Multiple discretization steps (dt's) for multiple domains?

[Potomac Firefly]

I wonder whether ANNarchy can be implemented to support multiple discretization steps for a network with multiple domains. For example, a network might be divided into two domains which are interconnected with synapses. For example, domain 1 may have dt=0.1ms, while domain 2 may have dt=0.01ms. Of course all dt's will be multiples of the shortest dt. Could this feature be implemented? Thanks!

[Julien Vitay]

That would be interesting. We have seen this feature in Brian, but could not really think of a realistic use-case, so we initially went for a single timestep for the whole network.

But it indeed makes sense for hybrid networks, where a rate-coded model (typical dt=1ms) interacts with a spiking one (typical dt=0.1ms). So it is something we should add soon (perhaps in 5.0)

We have to think about how hard it would be to add this feature. If it is just about calling the update method of some neural equations every 10 steps instead of 1, it should not be hard (there is already something similar for synaptic plasticity, where weight updates can be called less often than the regular clock), but we have to check how spike transmission works: the faster network should integrate spikes from the slower one at the right moment and only once. 

[Potomac Firefly]

Thank you for considering this feature. I am planning a simulation study with which this feature will be useful. Birdsong articulation may be implemented as a dynamic system described by differential equations. To support 10KHz (maybe as high as 20KHz) max frequency, the discretization steps must be much shorter than 0.1ms. A neuron network will drive this dynamic system. Thus multiple domains with different discretization steps will be helpful.