Towards real-time neuronal connectivity assessment: A scalable pipelined parallel generalized partial directed coherence engine

Abstract

The current paper introduces a real-time architecture for the computation of the Generalized Partial Directed Coherence (GPDC) of multiple signals. The motivating application is the localization and control of epileptic seizures where hitherto published results shown the effectiveness of exploiting Generalized Partial Directed Coherence to quantify and analyse connectivity and interaction of brain structures. To speed up GPDC computations we develop first, a parallelizing strategy leading to the high performance scalable architecture and second, a low-complexity fixed-point reciprocal square root module. We show that a real-time computation is feasible at a speed of 0.027ms for 16 channels and 1.637ms for 128 channels. Furthermore, the implementation results on Xilinx 7A35T, KC705, VC707, KU115 show that the power requirements are quite modest and allow for the embedded application of the engine