You are on page 1of 8

Evaluation of an FPGA-based optimal receiver for FSO laser

communications affected by jitter and scintillation


Oscar Cerdaa , Jaime Anguitab , and Marcelo Guarinic
a
Universidad de los Andes, Monseñor Álvaro del Portillo 12455, Santiago, Chile
b
MIRO (Millennium Institute of Research in Optics), Esteban Iturra s/n, Barrio Universitario,
Concepción, Chile

ABSTRACT
We propose an FSO receiver with adaptive-gain stage and associated filtering to reduce signal amplitude excursion
and noise for decoding the received symbols with low error rate. A digital synchronization method together with
a sampling strategy is developed for a FPGA-based receiver. The proposed techniques are evaluated through
simulation and laboratory experiments under real turbulence.
Keywords: Free-space optics, FPGA, optical turbulence, adaptive amplification, temporal correlation.

1. INTRODUCTION
Optical communication systems have been, since the 1990s, a fundamental pillar for the transfer of information
in the world, especially guided optical systems, which reach speeds of up to 40 Gb/s per channel. But it has
not been, until a few years ago, that unguided optical communication systems, or Free-Space Optical (FSO)
communication systems, have taken a more important role thanks to the great connectivity potential and its
high bandwidth wireless links over long distances, which is an alternative for radio frequency communications.
However, the effects caused by atmospheric turbulence on the communication channel greatly limit the oper-
ational availability of FSO communication systems. The main effects induced by turbulence in the transmission
of the laser beam are jitter, phase fluctuations and beam wandering. These effects result in a significant degra-
dation of the bit error rate (BER), so it is important to mitigate them in order to improve signal detection and
link performance.
Taking into account the above, we propose an FPGA-based system capable of adapting to these fluctuations
and delivering a robust estimate of the bit received.

2. USED MODULATION
The use of On-Off Keying (OOK) modulation is a common resource in free-space optical communications for
its simplicity. However, it delivers poor detection performance in the presence of amplitude fluctuations and
jitter. We propose an FPGA implementation of an optimal receiver based on binary pulse position modulation
(BPPM), in which each symbol is sampled and digitized N times to accommodate for jitter and provide an
effective synchronization.
BPPM is represented by the following symbols (which can be seen in time domain in the Fig. 1).

t ∈ [0, T2s ]

A if
s0 (t) =
0 otherwise

t ∈ [ −T

A if 2 , 0]
s
s1 (t) =
0 otherwise

Further author information: (Send correspondence to O.C.A.)


O.C.A.: E-mail: oacerda@miuandes.cl
J.A.: E-mail: janguita@miuandes.cl
Figure 1: Symbols of BPPM in time domain for our communication system.

where s1 (t) is transmitted if the sended bit is ”1” and s0 (t) is transmitted if the sended bit is ”0”, A is the
amplitude of the signal and Ts is the symbol time, given by 1/fc , with fc the transmission frequency. The most
important difference between OOK modulation and BPPM is that BPPM has same energy for both symbols
s1 (t) and s0 (t), where the energy of a symbol is given by

A2 Ts
Eb =
2

The main problem in FSO laser communications is that turbulence produce fluctuations in the energy of each
transmitted symbol, represented by an increase in the dynamic range of the signal amplitude.
As we can see in the Fig. 2, the received signal is affected in amplitude by an unknown signal, and also is
affected by white noise.

Figure 2: Transmitted symbols in laboratory, measured with an oscilloscope with fc = 1.25 MHz. Turbulence is
added with a heatgun.

This unknown signal can be represented by an ergodic random process with lognormal distribution. Then,
we can represent our received signal as r(t), like in Eq. 1.

r(t) = gc (t)sb (t) + n(t) (1)

where r(t) is the signal received at the photo-detector, sb (t) are our BPPM symbols, n(t) is a zero-mean
gaussian random process with a variance equal to N0 /2, and gc (t) is a lognormal random process with unit mean
and a variance equal to the Rytov variance of the turbulent channel.
3. OPTIMAL DIGITAL RECEIVER
In order to design an optimal receiver which be robust against the white noise, we will modify our model a bit.
Instead of using r(t) as the Eq. 1, we will use r(t) = sb (t) + n(t).

3.1 Discretization of the received signal


In order to analyze, filter and process the signal given by the photo-detector, the FPGA must be able to read
this data and process it, which is not possible in the analog domain. Because of that, an ADC will be used to
use discrete samples of our signal, which will be represented as the next discrete sequence

r[k] = r(kT ) = gc (kT )sb (kT ) + n(kT ) (2)

where T is the sampling time, given by the sampling frequency fs , and k = {0, 1, . . .}.
In order to simplify the discretization we’ll use N samples per symbol, so the sampling frequency is given by
fs = N fc , where N is an even number (N = {2, 4, 6, . . .}).

3.2 Optimal Symbol Detector


To demodulate the signal and estimate the received bit, we need to build an optimal symbol detector in the
FPGA environment.
As we said before, the FPGA use data in the digital domain, represented by Eq. 2. To design a robust
symbol detector, we’ll assume that our received signal in digital domain is given by r[k] = sb (kT ) + n(kT ).
With this in mind, using maximum likehood estimator we get the next expression for the best estimation of
a bit b̂ given N samples per symbol time Ts .
 
N/2−1 N −1
X X
b̂ = arg max  r[i]sb [i] + r[i]sb [i] (3)
b={0,1} i=0 i=N/2

We need to note that we have written this expression like this because we can see that this two terms can
be seen as a correlation between the received signal and our symbols. This can be noticed if we replace our two
symbols in the expression, giving two different terms.
If the sended bit is ”1”, then the Eq. 3 results in the sum of the first half of the sampled symbol, and if the
sended bit is ”0”, the equation results in the sum of the second half of the sampled symbol.
So we can see that the optimal symbol detector is given by a comparison of energy between the first half of
the sampled signal and the second half of it.
Now, we can obtain the error probability of our BPPM optimal symbol detector, knowing that our received
signal is a sum of one symbol and an additive gaussian noise process, resulting in the Fig. 3.

3.3 Equalizer
Until now, we’ve designed an optimal symbol detector for a received signal given the equation r[k] = sb [k] + n[k],
but we’ve not yet designed a module to reduce the effects produced by turbulence.
But what does the effect of turbulence on the time domain look like? Just like the red signal in the Fig. 4.
Now, as we’ve seen in the Eq. 2, the turbulence given by gc [k] affects the amplitude of the transmitted symbol,
where gc [k] is a lognormal random process. With this communication model, we can simulate the received digital
signal and, with a Rytov variance of 0.1, we can get a signal like the red one in the Fig. 5, with a generated
channel instant value gc [k] like the green curve.
With the presence of the turbulence, the SN R vs BER curve, in comparison to the theoretical one, are given
by the Fig. 6, where we can see a degradation of the BER metric after 5 dB of SN R.
(a) Eb /N0 vs BER graph. (b) SN R vs BER given N samples per symbol graph.
Figure 3: Signal-to-noise ratio versus bit error rate graph of our optimal digital symbol detector.

Figure 4: Effect of turbulence measured in laboratory, produced by a heatgun with a constant modulated signal
for the laser. This graph corresponds to the signal gc (kT ).

To design a module that reduces the dynamic range of amplitude of the received signal, we need to characterize
the channel instant value gc (kT ) using his normalized autocovariance function, which we can see in the Fig. 7.
As we can see, the half-width half-maximum is around the 10000th delay, that means that with a memory
of the last 10000 samples of the signal, we can estimate the next value of the lognormal random process or, on
other hand, we can use a moving average filter with a memory of 10000 samples to estimate the channel instant
value gc (kT ). So, the estimation of the channel instant value is given by the following equation.

10000
1 X
g˜c [k] = r[k − i] (4)
10000 i=1

This allows us to use this estimation of gc (kT ) as the inverse gain of a digital amplifier in the FPGA. So, the
output of the equalizer, that reduces the dynamic range of amplitude of our received digital signal is given by
the Eq. 5.

r[k] r[k]
r̃[k] = = 10000 P10000 (5)
g˜c [k] i=1 r[k − i]

4. EXPERIMENTAL AND SIMULATED MEASUREMENTS


4.1 Experimental measurements of untreated received signal
In order to see and measure the improvement in the quality of the signal, we’re going to use the eye diagram
method over the received signal at different propagation distances with the setup seen in Fig. 8.
Figure 5: Simulated received digital signal in the time domain (red), with his associated channel instant value
gc [k] (green).

Figure 6: Signal-to-noise ratio vs bit error rate in presence of a turbulent communication channel, compared to
the theoretical SNR vs BER curve.

The transmission frequency of the random generated symbols was of fc = 1.25 MHz, this conversion frequency
is due to limitations of the ADC used in the experiment. In future works, we’ll use an external ADC over 100
times more faster than this one.
Given the Fig. 9 we can see that, as the transmission distance increases, we have a smallest eye opening and
a greater dynamic range of amplitude. This effect is due to turbulence.
Please note that, for higher frequencies, is very difficult to our photo-detector to differentiate between one
symbol and another, like we can see at Fig. 9d.
With these eye diagrams we can conclude that we must include a filter for the gaussian noise and jitter, and
an equalizer to decrease the dynamic range of amplitude of the received signal.

4.2 Experimental measurements of treated received signal


Like we’ve said before, we need a filter to reduce the effect of the gaussian noise and have a cleaner signal
and, in order to reduce the effect of the turbulent channel, we also need to include the equalizer. So, the new
experimental setup is given by Fig. 10.
To measure the improvement of the signal quality we’ll use the same signal generator as the one used in Fig.
9c as an input for our new experimental setup.
Figure 7: Normalized Autocovariance function of channel instant value gc (kT ) and his half-width half-maximum
value, which represent the time constant of the measured turbulence.

Figure 8: Experimental setup used in laboratory to view the eye diagram of the received signal.

The lowpass Bessel filter is designed to have a cutoff frequency of 5 MHz, and it has been implemented in
an analog circuit, while the equalizer was implemented in the FPGA with an internal memory of 10000 samples,
just like in Eq. 5.
Given the above, we obtain the eye diagrams seen in Fig. 11.
We can already see a cleaner signal in Fig. 11a than in the input signal (given by Fig. 9c), but we still can
see the effect of turbulence in the dynamic range of amplitude.
The third one is the eye diagram of the filtered signal passed through the equalizer. We have a cleaner and
larger eye opening due to the filters, plus a smallest jitter.
Using this, we can obtain a simulated SN R vs BER curves to see how much improves our detection system.
This curves can be seen in Fig. 12
We can see that, with the addition of the filter alone, the system performance increases by 5 dB, and with
addition of the equalizer module, the system performance increases by 25 dB. However, in terms of BER, the
equalizer module only works when, in the filtered signal, we have a long series of consecutive errors, but when
the errors are sparse over time, BER of the filtered signal is equal to BER of the equalized signal.

5. CONCLUSIONS
In summary, we have seen that the BPPM is a best option for FSO laser communication (for binary symbols),
due to the equality of energy between the two symbols, and that our communication system model is given by
the Eq. 1. Also, we’ve obtained the math representation of the optimal receiver, given by the Eq. 3, and next
to that, his SN R vs BER curves.
We’ve seen that the turbulence degrade of the BER metric after 3 dB of SN R. So, we designed a system
to improve the performance of our receiver, passing it through a bessel lowpass filter and for our FPGA based
digital equalizer, giving us an improvement of performance of 25 dB.
(a) Eye diagram for a random symbols generated signal (b) Eye diagram for a random symbols generated signal
with propagation distance of 80 meters, with symbol with propagation distance of 160 meters, with symbol
time Ts = 800 ns. time Ts = 800 ns.

(c) Eye diagram for a random symbols generated sig- (d) Eye diagram for a random symbols generated signal
nal affected by simulated turbulence, with propagation with propagation distance of 160 meters and propaga-
distance of 160 meters, with symbol time Ts = 800 ns. tion frequency fc = 50 MHz.

Figure 9: Eye diagrams for random symbols, measured with the setup seen in Fig. 8. These eye diagrams are
normalized in amplitude.

Figure 10: Experimental setup used in laboratory to view the eye diagram of the received signal, including two
new blocks to clean the signal: a lowpass bessel filter and the equalizer.

Finally, we’ve seen that with the addition of lowpass filter abd the equalizer, our receiver gains the ability
of correct large series of consecutive errors, but when the errors are sparse over time, only the lowpass filter can
correct them. So, in future works, we’ll improve the equalizer performance to also correct those sparse errors.

ACKNOWLEDGMENTS
I would like to thank to Dr. Gustavo Funes and Dr. Eduardo Peters for helping in the design and measurement
of the performance of this FPGA-based method. Also, I would like to thank to Universidad de los Andes, MIRO
(a) Eye diagram of the filtered received signal, with symbol (b) Eye diagram of the equalized received signal after fil-
time Ts = 800 ns. tering it, with symbol time Ts = 800 ns.
Figure 11: Eye diagrams for random symbols, measured with the setup seen in Fig. 10. These eye diagrams are
normalized in amplitude.

Figure 12: Signal-to-noise ratio vs bit error rate for the treated signal. These results were obtained by simulation,
with transmission frequency fc = 100 MHz.

(Millennium Institute of Research in Optics), FONDEF (Fund for the promotion of scientific and technological
development of Chile) and Milenio iniciative of the chilean government.

REFERENCES
[1] Hyat, W.; Kemmerly, J. & Durbin, S. (2011). “Engineering Circuit Analysis” (8th ed.). New York: The
McGraw-Hill Companies, Inc.
[2] Haykin, S. (2000). “Communication Systems” (4th ed.). England: John Wiley & Sons, Inc.
[3] Sklar, B. (2001). “Digital communications: fundamental and applications” (2th ed.). New Jersey: Prentice-
Hall international edition.
[4] Proakis, J. & Salehi, M. (2008). “Digital Communications” (5th ed.). New York: The McGraw-Hill Com-
panies, Inc.
[5] Meyr, H.; Moeneclaey, M. & Fechtel, S. (1998). “Digital Communication Receivers: Synchronization, Chan-
nel Estimation, and Signal Processing” (1st ed.). New York: John Wiley & Sons, Inc.

You might also like