READ ME File For 'Research data for the paper: Arbitrarily Parallel Turbo Decoding for Ultra-Reliable Low Latency Communication in 3GPP LTE'

Dataset DOI: XXXX

ReadMe Author: Luping Xiang, Southampton Wireless Group, Electronics and Computer Science, University of Southampton

This dataset supports the publication:
Luping Xiang, Matthew Brejza, Robert G. Maunder, Bashir M. Al-Hashimi and Lajos Hanzo;
Arbitrarily Parallel Turbo Decoding for Ultra-Reliable Low Latency Communication in 3GPP LTE.
IEEE Journal on Selected Areas in Communications

This dataset contains which are used for generating Fig.6 Fig.10 Fig.11 and Fig.12. These figures are plotted using GLE (Graphics Layout Engine). The scripts of Gle are also included in the folds for each figures. In order to generate these figures, you should install Gle
http://glx.sourceforge.net/

The figures are as follows:

Fig. 6:  (a) The relationship between the number of activated processors $P$ and the frame length $N$ for various turbo decoders; (b) The relationship between window length L and frame length $N$, for various turbo decoders. Note that in the \ac{SOTA} turbo decoder, FPTD and  the first version of the APTD algorithms, all windows have the same length, since the number of activated processors $P$ is chosen as an integer factor of $N$ in these schemes. By contrast, some windows have a length $\lceil \frac{N}{P}\rceil$ and others have the length $\lfloor \frac{N}{P} \rfloor$ in the second version of APTD  algorithm. In this case, the average window lengths is plotted.
Fig. 10: FER performance of the APTD algorithm for the $N=64, 512$ and $6144$-bit LTE turbo code, employing $P=64$ processors, $I=8$ iterations, Radix-4 operation, (a) interleaved (\emph{In}) or non-interleaved (\emph{NIn}) systematic, upper-lower (U-L) or odd-even (O-E) operation, where the extrinsic LLRs are calculated four at a time once the forward and backward recursions have crossed over (Ext on Both); (b) \emph{NIn}, U-L or O-E operation, where the extrinsic LLRs are either obtained two at a time alongside the forward recursions (Ext on F), or four at a time once the forward and backward recursions have crossed over (Ext on Both).
Fig. 11: (a) FER performance of the conventional and  \ac{SOTA} LTE turbo decoder and both versions of the proposed APTD for the $N=64, 512$ and $6144$ bits LTE turbo code that employs $T=16,256$ and $4096$ clock cycles, respectively. The proposed APTD employs $P=56$ processors, Radix-4 operation,  \emph{NIn} systematic approach,  O-E operation, and the extrinsic values are obtained from forward recursions only (Ext on F); (b) FER performance for second version of the APTD algorithm for $N=64,512,6144$, with $T=16,256$ and $4096$ clock cycles, respectively, activating $P=56$ processors when different delays are imposed. Here, \emph{NIn} systematic and O-E operation are employed and the extrinsic values are obtained from forward recursions only (Ext on F).
Fig. 12: (a) The number of clock cycles required for all frame lengths $N\in [40,6144]$ to achieve a FER of $10^{-5}$ at the same $E_b/N_0$ as the conventional turbo decoder using $I=8$ iterations in different turbo decoders; (b) The reciprocal of the number of clock cycles required for all frame lengths $N\in [40,6144]$ in different turbo decoders; (c) The overall complexity in different turbo decoders as a function of the frame length $N$; (d) The overall computational efficiency in different turbo decoders as a function of the frame length $N$.



Date of data collection: from December 1th, 2018 to December 10st, 2018

Information about geographic location of data collection: University of Southampton, U.K.

Related projects:
the Royal Society Wolfson Research Merit Award


Date that the file was created: Jan 2019