A Simplified Approach For Digital Optical Receiver Design
A Simplified Approach For Digital Optical Receiver Design
A simplified theory for the performance of a digital optical receiver is developed. The receiver sensitivity
is calculated in terms of circuit parameters, received and equalized pulse shapes, photodiode parameters
and bit-rate. An excellent agreement between this theory and a more complicated analysis by
Personick [4] is demonstrated. It is shown that the receiver sensitivity may be improved by launching
reduced-width pulses into the fibre, particularly if fibre bandwidth is a significant limitation. Reduced-
width pulses bring benefits in source power consumption and lifetime, and in timing recovery.
1. Introduction
The sensitivity of an optical receiver using an avalanche photodiode depends on a balance between
signal-dependent and signal-independent noise. This problem has been discussed by several authors
[1-4], of whom Personick [4], gives a detailed and comprehensive analysis. Unfortunately, deriving
the shot noise at the decision time leads to expressions which are not simple to evaluate. In view of the
approximations used in expressing the excess multiplication noise and in calculating error rates it was
felt that a simplification of the treatment of shot noise would be justified and could lead to a theory of
optical receiver design which would be easier to apply to practical systems.
We have achieved such a simplification, with an inaccuracy which is negligible in view of other
assumptions made, by relating the shot noise to the average photocurrent over the bit-time. We compare
the results of our treatment with those of Personick [4] and show that the maximum disagreement is
only 0.6 dB in received optical power. We then use our theory to find the optimum launched pulse
width from a source such as a semiconductor laser, which degrades at a rate depending on the emitted
power. We show that using reduced-width pulses permits the noise-equivalent bandwidth of the receiver
to be reduced and consequently the receiver sensitivity is increased.
2. Theory
2.1. T h e optical receiver
The basic receiver is shown as a block diagram in Fig. 1, which also shows the important noise sources.
We can represent the binary digital pulse stream incident on the photodiode by [5 ] :
where p(t) is the received optical power, T is the bit-time and hp(t) is the pulse shape. The amplitude
parameter, bn, can take two values b e n and b o e r corresponding to ON and OFF pulses, respectively.
We take
;_~hp(t) dt= 1
so that bn is the energy in the n m pulse. The mean output current from the photodiode at time t is:
9 1978 Chapman and Hall Ltd. Printed in Great Britain. 211
D. R. Smith, L Garrett
hp(t)
Cd Rb ib(t)
r
RA CA
~ eA~
iA(t) ~
I "
Equaliser
r--~.-i v~
>
--haul(t)
Heq(f)
L HB If') ,
Figure 1 A schematic diagram of an optical receiver showing circuit components and noise sources.
T/e
ip(t) = ~-~#p(t)
where # is the mean avalanche gain, ~ is the quantum efficiency and h ~ is the photon energy. This
current causes a mean voltage at the output of the equalization network given by:
ATe
Vout(t) = ~ - gp(t)* h a (t)* heq(t)
where
ha(t) = g
[ '
( l / R ) +j27rfC
]
i.e., ha(t) is the impulse response of the bias circuit and amplifier, C being CA + Cb, ~: indicates Fourier
transform and * denotes convolution. Here heq(t ) is the impulse response of the equalization network.
Clearly Your(t) is of the form:
Yout = s bnhout(t--nT)
A~eg
where h~ = ~ [HP(f)HB(f)Heq(f)] ' h~2
Hp(f) is the Fourier transform of the received pulse shape hp(t), Hs(f) and Heq(f) are the transfer
functions of the bias circuit and the equalizer. Thus:
A~e~
212
A simplified approach to digital optical receiver design
_ 1 2
., _ I d:
The mean unity-gain photocurrent over the bit-time is, for an ON pulse:
~/e 1
<i0>~,oN : Y.. ,,..~b~ ~T/2 hp(t-- nT) dt
i.e., 3' is the fraction of the energy of a single pulse which is contained within its bit-time. These values of
<i0>T can be substituted in Equation 1 to calculate the worst-case mean square shot-noise voltages.
Personick [4] calculates the mean square shot-noise voltage as a function of time instead of approxi-
mating it to the shot noise on the mean unity-gain photocurrent over the bit-time. By making this
approximation we achieve substantial simplification of the final expressions and calculations for the
receiver sensitivity with negligible additional error.
For the thermal noise we use the normal expression
<~>__ {2xo+
~Rb
SOA=f= -= I Hp0O
[Hou,(f)2df + SEA2::~,Hea(f)[2df
where 0 is the absolute temperature, SI is the spectral density of the amplifier noise current source and
SE is the spectral height of the amplifier noise voltage source. The total mean square noise voltage at the
decision time is thus:
[
<v~> = A z e<io>Tg=+x + 2kO-t-Si]
Rb - I[H~
Hp(f) I 2 f: df+A2SEs (4)
We have used the approximation g2 : ~ ~+x. Alternatively, McIntyre's [6] more accurate expression could
be used at the expense of algebraic complication in finding the optimum avalanche gain.
We now normalize, following Personick [4], so that Oout(t = nT) = b,, which means putting
Arle~/hfZ = 1 and hout(t = 0) = 1. It is useful also to make the bandwidth integrals in Equation 4
independent of bit-time T, so that their numerical value depends only on the shape of the received and
equalized pulses and not on their scale. To this end we introduce the dimensionless time and frequency
variables z = tiT and r = fir. We want to find what functions H'out(~b) and H~,(4)) must replace Hout(f)
and Hv(J0 when r is substituted f o r f T and the integrations performed with respect to r One can easily
show, by considering the Fourier transforms of hout(z) and hp(r), that
1
H'out(r = ~-Hout(j0
:-I-oo,<:> = =
3--[HI,(:) d: - Tf_ dqL
213
D. R. Smith, L Garrett
and z = ~ I + -Rb
-+ / 2 + Te- 2 '3E13
i.e. Z is a dimensionless parameter representing the signal-independent noise terms, we obtain from
Equation 4:
and
hf2 g . Z
<v~r> = <to>TTI2+ 9
(s)
(bON--VD)/aON -- (V D - - b o F F ) / a O F F =
and
RE - 1 89
Then if bOFF is zero, the required energy per pulse to achieve the maximum error rate characterized
by Q is:
With a non-avalanche photodiode, ~ = 1 and the shot-noise terms are usually negligible in comparison
with the thermal noise. Thus
bON = 2 Q n a Z v Z , if= 1
rl
which is, of course, identical to Personick's [4] result for a non-avalanche photodiode.
govt u~ r~ 212 K
(7)
where
K = --1+ 1+ ~-~-]
From Equations 6 and 7 we obtain our main result for the minimum required pulse energy:
214
A simplified approach to digital optical receiver design
Though it is a cumbersome expression, L is a parameter which depends only on the fraction 7 of the
pulse energy contained within the bit-time and on the x-factor for the photodiode. Fig. 2 is a plot of L
versus 7 for three typical values of x, and may be used to find L for any received pulse shape.
The required energy per pulse at optimum gain is given by Equation 8, and can be calculated for any
received and equalized pulse shapes. The dependence on bit-rate is entirely contained in Z, which is
dominated by the term (2zrC) 2 SEI3/Te 2 except at very low bit-rates, and which enters to the power
0.12-0.17 (for x of 0.3-0.5). The required pulse energy is therefore very insensitive to bit-rate, decreas-
ing by a factor of 2-3 between 1 and 500 Mb s- 1, (assuming Sv. is constant with bit-rate).
3. Results
3.1. Dependence of required pulse energy on pulse width
The optimum received pulse shape is an impulse [4], since then [H~(~)12 = 1 for all 4) and the band-
width integrals I2 and 13 are as small as they can be made by varying H~(q~).
We can calculate the power penalty which has to be paid for having received pulses which are not
impulses. The penalty in dB is given by:
I IX/(2+2x)I1/(l+x)
3 2
L [
AboN = 101oglo } l ~ [
I: 3,opt 2,opt opt)
4.0
3.5
~c=1
3.2
28
sc=0.:
2/.
215
D. R. Smith, L Garrett
18
16
14 / //
m12
"o
S 10
g
Q..
8
/
//
from Equation 8, in which we have ignored the I2 term again in Z in comparison with the capacitative
term. The penalty then depends on the shape of the received pulse but not on the bit-rate. Fig. 3 shows
the penalty as a function of 7 for Gaussian received pulses, taking an x-factor of 0.5 for the photodiode.
The results of Personick [4] have been reproduced for comparison, and the agreement is excellent. One
can see from Fig. 3 that there is a severe penalty if the pulse energy spreads significantly beyond its
bit-time.
3.2. L a u n c h e d p o w e r r e q u i r e m e n t s
From Fig. 3, we deduce that in an optical fibre system one wants to use as narrow a launched pulse as
possible in order to minimize the penalty inflicted by dispersion and intersymbol interference of the
received pulses, and consequently reduce the noise equivalent bandwidth of the receiver. By contrast, in
a coaxial cable system, there is an optimum pulse width; a penalty accrues if too narrow pulses are
launched because of excessive attenuation of the high-frequency end of the spectrum.
However, in an optical fibre system, the practical narrowing of the launched optical pulse may be
limited by the peak output power which the optical source can emit without an unacceptable degra-
dation rate, for example, through facet damage in a semiconductor laser [8]. We do not have enough
information at relevant pulse repetition rates on the effect of peak power on laser degradation, so we
assume that the degradation rate rD is proportional to the emitted optical power P to some power z. For
rectangular launched pulses, the degradation rate during the pulse is
rD ~ (P,~)~
where PL is the peak emitted power. We will ignore the loss in coupling the laser to the fibre, as this does
not affect the argument, and take PL as the peak launched power also. We can now calculate the
necessary energy per pulse boN required to achieve a prescribed maximum error rate as a function of
launched pulse width, with the constraint that the total source degradation per unit time (say per bit-
time) is held constant at its maximum acceptable value RD. We have:
1 [ T/2
RD = -T" J-T~2 rD(t) dt"
For rectangular launched pulses of duration -- aLT/2 < t < aLT/2, and assuming zero output power
in the OFF state:
R D c c p~o~L
216
A simplified approach to digital optical receiver design
1c
=0.L
c~
&
0.3
0.2
We have also calculated the power penalty as a function of pulse width for the case of a rigid upper
limit to the peak power available from the laser, independent o f pulse duration and repetition rate, which
is the most pessimistic case. The results for several fibre bandwidths (~v) are shown in Fig. 4, normal-
ized to the case of full-width launched pulses and infinite fibre bandwidth (~L = 1, aF = 0). There is a
broad minimum, corresponding to the optimum trade-off between reducing the pulse overlap and reduc-
ing the pulse energy by narrowing the pulse. The benefit of using reduced-width pulses is a few decibels
if the fibre bandwidth is a significant limitation.
217
D. R. Smith, L Garrett
5
r
1:3
. /.
-6
#_
We have performed the calculations for several values of z with a fixed fibre transfer function a r .
Fig. 5 shows the results for infinite bandwidth fibre, a r = O. The case of z = 1 corresponds to constant
energy in the pulse. For larger values of z, there is a minimum in the power penalty. The curve marked
A is for a source with a fixed upper limit to the emitted optical power as in Fig. 4.
4. Discussion
4.1. Comparison with Personick's theory
Personick gives an expression for the mean square shot-noise voltage (Equation 17 of [4] p. 852):
where Xo is the 'dark current' contribution to the primary photocurrent, and, with r = 27rf,
1 Hout(Oa)
~: [hi(t -- t')] = gi(Lo ) = Heq((.o ) ( I / R ) +joaC Hp(cO)
In the case where all b n = boN (worst-case shot-noise for an ON pulse), and ignoring Xo, this expression
becomes
(n~(t = O)) = NON = e 2 g - ~ boN -- nT)h~ (t-- t') dt' .
NO N = e 2 ~2 fiY~ T
The agreement between the two theories thus depends on the shape of the received pulse. Disregarding
the constant multiplier, which is the same in both equations, the expression derived by Personick can
be written:
218
A simplified approach to digital optical receiver design
The quantity in the square brackets is a periodic waveform and so can be expressed as a Fourier series:
z
l f ~ hp(s)e2~mm/Tds
- -
T _
1 H {27rml = 1H;(m).
In particular, Co = 1/T.
Thus Personick's expression for the worst-case shot-noise during an ON pulse is:
= e Z g 2 17 b o N Z1
hgZ T
Thus the theory used here gives the same result as Personick's when 21 = / 2 . As/2 is the first term in
2;1, we expect good agreement when Hp(m) is small for [m[ ~> 1 which implies broad received pulses. In
fact, ifH~(m) = 0 for imj ~> 1 the received pulse shape must satisfy the Nyquist criterion, i.e. possess
skew symmetry about t = T. Unless such a pulse is rectangular, considerable energy is spread outside the
bit-time.
The case of an OFF pulse is perhaps best investigated by direct comparison of the results of the two
theories.
In Fig. 6 we show the error in boN resulting from the approximation used in this paper compared
with Personick's more detailed analysis, for Gaussian and rectangular received pulses and a raised-cosine
output pulse. The maximum error of 0.62 dB (the theory used here underestimates boN) occurs when
the received pulse is an impulse. For wider pulses the agreement is excellent. In fact the maximum error
involved is smaller than the error which could result from using the simple f'~ expression for the excess
0-7
0.6.
m 05
Z3g 0,4
c_
03
Ld
02
0I
Figure 6 The discrepancy between Qur
i theory and that derived by Personick [4]
0 0'1 0'2 0'3 0'~ 0'.~ 0'.6 0'7 0'8 0'-9 10 as a function of received pulse width
o~ for rectangular and Gaussian pulses.
219
D. R. Smith, L Garrett
multiplication noise instead of McIntyre's [6] more accurate form. We expect our simpler theory to
provide reasonable estimates of the performance of real systems, and to provide a sound basis for
comparison of systems.
5. Conclusions
We have described a simplified theory of noise in an optical receiver and hence derived the receiver
sensitivity. We have shown that this theory gives results very similar to the results of a more detailed
theory by Personick [4] for the types of received and equalized pulse shapes of practical interest. We
have used our simplified theory to calculate the optimum launched pulse width in a system with various
degrees of fibre dispersion, and also for various power-law dependences of laser degradation rate on
optical output power. We have shown that there is a small increase in receiver sensitivity to be gained by
using reduced-width pulses in all cases. Thus one may use return-to-zero-level coding, and possibly
reduce the mean power requirements of the laser significantly, with no penalty in receiver sensitivity or
system bandwidth. The methods used in this paper may be applied to the design of digital optical
receivers for any given received and equalized pulse shapes.
Acknowledgements
We are grateful to Dr J. E. Midwinter and other colleagues for many stimulating discussions, and to the
Director of Research of the Post Office for permission to publish this work.
220
A simplified approach to digital optical receiver design
References
1. w . M . HUBBARD, BellSyst. Tech. J. 52 (1973) 731-65.
2. R. DOGLIOTTI, A. GUARDINCERRI and A. LUVISON, Opt. Quant. Elect. 8 (1976) 343-53.
3. J.E. MIDWINTER, ibid 9 (1977) 299-304.
4. S. D. PERSONICK, Bell Syst. Tech. J. 52 (1973) 843-86.
5. A.B. CARLSON, 'Communication Systems', 2nd Edition, (McGraw-HiU, Kogakusha, 1975).
6. R.J. MCINTYRE, IEEE Trans. Electron. Devices ED-13 (1966) 164-8.
7. P. BALABAN,Bell Syst. Tech. J. 55 (1976) 745-66.
8. N. CHINONE, R. ITO and O. NAKADA, J. AppL Phys. 47 (1976) 785-6.
221