Most examples currently set velocity errors (sigma) to a fixed value of 30 km/s. I think an array is allowed, but somewhere it should be possible to intelligently set the noise level as a function of position. Faint regions will have lower S/N that should scale something like sqrt(flux in fiber). Ultimately we'll want to scale this to the integration time.
Most examples currently set velocity errors (sigma) to a fixed value of 30 km/s. I think an array is allowed, but somewhere it should be possible to intelligently set the noise level as a function of position. Faint regions will have lower S/N that should scale something like sqrt(flux in fiber). Ultimately we'll want to scale this to the integration time.