Spike generation

After summing the generator potential from the different temporal models, we apply firing rate gain (spikes/(second * unit contrast)), parameter A in the Victor 1987 model, values from Benardete and Kaplan 1999 for parasol units and Benardete and Kaplan 1997 for midget units. This gives us a distribution of gains. Next, this value is calibrated across the different temporal and spatial model combinations to give a threshold firing rate for 3.5% contrast (parasol units) or 11.4% contrast (midget units). We run separately luminance contrast calibration (following Lee et al. 1989) and drifting grating calibration (following Derrington and Lennie 1984).

Next, the signal goes through a threshold linear (rectified linear, ReLu) nonlinearity.

Finally, the spike generation comes from either a poisson process or from a refractory model. For refractory model the recovery function is from Berry and Meister 1998 and absolute and relative refractory parameters were estimated from Uzzell and Chichilnisky 2004, Fig 7B, bottom row, inset. Currently we have only one set of fixed refracory model parameters, shared across all GC units.

Noise model

Spontaneous firing rates in midgets and parasol units range from 5 to over 30 Hz (Appleby and Manookin 2019, Sinha et al. 2017); we fixed the cone noise to induce ~25Hz baseline firing rates for the ON parasol and midget units, and ~5Hz for the OFF parasol and midget units (Raunak Sinha, personal communication).

The noise type shared gets cone spectrum from Angueyra and Rieke 2013, and this drives the ganglion cell background firing rates. You can use the same cone noise for all ganglion cell types, hypothetically resulting in a shared downstream signal for a shared stimulus. The noise type independent fits Gaussian distribution to the GC noise firing rates from the shared noise, and then it samples indipendent noise for each GC unit from this distribution. Thus,with independent noise you lose both autocorrelation and correlation between units.