1

Subscriber Multiplexing
The recent WDM-based FTTH network uses 1550 nm wavelength for CATV video
stream, 1490nm for digital data downstream and 1310nm upstream TDMA as shown in
Figure 1-1. In terms of system design, this approach requires WDM filters, addi tional
lasers and photodiodes at central office ( CO) and end-users. It is not efficient for
bandwidth utilization and the difficulty of this architecture is the more demanding
1310nm TDMA upstream transmission resulting from the increase of the sharing rat io.
There goal another approach i.e. sub -carrier multiplexing (SCM) Optical Network as
shown in figure 1-2 and 2-1.
Because of the simplicity and stability of microwave and RF devices, SCM over WDM
can combine different RF channels (analog & digital signals) closely with each other in
electrical domain, and then modulate onto an optical carrier.
2
In this study, 78 NTSC standard analog video streams and 1Gb ps digital data are mixed
by different microwave frequencies and combined together in the electr ical domain
before modulating onto one wavelength using optical single sideband modulation. This
composite signal is modulated at the lower sideband of the optical carrier. In addition, a
microwave frequency is modulated at the upper sideband of optical ca rrier.
At the end-users, an optical filter (fabry-Perot Interferometer) and optical circulator can
be used to separate the optical subcarriers at the upper and lower sideband of optical
carrier. The optical subcarriers at the lower sideband of optical carrier will then
demodulate into electrical domain for CATV broadcasting and downstream digital data
transmission. The optical subcarriers at the upper sideband of optical carrier can be used
as an optical source for end-user upstream digital data transmi ssion.
Analog SCM Systems
Most CATV networks distribute television channels by using analog techniques based on
frequency modulation (FM) or amplitude modulation with vestigial sideband (AM -VSB)
formats. As the waveform of an analog signal must be preserve d during transmission,
analog SCM systems require a high SNR at the receiver and impose strict linearity
requirements on the optical source and the communication channel.
In analog SCM lightwave systems, each microwave subcarrier is modulated using an
analog format, and the output of all subcarriers is summed using a microwave power
3
combiner (see Fig. 8.29). The composite signal is used to modulate the intensity of a
semiconductor laser directly by adding it to the bias current.
In practice, the analog signal is distorted during its transmission through the fiber link.
The distortion is referred to as intermodulation distortion (IMD) and is similar in nature
to the FWM distortion. Any nonlinearity in the response of the semiconductor laser used
inside the optical transmitter or in the propagation characteristics of fibers generates new
frequencies of the form f
i
+ f
j
and f
i
+ f
j
f
k
, some of which lie within the transmission
bandwidth and distort the analog signal. The new frequencies are referred to as the
intermodulation products (IMPs).
These are further subdivided as two-tone IMPs and triple-beat IMPs. The triple-beat
IMPs tend to be a major source of distortion because of their large number. An N-channel
SCM system generates N(N-1)(N-2)/2 triple-beat terms compared with N(N -1) two-tone
terms. The second-order IMD must also be considered if sub carriers occupy a large
bandwidth.
Several other mechanisms, such as fiber dispersion, frequency chirp, and mode -partition
noise can cause IMD that induces degradation of the system performance.
Digital SCM Systems
The capacity of a digital SCM system is more that analog SCM systems. Moreover, a
single digital video channel requires a bit rate of more 100 Mb/s; a common technique
uses a multilevel QAM format is introduced to support high data rates. If M represents
the number of discrete levels used, the resulting non -binary digital signal is called M-ary
because each bit can have M possible amplitudes (typically M = 64). Such a signal can be
4
recovered at the receiver without using coherent detection and requires a lower CNR
compared with that needed for analog AM-VSB systems.
The capacity of an SCM system can be increased considerably by employing hybrid
techniques that mix analog and digital formats. The hybri d SCM systems can transmit a
large number of video channels over the same fiber simultaneously. Such hybrid SCM
systems can transport up to 80 analog and 30 digital channels using a single optical
transmitter. If only QAM format is employed, the number of digital channels is limited to
about 80.
The performance of such systems is affected by the clipping noise, multiple optical
reflections, and the nonlinear mechanisms such as self -phase modulation (SPM) and
SBS, all of which limit the total power and the number of channels that can be
multiplexed. Further increase in the system capacity can be realized by combining the
SCM and WDM techniques, a topic discussed next.
WDM- SCM Systems
Further, combining the SCM and WDM techniques can increase the system capa city. The
combination of WDM and SCM provides the potential of designing broadband passive
optical networks capable of providing integrated services (audio, video, data, etc.) to a
large number of subscribers. In this scheme, multiple optical carriers are launched into
the same optical fiber through the WDM technique. Each optical carrier carries multiple
SCM channels using several microwave subcarriers. The limiting factor for multi -
wavelength SCM networks is inter -channel crosstalk caused by SRS and XPM.
Multi-wavelength SCM systems are quite useful for LAN and MAN applications,
providing multiple services (telephone, analog and digital TV channels, computer data,
etc.) with only one optical transmitter and one optical receiver per user because different
services can use different microwave subcarriers. This approach lowers the cost of
terminal equipment in access networks.
Different services can be offered without requiring synchronization. The main advantage
of multi-wavelength SCM is that the network ca n serve NM users, where N is the number
of optical wavelengths and M is the number of microwave carriers. In another approach,
the hybrid fiber-coaxial (HFC) technology is used to provide broadband integrated
services to the subscriber.
5
Optical Code Division Multiple Access (CDMA)
Code Division Multiple Access (CDMA) is generically known as Spread Spectrum
transmission technique in the world of radio communications systems. In the optical
world, CDMA technology uses in two roles:
1. Optical shared medium LANs
2. Local access networks
In most communications systems, our objective is to fit the maximum amount of useful
signal into minimal bandwidth. In CDMA, a spread spectrum system, we use some
artificial technique to broaden the amount of bandwidth used a nd transmit multiple
signals over the same frequency band, using the same modulation techniques at the same
time to achieve the above-mentioned aim. This has the following effects:
Capacity Gain
Using the Shannon-Hartly law for the capacity of a band limit ed channel, it is easy to see
that for a given signal power, the wider the bandwidth used, the greater the channel
capacity. So if we broaden the spectrum of a given signal, we get an increase in channel
capacity and an improved signal -to-noise ratio is obtained.
Security
Military people, initially, use spread spectrum technique for security issues as spread
spectrum signals have an excellent rejection of intentional jamming. In addition, the
Direct Sequence (DS) technique results in a signal, which is ver y hard to distinguish from
background noise unless you know the random code sequence used to generate the signal.
Thus, not only are DS signals hard to jam, they are extremely difficult to decode (unless
you have the key) and quite hard to detect.
Immunity to Multipath Distortion
Some spectrum spreading techniques have a significantly better performance in the
presence of multipath spreading than any available narrowband technique.
Interference Rejection
Spread spectrum signals can be received even in th e presence of very strong narrowband
interfering signals (up to perhaps 30 dB above the wanted signal).
Direct Sequence Spread Spectrum (DSSS)
6
DSSS is a popular technique for spreading the spectrum. Figure 371 shows how the
signal is generated.
1. The binary user data is used to modulate a pseudo -random bit stream. The rate
of this pseudo-random bit stream is much faster (from 9 to 100 times) than the
user data rate. The bits of the pseudo -random stream are called chips. The ratio
between the speed of the chip stream and the data stream is called the spread ratio.
2. The output of the faster bit stream is used to modulate a radio frequency (RF) or
optical carrier.
3. Any suitable modulation technique, bi -polar phase shift keying (BPSK) is usually
adopted.
4. In optical systems, NRZ coding is typically used.
Whenever a carrier is modulated the result is a spread signal with two sidebands above
and below the carrier frequency. These sidebands are spread over a range plus or minus
the modulating frequency. The sideb ands carry the information and it is common to
suppress the transmission of the carrier (and sometimes one of the sidebands). It can be
easily seen that the width (spread) of each sideband has been multiplied by the spread
ratio.
1. The secret of DSSS is in the way the signal is received. The receiver knows the
pseudo-random bit stream (because it has the same random number generator).
Incoming signals are correlated with the known pseudo -random stream. Thus the
chip stream performs the function of a known wa veform against which we
correlate the input. Co-relational receivers can be constructed in several ways.
7
Code Division Multiple Access (CDMA)
The DSSS technique gives rise to a novel way of sharing the bandwidth. Multiple
transmitters and receivers are able to use the same frequencies at the same time without
interfering with each other. This is a by-product of the DSSS technique. The receiver
correlates its received signal with a known (only to it) random sequence - all other signals
are filtered out. This is interesting because it is really the same process as FDM.
When we receive an ordinary radio station (channels are separated by FDM), we tune to
that station. The tuning process involves adjusting a resonant circuit to the frequency we
want to receive. That circuit allows the selected frequency to pass and rejects all other
frequencies. What we are actually doing is selecting a sinusoidal wave from among many
other sinusoidal waves by selective filtering. If we consider a DSSS signal as a
modulated waveform, when there are many overlapping DSSS signals then the filtering
process needed to select one of them from among many is exactly the same thing as FDM
frequency selection except that we have waveforms that are not sinusoidal in shape.
However, the DSSS chipping sequences (pseudo -random number sequences) must be
orthogonal (unrelated). Fortunately there are several good simple ways of generating
orthogonal pseudo-random sequences.
For this to work, a receiving filter is needed which can select a s ingle DSSS signal from
among all the intermixed ones. In principle, you need a filter that can correlate the
8
complex signal with a known chipping sequence (and reject all others). There are several
available filtering techniques, which will do just this. T he usual device used for this
filtering process is called a Surface Acoustic Wave (SAW) filter. CDMA has a number of
very important characteristics:
Statistical Allocation of Capacity
Any particular DSSS receiver experiences other DSSS signals as noise. This means that
you can continue adding channels until the signal -to-noise ratio gets too great and you
start getting bit errors. The effect is like multiplexing packets on a link. You can have
many active connections and so long as the total (data traffic ) stays below the channel
capacity all will work well. For example, in a mobile telephone system, (using DSSS
over radio) only about 35% of the time on a channel actually has sound (the rest of the
time is gaps and listening to speech in the other directio n). If you have a few hundred
channels of voice over CDMA what happens is the average power is the channel limit -
so you can handle many more voice connections than are possible by FDM or TDM
methods. This also applies to data traffic on a LAN or access n etwork where the traffic is
inherently bursty in nature. However, it has particular application in voice transmission
because, when the system is over committed, there is no loss in service but only
degradation in voice quality. Degradation in quality (dro pping a few bits) is a serious
problem for data but not for voice.
No Guard Time or Guard Bands
In a TDM system when multiple users share the same channel there must be a way to
ensure that they don't transmit at the same time and destroy each other's sig nal. Since
there is no really accurate way of synchronizing clocks (in the light of propagation delay)
a length of time must be allowed between the end of one user's transmission and the
beginning of the next. This is called guard time. At slow data rate s it is not too
important but as speed gets higher it comes to dominate the system throughput. CDMA
of course does not require a guard time - stations simply transmit whenever they are
ready.
In FDM (and WDM) systems, unused frequency space is allocated b etween bands
because it is impossible to ensure precise control of frequency. These guard bands
represent wasted frequency space. Again, in CDMA they are not needed at all.
9
Requirement for Power Control
DSSS receivers can't distinguish a signal if its stre ngth is more than about 20 dB below
other similar signals. Thus if many transmitters are simultaneously active a transmitter
close to the receiver (near) will blanket out a signal from a transmitter which is farther
away. The answer to this is controlling the transmit power of all the stations so that they
have roughly equal signal strength at the receiver. In a reflective -star type optical LAN
topology this is not a problem since there will be very little variation in signal levels. But
in some possible access network configurations it could be a limitation.
Practical Optical CDMA
Optical CDMA is still very much a research technology. In the early 1990's it was
proposed as a technology for shared medium LANs but since then the shared medium
LAN itself has proven more costly than switch based star networks. Thus there isn't a lot
of interest in shared medium LANs in either the optical or electronic world. Today
however finding a low cost technology for the upstream transport in a passive optical
network is a significant and important challenge. CDMA might well be a good choice
here. Practical optical CDMA systems have some differences from RF ones:
- Instead of using a random number generator to generate the chipping sequence, a
fixed sequence only as long as one data bit is likely to be used. For example, you
might have 31 chips per bit and the chipping sequence would be the same for
every bit transmitted.
- Zero data bits are not transmitted at all. This nets out to saying that a 1 bit (for a
particular end-user) is transmitted as an invariant 31-chip sequence.
- The codings used in an optical system need to be different from those used in an
RF system as we don't have a negative signal state in optical communications. We
only have positive (or zero) states.
Optical Time Division Multiplexing (OTDM)
Time Division Multiplexing (TDM) provides a very simple and effective way of sub -
dividing a high-speed digital data stream into many slower speed data streams. Indeed
most optical communications links are really TDM data streams but the TDM is done
electronically rather than optically. SDH and SONET are standards for electronic TDM
10
over an optical carrier. The main objective of OTDM is to allow the optical signal stream
to run at speeds significantly in excess of the maxi mum speed of the electronics.
TDM Concept
Figure 373 illustrates the principle of time division multiplexing. In the illustration, there
are four slow-speed bit streams merged into a single high -speed stream at four times the
speed of any one of the component signals. Each input stream is assigned one bit in every
four in turn. There are a number of points to note:
- In the illustration, we are allocating time slots in the high -speed data stream at the
individual bit level, which is not necessary. In TDM electronic communications
system, time slots are often allocated on the basis of 8 -bit groups or even in
larger groupings. However, in optical TDM proposed systems use the bit -
interleaving technique almost exclusively.
- The data stream is arranged in r epeating patterns of time slots usually called
frames. In the example, a frame would be just four bits. Thus, input channel x
might be allocated bit number 3 in every frame.
- It is not necessary for each of the slow-speed streams to be the same. For
example, we could allocate three TDM signals at different rates by allocating a
different number of bits in each time frame to each stream. Thus, input stream 1
might be allocated bit numbers 1, 3, 5, 7..., stream 2 might be allocated bits 2,
11
6,10... and stream 3 bits 4, 8, 12... In this example stream 1 would be twice the
rate of either stream 2 or stream 3.
- There is very little delay experienced by the slow speed streams due to their travel
over the higher speed trunk. There will be a need for some speed -matching
buffering at the points of multiplexing and demultiplexing but this can usually be
limited to a single bit.
- Once the time slots are allocated each subordinate signal stream has a fixed and
invariant data rate. Re-allocating the bits can change this but this is difficult to do
dynamically, takes time and wastes resources.
- Each signal stream must be synchronized to the higher speed stream! This is the
most significant problem in TDM. Each slow speed stream must deliver its bits at
exactly the correct rate or there will be times when a bit needs to be transmitted
and it has not yet arrived or times where too many bits arrive and some must be
discarded. Neither of these situations is compatible with error -free transmission.
Of course at the destination each slow speed stream must be received at exactly
the rate that the bits are delivered from the high speed one. TDM takes no
prisoners!
TDM Network Principles
Figure 374 illustrates the general principle of a TDM network. For the sake of illustration
we will assume that data is multiplexed in units of a single byte. In the figure we have
illustrated a 1 Mbps synchronous connection between the two end users (User A and User
B). The network is configured as follows
12
- User A is connected on a dedicated link to Node B at a speed of 1 Mbps (125,000
bytes/sec). Note that Node B not by the end user provides the timing for the link.
This means that Node B sends a clock signal to User A each time that a bit is to
be sent.
- Node B has been set up with a rule that says whenever a byte is received on Link
1 place it into time-slot x of Link 2.
- Node B has a connection with Node A at a speed of 4 Mbps. Our connections
between end users is allocated to this link and so gets every fourth byte on Link 2.
- Node A has a rule that says whenever time-slot x of Link 2 arrives take the data
in it and place it into time-slot y of Link 3.
- Node A to Node C is a 10 Mbps link, which will also carry our end -user
connection. Thus only 1 byte in 10 on the link will belong to this particul ar end-
user connection.
- Node C to Node E is at 4 Mbps (same as B to A) and again we get every fourth
byte.
- Node E is connected directly to User B at a dedicated speed of 1 Mbps. Clocking
for this link is again provided by the network not the end user.
- Note that each connection is bi -directional (although strictly it doesn't need to be).
Thus we have an end-to-end connection where data is passed on from link -to-link one
byte at a time in a strictly controlled way. Of course, the 10 Mbps connection and the 4
Mbps connection must have a strictly identical timing source. If the timing relationship
13
between the two links was to vary even slightly loss or corruption of data would result.
Connections can be set up in two ways:
- By network management : In this case the node is called a cross-connect.
Connection setup or tear -down may take from a few minutes to a few days.
- By signalling :Signalling means by request in real -time by the end user.
Connection setup may take from about 10 ms up to about a second depending o n
the type of network and it's size. In this case the nodes are called switches. The
best-known switched TDM network is the telephone network. In the optical
world, operating at much higher speeds, it is likely the early networks will be
cross-connects only.
Optical TDM Principles
Figure 375 illustrates one particular proposed method for building an optical TDM
system. The system illustrated shows four streams merged into one. The modulation
technique used is RZ coding as discussed in 7.2.1.4, RZ Coding on page 305. RZ
coding is used because it alleviates the extremely difficult problem of synchronizing
different bit streams into adjacent time slots. (In RZ the laser ON state for only the first
half of the bit-time represents coding a 1-bit).
14
There will always be some jitter in the slow stream bit stream as it is mixed into the faster
stream. In addition any optical pulse will be bell shaped rather than square. Thus no
matter what we do there will be gaps between the bits. The system operates in the
following way:
- Each time slot (illustrated by the downward pointing arrows) is sub -divided into 4
bit times.
- Each bit-time (in conformity to the RZ code in use) is further divided into two
halves. For a 1 bit the first half of the bit time will be occupie d by an optical
pulse (and the second half will be dark). For a 0 bit the whole bit time will be
dark.
- A laser produces a short pulse (for half a bit time) at the beginning of each time
slot. In this example the laser is ON for one -eighth of the time slot. This can be
done in many ways. Self-pulsating laser diodes have been suggested (See 3.3.7,
Mode-Locking and Self-Pulsating Lasers on page 117.). However a standard
laser with an external modulator or an integrated modulator may be more
appropriate because we want to avoid laser chirp.
- The laser signal is split 4-ways. (There are many ways to do this - concatenated 3
dB couplers being the most obvious.) A planar free -space coupler will also do this
and would be used if the whole TDM device were built on a single planar
substrate.
- Each signal (except one) is then delayed by a fixed amount. Using a loop of
standard fiber easily and conveniently provides this delay. Of course each signal
is delayed by a different amount.
- Then each signal is separately modul ated to carry it's own unique information
stream. The trick here is to synchronize the modulators accurately given that their
response will be much slower than a single bit time (at the full link speed).
- The signals are then re-combined (perhaps using concatenated 3 dB splitters or a
free space coupler) to form a single data stream.
- During all this the original signal has lost a very large amount of power. Each
pulse will lose a minimum of 6 dB in each of the 4 -way splitter and the combiner.
In addition there will be loss in the modulator. It would be a very good modulator
15
if the insertion loss was only about 6 dB. So in total each output bit pulse will be
at least 18 dB (and maybe as much as 25 dB) less than the original pulse
amplitude as it left the transmit laser.
The whole stream then must be amplified to reach a strength suitable for transmission on
the link. Indeed, if soliton transmission is to be used (and we can't go at 100 Gbps rates
any other way), there will need to be a very high level of ampli fication. The power level
needs to be around 3 mw or above for a soliton to form. Hence we will probably be
looking for around 40 dB or more of gain from the amplifier! (This is no problem for a
multi-stage EDFA but there is an interesting challenge here i n amplifier design to manage
the amplified spontaneous emission.)
Sources of Power Penalty
Besides fiber dispersion, there are several physical phenomena that degrade the receiver
sensitivity like modal noise, dispersion pulse broadening and intersymbol i nterference,
mode-partition noise, frequency chirp, and reflection feedback. In this section, we discuss
how the system performance is affected by considering these phenomena.
Modal Noise
Modal noise originates from the interference among the various propa gating modes in
multimode fiber, creating speckle pattern at the receiver. Any fluctuations in this speckle
pattern with time leads to the fluctuations in the received power. Such fluctuations are
termed as Modal noise. Modal noise can cause a serious prob lem in optical transmission
by producing an important degradation of the bit -error-rate. The measurement of the bit -
error-rate in optical links is a difficult task because modal noise tends to group errors in
packets. Modal noise is strongly affected by so urce spectral width as mode interference
occurs if the coherent time (duration during which the source phase remains relatively
stable and inversely proportional to the source spectral width) is lon ger than inter modal
delay time. That is why the designer to avoid modal noise over lasers mostly adopts
LEDs. The following Figure reveals that as the number of propagating modes increases,
the power penalty at the receiver to combat with modal noise increases. Modal noise is
not only associated with multimode fiber but also occurs in single mode fiber if small
sections of fibers i.e. below 2mm are installed between two connectors or splices.
16
Dispersive pulse broadening
The use of single-mode fibers for light wave systems nearly avoids the problem of
intermodal dispersion and the associated modal noise. The group -velocity dispersion
(GVD) still limits the bit ratedistance product BL by broadening optical pulses beyond
their allocated bit slot and depends on th e source spectral width v A . Dispersion-induced
pulse broadening affects the receiver performance in two ways.
- First, a part of the pulse energy spreads beyond the allocated bit slot and leads to
ISI (discussed previously).
- Second, the pulse energy within the bit slot is reduced when the optical pulse
broadens. Such a decrease in the pulse energy reduces the SNR at the decision
circuit.
Since the SNR should remain constant to maintain the system performance, the receiver
requires more average power. This is the origin of dispersion induced power
penalty
d
PP , which is given as.
, )
10
5log 1 4
d
PP BLD v = A
]
Figure shows the power penalty as a function of the dimensionless parameter
combination BLD v A . Although the power penalty is negligible (
d
PP = 0.38 dB) for
17
BLD v A = 0.1, it increases to 2.2 dB when BLD v A = 0.2 and becomes infinite when
BLD v A = 0.25.
Mode Partition Noise
Mode partition noise is a problem in single -mode fiber operation. In multimode fiber
modal noise and intermodal dispersion dominate. Mode -partition noise (MPN) occurs
because of an anti-correlation among pairs of longitudinal modes. In particular, various
longitudinal modes fluctuate in such a way that individual modes exhibit large intensity
fluctuations but the total intensity remains relatively constant. MPN is harmless in the
absence of fiber dispersion, as all modes remain synchronized during transmission and
detection. But practically, different modes travel at slightly different speeds inside the
fiber because of group-velocity dispersion and become unsynchronized, As a result of
such de-synchronization; the receiver current exhibits additional fluctuations that reduce
the SNR at the decision circuit. A power penalty must be paid to improve the SNR to
achieve the required BER, which is calculated as
, )
2 2
10
5log 1
MPN MPN
PP Q r =
Here
MPN
r is the relative noise level of the received power in the presence of MPN
. , )
, ,
2
1 exp
2
MPN
k
r BLD

=
]
18
Here the mode-partition coefficient 1
cc
k = = with values in the range 01 and is
likely to vary from laser to laser. And BLD

is normalized dispersion parameter. The
following figure shows the power penalty at a BER of 10
-9
(Q = 6) as a function of the
normalized dispersion parameter BLD

for several values of the mode-partition
coefficient k.
Frequency Chirping
Frequency chirping is a phenomenon that limits the performance of light wave systems
operating near 1.55 m. The amplitude modulation in semiconductor lasers is
accompanied by phase modulation, which introduces transient changes in the refractive
index governed by the linewidth enhancement factor. Such a pulse is called chirped. As a
result, this frequency chirp, imposed on an optical pulse, broadens its spectrum
considerably. Such spectral broadening affects the pulse shape at the fiber output because
of fiber dispersion and degrades system performance. An exact calculation of the chirp -
induced power penalty
Chirp
PP is difficult because frequency chirp depends on both the
shape and the width of the optical pulse. In a simple model the chirp -induced power
penalty is given by
j
10
10log 1 4
Chirp c
PP BLD = A
19
Here,
c
A is the spectral shift associated with frequency chirping. The following figure
shows the power penalty as a function of the normalized dispersion parameter
c
Bt , here
c
t is the chirp duration (100200 ps).
Reflection Feedback and Noise
Control and minimization of reflections is a key issue in every optical communication
system. Of course, there are many instances where we create reflections intentionally: for
example at the end facets of a las er. The reflections discussed here are unintended ones
that occur at connectors, joins and in some devices. These unwanted reflections could
have many highly undesirable effects. Among the most important of these are:
- Disruption of laser operation
Reflections entering a laser disturb its stable operation adding noise and shifting
the wavelength.
- Return Loss
Reflections can vary with the signal and produce a random loss of signal power.
This is termed return loss and is further described in 2.4.4, Refl ections and
Return Loss Variation on page 67.
- Amplifier operation
Reflections returning into an optical amplifier can have two main effects:
In the extreme case of reflections at both ends the amplifier becomes a laser and
produces significant power of its own. (In a simple EDFA with only Ge as co -
dopant this would happen at the ASE wavelength of erbium, which is 1553 nm.
20
However, with other co-dopants present the lasing wavelength will often be
between 1535 nm and 1540 nm.
In lesser cases reflections can cause the amplifier to saturate (by taking away
power) and again introduce noise to the signal.
Reflections can be created at any abrupt change in the refractive index of the optical
material along the path. The major causes are:
- Joins between high RI material and fiber (such as at the junction between a laser
or LED and a fiber or between any planar optical component and a fiber).
- Joins between fibers of different characteristics. This is a bit unusual but there are
some cases where this has to happen . For example where a Pr doped amplifier
employing ZBLAN host glass is coupled to standard fiber for input and output.
- Any bad connector produces significant reflections. For that matter most good
connectors produce some reflection albeit slight.
- Some optical devices such as Fabry-Perot filters reflect unwanted light as part of
their design.
Reflections need to be kept in mind and can be controlled by one or more of the
following measures:
1) Taking care with fiber connectors and joins to ensure that they are made correctly
and produce minimum reflections. This can be checked using an OTDR.
2) By inclusion of isolators in the packaging of particularly sensitive optical
components (such as DFB lasers and amplifiers). The use of isolators is important
but these devices (of course) attenuate the signal and are polarization sensitive.
They can also be a source of polarization modal noise. Their use should be
carefully planned and in general, minimized.
3) In critical situations a diagonal splice in the fiber can be made o r a connector
using a diagonal fiber interface can be employed. The use of a diagonal join
ensures that any unwanted reflections are directed out of the fiber core.
Nevertheless, diagonal joins are difficult to make in the field due to the tiny
diameter of the fiber and the high precision required.
Anti-reflection coatings are very important where the reflection is due to an RI
difference. This may be at the edge of a planar waveguide for example. The fiber or
21
waveguide end is coated with a 1/4 wave thick l ayer of material of RI intermediate
between the device material and the air (if air is the adjoining material). The principle
involved here was discussed in 2.1.3.2, Transmission through a Sheet of Glass on page
22. In many systems it is critical to ensu re that reflections are considered in the system
design and that links are tested after installation to ensure that reflections are minimized.