• Diese Seite als PDF öffnen.

Protograph-based LDPC Code Design for Finite Number of Iterations

Description

Low-density parity-check (LDPC) codes are usually designed using an extrinsic information transfer (EXIT) analysis to compute the asymptotic iterative decoding threshold for a given code ensemble. The iterative decoding threshold over an additive white Gaussian noise (AWGN) channel is thereby defined as the minimum signal-to-noise ratio (SNR) for which the a posteriori mutual information converges to one as the number of decoding iterations goes to infinity.

For practical applications, however, decoding is often performed using only a small number of iterations. Mulholland et al. investigated a code design for protograph-based LDPC codes where the protograph-based EXIT (PEXIT) analysis [1] is run for the targeted number of decoding iterations only [2]. They showed that for the binary erasure channel (BEC), the obtained iteration-constrained threshold does not match the bit error rate (BER) threshold of finite-length codes.

In this thesis, the student will understand and reproduce the results from [2], apply these tools to the AWGN channel, and compare them to code design tools developed at the institute.

[1] G. Liva and M. Chiani, “Protograph LDPC codes design based on EXIT analysis,” in Proc. IEEE Global Telecommun. Conf. (GLOBECOM), Nov. 2007, pp. 3250–3254.

[2] I. P. Mulholland, E. Paolini, and M. F. Flanagan, “Design of protograph-based LDPC code ensembles with fast convergence properties,” in Proc. Int. ITG Conf. Syst. Commun. Coding (SCC), Feb. 2017, pp. 1–6.

Prerequisites

Supervisor:

• Diese Seite als PDF öffnen.

Efficient Syndrome Distribution Matching

Description

Probabilistic parity shaping (PPS) has recently been proposed for energy-efficient transmission where the capacity-achieving distribution requires shaped parity bits, e.g., when applying intensity modulation or on-off keying (OOK). A key component of this transmission scheme is the syndrome distribution matcher (SDM), which maps the syndrome of the systematic part of the code to a valid parity sequence under a cost constraint. Since this problem is equivalent to the syndrome decoding problem, syndrome distribution matching can be performed efficiently for code classes having an efficient syndrome decoding algorithm.

In this thesis, the student will review syndrome decoding algorithms for different code classes, identify the most promising code classes and decoding algorithms for syndrome distribution matching, implement these algorithms, and analyze their performance.

[1] G. Böcherer, D. Lentner, A. Cirino, F. Steiner, “Probabilistic Parity Shaping for Linear Codes," Feb. 2019. Available: https://arxiv.org/abs/1902.10648

Prerequisites

Supervisor:

• Diese Seite als PDF öffnen.

Code construction for quasi-cyclic LDPC codes based on PEG and ACE

Description

Finite length low-density parity-check codes can be constructed based on different approaches. The most obvious approach is a random, unconstrained construction, where the edges in the parity-check matrix are placed based on the desired edge distribution pairs.

In general, this approach leads to suboptimal performance, as the codes have  a lot of small cycles, which deteriorate the belief propagation decoding. To mitigate this, greedy approaches like progressive edge (PEG) have been proposed. PEG can further be complemented by taking the extrinsic cycle degree into accout [2], which also provides an additional measure for the "harmfullness" of a cycle in the Tanner graph.

In this research internship, the student is asked to implement a tool for building finite length QC LDPC codes which uses both PEG and ACE for girth optimization and conditioning.

[1] X.-Y. Hu, E. Eleftheriou, and D. M. Arnold, “Regular and irregular progressive edge-growth Tanner graphs,” IEEE Trans. Inf. Theory, vol. 51, no. 1, pp. 386–398, Jan. 2005.

[2] T. Tian, C. R. Jones, J. D. Villasenor, and R. D. Wesel, “Selective avoidance of cycles in irregular LDPC code construction,” IEEE Trans. Commun., vol. 52, no. 8, pp. 1242–1247, 2004.

Prerequisites

Supervisor:

• Diese Seite als PDF öffnen.

Vorverzerrung des Basisbandsignals mit Hilfe eines Neuronales Netzwerkes

Description

In terrestrischen Sendern werden zur Verstärkung des Signals Verstärker der Klasse AB eingesetzt. Diese Verstärker werden im hocheffizienten Arbeitsbereich betrieben, so dass neben den AM-AM- und AM-PM-Verzerrungen mit Memoryeffekten zu rechnen ist. Diese müssen entzerrt werden. Zusätzlich ist es mit Sättigungseffekten zurechnen. Die Entzerrung erreicht man durch das Vorverzerren des Signals im Basisband. Zur Vorverzerrung des Signals werden Memory Polynom Algorithmen eingesetzt. Die bekannten Memory Polynom Algorithmen sind sehr rechenintensiv und können nicht zur Echtzeitentzerrung mit zur Verfügung stehenden Ressourcen in der SW umgesetzt werden. Ziel der Masterarbeit ist ein Neuronales Netzwerk aufzubauen, das ressourcenschonend in der SW mit Echtzeitanforderung umgesetzt werden kann.

Aufgaben:

•  Recherche der bestehenden Deep und Machine Learning Mechanismen zur Vorverzerrung des Basisbandsignals.
•  Beurteilung dieser Mechanismen.
•  Design und Performanz Vergleich mehrerer NN-Modelle.
•  Optimierung der Modelle.
•  Dokumentation der Ergebnisse.

Ressourcen: unterschiedliche Verstärker-Typen/-Leistungsklassen; Messaufbau mit Signalgenerator, Verstärker, Signalanalysator.

Entwicklungsumgebung: Matlab, Python, Tensorflow.

Supervisor:

Student