We propose a n e w algorithm for T u r b o code interleaver design, which is based on the conventional s-random approach and whose complexity grows only linearly w i t h the interleaver length.Designing the interleaver x = ( T I ; ..; T K ) of length K of a n r b o code serves to increase the code's minimum distance &,,in and hence to lower the error floor of the Word and Bit Error Rates (WER/BER). An efficient method was presented in [l]. Examinations show that for so-designed interleavers, the codeword at bmin is mainly caused by a combination of an input word u(l) of the first component encoder (identical to the Turbo encoder input U) and a second component input word U(') as shown in Fig.

The class of interleavers is identi ed, which possesses the following property: both Turbo Code component scramblers are terminated in the same state, independent of the encoder information word. It is proven that the presented class contains all interleavers with the above property. Introduction: Turbo Codes 4] represent the most power-e cient binary channel codes of the present time. Their encoder consists of three parallel branches: one systematic branch and two (recursive) scramblers (in the following denoted by scr1 and scr2, respectively), both of which are joined by an interleaver. In the article we always assume identical scramblers of memory. Scrambler properties: Throughout the article, we use the following terminology. Variables with a superscript index (1) and (2) are associated with scr1 and scr2, respectively. The input of length K to the scramblers is denoted by the vector u (j) = (u (j) 0 ; u (j) 1 ; ::; u (j) K?1); j 2 f1; 2g with u (j) i 2 GF(2) = f0; 1g. The interleaver function u (2) = (u (1)) performs transpositions i ! i of the input bit positions i = 0::K ? 1 of scr1 to those of scr2, i.e. u (2) i = u (1)

Zusammenfassung Hintergrund Analysiert wurden die Patientencharakteristika und Krankheitsverläufe aller Patienten, die mit COVID-19 in der 1. und 2. Welle im HELIOS-Klinikum Krefeld behandelt wurden. Methoden Eingeschlossen wurden 84 Patienten aus der 1. Welle (11.03.2020–30.06.2020) und 344 Patienten aus der 2. Welle (01.07.2020–31.01.2021). Ergebnisse Alter, Geschlecht und Komorbiditäten der Patienten waren ähnlich, mit Ausnahme der venösen Thrombose in der Anamnese. Diese lagen in der 1. Welle häufiger vor als in der 2. Welle (6 % vs. 0,3 %, p = 0,001). Bei der Aufnahme gab es keine Unterschiede in den Ergebnissen der initialen Laborwerte (C-reaktives Protein, Leukozyten) und Blutgasanalysen zwischen beiden Gruppen. Die Behandlung unterschied sich in der Anwendung von Dexamethason und Antikoagulation. In der 1. Welle erhielt niemand Dexamethason, in der 2. Welle jedoch 52,6 % der Patienten für eine durchschnittliche Dauer von 3,6 ± 4,1 Tagen. Eine Antikoagulation mit doppelter Standardprophylaxe (2 × 40 mg niedermolekulares Heparin, subkutan) wurde in der 1. Welle bei 7,1 % und in der 2. Welle bei 30,2 % der Patienten (p = 0,002) durchgeführt. In der 1. Welle wurden mehr thromboembolische Ereignisse nach der Aufnahme diagnostiziert (19,0 % gegenüber 7,0 %, p = 0,001). Die Sterblichkeitsrate im Krankenhaus lag in der 1. Welle bei 26,2 % und in der zweiten Welle bei 15,4 % (p = 0,0234). Die meisten Todesfälle waren auf das akute Atemnotsyndrom (ARDS) zurückzuführen. Schlussfolgerung Die Patientencharakteristika unterschieden sich in der 1. und 2. COVID-19-Welle nicht, aber Antikoagulation und Dexamethason wurden in der 2. Welle häufiger eingesetzt. Darüber hinaus traten in der 2. Welle weniger thromboembolische Komplikationen auf.

Like [1], we present an algorithm to compute the simulation of a query pattern in a graph of labeled nodes and unlabeled edges. However, our algorithm works on a compressed graph grammar, instead of on the original graph. The speed-up of our algorithm compared to the algorithm in [1] grows with the size of the graph and with the compression strength.

Performance prediction for hardware-software configurations is a relevant and practically important problem. With an increasing availability of data in the form of performance measurements, this problem becomes amenable to machine learning, i.e., the data-driven construction of predictive models. In this paper, we propose a learning method that is specifically tailored to the task of performance prediction and takes two important characteristics of this problem into account: (i) prior knowledge in the form of monotonicity constraints, suggesting that certain properties of hard-or software can influence performance only positively or negatively, and (ii) strong differences in the precision and reliability of performance measurements available as training data. We evaluate our method on a real-world dataset from the domain of performance prediction in video games, which we specifically collected for this purpose.