A 40 Gb/s Asynchronous Optical Packet Buffer Based on an SOA Gate Matrix for Contention Resolution
ABSTRACT We demonstrate a 40 Gb/s optical packet buffer with asynchronous and autonomous control for use in contention resolution. Layer-2 (packet recovery) measurements are presented with less than 1 dB power penalty up to 10 circulations.
A 40 Gb/s Asynchronous Optical Packet Buffer Based
on an SOA Gate Matrix for Contention Resolution
J.P. Mack, H.N. Poulsen, E.F. Burmeister, J.E. Bowers, and D.J. Blumenthal
Department of Electrical and Computer Engineering, University of California, Santa Barbara, CA 93106, U.S.A.
Abstract: We demonstrate a 40 Gb/s optical packet buffer with asynchronous and
autonomous control for use in contention resolution. Layer-2 (packet recovery)
measurements are presented with less than 1 dB power penalty up to 10 circulations.
©2006 Optical Society of America
OCIS Codes: (060.2330) Fiber optics communications, (060.4250) Networks
Optical packet switching provides a means of communication that is flexible, scalable, fast switching, and
high capacity . Optical buffers are a necessary component to resolve contention and congestion.
Practical buffers to date have relied on fiber delay lines that provide low-loss and large delays, but are fixed
in size. Most optical buffers can be characterized as either feed-forward or feed-back. Feed-forward
buffers commonly utilize wavelength routing schemes which lessen the number of active components, but
require many delay lines in order to achieve a variable delay. Feed-back buffers offer a large variable delay
with the use of few delay lines . Re-circulating loop buffers offer the advantage of being compact and
thus amenable to integration. They are commonly composed of only a switch and delay line. Several types
of switches have been developed and show promising results towards integration of optical buffers [3, 4].
Further integration of buffers to include on-chip delay lines has been proposed through the use of hybrid
integrated switches with low loss waveguides . A re-circulating loop buffer is studied because of its
potential for integration on chip and practical application in future optical routers. In this work, a re-
circulating buffer with asynchronous and autonomous control at a data rate of 40 Gb/s is investigated. A
dynamic range in excess of 20 dB and a power penalty less than 1dB is achieved up to 10 circulations,
corresponding to 1.92 µs of delay.
A 2x2 semiconductor optical amplifier (SOA) gate matrix is used here as the switching mechanism for a re-
circulating buffer. The major criteria that an optical buffer must meet is bit rate scalability up to and in the
future beyond 40 Gb/s, low crosstalk (<-40 dB) and high extinction (> 40dB) for cascadability, and fast
switching times (1-5 ns) . The 2x2 SOA gate matrix achieves all the aforementioned requirements but
has the disadvantage of introducing amplified spontaneous emission (ASE) which degrades the optical
signal to noise ratio (OSNR). This necessitates the use of a band pass filter (BPF) in the loop to suppress
noise. However, the filter cannot remove noise at the signal wavelength; therefore, the number of
circulations in a re-circulating buffer based on SOAs is limited by accumulated ASE. In order to increase
the number of circulations, 2R or 3R techniques must be incorporated in the loop.
This type of buffer can be incorporated into an optical packet switch based on optical label swapping .
A label is used to determine where to route packets while the payload information is not processed
electronically. This allows the payload to be at a high bit rate while the label can be at a lower bit rate for
the electronics. An important requirement of a packet switch is the ability to account for the asynchronous
nature between the optical data and electronic control plane . A major component to resolving this issue
is the payload envelope detector (PED) . The control plane uses the payload envelope to obtain a
precise time reference to the start and end of packets for control signal generation in the packet switch. A
PED and a field programmable gate array (FPGA) are used in this experiment to detect and forward
packets, completely autonomous and self-controlled. Contention resolution is demonstrated by using the
ptical buffer to avoid temporal collisions.
3. Experimental Setup and Results
A re-circulating buffer based on a 2x2 SOA gate matrix is depicted in Fig 1. Four SOAs were used as the
gates for the buffer and to provide gain to compensate for losses. Four 3dB splitters were placed on the
ports of the buffer and attenuators were used to balance gain and loss. In order to suppress accumulated
ASE, a BPF with a bandwidth of 2.4 nm was placed in the loop. The delay through any port configuration
of the switch, Tpath, is 120ns and the delay through the loop, Tloop, is 60 ns. The total delay of the buffer per
number of circulations can be given as
T (n) = (n+1)Tpath + nTloop (1)
Packets entering the buffer can either be placed in the loop by way of path13 or pass through the buffer via
path14. Once a packet is in the loop it can either re-circulate by way of path23 or exit via path24. Packets can
simultaneously circulate and pass through the buffer.
Fig. 1. Schematic of experimental setup
A schematic illustrating contention resolution is shown in Fig. 2. The setup consists of an optical buffer
and a contention path. The contention path simulates an input port configured to send traffic to the same
output port as the buffer. The relative delay of the buffer to the contention path is the same as stated in
Eqn. 1. Fig. 2(a) shows two transmitted packets, labeled 1 and 0. The transmitted signal is split equally in
power to the buffer and the contention path. Fig. 2(b) shows the packets located in the buffer and
contention path. The first packet, labeled 1, going to the buffer must circulate because there is a second
packet, labeled 0, it will collide with in the contention path. The second packet going to the buffer passes
straight through with a delay of T (0) since it will not collide with any packets. The packet in the loop exits
after one circulation, T (1), because there is no packet to contend with at the output. Fig. 2(c) shows
successful rearrangement of packets at the receiver labeled 1, 0, 0, and 1.
Fig. 2 (a) Packets from transmitter
In this experiment, 40 byte RZ packets at 40 Gb/s are used as data with 60 ns timeslots for switching.
The transmitted signal is tapped for electronic control processing. The PED is used to detect incoming
packets and generate packet envelopes. The envelopes enter an FPGA based controller board where they
are processed and the behavior of the buffer is determined. The FPGA board operates asynchronously to
the transmitter and therefore takes into account any relative drift between clocks. Moreover, it must also
cope with the fact that the data rate is not an integer multiple of the 100 MHz clock running the FPGA
board. On entering the FPGA board, the envelopes are expanded to 60 ns and aligned to the local clock
through an SR latch and a series of D flip-flops. Using delayed versions of the detected envelopes, the
controller determines how to forward packets. The processing time in the electronic control plane is
compensated for by adding 750m of fiber, corresponding to 3.75 µs delay. The transmitted signal is split
equally after the processing delay to the buffer and the contention path. An SOA is used in the contention
path to gate the incoming packets and thus avoid issues pertaining to intraband crosstalk. All packets that
are transmitted pass through the contention path. The packet stream is engineered to exercise all states of
the buffer. The first packet circulates once, the third twice, the sixth thrice, and so on until ten circulations.
Fig. 3(a) shows the first five packets of the input data stream. The packets are labeled according to the
number of circulations i.e. 0, 1 and 2 circulations. Fig. 3(b) shows the packets from the buffer and
contention path. The first packet, labeled 1, must circulate because a second packet, labeled 0, is located
120 ns later. The second packet can pass through the buffer since there is no packet 120ns later. Since
there is no packet 300ns after the first packet, the buffered packet can exit. The third packet, labeled 2,
(b) Packets in buffer and contention path
(c) Packets at receiver
must circulate twice because it is blocked by packets at 120ns and 300ns later. The fourth and fifth
packets, both labeled 0, pass through the buffer because they will not collide with other packets. The
packet labeled 2 exits the buffer 480 ns after it entered. Fig. 3(c) shows that the first five packets are
rearranged properly thus avoiding collisions.
Fig. 3(a) Packets from transmitter
Layer-2 (packet recovery) measurements were conducted at various input powers for the contention path
(CP) and 0, 2, 6, and 10 circulations. Fig. 4(a) shows results at low input power (-29 dBm). The signal
degrades as the number of circulations increases due to accumulated noise. Fig. 4(b) shows measurements
at an optimal input power (-24dBm). The power penalty is less than 1 dB at this input power because
neither noise nor saturation is a major issue. Fig. 4(c) shows results at high input power (-3 dBm).
Saturation distorts the signal as well as causes the gain to decrease in the SOAs which leads to poor signal
quality. Nonetheless, the buffer offers a large dynamic range in excess of 20 dB, which makes it suitable
for an optical packet switch.
(b) Packets from buffer and contention path (c) Packets at receiver
Fig. 4(a) Packet recovery at Pin= -29dBm
A fiber loop re-circulating buffer is investigated and less than 1 dB of power penalty is achieved at
40Gb/s for 10 circulations, equivalent to 1.92 µs of delay. A payload envelope detector and FPGA are used
asynchronous to the optical data to control the buffer and forward packets. Successful rearrangement of 65
ackets in a data stream with 195 timeslots is shown which demonstrates contention resolution.
his work is supported by DARPA/MTO and ARL under LASOR award #W911NF-04-9-0001.
 G.I. Papadimitriou et al., “Optical switching: switch fabrics, techniques, and architectures,” J. Lightw. Techno., vol. 21, no. 2,
pp.284-404, Feb. 2003.
 R. Langenhorst et al., “Fiber loop optical buffer,” J. Lightw. Technol., vol. 15, no. 3, pp. 324-335, Mar. 1996.
 E.F. Burmeister et al., “Integrated Gate Matrix for Optical Packet Buffering,” IEEE Photon. Technol. Lett.,, vol. 18, no. 1,
pp. 103-105, Jan. 2006.
 N. Chi, et al.,“A large variable delay, fast reconfigurable optical buffer based on multi-loop configuration and an optical crosspoint
switch matrix,” in Proc. OFC 2006, Paper OFO7.
 H.D. Park et al., “40-Gb/s optical buffer design and simulation,” Proc. 4th Int. Conf. Numerical Simulation of Optoelectronic
Devices 2004, Paper TuC2, pp. 19-20
 D. Wolfson et al., “All-optical asynchronous variable-length optically labeled 40 Gb/s switch,” in Proc. ECOC 2005, Paper Th.
 D. Wolfson et al., “Synchronizing optical data and electrical control planes in asynchronous optical packet switches,” in Proc.
OFC 2006, Paper OThM3.
 Z. Hu et al., “Optical label swapping using payload envelope detection circuits,” IEEE Photon. Technol. Lett., vol. 17, no. 7, pp.
1537-1539, July 2005.
(b) Packet recovery at Pin= -24 dBm (c) Packet recovery at Pin= -3 dBm