Semiconductor optical amplifier based nonlinear optical loop mirror with feedback: two modes of operation at high switching rates

Abstract

We describe the operating conditions under which a semiconductor optical amplifier based nonlinear loop mirror with optical feedback can have two stable modes of operation at switching rates faster than the gain recovery rate of the semiconductor optical amplifier. It can act either as an all-optical circulating shift register with inverter or as an all-optical clock divider, depending on the relative timing of the switching pulses. q 1998 Elsevier Science B.V. All rights reserved.

Full text

Page 1

1 December 1998
Ž.
Optics Communications 157 1998 45–51
Semiconductor optical amplifier based nonlinear optical loop
mirror with feedback: two modes of operation
at high switching rates
R.J. Manning
), A.E. Kelly, K.J. Blow, A.J. Poustie, D. Nesset
BT Laboratories, Martlesham Heath, B55, Room 131, Ipswich, Suffolk IP5 3RE, UK
Received 29 June 1998; revised 11 September 1998; accepted 11 September 1998
Abstract
We describe the operating conditions under which a semiconductor optical amplifier based nonlinear loop mirror with
optical feedback can have two stable modes of operation at switching rates faster than the gain recovery rate of the
semiconductor optical amplifier. It can act either as an all-optical circulating shift register with inverter or as an all-optical
clock divider, depending on the relative timing of the switching pulses. q1998 Elsevier Science B.V. All rights reserved.
Keywords: Modes of operation; Nonlinear optical loop; Switching rates
1. Introduction
Considerable progress has been made in recent years in
the field of all-optical processing with the use of semicon-
Ž
ductor optical amplifiers SOAs as non-linear optical ele-
ments in interferometric structures 1 . The SLALOM 3
w x
or TOAD 4 is an example of a nonlinear loop mirror
w x
configuration 2 having a SOA as the nonlinear material.
The low switching energy of these devices -1 pJ and
the ability to switch at repetition rates up to 100 Gbitrs
wx
1,5,6 has led to demonstrations of all-optical demultiplex-
Ž
w x.
ing e.g., Ref. 7and addrdrop multiplexing for telecom-
municationsapplicationsusing
pulses. In addition, more sophisticated all-optical process-
ing functions which use the TOAD as the basic switching
element have been reported recently. These include a
regenerative optical memory 8 and a circulating shift
wx
register with inverter 9,10 .
This paper describes the behaviour of a circulating shift
register with inverter architecture 11 using a SOA oper-
ated at high switching rates. Fig. 1 is a schematic of the
.
w xw x
Ž.
return-to-zerooptical
w x
wx
)Corresponding author. E-mail: bob.manning@bt-sys.bt.co.uk
experimental arrangement, which employs a TOAD with
optical feedback to realise the circulating shift register
with inverter functionality. The TOAD consists of a SOA
placed asymmetrically in a fibre loop mirror where the
offset of the SOA from the loop centre defines a temporal
w x
switching window 4 . A pulse input via the 50:50 coupler
into the loop is split into counter-clockwise and clockwise
travelling pulses, and these arrive at the SOA at different
times due to its offset from loop centre. In the absence of
switching pulses both pulses experience a similar phase
delay on passing through the SOA, recombine at the
coupler with no phase difference between them, and are
reflected. If a switching pulse is used to alter the refractive
index of the SOA, then one pulse suffers a different phase
lag compared to the other, and upon recombination of the
pulse pair at the coupler, the pulse may be transmitted.
With the feedback arrangement of Fig. 1, the reflected
Ž.
unswitched pulses from the loop are selected by a circu-
lator and amplified in an erbium doped fibre amplifier
Ž.
EDFA before being reinjected into the TOAD as orthog-
onally polarised switching pulses. These pulses cause the
loop to switch to transmission mode and so the feedback
loop empties of pulses. When the last fed-back pulse has
switched the SOA, the feedback loop is empty and the
0030-4018r98r$ - see front matter q 1998 Elsevier Science B.V. All rights reserved.
Ž.
PII: S0030-4018 98 00513-6

Page 2

()
R.J. Manning et al.rOptics Communications 157 1998 45–51
46
Fig. 1. Schematic diagram of the experimental arrangement. The loop mirror is biased for reflection, and pulses are fed back, amplified and
injected into the loop as switching pulses via polarisation selective couplers.
loop mirror reverts to reflection mode once more. The
temporal output hence contains alternating blocks of pulses
Ž.
‘ones’ , followed by blocks of ‘zeros’.
A key parameter determining the overall behaviour of
this all-optical architecture is the ratio of the recovery time
of the SOA to the pulse repetition rate 12 . For optical
pulses whose repetition rate is less than the gain recovery
rate of the SOA, the ‘block behaviour’ described above
occurs. However, if the repetition rate exceeds the SOA
recovery rate, then spontaneous clock division can occur,
where the output pulse stream is at a repetition rate equal
w
to half the input rate 12,13 . The clock division arises
because of a combination of the non-linear gain saturation
in the SOA and the memory due to the optical feedback.
The first switching pulse in the feedback sequence experi-
ences gain in an unsaturated SOA. If the repetition rate of
the pulse train is faster than the recovery rate of the SOA,
then the second switching pulse experiences less gain, and
Ž
so imparts a different smaller phase change on the corre-
sponding signal pulses in the TOAD. The differential
phase change produced by the SOA gain saturation is
enhanced on each circulation of the shift register and
propagates through the ‘block’ until every second pulse in
the block is either switched or not. This can result in pure
clock division of the pulse train if the feedback loop
contains an odd number of pulses 13 .
In this paper we describe the circulating shift register
with inverter architecture operating at switching rates faster
than the SOA recovery rate and find that it can be config-
ured to support both ‘block’ and ‘clock division’ modes of
wx
x
.
wx
operation, depending on the relative arrival time of the
switching pulses.
2. Theoretical background
We model the basic behaviour of the SOA being
switched at repetition rates exceeding the SOA lifetime
with the simplest rate equation model for the carrier
density in the conduction band, N, at point z from the
input facet of the SOA, at time t 3 :
w x
EN z,t
Ž
Et
IN z,t
Ž
t
.
hnA
..
sy
eAl
P z,t G g
Ž
y
N z,t yN
Ž
,1
.Ž .
Ž.
T
where I is the effective injection current, e is the unit
electric charge, l is the SOA length, t is the carrier
lifetime, P is the optical power, G is the mode confine-
ment factor, g is the gain coefficient, hn is the photon
energy, A is the cross-sectional area of the active region,
and N
is the carrier density at transparency. We ignore
T
high speed effects such as carrier heating and spectral hole
burning, which have very small contributions for pulses
longer than 1–2 ps. The equilibrium carrier density N in
the absence of any optical pumping, is given by:
0
N sItreAl.
0
2 Ž .

Page 3

()
R.J. Manning et al.rOptics Communications 157 1998 45–51
47
Ž .
This may be used to express Eq. 1 in the following
form:
wx
Ž.Ž
0
sy
Et
t
ENN yN z,tP z,t G g
hnA
.
N z,t yN
Ž
.
.
Ž.
T
3 Ž .
The optical power P varies with distance z down the
SOA according to:
EP z,t
Ž
Ez
.
s gG NyN
w
ya
P z,t ,
x Ž
4
.Ž .
Ž.
T
int
where a
We use these equations to calculate the phase shift
imparted on the two counterpropagating signal pulses in
the loop by the switching pulse in the SOA. In semicon-
ductors, the refractive index change Dn caused by a
change in carrier density DN via electron-hole recombina-
tion is given by:
is the internal waveguide loss coefficient.
int
Dnsn DN,5 Ž .
eh
where n
Ž
;2=10
by a pulse due to this carrier density change is given by:
is the refractive index change per carrier pair
y20
cm. Hence the phase shift experienced
eh
y3.
DFskn DNl,6 Ž .
eh
where l is the length of the SOA, and k is the wavevector.
In our modelling, we assume the SOA has a lifetime of
80 ps, a cross-sectional area A of 0.2 mm2, and an a
y1
30 cm . The 3 dB saturation energy E s Ahn rG g,
is taken to be 2 pJ. The numerical calculation accounts for
length effects in the 1 mm long SOA by dividing the SOA
into 20 sections, each 50 mm long. The phase changes are
accompanied by gain changes which affect the relative
amplitudes of the pulse pair recombining at the coupler.
The amplitude modulation is typically 3 dB 14 , reducing
the expected contrast ratio to ;15 dB, which is observed
experimentally.
Fig. 2 shows the phase change produced by an input
switching pulse train at 10 GHz having individual pulse
energies of 100 fJ and pulse durations of 5 ps. Note that
the initial switching pulse causes a larger phase change
than subsequent pulses, due to the initial unsaturated state
of the SOA. The SOA lifetime is too long to allow full
recovery before the second switching pulse arrives, and
quasi equilibrium state is quickly established within 1–2
.
pulses , where the SOA is always quasi-saturated. It is this
initial difference in phase change at the beginning of the
switching sequence which can lead to clock division. It is
very important to distinguish two operating modes for the
TOAD, which are defined by the arrival time of the
switching pulses with respect to the pulse pairs in the loop
w x
3 . We show the two regimes in Fig. 2 by loop signal
Ž
pulse pairs closed and open circles placed at the appro-
priate arrival times relative to a fixed switching phase
of
int
Ž.
sat
wx
Ž
Ž.
.
Fig. 2. Phase evolution induced on the signal pulses in a TOAD
with the SOA subject to a 10 GHz switching pulse train. The two
possible switching windows are shown in a and b . a Window
1. The counter-clockwise travelling loop pulse full circle arrives
before the switching pulse, and the clockwise travelling loop pulse
Ž.
open circle arrives after. The first four pulse pairs are shown. b
Window 2. Both pulse pairs arrive after the switching pulse. The
first three pulse pairs are shown. In both cases, there is approxi-
mately a p radians phase difference induced between the signal
pulses, except for the initial phase difference for the first pulse
pair in the window 1 regime, which is different from all the rest.
Ž . Ž . Ž .
Ž.
Ž .
Ž
response. We assume a switching window of 40 ps an
offset of the SOA from loop centre of 20 ps . One
operating regime, window 1, is shown in Fig. 2 a , where
Ž
counter-clockwise closed circle and clockwise open cir-
.
cle loop pulses arrive either side of the switching pulse,
and their phase difference is determined by the phase
change caused by the switching pulse. The other regime,
Ž .
window 2, shown in Fig. 2 b , is where both pulses arrive
after the switching pulse, and their phase difference is
simply determined by the recovery rate of the SOA. We
use the phase difference between loop pulses to calculate
the loop power reflection coefficient R, where
Ž.
and DF is the relative phase between loop pulses 12 .
The experimentally measured contrast ratios of ;15 dB
Ž .
justify not modifying Eq. 7 to account for amplitude
modulation effects in the loop. The two timing regimes
.
Ž .
Ž.
Rs0.5 1qcos DF ,7
Ž .
wx

Page 4

()
R.J. Manning et al.rOptics Communications 157 1998 45–51
48
Fig. 3. Transmission of the loop for the two switching windows.
Ž . a In window 1, the transmission for the first pulse is different to
Ž .
that for subsequent pulses. b In window 2, all pulses experience
a similar transmission coefficient.
give two different reflection and hence transmission val-
ues, as shown in Fig. 3. The evolution of the all-optical
shift register dynamics depends upon which of the two
windows is used. If window 1 is used, Fig. 3 a , then the
loop transmission caused by the initial switching pulse is
significantly different to that induced by subsequent
switching pulses. This difference is remembered by the
system because of the feedback, and critically determines
the system evolution. This ultimately leads to a clock
divided output, as discussed above. However, if window 2
is used then the transmission induced by each switching
ŽŽ ..
pulse is very similar Fig. 3 b , and the evolution of the
system is not affected by the initial switching pulse. The
stable output consists of a block of pulses, followed by a
block of zeros, similar to the temporal behaviour when the
SOA fully recovers between switching pulses as described
previously. However in this regime, at the end of the
switching sequence when the feedback path has emptied,
further signal pulses may still be switched out of the loop.
This can occur if the SOA has a slow recovery rate
compared to the bit rate. In this case, signal pulses can
experience a substantial differential phase shift due to the
gain recovery of the SOA which continues in the absence
of switching pulses.
We numerically modelled the all-optical circulating
shift register with inverter architecture operating in these
two switching regimes. The initial reflected signal pulse
train was input to the SOA as a new train of switching
pulses and subsequent values of R were re-calculated by
ŽŽ ..
Ž .Ž .
Fig. 4. Results of a numerical model showing the output of the shift register after 30 circulations. a Input block of five pulses; b clock
Ž. Ž .Ž.
divided output window 1 ; c output block window 2 .

Page 5

()
R.J. Manning et al.rOptics Communications 157 1998 45–51
49
Ž .Ž .Ž
Fig. 5. Experimental results for 10 GHz repetition rate input pulses. a Input pulses, timescale: 50 psrdiv. b Clock divided output using
. Ž .Ž.
window 1 , timescale: 50 psrdiv. c Block output using window 2 , timescale: 500 nsrdiv. Blocks are approx. 300 ns long, and individual
Ž .
pulses are not resolved. d Eye diagram for block mode. Timescale 20 psrdiv.
re-calculating N and DF. This sequence was repeated 30
times, corresponding to 30 circulations around the feed-
back path in the shift register, so that the steady state
behaviour was obtained. Fig. 4 shows the numerical results
of operating in the two switching regimes. Only the peak
Ž
pulse amplitude is plotted as a square in the graphs. Fig.
Ž .
4 a shows the input block of 5 switching pulses at 10
GHz used in each calculation. For switching in window 1,
a clock divided output is obtained as shown in Fig. 4 b .
For switching in window 2, a stable block output is
.
Ž .
obtained but note that the block output has some amplitude
modulation on the first pulse.
3. Experimental
We experimentally verified the two-mode behaviour of
the all-optical circulating shift register with inverter at high
switching rates. The optical pulse source was derived from
Ž
a tuneable continuous wave cw laser which was modu-
.
Ž . Ž .Ž .
Fig. 6. Experimental results for 20 GHz pulses. a Input pulses, timescale 20 psrdiv; b clock divided output, timescale 20 psrdiv; c
Ž .
block output, timescale: 200 nsrdiv; d eye diagram for block mode, timescale 20 psrdiv.

Page 6

()
R.J. Manning et al.rOptics Communications 157 1998 45–51
50
Fig. 7. RF spectrum analyser trace of block output for a 20 GHz
input train.
Ž. wx
lated using electro-absorption modulators EAMs
provided a low jitter and low noise pulse train at a
wavelength of 1558 nm with a pulse width of 4.5 ps, and
was run at both 10 GHz and 20 GHz repetition rates. The
pulse train was amplified in an EDFA and injected into a
TOAD in which the 1 mm long SOA was offset from loop
centre to give a switching window of ;40 ps for the 10
GHz pulse train, and ;25 ps for the 20 GHz pulse train.
The offset of the SOA was controlled using a variable
optical delay stage. The SOA was designed to have a
larger band-gap than usual, having a gain peak at 1500 nm
wx
16 at a bias current of 200 mA. The SOA was kept at a
constant temperature of 258C using a Peltier cooler, and
was run at 300 mA in most experiments. It had an
alpha-factor of ;9 at a wavelength of 1558 nm and a
wx
lifetime of ;80 ps 14 . The TOAD was biased for
reflection and the reflected pulses were fed back as orthog-
onally polarised switching pulses using a circulator. A
second EDFA amplified the pulses to the required switch-
ing energy of ;300 fJ per pulse at the input facet of the
Ž
SOA there was ;4 dB coupling loss at each facet, giving
an input switching energy of ;100 fJ, as used in the
.
modelling . The energy of the loop pulses the switched
.
pulses were between 10–20 fJ. The arrival time of the
switching pulses was controlled using another variable
optical delay stage having a range of 200 ps, i.e., two pulse
periods at 10 GHz, four pulse periods at 20 GHz. The
exact arrival time was important for two reasons: a it
determined which of the two switching windows were
used and hence the mode of operation of the system. b
Pure clock division only occurs when there is an odd
number of pulses in the shift register 12,13 .
The temporal output of the shift register was observed
with a 35 GHz photodiode on a 50 GHz sampling oscillo-
scope and a 1 GHz realtime oscilloscope.
15 . It
Ž
Ž .
Ž .
wx
4. Results
Ž .
Fig. 5 a shows the input 10 GHz pulse train, and Fig.
Ž .
5 b the clock divided output at 5 GHz, where the switch-
ing pulses are timed to arrive in between the loop pulses.
The block output mode, as shown from the realtime
Ž .
oscilloscope trace in Fig. 5 c , was observed by retarding
the switching pulse arrival time by 25 ps. Fig. 5 d shows
the eye diagram obtained when the system ran in block
mode and the measured switching contrast was )15 dB.
This was an extremely stable mode of operation, once
established. It required initially a slightly higher switching
Ž
energy ;400 fJ per pulse before the input facet of the
.
SOA than the clock division mode, consistent with the
fact that slightly lower transmission is obtained in the
Ž.
second window Fig. 3 . Similar behaviour was also ob-
served at 20 GHz, where the input pulse train at 20 GHz
and clock divided output at 10 GHz are shown in Fig. 6 a
Ž .
and b . Retarding the switching pulse arrival time by 25
ps again gave block mode behaviour, as shown in Fig.
Ž .
6 c . The eye diagram for the blocks is shown in Fig. 6 d
and the measured switching contrast was )10 dB. Both
modes of operation were stable to a "12 ps variation in
switching pulse arrival time, and could be observed for
many hours in the laboratory without the need for active
stabilisation. Fig. 7 shows the microwave spectrum anal-
yser trace of the block output for a 20 GHz input pulse
train. It has the expected frequency spectrum for a square
wave of period ;600 ns.
Ž .
Ž .
Ž .
5. Conclusions
We have shown that the all-optical circulating shift
register with inverter architecture is capable of running in
two distinct stable modes, even when the repetition rate of
the switching exceeds the SOA gain recovery time. It was
possible to experimentally observe clock division and
‘block’ modes at optical pulse repetition rates up to 20
GHz because of the short carrier lifetime ;80 ps of the
SOA used in the TOAD. We expect that the operational
principles described here should be extendible to much
higher pulse repetition rates - 100 GHz by using an
wx
optical holding beam 17 to modify the effective gain
lifetime of the SOA.
Ž.
Ž.
References
w x 1 R.J. Manning, A.D. Ellis, A.J. Poustie, K.J. Blow, J. Opt.
Ž.
Soc. Am. B 14 1997 3204.
w x
2 N.J. Doran, D. Wood, Optics Lett. 13 1988 56.
w x 3 M. Eiselt, W. Pieper, H.G. Weber, J. Lightwave Technol. 13
Ž.
1995 2099.
w x 4 J.P. Sokoloff, P.R. Prucnal, I. Glesk, M. Kane, IEEE Photon-
Ž.
ics Technol. Lett. 5 1993 787.
w x 5 K.L. Hall, K.A. Rauschenbach, Paper PD5-1 Optical Fibre
Communications Conference, 1998.
w x
6 R.J. Manning, G. Sherlock, Electron. Lett. 31 1995 307.
w x 7 R. Hess, M. Caraccia-Gross, W. Vogt, E. Gamoer, P.A.
Ž.
Ž.

Page 7

()
R.J. Manning et al.rOptics Communications 157 1998 45–51
51
Besse, M. Duelk, E. Gini, H. Melchior, B. Mikkelsen, M.
Vaa, K.E. Jepsen, K.E. Stubjkaer, S. Bouchoule, IEEE Phot.
Ž.
Technol. Lett. 10 1998 165.
w x 8 A.J. Poustie, K.J. Blow, R.J. Manning, Optics Commun. 140
Ž.
1997 184.
w x 9 A.J. Poustie, R.J. Manning, K.J. Blow, Electron. Lett. 32
Ž.
1996 1215.
x
10 K.L. Hall, J.P. Donnelly, S.H. Groves, C.I. Fennelly, R.J.
Bailey, A. Napoleone, Optics Lett. 22 1997 1479.
wx
11 N.A. Whitaker Jr., M.C. Gabriel, H. Avramopoulos, A.
Ž
Huang, Optics Lett. 16 1991 1999.
w
Ž.
.
wx
12 K.J. Blow, R.J. Manning, A.J. Poustie, Optics Commun. 134
Ž.
1997 43.
wx
13 R.J. Manning, A.J. Poustie, K.J. Blow, Electon. Lett. 32
Ž.
1996 1504.
wx
14 R.J. Manning, A.E. Kelly, A.J. Poustie, K.J. Blow, Electron.
Ž.
Lett. 34 1998 916.
wx
15 D.D. Marcenac, A.D. Ellis, D.G. Moodie, Electron. Lett. 34
Ž.
1998 101.
wx
16 A.E. Kelly, D.D. Marcenac, D. Nesset, Electron. Lett. 33
Ž.
1997 2123.
wx
17 R.J. Manning, D.A.O. Davies, Optics Lett. 19 1994 889.
Ž.