Mapping the Information Trace in Local Field Potentials by a Computational Method of Two-Dimensional Time-Shifting Synchronization Likelihood Based on Graphic Processing Unit Acceleration

SL processing routine with GPU acceleration. A Flowchart of SL computing procedures. B Raw LFP traces from channel A (blue) and channel B (red). The SL calculating windows (shaded) are shifted by start time and delay from the beginnings of data segments. C The SL calculation process of a single computing step. Upper two rows: normalized LFP traces of selected time window; lower two rows: normalized Euclidean distance between each state vector and its reference vector in channel A (blue) and channel B (red). Shaded areas show the recurrent events in both channels. D Demonstration of SL calculating steps (Vec, vector; ST, start time). The current page is marked under the red tab to show a batch of parallel processing tasks simultaneously carried out by the GPU for the red data trace in C as time-shifting of 25 resampled data points at a delay. Blue shaded area in current page shows the parallel processing task carried out by the GPU for the blue data trace in C with 50 resampled data points from the initial calculating start point. E Task assignment for the GPU acceleration procedure. Each block is assigned to execute parallel processing tasks of a same start time. Every thread within a block is assigned to calculate the Euclidean distance between a state vector and its reference vector. For the same value in delay, all threads work simultaneously to cover all values in start time and therefore to realize GPU acceleration. F Synchronization matrix of LFP data of channels A and B. Grey row at a delay of 25 (black arrow) indicates the batch of parallel processing tasks shown in D. Red arrow, processing task shown within the blue shaded area in D