Volker Turau

Abstract: In this paper we investigate the usage of machine learning for interpreting measured sensor values in sensor modules. In particular we analyze the potential of artificial neural networks (ANNs) on low-cost microcontrollers with a few kilobytes of memory to semantically enrich data captured by sensors. The focus is on classifying temporal data series with a high level of reliability. Design and implementation of ANNs are analyzed considering Feed Forward Neural Networks (FFNNs) and Recurrent Neural Networks (RNNs). We validate the developed ANNs in a case study of optical hand gesture recognition on an 8-bit microcontroller. The best reliability was found for an FFNN with two layers and 1493 parameters requiring an execution time of 36 ms. We propose a workflow to develop ANNs for embedded devices.

Abstract: In 2015, the IEEE 802.15.4 standard was expanded by theDeterministic and Synchronous Multi-Channel Extension (DSME) toincrease reliability, scalability and energy-efficiency in industrial appli-cations. The extension offers a TDMA/FDMA-based channel access,where time is divided into two alternating phases, a contention accessperiod (CAP) and a contention free period (CFP). During the CAP, transmission slots can be allocated offering an exclusive access to theshared medium during the CFP. The fractionτof CFP’s time slots ina dataframe is a critical value, because it directly influences agility andthroughput. A high throughput demands that the CFP is much longerthan the CAP, i.e., a high value ofτ, because application data is only sentduring the CFP. High agility is given if the expected waiting time to senda CAP message is short and that the length of the CAPs are long enoughto accommodate necessary GTS negotiations, i.e., a low value ofτ. OnceDSME is configured according to the needs of an application,τcan onlyassume one of two values and cannot be changed at run-time. In thispaper, we propose two extensions of DSME that allow to adoptτto thecurrent traffic pattern. We show theoretically and through simulationsthat the proposed extensions provide a high degree of responsiveness totraffic fluctuations while keeping the throughput high.

Abstract: It is known for some time that a random graph G(n,p) contains w.h.p. a Hamiltonian cycle if p is larger than the critical value pcrit = (logn + log log n+ωn)/n. The determination of a concrete Hamiltonian cycle is even for values much larger than pcrit a nontrivial task. In this paper we consider random graphs G(n,p) with p in ̃Ω(1/√n), where ̃Ω hides poly-logarithmic factors in n. For this range of p we present a distributed algorithm AHC that finds w.h.p. a Hamiltonian cycle in O(logn) rounds. The algorithm works in the synchronous model and usesmessages of size O(logn) and O(logn) memory per node.