Chair of Communications


Offered Theses

Additional topics for Master Theses are available. Please contact Prof. Dr.-Ing. Stephan Pachnicke or the assistants directly.

  1. Compensation of Transmission Impairments in Optical Transmission Systems with Direct Detection

    Internet service providers and content providers have to extend their networks within and between data centers due to increasing data traffic generated by cloud based services or multimedia content delivery. The focus of research lies on optical transmission systems with a reach of up to 100 km, and date rates of more than 400 Gbit/s. The large number of optical links implicates a high cost pressure, so that low-complexity transmission systems with intensity modulation at the transmitter side and direct detection (IM/DD) in the receiver play a significant role. Nevertheless, the bandwidth limitations of the electrical components and the square-law characteristic of the photodiode can distort the transmission (power-fading, signal-signal beating interference). These impairments can be avoided by proper system design or can be compensated in the digital signal processing.

     Keywords: Direct Detection, Signal-signal beating interference, Modelling, Data Center
     Supervisor: Simon Ohlendorf, Room C-014, Phone: 880-6311,


  2. GPU assisted computation in the nonlinear frequency domain

    The Nonlinear Fourier Transform (NFT) has the potential to overcome the nonlinear barrier which prevents the increase of spectral efficiency by increasing the SNR. Recent works show a significant progress in terms of modulation formats used in NFT-based transmission. However, because of the numerical complexity the computational time to generate (Inverse NFT) and demodulate (NFT) signals is still not practical for real world usage. It is therefore imperative to find ways to either decrease the complexity or increasing the computational efficiency. In order to reduce the computational time one often looks at the parallel processing capabilities of the GPU inside graphic cards. It has been shown in the past that the calculation of the Split-Step Fourier method for the wave propagation inside the optical fiber can be accelerated by using the GPU considerably.
    During this master thesis, the existing NFT algorithm in Matlab is to be converted into the CUDA programming code for NVidia graphic cards. The efficiency gain shall be proven in a real lab setup afterwards.

     Keywords: Nonlinear Fourier Transform, GPU Processing, CUDA Programming
     Supervisor: Shi Li, Room C-031, Phone: 880-6304,


  3. Information Processing using semiconductor LASER

    The requirement for processing time dependent data (speech recognition, -processing, weather or general dynamic systems) demand novel and efficient methods for the increasing amount of data being generated in different areas of science and technology. One possibility is the bio-inspired neural network approach. However, the machine learning of such artificial neural networks for time dependent dynamic data processing is very complex. For such “real-world” tasks, the feed-forward architecture has to be adapted by using recurrent connections (Recurrent Neural Networks [RNN]). A further improvement are the nonlinear nodes in a reservoir. It turns out that the nonlinear transient dynamic of a semiconductor LASER with a delayed optical feedback shows promising results for such concept (Reservoir Computing [RC]). During this master thesis, this concept is to be modeled and if possible experimentally proven.

     Keywords: Information Processing, Semiconductor LASER, machine learning, Reservoir Computing, Modeling
     Supervisor: Shi Li, Room C-031, Phone: 880-6304,


  4. Compensation of fiber nonlinearities in the nonlinear Fourier spectrum

    The Nonlinear Fourier Transform (NFT) has the potential to remove the nonlinear limitations for higher capacity with increasing signal to noise ratio (SNR) in the optical fiber transmission system. To that end, new system components are needed for the actual implementation. However, even existing and currently deployed optical transmission systems can profit from the NFT. The evolution in the nonlinear spectrum for any distances is just a linear phase shift, which could be used as an alternative to the digital back-propagation (DBP) scheme in the digital signal processing (DSP) unit. During this master thesis, the feasibility of such application are evaluated in comparison to exiting DBP schemes. A model is then to be created for later experimental testing and usage.

     Keywords: Nonlinear Fourier Transform (NFT), Digital Back-Propagation (DBP), Modeling
     Supervisor: Shi Li, Room C-031, Phone: 880-6304,


  5. Progress in Digital Back-Propagation

    Equalization of nonlinear impairments in fiber-optic transmission systems is becoming crucial for future system design. The Digital back-propagation (DBP) method can jointly compensate for linear and nonlinear fiber impairments by digitally creating a "mirror image" of the transmission link. This can be achieved with the implementation of an inverse fiber by solving numerically the nonlinear Schrödinger equation (NLSE), as similarly done for simulation purposes. Along its many advantages, such as high compensation potential and transparency, DBP suffers from a very high computational complexity, which is prohibited in high-speed transmission systems. Since 2008, numerous research groups have investigated DBP, in order to overcome its disadvantages.

     Keywords: Nonlinear Compensation, Digital Back-Propagation, Coherent Detection
     Supervisor: Dennis Clausen, Room C-029C, Phone: 880-6305,


  6. Analysis of the Energy Efficiency of Hybrid Optical-Electrical Data Center Connects

    In next generation data centers thousands of servers need to be interconnected with very high bit rates. Current architectures mainly rely on so called “fat-tree” approaches using a multi-layer electrical switching matrix. For further increasing data rates such architectures become more and more inefficient and consume a tremendous amount of electrical energy. A potential solution is to use additional optical connects between blades or racks purely in the optical layer without any electrical components in between. The optical switching of data connects will allow a significant reduction of the energy consumption because no optical-electrical conversion is required along the path. As an example an electrical 10 Gb/s Ethernet connection requires approx. 10 W of power whereas a pure photonic solution can be realized with only a fraction of the energy consumption.

     Keywords: Modelling, Optimization, Data Center
     Supervisor: Mihail Balanici, Raum C-014, Telefon: 880-6311,