Detection of Optical Radiation - Chapter-11

11
Detection of
Optical Radiation
11.0 INTRODUCTION
The detection of optical radiation is often accomplished by converting the

radiant energy into an electric signal whose intensity is measured by con-
ventional techniques. Some of the physical mechanisms that may be involved
in this conversion include
1. The generation of mobile charge carriers in solid-state photoconductive

detectors
2. Changing through absorption the temperature of thermocouples, thus
causing a change in the junction voltage
3. The release by the photoelectric effect of free electrons from photo-
emissive surfaces
In this chapter we consider in some detail the operation of four of the

most important detectors:
1. The photomultiplier
2. The photoconductive detector
3. The photodiode
4. The avalanche photodiode
The limiting sensitivity of each is discussed and compared to the theoretical

limit. We will find that by use of the heterodyne mode of detection the
theoretical limit of sensitivity may be approached.
399
400 DETECTION OF OPTICAL RADIATION
11.1 OPTICALLY INDUCED TRANSITION RATES
A common feature of all the optical detection schemes discussed in this

chapter is that the electric signal is proportional to the rate at which electrons
are excited by the optical field. This excitation involves a transition of the
electron from some initial bound state, say a, to a final state (or a group of
states) b in which it is free to move and contribute to the current flow. For
example, in an n-type photoconductive detector, state a corresponds to
electrons in the filled valence band or localized donor impurity atoms, while
state b corresponds to electrons in the conduction band. The two levels
involved are shown schematically in Figure 11-1. A photon of energy hv is
absorbed in the process of exciting an electron from a "bound" state a to
a "free" state b in which the electron can contribute to the current flow.
An important point to understand before proceeding with the analysis
of different detection schemes is the manner of relating the transition rate
per electron from state a to b to the intensity of the optical field. This rate
is derived by quantum mechanical considerations.1 In our case it can be
stated in the following form: Given a nearly sinusoidal optical field2
(11.1-1)
where V(t) = E(t) exp(iω0t),3 the transition rate per electron induced by this
field is proportional to V(t)V*(t). Denoting the transition rate as Wa→b, we
have
(11.1-2)
1More specifically, from first order time-dependent perturbation theory; see, for example.
Reference [1].
2By "nearly sinusoidal" we mean a field where E(t) varies slowly compared to exp(iω0t) or,
equivalently, where the Fourier spectrum of E(t) occupies a bandwidth that is small compared
to ω0. Under these conditions the variation of the amplitude E(t) during a few optical periods
can be neglected.
3V(t) is referred to as the "analytic signal" of e(t). See Problem 1.1.
Figure 11-1 Most high-speed optical detectors depend on absorption of photons of

energy hv accompanied by a simultaneous transition of an electron (or hole) from
a quantum state of low mobility (a) to one of higher mobility (b).
PHOTOMULTIPLIER 401
We can easily show that V(t)V*(t) is equal to twice the average value of
e2(t), where the averaging is performed over a few optical periods.
To illustrate the power of this seemingly simple result, consider the
problem of determining the transition rate due to a field
(11.1-3)
taking ω1 - ω0 ≡ ω ≪ ω0. We can rewrite (11.1-3) as
(11.1-4)
and, using (11.1-1), identify V(t) as
thus, using (11.1-2), we obtain
(11.1-5)
This shows that the transition rate has, in addition to a constant term
0 + E21
E2 , a component oscillating at the difference frequency ω with a phase
equal to the difference of the two original phases. This coherent "beating"
effect forms the basis of the heterodyne detection scheme, which is discussed
in detail in Section 11.4.
11.2 PHOTOMULTIPLIER
The photomultiplier, one of the most common optical detectors, is used to

measure radiation in the near ultraviolet, visible, and near infrared regions
of the spectrum. Because of its inherent high current amplification and low
noise, the photomultiplier is one of the most sensitive instruments devised
by man and under optimal operation—which involves long integration time,
cooling of the photocathode, and pulse-height discrimination—has been used
to detect power levels as low as about 10-19 watt [2].
A schematic diagram of a conventional photomultiplier is shown in Fig-
ure 11-2. It consists of a photocathode (C) and a series of electrodes, called
dynodes, that are labeled 1 through 8. The dynodes are kept at progressively
higher potentials with respect to the cathode, with a typical potential dif-
ference between adjacent dynodes of 100 volts. The last electrode (A), the
anode, is used to collect the electrons. The whole assembly is contained
within a vacuum envelope in order to reduce the possibility of electronic
collisions with gas molecules.
The photocathode is the most crucial part of the photomultiplier, since
it converts the incident optical radiation to electronic current and thus de-
termines the wavelength-response characteristics of the detector and, as will
Figure 11-2 Photocathode and focusing dynode configuration of a typical

commercial photomultiplier. C = cathode; 1-8 = secondary-emission dynodes;
A = collecting anode. (After Reference [3].)
be seen, its limiting sensitivity. The photocathode consists of materials with

low surface work functions. Compounds involving Ag-O-Cs and Sb-Cs are
often used; see References [2, 3]. These compounds possess work functions
as low as 1.5 eV, as compared to 4.5 eV in typical metals. As can be seen
in Figure 11-3, this makes it possible to detect photons with longer wave-
lengths. It follows from the figure that the low-frequency detection limit
corresponds to hv = φ. At present the lowest-work-function materials make
possible photoemission at wavelengths as long as 1-1.1 μm.
Spectral response curves of a number of commercial photocathodes are
shown in Figure 11-4. The quantum efficiency (or quantum yield as it is
often called) is defined as the number of electrons released per incident
photon.
The electrons that are emitted from the photocathode are focused elec-
trostatically and accelerated toward the first dynode, arriving with a kinetic
Figure 11-3 Photomultiplier photocathode. The vacuum level corresponds to the

energy of an electron at rest an infinite distance from the cathode. The work
function φ is the minimum energy required to lift an electron from the metal into
the vacuum Ievel, so only photons with hv > φ can be detected.
PHOTOMULTIPLIER 403
Figure 11-4 Photoresponse versus wavelength characteristics and quantum

efficiency of a number of commercial photocathodes. (After Reference [3], p. 228.)
energy of, typically, about 100 eV. Secondary emission from dynode surfaces
causes a multiplication of the initial current. This process repeats itself at
each dynode until the initial current emitted by the photocathode is amplified
by a very large factor. If the average secondary emission multiplication at
each dynode is δ (that is, δ secondary electrons for each incident one) and
the number of dynodes is N, the total current multiplication between the
cathode and anode is
which, for typical values4*of δ = 5 and N = 9, gives G — 2 × 106.
4The value of δ depends on the voltage V between dynodes, and values of δ ≃ 10 can be
obtained (for V ≃ 400 volts). In commercial tubes, values of δ ≃ 5, achievable with V ≃ 100
volts, are commonly used.
11.3 NOISE MECHANISMS IN PHOTOMULTIPLIERS
The random fluctuations observed in the photomultiplier output are due to
1. Cathode shot noise, given according to (10.4-9) by
(11.3-1)
where ic is the average current emitted by the photocathode due to the

signal power that is incident on it. The current id is the so-called dark
current, which is due to random thermal excitation of electrons from
the surface as well as to excitation by cosmic rays and radioactive
bombardment.
2. Dynode shot noise, which is the shot noise due to the random nature
of the secondary emission process at the dynodes. Since current origi-
nating at a dynode does not exercise the full gain of the tube, the con-
tribution of all the dynodes to the total shot noise output is smaller by
a factor of ~δ-1 than that of the cathode; since δ — 5 it amounts to a
small correction and will be ignored in the following.
3. Johnson noise, which is the thermal noise associated with the output
resistance R connected across the anode. Its magnitude is given by
(10.5-9) as
(11.3-2)
Minimum Detectable Power in Photomultipliers—Video Detection
Photomultipliers are used primarily in one of two ways. In the first, the
optical wave to be detected is modulated at some low frequency ωm before
impinging on the photocathode. The signal consists then, of an output current
oscillating at ωm, which, as will be shown below, has an amplitude propor-
tional to the optical intensity. This mode of operation is known as video, or
straight, detection.
In the second mode of operation, the signal to be detected, whose optical
frequency is ωs, is combined at the photocathode with a much stronger optical
wave of frequency ωs + ω. The output signal is then a current at the offset
frequency ω. This scheme, known as heterodyne detection, will be consid-
ered in detail in Section 11-4.
The optical signal in the case of video detection may be taken as
(11.3-3)
where the factor (1 + m cos ωmt) represents amplitude modulation of the

NOISE MECHANISMS IN PHOTOMULTIPLIERS 405
carrier.5 The photocathode current is given, according to (11.1-2), by
(11.3-4)
To determine the proportionality constant involved in (11.3-4), consider the

case of m = 0. The average photocathode current due to the signal is then6
(11.3-5)
where vs = ωs/2π, P is the average optical power, and η (the quantum

efficiency) is the average number of electrons emitted from the photocathode
per incident photon. This number depends on the photon frequency, the
photocathode surface, and in practice (see Figure 11-4) is found to approach
0.3. Using (11.3-5), we rewrite (11.3-4) as
(11.3-6)
The signal output current at ωm is
(11.3-7)
If the output of the detector is limited by filtering to a bandwidth Δv centered

on ωm, it contains a shot-noise current, which, according to (11.3-1), has a
mean-squared amplitude
(11.3-8)
where ic is the average signal current and id is the dark current.
The noise and signal equivalent circuit is shown in Figure 11-5, where
for the sake of definiteness we took the modulation index m = 1. R represents
the output load of the photomultiplier. Te is chosen so that the term 4kTeΔv/R
accounts for the thermal noise of R as well as for the noise generated by
the amplifier that follows the photomultiplier.
The signal-to-noise power ratio at the output is thus
(11.3-9)
5The amplitude modulation can be due to the information carried by the optical wave or, as
an example, to chopping before detection.
6P/hvs is the rate of photon incidence on the photocathode; thus, if it takes 1/η photons to
generate one electron, the average current is given by (11.3-5).
Figure 11-5 Equivalent circuit of a photomultiplier.
Due to the large current gain (G ≃ 106), the first term in the denominator
of (11.3-9), which represents amplified cathode shot noise, is much larger
than the thermal and amplifier noise term 4kTeΔv/R. Neglecting the term
4kTeΔv/R, assuming id ≫ ic, and setting S/N = 1, we can solve for the
minimum detectable optical power.
(11.3-10)
Numerical Example: Sensitivity of Photomultiplier
Consider a typical case of detecting an optical signal under the following

conditions:
vs = 6 × 1014 Hz (λ = 0.5 μm)
η = 10 percent
Δv = 1 Hz
id = 10-15 ampere (a typical value of the dark photocathode current)
Substitution in (11.3-10) gives
Pmin = 3 × 10-16 watt
The corresponding cathode signal current is ic ~ 10-17 ampere, so the as-

sumption id ≫ ic is justified.
Signal-Limited Shot Noise
If one could, somehow, eliminate the Johnson noise and the dark current
altogether, so that the only contribution to the average photocathode current
is ic, which is due to the optical signal, then, using (11.3-5) and (11.3-9) to
solve self-consistently for Pmin,
(11.3-11)
HETERODYNE DETECTION WITH PHOTOMULTIPLIERS 407
This corresponds to the quantum limit of optical detection. Its significance

will be discussed in the next section. The practical achievement of this limit
in video detection is nearly impossible since it depends on near total suppres
sion of the dark current and other extraneous noise sources such as back-
ground radiation reaching the photocathode and causing shot noise.
The quantum detection limit (11.3-11) can, however, be achieved in the
heterodyne mode of optical detection. This is discussed in the next section.
11.4 HETERODYNE DETECTION WITH PHOTOMULTIPLIERS
In the heterodyne mode of optical detection, the signal to be detected

Es cos ωst is combined with a second optical field, referred to as the
local-oscillator field, EL cos(ωs + ω)t, shifted in frequency by ω(ω ≪ ωs).
The total field incident on the photocathode is therefore given by
(11.4-1)
The local-oscillator field originates usually at a laser at the receiving end,

so that it can be made very large compared to the signal to be detected. In
the following we will assume that
(11.4-2)
A schematic diagram of a heterodyne detection scheme is shown in Figure

11-6. The current emitted by the photocathode is given, according to (11.1-2)
and (11.4-1), by
Figure 11-6 Schematic diagram of a heterodyne detector using a photomultiplier.

which, using (11.4-2) can be written as
(11.4-3)
where Ps and PL are the signal and local-oscillator powers, respectively. The
proportionality constant a in (11.4-3) can be determined as in (11.3-6) by
requiring that when Es = 0 the direct current be related to the local-oscillator
power PL by ic = PLηe/hvL,7 so taking v ≈ vL
(11.4-4)
The total cathode shot noise is thus
(11.4-5)
where id is the average dark current while PLeη/hv is the dc cathode current
due to the strong local-oscillator field. The shot-noise current is amplified
by G, resulting in an output noise
(11.4-6)
The mean-square signal current at the output is, according to (11.4-4),
(11.4-7)
The signal-to-noise power ratio at the output is given by
(11.4-8)
where, as in (11.3-9), the last term in the denominator represents the Johnson
(thermal) noise generated in the output load, plus the effective input noise
of the amplifier following the photomultiplier. The big advantage of the
heterodyne detection scheme is now apparent. By increasing PL the S/N
ratio increases until the denominator is dominated by the term G22ePLeη/hv.
This corresponds to the point at which the shot noise produced by the local
oscillator current dwarfs all the other noise contributions. When this state
of affairs prevails, we have, according to (11.4-8),
(11.4-9)
7This is just a statement of the fact that each incident photon has a probability η of releasing
an electron.
HETERODYNE DETECTION WITH PHOTOMULTIPLIERS 409
which corresponds to the quantum-limited detection limit. The minimum

detectable signal—that is, the signal input power leading to an output sig
nal-to-noise ratio of 1—is thus
(11.4-10)
This power corresponds for η = 1 to a flux at a rate of one photon per (Δ v)-1
seconds—that is, one photon per resolution time of the system.8
Numerical Example: Minimum Detectable Power with a Heterodyne System
It is interesting to compare the minimum detectable power for the heterodyne

system as given by (11.4-10) with that calculated in the example of Section
11.3 for the video system. Using the same data,
v = 6 × 1014 Hz(λ = 0.5 μm)
η = 10 percent
Δv = 1 Hz
we obtain
(Ps)min ≃ 4 × 10-18 watt
to be compared with Pmin ≃ 3 × 10-16 watt in the video case.
Limiting Sensitivity as a Result of the Particle Nature of Light
The quantum limit to optical detection sensitivity is given by (11.4-10) as
(11.4-11)
This limit was shown to be due to the shot noise of the photoemitted current.
We may alternatively attribute this noise to the granularity—that is, the
particle nature—of light, according to which the minimum energy increment
of an electromagnetic wave at frequency v is hv. The power average P of
an optical wave can be written as
(11.4-12)
where N is the average number of photons arriving at the photocathode per

second. Next assume a hypothetical noiseless photomultiplier in which ex-
8A detection system that is limited in bandwidth to ∆v cannot resolve events in time that are
separated by less than ~(∆v)-1 second. Thus (Δv)-1 is the resolution time of the system.
actly one electron is produced for each η-1 incident photon. The measure
ment of P is performed by counting the number of electrons produced during
an observation period T and then averaging the result over a large number
of similar observations.
The average number of electrons emitted per observation period T is
(11.4-13)
If the photons arrive in a perfectly random manner, then the number of

photons arriving during the fixed observation period obeys Poissonian
statistics9. Since in our ideal example, the electrons that are emitted mimic
the arriving photons, they obey the same statistical distribution law. This
leads to a fluctuation
Defining the minimum detectable number of quanta as that for which the
rms fluctuation in the number of emitted photoelectrons equals the average
value, we get
or
(11.4-14)
If we convert the last result to power by multiplying it by hv and recall that

T-1 ≃ Δv, where Δ v is the bandwidth of the system, we get
(11.4-15)
in agreement with (11.4-10).

The above discussion points to the fact that the noise (fluctuation) in
the photo current can be blamed on the physical process that introduces the
9This follows from the assumption that the photon arrival is perfectly random, so the probability
of having N photons arriving in a given time interval is given by the Poisson law
The mean-square fluctuation is given by
where
is the average N.
PHOTOCONDUCTIVE DETECTORS 411
randomness. In the case of Poissonian photon arrival statistics (as is the

case with ordinary lasers) and perfect photon emission (η = 1), the fluctua
tions are due to the photons. The opposite, hypothetical, case of no photon
fluctuations but random photoemission (η < 1) corresponds to pure shot
noise. The electrical measurement of noise power will yield the same result
in either case and cannot distinguish between them.
11.5 PHOTOCONDUCTIVE DETECTORS
The operation of photoconductive detectors is illustrated in Figure 11-7. A

semiconductor crystal is connected in series with a resistance R and a supply
voltage V. The optical field to be detected is incident on and absorbed in
the crystal, thereby exciting electrons into the conduction band (or, in p-type
semiconductors, holes into the valence band). Such excitation results in a
lowering of the resistance Rd of the semiconductor crystal and hence in an
increase in the voltage drop across R, which, for ΔRd/Rd ≪ 1, is proportional
to the incident optical intensity.
To be specific, we show the energy levels invoked in one of the more
popular semiconductive detectors—mercury-doped germanium [7]. Mercury
atoms enter germanium as acceptors with an ionization energy of 0.09 eV.
It follows that it takes a photon energy of at least 0.09 eV (that is, a photon
with a wavelength shorter than 14 μm) to lift an electron from the top of the
valence band and have it trapped by the Hg (acceptor) atom. Usually the
germanium crystal contains a smaller density ND of donor atoms, which at
low temperatures find it energetically profitable to lose their valence elec-
trons to one of the far more numerous Hg acceptor atoms, thereby becoming
positively ionized and ionizing (negatively) an equal number of acceptors.
Since the acceptor density NA ≫ ND, most of the acceptor atoms remain
neutrally charged.
An incident photon is absorbed and lifts an electron from the valence
band onto an acceptor atom, as shown in process A in Figure 11-8. The
electronic deficiency (that is, the hole) thus created is acted upon by the
electric field, and its drift along the field direction gives rise to the signal
Figure 11-7 Typical biasing circuit of a photoconductive detector.

LEGEND: electron
hole (an electron vacancy)
neutral donor atom (still in possession of its electron)
neutral acceptor
positively ionized donor atom (stripped of its valence electron)
negatively ionized acceptor atom (which has trapped an electron)
Figure 11-8 Donor and acceptor impurity levels involved in photoconductive

semiconductors.
current. The contribution of a given hole to the current ends when an electron
drops from an ionized acceptor level back into the valence band, thus elim
inating the hole as in B. This process is referred to as electron-hole recom-
bination or trapping of a hole by an ionized acceptor atom.
By choosing impurities with lower ionization energies, even lower-energy
photons can be detected, and, indeed, photoconductive detectors commonly
operate at wavelengths up to λ = 50 μm. Cu, as an example, enters into Ge
as an acceptor with an ionization energy of 0.04 eV, which would correspond
to long-wavelength detection cutoff of λ ≃ 32 μm. The response of a number
of commercial photoconductive detectors is shown in Figure 11-9.
It is clear from this discussion that the main advantage of photocon
ductors compared to photomultipliers is their ability to detect long-wavelength
radiation, since the creation of mobile carriers does not involve overcoming
the large surface potential barrier. On the debit side we find the lack of
current multiplication and the need to cool the semiconductor so that pho-
toexcitation of carriers will not be masked by thermal excitation.
Consider an optical beam, of power P and frequency v, that is incident
on a photoconductive detector. Taking the probability for excitation of a
carrier by an incident photon—the so-called quantum efficiency—as η, the
carrier generation rate is G = Pη/hv. If the carriers last on the average τ0
seconds before recombining, the average number of carriers Nc is found by
Figure 11-9 Relative sensitivity of a number of commercial photoconductors.

(Courtesy Santa Barbara Research Corp.)
equating the generation rate to the recombination rate (Nc/τ0), so
(11.5-1)
Each one of these carriers drifts under the electric field influence10 at a
velocity v giving rise, according to (10.4-1), to a current in the external
circuit of ie = ev/d, where d is the length (between electrodes) of the semi
conductor crystal. The total current is thus the product of ie and the number
of carriers present, or, using (11.5-1),
(11.5-2)
10The drift velocity is equal to μE, where μ is the mobility and E is the electric field.
where τd = d/v is the drift time for a carrier across the length d. The factor
(τ0/τd) is thus the fraction of the crystal length drifted by the average excited
carrier before recombining.
Equation (11.5-2) describes the response of a photoconductive detector
to a constant optical flux. Our main interest, however, is in the heterodyne
mode of photoconductive detection, which, as has been shown in Section
11.4, allows detection sensitivities approaching the quantum limit. In order
to determine the limiting sensitivity of photoconductive detectors, we need
first to understand the noise contribution in these devices.
Generation Recombination Noise in Photoconductive Detectors
The principal noise mechanism in cooled photoconductive detectors reflects

the randomness inherent in current flow. Even if the incident optical flux
were constant in time, the generation of individual carriers by the flux would
constitute a random process. This is exactly the type of randomness involved
in photoemission, and we may expect, likewise, that the resulting noise will
be shot noise. This is almost true except for the fact that in a photoconductive
detector a photoexcited carrier lasts τ seconds11 (its recombination lifetime)
before being captured by an ionized impurity. The contribution of the carrier
to the charge flow in the external circuit is thus e(τ/τd), as is evident from
inspection of (11.5-2). Since the lifetime τ is not a constant, but must be
described statistically, another element of randomness is introduced into the
current flow.
Consider a carrier excited by a photon absorption and lasting τ seconds.
Its contribution to the external current is, according to (10.4-1)
(11.5-3)
which has a Fourier transform
(11.5-4)
so that
(11.5-5)
According to (10.3-10) we need to average |Ie(ω, τ)|2 over τ. This is done in

a manner similar to the procedure used in Section 10.5. Taking the probability*
11The parameter τ0 appearing in (11.5-2) is the value of τ averaged over a large number of
carriers.
function12 g(τ) = τ0-1 exp(-τ/τ0), we average (11.5-5) over all the possible
values of τ according to
(11.5-6)
The spectral density function of the current fluctuations is obtained using

Carson's theorem (10.3-10) as
(11.5-7)
where we used τd = d/v and where N, the average number of carriers

generated per second, can be expressed in terms of the average current I
by use of the relation13
(11.5-8)
leading to
Therefore, the mean-square current representing the noise power in a fre-

quency interval v to v + Δv is
(11.5-9)
which is the basic result for generation-recombination noise.
Numerical Example: Generation Recombination Noise in Hg Doped Germanium

Photoconductive Detector
To better appreciate the kind of numbers involved in the expression for i2N
we may consider a typical mercury-doped germanium detector operating at
20 K with the following characteristics:
d = 10-1 cm
τ0 = 10-9 s
12g(τ) dτ is the probability that a carrier lasts between τ and τ + dτ seconds before recombining.
13This relation follows from the fact that the average charge per carrier flowing through the
external circuit is e(τ0/τd), which, when multiplied by the generation rate N, gives the current.
V (across the length d) = 10 volts ⇒ E = 102 V/cm
μ = 3 × 104 cm2/V-s
The drift velocity is v = μE = 3 × 106 cm/s and τd = d/v = 3.3 × 10-8

second, and therefore τ0/τd = 3 × 10-2. Thus, on the average, a carrier
traverses only 3 percent of the length (d = 1 mm) of the sample before
recombining. Comparing (11.5-9) to the shot-noise result (10.4-9), we find
that for a given average current I the generation recombination noise is
reduced from the shot-noise value by a factor
(11.5-10)
which, in the foregoing example, has a value of about 1/15. Unfortunately,

as will be shown subsequently, the reduced noise is accompanied by a
reduction by a factor of (τ0/τd) in the magnitude of the signal power, which
wipes out the advantage of the lower noise.
Heterodyne Detection in Photoconductors
The situation here is similar to that described by Figure 11-6 in connection

with heterodyne detection using photomultipliers. The signal field
is combined with a strong local-oscillator field
so the total field incident on the photoconductor is
(11.5-11)
The rate at which carriers are generated is taken, following (11.1-2), as

aV(t)V*(t) where a is a constant to be determined. The equation describing
the number of excited carriers Nc is thus
(11.5-12)
where τ0 is the average carrier lifetime, so Nc/τ0 corresponds to the carrier's

decay rate. We assume a solution for Nc(t) that consists of the sum of dc
and a sinusoidal component in the form of
(11.5-13)
where c.c. stands for "complex conjugate."

Substitution in (11.5-12) gives
(11.5-14)
where we took Es and EL as real. The current through the sample is given
by the number of carriers per unit length Nc/d times ev, where v is the drift
velocity
(11.5-15)
which, using (11.5-14), gives
(11.5-16)
where φ = tan-1(ωτ0).
The current is thus seen to contain a signal component that oscillates at
ω and is proportional to Es. The constant a in (11.5-16) can be determined
by requiring that, when Ps = 0, the expression for the direct current predicted
by (11.5-16) agree with (11.5-2). This condition is satisfied if we rewrite
(11.5-16) as
(11.5-17)
where Ps and PL refer, respectively, to the incident-signal and local-oscillator

powers and η, the quantum efficiency, is the number of carriers excited per
incident photon. The signal current is thus
(11.5-18)
while the dc (average) current is
(11.5-19)
Since the average current I appearing in the expression (11.5-9) for the
generation recombination noise is given in this case by
we can, by increasing PL, increase the noise power i2N and at the same time,
according to (11.5-18), the signal i2suntil the generation recombination noise
(11.5-9) is by far the largest contribution to the total output noise. When this
condition is satisfied, the signal-to-noise ratio can be written, using (11.5-9),
(11.5-18), and (11.5-19) and taking PL ≫ Ps, as
(11.5-20)
The minimum detectable signal—that which leads to a signal-to-noise ratio

of unity—is found by setting the left side of (11.5-20) equal to unity and
solving for Ps. It is
(11.5-21)
which, for the same η, is twice that of the photomultiplier heterodyne de-
tection as given by (11.4-10). In practice, however, η in photoconductive
detectors can approach unity, whereas in the best photomultipliers η ≃ 30
percent.
Numerical Example: Minimum Detectable Power of a Heterodyne Receiver Using a

Photoconductor at 10.6 μm
Assume the following:

λ = 10.6 μm
Δv = 1 Hz
η ≃ 1
Substitution in (11.5-21) gives a minimum detectable power of
(Ps)min ≃ 10-19 watt
Experiments ([8, 9]) have demonstrated that the theoretical signal-to-
noise ratio as given by (11.5-20) can be realized quite closely in practice;
see Figure 11-10.
Figure 11-10 Signal-to-noise ratio of heterodyne signal to Ge:Cu detector at a

heterodyne frequency of 70 MHz. Data points represent observed values. (After
Reference [8].)
THE p-n JUNCTION 419
11.6 THE p-n JUNCTION
Before embarking on a description of the p-n diode detector, we need to

understand the operation of the semiconductor p-n junction. Consider the
junction illustrated in Figure 11-11. It consists of an abrupt transition from
a donor-doped (that is, n-type) region of a semiconductor, where the charge
carriers are predominantly electrons, to an acceptor-doped (p-type) region,
where the carriers are holes. The doping profile—that is, the density of
excess donor (in the n region) atoms or acceptor atoms (in the p region)—
is shown in Figure 11-11(a). This abrupt transition results usually from dif
fusing suitable impurity atoms into a substrate of a semiconductor with the
opposite type of conductivity. In our slightly idealized abrupt junction we
assume that the n region (x > 0) has a constant (net) donor density ND and
the p region (x < 0) has a constant acceptor density NA.
The energy-band diagram at zero applied bias is shown in Figure 11-11(b).
The top (or bottom) curve can be taken to represent the potential energy of
an electron as a function of position x, so the minimum energy needed to
take an electron from the n to the p side of the junction is eVd. Taking the
separations of the Fermi level from the respective band edges as φn and φp
as shown, we have
Vd is referred to as the "built-in" junction potential.

Figure 11-11(c) shows the potential distribution in the junction with an
applied reverse bias of magnitude Va. This leads to a separation of eVa
between the Fermi levels in the p and n regions and causes the potential
barrier across the junction to increase from eVd to e(Vd + Va). The change
of potential between the p and n regions is due to a sweeping of the mobile
charge carriers from the region -lp < x < ln, giving rise to a charge double
layer of stationary (ionized) impurity atoms, as shown in Figure 11-11(d).
In the analytical treatment of the problem we assume that in the depletion
layer (-lp < x < ln) the excess impurity atoms are fully ionized and thus,
using ∇⋅E = ρ/ε and E = -∇V, where V is the potential, we have
(11.6-1)
(11.6-2)
where the charge of the electron is -e and the dielectric constant is ε. The
boundary conditions are
(11.6-3)
Figure 11-11 The abrupt p-n junction. (a) Impurity profile. (b) Energy-band diagram
with zero applied bias. (c) Energy-band diagram with reverse applied bias. (d) Net
charge density in the depletion layer. (e) The electric field. The circles in (b) and
(c) represent ionized impurity atoms in the depletion layer.
THE p-n JUNCTION 421
(11.6-4)
(11.6-5)
The solutions of (11.6-1) and (11.6-2) conforming with the arbitrary choice
of V(0) = 0 are
(11.6-6)
(11.6-7)
which, using (11.6-4), gives
(11.6-8)
so the double layer contains an equal amount of positive and negative charge.
Condition (11.6-5) gives
(11.6-9)
which, together with (11.6-8) leads to
(11.6-10)
(11.6-11)
and, therefore, as before,
(11.6-12)
Differentiation of (11.6-6) and (11.6-7) yields
(11.6-13)
The field distribution of (11.6-13) is shown in Figure 11-11(e). The maximum

field occurs at x = 0 and is given by
(11.6-14)
Figure 11-12 Equivalent circuit of a p-n junction. In typical back-biased diodes,

Rd ≫ Rs and RL and RL ≫ Rs, so the resistance across the junction can be taken
as equal to the load resistance RL.
The presence of a charge Q = -eNAlp per unit junction area on the p side
and an equal and negative charge on the n side leads to a junction capacitance.
The reason is that lp and ln depend, according to (11.6-10) and (11.6-11), on
the applied voltage Va, so a change in voltage leads to a change in the charge
eNAlp = eNDln and hence to a differential capacitance per unit area,14 given
by
(11.6-15)
which, using (11.6-10) and (11.6-11), can be shown to be equal to
(11.6-16)
as appropriate to a parallel-plate capacitance of separation l = lp + ln. The

equivalent circuit of a p-n junction is shown in Figure 11-12. The capacitance
Cd was discussed above. The diode shunt resistance Rd in back-biased junc-
tions is usually very large (>106 ohms) compared to the load impedance RL
and can be neglected. The resistance Rs represents ohmic losses in the bulk
p and n regions adjacent to the junction.
11.7 SEMICONDUCTOR PHOTODIODES
Semiconductor p-n junctions are used widely for optical detection: see Ref-
erences [10-12]. In this role they are referred to as junction photodiodes.
The main physical mechanisms involved in junction photodetection are il
14The capacitance is defined by C = Q/Va, whereas the differential capacitance Cd = dQ/dVa

is the capacitance "seen" by a small ac voltage when the applied bias is Va.
SEMICONDUCTOR PHOTODIODES 423
lustrated in Figure 11-13. At A, an incoming photon is absorbed in the p

side creating a hole and a free electron. If this takes place within a diffusion
length (the distance in which an excess minority concentration is reduced
to e-1 of its peak value, or in physical terms, the average distance a minority
carrier traverses before recombining with a carrier of the opposite type) of
the depletion layer, the electron will, with high probability, reach the layer
boundary and will drift under the field influence across it. An electron tra-
versing the junction contributes a charge e to the current flow in the external
circuit, as described in Section 10.4. If the photon is absorbed near the n
side of the depletion layer, as shown at C, the resulting hole will diffuse to
the junction and then drift across it again, giving rise to a flow of charge e
in the external load. The photon may also be absorbed in the depletion layer
as at B, in which case both the hole and electron that are created drift (in
opposite directions) under the field until they reach the p and n sides, re-
spectively. Since in this case each carrier traverses a distance that is less
than the full junction width, the contribution of this process to charge flow
in the external circuit is, according to (10.4-1) and (10.4-7), e. In practice
this last process is the most desirable, since each absorption gives rise to a
charge e, and delayed current response caused by finite diffusion time is
avoided. As a result, photodiodes often use a p-i-n structure in which an
intrinsic high resistivity (i) layer is sandwiched between the p and n regions.
The potential drop occurs mostly across this layer, which can be made long
enough to ensure that most of the incident photons are absorbed within it.
Typical construction of a p-i-n photodiode is shown in Figure 11-14.
Figure 11-13 The three types of electron-hole pair creation by absorbed photons
that contribute to current flow in a p-n photodiode.
Figure 11-14 A p-i-n photodiode. (After Reference [13].)
It is clear from Figure 11-13 that a photodiode is capable of detecting

only radiation with photon energy hv > Eg, where Eg is the energy gap of
the semiconductor. If, on the other hand, hv ≫ Eg, the absorption, which
in a semiconductor increases strongly with frequency, will take place entirely
near the input face (in the n region of Figure 11-14) and the minority carriers
generated by absorbed photons will recombine with majority carriers before
diffusing to the depletion layer. This event does not contribute to the current
l
f ow and, as far as the signal is concerned, is wasted. This is why the
photoresponse of diodes drops off when hv>Eg. Typical frequency response
curves of photodiodes are shown in Figure 11-15. The number of carriers
flowing in the external circuit per incident photon, the so-called quantum
efficiency, is seen to approach 50 percent in Ge.
Frequency Response of Photodiodes
One of the major considerations in optical detectors is their frequency re

sponse—that is, the ability to respond to variations in the incident intensity
such as those caused by high-frequency modulation. The three main mech
anisms limiting the frequency response in photodiodes are:
1. The finite diffusion time of carriers produced in the p and n regions.

This factor was described in the last section, and its effect can be min
imized by a proper choice of the length of the depletion layer.
2. The shunting effect of the signal current by the junction capacitance Cd
shown in Figure 11-12. This places an upper limit of
(11.7-1)
on the intensity modulation frequency where Re is the equivalent re-

sistance in parallel with the capacitance Cd.
3. The finite transit time of the carriers drifting across the depletion layer.
Figure 11-15 Quantum efficiencies for silicon and germanium photodiodes

compared with the efficiency of the S-1 photodiode used in a photomultiplier tube.
Emission wavelengths for various lasers are also indicated. (After Reference [13].)
To analyze first the limitation due to transit time, we assume the slightly
idealized case in which the carriers are generated in a single plane, say point
A in Figure 11-13, and then drift the full width of the depletion layer at a
constant velocity v. For high enough electric fields, the drift velocity of
carriers in semiconductors tends to saturate, so the constant velocity as
sumption is not very far from reality even for a nonuniform field distribution,
such as that shown in Figure 11-11(e), provided the field exceeds its satu-
ration value over most of the depletion layer length. The saturation of the
whole velocity in germanium, as an example, is illustrated by the data of
Figure 11-16.
The incident optical field is taken as
(11.7-2)
where
(11.7-3)
Thus, the amplitude is modulated at a frequency ωm/2π. Following the dis-

cussion of Section 11.1 we take the generation rate G(t); that is, the number
of carriers generated per second, as proportional to the average of e2(t) over
a time long compared to the optical period 2π/ω. This average is equal to
Figure 11-16 Experimental data showing the saturation of the drift velocity of holes
in germanium at high electric fields. (After Reference [14].)
½V(t)V*(t), so the generation rate is taken as
(11.7-4)
where a is a proportionality constant to be determined. Dropping the term

involving cos 2ωmt and using complex notation, we rewrite G(t) as
(11.7-5)
A single carrier drifting at a velocity v contributes, according to (10.4-1), an

instantaneous current
(11.7-6)
to the external circuit, where d is the width of the depletion layer. The
current due to carriers generated between t' and t' + dt' is (ev/d)G(t') dt'
but, since each carrier spends a time τd = d/v in transit, the instantaneous
current at time t is the Sum of contributions of carriers generated between
t and t - τd
and, after integration,
(11.7-7)
The factor (1 - e-iωmτd)/iωmτd represents the phase lag as well as the reduction
in signal current due to the finite drift time τd. If the drift time is short
compared to the modulation period, so ωmτd ≪ 1, it has its maximum value
of unity, and the signal is maximum. This factor is plotted in Figure 11-17
as a function of the transit phase angle ωmτd. We can determine the value
of the constant a in (11.7-7) by requiring that (11.7-7) agree with the ex-
perimental observation according to which in the absence of modulation,
m = 0, each incident photon will create η carriers. Thus the dc (average)
current is
(11.7-8)
where P is the optical (signal) power when m = 0. Using (11.7-8), we can

rewrite (11.7-7) as
(11.7-9)
To evaluate the effect of the other limiting factors on the modulation fre-
quency response of a photodiode, we refer to the diode equivalent ac circuit
in Figure 11-18. Here Rd is the diode incremental (ac) resistance, Cd the
junction capacitance, Rs represents the contact and series resistance, Lp the
Figure 11-17 Phase and magnitude of the transit-time reduction factor

(1 — e-iωmτd)/iωmτd.
Figure 11-18 The equivalent high-frequency circuit of a semiconductor photodiode.
parasitic inductance associated mostly with the contact leads, and Cp, the
parasitic capacitance due to the contact leads and the contact pads.
Recent advances [20-22, 24] have resulted in metal-GaAs (Schottky)
diodes with frequency response extending up to 10" Hz. Figure 11-19 shows
a schematic diagram of such a diode. This high-frequency limit was achieved
by using a very small area (5 μm × 5 μm) that minimizes Cd, by using
extremely short contact leads to reduce Rs and Lp, by fabricating the diode
on semi-insulating GaAs substrate [20] to reduce Cp, and by using a thin
(0.3 μm) n- GaAs drift region to reduce the transit time. The resulting mea
sured frequency response is shown in Figure 11-20. The measurement of the
frequency response up to 100 GHz is by itself a considerable achievement.
This was accomplished by first obtaining the impulse response of the pho-
todiode by exciting it with picosecond pulses (which, for the range of fre-
Figure 11-19 (a) Planar GaAs Schottky photodiode. (b) Cross section along A-A.
The n- GaAs layer (0.3 μm thick) and the n+ GaAs (0.4 μm thick) are grown by
liquid-phase epitaxy on semi-insulating GaAs substrate. The semitransparent
Schottky consists of 100 Å of Pt (After Reference [22].)
Figure 11-20 The modulation frequency response of the Schottky photodiode

shown in Figure 11-19. (After Reference [22].)
quencies of interest, may be considered as delta functions) from a mode-locked

laser [21]. The diode response, which is only a few picoseconds long, is
measured by a new electrooptic sampling technique [23, 24]. The frequency
response, as plotted in Figure 11-20, is obtained by taking the Fourier trans-
form of the measured impulse response.
Numerical Example: Modulation Response of a GaAs p-n Junction Photodiode
Let us calculate the upper limit on the frequency response of the diode
shown in Figure 11-19. The following data apply.
Area = 5 μm × 5 μm
ε = 12.25 ε0
d = 0.3 μm (= thickness of drift region)
v = 107 cm/s (saturation velocity of electron in GaAs)
Rs ≈ 10 ohms
The transit-time limit fm is obtained from the condition 2πfmτd = 2. This,

according to Figure 11-17, is the frequency where the response is down to
84 percent of its maximum (zero-frequency) value. The result is
The junction capacitance, based on the above data, is ~10-14 farad. The
parasitic capacitance can be kept in this case to ~10-13 farad. Since the
resistance Rd of the reverse-biased junction is very large, it is usually ne-
glected.
The circuit limit to the frequency response is fm ~ 1/(2πRsCp) =
1.59 × 1011 Hz. Since this value is larger than the transit-time limit, we
conclude that the frequency response is transit-time limited to a value ~1011
Hz, which is in agreement with the value obtained from Figure 11-20.
Detection Sensitivity of Photodiodes
We assume that the modulation frequency of the light to be detected is low

enough that the transit time factor is unity and that the condition
(11.7-10)
is fulfilled and, therefore, according to (11.7-1), the shunting of signal current

by the diode capacitance Cd can be neglected. The diode current is given
by (11.7-9) as
(11.7-11)
The noise equivalent circuit of a diode connected to a load resistance RL is

shown in Figure 11-21. The signal power is proportional to the mean-square
value of the sinusoidal current component, which, for m = 1, is
(11.7-12)
Two noise sources are shown. The first is the shot noise associated with the
random generation of carriers. Using (10.4-9), this is represented by a noise
generator i2N1 = 2eI Δv, where I is the average current as given by the first
Figure 11-21 Noise equivalent circuit of a photodiode operating in the direct (video)
mode. The modulation index m is taken as unity, and it is assumed that the
modulation frequency is low enough that the junction capacitance and transit-time
effects can be neglected. The resistance RL is assumed to be much smaller than
the shunt resistance Rd of the diode, so the latter is neglected. Also neglected is
the series diode resistance, which is assumed small compared with RL.
term on the right side of (11.7-11). Taking m = 1, we obtain
(11.7-13)
where PB is the background optical power entering the detector (in addition
to the signal power) and id is the "dark" direct current that exists even when
Ps = Pb = 0. The second noise contribution is the thermal (Johnson noise)
generated by the output load, which, using (10.5-9), is given by
(11.7-14)
where Te is chosen to include the equivalent input noise power of the amplifier
following the diode.15 The signal-to-noise power ratio at the amplifier output
is thus
(11.7-15)
In most practical systems the need to satisfy Equation (11.7-10) forces one
to use small values of load resistance RL. Under these conditions and for
values of P that are near the detectability limit (S/N = 1), the noise term
(11.7-14) is much larger than the shot noise (11.7-13) and the detector is
consequently not operating near its quantum limit. Under these conditions
we have
(11.7-16)
15In practice it is imperative that the signal-to-noise ratio take account of the noise power
contributed by the amplifier. This is done by characterizing the "noisiness" of the amplifier
by an effective input noise "temperature" TA. The amplifier noise power measured at its output
is taken as GkTA∆v, where G is the power gain. (A hypothetical noiseless amplifier will thus
be characterized by TA = 0.) This power can be referred to the input by dividing by G, thus
becoming kTA∆v. The total effective noise power at the amplifier input is the sum of this power
and the Johnson noise kTΔv due to the diode load resistance; that is, k(T + TA)Δv ≡ kTeΔv.
The amplifier noise temperature TA is related to its "noise figure" F by the definition
It follows that the noise power generated within the amplifier and measured at its output is
where T0 = 290. The ratio of the signal-to-noise power ratio at the input of the amplifier to the
same ratio at the output is thus
This ratio becomes equal to the "noise figure" F when the temperature T of the detector output
load is equal to T0. (Note that the choice T0 = 290 is a matter of, universal, convention.)
The "minimum detectable optical power" is by definition that yielding

S/N = 1 and is, from (11.7-16),
(11.7-17)
which is to be compared to the theoretical limit of hv∆v/η, which, according

to (11.3-11), obtains when the signal shot-noise term predominates. In prac-
tice, the value of RL is related to the desired modulation bandwidth Δv and
the junction capacitance Cd by
(11.7-18)
which, when used in (11.7-16), gives
(11.7-19)
This shows that sensitive detection requires the use of small area junctions
so that Cd will be at a minimum.
Numerical Example: Minimum Detectable Power in the Case of Amplifier Limited

Detection
Assume a typical Ge photodiode operating at λ = 1.4 μm with Cd = 1 pF,

Δv = 1 GHz, and η = 50 percent. Let the amplifier following the diode have
an effective noise temperature Te = 1200 + 290 = 1490 K (see footnote 15)
[14-15]. Substitution in (11.7-19) gives
Pmin ≃ 3.34 × 10-7 watt
for the minimum detectable signal power.
11.8 THE AVALANCHE PHOTODIODE
By increasing the reverse bias across a p-n junction, the field in the depletion
layer can increase to a point at which carriers (electrons or holes) that are
accelerated across the depletion layer can gain enough kinetic energy to
"kick" new electrons from the valence to the conduction band, while still
traversing the layer. This process, illustrated in Figure 11-22, is referred to
as avalanche multiplication. An absorbed photon (A) creates an elec
tron-hole pair. The electron is accelerated until at point C it has gained
sufficient energy to excite an electron from the valence to the conduction
THE AVALANCHE PHOTODIODE 433
Figure 11-22 Energy-position diagram showing the carrier multiplication following

a photon absorption in a reverse-biased avalanche photodiode.
band, thus creating a new electron-hole pair. The newly generated carriers
drift in turn in opposite directions. The hole (F) can also cause carrier
multiplication as in G. The result is a dramatic increase (avalanche) in junc-
tion current that sets in when the electric field becomes high enough. This
effect, discovered first in gaseous plasmas and more recently in p-n junctions
(References [15, 16]), gives rise to a multiplication of the current over its
value in an ordinary (nonavalanching) photodiode. An experimental plot of
the current gain M as a function of the junction field is shown in Figure
11-23.16
Avalanche photodiodes are similar in their construction to ordinary pho-
todiodes except that, because of the steep dependence of M on the applied
field in the avalanche region, special care must be exercised to obtain very
uniform junctions. A sketch of an avalanche photodiode is shown in Figure
11-24.
Since an avalanche photodiode is basically similar to a photodiode, its
equivalent circuit elements are given by expressions similar to those given
above for the photodiode. Its frequency response is similarly limited by
diffusion, drift across the depletion layer, and capacitive loading, as dis-
cussed in Section 11.7.
A multiplication by a factor M of the photocurrent leads to an increase
by M2 of the signal power S over that which is available from a photodiode
16If the probability that a photo-excited electron-hole pair will create another pair during its
drift is denoted by p, the current multiplication is
Figure 11-23 Current multiplication factor in an avalanche diode as a function of

the electric field. (After Reference [16].)
Figure 11-24 Planar avalanche photodiode. (After Reference [13].)

THE AVALANCHE PHOTODIODE 435
so that, using (11.7-12), we get
(11.8-1)
where P is the optical power incident on the diode. This result is reminiscent
of the signal power from a photomultiplier as given by the numerator of
(11.3-9), where the avalanche gain M plays the role of the secondary electron
multiplication gain G. We may expect that, similarly, the shot-noise power
will also increase by M2. The shot noise, however, is observed to increase
as Mn, where 2 < n < 3.17 Experimental observation of a near ideal M2.1
behavior is shown in Figure 11-25.
The signal-to-noise power ratio at the output of the diode is thus given,
following (11.7-15), by
(11.8-2)
The advantage of using an avalanche photodiode over an ordinary photo-

diode is now apparent. When M = 1, the situation is identical to that at the
photodiode as described by (11.7-15). Under these conditions the thermal
term 4kTe∆v/RL in the denominator of (11.8-2) is typically much larger than
17A theoretical study by McIntyre [17] predicts that if the multiplication is due to either holes
or electrons, n = 2, whereas if both carriers are equally effective in producing electron-hole
pairs, n = 3.
Figure 11-25 Noise power (measured at 30 MHz) as a function of photocurrent

multiplication for an avalanche Schottky-barrier photodiode. (After Reference
[18].)
the shot-noise terms. This causes S/N to increase with Μ. This improvement
continues until the shot-noise terms become comparable with 4kTe∆v/RL.
Further increases in M result in a reduction of S/N since n > 2, and the
denominator of (11.8-2) grows faster than the numerator. If we assume that
M is adjusted optimally so that the denominator of (11.8-2) is equal to twice
the thermal term 4kTe∆v/RL, we can solve for the minimum detectable power
(that is, the power input for which S/N = 1) obtaining
(11.8-3)
where M' is the optimum value of M as discussed previously. The improve

ment in sensitivity over the photodiode result (11.7-16) is thus approximately
M'. Values of M' between 30 and 100 are commonly employed, so the use
of avalanche photodiodes affords considerable improvement in sensitivity
over that available from photodiodes.
11.9 POWER FLUCTUATION NOISE IN LASERS
The power output from lasers is ever fluctuating. This fluctuation may be
due to temperature variations, acoustic vibrations, and other man-made
causes. Even if all of these extraneous effects are eliminated, there remains
a basic (quantum mechanical) contribution that is due to spontaneous emis-
sion of radiation into the laser mode by atoms dropping from the upper
transition level into the lower levels. The field due to this spontaneous
emission is not coherent with that of the laser mode, thus causing phase and
amplitude fluctuations [1]. Since these fluctuations are random, they are
described and quantified in terms of the statistical noise tools developed
earlier in this chapter and in Chapter 10.
Let the power output of the laser be
(11.9-1)
where the time-averaging value of the fluctuation is zero.
so that P0 is the average optical power. Using (10.2-6) and (10.2-8) we char
acterize the "power" of the fluctuation via the mean of the squared deviation18
(11.9-2)
S∆P(f) is related to the spectral density function S∆P(ω), defined by (10.2-8)

and (10.2-14), by
(11.9-3)
18In this section, we will usef to denote "low" (rf) frequencies and vfor optical frequencies.
POWER FLUCTUATION NOISE IN LASERS 437
If an optical field at frequency v with a power P(t) is incident on a detector

whose quantum efficiency (electrons per photon) is η, the output current is
so that according to Equation (11.9-1) the optical power fluctuation ΔP(t)

causes a fluctuating current component Δi(t) = eη∆P(t)/hv with a mean
square
(11.9-4)
where Δf is the bandwidth of the electronic detection circuit.

The relative intensity noise (RIN), is defined as the relative fluctuation
"power" in a Δf = 1 Hz bandwidth
(11.9-5)
A single-mode semiconductor laser might possess a value of RIN ≈ 10-16

(or -160 db). Assuming that the detector circuit has a bandwidth of, say,
Δf = 109 Hz, the relative mean-squared fluctuation in the detected current
is
The RMS value of the power fluctuation is thus
The mean-squared noise current in the output of the detector due to these
fluctuations is given by (11.9-4)
(11.9-6)
Assuming as an example that λ = 1.3 μm, P0 = 3 mW, RIN = 10-16 Hz-1,

Δf = 109 Hz, and η = 0.6, we obtain
Example: Optical Fiber Link Design
Our task here is to determine the maximum allowed repeater spacing for an
optical fiber communication link. We will assume that the optical source is
a 1.3 μm GaInAsP laser (v = c/λ = 2.31 × 1014 Hz) and that the fiber
possesses an attentuation of 0.3 db/km (corresponding to an attenuation

constant α = 0.3/4.343 = 0.0691 (km)-1). The optical power launched into
the fiber is P0 = 3 mW. The channel is to transmit 109 bits/s so that the
bandwidth of the detector circuit is taken as Δf = 1/period = 109 Hz. The
system considerations dictate that the bit error probability at the detector
output not exceed 10-10. The detector output impedance is Rl = 1,000 Ω,
and the amplifier (following the detector) noise figure is 6 db, i.e., F = 4
(see footnote 15).
From Figure 10-20 we determine that the signal-to-noise power ratio at
the amplifier output must exceed 22 db to assure a bit error probability upon
detection that is smaller than 10-
10 . Our task is thus to calculate the signal
power i2s and the total noise power i2N at the output of the detector as a
function of the length L of the link.
The signaI power is obtained from (11.7-11), assuming a modulation index
m = 0.5, as
(11.9-7)
The total noise power at the output of the amplifier referred to its input is
(11.9-8)
The first noise term is that due to power fluctuation (11.9-6); the second is
the shot noise associated with the average current at the output of the
detector Id0 = ηP0e exp(-αL)/(hv). The third term represents, as in (11.7-15),
both the Johnson noise of the output resistor Rl as well as the amplifier
output noise power (referred to its input, see footnote 15). If the temperature
of the output resistor RL is T = 290 K, TE = T + (F - 1)290 = 1160 K.
Figure 11-26(a) shows the main elements of an optical fiber link. Figure
, i2NL, i2
11-26(b) shows a plot of i2s NS, and i2NA as well as the total noise power
Figure 11-26 (a) An optical fiber communication link consisting of a laser, an

optical coupling system c, a fiber L/(km long), a detector D, an output resistance
RL and an amplifier A with a current gain G and a noise figure F. (b) The signal
) laser fluctuation (i2NL) detector shot noise (i2NS), combined Johnson-amplifier
(i2s
noise (i2NA), and the total noise
currents as a function of the link length L. The currents are referred to the
amplifier input plane S, i.e., they correspond to output currents divided by the
current gain G of the output amplifier.
439
as a function of the link length L. The important thing to note is the relative
change of the various powers with distance. The distance L0, where the
detected signal-to-noise power ratio is down to 22 db, is read off as L0 = 87
km.19 This distance is thus chosen as the link length. Notice, as an example,
that the dominant noise contribution at L > 33 km is the amplifier-detector
noise i2NA. If the latter were reduced by, say, 3 db, the link length could be
increased by 5 km, as indicated by the dashed line.
The signal-to-noise power ratio of a p-n diode detector is given by (11.7-16)
in the case where the dominant contributions to the noise power are the
amplifier noise and the Johnson (thermal) noise of the load resistance RL in
the diode output circuit. The mean-square noise current is then
(11.9-9)
The signal peak current is given by (11.7-8) as
(11.9-10)
where Ps is the peak pulsed optical power incident on the detector. The
signal-to-noise current ratio at the amplifier output (see footnote 15) is thus
(11.9-11)
Our next prob1em_is that of finding the minimum value of the signal
power Ps so that is/(i2N)1/2 in (11.9-11) exceeds the needed value of 12.59.
We thus need to know Te, RL, and Δf. Te is obtained from the given value
of the amplifier noise figure (F = 6 dB). Taking T = 290 K, we obtain, using
footnote 15, Te = 290 + (4 - 1)290 = 1160. In orderto achieve this bandwidth,
the load resistance Rl must not exceed (see 11.7-18) the value
(11.9-12)
where C is the total output capacitance given as 3 × 10-12 f. Using the

above value of Δf and C, we obtain
We return now to (11.9-11), which, using η = 0.5, λ = 1.35 μm, iS/(i2N)1/2 =

12.59, yields
for the minimum power input to the photodiode.

The total transmission loss in the 50 km fiber is 20 dB. We will assume
19That is, 10 log (i2S/i2N) = 22.

INFRARED IMAGING AND BACKGROUND-LIMITED DETECTION 441
that an additional 4 dB loss is caused by coupling the laser output to the

fiber and at the fiber output so that the total loss is 24 dB (that is, 251). The
laser power output must thus exceed
Plaser = 6 × 10-3 watt
which is a reasonable power level for CW diode lasers.

If the fiber had been substantially lossier than in the above example, we
could still have met our design specifications by using an avalanche pho-
todiode.
11.10 INFRARED IMAGING AND BACKGROUND-LIMITED DETECTION [25—28]
Arrays of cooled infrared detectors based mostly on photoconductive semi-

conductors such as mercury cadmium telluride (HgCdTe) have become in
creasingly important elements in the fast developing technology of infrared
imaging and detection. The application areas served by this new technology
include tumor detection, the mapping of earth resources by orbiting satellites,
"spy" satellites, and nighttime "seeing." We will not concern ourselves
here with the system aspects of these applications but rather with the basic
noise physics of a single element that is prerequisite to system considera-
tions. The concepts involved here are the same as those we have encountered
in the early sections of this chapter, but the operational considerations merit
a dedicated treatment. To be specific, we will focus our discussion to doped,
say, n-type, photoconductors, such as HgCdTe, in which the optical input
field causes excitation of electrons to the conduction band so that the signal
current is due to the drift of the excited carriers.
Consider the photoconductive detector shown in Figure 11-27. The de-
tected radiation is incident on the "face" whose area is A. The thickness
of the detector is t. The optical signal power input to the detector is Ps. This
power gives rise to a signal current given according to (11.5-2) by
(11.10-1)
where τ0 is the lifetime of the photo-excited carriers, is the transit
time for a carrier with a velocity v. vopt is the frequency of the optical beam.
η is the fraction of the incoming photons that are usefully absorbed in the
photoconductor. τ0 is the average lifetime of a photo-excited carrier. In
addition to the signal current, we have two other major sources of noise
currents that are not related to the signal.
The first is the shot noise associated with the drift under the influence
of the applied external field of thermally excited (minority) carriers, while
the second is the shot noise due to carriers excited by the ever present
incoming background optical radiation. Referring to Figure 11-27(b), we
Figure 11-27 (a) A schematic diagram of an infrared detecting element intercepting

radiation from an acceptance angle θ. (b) A detailed view of the photoconductor.
distinguish in our photoconductor three populations of carriers: ns (cm-3)

due to the signal, nt due to thermal excitation, and nB due to incoming
background radiation. The density nB of carriers excited by the background
radiation is
(11.10-2)
This last expression can be obtained by setting the rate of photon absorption
per unit volume equal to the rate of minority carrier recombi-
nation. The incident background power PB is most often that of the back-
ground blackbody radiation, in which case
(11.10-3)
where A is the cross-sectional area of the detector, Δv is the optical band-

width of the radiation allowed into the detector, -TB is the background
temperature, white θis the angle [see Figure 11-27(a)] within which radiation
is accepted by the detector. In practice the acceptance angle θ and the
background temperature T are dictated by the application. A reasonable
strategy in such a case is to cool the detector to the point where nt < nB.
This renders the contribution of n, to the shot noise equal to that of nB so
that additional cooling will not materially improve the signal-to-noise ratio
at the output. Since the signal-to-noise ratio under this condition is deter-
mined by the background radiation, it is referred to as background limited
infrared performance (BLIP).
Let us assume that a BLIP condition has been achieved and calculated
the resulting detector performance. From (11.5-9) the mean-squared output
noise current is (in the limit vτ0 ≪ 1)
(11.10-4)
Where IB = nBevA is the average current due to the background radia-

tion-excited carriers and Δf is the bandwidth of the (electronic) detection
circuit. Substituting for nB from (11.10-2) gives
(11.10-5)
The minimum detectable signal, also known as the noise equivalent power
(NEP) of the detector, is that value of the signal power Ps for which
Using (11.10-1) and (11.10-5) we can solve for the minimum detectable power
when the main noise contribution is due to the background radiation, i.e.,
nt < nB
(11.10-6)
A common figure of merit used in the infrared imaging community to describe

detector sensitivity is the specific peak detectivity D* ("Dee" star) defined
as
When the detector is coded sufficiently so that it is background limited, D*

becomes
(11.10-6a)
where, to remind us, the B subscript stands for the background limited
condition and Δf is bandwidth of the detection circuit including the pho-
toconductive element.
In a detector limited by thermal excitation of carriers, i.e., one where
nt > nB we have
(11.10-7)
(11.10-8)
Equating i2Nt to i2sas in (11.10-6), we obtain
and
(11.10-9)
where, in the last expression, we used τd ≡ t/v.

It is obvious that the condition nt ≤ nB is equivalent to
(11.10-10)
A key issue in infrared detection is to determine to what temperature a

detector element need be cooled to be background limited. To answer this
question we need to know nB and the dependence of nt on the material
parameters and the temperature.
As an example consider an infrared detector as shown in Figure 11-28(a)
in which the photoconductive medium is a GaAs/GaAlAs superlattice [29].
It is based on excitation of electrons from a confined "quantum-well" state
(see Section 16.1) to continuum (unconfined) states where they are free to
conduct. These wells consist of thin layers (~100 Å) of crystalline GaAs
layers sandwiched between higher energy gap Ga1-xAlxAs crystalline layers.
In this case [25],
(11.10-11)
where L is the width of the unit cell in Figure 11-28(a), m* is the carrier
(electron) effective mass, EF the Fermi energy of the material, and V the
depth of the quantum well. Using the data of Figure 11-28(b) and m* =
Figure 11-28 Schematic illustration of a typical quantum well detector structure

(only two wells are shown) with the relavant energies. In the upper part of the
figure (a), the quantum well detector under bias is shown. The second part of the
figure (b) shows the relevant distances and energies. The sub-band levels are given
by E1 and E2.
, we can use (11.10-9) to plot D*t vs.
the detector temperature T. The result is shown in Figure 11-29(a).

To use this curve we need first to obtain a value for D*B using (11.10-3)
and (11.10-6a). We then find the temperature in Figure 11-29(a) where
D*t = D*B. As an example, given D*B = 1012 we find from the figure that the
quantum-well detector needs to be cooled to T < 48 K in order to become
background limited.20 Figure 11-29 curve (b) is a plot based on Equations
(11.10-3, 11.10-6, 11.10-9) of the acceptance angle θB for which D*B is equal
to D*t. Further increases of θB will thus cause the detector to be background
limited. In the figure we used a background temperature TB = 300 K,
20In practice cooling the detector below the liquid N2 temperature, 77.7oK, is expensive and is
reserved to very demanding applications.
Figure 11-29 Curve (a) is a plot of the thermal-excitation-limited detectivity D*tat

λ = 10 μm vs. temperature of the GaAs quantum well detector. Curve (b) is a plot
of the acceptance angle θB which results in D*B = D*t (BLIP) for the background
conditions stated in the figure.
vopt = 3 × 1013 (λ = 10 μm), Δv = 0.13vopt. As an example, if our detector

has a D*t of 1012, then the BLIP condition obtains when θB = 4.6°.
We can also use the curves of Figure 11-29 in reverse to find D*B for
a given θB. For instance, for θB = 4.6° we follow the sequence W → X →
Y → Z to find D*B = 1012; in the process we also learn that the detector
temperature must be below 48 K in order to be background limited.
The most widely used material for infrared imaging near λ = 10 μm is
the semiconductor HgCdTe [28] (MCT) whose composition can be adjusted
to yield an energy of a photon with λ ~ 10 μm. The photoconduction in this
case is due to excitation across the energy gap of the semiconductor. Typical
MCT photoconductive detectors use N-type material so that the carriers
responsible for the signal (and noise) are the (minority) holes. Our theoretical
discussion up to this point applies if we merely take nB, nt, and ns, respec-
tively, as the density of holes excited by the background radiation, thermal
process, and the "signal" radiation.
The most important task that confronts the infrared detector scientist is
to develop materials that enable background-limited performance at the high
est possible temperature. The background limit condition nt = nB can be
written using (11.10-2) as
(11.10-12)
Since nt invariably increases with T, it follows from (11.10-12) that the tem
perature for background-limited operation increases with the carrier lifetime
τ0. In the example given above, the temperature T for background-limited
detection condition nt = nB is given, according to (11.10-11), by the condition
nB = nt i.e., by the value of T satisfying
(11.10-13)
In a typical HgCdTe at 77 K, the carrier lifetime is τ0 ~ 10-6 s while in

our quantum well detector τ0 ~ 10-11 s [the time for an excited carrier to
drop in energy below the top of the well thus becoming immobile (trapped).
This happens after the emission of only a few optical-branch phonons by
the excited carrier.]
It follows that HgCdTe is background limited and thus has an NEP,
described by (11.10-6), at a higher temperature than aGaAs/GaAlAs detector
used in the above example. To illustrate this point, we show in Figure 11-30
a plot of the thermal generation current that is the rate of decay (per unit
of incidence area) of thermally excited carriers21
(11.10-14)
where nt is the density of thermally excited carriers (for HgCdTe it is the

minority carrier density [26-28]). It-g is commonly used in the infrared
imaging community to compare different materials since at the background
limit (BLIP) it is equal to the rate ηℐB/hvopt of (absorbed) background pho-
tons (per cm2) incident on the detector. Since the latter rate is determined
by system considerations (see Equation 11.10-3 and the following discus-
sion), given the background absorbed photon flux ηℐB/hvopt we can deter
mine at a glance the temperature to which our detector needs to be cooled
to achieve BLIP condition or, equivalently, the temperature to which the
detector needs to be cooled for BLIP operation at a given background photon
f ux.
l
We note by comparing Figure 11-30(a) to 11-30(b) that at given T, It-g
in HgCdTe is ~6 orders of magnitudes smaller than in the GaAs/GaA1As
detector, reflecting mainly the difference in carrier lifetime τ0. It is
~10-11 s in GaAs/GaAlAs [25] and ~10-6 s in HgCdTe. As an example con-
sider a system subject to a background photon flux near λ = 10 μm of 3 ×
21This rate is equal, at thermal equilibrium, to the rate at which the "thermal" carriers are
generated.
Figure 11-30 (a) Thermal equivalent current for HgCdTe IR devices. On the left
ordinate it is expressed in A/cm2 while on the right ordinate it is expressed in the
equivalent arrival rate of (10 μm) photons/cm2. (b) Thermal equivalent generation
current for GaAs/AlGaAs.
1013 cm-2 - s-1. The HgCdTe detector becomes, according to Figure 11-30(a),
background limited at ~80 K while the GaAs/GaAlAs detector [Figure (b)]
needs to be cooled to ~45 K. If we calculate the D*t corresponding to this
incoming background flux, we obtain using, for example, Figure 11-29(a) at
45 K, D*t = 2.2 × 1013 cm-Hz1/2-W-1.
OPTICAL AMPLIFICATION IN FIBER LINKS 449
11.11 OPTICAL AMPLIFICATION IN FIBER LINKS
Optical amplification in fiber links [31-33] has recently been recognized as

having major system implications for very long distance transmission of
information (>1000 km) using optical fibers and for distribution systems
involving a large number of subscribers. These purely optical repeaters may,
in most cases, obviate the need for the repeater stations currently used that
involve detection, electronic amplification, and remodulation of a (new)
launched optical beam.
The raison d'être for the optical amplifiers is that they make it possible
to maintain the optical power at sufficiently high levels along the path so
that the signal-to-noise ratio (SNR) degradation due to signal shot noise and
receiver noise is reduced to practical inconsequence.
A new and dominant noise source, amplified spontaneous emission,
however, is introduced by the optical amplifier [35], and its effect on the
signal-to-noise ratio (SNR) of the detected signal current will be considered
below. Before doing so, we will briefly review the relevant physics of the
amplifier.
The most common amplifier uses a transition at λ = 1.535 μm in an
E3+r ion introduced as a dopant into the present in a silica fiber [34, 35]. The
pertinent energy levels are shown in Figure 11-31(a).
The laser transition can be pumped by radiation at λ ~ 0.98 μm or λ ~
1.49 μm as shown. This pumping field is usually obtained from semiconductor
lasers and is coupled into the amplifying fiber whose length is typically
between a few meters and a few tens of meters. A schematic diagram of the
amplifier configuration is shown in Figure 11-31(b). The fiber amplifier section
can be spliced smoothly into the fiber transition. A plot of the gain vs. signal
wavelength is shown in Figure 11-32.
The main effect of the optical amplifier on the SNR of the detected signal
is to add, upon detection, a noise current component, at frequencies near
that of the signal current. This noise is due to beating between the amplified
(optical) spontaneous emission (ASE) power of the amplifier and the signal
optical field. We first need to obtain an expression for the optical spontaneous
emission power at the output of an optical amplifier. This topic is the subject
of Appendix D. The main result, Equation D-8, is that the (amplified) spon-
taneous emission power in a single mode with a spectral bandwidth Δvopt
at the output of an optical amplifier is [36]
(11.11-1)
where G is the power gain of the optical amplifier and
(11.11-2)
is the atomic inversion factor of the transition. It accounts for the larger
Figure 11-31 (a) The pertinent energy level diagram of E3+r in silica. For pumping at
λ = .98 μm (preferred). (b) A schematic diagram showing the amplifying fiber
spliced into the transmission fiber and the method for coupling the pump radiation
into the fiber.
Figure 11-32 Noise factor and gain spectrum of the silica E3+r fiber amplifier for a
constant pump power of 34.2 mW at 0.98 μm. (After Reference [35].)
value of N2, and hence larger spontaneous emission power, in atomic (am
plifier) systems in which N1 ≠ 0.22
If we denote the optical power at ω2 as S and that of the spontaneous
emission at ω1 as F, then the beat current component with a frequency
ωm = ω2 - ω1 is (see Equation 11.4-4)
(11.11-3)
where Φase and φs are the phases of the ASE field and the signal optical
field, respectively. The mean-squared beat current is then
(11.11-4)
which, using (11.11-1) and putting Δvopt = 2Δfsig, yields23
(11.11-5)
In the remainder we will drop the subscript "sig" and use Δf only.
Consider an optical in-line amplifier as shown in Figure 11-33. The input
signal power is S0, and it enters the amplifier in a single transverse (usually
the fundamental) fiber mode. The amplified output signal is GS0, while F0,
as given by ( 11.11-1 ), represents the (optical) amplified spontaneous emission
power at the output, which is generated within the amplifier in a band Δv.
If we were to detect the signal at the input to the amplifier, the main noise
contribution would, in an ideal case, i.e., a noiseless receiver, be that of the
signal shot noise so that the signal-to-noise power ratio (SNR) at the input
22In a laser the gain per pass is given by G = exp[a(N2 - N1)Lamp] where Lamp is the length
and a is a constant depending on the atoms. A large N1 thus causes a larger N2 for a given
gain. The SE power is proportional to N2.
23Two ASE frequency bands, each with a width ∆vsig, one above and one below the signal
frequency contribute incoherently to the beat power so that the effective Δ vopt = 2∆fsig.
Figure 11-33 An optical amplifier with a power gain G and an input signal power
S0. F is the total power of the amplified spontaneous emission (ASE) at the output
of the amplifier in the appropriate bandwidth ∆v.
to the amplifier is
(11.11-6)
The detected signal "power"24 at the output is
(11.11-7)
while the noise power is that of the ASE-signal noise (11.11-5) and the shot
noise
(11.11-8)
The noise current component that is due to beating of ASE frequencies with
themselves is proportional to F2 0 and can be made to be negligible compared
to the ASE-signal current if the signal power S(z) is not allowed to drop too
far and/or by optical filtering. We have neglected for similar reasons the shot
noise due to the ASE. The (S/N) ratio at the output of the amplifier is thus
(11.11-9)
where we assumed a 100 percent detector quantum efficiency. For large gain
G ≫ 1, the second term in the denominator of (11.11-9), dominates, and
(11.11-10)
The ratio of the input (SNR) to the output value is thus
which in an ideal, four-level (N1 = 0, μ = 1) amplifier is equal to 2. The

single high-gain optical amplifier will thus degrade the SNR of the detected
output by a factor of 2 (3 db). We recall that this degradation is tolerated
only in order to save the signal from the, far worse, fate of succumbing, in
its attenuated state, to the noise of the receiver. An experimental verification
of the SNR result is shown in Figure 11-32.
24The "power" everywhere is taken as the mean square of the current. Since our final results
involve only (signal-to-noise) power ratios, this procedure is justified.
In a very long (100 km) fiber link, we will need to amplify the signal a
number of times. We will consequently develop in what follows a formalism
for treating systematically cascades of amplifiers.
A generalization of the expression (11.11-9) for the SNR of the detected
signal at an arbitrary point z along the link is to write
(11.11-11)
where the last term in the denominator represents the mean-squared thermal
noise current of the receiver (at point z) whose effective noise temperature
is Te. R is the output impedance of the detector including the receiver's
input impedance. Equation (11.11-11) neglects, again, the shot noise due to
the ASE, the ASE-ASE beat noise, and intensity fluctuation noise of the
source laser. If the signal power S(z) can be maintained above a certain level
by repeated amplification, we can neglect the receiver noise term. Under
these realistic circumstances, the SNR expression (11.11-11) becomes
(11.11-12)
S(z) is the signal power at z, while F(z) is the total ASE power at z originating
in all the preceding amplifiers (z' < z).
Let us next consider the realistic scenario of a long fiber with amplifiers
employed serially at fixed and equal intervals (z0), as illustrated in Figure
(11-34).
The signal power level S(z) at the fiber input and at the output of each
amplifier is S0. The signal is attenuated by a factor of L ≡ exp(-αz0) in
the distance z0 between amplifiers and is boosted back up by the gain G =
L-1 = eαz0 at each amplifier to the initial level S0. The spontaneous emission
power F(z) is attenuated by a factor L between two neighboring amplifiers
and increases by an increment of F0 at the output of each amplifier. We
employ Equation 11.11-1 to calculate the SNR of the detected current at the
output of the nth amplifier. Assuming G ≫ 1, the result is
(11.11-13)
Figure 11-34 A fiber link with periodic amplification.

where, because of the high signal and ASE levels, we neglected the thermal
receiver noise. When exp(αz0) = G ≫ 1, we find az-1 (more exactly an n-1)
dependence of the SNR rather than the exp(-αz) dependence of a fiber
without amplification in which the main noise mechanism is shot noise. The
physical reason for this difference is that the repeated amplification keeps
the signal level high as well as the level of the signal-ASE beat noise. The
latter is kept well above the signal shot noise. A fixed amount of beat noise
power is thus added at each stage leading to the inverse distance dependence
of the SNR.
Equation (11.11-13) suggests that the SNR at z can be improved by
reducing z0, i.e., by using smaller intervals between the amplifiers which,
of course, entails reducing the gain G = exp(αz0) of each. Let us take the
limit of Equation (11.11-13) as z0→ 0, i.e., the separation between amplifiers
tends to zero. In this limit the whole length of the fiber acts as a distributed
amplifier with a gain constant g = α, just enough to maintain the signal at
a constant value. Since S(z) is a constant, we need only evaluate the ASE
optical power F(z) in order to obtain, using (11.11-12), an expression for the
SNR at z. To find how much noise power is added by the amplifying fiber,
we consider a differential length dz. It may be viewed as a discrete amplifier
with a gain of exp(gdz) so that its contribution to F(z) is given by (11.11-1)
as
(11.11-14)
or
(11.11-15)
where, since no spontaneous emission is present at the input, we used

F(0) = 0. Using 11.11-15 in (11.11-11) and taking S(z) = S0, g = a results
in
(11.11-16)
We can also obtain (11.11-16) as the limit of (11.11-13) when z0 → 0. It is

interesting to compare the (ideal) distributed amplifier to the discrete am
plifier case of Equation (11.11-13)
(11.11-17)
where we used G = exp(αz0) and n = z/z0.

Figure 11-35 shows plots of the ideal continuous amplification case de-
scribed by Equation (11.11-16) as well as two cases of discrete amplifier
cascades [Equation (11.11-13)]. The advantage of continuous amplification
Figure 11-35 A universal plot of the degradation of the SNR compared to the
initial (z = 0) value in the cases of (a) continuous amplification (g = α), (μ = 1);
(b) periodic amplification every z0 = α-1 (z' = 1, 2, 3, . . .), (μ = 1), (curve is to
be read only at z' = 1, 2, 3 . . .); and (c) periodic amplification every z0 = 2α-1
(z' = 2, 4, 6, . . .), (μ = 1), (curve is to be read only at z' = 2, 4, 6, . . .).
compared to, say, amplification every α-1 is seen to be less than 2 db so

that the latter may be taken as a practical optimum configuration. In a low-
loss optical fiber, say with a = 0.2 db/km, the distance between amplifiers
that are placed every α-1 km would be 21.7 km. Figure 11-36 shows the
SNR of the detected signal along a realistic link for the case of (a) continuous
amplification; (b) discrete amplifiers spaced by z0 = α0-1: and (c) for the case
of no amplification at all. The launched power is P0 = ρ mW, λ = 1.55 μm,
Δf = 109 Hz, and α = 0.2 db/km. Curve (b) is to be read only at multiples
of z = α-1 = 21.7 km, which are the output planes of the optical amplifiers.
Curve (c) assumes detection with a receiver with Te = 725 K (F = 4 db)
and an input impedance of 1000 Ω.
We note that if, for example, we need to maintain a SNR exceeding 50
db, we must use a fiber link shorter than 100 km if no amplifier is used, but
if laser amplifiers are used every, say, z0 = α-1 ( = 21.7 km), fiber length in
excess of 1000 km can be employed.
Serious consideration has also been devoted to the use of semiconductor
(SC) laser amplifiers [31]. These are identical in their construction to semi-
conductor laser oscillators, which are discussed in Chapter 15, except that
the facets are coated with antireflection layers to reduce optical feedback
and thus prevent oscillation from taking place. The main advantage is the
possibility of very large gains, > 20 db in a short (< ~400 μm) semiconductor
chip. The main disadvantages of the SC amplifier compared to the fiber
Figure 11-36 SNR of detected signal in a fiber link with (a) a continuous amplifier
g = α, (μ = 1); (b) discrete amplifiers employed every absorption length α-1 =
21.7 km (0.2 db/km fiber loss), (μ = 1) (curve is to be read only at multiples of
21.7 km); and (c) no optical amplification and detection with a receiver with a
noise figure of 4 db. The power launched into the fiber is 5 mW, the fiber loss is
0.2 db/km, λ = 1.55 μm, the detection bandwidth is Δf = 109, and the detector
load impedance is 1000 ohms.
amplifier is the presence of residual reflections and the resulting need for
optical isolators. The presence of even minute reflection (R < 10-5) can
give rise to instabilities and excess noise in the source laser oscillator. Im-
pressive results, however, have been demonstrated [36].
The above discussion centers on the use of optical amplification in long
distance transmission of data. A second class of applications, no less im-
portant, is that of distribution systems with a very large number of sub
scribers. The use of optical amplifiers makes it possible to maintain the power
arriving at a subscriber's premises at sufficiently high levels so as not to be
degraded by the receiver noise. The number of subscribers that can thus be
served by a single laser can be increased by anywhere from 1 to 3 orders
of magnitude. This topic is the subject of Problem 11.13.
Problems
11.1 Show that the total output shot-noise power in a photomultiplier in-
cluding that originating in the dynodes is given by
where δ is the secondary-emission multiplication factor and N is the number

of stages.
PROBLEMS 457
11.2 Calculate the minimum power that can be detected by a photoconductor

in the presence of a strong optical background power PB. Answer:
11.3 Derive the expression for the minimum detectable power using a pho
toconductor in the video mode (that is, no local-oscillator power) and as
suming that the main noise contribution is the generation-recombination
noise. The optical field is given by e(t) = E(1 + cos ωmt) cos ωt, and the
signal is taken as the component of the photocurrent at ωm.
11.4 Derive the minimum detectable power of a Ge:Hg detector with char-
acteristics similar to those described in Section 11.7 when the average cur-
rent is due mostly to blackbody radiation incident on the photocathode.
Assume T = 295 K, an acceptance solid angle Ω = π and a photocathode
area of 1 mm2. Assume that the quantum yield η for blackbody radiation at
λ < 14 μm is unity and that for λ > 14 μm, η = 0. [Hint: Find the flux of
photons with wavelengths 14 μm > λ > 0 using blackbody radiation formulas
or, more easily, tables or a blackbody "slide rule."]
11.5 Find the minimum detectable power in Problem 11.4 when the input
field of view is at T = 4.2 K.
11.6 Derive Equations (11.6-15) and (11.6-16).
11.7 Show that the transit time reduction factor (1 - e-iωmτd)/iωmτd in Equa
tion (11.7-7) can be written as
where
Plot a and β as a function of ωmτd.
11.8 Derive the minimum detectable optical power for a photodiode operated
in the heterodyne mode. (Answer: Pmin = hv Δv/η.)
11.9 Discuss the limiting sensitivity of an avalanche photodiode in which

the noise increases as M2. Compare it with that of a photomultiplier. What
is the minimum detectable power in the limit of M ≫ 1, and of zero back-
ground radiation and no dark current?
11.10 Derive an expression for the magnitude of the output current in a

heterodyne detection scheme as a function of the angle θ between the signal
and local-oscillator propagation directions. Taking the aperture diameter (see
Figure 11-6) as D, show that if the output is to remain near its maximum
(θ = 0°) value, θ should not exceed λ/D. [Hint: You may replace the lens
in Figure 11-6 by the photoemissive surface.] Show that instead of Equation

(11.4-4) the current from an element dx dy of the detector is
The propagation directions lie in the z-x plane. The contribution of dx dy to

the (complex) signal current is thus
11.11 Show that for a Poisson distribution (footnote 9)(ΔN)2 = N.
11.12 Calculate the smallest temperature microment that can be measured

by an infrared detector "looking" at an object at T = 350 K with a back-
ground temperature of T = 300 K. The detector has a D*λ = 1011 cm
(Hz)1/2/W and responds to Δλ ~ 0.1λ centered on λ = 10 μm. The output
circuit bandwidth is Δf = 103 Hz.
11.13 Assume a fiber distribution network fed by a single semiconductor laser

at λ = 1.55 μm with a power output P0 = 10 mW. The power is divided
into N branches, amplified (in each branch) and then divided again into M
branches.
Determine the maximum number of "subscribers" NM that can be ser-
viced by the system assuming: Δf = 109 Hz; R (receiver input impedance)
is 103 ohms, Te = 1000 K; and a minimum SNR at the subscriber of 42 db.
The maximum power level at the output of the amplifiers is 10 mW.
References
1. Yariv, A., Quantum Electronics, 3d ed. New York: Wiley, 1988, p. 54.
2. Engstrom, R. W., "Multiplier phototube characteristics: Application to
low light levels," J. Opt. Soc. Am. 37:420, 1947.
3. Sommer, A. H., Photo-Emissive Materials. New York: Wiley, 1968.
4. Forrester, A. T., "Photoelectric mixing as a spectroscopic tool," J. Opt.
Soc. Am. 51:253, 1961.
5. Siegman, A. E., S. E. Harris, and B. J. McMurtry, "Optical hetero-
dyning and optical demodulation at microwave frequencies." In Optical
Masers, J. Fox, ed. New York: Wiley, 1963, p. 511.
6. Mandel, L., "Heterodyne detection of a weak light beam," J. Opt. Soc.
Am. 56:1200, 1966.
7. Chapman, R. A., and W. G. Hutchinson, "Excitation spectra and pho
toionization of neutral mercury centers in germanium," Phys. Rev.
157:615, 1967.
8. Teich, M. C., "Infrared heterodyne detection," Proc. lEEE 56:37, 1968.
9. Buczek, C., and G. Picus, "Heterodyne performance of mercury doped
germanium," Appl. Phys. Lett. 11:125, 1967.
REFERENCES 459
10. Lucovsky, G., M. E. Lasser, and R. B. Emmons, "Coherent light de

tection in solid-state photodiodes," Proc. IEEE 51:166, 1963.
11. Riesz, R. P., "High speed semiconductor photodiodes," Rev. Sci. Instr.
33:994, 1962.
12. Anderson, L. K., and B. J. McMurtry, "High speed photodetectors,"
Appl. Opt. 5:1573, 1966.
13. D'Asaro, L. A., and L. K. Anderson, "At the end of the laser beam, a
more sensitive photodiode," Electronics, May 30, 1966, p. 94.
14. Shockley, W., "Hot electrons in germanium and Ohm's law," Bell Syst.
Tech. J. 30:990, 1951.
15. McKay, K. G., and K. B. McAfee, "Electron multiplication in silicon
and germanium," Phys. Rev. 91:1079, 1953.
16. McKay, K. G., "Avalanche breakdown in silicon," Phys. Rev. 94:877,
1954.
17. McIntyre, R., "Multiplication noise in uniform avalanche diodes," IEEE
Trans. Elect. Devices ED-13:l64, 1966.
18. Lindley, W. T., R. J. Phelan, C. M. Wolfe, and A. J. Foyt, "GaAs
Schottky barrier avalanche photodiodes," Appl. Phys. Lett. 14:197, 1969.
19. Nahory, R. E., M. A. Pollack, E. D. Beebe, and J. C. DeWinter, "Con
tinuous operation of a 1.0 μm wavelength GaAs1-xSbx/AlyGa1-y-As1-xSbx
double-heterostructure injection laser," Appl. Phys. Lett. 28:19, 1976.
20. Bar-Chaim, N., K. Y. Lau, I. Ury, and A. Yariv, "High speed
GaAlAs/GaAs photodiode on a semi-insulating GaAs substrate," Appl.
Phys. Lett. 43:261, 1983.
21. Wang, S. Y., D. M. Bloom, and D. M. Collins, "20-GHz bandwidth
GaAs photodiode," Appl. Phys. Lett. 42:190, 1983.
22. Wang, S. Y., and D. M. Bloom, "100 GHz bandwidth planar GaAs
Schottky photodiode," Elec. Lett. 19:554, 1983.
23. Valdmanis, J. A., G. Mourou, and C. W. Gabel, "Picosecond electro-
optic sampling system," Appl. Phys. Lett. 41:211, 1982.
24. Kolner, B. H., D. M. Bloom, and P. S. Cross, "Characterization of high
speed GaAs photodiodes using a 100-GHz electro-optic sampling sys-
tem," 1983 Conference on Lasers and Electro-optics, paper ThGI.
25. Kinch, M. A. and A. Yariv, "Performance limitations of GaAs/GaAlAs
infrared superlattices," Appl. Phys. Lett. 55:2093, 1990.
26. R. E. Burgess, "Fluctuations in the number of electrons and holes in
semiconductors." Proc. Phys. Soc. B 68:661, 1955.
27. Van Der Ziel, A., Fluctuation Phenomena in Semiconductors Ch. 4,
New York: Academic Press, 1959.
28. Long, D. "On generation-recombination noise in infrared detector ma-
terials," Infrared Phys. 7:167, Pergamon Press, 1967.
29. L. C. Chiu, J. S. Smith, S. Margalit, A. Yariv, and A. Y. Cho, "Ap
plication of internal photoemission from quantum well heterojunction
superlattices to infrared detectors" Infrared Phys. 23(2):93, 1983.
30. B. F. Levine, C. G. Bethea, G. Hasnain, J. Walker, and R. J. Malek,
"High detectivity D* = 1010 cm√Hz/W GaAs/AlGaAs multiquantum

well λ = 8.3 μm Infrared Detector," Appl. Phys. Lett. 53:2196, 1988.
31. Simon J. C., "Semiconductor laser amplifier for single mode optical
fiber communications," J. Opt. Commun. 4:51, 1983.
32. Mears, R. J., L. Reekie, I. M. Jauncey, and D. N. Payne, "Low noise
erbium-doped fiber amplifier operating at 1.54 mm," Elec. Lett. 23:1026,
1987.
33. Hagimoto, K., et al. "A 212 Km non-repeatered transmission experi
ment at 1.8 Gb/s using LD pumped Er3+ -doped fiber amplifiers in an
Im/direct-detection repeater system," in Proceedings of the Optical
Fiber Conference, Houston, TX, postdeadline Paper PD15, 1989.
34. Olshansky, R., "Noise figure for Er-doped optical fibre amplifiers,"
Elec. Letts. 24:1363, 1988.
35. Payne, David N., "Tutorial session abstracts," Optical Fiber Com
munication (OFC 1990) Conference, San Francisco 1990. Published by
Opt. Soc. of Am., Washington, D.C.
36. See, for example, Eisenstein, G., U. Koren, G. Raybon, T. L. Koch,
M. Wiesenfeld, M. Wegener, R. S. Tucker, and B. I. Miller, "Large-
signal and small-signal gain characteristics of 1.5 mm quantum well
optical amplifiers," Appl. Phys. Lett. 56:201, 1990.
Supplementary Reference
37. Boyd, R. W., Radiometry and the Detection of Optical Radiation. New
York: Wiley, 1983.

Detection of Optical Radiation - Chapter-11

Uploaded by

Copyright:

Available Formats

Detection of Optical Radiation - Chapter-11

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Detection of Optical Radiation - Chapter-11

Uploaded by

Copyright:

Available Formats

11

The detection of optical radiation is often accomplished by converting the

1. The generation of mobile charge carriers in solid-state photoconductive

In this chapter we consider in some detail the operation of four of the

The limiting sensitivity of each is discussed and compared to the theoretical

11.1 OPTICALLY INDUCED TRANSITION RATES

A common feature of all the optical detection schemes discussed in this

Figure 11-1 Most high-speed optical detectors depend on absorption of photons of

taking ω1 - ω0 ≡ ω ≪ ω0. We can rewrite (11.1-3) as

and, using (11.1-1), identify V(t) as

thus, using (11.1-2), we obtain

The photomultiplier, one of the most common optical detectors, is used to

Figure 11-2 Photocathode and focusing dynode configuration of a typical

be seen, its limiting sensitivity. The photocathode consists of materials with

Figure 11-3 Photomultiplier photocathode. The vacuum level corresponds to the

Figure 11-4 Photoresponse versus wavelength characteristics and quantum

which, for typical values4*of δ = 5 and N = 9, gives G — 2 × 106.

11.3 NOISE MECHANISMS IN PHOTOMULTIPLIERS

The random fluctuations observed in the photomultiplier output are due to

1. Cathode shot noise, given according to (10.4-9) by

where ic is the average current emitted by the photocathode due to the

Minimum Detectable Power in Photomultipliers—Video Detection

where the factor (1 + m cos ωmt) represents amplitude modulation of the

carrier.5 The photocathode current is given, according to (11.1-2), by

To determine the proportionality constant involved in (11.3-4), consider the

where vs = ωs/2π, P is the average optical power, and η (the quantum

The signal output current at ωm is

If the output of the detector is limited by filtering to a bandwidth Δv centered

Figure 11-5 Equivalent circuit of a photomultiplier.

Numerical Example: Sensitivity of Photomultiplier

Consider a typical case of detecting an optical signal under the following

vs = 6 × 1014 Hz (λ = 0.5 μm)

id = 10-15 ampere (a typical value of the dark photocathode current)

Substitution in (11.3-10) gives

Pmin = 3 × 10-16 watt

The corresponding cathode signal current is ic ~ 10-17 ampere, so the as-

Signal-Limited Shot Noise

This corresponds to the quantum limit of optical detection. Its significance

11.4 HETERODYNE DETECTION WITH PHOTOMULTIPLIERS

In the heterodyne mode of optical detection, the signal to be detected

The local-oscillator field originates usually at a laser at the receiving end,

A schematic diagram of a heterodyne detection scheme is shown in Figure

Figure 11-6 Schematic diagram of a heterodyne detector using a photomultiplier.

which, using (11.4-2) can be written as

The total cathode shot noise is thus

The mean-square signal current at the output is, according to (11.4-4),

The signal-to-noise power ratio at the output is given by

which corresponds to the quantum-limited detection limit. The minimum

Numerical Example: Minimum Detectable Power with a Heterodyne System

It is interesting to compare the minimum detectable power for the heterodyne

v = 6 × 1014 Hz(λ = 0.5 μm)

(Ps)min ≃ 4 × 10-18 watt

to be compared with Pmin ≃ 3 × 10-16 watt in the video case.

Limiting Sensitivity as a Result of the Particle Nature of Light

The quantum limit to optical detection sensitivity is given by (11.4-10) as

where N is the average number of photons arriving at the photocathode per

If the photons arrive in a perfectly random manner, then the number of

If we convert the last result to power by multiplying it by hv and recall that

in agreement with (11.4-10).

One of the major considerations in optical detectors is their frequency re