3-layer propagation-readout architecture
MATLAB R2018a / CUDA GPU
Run rwave/main.m -> wave data will be automatically saved in (time steps X ON/OFF RGC states) matrix format under rwave/export/yyyy-mm-dd,HH-MM
ON/OFF RGC positions are saved as well, for later reconstruction of waveshapes.
Simplified 3-layer readout architecture: ON retinal ganglion cells -> AII amacrine cells -| OFF retinal ganglion cells (Kerschensteiner, 2016).
Superposition of ON/OFF RGC hexagonal mosaic results in a periodic moiré interference pattern, which later seeds V1 orientation tuning (Paik, 2011).
An ON RGC receives input from other nearby ON RGCs of distance < 120μm.
An AII AC receives input from nearby ON RGCs of distance < 12μm.
An OFF RGC gets inhibitory input from nearby ACs of distance < 12μm.
The simulation starts in a state where a randomly selected fraction f>0.3 of the cells are assigned to be recruitable, and the remaining 1-f are assumed to be inactive for the duration (Butts, 1999).
A wave is initiated at t = 0 by a local stimulus and is allowed to propagate in ON RGC layer.
ON RGC: Recruitable -> Bursting -> Inactive
Upon receiving over-threshold input, an ON RGC fires for T = 1s. Then it becomes inactive during the rest of the event.
AC: Recruitable <-> Active
Upon receiving input, an AC suppresses nearby OFF RGC by giving negative input.
OFF RGC: Recruitable -> Hyperpolarized -> Bursting -> Inactive
When the inhibitory input ends (input returning to some <0 threshold), OFF RGCs fire for T = 1s. Then they become inactive for the rest of the event.
Se-Bum Paik & Dario L Ringach (2011) Retinal origin of orientation maps in visual cortex