|Lecturers:||Gerhard Schmidt and Thorben Kaak|
|Target group:||Students in electrical engineering and computer engineering|
|Prerequisites:||Skills in C programming language (for the DSS part), basic MATLAB knowledge (for the LNT part), diverse coding skills (for the ICT part)|
If you want to sign up for this laboratory, you need to register with the following information in the registration form
Please note, that the registration period starts the 11.03.2019 at 11:00 am and ends the 01.04.2019 at 23:59 am. All applications before and after this registration period, will not be taken into account.
Registration will be possible within the before mentioned time under the following subsite - Lab Registration.
Requirements, rules and commitments depend on the chosen topic and will be announced at the preliminary meeting (attendance is mandatory).
Attendance at all of the final presentations is mandatory as well to pass the lab.
|Contents:||See the detailed introduction of the topics below.|
ICT.1: MIMO LED-to-Camera Transmission (1 group of 3 students, Prof. Dr.-Ing. P.A. Hoeher)
LED-to-camera data transmission is a special case of Visible Light Communication (VLC). LEDs are used as transmitters and a camera as receiver. The LEDs can be modulated with different schemes like On-Off Keying (OOK) or in case of RGB LEDs Color Shift Keying (CSK) is also possible.
As cameras provides only low frame rates in the range of several Hz, the sampling rate of the LEDs is also quite low. To increase the data rates there are some strategies like exploiting the rolling shutter effect or to use multiple transmitter LEDs, which are decoded simultaneously by the camera.
In this lab it should be evaluated how an MIMO system with several LEDs could be established in theory and also in an experimental setup. For the latter part programming skills in C (Arduino) and Python are necessary.
ICT.2: Construction of a Magneto-OFDM System using STM32 Microcontrollers (1 group of 3 students and 1 group of 2 students, Prof. Dr.-Ing. P.A. Hoeher)
In the field of magnetic induction (MI) communications, the achievable bandwidths of single-coil systems are limited due to the quality factor of the coil. In order to increase the channel capacity, different resonant circuits employing different resonant frequencies corresponding to the sub-carriers of an OFDM-like system can be used.
During this lab, the participants shall construct an OFDM system employing multiple resonant circuits (e.g. 4). Further on, the provided micro controllers shall be used on receiver and transmitter side. An OFDM precoder and decoder shall be implemented. As a result, the data rate between a conventional MI communication system and the OFDM system approach employed shall be compared.
NT.1: Photonic Reservoir Computing (1 group of 2 students, Prof. Dr.-Ing. S. Pachnicke)
As an extension of the recurrent neural network (RNN) framework, reservoir computing (RC) is particularly well suited for processing chaotic time series such as a distorted information signal in optical communication systems. As with other artificial neural networks (ANNs), artificial neurons, called virtual nodes (VNs), are also used here. The difference, however, lies in their weighted interconnection, which is predefined and once initialized cannot be modified. This makes it possible to implement the reservoir with the VNs in hardware such as a photonic delay system. The input signal is mapped onto a higher dimensional space corresponding to the nonlinear activation function of conventional ANN approaches. Since the nonlinear functional principle is in the optical domain, the readout in the electrical domain remains linear. RC provide therefore a good balance between performance and feasibility for ANNs. There are, however, several issues that remains to be solve. E.g., current implementations map (masking) the input samples to multidimensional space digitally.
The aim of this project is to investigate and implement an analogue masking and reservoir scheme using Mach-Zehnder modulators for nonlinear channel equalization. Subsequently, a qualitative comparison between the analogue and digital masking is to be conducted.
NT.2: Simulation of an Optical Intra-Data Center Network (1 group of 2 students, Prof. Dr.-Ing. S. Pachnicke)
Modern intra-data center networks (DCNs) use electronic packet switching (EPS) as a main technique for data transmission among multiple source-destination servers hosted by the data center. On the other hand, optical fiber links are currently replacing the copper cables as the preferred transmission medium due to their much larger bandwidth, lower susceptibility to external factors, and compact packaging. As a consequence, current DCNs often find themselves in the state where data traffic is transmitted optically, but switched and processed electronically. This operational difference results in a constant optical-electrical-optical conversion at each switching point, which implies high operational costs and additional delays. One proposed solution to address this problem has been the migration of traffic switching into the optical domain, which on the downside is characterized by large switching times. Therefore, another possible solution that is currently being developed, is the application of a multi-ring network architecture, where the server racks share the same optical transmission medium using the advantages of wavelength division multiplexing (WDM) technology.
In this lab, the performance of a multi-ring DCN architecture is to be simulatively tested and analyzed, where the routing and wavelength assignment problem has to be addressed. Moreover, the limitations of the WDM technique (such as the maximal number of transmitted wavelengths) have to be also considered.
NT.3: Neural Network (1 group of 2 students, Prof. Dr.-Ing. S. Pachnicke)
Machine learning (ML) algorithms have been successfully applied to a variety of problems in the past. Especially the idea of deep learning based on artificial neural networks (ANN) achieved a breakthrough for statistical learning in language and image processing. The utilization of ML in optical communication technology can also be of great use for many application areas, such as nonlinear equalization, polarization recovery, carrier recovery, etc.
The main purpose of this lab is to develop a neural network for data classification applied in the compensation of nonlinear distortions, such as nonlinear phase noise. Without any prior knowledge, the NN should be able to learn and capture the link properties from only a short training data set. The NN should be able to calculate the non-linear decision thresholds and demodulate the received data.
DSS.1-3: Real-Time Audio Processing (3 groups of 3 students, Prof. Dr.-Ing. G. Schmidt)
In this project, students are going to implement a speech enhancement system in the Kiel Real-Time Audio Toolkit (KiRAT). Algorithms within this framework are to be programmed in C language, the graphical user interface is written in C++ using the QT framework. Thus, it is expected that the participants have programming skills in C/C++. There will be up to three groups of three students that will create their own speech enhancement systems. Each group will specialize on one of the following algorithmic components:
- Analysis and synthesis filterbanks,
- Noise estimation, and
- Noise reduction.
Schedule of talks
Attendance during all presentations as well as active paticipation in the discussions is mandatory to pass the lab.
The schedule can be found below:
|xx.xx 2019||Group||Topic||Lecturer(s)||Talk duration|