Worst-case latency analysis of SDF-based parametrized dataflow MoCs

Modern-day streaming digital signal processing (DSP) applications are often accompanied by real-time requirements. In addition, they expose increasing levels of dynamic behavior. Dynamic dataflow models of computation (MoCs) have been introduced to model and analyze such applications. Parametrized dataflow MoCs are an important subclass of dynamic dataflow MoCs because they integrate dynamic parameters and run-time adaptation of parameters in a structured way. However, these MoCs have been primarily analyzed for functional behavior and correctness while the analysis of their temporal behavior has received little attention. In this work, we present a new analysis approach that allows analysis of worst-case latency for dynamic streaming DSP applications that can be captured using parametrized dataflow MoCs based on synchronous dataflow (SDF). We show that in the presence of parameter inter-dependencies our technique can yield tighter worst-case latency estimates than the existing techniques that operate on SDF structures that abstract the worst-case behaviour of the initial parametrized specifications. We base the approach on the (max,+) algebraic semantics of timed SDF and on its non-parametric generalization known as FSM-based scenario-aware dataflow (FSM-SADF). We evaluate the approach on a realistic case study from the multimedia domain.