poisson_generator - imulate neuron firing with Poisson processes
statistics.
The poisson_generator simulates a neuron that is firing with Poisson
statistics, i.e. exponentially distributed interspike intervals. It will
generate a _unique_ spike train for each of it's targets. If you do not want
this behavior and need the same spike train for all targets, you have to use a
parrot neuron inbetween the poisson generator and the targets.
The following parameters appear in the element's status dictionary:
\verbatim embed:rst
======= ======== =========================================================
rate spikes/s Mean firing rate
origin ms Time origin for device timer
start ms Begin of device application with resp. to origin
stop ms End of device application with resp. to origin
======= ======== =========================================================
\endverbatim
SpikeEvent
A Poisson generator may, especially at high rates, emit more than one
spike during a single time step. If this happens, the generator does
not actually send out n spikes. Instead, it emits a single spike with
n-fold synaptic weight for the sake of efficiency.
The design decision to implement the Poisson generator as a device
which sends spikes to all connected nodes on every time step and then
discards the spikes that should not have happened generating random
numbers at the recipient side via an event hook is twofold.
On one hand, it leads to the saturation of the messaging network with
an enormous amount of spikes, most of which will never get delivered
and should not have been generated in the first place.
On the other hand, a proper implementation of the Poisson generator
needs to provide two basic features: (a) generated spike trains
should be IID processes w.r.t. target neurons to which the generator
is connected and (b) as long as virtual_num_proc is constant, each
neuron should receive an identical Poisson spike train in order to
guarantee reproducibility of the simulations across varying machine
numbers.
Therefore, first, as Network::get_network().send sends spikes to all the
recipients, differentiation has to happen in the hook, second, the
hook can use the RNG from the thread where the recipient neuron sits,
which explains the current design of the generator. For details,
refer to:
http://ken.brainworks.uni-freiburg.de/cgi-bin/mailman/private/nest_developer/2011-January/002977.html
/var/www/debian/nest/nest-simulator-2.20.0/models/poisson_generator.h