For a single raw WIRE frame the dominant source of noise in a pixel is Poisson noise or "counting statistics" noise. This says that the when counting N indivisible objects (like photons or electrons) the variance is equal to N, which means the signal-to-noise goes as sqrt(N).
In a WIRE frame, each digital number (DN) counts a certain number of electrons (assume 1 electron = 1 detected photon) given by the inverse gain (IG) in units of electrons (e-) per DN. The nominal WIRE values are 85 e-/DN (12 micron) and 42.5 e-/DN (25 micron).
A second noise source is the readout noise which contributes a given amount of gaussian for each pixel for each subframe (i.e. each readout). A calculation of the noise in a pixel in a subframe is given here . For a whole frame made up of N subframes the formula is:
(pixel noise in DN) = sqrt[ ( DN level / IG ) + N*(Re/IG)^2 ]Where "Re" is the readout noise in electrons, which is nominally 60 electrons for both bands.
A third source of noise is the "confusion noise" which is due to a sea of faint unresolved, undistinguishable sources which cover the WIRE fields. We assume that this is negligible for a single WIRE frame. (Note however that since the sources are fixed on the sky the constructively interfere between exposures, unlike the Poisson and readout noise, so that confusion noise will become dominant when many (about 100 or more) frames are added together.)
Terminated WIE Box Noise Pattern: It has been noticed that there is a faint pattern which appears on "blank" or "terminated" images. See here for details.
Effective Readout Noise Using Linearity Test Data: One can estimate an effective readout noise using linearity test images taken with the actual detector. See here for details. (Note that a similar estimate should be made during the mission.)
This is a rather complex issue. See Tom's page for discussion on noise and effective gain: Tom's Page .
See Joe's page for other discussions: Joe's Page .