Chapter 12: Optical Communications

12.1 Introduction to optical communication links

12.1.1 Introduction to optical communications and photonics

Optical communications is as ancient as signal fires and mirrors reflecting sunlight, but it is
rapidly being modernized by photonics that integrate optics and electronics in single devices.
Photonic systems are usually analyzed in terms of individual photons, although wave methods
still characterize the guidance of waves through optical fibers, space, or other media. This
chapter introduces optical communications and applications of photonics in Section 12.1. It then
discusses simple optical waveguides in Section 12.2, lasers in Section 12.3, and representative
components of optical communications systems in Sections 12.4, including photodetectors in
12.4.1-2, multiplexers in 12.4.3, interferometers in 12.4.4, and optical switches in 12.4.5.

12.1.2 Applications of photonics

Perhaps the single most important application of photonics today is to optical communications
through low-loss glass fibers. Since 1980 this development has dramatically transformed global
communications. The advantage of an optical fiber for communications is that it has a
bandwidth of approximately one terahertz, and can propagate signals over continental and even
global distances when assisted by optical amplifiers. These amplifiers are currently separated
more than ~80 km, and this separation is steadily increasing as technology improves. In contrast,
coaxial cable, wire-pair, and wireless links at radio frequencies still dominate most
communication paths of bandwidth < ~2 MHz, provided the length is less than ~1–50 km.

One broadband global wireless alternative to optics is microwave communications satellites

in geosynchronous orbit66 that can service ships at sea and provide moveable capacity addressing
transient communications shortfalls or failures across the globe; the satellites simply point their
antenna beams at the new users, who can be over 10,000 km apart. The greatest use of satellites,
however, is for broadcast of entertainment over continental areas, either to end-users or to the
head ends of cable distribution systems. In general, the limited terrestrial radio spectrum is more
efficiently used for broadcast than for one-to-one communications unless there is re-use of
spectrum as described in Section 10.4.6. Optical techniques are disadvantaged for satellite-
ground links or ground-to-ground links through air because of clouds and fog, which restrict
such links to very short distances or to cases where spatial diversity67 offers clear-air alternatives.

Optical links also have great potential for very broadband inter-satellite or diversity-
protected satellite-earth communications because small telescopes easily provide highly focused
antenna beams. For example, beamwidths of telescopes with 5-inch apertures are typically one

66
Geosynchronous satellites at 22,753-mile altitude orbit Earth once every 24 hours and can therefore hover
stationary in the sky if they are in an equatorial orbit.
67
Spatial diversity involves use of spatially distinct communications links that suffer any losses independently;
combining these signals in non-linear ways improves overall message reliability.

- 369 -
arc-second68, corresponding to antenna gains of ~4π×(57×3600)2 ≅ 5×1011, approximately 5000
times greater than is achievable by all but the very best radio telescopes. Such optical links are
discussed in Section 12.1.4.

Optical fibers are increasingly being used for much shorter links too, simply because their
useable bandwidth can readily be expanded after installation and because they are cheaper for
larger bandwidths. The distance between successive amplifiers can also be orders of magnitude
greater (compare the fiber losses of Figure 12.2.6 with those of wires, as discussed in Section
7.1.4 and Section 8.3.1). The bandwidth per wire is generally less than ~0.1 GHz for distances
between amplifiers of 1 km, whereas a single optical fiber can convey ~1 THz for 100 km or
more. Extreme data rates are now also being conveyed optically between and within computers
and even chips, although wires still have advantages of cost and simplicity for most ultra-short
and high-power applications.

Optical communication is not the only application for photonics, however. Low-power
lasers are used in everyday devices ranging from classroom pointers and carpenters’ levels to
bar-code readers, laser copiers and printers, surgical tools, medical and environmental
instruments, and DVD players and recorders. Laser pulses lasting only 10-15 second (0.3 microns
length) are used for biological and other research. High power lasers with tens of kilowatts of
average power are used for cutting and other manufacturing purposes, and lasers that release
their stored energy in sub-picosecond intervals can focus and compress their energy to achieve
intensities of ~1023 W/m2 for research or, for example, to drive small thermonuclear reactions in
compressed pellets. Moreover, new applications are constantly being developed with no end in
sight.

12.1.3 Link equations

The link equations governing through-the-air optical communications are essentially the same as
those governing radio, as described in Section 10.3. That is, the received power Pr is simply
related to the transmitted power Pt by the gain and effective area of the transmitting and
receiving antennas, Gt and Ae:

Pr = (G t Pt 4πr 2 )A e [W] (optical link equation) (12.1.1)

The gain and effective area of single-mode optical antennas are related by the same equation
governing radio waves, (10.3.23):

G = 4πA λ 2 (12.1.2)

Some optical detectors intercept multiple independent waves or modes, and their powers add. In
this case, the gain and effective area of any single mode are then less relevant, as discussed in
Section 12.1.4.

68
One arc-second is 1/60 arc-minutes, 1/602 degrees, 1/(57.3×3600) radians, or 1/60 of the largest apparent
diameters of Venus or Jupiter in the night sky.

- 370 -
 The maximum bit rate that can be communicated over an optical link is not governed by the
Eb > ~10-20 Joules-per-bit limit characteristic of radio systems, however, but rather by the
number of photons the receiver requires per bit of information, perhaps ~10 for a typical good
system. Each photon has energy E = hf Joules. Thus to receive R bits/second might require
received power of:

Pr = E b R ≅ 10hfR [W] (optical rate approximation) (12.1.3)

where h is Planck’s constant (6.624×10-34) and f is photon frequency [Hz]. Clever design can
enable many bits to be communicated per photon, as discussed in the following section.

12.1.4 Examples of optical communications systems

Three examples illustrate several of the issues inherent in optical communications systems: a
trans-oceanic optical fiber cable, an optical link to Mars, and an incoherent intra-office link
carrying computer information.

First consider a trans-oceanic optical fiber. Section 12.2.2 discusses losses in optical fibers,
which can be as low as ~0.2 dB/km near 1.5-micron wavelength (f ≅ 2×1014 Hz). To ensure the
signal (zeros and ones) remains unambiguous, each link of an R = 1-Gbps fiber link must deliver
to its receiver or amplifier more than ~10hfR watts, or ~10×6×10-34×2×1014×109 ≅ 1.2×10-9
watts; a more typical design might deliver ~10-6 watts because errors accumulate and equipment
can degrade. If one watt is transmitted and10-6 watts is received, then the associated 60-dB loss
corresponds to 300 km of fiber propagation between optical amplifiers, and perhaps ~20
amplifiers across the Atlantic Ocean per fiber. In practice, erbium-doped fiber amplifiers,
discussed in Section 12.3.1, are now spaced approximately 80 km apart.

Next consider an optical link communicating between Earth and astronauts on Mars.
Atmospheric diffraction or “seeing” limits the focusing ability of terrestrial telescopes larger than
~10 cm, but Mars has little atmosphere. Therefore a Martian optical link might employ the
equivalent of a one-square-meter optical telescope on Mars and the equivalent of 10-cm-square
optics on Earth. It might also employ a one-watt laser transmitter on Earth operating at 0.5­
micron wavelength, in the visible region. The nominal link and rate equations, (12.1.1) and
(12.1.3), yield the maximum data rate R possible at a range of ~1011 meters (approximate closest
approach of Mars to Earth):

R = Pr E b ≅ (G t Pt 4πr 2 )A e 10hf [bits s-1] (12.1.4)

The gain Gt of the transmitter given by (12.1.2) is Gt ≅ 4πA/λ2 ≅ 5×1011, where A ≅ (0.1)2 and λ
≅ 5×10-7 [m]. The frequency f = c/λ = 3×108/[5×10-7] = 6×1014. Therefore (12.1.4) becomes:

{ }
R ≅ [5×1011 ×1] ⎣⎡4π(1011 ) ⎦⎤ {1 [10 × 6.624 ×10−34 × 6×1014 ]} ≅ 1 Mbps
2
(12.1.5)

- 371 -
 Table 11.4.1 suggests that this data rate is adequate for full-motion video of modest quality.
The delay of the signal each way is τ = r/c = 1011/[3×108] seconds ≅ 5.6 minutes, impeding
conversation. This delay becomes several times greater when Mars is on the far side of the sun
from Earth, and the data rate R would then drop by more than a factor of ten.

This 1-Mbps result (12.1.5) assumed 10 photons were required per bit of information.
However this can be reduced below one photon per bit by using pulse-position modulation.
Suppose ~106 1-nsec 10-photon pulses were received per second, where each pulse could arrive
in any of 1024 time slots because the ratio of pulse width to average inter-pulse spacing is 1024.
This timing information conveys ten bits of information per pulse because log21024 = 10. Since
each 10-photon pulse conveys 10 bits of information, the average is one bit per photon received.
With more time slots still fewer photons per bit would be required. If a tunable laser can
transmit each pulse at any of 1024 colors, for example, then another factor of 10 can be achieved.
Use of both pulse position and pulse-frequency modulation can permit more than 10 bits to be
communicated per photon on average.

The final example is that of a 1-mW laser diode transmitting digitally modulated light at λ =
5×10-7 [m] isotropically within a large office over ranges r up to 10 meters, where the light might
travel directly to the isotropic receiver or bounce off walls and the ceiling first. Such optical
communications systems might link computers, printers, personal digital assistants (pda’s), and
other devices within the room. In this case Gt = 1 and Ae = Gλ2/4π = (5×10-7)2/4π ≅ 2×10-14
[m2]. The maximum data rate R can again be found using (12.1.4):

( )
R = Pr E b ≅ 1×10-3 4π102 (2×10-14 ) (10×6.6×10-34 ×6×1014 ) ≅ 0.004 ⎡⎣bits s-1⎤⎦ (12.1.6)

The fact that we can send 106 bits per second to Mars with a one-watt transmitter, but only 4
millibits per second across a room with a milliwatt, may conflict with intuition.

The resolution of this seeming paradox lies in the assumption that the receiver in this
example is a single mode device like that of typical radio receivers or the Martian optical
receiver considered above. If this room-link receiver were isotropic and intercepted only a single
mode, its effective area Ae given by (12.1.2) would be 2×10-14 [m2]. The tiny effective area of
such low-gain coherent optical antennas motivates use of incoherent photodetectors instead,
which respond well to the total photon flux from all directions of arrival. For example, intra-
room optical links of this type are commonly used for remote control of many consumer
electronic devices, but with a much larger multimode antenna (photodiode) of area
A ≅ 2×10-6 [m2] instead of 2×10-14. This “antenna” is typically responsive to all photons
impacting its area that arrive within roughly one steradian. That is, a photodetector generally
intercepts all photons impacting it, even though those photons are incoherent with each other.
Thus the solution (12.1.6) is increased by a factor of 10-6/10-14 if a two-square-millimeter
photodetector replaces the single-mode antenna, and R then becomes 0.4 Mbps, which is more
capacity than normally required. In practice such inexpensive area detectors are noisier and
require orders of mangitude more photons per bit. Better semiconductor detectors can achieve
10 photons per bit or better, however, particularly at visible wavelengths and if stray light at
other wavelengths is filtered out.

- 372 -
12.2 Optical waveguides

12.2.1 Dielectric slab waveguides

Optical waveguides such as optical fibers typically trap and guide light within rectangular or
cylindrical boundaries over useful distances. Rectangular shapes are easier to implement on
integrated circuits, while cylindrical shapes are used for longer distances, up to 100 km or more.
Exact wave solutions for such structures are beyond the scope of this text, but the same basic
principles are evident in dielectric slab waveguides for which the derivations are simpler.
Dielectric slab waveguides consist of an infinite flat dielectric slab of thickness 2d and
permittivity ε imbedded in an infinite medium of lower permittivity εo, as suggested in Figure
12.2.1(a) for a slab of finite width in the y direction. For simplicity we here assume μ = μo
everywhere, which is usually the case in practice too.

Ey(x)
(a) (b)
x
TE1 TE2 TE3
+d
x
θ > θc
y z 2d 0 0 0
Slab ε > εo
ε > εoutside
-d
Ey εoutside

Figure 12.2.1 Dielectric slab waveguide and TE mode structure.

As discussed in Section 9.2.3, uniform plane waves within the dielectric are perfectly
reflected at the slab boundary if they are incident beyond the critical angle θc = sin-1(cε/co), where
cε and co are the velocities of light in the dielectric and outside, respectively. Such a wave and
its perfect reflection propagate together along the z axis and form a standing wave in the
orthogonal x direction. Outside the waveguide the waves are evanescent and decay
exponentially away from the guide, as illustrated in Figure 12.2.2. This figure portrays the fields
inside and outside the lower half of a dielectric slab having ε > εo; the lower boundary is at x = 0.
The figure suggests two possible positions for the upper slab boundary that satisfy the boundary
conditions for the TE1 and TE2 modes. Note that the TE1 mode waveguide can be arbitrarily thin
relative to λ and still satisfy the boundary conditions. The field configurations above the upper
boundary mirror the fields below the lower boundary, but are not illustrated here. These
waveguide modes are designated TEn because the electric field is only transverse to the direction
of propagation, and there is part of n half-wavelengths within the slab. The orthogonal modes
(not illustrated) are designated TMn.

- 373 -
 x standing wave in x direction

upper boundary × • × •
× × • • • × × × • • direction of
for TE2 mode
× • × • propagation
E(t)
upper boundary
for TE1 mode • × • ×
• • × × × • • • × × ε, μ z
lower boundary • y × • × εo, μo
R e {S}

H(t) × • × I m {S}
λz/2 evanescent wave

Figure 12.2.2 Fields in dielectric slab waveguides for TEn modes.

The fields inside a dielectric slab waveguide have the same form as (9.3.6) and (9.3.7) inside
parallel-plate waveguides, although the boundary positions are different; also see Figures 9.3.1
and 9.3.3. If we define x = 0 at the axis of symmetry, and the thickness of the guide to be 2d,
then within the guide the electric field for TE modes is:

E = ŷ Eo {sink x x or cosk x x}e− jk z z for x ≤ d (12.2.1)

The fields outside are the same as for TE waves incident upon dielectric interfaces beyond the
critical angle, (9.2.33) and (9.2.34):

E = ŷ E1e−αx − jk z z for x ≥ d (12.2.2)

E = {- or +} ŷ E1 +αx − jk z z for x ≤ −d (12.2.3)

The first and second options in braces correspond to anti-symmetric and symmetric TE modes,
respectively. Since the waves decay away from the slab, α is positive. Faraday’s law in
combination with (12.2.1), (12.2.2), and (12.2.3) yields the corresponding magnetic field inside
and outside the slab:

H = ⎡⎣ x̂ k z {sin k x x or cos k x x}
for x ≤ d (12.2.4)
+ ẑ jk x {cos k x x or sin k x x}⎤⎦ ( E o ωμo ) e− jk z z

H = −( x̂ k z + ẑ jα)(E1 ωμo )e−αx − jk z z for x ≥ d (12.2.5)

- 374 -
 H = {+ or -}( xˆ k z − zˆ jα)(E1 ωμo )eαx − jk z z for x ≤ −d (12.2.6)

The TE1 mode has the interesting property that it approaches TEM behavior as ω → 0 and
the decay length approaches infinity; most of the energy is then propagating outside the slab
even though the mode is guided by it. Modes with n ≥ 2 have non-zero cut-off frequencies.
There is no TM mode that propagates for f→0 in dielectric slab waveguides, however.

Although Figure 12.2.1(a) portrays a slab with an insulating medium outside, the first option
in brackets {•} for the field solutions above is also consistent for x > 0 with a slab located 0 < x
< d and having a perfectly conducting wall at x = 0; all boundary conditions are matched; these
are the anti-symmetric TE modes. This configuration corresponds, for example, to certain
optical guiding structures overlaid on conductive semiconductors.

To complete the TE field solutions above we need additional relations between Eo and E1,
and between kx and α. Matching E at x = d for the symmetric solution [cos kxx in (12.2.1)]
yields:

ŷ Eo cos(k x d)e− jk z z = ŷ E1e−αd − jk z z (12.2.7)

Matching the parallel ( ẑ ) component of H at x = d yields:

− zˆ jk x sin(k x d)(Eo ωμo )e− jk z z = − ẑ jα(E1 ωμo )e−αd − jk z z (12.2.8)

The guidance condition for the symmetric TE dielectric slab waveguide modes is given by the
ratio of (12.2.8) to (12.2.7):

k x d tan(k x d) = αd (slab guidance condition) (12.2.9)

Combining the following two dispersion relations and eliminating kz can provide the needed
additional relation (12.2.12) between kx and α:

k z2 + k 2x = ω2μo ε (dispersion relation inside) (12.2.10)

k 2z − α 2 = ω2μo εo (dispersion relation outside) (12.2.11)

k 2x + α 2 = ω2 (μo ε − μo εo ) > 0 (slab dispersion relation) (12.2.12)

By substituting into the guidance condition (12.2.9) the expression for α that follows from
the slab dispersion relation (12.2.12) we obtain a transcendental guidance equation that can be
solved numerically or graphically:

- 375 -
 ( )
0.5
tank x d = ⎡⎣ω2μo (ε − εo )d 2 k 2x d 2 ⎤⎦ −1 (guidance equation) (12.2.13)

Figure 12.2.3 plots the left- and right-hand sides of (12.2.13) separately, so the modal solutions
are those values of kxd for which the two families of curves intersect.

tan kxd = ([ω2μo(ε - εo)d2/kx2d2] – 1)0.5

Increasing ω

No trapped
solutions if
α<0
kxd
0 π 2π
No trapped
solutions
TE1 TE3 TE5
if α > 0
mode mode mode

Figure 12.2.3 TE modes for a dielectric slab waveguide.

Note that the TE1 mode can be trapped and propagate at all frequencies, from nearly zero to
infinity. At low frequencies the waves guided by the slab have small values of α and decay very
slowly away from the slab so that most of the energy is actually propagating in the z direction
outside the slab rather than inside. The value of α can be found from (12.2.12), and it
approaches zero as both kxd and ω approach zero.

The TE3 mode cannot propagate near zero frequency however. Its cutoff frequency ωTE3
occurs when kxd = π, as suggested by Figure 12.2.3; ωTE3 can be determined by solving (12.2.12)
for this case. This and all higher modes cannot be trapped at low frequencies because then the
plane waves that comprise them impact the slab wall at angles beyond θc that permit escape. As
ω increases, more modes can propagate. Figures 12.2.2 and 12.2.1(b) illustrate symmetric TE1
and TE3 modes, and the antisymmetric TE2 mode. Similar figures could be constructed for TM
modes.

These solutions for dielectric-slab waveguides are similar to the solutions for optical fibers,
which instead take the form of Bessel functions because of their cylindrical geometry. In both
cases we have lateral standing waves propagating inside and evanescent waves propagating
outside.

12.2.2 Optical fibers

An optical fiber is generally a very long solid glass wire that traps lightwaves inside as do the
dielectric slab waveguides described in Section 12.2.1. Fiber lengths can be tens of kilometers or

- 376 -
more. Because the fiber geometry is cylindrical, the electric and magnetic fields inside and
outside the fiber are characterized by Bessel functions, which we do not address here. These
propagating electromagnetic fields exhibit lateral standing waves inside the fiber and
evanescence outside. To minimize loss the fiber core is usually overlaid with a low-permittivity
glass cladding so that the evanescent decay also occurs within low-loss glass.

A typical glass optical fiber transmission line is perhaps 125 microns in diameter with a
high-permittivity glass core having diameter ~6 microns. The core permittivity ε + Δε is
typically ~2 percent greater than that of the cladding (ε). If the lightwaves within the core
impact the cladding beyond the critical angle θc, where:

θc = sin −1(ε (ε + Δε)) (12.2.14)

then these waves are perfectly reflected and trapped. The evanescent waves inside the cladding
decay approximately exponentially away from the core to negligible values at the outer cladding
boundary, which is often encased in plastic about 0.1 mm thick that may be reinforced. Graded-
index fibers have a graded transition in permittivity between the core and cladding. Some fibers
propagate multiple modes that travel at different velocities so as to interfere at the output and
limit information extraction (data rate). Multiple fibers are usually bundled inside a single cable.
Figure 12.2.4 suggests the structure of a typical fiber.

glass core ε2 = ε1 + Δε 125 μm

6 μm glass cladding ε1 Δε / ε1 ≅ 0.02

Figure 12.2.4 Typical clad optical fiber.

Figure 12.2.5 shows four common forms of optical fiber; many others exist. The multimode
fiber is thicker and propagates several modes, while the single-mode fiber is so thin that only one
mode can propagate. The diameter of the core determines the number of propagating modes. In
all cylindrical structures, even single-mode fibers, both vertically and horizontally polarized
waves can propagate independently and therefore may interfere with each other when detected at
the output. If a single-mode fiber has an elliptical cross-section, one polarization can be made to
escape so the signal becomes pure. That is, one polarization decays more slowly away from the
core so that it sees more of the absorbing material that surrounds the cladding.

Multimode Single-mode Clad Single polarization

Figure 12.2.5 Types of optical fiber.

- 377 -
 The initial issue faced in the 1970’s by designers of optical fibers was propagation loss.
Most serious was absorption due to residual levels of impurities in the glass, so much research
and development involved purification. Water posed a particularly difficult problem because
one of its harmonics fell in the region where attenuation in glass was otherwise minimum, as
suggested in Figure 12.2.6.

Attenuation (dB km-1)

10 Infrared absorption
dominates loss
H2O
1
Rayleigh 20 THz
scattering
dominates λ(microns)
0.1
1 1.8

Figure 12.2.6 Loss mechanisms in optical fibers.

At wavelengths shorter than ~1.5 microns the losses are dominated by Rayleigh scattering of the
waves from the random fluctuations in glass density on atomic scales. These scattered waves
exit the fiber at angles less than the critical angle. Rayleigh scattering is proportional to f4 and
occurs when the inhomogeneities in ε are small compared to λ/2π. Inhomogeneities in glass
fibers have near-atomic scales, say 1 nm, whereas the wavelength is more than 1000 times larger.
Rayleigh scattering losses are reduced by minimizing unnecessary inhomogeneities through glass
purification and careful mixing, and by decreasing the critical angle. Losses due to scattering by
rough fiber walls are small because drawn glass fibers can be very smooth and little energy
impacts the walls.

At wavelengths longer than ~1.5 microns the wings of infrared absorption lines at lower
frequencies begin to dominate. This absorption is due principally to the vibration spectra of
inter-atomic bonds, and is unavoidable. The resulting low-attenuation band centered near 1.5­
microns between the Rayleigh and IR attenuating regions is about 20 THz wide, sufficient for a
single fiber to provide each person in the U.S.A. with a bandwidth of 20×1012/2.5×108 = 80 kHz,
or 15 private telephone channels! Most fibers used for local distribution do not operate
anywhere close to this limit for lack of demand, although some undersea cables are pushing
toward it.

The fibers are usually manufactured first as a preform, which is a glass rod that
subsequently can be heated at one end and drawn into a fiber of the desired thickness. Preforms
are either solid or hollow. The solid ones are usually made by vapor deposition of SiO2 and
GeO2 on the outer surface of the initial core rod, which might be a millimeter thick. By varying
the mixture of gases, usually Si(Ge)Cl4 + O2 ⇒ Si(Ge)O2 + 2Cl2, the permittivity of the
deposited glass cladding can be reduced about 2 percent below that of the core. The boundary
between core and cladding can be sharp or graded in a controlled way. Alternatively, the
preform cladding is large and hollow, and the core is deposited from the inside by hot gases in

- 378 -
the same way; upon completion there is still a hole through the middle of the fiber. Since the
core is small compared to the cladding, the preforms can be made more rapidly this way. When
the preform is drawn into a fiber, any hollow core vanishes. Sometimes a hollow core is an
advantage. For example, some newer types of fibers have cores with laterally-periodic lossless
longitudinal hollows within which much of the energy can propagate.

Another major design issue involves the fiber dispersion associated with frequency-
dependent phase and group velocities, where the phase velocity vp = ω/k. If the group velocity
vg, which is the velocity of the envelope of a narrowband sinusoid, varies over the optical
bandwidth, then the signal waveform will increasingly distort as it propagates because the faster
moving frequency components of the envelope will arrive early. For example, a digital pulse of
light that lasts T seconds is produced by multiplying a boxcar modulation envelope (the T-
second pulse shape) by the sinusoidal optical carrier, so the frequency spectrum is the
convolution of the spectrum for the sinusoid (a spectral impulse) and the spectrum for a boxcar
pulse (∝ [sin(2πt/T)]/[2πt/T]). The outermost frequencies suffer from dispersion the most, and
these are primarily associated with the sharp edges of the pulse.

The group velocity vg derived in (9.5.20) is the slope of the dispersion relation at the optical
frequency of interest:

−1
v g = (∂k ∂ω) (12.2.15)

Figure 12.2.7 illustrates the dispersion relation for three different modes; the higher order modes
propagate information more slowly.

ω TE5
ωco_TE5
TE3 slope
slope = vgcladding inflection
point
ωco_TE3 slope = vgcore
TE1

radiation mode slope = vg

0
wavenumber kz

Figure 12.2.7 Group velocities for optical fiber modes.

The group velocity vg is the slope of the ω(k) relation and is bounded by the slopes associated
with the core (vgcore) and with the cladding (vgcladding), where the cladding is assumed to be
infinite. The figure has greatly exaggerated the difference in the slope between the core and
cladding for illustrative purposes.

- 379 -
 A dispersive line eventually transforms a square optical pulse into a long “frequency
chirped” pulse with the faster propagating frequencies in the front and the slower propagating
frequencies in the back. This problem can be minimized by carefully choosing combinations of:
1) the dispersion n(f) of the glass, 2) the permittivity contour ε(r) in the fiber, and 3) the optical
center frequency fo. Otherwise we must reduce either the bandwidth of the signal or the length
of the fiber. To increase the distance between amplifiers the dispersion can be compensated
periodically by special fibers or other elements with opposite dispersion.

Pulses spread as they propagate over distance L because their outermost frequency
components ω1 and ω2 = ω1+Δω have arrival times at the output separated by:

( ) (
Δt = L vg1 − L vg2 = L⎡d vg−1 dω⎤ Δω = L d 2 k dω2 Δω
⎣ ⎦
) (12.2.16)

where vgi is the group velocity at ωi (12.2.15). Typical pulses of duration Tp have a bandwidth
Δω ≅ Tp-1, so brief pulses spread faster. The spread Δt is least at frequencies where d2k/dω2 ≅ 0,
which is near the representative slope inflection point illustrated in Figure 12.2.7.

This natural fiber dispersion can, however, help solve the problem of fiber nonlinearity.
Since attenuation is always present in the fibers, the amplifiers operate at high powers, limited
partly by their own nonlinearities and those in the fiber that arise because ε depends very slightly
on the field strength E. The effects of non-linearities are more severe when the signals are in the
form of isolated high-energy pulses. Deliberately dispersing and spreading the isolated pulses
before amplifying and introducing them to the fiber reduces their peak amplitudes and the
resulting nonlinear effects. This pre-dispersion is made opposite to that of the fiber so that the
fiber dispersion gradually compensates for the pre-dispersion over the full length of the fiber.
That is, if the fiber propagates high frequencies faster, then those high frequency components are
delayed correspondingly before being introduced to the fiber. When the pulses reappear in their
original sharp form at the far end of the fiber their peak amplitudes are so weak from natural
attenuation that they no longer drive the fiber nonlinear.

Example 12.2A
If 10-ps pulses are used to transmit data at 20 Gbps, they would be spaced 5×10-10 sec apart and
would therefore begin to interfere with each other after propagating a distance Lmax sufficient to
spread those pulses to widths of 50 ps. A standard single-mode optical fiber has dispersion
d2k/dω2 of 20 ps2/km at 1.5 μm wavelength. At what distance Lmax will such 10-ps pulses have
broadened to 50 ps?

Solution: Using (12.2.16) and Δω ≅ Tp-1 we find:

Lmax = Δt/[Δω(d2k/dω2)] = 50 ps × 10 ps /(20 ps2/km) = 25 km

Thus we must slow this fiber to 10 Gbps if the amplifiers are 50 km apart.

- 380 -
12.3 Lasers

12.3.1 Physical principles of stimulated emission and laser amplification

Lasers (Light Amplification by Stimulated Emission of Radiation) amplify electromagnetic

waves at wavelengths ranging from radio to ultraviolet and x-rays. They were originally called
masers because the first units amplified only microwaves. Lasers can also oscillate when the
amplified waves are reflected back into the device. The physical principles are similar at all
wavelengths, though the details differ. Laser processes can occur in solids, liquids, or gases.

Lasers have a wide and growing array of applications. For example, optical fiber
communications systems today commonly use Erbium-doped fiber amplifiers (EDFA’s) that
amplify ~1.5-micron wavelength signals having bandwidths up to ~4 THz. Semiconductor, gas,
and glass fiber laser amplifiers are also used to communicate within single pieces of equipment
and for local fiber or free-space communications. Lasers also generate coherent beams of light
used for measuring distances and angles; recording and reading data from memory devices such
as CD’s and DVD’s; and for cutting, welding, and shaping materials, including even the human
eye. Laser pointers have been added to pocket pens while higher-power industrial units can cut
steel plates several inches thick. Weapons and laser-driven nuclear fusion reactions require still
higher-power lasers. Peak laser pulse powers can exceed 1015 watts, a thousand times the total
U.S. electrical generating capacity of ~5×1011 watts. The electric field strengths within a focal
spot of <100-micron diameter can strip electrons from atoms and accelerate them to highly
relativistic velocities within a single cycle of the radiation. The roles of lasers in science,
medicine, industry, consumer products, and other fields are still being defined.

Laser operation depends intimately upon the quantum nature of matter and the fact that
charges trapped in atoms and molecules generally move at constant energy without radiating.
Instead, transitions between atomic or molecular energy states occur abruptly, releasing or
absorbing a photon.69 This process and lasers can fortunately be understood semi-classically
without reference to a full quantum description.

Electrons within atoms, molecules, and crystals occupy discrete energy states; the lower
energy states are preferentially occupied. Energy states can also be vibrational, rotational,
magnetic, chemical, nuclear, etc.70 The number of possible states greatly exceeds those that are
occupied.

69
Alternatively, acoustic phonons with energy hf can be released or absorbed, or an additional molecular or atomic
state transition can occur to conserve energy. Phonons are acoustic quanta associated with mechanical waves in
materials. Optical transitions can also absorb or emit two photons with total energy equal to Ε2 − Ε1, although such
two-photon transitions are much less likely.
70
The distances between adjacent nuclei in molecules can oscillate sinusoidally with quantized amplitudes and
frequencies characterisitic of each vibrational state. Isolated molecules can spin at specific frequencies
corresponding to various rotational energy states. Electron spins and orbits together have magnetic dipole moments
that align with or oppose an applied magnetic field to a quantized degree. Atoms bond to one another in quantized
ways having specific chemical consequences. Nuclear magnetic moments can also align with other atomic or
molecular magnetic moments in quantized ways corresponding to discrete energy states.

- 381 -
 For example, as illustrated in Figure 12.3.1(a), an electron trapped in an atom, molecule, or
crystal with energy E1 can be excited into any vacant higher-energy state (E2) by absorbing a
photon of frequency f and energy ΔE where:

ΔE = E 2 − E1 = hf [J] (12.3.1)

The constant h is Planck’s constant (6.625×10-34 [Js]), and the small circles in the figure
represent electrons in specific energy states.

(a) (b) (c)

E2 E2 E2
Before:
E1 E1 E1

E2 E2 E2
After:
E1 E1 E1
Photon Spontaneous Stimulated
absorption emission emission
Figure 12.3.1 Photon absorption, spontaneous emission, and stimulated emission.

Figures 12.3.1(b) and (c) illustrate two additional basic photon processes: spontaneous emission
and stimulated emission. Photon absorption (a) occurs with a probability that depends on the
photon flux density [Wm-2], frequency [Hz], and the cross-section for the energy transition of
interest. Spontaneous emission of photons (b) occurs with a probability A that depends only on
the transition, as discussed below. Stimulated emission (c) occurs when an incoming photon
triggers emission of a second photon; the emitted photon is always exactly in phase with the first,
and propagates in the same direction. Laser action depends entirely on this third process of
stimulated emission, while the first two processes often weaken it.

The net effect of all three processes—absorption, spontaneous emission, and stimulated
emission—is to alter the relative populations, N1 and N2, of the two energy levels of interest. An
example exhibiting these processes is the Erbium-doped fiber amplifiers commonly used to
amplify optical telecommunications signals near 1.4-micron wavelength on long lines. Figure
12.3.2 illustrates how an optical fiber with numerous atoms excited by an optical pump
(discussed further below) can amplify input signals at the proper frequency. Since the number of
excited atoms stimulated to emit is proportional to the input wave intensity, perhaps only one
atom might be stimulated to emit initially (because the input signal is weak), producing two in­
phase photons—the original plus the one stimulated. These two then propagate further
stimulating two emissions so as to yield four in-phase photons.

- 382 -
 Optical pump (repopulates level 2)
Optical fiber

2
hf hf ⇒ 2hf ⇒ 4hf ⇒ 6hf
1
input amplification, exponential growth amplification, linear
growth
[Each is a separate atom or molecule;
need N2 > N1 for amplification] Limited by n2 replacement-rate

Figure 12.3.2 Optical fiber amplifier with exponential and linear growth.

This exponential growth continues until the pump can no longer empty E1 and refill E2 fast
enough; as a result absorption [m-1] approaches emission [m-1] as N1 approaches N2 locally. In
this limit the increase in the number of photons per unit length is limited by the number np of
electrons pumped from E1 to E2 per unit length. Thereafter the signal strength then increases
only linearly with distance rather than exponentially, as suggested in Figure 12.3.3; the power
increase per unit length then approaches nphf [Wm-1].

P(z)

exponential growth

linear growth
z

Figure 12.3.3 Exponential and linear growth regimes in optical fiber amplifiers.

Simple equations characterize this process quantitatively. If E1 < E2 were the only two
levels in the system, then:

dN 2 dt = −A 21N 2 − I21B21( N 2 − N1) ⎡s −1⎤ (12.3.2)

⎣ ⎦

The probability of spontaneous emission from E2 to E1 is A21, where τ21 = 1/A21 is the 1/e
lifetime of state E2. The intensity of the incident radiation at f = (E2-E1)/h [Hz] is:

I21 = F21hf ⎡Wm −2 ⎤ (12.3.3)

⎣ ⎦

where F21 is the photon flux [photons m-2s-1] at frequency f. The right-most term of (12.3.2)
corresponds to the difference between the number of stimulated emissions (∝ N2) and
absorptions (∝ N1), where the rate coefficients are:

- 383 -
 ( ⎣ )
B21 = A 21 π2c 2 hω3n 2 ⎡m 2 J −1⎤
⎦
(12.3.4)

A 21 = 2ω3D221 hεc3 ⎡s −1⎤ (12.3.5)

⎣ ⎦

In these equations n is the index of refraction of the fiber and D21 is the quantum mechanical
electric or magnetic dipole moment specific to the state-pair 2,1. It is the sharply varying values
of the dipole moment Dij from one pair of levels to another that makes pumping practical, as
explained below.

Laser amplification can occur only when N2 exceeds N1, but in a two-level system no pump
excitation can accomplish this; even infinitely strong incident radiation I21 at the proper
frequency can only equalize the two populations via (12.3.2).71 Instead, three- or four-level
lasers are generally used. The general principle is illustrated by the three-level laser of Figure
12.3.4(a), for which the optical laser pump radiation driving the 1,3 transition is so strong that it
roughly equalizes N1 and N3. The key to this laser is that the spontaneous rate of emission A32
>> A21 so that all the active atoms quickly accumulate in the metastable long-lived level 2 in the
absence of stimulation at f21. This generally requires D32 >> D21, and finding materials with such
properties for a desired laser frequency can be challenging.

(a) (b) (c) A32

3 3 3
Large A32 A31 A32 2
Pump 2 Pump Pump A41 4
lasing 2
1 1
1
A21 ~B21I21 lasing

Figure 12.3.4 Energy diagrams for three- and four-level lasers.

Since it requires hf13 Joules to raise each atom to level 3, and only hf21 Joules emerges as
amplified additional radiation, the power efficiency η (power out/power in) cannot exceed the
intrinsic limit ηI = f21/f31. In fact the efficiency is lowered further by a factor of ηA
corresponding to spontaneous emission from level 3 directly to level 1, bypassing level 2 as
suggested in Figure 12.3.4(b), and to the spontaneous decay rate A21 which produces radiation
that is not coherent with the incoming signal and radiates in all directions. Finally, only a
fraction ηp of the pump photons are absorbed by the transition 1→2. Thus the maximum power
efficiency for this laser in the absence of propagation losses is:

η = ηIηA ηp (12.3.6)

71
Two-level lasers have been built, however, by physically separating the excited atoms or molecules from the
unexcited ones. For example, excited ammonia molecules can be separated from unexcited ones by virtue of their
difference in deflection when a beam of such atoms in vacuum passes through an electric field gradient.

- 384 -
 Figure 12.3.4(c) suggests a typical design for a four-level laser, where both A32 and A41 are
much greater than A24 or A21 so that energy level 2 is metastable and most atoms accumulate
there in the absence of strong radiation at frequency f24 or f21. The strong pump radiation can
come from a laser, flash lamp, or other strong radiation source. Sunlight, chemical reactions,
nuclear radiation, and electrical currents in gases pump some systems.

The ω3 dependence of A21 (12.3.5) has a profound effect on maser and laser action. For
example, any two-level maser or laser must excite enough atoms to level 2 to equal the sum of
the stimulated and spontaneous decay rates. Since the spontaneous decay rate increases with ω3,
the pump power must also increase with ω3 times the energy hf of each excited photon. Thus
pump power requirements increase very roughly with ω4, making construction of x-ray or
gamma-ray lasers extremely difficult without exceptionally high pump powers; even ultraviolet
lasers pose a challenge. Conversely, at radio wavelengths the spontaneous rates of decay are so
extremely small that exceedingly low pump powers suffice, as they sometimes do in the vast
darkness of interstellar space.

Many types of astrophysical masers exist in low-density interstellar gases containing H2O,
OH, CO, and other molecules. They are typically pumped by radiation from nearby stars or by
collisions occurring in shock waves. Sometimes these lasers radiate radially from stars,
amplifying starlight, and sometimes they spontaneously radiate tangentially along linear
circumstellar paths that have minimal relative Doppler shifts. Laser or maser action can also
occur in darkness far from stars as a result of molecular collisions. The detailed frequency,
spatial, and time structures observed in astrophysical masers offer unique insights into a wide
range of astrophysical phenomena.

Example 12.3A
What is the ratio of laser output power to pump power for a three-level laser like that shown in
Figure 12.3.4(a) if: 1) all pump power is absorbed by the 1→3 transition, 2) N2 >> N1, 3)
A21/I21B21 = 0.1, 4) A31 = 0.1A32, and 5) f31 = 4f21?

Solution: The desired ratio is the efficiency η of (12.3.6) where the intrinsic efficiency is
ηI = f21/f31 = 0.25, and the pump absorption efficiency ηp = 1. The efficiency ηA is
less than unity because of two small energy losses: the ratio A31/A32 = 0.1, and the
ratio A21/I21B21 = 0.1. Therefore ηA = 0.92 = 0.81, and η = ηIηAηp = 0.25 × 0.81 ≅
0.20.

12.3.2 Laser oscillators

Laser amplifiers oscillate nearly monochromatically if an adequate fraction of the amplified

signal is reflected back to be amplified further. For example, the laser oscillator pictured in
Figure 12.3.5 has parallel mirrors at both ends of a laser amplifier, separated by L meters. One
mirror is perfect and the other transmits a fraction T (say ~0.1) of the incident laser power. The
roundtrip gain in the absence of loss is e2gL. This system oscillates if the net roundtrip gain at
any frequency exceeds unity, where round-trip absorption (e-2αL) and the partially transmitting
mirror account for most loss.

- 385 -
 Perfect mirror
Partially reflecting mirror,
Amplifier power transmittance T
P+ Output laser
beam, Pout [W]
L

Figure 12.3.5 Laser oscillator.

Amplifiers at the threshold of oscillation are usually in their exponential region, so this net
roundtrip gain exceeds unity when:

(1 − T)e2(g −α)L > 1 (12.3.7)

Equation (12.3.7) implies e2(g - α)L ≥ (1-T)-1 for oscillation to occur. Generally the gain g per
meter is designed to be as high as practical, and then L and T are chosen to be consistent with the
desired output power. The pump power must be above the minimum threshold that yields g > α.

The output power from such an oscillator is simply Pout = TP+ watts, and depends on pump
power Ppump and laser efficiency. Therefore:

P+ = Pout T = ηPpump T (12.3.8)

Thus small values of T simply result in higher values of P+, which can be limited by internet
breakdown or failure.

One approach to obtaining extremely high laser pulse powers is to abruptly increase the Q
(reverberation) of the laser resonator after the pump source has fully populated the upper energy
level. To prevent lasing before that level is fully populated, strong absorption can be introduced
in the round-trip laser path to prevent amplification of any stimulated emission. The instant the
absorption ceases, i.e. after Q-switching, the average round-trip gain g of the laser per meter
exceeds the average absorption α and oscillation commences. At high Q values lasing action is
rapid and intense, so the entire upper population is encouraged to emit instantly, particularly if
the lower level can be rapidly emptied. Such a device is called a Q-switched laser. Resonator Q
is discussed further in Section 7.8.

The electronic states of glass fiber amplifiers are usually associated with quantized electron
orbits around the added Erbium atoms, and state transitions simply involve electron transfers
between two atomic orbits having different energies. In contrast, the most common lasers are
laser diodes, which are transparent semiconductor p-n junctions for which the electron energy
transitions occur between the conduction and valence bands, as suggested in Figure 12.3.6.

- 386 -
 (a) (b)
E electrons Electron energy E
Conduction
p-type band Conduction
EF band
EF
Active region E2 - E1 = hf

holes n-type
Valence band z Valence band
k

Figure 12.3.6 Laser diode – a forward-biased p-n junction

bounded by mirrors promoting oscillation.

Parallel mirrors at the sides of the p-n junction partially trap the laser energy, forming an
oscillator that radiates perpendicular to the mirrors; one of the mirrors is semi-transparent.
Strong emission does not occur in any other direction because without the mirrors there is no
feedback. Such lasers are pumped by forward-biasing the diode so that electrons thermally
excited into the n-type conduction band diffuse into the active region where photons can
stimulate emission, yielding amplification and oscillation within the ~0.2-μm thick p-n junction.
Vacancies in the valence band are provided by the holes that diffuse into the active region from
the p-type region. Voltage-modulated laser diodes can produce digital pulse streams at rates
above 100 Mbps.

The vertical axis E of Figure 12.3.6(a) is electron energy and the horizontal axis is position z
through the diode from the p to n sides of the junction. The exponentials suggest the Boltzmann
energy distributions of the holes and electrons in the valence and conduction bands, respectively.
Below the Fermi level, EF, energy states have a high probability of being occupied by electrons;
EF(z) tilts up toward the right because of the voltage drop from the p-side to the n-side. Figure
12.3.6(b) plots electron energy E versus the magnitude of the k vector for electrons (quantum
approaches treat electrons as waves characterized by their wavenumber k), and suggests why
diode lasers can have broad bandwidths: the energy band curvature with k broadens the laser
linewidth Δf. Incoming photons can stimulate any electron in the conduction band to decay to
any empty level (hole) in the valence band, and both of these bands have significant energy
spreads ΔE, where the linewidth Δf ≅ ΔE/h [Hz].

The resonant frequencies of laser diode oscillators are determined by E2 - E1, the linewidth
of that transition, and by the resonant frequencies of the TEM mirror cavity resonator. The width
Δω of each resonance is discussed further later. If the mirrors are perfect conductors that force
E // = 0 , then there must be an integral number m of half wavelengths within the cavity length L
so that mλm = 2L. The wavelength λm' is typically shorter than the free-space wavelength λm due
to the index of refraction n of the laser material. Therefore λm = 2Ln/m = c/fm, and:

f m = cm 2Ln (12.3.9)

- 387 -
For typical laser diodes L and n might be 0.5 mm and 3, respectively, yielding a spacing between
cavity resonances of: c/2Ln = 3×108/(2×10-3×1.5) = 100 GHz, as suggested in Figure 12.3.7(a).
The figure suggests how the natural (atomic) laser line width might accommodate multiple
cavity resonances, or possibly only one.

(a) (b)
Cavity resonances
Laser line shape Saturated,
~100 GHz for line narrowing
diode lasers

f f
Unsaturated
Figure 12.3.7 Line widths and frequencies of the resonances of a cavity laser.

If the amplifier line shape is narrow compared to the spacing between cavity resonances,
then the cavity length L might require adjustment in order to place one of the cavity resonances
on the line center before oscillations occur. The line width of a laser depends on the widths of
the associated energy levels Ei and Ej. These can be quite broad, as suggested by the laser diode
energy bands illustrated in Figure 12.3.6(b), or quite narrow. Similarly, the atoms in an EDFA
are each subject to slightly different local electrical fields due to the random nature of the glassy
structure in which they are imbedded. This results in each atom having slightly different values
for Ei so that EFDA’s amplify over bandwidths much larger than the bandwidth of any single
atom.

Lasers for which each atom has its own slightly displaced resonant frequency due to local
fields are said to exhibit inhomogeneous line broadening. In contrast, many lasers have no such
frequency spread induced by local factors, so that all excited atoms exhibit the same line center
and width; these are said to exhibit homogeneous line broadening. The significance of this
difference is that when laser amplifiers are saturated and operate in their linear growth region,
homogeneously broadened lasers permit the strongest cavity resonance within the natural line
width to capture most of the energy available from the laser pump, suppressing the rest of the
emission and narrowing the line, as suggested in Figure 12.3.7(b). This suppression of weak
resonances is reduced in inhomogeneously broadened lasers because all atoms are pumped
equally and have their own frequency sub-bands where they amplify independently within the
natural line width.

In gases the width of any spectral line is also controlled by the frequency of molecular
collisions. Figure 12.3.8(b) illustrates how an atom or molecule with sinusoidal time variations
in its dipole moment might be interrupted by collisions that randomly reset the phase. An
electromagnetic wave interacting with this atom or molecule would then see a less pure sinusoid.
This new spectral characteristic would no longer be a spectral impulse, i.e., the Fourier transform
of a pure sinusoid, but rather the transform of a randomly interrupted sinusoid, which has the
Lorentz line shape illustrated in Figure 12.3.8(a). Its half-power width is Δf, which is
approximately the collision frequency divided by 2π. The limited lifetime of an atom or

- 388 -
molecule in any state due to the probability A of spontaneous emission results in similar
broadening, where Δf ≅ A/2π; this is called the intrinsic linewidth of that transition.

(a) (b)

absorption/emission
Collisions ⇒ phase changes
Δf

f
fo

Figure 12.3.8 Lorentzian line shape and origins of intrinsic line width.

Example 12.3B
A Q-switched 1-micron wavelength laser of length L = 1 mm is doped with 1018 active atoms all
pumped to their upper state. When the Q switches instantly to 100, approximately what is the
maximum laser power output P [W]? Assume ε = 4εo.

Solution: The total energy released when the Q switches is 1018hf ≅ 1018 × 6.6×10-34 × 3×1014 =
0.20 Joules. If the laser gain is sufficiently high, then a triggering photon originating
near the output could be fully amplified by the time the beam reaches the rear of the
laser, so that all atoms would be excited as that reflected pulse emerges from the front
of the laser. A triggering photon at the rear of the laser would leave some atoms
unexcited. Thus the minimum time for full energy release lies between one and two
transit times τ of the laser, depending on its gain; τ = L/c' = 2L/c = 6.7×10-12. Lower
laser gains may require many transit times before all atoms are stimulated to emit.
Therefore P < ~0.2/(6.7×10-12) ≅ 30 GW.

12.4 Optical detectors, multiplexers, interferometers, and switches

12.4.1 Phototubes

Sensitive radio-frequency detectors typically require at least 10-20 Joules per bit of information,
which roughly corresponds to thousands of photons of energy hf, where Planck’s constant h =
6.625×10-34 Joules Hz-1. This number of photons is sufficiently high that we can ignore most
quantum effects and treat the arriving radio signals as traditional waves. In contrast, many
optical detectors can detect single photons, although more than five photons are typically used to
distinguish each pulse from interference; this requires more energy per bit than is needed at radio
wavelengths. The advantage of long-range optical links lies instead in the extremely low losses
of optical fibers or, alternatively, in the ability of relatively small mirrors or telescopes to focus

- 389 -
energy in extremely small beams so as to achieve much higher gains than can practical radio
antennas.

Typical photon detectors include phototubes and semiconductors. A phototube detects

photons having energies hf > Φ using the photoelectric effect, where Φ is the work function [J] of
the metal surface (cathode) that intercepts the photons. Photons with energies above this
threshold eject an electron from the cathode with typical probabilities η (called the quantum
efficiency) of ~10-30 percent. These ejected electrons are then pulled in vacuum toward a
positively charged anode and contribute to the current I through the load resistor R, as illustrated
in Figure 12.4.1(a). Although early phototubes ejected electrons from the illuminated surface, it
is now common for the metal to be sufficiently thin and transparent that the electrons are emitted
from the backside of the metal into vacuum; the metal is evaporated in a thin layer onto the
interior surface of the tube’s evacuated glass envelope.

(a) Phototube (b) Photomultiplier Tube

~×54 electrons
anode per photon
photon -e
vout ~×5 vout
- -
R ~×5 R
cathode + + -V + +
-V dynodes

Figure 12.4.1 Phototube and photomultiplier tube detectors.

The current I is proportional to the number N of photons incident per second with energies
above Φ:

I = −ηNe [ A ] (12.4.1)

The work functions of most metals are ~2-6 electron volts, where the energy of one electron volt
(e.v.) is –eV = 1.602×10-19 Joules72. Therefore phototubes do not work well for infrared or
longer wavelengths because their energy hf is too small; 2 e.v. corresponds to a wavelength of
0.62 microns and the color red.

Because the charge on an electron is small, the currents I are often too small to induce
voltages across R (see Figure 12.4.1) that exceed the thermal noise (Johnson noise) of the
resistor unless the illumination is bright. Photomultiplier tubes release perhaps 104 electrons per
detected photon so as to overcome this noise and permit each detected photon to be
unambiguously counted. The structure of a typical photomultiplier tube is illustrated in Figure
12.4.1(b). Each photoelectron emitted by the cathode is accelerated toward the first dynode at
~50-100 volts, and gains energy sufficient to eject perhaps five or more low energy electrons
from the dynode that are then accelerated toward the second dynode to be multiplied again. The
72
Note that the energy associated with charge Q moving through potential V is QV Joules, so QV = 1 e.v. = e×1 =
1.602×10-19 Joules.

- 390 -
illustrated tube has four dynodes that, when appropriately charged, each multiply the incident
electrons by ~5 to yield ~54 ≅ 625 electrons at the output for each photon detected at the input.
Typical tubes have more dynodes and gains of ~104-107. Such large current pulses generally
overwhelm the thermal noise in R, so random electron emissions induced by cosmic rays or
thermal effects at the cathode dominate the detector noise. The collecting areas of such tubes
can be enhanced with lenses or mirrors.

12.4.2 Photodiodes

Phototubes are generally large (several cubic inches), expensive, and fragile, and therefore
semiconductor photodiodes are more commonly used. Photodiodes also respond better to visible
and infrared wavelengths and operate at much lower voltages. Figure 12.4.2(a) illustrates the
energy diagram for a typical short-circuited p-n junction between p-type and n-type
semiconductors, where the vertical axis is electron energy E and the horizontal axis is distance z
perpendicular to the planar junction.

(a) E no net current (b) E

backcurrent
donor
acceptor p-type atoms photoelectron
atoms
Fermi
- hf +
n-type vB
level
z photon
Short-circuited diode
i z
(c)
i photocurrent
-vB
v -
+ v = - vB
operating Back-biased diode (vB > 0)
point, sensitive
avalanche
to photons
region

Figure 12.4.2 Semiconductor photodiodes.

The lower cross-hatched area is the valence band within which electrons are lightly bound to
ions, and the upper area is the conduction band within which electrons are free to move in
response to electric fields. The band gap between these regions is ~1.12 electron volts (e.v.) for
silicon, and ranges from 0.16 e.v. for InSb (indium antimonide to ~7.5 e.v. for BN (boron
nitride). In metals there is no such gap and some electrons always reside in the conduction band
and are mobile. Additional discussion of p-n junctions appears in Section 8.2.4.

Electrons move freely in the conduction band, but not if they remain in the valence band.
Most photons entering the junction region with energy greater than the bandgap between the
Fermi level and the lower edge of the conduction band can excite electrons into the conduction

- 391 -
band to enhance device conductivity. In semiconductors the Fermi level is that level
corresponding to the nominal maximum energy of electrons available for excitation into the
conduction band. The local Fermi level is determined by impurities in the semiconductors that
create electron donor or acceptor sites; these sites easily release or capture, respectively, a free
electron. The Fermi level sits just below the conduction band for n-type semiconductors because
donor atoms easily release one of their electrons into the conduction band, as illustrated in Figure
12.4.2(a) and (b). The Fermi level sits just above the valence band for p-type semiconductors
because acceptor atoms easily capture an extra electron from bound states in nearby atoms.

If the p-n junction is short-circuited externally, the Fermi level is the same on both halves,
as shown in Figure 12.4.2(a). Random thermal excitation produces an exponential Boltzmann
distribution in electron energy, as suggested in the figure, the upper tails of which lie in the
conduction band on both halves of the junction. When the device is short-circuited these current
flows from thermal excitations in the p and n halves of the junction balance, and the external
current is zero. If, however, the diode is back-biased by VB volts as illustrated in Figure
12.4.2(b), then the two exponential tails do not balance and a net back-current current flows, as
suggested by the I-V characteristic for a p-n junction illustrated in Figure 12.4.2(c). The back
current for an un-illuminated photodiode approaches an asymptote determined by VB and the
number of thermal electrons excited per second into the conduction band for the p-type
semiconductor. When an un-illuminated junction is forward biased, the current increases
roughly exponentially.

When a p-n junction is operated as a photodiode, it is back-biased so that every detected

photon contributes current flow to the circuit, nearly one electron per photon received. By
cooling the photodiode the thermal contribution to diode current can be reduced markedly so that
the diode becomes more sensitive to dim light. Cooling is particularly important for photodiodes
with the small bandgaps needed for detecting infrared radiation; otherwise the detected infrared
signals must be bright so they exceed the detector noise.

If photodiodes are sufficiently back-biased, they can enter the avalanche region illustrated in
Figure 12.4.2(c), where an excited electron is accelerated sufficiently as it moves through the
semiconductor that it can impact and excite another electron into the conduction band; both
electrons can now accelerate and excite even more electrons, exponentially, until they all exit the
high-field zone so that further excitations are not possible. In response to a single detected
photon such avalanche photodiodes (APD’s) can produce an output pulse of ~104 electrons that
stands out sufficiently above the thermal noise that photons can again be counted individually.
The number of photons detected per second is proportional to input power, and therefore to the
square of the incident electric field strength.

12.4.3 Frequency-multiplexing devices and filters

The major components in fiber-optic communications systems are the fibers themselves and the
optoelectronic devices that manipulate the optical signals, such as detectors (discussed in
Sections 12.4.1–2), amplifiers and sources (Section 12.3), multiplexers and filters (this section),
modulators, mixers, switches, and others (Section 12.4.4). These are assembled to create useful
communications, computing, or other systems.

- 392 -
 A typical wave-division multiplexed (WDM) amplifier is pictured in Figure 12.4.3;
narrowband optical signals of different colors are aggregated at a point of departure and merged
onto a single long fiber by a frequency multiplexer (MUX). Along this fiber extremely
broadband optical amplifiers (OAMPs) are spaced perhaps 80 km apart to sustain the signal
strength. OAMPs today are typically erbium-doped fiber amplifiers (EFDA’s). At the far end
the signal is de-multiplexed into its spectral components, which are then directed appropriately
along separate optical fibers. Before broadband amplifiers were available, each narrow band had
to have separate amplifiers and often separate fibers.

λ� ~75 km λ1
D
λ2 M E λ2
λ3 U M λ3
X U
OAMP OAMP OAMP
λn X λn

Figure 12.4.3 Wave-division multiplexed amplifier.

Such multiplexers can be made using prisms or diffraction gratings that refract or diffract
different colors at different angles, as suggested in Figure 12.4.4(a) and (b); by reciprocity the
same devices can be used either for multiplexing (superimposing multiple frequency bands on
one beam) or demultiplexing (separation of a single beam into multiple bands), depending on
which end of the device receives the input.

The diffraction grating of Figure 12.4.4(b) is typically illuminated by normally incident

uniform plane waves, and consists of closely spaced ruled straight lines where equal-width
stripes typically alternate between transmission and reflection or absorption. Alternate stripes
sometimes differ only in their phase. Each stripe must be more than λ/2 wide, and λ is more
typical. In this case the rays from each transparent stripe 2λ apart will add in phase straight
forward (θ = 0) and at θ = sin-1(λ/2λ) = 30o, exactly analogous to the grating lobes of dipole
array antennas (Section 10.4). Since the stripe separation (2λ here) is fixed, as the frequency f =
c/λ varies, so does θ, thus directing different frequencies toward different angles of propagation,
much like the prism.

Another useful optical device is the Fabry-Perot resonator, which is the optical version of
the TEM resonator illustrated in Figure 7.4.3(a) and explained in Section 7.4.3. For example, an
optical TEM resonator can be fabricated using parallel mirrors with uniform plane waves trapped
between them; the allowed resonator modes have an integral number n of half-wavelengths in the
distance L between the parallel conductors:

nλ n 2 = L (12.4.2)

- 393 -
 (a) (b) first-order
diffraction lobe
f1 input θ(λ)
f2 wave
prism f3 λ
diffraction
grating
(c) mirror 1 (d)
mirror 2 diplexer
modulator mirror 2
x
input
input
output "B" output "A"
beam mirror 1
splitter output "A" output "B"

(e)
++++++ V
input - - - - - - output "A"
modulator

Figure 12.4.4 Optical frequency multiplexers, interferometers, and modulators:

(a) prism, (b) diffraction grating, (c) Michelson interferometer, (d) Mach-Zehnder interferometer,
(e) waveguide Mach-Zehnder interferometer.

Thus the frequency fn of the nth TEM resonance is:

f n =c λ n = nc 2L [Hz] (12.4.3)

and the separation between adjacent resonances is c/2L [Hz]. For example, if L is 1.5 mm, then
the resonances are separated 3×108/0.003 = 100 GHz.

If the input and output mirrors transmit the same small fraction of the power incident upon
them, then the “internal” and external Q’s of this resonator are the same, where the “internal Q”
here (QI) is associated with power escaping through the output mirror and the “external Q” (QE)
is associated with power escaping through the input mirror. As suggested by the equivalent
circuits in Figures 7.4.4–5, there is perfect power transmission through the resonator at resonance
when the internal and external Q’s are the same, provided there are no dissipative losses within
the resonator itself. The width of the resonance is then found from (7.4.45–6):

Δω = ωo QL = 2ωo QE = 2PE w T (12.4.4)

- 394 -
The loaded QL = QE/2 when QE = QI, and PE [W] is the power escaping through the input mirror
when the total energy stored in the resonator is wT [J]. With high reflectivity mirrors and low
residual losses the bandwidth of such a resonator can be made almost arbitrarily narrow. At
optical frequencies the ratio of cavity length L to wavelength λ is also very large. This increases
the ratio of wT to PE proportionally and leads to very high QL and narrow linewidth.

If the medium in a Fabry-Perot interferometer is dispersive, then it can be shown that the
spacing between resonances is vg/2L [Hz], or the reciprocal of the round-trip time for a pulse.
Thus such a resonator filled with an active medium could amplify a single pulse that rattles back
and forth in the resonator producing sharp output pulses with a period 2L/vg. The Fourier
transform of this pulse train is a train of impulses in the frequency domain with spacing vg/2L,
i.e., representing the set of resonant frequencies for this resonator. The resonant modes of such a
mode-locked laser are synchronized, so they can usefully generate pulse trains for subsequent
modulation.

Example 12.4A
What is the ratio of the width Δω of the passband for a Fabry-Perot resonator relative to the
spacing ωi+1 - ωi between adjacent resonances? What power transmission coefficient
T2 = |Et|2/|Ei|2 is required for each mirror in order to produce isolated sharp resonances? What is
the width Δf [Hz] of each resonance?

Solution: The resonance width Δω and spacing are given by (12.4.4) and (12.4.2), respectively,
so:
( )
Δω (ωi +1 − ωi ) = (2PE w T ) (πc L) = ⎡⎣ 2P+ T 2 (2LP+ c)⎤⎦(L πc) = T 2 π < ~ 0.3
Therefore we require T2 < ~1 so that Δω <~ 0.3(ωi +1 − ωi ) .
Δf = Δω 2π = (T 2 π)(ωi+1 − ωi ) 2π = (T 2 π)c 2L [Hz]; it approaches zero as T2/L does.

12.4.4 Interferometers

The Michelson interferometer and the Mach-Zehnder interferometer are important devices
illustrated in Figure 12.4.4(c) and (d), respectively. In both cases an input optical beam is split
by a beam-splitter into two coherent beams that are reflected by mirrors and then recombined
coherently in a second beam-splitter to form two output beams. The intensity of each output
beam depends on whether its two input components added in-phase or out-of-phase. The beam-
splitters are typically dielectric mirrors coated so that half the power is reflected and half
transmitted from the front surface; the rear surface might be anti-reflection coated. Half-silvered
mirrors can also be used.

As the position x of a Michelson mirror varies, the output power varies sinusoidally from
zero, which results when the two beams cancel at the output, to its peak value when the two
beams add in phase. Cancellation requires that the two beams have equal strength. It is
interesting to ask where the input power goes when output “A” is zero; the figure suggests the
answer. The missing power emerges from the other output; the sum of the powers emerging

- 395 -
from the two outputs equals the input power, less any dissipative losses. This requirement for
power conservation translates into a requirement for a specific phase relationship between the
various beams.

Consecutive peaks in output strength occur as the mirror moves λ/2 (typically ~3×10-7 m);
the factor of 1/2 arises because of the round trip taken by the reflected beam. The sinusoidal
output power can generally be measured with sufficient accuracy at optical wavelengths to
determine relative mirror positions x with accuracies of an angstrom (10-10 m), or tiny fractions
thereof; thus the Michelson interferometer is a powerful tool for measuring or comparing
wavelengths and distances. Another important application is measurement of optical spectra.
Since each optical wavelength λ in the input beam produces an additive sinusoidal contribution
to the output power waveform A(x) of period λ/2, the input optical power spectrum is the Fourier
transform of A(2x). Because this technique was first used to analyze infrared spectra, it is called
Fourier transform infrared spectroscopy (FTIR).

If the two output beams in a Mach-Zehnder interferometer add in phase, the output power is
maximized and equals the input power, much like the Michelson interferometer. In either type
of interferometer the phase of the optical beam in one arm can be modulated by varying the
effective dielectric constant and delay of its propagation medium; certain dielectrics are tunable
when biased with large electric fields. In this fashion the output beam power can be modulated
by varying the voltage V across the propagation medium, as illustrated in Figure 12.4.4(d) and
(e). If the device operates near a transmission null, very little change in refractive index is
required to produce a large increase in output power. Such devices can modulate optical power
at frequencies of 10 GHz or more.

The Mach-Zehnder interferometer configuration in Figure 12.4.4(e) is widely used for

modulators because the waveguides can be integrated on a chip together with other optical
components. The output is maximum when the two arms have equal phase delays. When the
two merging beams are out of phase the excited waveguide mode is not trapped in the output
waveguide but radiates away; the radiated wave corresponds to output B in Figure 12.4.4(d).
The same integrated configuration can alternatively be used as a notch filter, eliminating an
undesired optical wavelength for which the two arm lengths differ by exactly λ/2, while passing
nearby wavelengths.

12.4.5 Optical switches

Optical switches redirect optical beams just as electrical switches redirect currents. One
approach is to use MEMS devices that mechanically move mirrors or shutters to redirect the light
beams, which usually are narrow, coherent, and laser-produced. Such devices can switch light
beams at rates approaching 1 MHz.

Another approach is to direct the light beam at right angles to a dielectric within which
ultrasonic acoustic waves (at radio frequencies) are propagating transverse to the light so as to
produce a dynamic phase grating through which the light propagates and diffracts. The
configuration is that of Figure 12.4.4(b). The acoustic waves compress and decompress the
medium in a wavy pattern; the compressed regions have a slightly higher permittivity and

- 396 -
therefore a slightly lower velocity of light. By making the dielectric sufficiently thick, the
cumulative phase variation of the light passing through the device can be λ/2 or more, thus
producing strong diffraction at an angle θ corresponding to the wavelength of light λ and the
acoustic wavelength λa, where θ = sin-1(λ/λa) and λa ≅ 2λ. In practice the cumulative phase
variation is often much less than λ/2 because of improved simplicity, linearity, and the
availability of high input powers that can compensate for the reduced diffractive power
efficiency. In this fashion the diffracted beam can be steered among several output ports at rates
up to ~1 MHz or more, limited largely by the time it takes the acoustic wave to traverse the
diffraction zone. Acoustic velocities in solids are roughly 1000-3000 m/s.

A more important method, however, is the use of Mach-Zehnder interferometers (see

Section 12.4.4) to modulate input optical streams, varying their intensity by more than 15 dB at
rates up to ~10 GHz, limited by the time it takes the signals modulating the electrical phase
length modulator of Figure 12.4.4(e) to propagate across that modulator (e.g. nanoseconds). The
detected modulator output signal is the product of the optical and modulator signals. The
spectrum of this product contains the convolution of the two input spectra, which exhibits upper
and lower sidebands that correspond to the radio frequency signal being communicated.

- 397 -
- 398 -
MIT OpenCourseWare
http://ocw.mit.edu

6.013 Electromagnetics and Applications

Spring 2009

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.