4.3.3. Data Recovery: March 5, 2008
4.3.3. Data Recovery: March 5, 2008
I(t) = Ip + is (t)
where Ip = RPin is the average current, and is (t) is the noise current. For shot noise, is is a
stationary random process with Poisson statistics. We will approximate is with Gaussian statistics.
The autocorrelation of the noise current is
∞
is (t)is (t + τ )= Ss (f )e2πif τ df,
autocorreclation
−∞
where Ss = qIp is the power spectral density. Shot noise is white noise, which means that the
noise power is constant across frequency. The noise variance is obtained by using τ = 0, and is
written
∞
σ 2s = i2s (t) = Ss (f )df = 2qIp Δf.
−∞
Here, Δf is the effective noise bandwidth of the receiver, and depends on the receiver design. In
general, the effective noise bandwidth is given by
∞
Δf = 2qIp |HT (f )|2 df.
0
The photodetector dark current also has shot noise, so that the total RMS noise current due to
shot noise is given by
σ 2s = 2q (Ip + Id ) Δf.
S. Blair March 5, 2008 92
I(t) = Ip + iT (t)
This expression is valid for a receiver using a pin photodetector. For an uncooled pin detector
at room temperature, thermal noise usually dominates (unless the incident optical power is very
large). Under thermal noise only,
RL R2 Pin
2
SNR = ,
4kB T Fn Δf
and it is desirable to have large RL . However, large RL decreases receiver bandwidth.
The noise equivalent power, or NEP, is the minimum optical power per unit bandwidth that
achieves SNR=1. In the case of thermal noise,
RL R2 Pin
2
P2 4kB T Fn
1= ⇒ in =
4kB T Fn Δf Δf RL R 2
Pin 2 kB T Fn
NEP = √ = .
Δf R RL
√
The detectivity is defined as NEP−1 . For a typical pin receiver, the NEP∼ 1 − 10 pW/ Hz. As
an example, for 1 MHz receiver bandwidth, the incident optical power must be in the 1-10 nW
range.
For cooled pin detectors or in situatations where the incident optical power is large, then shot
noise dominates. This is called the shot-noise limit.
R2 Pin
2
RPin
SNR = = ,
2qRPin Δf 2qΔf
where detector dark current was ignored. We can also express SNR in terms of the number of
incident photons comprising a “1” bit Np . The energy in a bit of duration 1/B is Ep = Pin /B.
The number of photons is then
Ep Pin
Np = = .
hν Bhν
Using a receiver bandwidth Δf = B/2, the signal to noise ratio becomes
Ip = M RPin ,
and we would expect an increase in SNR by the factor M 2 for thermal noise, but no change for
shot noise. However, due to the avalanch process, shot noise is enhanced. This effect is described
by
σ 2s = 2qM 2 FA (RPin + Id ) Δf,
where FA is the APD excess noise factor defined by
FA = kA M + (1 − kA ) (2 − 1/M ) .
When FA = 1, Ip and σ s are increased by same factor and the SNR is unchanged in the shot noise
limit.
In general, the SNR for APD receivers is worse than that for pin receivers when shot noise
dominates, but can be considerably better when thermal noise dominates (because Ip ∝ M ). The
total SNR for an APD receiver is
Ip2
SNR =
σ 2s + σ 2T
(M RPin )2
= .
2qM 2 FA (RPin + Id ) Δf + 4 (kB T /RL ) Fn Δf
(M RPin )2 RL R2 Pin
2
SNR = = M 2,
4 (kB T /RL ) Fn Δf 4kB T Fn Δf
3 4kB T Fn
kA Mopt + (1 − kA ) Mopt = .
qRL (RPin + Id )
For Si APD’s for which kA 1, Mopt ∼ 100, while for InGaAs APD receivers for which kA ∼ 0.7,
Mopt ∼ 10.