Linear Programming in Embedded Systems

In the past signal processing algorithms for mobile communications systems were implemented in hardware using a single chip solution, usually a Digital Signal Processor (DSP). But due to the increased and still increasing complexity of these signal processing algorithms, the computational power of single chip solutions is not sufficient any more. In addition the number of algorithms to be processed simultaneously is rising. For example owners of mobile phones nowadays want to use WLAN, WiMAX and DVB-H at the same time. To tackle this increasing computational demand, there are two competing approaches in chip design. On the one hand dedicated hardware design like application specific integrated circuits (ASICs), on the other hand so-called systems-on-chip (SoC) design. SoCs can consist of several general purpose processors as well as DSPs and ASICs. In general terms ASIC-based solutions are smaller and more energy efficient [1] than SoC-based solutions, but not as flexible, because once the chip is developed, the possibility to modify the behavior of the system is very limited. Another advantage of SoCs is, that components can be reused from existing processor designs. Thus, from now on we will concentrate on SoCs. There are several contradicting objectives for designing SoCs. These objectives can be area (die size), performance, energy consumption as well as rather soft defined goals like programmability or flexibility. Since chip design and verification is very time consuming, the designer wants to know at a very early design stage, how the system should look like, how the application behaves on the determined architecture and what the performance figures would be. An idea presented in [2] is to determine a first system at a very high abstraction level neglecting some details and in this way reducing the set of remaining candidate systems. With the gained solution the designer can proceed with the next lower abstraction level and refine the system step by step. Figure 1 illustrates this method.