web viewwavelength division multiplexing (wdm) give each message a different wavelength ... the...
TRANSCRIPT
CONTENTS
1
Chapters Page
nos.
1. INTRODUCTION 1
1.1 Different types multiplexing techniques 2
1.1.1 Frequency Division Multiplexing 2
1.1.2 Time Division Multiplexing3
1.1.2.1 Synchronous Time Division Multiplexing 3
1.1.2.2 Statistical Time Division 3
1.1.3 Wavelength Division Multiplexing (WDM) 4
1.1.4 Code Division multiplexing(CDM) 4
2. PRINCIPLES OF OPTICAL TIME-DIVISION 5
MULTIPLEXING AND DEMULTIPLEXING
2.1Electrical and Optical Multiplexed Systems 5
2.2. Optical Time-Division Multiplexing 7
2.3. Optical Switching and Demultiplexing 12
2.4 Timing Recovery 19
3 .EXPERIMENTS 20
3.1Multiple-Laser Systems 21
3.2 Single-Laser System 31
4 CONCLUSIONS AND OUTLOOK 32
5.REFERENCES 35
2
Chapetr-1
1. IntroductionProgress in very high bit-rate light wave systems has been stimulated by consistent demands for
expanded transmission capacity. These demands have led to increased interest in multi giga bit
per-second pulse-code modulated (PCM) systems, and have put an emphasis on the need for
high-speed and wide-band electronics in light wave transmitters and receivers. To date, it has
generally been possible to meet these needs with high-speed Si and GaAs circuits, but at gigabit-
per-second bit rates it becomes increasingly difficult to develop the necessary digital electronic
circuits. One method to relieve this electronic speed bottleneck is to extend the well-known
techniques of electrical multiplexing into the optical domain. The two main approaches to optical
multiplexing are optical wavelength-division (or frequency-division) multiplexing and optical
time-division multiplexing. This presentation concentrates on optical time division multiplexing.
In optical time-division multiplexing (OTDM), a high bit-rate data stream is constructed directly
by time-multiplexing several lower bit-rate optical streams. Similarly, at the receiver end of the
system, the very high bit-rate optical signal is demultiplexed to several lower bit-rate optical
signals before detection and conversion to the electrical domain. This approach to optical time-
division multiplexing and demultiplexing moves the demand for high-speed performance away
from electronic devices such as transistors, and places it on optical and optoelectronic devices
such as pulsed semiconductor lasers and optical switches. The time-division multiplexing
approach is a purely digital technique and is therefore compatible with the concept of an
all-digital network that combines switching and transmission. In addition, optical time-
division multiplexing offers system design flexibility, including the possibility of adjustable
bandwidth allocation in different baseband channels and the possibility of simple system
hardware in which only a single transmitter laser is required for all channels. The potential of
optical time division multiplexing and demultiplexing for very high bit-rate PCM systems
has been recognized for more than two decades but until recently there have been few
system-level demonstrations of the technique at multi gigabit-per- second bit rates. The
implementation of very high bit-rate. OTDM systems has been slow because electronic
multiplexing has usually served adequately and because the necessary hardware, such as high-
3
speed optical switches and compact pulsed semiconductor lasers, has only recently reached a
sufficient state of refinement. This paper describes recent experiments in optical time-division
multiplexing and demultiplexing that have been made possible by improvements in lasers and
switch /modulators. Our emphasis is on very high bit-rate point-to-point transmission systems
but many of the concepts are also relevant to multiuser systems and time-multiplexed photonic
switching networks. We review system architectures, describe the requirements on individual
system components, and give examples of transmission system experiments operating at bit rates
up to 16 Gbit /s. One of the key factors affecting the performance of multiplexed systems is
crosstalk between baseband channel . In presentation explore sources of crosstalk in optical time-
division multiplexing and demultiplexing, and describe how the main system components affect
the overall crosstalk performance. Under the simplest conditions, a medium can carry only one
signal at any moment in time. For multiple signals to share one medium, the medium must
somehow be divided, giving each signal a portion of the total bandwidth.The current techniques
that can accomplish this include
1.1 Different types multiplexing techniques
1.frequency division multiplexing (FDM)
2.Time division multiplexing (FDM)
3.wavelength division multiplexing (WDM)
4.code division multiplexing (CDM)
1.1.1Frequency Division Multiplexing
Assignment of non-overlapping frequency ranges to each “user” or signal on a medium. Thus,
all signals are transmitted at the same time, each using different frequencies.A multiplexor
accepts inputs and assigns frequencies to each device. The multiplexor is attached to a high-
speed communications line.A corresponding multiplexor, or demultiplexor, is on the end of the
4
high-speed line and separates the multiplexed signals.
1.1.2Time Division Multiplexing:
Sharing of the signal is accomplished by dividing available transmission time on a medium
among users.Digital signaling is used exclusively.
Time division multiplexing comes in two basic forms:
1. Synchronous time division multiplexing, and
2. Statistical, or asynchronous time division multiplexing
1.1.2.1 Synchronous Time Division Multiplexing
The original time division multiplexing.The multiplexor accepts input from attached devices in a
round-robin fashion and transmit the data in a never ending pattern.T-1 and ISDN telephone
lines are common examples of synchronous time division multiplexing.
1.1.2.2 Statistical Time Division:
MultiplexingA statistical multiplexor transmits only the data from active workstations (or why
work when you don’t have to).If a workstation is not active, no space is wasted on the
5
multiplexed stream.A statistical multiplexor accepts the incoming data streams and creates a
frame containing only the data to be transmitted. A statistical multiplexor does not require a line
over as high a speed line as synchronous time division multiplexing since STDM does not assume all
sources will transmit all of the time!Good for low bandwidth lines (used for LANs).
1.1.3Wavelength Division Multiplexing (WDM)
Give each message a different wavelength (frequency)Easy to do with fiber optics and optical
sources, Dense wavelength division multiplexing is often called just wavelength division
multiplexing .Dense wavelength division multiplexing multiplexes multiple data streams onto a
single fiber optic line. Different wavelength lasers (called lambdas) transmit the multiple
signals.Each signal carried on the fiber can be transmitted at a different rate from the
othersignals.Dense wavelength division multiplexing combines many (30, 40, 50, 60,
more?)onto one fiber.
1.1.4 Code Division Multiplexing (CDM)
Also known as code division multiple access (CDMA),An advanced technique that allows
multiple devices to transmit on the same frequencies at the same time using different codes .Used
for mobile communications. time. Each mobile device is assigned a unique 64-bit code (chip
spreading code)
6
Chapter-2
2. PRINCIPLES OF OPTICAL TIME-DIVISION
M UL TIPLEXING AND DEMUL TIPLEXING
2.1Electrical and Optical Multiplexed Systems
The basic principle of time-division multiplexing and demuItiplexing is that each of the
baseband data streams is allocated a series of time slots on the multiplexed channel.
A multiplexer (MUX) assembles the h higher bit-rate bit stream from the baseband streams and
a demultiplexer(DEMUX) reconstructs bit streams at the original lower bit rate by separating bits
in the multiplexed stream . The techniques for this process are well established for electrical
time-division multiplexing and demultiplexing but are only now emerging in optical
systems .Fig. I highlights the similarities and differences between electrically time-multiplexed
and optically time multiplexed light wave systems . In this figure and in subsequent figures ,
thick lines are used for optical (fiber) signal paths and thin lines are used for electrical signal
paths .In an electrically time-multiplexed system , Fig . lea),multiplexing is carried out in the
electrical domain , before the electrical-to-optical (E/O) conversion . Demultiplexing is carried
out after the optical-to-electrical (0 / E) conversion. For n baseband channels, each of bit rate B,
the multiplexed bit rate is nB. Potential electronic bottle necks occur in the MUX and the E /0
converter, and in the 0/ E converter and the DEMUX , where the electronics must operate at the
full multiplexed bit rate . These bottlenecks arise from a) speed limitations of digital integrated
circuits, b) speed limitations of high-power and low-noise linear amplifiers used to drive the
laser or modulator in the E/O converter and in the O/E converter , c) limited modulation
bandwidths of lasers and modulators , and d)the fact that the receiver sensitivity offered by an
avalanche photodiode degrades by more than 3 dB for every octave increase in receiver
bandwidth. These problems have so far limited the maximum b it rate for electrically
multiplexed systems to 10 Gbit/s. In the optically multiplexed system, Fig. l (b) , the electronic
bottlenecks are removed by moving the E / 0 and o/ E converters (i.e, the transmitters and
receivers) into the baseband channels. Multiplexing is carried out after the E /0 conversion and
7
demultiplexing is carried out before the O/E conversion. All electronics associated with signal
processing operate only at the baseband bit rate .Note that a control signal is needed to drive the
demultiplexer. In general , this control signal could be either electrical or optical , depending on
the demultiplexer technology. At p resent the most practical optical demultiplexers are based on
electrooptic switches, which use electrical control signals . It will be shown later that the
bandwidth of this electrical control signal need not be large for demultiplexing in an OTDM
system. An important difference between electrically multiplexed and optically multiplexed
systems is that in electrical systems the multiplexing and demultiplexing can be carried out at
points in the system where the signal has been amplified to large levels . The signal-to-noise
ratio is determined by the receiver and its associated low-noise front end,and is not affected by
loss in the multiplexing or demultiplexing operations. In an OTDM system, on the other hand ,
the multiplex ing and demultiplexing is carried out on the optical signal, Thus optical losses
reduce the signal level relative to the receiver noise, and losses must be kept small . It may be
feasible, however, to place optical amplifers at selected points in the system to compensate for
some of these losses .
8
2.2. Optical Time-Division MultiplexingIn this section we consider optical waveforms and timing requirements for optical multiplexing,
and examine topologies for transmitters and multiplexers. The operation of time-multiplexing
several lower bitrate baseband channels onto a higher bit-rate channel can be divided into three
sub functions: sampling, timing, and combining. The sampling function takes samples of the
incoming baseband data stream, thereby identifying the value of each incoming bit. The timing
function ensures that the samples are available at the correct time slots on the multiplexed
9
channel. The combining function assembles all the sampled baseband data streams to generate
the higher bit-rate multiplexed data stream. In multi gigabit-per-second electrically multiplexed
systems it is convenient to sample each of the input data streams using short sampling pulses that
are timed to correspond to the appropriate time slots on the multiplexed bit stream. If the
sampling pulse widths are less than one bit-period of the high bit-rate multiplexed signal, the
combiner can be a simple summing circuit. A similar strategy is a preferred approach to optical
time-division multiplexing because it can capitalize on mode-locked and gain switched
semiconductor lasers, which are capable of generating pulses more than ten times shorter than
electrical pulses. In this approach to multiplexing the sampling function is carried out in the E /0
converters (i.e., the system transmitters). Consequently, the mux in Fig. l(b) is required only to
do the combining function. The following concentrates on this method of multiplexing. Fig. 2
shows the schematic of an E /O converter (transmitter) that can be used to sample the input data
before optical combining.
Short optical pulses from a laser are incident on an optical modulator, which is driven by
an input electrical data stream. The electrical data stream could be either in the return-to-zero
(RZ) or the non-retum-to-zero (NRZ) format, but NRZ is usually preferable because it minimizes
the bandwidth requirements of the baseband digital electronics, the modulator, and its drive
amplifiers. The optical pulse train from the laser samples the electrical input data via the
modulator, thereby converting it from NRZ in the electrical domain to RZ in the optical domain.
An important feature of the laser-modulator combination in Fig. 2 is that when the laser and
input data are correctly timed, the modulator is either fully "on" or fully "off" when the optical
pulse passes through it. This means that the modulation process does not cause the optical signal
to chirp. Sampling the baseband data in the E / 0 converter enables the optical combining
function to be carried out using a passive device such as a fiber directional coupler power
combiner. Another advantage of this approach is that independent of the type of combiner used
in the system, the RZ output format from the E / 0 converters leads to low multiplexing crosstalk
(see below). Furthermore, the optical power from the lasers in the E /0 converters is used
efficiently since the signal in each baseband channel is zero during time slots to be occupied by
other channels in the multiplexed data stream. This is a significant practical consideration
because semiconductor lasers are average-power limited devices. The timing scheme for a
general n-channel optical time division multiplexed system is shown in Fig. 3. The n optical
10
signals incident on the combiner are RZ pulse trains with repetition rates B and with pulsewidths
T(measured at the baseline). The incoming bit streams are temporally offset from one another by
delays D. Data are encoded on each pulse train, before combining, so in general some of the
individual pulses will have zero amplitude. However, in Fig. 3 all bits are shown as "ones" for
clarity. When the pulse spacing in Fig. 3 is adjusted for maximum multiplexed hit rate, each
pulse in the multiplexed bit stream just comes into contact with its nearest neighbors. Under
these circumstances D = T, and the multiplexed bit rate is 1 / T. With laser pulses that are 10 ps
wide at the baseline, for example, the multiplexed bit rate could be as high as 100 Gbit / s.
Fig. 3 shows that the RZ signal format provides low system crosstalk. Since each baseband
signal is always nominally zero except in its allotted time slot on the multiplexed bit stream, it
cannot interfere with other channels. In practice, the pulse stream will have a finite on / off ratio
and the resulting baseline light signal between pulses will cause a component of crosstalk.
Furthermore, leading and trailing tails on the pulses will also cause cross talk if the pulses are
spaced such that tails overlap on adjacent pulses.
To avoid possible overlap problems it is usually necessary to ensure that the pulses are somewhat
shorter than the one bit period of the multiplexed bit stream. It is shown in Section II-C below
that shorter pulses also help to reduce crosstalk in the demultiplexer. Since shorter pulses occupy
a wider optical spectrum, reducing the pulse width may increase the dispersion penalty, even if
the lasers operate at or near the wavelength of zero first order chromatic dispersion in the fiber.
Thus, choosing the optimum pulse width may entail a compromise between system crosstalk and
pulse spreading caused by fiber dispersion. For example, in local applications requiring very
high bit rates over short lengths of fiber, optimum system performance would be obtained by
minimizing crosstalk. This could be achieved using optical pulses that are significantly shorter
than the multiplexed bit period. In longer distance applications, on the other hand, it may be
necessary to put more emphasis on minimizing the dispersion penalty rather than crosstalk. Thus
a long distance system might achieve optimum performance with longer laser pulses. Block
schematics showing two possible configurations for the E /0 converters (transmitters) and
combiner for an n-channel OTDM system are presented in Fig. 4. The first configuration, Fig.
4(a), uses n optical pulse generators, all driven by the same master clock. These pulse generators
could be mode-locked or gain-switched [semiconductor lasers. The pulse streams are delayed
with respect to one another using delay elements either in the electrical clock paths (as shown
11
here) or in the optical signal paths. Electrical delays would usually be preferred because they are
easily made adjustable. Data encoding and sampling can bc carried out using optical modulators
such as waveguide electro
optic devices at the outputs of the pulse generators. An alternative to using external modulators
would be to directly encode data on gain-switched lasers. The data-encoded pulse streams in Fig.
4(a) are brought together in the combiner. A potential disadvantage of the transmitter- combiner
arrangement in Fig . 4(a) is that the wavelengths of all lasers need to be closely matched to avoid
pulse overlap at the receiver end of the fiber caused by different propagation times.
12
Fig. 4. Schematics of two configurations for E/O converters and combiner.
(a) Multiple optical pulse generators. (b) Single optical pulse generator.
One approach to circumvent this problem is to use an active control mechanism that detects
pulse overlap at the receiver end of the system and continuously adjusts the timing of the
transmitter pulses accordingly . This type of control scheme could also be used to adjust the
timing of bit streams originating in different locations.The second OTDM transmitter
configuration, Fig. 4(b),uses a single optical pulse generator . The output of the generator is split
passively into 11 channels, which are then encoded with data and properly delayed with respect
to one another. This arrangement requires only one laser a feature that cannot be achieved with
other optical multiplexing methods such as wavelength-division multiplexing. In addition, all
channels operate at precisely the same optical frequency and will therefore propagate through a
fiber with the same delay. The main disadvantage of the scheme in Fig. 4(b) is that the total
13
transmitted power will be less than in Fig. 4(a) because there is only one laser. However, the
transmitted power level could be increased using optical amplification .Up to this point, we have
assumed that the baseband modulation in both systems in Fig. 4 is synchronous with the pulse
stream from the lasers. This is desirable for best performance in very high bit-rate systems but
the requirement of synchronous operation can be removed if the modulation bit rate is
substantially less than the repetition rate of the lasers . Thus, one can trade off capacity for
increased system flexibility. Asynchronous operation is particularly attractive for systems where
the baseband channels originate in different locations. The optical combiner in Fig . 4(a) and (b)
can, in general, be either passive or active. A passive combiner would incorporate devices such
as fiber directional couplers, while active combining would use active devices such as optical
switches .
The passive combiner option has the advantage of simplicity but its losses can become
large because each 3-dB directional coupler introduces 3-dB combining loss. The losses in an
active combiner should, in principle, be smaller because the loss per device is potentially low.
An integrated array of switches on a single chip promises the lowest combining loss for systems
with more than two baseband channels. In either case, optical amplification may counter the
power penalties incurred. An active combiner has the potential of reducing crosstalk, by
decreasing pulse overlap caused by finite on I off ratios and leading and trailing tails on the
pulses .In a passive combiner, a tail extending into the time slot of a neighboring bit would be un
attenuated, but in an active combiner it would be reduced by the (time-dependent)switching
function . In addition to combining the baseband signals, an active combiner could also perform
the sampling function required in optical multiplexing. An active multiplexer of this type could,
in principle, eliminate the need for pulsed lasers. However, CW laser operations would place
very stringent requirements on the switch speed and extinction if low crosstalk is to be achieved.
In addition, CW operation would not be efficient because of the average power limit of
semiconductor lasers and the low duty cycle of each baseband optical signal.
2.3. Optical Switching and Demultiplexing
The demultiplexer is the most critical element of an OTDM system. Its purpose is to
direct each bit of the arriving multiplexed bit stream to the appropriate O/E converter, as shown
14
in F ig . l (b).For good system sensitivity , it is necessary that most photons from each in coming
bit are transferred to the appropriate 0 /E converter. Thus the optical demultiplexer should switch
the entire bit rather than just sample part of it. In addition, it is important that crosstalk between
channels is small. The basic building block of an optical demultiplexer is a 1 x 2 switch.
Typically, optical switches such as directional coupler devices are fabricated as2 x 2 cross points,
which can serve as I X 2 switches by simply terminating one port. However, waveguide devices
specifically designed as 1 X 2 switches also are feasible. We consider here a model of analog
1X2 optical switch and the functional switching properties that influence the design of the
demultiplexer. In Section III we will give specific details of the switches used in our system
experiments. Fig. 5 shows the optical and electrical connections to a 1x 2 switch and the form of
the switching characteristic as a function of the applied electrical control voltage V. For
simplicity, the excess loss of the switch is taken to be zero but nonzero loss can be included by
scaling the output power axes. When used as a demultiplexer, the input power PI is applied at
port I and the output powers P2 and P3 appear at ports 2 and 3 . In general, the output powers are
nonlinear functions of the control voltage V. The characteristics shown here are for a reactive
switch, such as a directional coupler switch, where all the input power is transmitted to the
outputs and is not intentionally dissipated in the device. The switching characteristics can be
written as
where f( V) is the (nonlinear) switching function. The switch is an analog device, which means
that the optical power at each output port changes smoothly and continuously with control
voltage. There is significant power at both outputs for most of the control voltage range of
interest and the output powers are equal when V = V-3dB• We show below that this analog
switching characteristic strongly influences crosstalk. The switching function in Fig. 5 reaches its
maximum and minimum values at V +and V _, respectively; for optimum switch performance
the control voltage should swing dynamically between these voltage limits. Note that, in general,
15
the switch extinction is not perfect, even at V = V+ and V = V_ : the output power at ports 2 and
3 have minima given by the extinction parameters β- and β+, respectively. Crosstalk in the
demultiplexer is influenced by the residual nonzero extinction parameters β- and β of the
switch, the pulse width of the bits on the multiplexed stream, and by the shape of the waveform
of the switching function f ( V). The pulse widths and the switching function waveform are
important because they both affect crosstalk that occurs while the switch is in transition from one
state to another. The optical receiver following the demultiplexer will generally be designed with
limited bandwidth (to maximize sensitivity) and will integrate all photo electrons detected in a
baseband bit period, including those detected while the switch is in transition between states.
Fig. 6 shows the multiplexed bit stream at the input to the demultiplexer and the control voltage
used to separate it into two bit streams each at half the bit rate of the input stream. The switch is
driven by a periodic switch control voltage V with a frequency equal to half of the input bit rate.
The control voltage swings between the values V_and V+ in each cycle of the drive signal. To
minimize crosstalk caused by the nonzero extinctions β- and β+, the phase shift of the control
voltage V is adjusted such that the switching function f( V) reaches its maximum and minimum
values at the instants the input optical pulses reach their peak powers. However, even under these
conditions, a small amount of crosstalk is inevitable because of the limited extinction. This is
illustrated in Fig. 6(d)and (e) as small unwanted pulses between the main output pulses on each
of the two output channels.
16
Fig. 5. (a) 1 X 2 optical switch. (b)Switching characteristic of 1 X 2optical switch.
The waveform of the switching function affects crosstalk because the switch can deliver
significant power to both output ports over a range of values of f( V). To minimize crosstalk
caused by this property of the switch, it is necessary, in principle, to use a square-wave switching
function f( V). Since the f( V) is monotonic between V = V_ and V = V+, a square-wave f( V)
implies a square-wave control voltage V, as shown in Fig. 6(b). For multi gigabit-per-second
multiplexed bit rates it is difficult to generate a square voltage waveform, because of the finite
bandwidth of the drive electronics. A practical and desirable solution is to use a sinusoidal V
instead of the square waveform and to use optical pulses that are significantly shorter than a bit
period on the multiplexed stream. A sinusoidal control voltage waveform requires high
frequency electronics, but only with narrow bandwidth.
17
Fig. 6. Timing scheme for demultiplexing with a 1 X 2 switch.
The resulting crosstalk is poorer than would be with a square-wave voltage, but this problem is
mitigated to a degree because the inherent nonlinearity off ( V) tends to make the f ( V)
waveform more square than sinusoidal. This squaring of the switching function waveform is
illustrated in Fig. 7, which shows the calculated power atone output port of a I-em-long reversed
∆β. switch the input port of the switch is driven by CW light and the control voltage is a sinusoid
at 4 GHz with an amplitude of 1.4 times the de switching voltage. For systems with more than
two baseband channels, larger demultiplexer switching networks can be constructed by
interconnecting 1 x 2 building blocks .There are two main classes of demultiplexer network: the
binary tree and the linear bus. These are shown in Fig. 8.The binary tree demultiplexer, Fig. 8(a),
uses (n-1) switches. Its first stage is a high-speed switch which demultiplexes the data stream to
two data streams, each at half the multiplexed bit rate. This switch needs to operate at a high
frequency but, as explained earlier, it can be driven by a sinusoid from narrow-band electronics.
18
Subsequent switches in the binary tree structure of Fig. 8(a) also can be driven by sinusoids, but
operate at sub harmonics of the drive frequency of the first switch. These lower frequency
switches are synchronized to the timing of the first switch. In some applications it may be
feasible to use a configuration in which all switches operate at the same frequency r 181. The
binary tree demultiplexer would find application where all baseband receivers are located
together, such as in a point-to-point trunk system. Note that in order to achieve low
demultiplexing crosstalk on all output ports of the binary tree it is necessary that each switch in
the tree provides low crosstalk on both of its outputs . This means that both extinction parameters
(β- and β+ ) should be small in each switch. The linear bus arrangement, Fig. 8(b), uses switches
connected serially, and enables the demultiplexers and base band receivers to be placed in
separate locations such as in a local loop. The demultiplexed bit rates can be made different in
each baseband by appropriate selection of the switching rate for the switches. There are two main
disadvantages of the linear bus demultiplexer arrangement. First, the switches should ideally be
driven by pulses rather than sinusoids. Wide-band drive electronics and switching characteristics
are required, since higher harmonics of the pulse repetition rate are needed to obtain pulse-like
characteristics. Second, data from outputs towards the end of the bus suffer the accumulated
insertion loss of many devices. The loss for the linear bus increases with n, while the loss for the
binary tree increases at the slower rate of 2 logz n. This problem of increased loss could be
alleviated using in-line optical amplifiers, providing spontaneous emission noise does not
degrade the receiver sensitivity . An advantage of the linear bus demultiplexer is that it places
less stringent requirements on one extinction parameter of each of the 1 x 2 switches. For low
crosstalk, good extinction is required only on the demultiplexed output of each switch. Crosstalk
onto the bus from a demultiplexed channel is reduced by subsequent switches. If all switches in
the bus have good extinction in both switch state s then (n - I) switches are required, as shown in
Fig. 8(b). However, if the extinction is good only on the demultiplexed outputs, the "bus" output
on the final switch will be unusable because of crosstalk and a total of n switches will be
required in the demultiplexer.
19
20
2.4 Timing Recovery
As explained earlier, successful demultiplexing depends on correct timing of the demultiplexer
switches. In an optically time-division multiplexed system there is generally no electrical signal
available at the multiplexed bit rate. It is therefore necessary, in systems using electrically driven
demultiplexer switches, to generate an electrical clock signal for the demultiplexer. One solution
to this problem would be to tap off a small component of the multiplexed optical signal at the
input to the demultiplexer and to feed this signal to a separate receiver. The output of this
receiver would then be used to drive a microwave phase-locked loop that generates a phase-
stable clock signal. The receiver could be narrow-band and optimized for the frequency
component of interest.
An alternative to the above scheme would be to obtain a signal for the
timing recovery circuit from a demultiplexed electrical signal at a receiver output. Such a method
has been used in the experiments described in Section III. For this technique to produce low
crosstalk demultiplexing, it is important that timing information contained in the pulse position
of the incoming multiplexed bits is conveyed without distortion to the timing recovery circuit.
Any pulse shaping caused by the demultiplexer will tend to corrupt this timing information. This
requirement of demultiplexing with minimum pulse distort ion of the optical data bits reinforces
the need for optical pulse widths that are significantly shorter than a bit period on the
multiplexed data channel. It also re in forces the need for a switching function with a waveform
that approximates a square wave.
21
Chaper-3
3 EXPERIMENTS
In this section we describe a series of system experiments we have carried out to illustrate
some of the principles presented in the previous sections. The various experiments differ from
one another in the number of channel s multiplexed, the type of lasers used, and in the
multiplexing architecture. In all experiments the baseband electrical data at both the transmitter
and receiver ends of the system are at 4 Gbit/s and are in the NRZ format. Key components in
the experiments are actively mode locked and gain-switched semiconductor lasers, which
generate optical pulses for the transmitters, and Ti: LiNb03 waveguide electro optic
switch/modulators, which are used in the transmitters and demultiplexer. All were designed for
operation at 1.3-p,m wavelength, where the experiments were carried out . Before giving details
of the experiments we briefly describe these key components. The actively mode-locked
semiconductor lasers use antireflection coated CSBH laser chips and 5-cm-longfiber extended
cavities having microwave resonance frequencies of 2 GHz. To general pulses at 4 GHz (the
baseband bit rate), the lasers were mode locked at the second harmonic of their microwave
resonance frequency. Up to four lasers of this kind were operated simultaneously from a
common clock. It was necessary to control the cavity lengths of all four lasers to within 0.5 mm
to ensure that all mode-locking bandwidths overlapped. The pulse widths of the fiber extended
cavity lasers are - 15ps FWHM, or - 30 ps at the baseline. The on/off ratio of the pulses is better
than 20 dB and the output power delivered to the fiber is in the range 5-7 dBm. If we were to
place the pulses with minimum spacing, such that D =T, the multiplexed bit rate would be
greater than 30Gbit/s. In the pre sent experiments, we have increased the spacing to larger values
than this to ensure low crosstalk, and have achieved bit rates up to 16 Gbit/s.
The fiber extended-cavity laser structure has no frequency-selective element and
consequently each laser operates at a slightly different wavelength. In addition, each laser has an
optical spectrum that is wider than the transform-limited bandwidth determined by the pulse
width. This no optimum spectrum is mainly caused by residual short cavity modes of the
antireflection coated laser chips. The overall spectral width of each laser was - 5 nm. In some of
22
our experiments we have used gain-switched DFB lasers in place of the mode-locked lasers. The
gain-switched laser s were also operated at 4 GHz by driving them with a 4-GHz sinusoid. The
pulse widths of these lasers were - 25 ps FWH M, which is not as short as the mode-locked
lasers, but the gain-switched laser offers the advantage of simplicity since no external cavity is
required. The spectral widths of the gain-switched pulse trains were - I nm. We were able to
select DFB lasers which operated within a few nanometers of the wave length of zero first-order
fiber dispersion.
The optical modulators and demultiplexing switches used in the experiments are high-
speed waveguide electro optic directional coupler devices. Uniform Δβ traveling-wave electrodes
were used for the modulators, which are similar in design and performance to the modulators
used in previous fiber transmission experiments. Tn the demultiplexer, high- speed Δβ-reversal
switches are used to en sure low crosstalk in both switch states. To obtain a low drive voltage ,
these devices use a polarization dependent design. Thu s, a fiber in-line compensator was used to
adjust the polarization state at the input to the switches.
3.1Multiple-Laser Systems
We present data obtained from two experimental systems using the multiple-laser
architecture of Fig. 4(a).One of these systems uses two baseband channels and one uses four
baseband channels. Tn both systems the baseband data rate is 4 Gb it /s. A block diagram of the
four channel system is shown in Fig. 9. A common 4-GHz clock drives the four transmitters via
a series of microwave delay lines that are adjusted to provide correct timing of the optical
pulses. In the two-channel system, only two transmitters were used. A schematic of the optical
transmitters is shown in Fig. 10. The clock drives both the word generator and the laser, thereby
ensuring that the electrical data is synchronized to the optical pulses. The lasers used in these
experiments were of the fiber extended-cavity type described above . All lasers were within 8
nm of the wavelength of zero first-order chromatic dispersion in the fiber . The total spectral
width off our multiplexed lasers was 10 nm . For reasons of economy, only one channel was
modulated with pseudorandom data, Data on the other three channels were simulated using
various combinations of periodic 2-GHz and4-GHz pulse trains.
23
Fig. 9, Block schematic of 4-channel aTDM system, The electrical time
delays 7 are 62.5 ps, corresponding to one bit period at 16 Gbit/s .
The combiner architecture used in Fig. 9 was a binary tree arrangement of passive 3-dB
fiber directional couplers. The demultiplexer was a binary tree of active Ti: LiNb03 direction
coupler switches . A block diagram of the demultiplexer and associated electronics is shown in
Fig. 1 1. The first switch is a 5-mm-long high-speed fiber-pigtailed traveling-wave ∆β reversal
directional coupler switch , and is driven by a sinusoidal control voltage at 8 GHz. The two
outputs from this switch are further de multiplexed in two additional fiber-pigtailed switches,
each 10 mm long and driven at 4GHz. Microwave delay liners TI through T3are used to obtain
the correct timing for all switches. The Ti: LiNb03switches used in t he demultiplexer are
discrete devices that are interconnected using fiber pigtails. For the present prototype
experiments, the emphasis was on switching speed and no special care was taken to obtain
minimum insertion loss. As a result, the excess loss of the complete1 x 4 demultiplexer was 12
dB. This loss could be reduced to - 5 dB using an integrated 1 X 4 demultiplexer on a single
chip. For the two-channel system, only one switch was necessary. Its control signal was a
24
sinusoid at 4 GHz . The receivers in both systems used an avalanche photodiode coupled to a
low-noise GaAs FET .
The four-channel 16-Gbit/s system in Fig. 9 serves as a good example of how optical
multiplexing and demultiplexing can remove t he need for very wideband electronics. This
system transmits data over 8 km of fiber at16 Gbit/s, but uses a maximum electronic bandwidth
of only 2.6 GHz (in the drive electronics for the modulator and in the receiver). Since the
demultiplexer switches are driven by sinusoids, There is negligible electrical bandwidth required
in the demultiplexer.. Examples of bit patterns for the two-channel 8-Gbit/s system are shown in
Fig. 12. Fig. 1 2(a) shows the RZ bit pattern "0100110" at the output of transmitter 1, while Fig.
1 2(b) is the RZ bit pattern" 11111111" at the output of transmitter 2. Fig. 12(c) is the
corresponding multiplexed8-Gbit/s RZ signal at the input to the transmission fiber and Fig. 12(d)
is the demultiplexed and re-shaped 4Gbit/s NRZ "01001100" at the output of receiver 1.Fig. 13
illustrates the operation of the four-channel 16-Gbit/s system. The multiplexed 16 Gbit/s RZ
bitpattern"100111111001111" at the output of the multiplexer is shown in Fig. 13(a) . Note that
the bits in this oscilloscope trace are not completely resolved because of the limited bandwidth of
the photo detector (25 GHz) and sampling oscilloscope ( 14 GHz) used to monitor the
multiplexed1 6-Gbit/s signal. The demultiplexed and received 4-Gbit/s NRZ eye pattern
corresponding to Fig. 13(a) is shown in Fig. 1 3(b). This skewing is visible in Fig. 15(c)-(e) for
different levels of intentionally induced crosstalk, obtained by incorrect adjustment of the bias
point and timing of the demultiplexer control voltage away from their optimum values
The pseudorandom modulation on channel 1 combined with the "101010" bit
pattern on channel 2 is a convenient combination of signal formats for assessing and minimizing
crosstalk. It provides a graphic illustration of the effects of crosstalk, it enables the
demultiplexer timing and drive conditions to be adjusted for minimum crosstalk, and it enables
an upper limit on the crosstalk to be directly measured using a microwave spectrum
analyzer .Fig. 16 shows the microwave spectrum of the signals on channel l and channel 2 when
the bias and timing is adjusted for minimum skewing and hence minimum crosstalk. The 2-GHz
components of the two signals differ by-23 dB (electrical), indicating that the crosstalk is - 23dB
or better. This translates to a crosstalk of - ll. 5 dB or better in the optical domain. Note that this
crosstalk is an overall system crosstalk, which reflects contributions both from multiplexing and
25
Fig. 11. Block schematic of demultiplexer and associated electronics for
4-channel system.
26
.
demultiplexing. In the present system the multiplexing crosstalk is estimated to be less than -20
dB (optical). Therefore, the system crosstalk is dominated by crosstalk in the demultiplexer.
27
From above measured bit error rate curves for the two-channel 8-Gbit/s system and the four-
channel 16-Gbit/s system in Fig. 17. These curves show the error rate as a function of the
received 4-Gbit/s power at the input to receiver 1. Four curves are shown: two for 8Gbit/s and
two for 16 Gbit/s. At each bit rate, one curve is presented for transmission through 8 km of fiber
and one curve is presented for the system with the fiber removed ( 0 km ). This pennits an
assessment of various sources of system penalty.
28
Fig. 14. Ey e patterns in 8-Gbit/s system
Fig. 15. Effect of crosstalk on received eye patterns in 8-Gbit/s system
The best receiver sensitivity in Fig. 17 was obtained at8 Gbit/s with no fiber. Under these
conditions there is no fiber dispersion penalty. it shown separately that there is negligible
crosstalk penalty with no fiber at 8Gbit/s by observing that the receiver sensitivity is unchanged
if one of the two transmitters is switched off . The data therefore show that there is negligible
crosstalk caused by the laser pulse on/off ratio or the extinction of the demultiplexer switch.
When the fiber is included in the 8-Gbit /s system, The receiver sensitivity for 10 -9 error rate
29
degrades by 2.5 dB. This is a dispersion penalty attributed to the non optimum spectrum of the
mode-locked lasers. The system margin (i.e., the power available at the receiver above the
receiver sensitivity) is 13.9 dB for the8-Gbit/s system, estimate that the maximum transmission
distance for this dispersion-limited system would be about 16 km. A transmission distance of
over 40 km is possible using lasers with improved spectra (see Section III-B below).The receiver
sensitivity at 10-9 error rate for the fourchannel16-Gbit/s system with no fiber is 2.9 dB poorer
than that of the 8-Gbit/s system, indicating that there is a 2. 9-dB crosstalk penalty at 16 Gbit / s.
Fig. 16. Measurement of crosstalk in 8-Gbit/s demultiplexer. The horizontal
scale on the microwave spectrum analyzer is 10 MHz/div.
30
Fig. 17. Measured error rates of 8-Gbit/s and 16-Gbit/s systems against
4-Gbit/s power level at input to receiver
This crosstalk is dominated by the contribution of the high-speed ( 8 GHz)reverse Δβ switch at
the input of the demultiplexer tree. It is partly caused by increased effective values of the
extinction parameters β- and β+ which is, in turn, caused by under modulation of the switch at
high frequencies. In addition to the above crosstalk penalty, a dispersion penalty of 3.6 dB is
introduced when the transmission fiber is included in the system. The dispersion penalty arises
from two separate effects. The first effect is the well known pulse-broadening associated with the
spectral width of each laser. The second effect is an overlap between adjacent pulses, caused by
the small fiber propagation time differences associated with wavelength differences between
lasers. The dispersion penalty can be viewed as an additional crosstalk penalty but we use the
tern "dispersion penalty" here to separate clearly the crosstalk penalty inherent in the multiplexer
31
and demultiplexer from the additional penalty caused by the dispersive characteristics of the
fiber.
Fig. 18 traces the losses and power levels of a single 4-Gbit/s signal as it passes through
the 16-Gbit/s system from the laser to the receiver APD. Losses and penalties are shown at the
top of the figure and optical power levels are shown at the bottom.
The penalties include a 3-dBduty cycle penalty that occurs with external modulation, the 3. 6-dB
dispersion penalty associated with the 16-Gbit/s signal in the 8-km length of fiber, and the 2.9-
dBcrosstalk penalty associated with the demultiplexer. It can be seen in Fig. 18 that most of the
losses in the system are in the modulator, the multiplexer, and the demultiplexer. As pointed out
earlier, we expect that the demultiplexer loss could be reduced to 5 dB by monolithic integration
on a single chip . Similarly, the modulators could be integrated with an active multiplexer to
give a combined loss also of 5 dB-an improvement of almost 7 dB on the present system . The
system is dispersion-limited in its present form, but with reduced switch losses and improved
laser spectra, we estimate that the transmission distance could be increased to more than
40km .The receiver sensitivities given in the preceding paragraphs were measured after the
demultiplexer , at the input to the 4-Gbit/s baseband receiver . It is also of interest to consider the
32
overall receiver sensitivity of the 8-Gbit/s system at the input to the demultiplexer , and to
compare this number with the best reported receiver sensitivity for a conventional 8-Gbit/s direct
detection system with electrical demultiplexing . The two sensitivities are - 24. 1dBm for the
OTOM system and - 2 5.8 dBm for a state of-the-art 8-Gbit/s APO-FET receiver [4 1 ] . The
present OTOM system is therefore only 1.7 dB less sensitive than the APO-FET receiver . This
difference could easily be eliminated by optimizing the demultiplexer switch and reducing its
loss to a value of 3.0 dB. The overall 1 6-Gbit/s receiver sensitivity of the 16-Gbit/s system is -
10 . 9dBm. We cannot compare this with a conventional direct detection receiver because there
are no published reports of receivers operating at bit rates as high as 16 Gbit/s. If the
demultiplexer loss in the 16-Gbit/s system was reduced to 5.0 dB by integration and if the
crosstalk penalty was made negligible by improving the h igh-frequency switch extinction , the
overall 16-Gbit/s receiver sensitivity would become -20.8 dBm.
3.2 Single-Laser System
To demonstrate that the fiber dispersion penalty can be reduced using lasers with
improved spectra , we briefly describe a two -channel 8-Gbit/s system experiment using the gain-
switched lasers described earlier. The spectral width of the gain-switched lasers is - 1 nm , and
the emission wavelength is within a few nanometers of the wavelength I)f zero first-order fiber
dispersion. In addition, we have ensured that both baseband channels operate at exactly the same
wavelength by using the single-laser topology of F ig . 4(b). The laser output is split into two
using a fiber directional coupler and the optical delays in Fig .
4(b) are introduced using short lengths of fiber . The two channels are combined using a second
fiber directional coupler . The improved spectrum has enabled us to increase the transmission
fiber length to 40 km . The demultiplexer and receiver section of the system is identical to the
multiple-laser 8-Gbit/s system described above. Fig. 19 shows measured error-rate curves for the
8-Gbit/s two-channel single-laser system. A baseline curve is given for 0 km of fiber and with
only one of the two 4-Gbit/s channels operating . The full 8-Gbit/s system error rate curve is
given for a fiber transmission distance of 40 km .The offset between the two curves is 1 .2 dB,
indicating a total penalty (crosstalk plus dispersion) of 1.2 dB.
33
Chaper-4
4 CONCLUSIONS AND OUTLOOK
The system experiments described here have illustrated the potential of optical time-
division multiplexing for multigigabit-per- second point-to point transmission systems. We have
shown that multiplexing, demultiplexing, And timing recovery can be achieved without the need
for very wide bandwidth electronics, and have demonstrated a transmission bit rate that is higher
than in any previous time-multiplexed system. The experiments have also served to highlight the
limitations of OTOM, and help to identify areas in which device improvement s are needed
before the full potential of optical time-division multiplexing can be realized. The two key
limitations in very high bit-rate OTOM are pulse-spreading caused by fiber dispersion and
crosstalk in the demultiplexer. It is desirable, therefore, to optimize those components which
influence these factors. To minimize the dispersion penalty, the semiconductor laser(s) in the
transmitters should have minimum optical spectral width ( i.e. , be Fourier-transform limited). It
is also desirable that the laser wavelength can be adjusted to a value that is close to the
wavelength of the other transmitters (so that all pulses experience the same propagation time),
and close to the zero first-order dispersion wavelength of the fiber (to minimize pulse spreading).
For a four-channel system with transform limited pulses and laser wavelengths within 2 nm of
the zero first-order dispersion wavelength , we estimate that the dispersion penalty will be
sufficiently small to permit loss-limited operation and to allow transmission distances of more
than50 km at 16 Gbit/s. A bit rate of 16 Gbit/s is about the maximum speed at which the present
demultiplexer can operate without significant increases in crosstalk penalty. To improve the
demultiplexer crosstalk performance at 16 Gbit/ s and to extend OTDM methods to higher bit
rates it will be necessary to introduce new approaches to switch design. It will be necessary to
reduce the required drive power to take advantage of the inherent speed capability of technology.
Several complementary avenues to be explored include electrical/optical velocity matching
techniques, optimization tradeoffs to minimize RF attenuation, and resonant microwave
34
techniques. Optoelectronic integration in ш-V semiconductors of the functions and devices
employed here may also permit further optimization of OTDM systems. In addition, all-optical
techniques may lead to the achievement of unforeseen bit rates.
Fig . 1 9 . Measured error rate of single-laser 8-Gbit/s system against
4-Gbit/s power level at input to receiver
In this presentation briefly touched on the possibility of exploiting optical amplifiers in
OTDM systems. Optical amplifiers promise exciting opportunities for improving system
performance and flexibility by compensating for losses. They appear to be particularly attractive
for increasing the transmitted power in systems using single-laser transmitters with a split-and-
recombine architecture to multiplex many baseband channels. Optical amplifiers are also
35
attractive for compensating the accumulated losses in demultiplexers using the linear bus
configuration. It has been shown that traveling-wave optical amplifiers can amplify high-
repetition-rate pico second pulses without distortion, but there are several areas that need to be
investigated before the full potential and limitations of optical amplifiers in OTDM systems can
be gauged. In particular, it is necessary to consider possible degradation of crosstalk performance
caused by spontaneous noise generated in the amplifiers , and the introduction of pattern effects
into high bit-rate optical data streams by carrier storage effects in the optical amplifier. These
effects may introduce some limitations, but there seems no doubt that optical amplifiers will
greatly enhance the design flexibility of OTDM systems and will lead to improved system
performance.
36
Chaper-5
5.REFERENCES
[ 1 ] H. Kobrinski, R. M. Bulky, M. S. Goodman, M. P. Vecchl. C. A .Brackett, L. Curtis, and 1.
L. Gimlet, " Demonstration o f high capacity in the LAMBDANET architecture: A
multiwavelength optical network . " Electron. Lell. , vol . 2 3 , pp. 824-826, 1 987.
[21 O. E. De Lange, " Wide-band optical communications systems: PartII-Frequency division
multiplexing," Proc. IEEE. , vol . 58, pp.1 683- 1 690, 1970; see also: B . Glance, 1. Stone. M.
A . Pollock, P .J . Fitzgerald, C . A . Burrus, B . L. Kasper, and L . w. Stulz,
"DenselyspacedFDM coherent star network with optical signals confined toequally-spaced
frequencies , " 1. Lightwave Technol. , this issue, pp.1770-1786.
[3] T. S. Kinsel and R . T . Denton, "Terminals for a high-speed opticalpulse code modulation
communication system: II. Optical multiplexingand demu1tiplexing , " Proc. IEEE. , vol . 56, pp.
146- 154, 1968.
[4] F . S. Chen, "Demultiplexers for high-speed optical PCM , " IEEE 1.Quantum Electron. , vol.
QE-7. pp. 24-29, 1 97 1 .
[5] S . 1 . Buchsbaum and R . Kompfner, "Time-division multiplex opticaltransmission system ,
" U . S . Patent 3 506 834 , 1970.
[6] T. S. Kinsel and F . S . Chen, "Experimental evaluation of an opticaltime division
demultiplexer for twenty-four channel s , " Appl. Opt. ,vol. 1 1 , pp. 1 4 1 1 - 14 1 8, 1 972.
[71 M . Thewalt, "Time domain multiplexing of signals on an optical fiberusing mode-locked
laser pulses , " IBM Tech. Disclosure Bull. ,vol. 24, pp. 2473-2475, 1 98 1 .
[8] A . Alping, T . Andersson, R . Tell, and S . T . Eng. "20-Gbit/s opticaltime multiplexing with
TJS GaA1As lasers , " Electron. Lett. , vol. 1 8 ,pp. 422-424. 1982.
[9] S . K . Korotky , G. Eisenstein, R . C . Alferness, 1. J. Veselka, L . L .Buhl, G . T . Harvey,
and P. H . Read, "Fully connectorized highspeedTi : LiNbO, switch/modulator for time-
divisionmultiplexing anddata encoding," 1. Lightwave Technol . . vol. LT-3, pp. 1 -6, 1985.
[ 1 0] P. R. Prucnal . M . A. Santoro, S . K . Sehgal, and I . P. Kaminow ," TDMA fiber-optic
network with optical processing," Electron.
37
Lett. , vo\. 22. pp. 1 2 1 8 - 1 2 1 9 , 1986.
[ 11] R. S. Tucker, G . Eisenstein, S. K . Korotky , U. Koren, G . Raybon,
1. 1. Veselka, L. L. Buhl, B. L. Kasper, and R. C . Alferness, "Opticaltime-division multiplexing
and demultiplexing in a mUltigigabit/second fiber transmission systems , " Electron. Lett. , vol.
23, pp.208-209, 1 987 .
[ 1 2] L. C. Blank, E. G. Bryant, A. Lord, J . M . Boggis, and W . A. Stallard, " 1 50-km optical
fiber transmission network experiment with 2-Gbit/s throughput," Electron. Lett. , vol. 2 3 , pp.
977-978. 1 987 .
[ l 3] P. J . Chidgey and D. W . Smith, "Sampled optical time-division multiplexing of
asynchronous data, ' . Electron. Lett" vol . 2 3 , pp. 1 228-1 229, 1 987 .[ 14] P. R. Prucnal, D. J.
Blumenthal, and M . A. Santoro, " 1 2 . 5-Gbit/sfiber-optic network using all-optical processing,"
EleCiron. Lett. , vol.23, pp. 629-630, 1 987.
[ 15] G. Eisenstein, R . S . Tucker, S. K . Korotky , G . Raybon, J. J. Veselka,L L. Bu hl , A. H.
Gnauck, B. L. Kasper, and R. C. Alferness,.'Optical time-division multiplexed transmission
system experimentat 8 Gbit/s, " Electron. Lett. , vol . 2 3 , pp. 1 1 1 5- 1 1 1 6 , 1 987 .
[ 1 6] R. S. Tucker. G. Eisenstein, S. K. Korotky, L. L . Buhl, J . J. Veselka, G. Raybon. B. L.
Kasper, and R. C. Alferness, " 1 6-Gbit/sfiber transmission experiment using optical time-
division multiplexing," EleCiroYl. Lett. , vol. 2 3 , pp. 1 270- 1 27 1 , 1987 .
[ 1 7] G. Eisenstein, R. S . Tucker, U. Koren, and S. K. Korotky. " Activemode-lucking
characteristics of InGaAsP-sing1e mode fiber compositecavity lasers , " IEEE J. Quantum
Electron. , vol. QE-22, pp. 142-148, 1985.
[ 1 8] H . Haga, M. Izutsu, a n d T . Sueta, "An integrated 1 x 4 high-speedoptical switch and its
applications to a time demultiplexer," J.Lightwave Technol. , vol . LT-3, pp. 1 1 6- 1 20 , 1 9 8
5 .
[ 1 9] R. C. Alferness, L . L . Buhl, S. K . Korotky, and R . S . Tucker, " HighspeedLl./3-reversal
directional coupler switch," presented at Top.Meet. Photonic Switching, Incline Village. NV,
Mar. 1987, pap.PhD6.
[20] P. R . Prucnal, M. A. Santoro, and S. K. Sehgal, "Ultrafast alI-opticalsynchronous multiple
access fiber networks , " IEEE J. SelectedAreas Commun. , vol . SAC-4, pp. 1484- 1493, 1 986.
38
39