Experiment FindingsPDF Available

Abstract

We compared the BER using a conventional receiver for 10G OOK by the BER using a receiver based on machine learning. We found out that the max. reach can be increased considerably. Eye diagrames where no signal is visibel anymore, can still be analyzed by machine learning.
Preliminary results on
machine learning based
receiver in optical
communication
Prof. Dr.-Ing. Arne G. Striegler
Laboratory of optical communication - LON
University of Applied Science Munich
April 9th 2019
2
- OOK, 10Gbit/s, One polarisation modulated
- Number of Bits : 212
- Number of wavelength channels: 1
- SSMF fiber : D= 16ps/(nm km), S= 0.058 ps/(nm2km), a=0.25dB/km,
g= 1.3 1/(W km)
- Random polarisation rotation every 1km
- Dispersion map: 3x80 km SSMF followed by DCF
Simulation setup
Accumulated
chromatic dispersion
Transmission
distance
80 km 80 km 80 km
3
We compared the BER using a conventional receiver setup with the BER
using an receiver setup based on machine learning.
For this we tracked the BER along the transmission distance for different
channel input powers -3dBm, 0 dBm and +3dBm.
Details of the machine learning based receiver concepted will be presented
in the next presentation or publication.
Comparision
4
Priliminary results
Number of spans à 80km
log(BER)
Transmission distance in km
1 600 3 200 8 000 9 600
4 800 6 400
-3dBm, machine learning
0dBm, machine learning
+3dBm, machine learning
-3dBm, conventional receiver
0dBm, conventional receiver
+3dBm, conventional receiver
11 200
5
OSNR= 16.5dB
Conventional receiver: BER = 10-2.87
Machine learning based receiver: BER = 10-error free
Eye diagrammes and results in detail
0100
Time in ps
Optical power a.u.
Optical channel power= 0dBm, 45 spans (3600 km)
OSNR= 12.2dB
Conventional receiver: BER = 10-0.7
Machine learning based receiver: BER = 10-4.3
0100
Time in ps
Optical power a.u.
Optical channel power= 0dBm, 120 spans (9600 km)
6
- A receiver setup optimized by machine learning increases the
maximum reach considerably
- Eye diagrames where no signal is visibel for a human being
anymore, can still be analyzed by machine learning
Conclusion
- Simulations with polarisation multiplexing are done and show good
results
- Next 4-QAM and 8 QAM modulated signals will be investigated
- Simulation with several WDM neighbour channels
Current work & outlook
ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.