Photon Counting
Photon Counting in EMCCDs
Photon Counting in EMCCDs is a way to overcome the multiplicative noise associated
with the amplification process, thereby increasing the signal to noise ratio by a factor of
root 2 (and doubling the effective Quantum Efficiency of the EMCCD). Only EMCCDs
with a low noise floor can perform photon counting. The approach can be further enhanced
through innovative ways to post process kinetic data.
The industry-leading darkcurrent and Clock Induced Charge (CIC)
specification of Andor’s back-illuminated iXon Ultra 897 and 888 models render them uniquely suited to imaging by Photon
Counting.
Photon Counting can only be successfully carried out with very weak
signals because, as the name suggests, it involves counting only single
photons per pixel. If more than one photon falls on a pixel during the
exposure, an EMCCD (or an ICCD for that matter) cannot distinguish
the resulting signal spike from that of a single photon event, and thus
the dynamic range of a single frame exposure is restricted to one
photon.
To successfully photon count with EMCCDs, there
has to be a significantly higher probability of seeing a
'photon spike' than seeing a darkcurrent/CIC 'noise
spike'. The iXon Ultra 897 and 888 have the
lowest darkcurrent/CIC performance on the market,
yielding both lower detection limits and higher
contrast images.
Under such ultra-low light conditions, 'photon counting mode'
imaging carries the key benefit that it is a means to circumvent the
Multiplicative Noise, also know as 'Noise Factor'. Multiplicative
noise is a by-product of the Electron Multiplication process and
affects both EMCCDs and ICCDs. In fact, it has been measured to
be significantly higher in ICCDs. The noise factor of EMCCDs is
well theorized and measured; to account for it you increase the shot
noise of the signal by a factor of square root 2 (~x1.41). This gives the
new 'effective shot noise' that has been corrected for multiplicative
noise. The effect of this additional noise source on the overall Signal
to Noise ratio can be readily viewed in the S/N plots in the technical
note entitled 'EMCCD signal to noise plots'.
Photon Counting Mode does not measure the exact intensity of a
single photon spike, it merely registers its presence above a threshold
value. It does this for a succession of exposures and combines the
individual 'binary' images to create the final image. As such, this
mode of operation is not affected by the multiplication noise (which
otherwise describes the distribution of multiplication values around
the mean multiplication factor chosen). The end result is that low
light images acquired through this mode of acquisition are improved
by a factor of ~x1.41 Signal to Noise, compared to a single integrated
image with the same overall exposure time.
To successfully photon count with EMCCDs, there has to be a
significantly higher probability of seeing a 'photon spike' than seeing
a darkcurrent/CIC 'noise spike'. The lower the contribution of this
'spurious' noise source to a single exposure within the accumulated
series, the lower the detection limit of photon counting and the cleaner
the overall image will be, as demonstrated in Figure 1.
The iXon Ultra 897 and 888 have the most effective combined
cooling/CIC minimization on the market, lower than other competing
EMCCDs utilizing the same 512 x 512 or 1024 x 1024 sensors. As such, the detection
limit for Photon Counting is markedly lower. The iXon intuitively
offers Photon Counting modes, either as a real time acquisition or as
a post-processing step. OptAcquire can be used to first optimize the
camera for Photon Counting acquisition.
Photon Counting by Post-Processing
As a post-processing analysis, the user holds the flexibility to 'trial
and error' photon count a pre-recorded kinetic series, trading-off
temporal resolution vs SNR by choosing how many images should
contribute to each photon counted accumulated image. For example,
a series of 1000 images could be broken down into groups of 20
photon counted images, yielding 50 time points. If it transpires that
better SNR is required, the original dataset could be re-treated using
groups of 50 photon counted images, yielding 20 time points.