
ground plane (CPWG) on the PCB. The measured delay
deviation from the target value is within ±2ps which is
reasonable considering the skew value (±3ps) specified in the
HMC940LC4B datasheet. Four phase delayed copies of the
data are recorded as show in Fig. 4
EXPERIMENT AND DISCUSSION
t
t
t
50:50
FDL
IM EDFA
RF
driver
O/E
10G BMCDR
50:50
TDL
t
PPG
CLK
10GHz
Figure 5. Example of a figure caption. (figure caption). IM: Intensity
modulator; TLD: tunable laser didoe; PPG: pulsed pattern generator;
FDL:fixed delay liner; TDL: tunable delay line; CLK: Clock; 50:50: 50:50
fiber coupler
0.5
0
-0.5
-135 -45 45 135
Phase delay (Degree)
Figure 3. Mixer output as a function of phase delay (a) best case (b) worst
case
Theoretically the phase delay between the extracted clock
and the local clock can be uniquely identified by frequency
mixing of the extracted clock with two phase delayed copies of
the local clock [5]. But low speed (<1Gsps) analog-to-digital
converters (ADC) which are needed later on to digitize the
mixer outputs will inevitably introduce significant conversion
latency. So there is a trade-off between cost, resolution and
Setup in Fig. 5 is used to generate burst-mode data with
varying phases. The pulse pattern generator (PPG) generates a
periodic 1024-bit long sequence (constructed using 27-1 PRBS)
followed by a 1546-bit long guard time. The electrical
sequence is amplified and fed into a 10-Gbps optical intensity
modulator (IM) to modulate the output of a tunable laser diode
(TLD) at 1545 nm. The modulated optical signal is amplified
and split into two by a 50:50 fiber coupler. After proper delay
adjustment, the optical signals in the two branches are
interleaved by another 50:50 fiber coupler. The guard time
between the adjacent packets either 23.5 ns (235 bits '0') or
27.7 ns (277 bits '0'). In the end, the interleaved signal is O/E
converted and sent to the proposed BM-CDR.
The minimum time needed for synchronization under
current configuration is approximately 20ns which is
determined mainly by the mixer output rising time (~10ns
when using a BPF with a Q of 200 as shown in Fig. 6), the
FPGA decision time (<4ns, 2 clock cycles when clocking
@500MHz) and the 4:1 selector select time (< 100ps).
Figure 6. Rising/falling edges of the mixer outputs when using bandpass
filter with a Q of 200 for clock extraction. Top: best case, Bottom: worst case
Mixer output rising/falling time depends on the BPF used
to extract the data clock. A BPF with smaller Q can be used to
reduce the rising/falling time of the mixer output. The optimum
Q for this particular application is still under investigation. The
scope trace of the extracted clock and the reference clock is
illustrated in Fig.7 showing no degradation in terms of jitter.
Error free Operation is achieved using the configuration as
shown in Fig. 5 under various phase delay between adjacent
packets. This implies the delay of the clock path tracks that of
the data path very well and a retiming DFF in the next stage
will always sample in the vicinity of the data eye. The
amplitude electrical input of the BM-CDR is kept above 100
mV which is specified as the minimum input voltage for the
1:2 fanout buffer (HMC720LC3C) in its datasheet. To measure
BER as a function of input optical power requires additional
BM-LA as shown in Fig. 1 which is not available at this time.
The proposed BM-CDR technique assumes that the
extracted clock and the local reference clock are matched in
frequency since the same clock source is used for the 10Gbps
RZ-OOK transmitter and the BM-CDR in the experiment.
However, error will occur if the frequency walk-off between
the two clocks is greater than 1/4 of the bit time cross the entire
packet. The commercially available electrical 10-GHz local
oscillators (from Microwave Dynamics) have a typical
frequency stability of ±5 part-per-million (ppm). Therefore
without any modification, the proposed BM-CDR circuit can
accommodate packet length up to 25,000 bits without error.
For longer packets, the selection of the data path with optimal
phase alignment can be achieved by tracking the phase delay
variation over the entire packet length and making necessary
changes to the 4:1 selector on-the-fly. Reconfiguring the 4:1
selector during the middle of data transmission is expected to
introduce negligible penalty since the selector (HMC958LC5)
can support a maximum select rate of 14 GHz.
IV.
CONCLUSION
A 10Gbps burst-mode clock and data recovery (BM-CDR)
circuit based on phase picking method is presented.
Experiment demonstrates the proposed BM-CDR circuit is able
to align the burst mode incoming packets to the local clock
with a maximum phase misalignment of ±π/8 within 20 ns.
REFERENCES
[1]
[2]
[3]
[4]
[5]
Figure 7. Recorded scope traces and RMS Jitter by Agilent 86100A. Top:
reference clock, Bottom: extracted clock. The Q of the BPF is 200.
S. J. B. Yoo, "Optical packet and burst switching technologies for the
future photonic Internet," Journal of Lightwave Technology, vol. 24, pp.
4468-92, 2006
M. Nogawa et al., “A 10 Gb/s burst-mode CDR IC in 0.13
µmCMOS,”in IEEE ISSCC Dig. Tech. Papers, Feb. 2005, pp. 228–229.
J. Lee and M. Liu, “A 20-Gb/s burst-mode clock and data recovery
circuit using injection-locking technique,” IEEE J. Solid-State Circuits,
vol. 43,no. 3, pp. 619–630, Mar. 2008.
B. J. Shastri and D. V. Plant, “5/10-Gb/s burst-mode clock and data
recovery based on semiblind oversampling for PONs: Theoretical and
experimental,” IEEE J. Sel. Topics Quantum Electron., vol. 16, no. 5,
pp. 1298–1320, Sep./Oct. 2010.
T. Sangsiri, S. A. Havstad, C. Kim, X. Jiang, B. Hoanca, and A.
E.Willner, “Bit synchronization using subcarriers for control
signaling,”IEEE Photon. Technol. Lett., vol. 11, pp. 602–604, May 1999.