Spectral partitioning for modular software defined radio

Guidelines for designing and operating modular software defined radio (Mod-SDR) devices gain in importance as the interworking of different radio technologies emerges as a potential solution to achieving short time-to-market and to offering flexible services to mobile users. We continue to study the partitioning and scheduling of software modules in multiprocessor devices. In particular, we focus on spectral partitioning, which is known to yield good partitions for general-purpose parallel computing applications. We discuss a spectral partitioning algorithm in the context of Mod-SDR, comparing it to implicit partitioning results produced by a variant of Hu's algorithm under realistic operating conditions. Test settings include two identical processors connected through either one or two data buses, and pipelining as a signal processing option for delay-insensitive transmission modes. Furthermore, we show how close both approaches come to the theoretical upper bound on two-processor speedup. Based on computer simulations, we can conclude on an enhanced set of guidelines for operating a modular software defined radio.