Peter Oppermann received a second award of the Werner Petersen Technology Awards (Prof. Dr. Werner Petersen-Preise der Technik)  for his master thesis Implementation of Information Optimum Decoders for Low-Density Parity-Check Codes on a Digital Signal Processor which he had done at the Institute of Communications (INT) under the supervision of Jan Lewandowsky in cooperation with the Fraunhofer Institute for Communication, Information Processing and Ergonomics (FKIE), Wachtberg. The award ceremony with participation of the Petersen foundation board members Dr. Christian W. Zöllner and Dr. Klaus-Jürgen Wichmann took place on January 19, 2018 at the Christian-Albrechts-Universität in Kiel.

(Foto: fabianfrühling.com)

The thesis considers an innovative and unconventional approach to information-optimum digital signal processing for the example of decoding an error correcting LDPC code.

State-of-the-art discrete signal processing typically is based on an optimum or close-to-optimum algorithm which is designed for unquantized signals, at least for high resolution such as double precision. The optimum algorithm is then made implementable by approximations and reduced resolution, i.e. a coarser quantization with a limited number of bits per sample. The approximations are usually “local” approximations: In each stage of the signal processing chain, the current sample values and operations are approximated as good as possible. Optimality of an algorithm is usually lost in such a quantized implementation which often results in a significant performance degradation compared to a high precision implementation.

In contrast, Peter Oppermann pursues the innovative concept of “relevant information”. Quantized signal representation and operations are designed such that as much information about the actual variable of interest in the sense of Shannon’s information theory is preserved. This is a fundamentally different approach. The so-called information bottleneck method is used which was previously applied in the field of machine learning but is hardly known in signal processing. 

The contribution of the thesis is that it not only designs and analyzes such an information-optimum decoder but also develops several variants of a highly efficient implementation on a digital signal processor (DSP) bringing the theoretical idea into practice. In comparison to state-of-the-art decoders, the throughput can be improved by about 30 % while – a remarkable result – almost optimum performance is obtained.

(Foto: privat)

The Petersen Technology Awards honor outstanding bachelor and master theses in technology and natural science from northern German universities. They are the highest remunerated German awards of that kind. Special attention is paid to high industrial, medical, economical or social significance.  The founder Prof. Dr. Werner Petersen (1910-2003) was a German physicist, professor and publisher.

Link Petersen Foundation

http://www.petersen-stiftung.de

Press release

http://www.uni-kiel.de/pressemeldungen/index.php?pmid=2018-010-preisverleihung-informatik&pr=1