Vortrag von Dr. Alexios Balatsoukas-Stimming
Datum: 17.04.19 Zeit: 15.00 - 16.00 Raum: UNINE, B217
During the so-called happy scaling era of integrated circuits, the increasing transistor count per unit area and the increasing energy-efficiency were able to offset the ever-increasing complexity of communications and digital signal processing algorithms. Unfortunately, the gains from circuit technology scaling have slowed down significantly in the last few years. One of the promising techniques that have been proposed to overcome this hurdle is approximate computing, where the reliability constraints of integrated circuits are relaxed significantly. In this presentation, we will first explain how well-known methods, such as density evolution, can be adapted in order to analyze LDPC and polar decoders that operate with approximate memory. For polar codes in particular, we will also describe efficient techniques that can be used to mitigate the effect of memory faults. Finally, we will explain why most approximate channel decoders that have been proposed in the literature so far (including our own) are impractical, and we will present some low-complexity hardware methods that can be used to partially overcome this obstacle.