Institute of Photonics and Quantum Electronics
KIT – The Research University in the Helmholtz Association
Bachelor / Master Thesis:
Neural Network Pre-Distortion
The performance of high-speed optical transmission links is strongly impaired
by a variety of impairments induced by the transmitter components (digital-to-
analog converter, electrical amplifier, electro-optic modulator). Linear digital
pre-distortion is an established concept to reduce the impact of transmitter
bandwidth limitations. For modulation formats of high order, e.g., 64QAM,
however, nonlinear effects become limiting. Neural networks (NNs) are a
promising approach to adress nonlinear distortions, especially nonlinear
intersymbol interference (ISI). Due to their generalization capability, they
theoretically can model and invert any kind of nonlinearity.
In this thesis, novel concepts for a NN based pre-distorter shall be
investigated. For later use in commercial transceivers, an efficient real-time
implementation in an CMOS ASIC is key. Therefore, a detailed comparison of
the NN approach with alternative nonlinear compensation techniques
regarding performance and implementation complexity is one main focus of
this work. Look-up tables, Volterra equalizers, Wiener-Hammerstein systems,
decision-feedback or MLSE equalizers are examples for alternative concepts.
Fig. 1: Principle structure of a neural network
with delay elements (z-1), weights (colored),
sums, and nonlinear functions (𝜎).
For detailed information contact:
M. Sc. Christoph Füllner
Tel. 0721-608-47173
Prof. Dr. Sebastian Randel
Tel. 0721-608-42490
Your tasks:
Research on existing ideas and developing of
new pre-distortion NN-based concepts
Implementation of the NN and some alternative
methods in Matlab or Python
Evaluation of the perfomance in simulations
and on experimental data
Analysis of the implementation complexity
z-1
z-1
z-1
z-1
x(n)
.
..
1
1
1
1
1
2
2
2
2
y(n)
x(n-1)
x(n-2)
x(n-3)
x(n-4)
Fig. 2: Constellation diagram for 64QAM. The
right-hand one clearly is nonlinearly distorted.