• Global Sites

Relative SNR—How does your camera compare?

THIS GUIDE IS FOR:

Biologists who want to understand why camera makers throw around terms like median and r.m.s.

THIS GUIDE OFFERS:

An explanation, with visuals.

Learn more...

Learn more about noise and CMOS technology, read Changing the Game.

Download a PDF of the camera specs used in this tool.

Want to see the math behind the SNR calculator? Click here.

In an ideal imaging demo, every piece of equipment would arrive on time, all software and hardware would integrate seamlessly, biological samples would expresses their fluorescent proteins as expected, and they would glow brilliantly. The microscopy system would be perfectly aligned, and designed with optics and a camera to maximize capture of every photon, providing astounding resolution due to perfect compliance with Nyquist.

Yeah, right.

Demos will always be demos, but we believe that you can take steps to inform application-specific camera or camera mode choice beyond the cut and dry data sheet and before the orchestration of a demo by using this handy absolute and relative signal-to-noise ratio (rSNR) calculator.

At this point you are probably asking “What is rSNR and why should I care?” A standard, absolute SNR plot looks at the SNR of a single pixel over a range of light levels. The relative part simply means that we’ve plotted that SNR relative to a perfect camera making it easier to see differences at various light levels. In addition to handling the SNR math, we’ve also set this calculator up to automatically adjust pixel size, since this is difficult to do with optics in a demo. In other words, all comparisons are presented as if you’ve set up your optics for a 16 μm2 pixel (just the size of a typical EM-CCD camera pixel). We’ve chosen this size because EM-CCDs are typically the biggest pixel size used in microscopy and bigger pixels have a huge advantage in SNR camera comparisons simply because they can collect more photons, but with lower spatial result; a bigger bucket collects more rain, but you don’t know exactly where it fell. To really know how two cameras’ pixel SNRs compare, pixel area must be matched.

To put it very simply, assuming that SNR is a relevant performance indicator, this calculator can help answer the age old question “Which camera is best for my application?”

Using the interactive graph

Here are a few quick pointers on getting the most from this tutorial:

  1. Read noise. When you enter read noise, make sure it’s the read noise specified for how you run your camera and defined as “rms” noise.
  2. QE. Enter the QE of your camera at your wavelength of interest. All other cameras will automatically adjust to that wavelength with the appropriate QE.
  3. Pixels. Once you enter your camera’s pixel size, we calculate the correct factor to compare all cameras to a 16 μm2 pixel.
  4. Binning. If you choose 2×2 binning, all cameras are automatically set to 2×2 and the optical matching is updated.

Background photons. Although you may not know exactly how many background photons your sample has, this variable is included so that you can observe the dramatic effect of even small background (one photon!) on SNR.

A little bit more background.

 

  1. Read noise. This specification of camera noise is the bees’ knees of camera low-light performance. However, to get the correct results, be sure to enter your read noise in your operating mode and that it is specified as r.m.s., not median. The difference is subtle, but meaningful. For CCDs, typical r.m.s. read noise values are 6 -10 e-, for sCMOS 1.5 – 2 e-; for EM-CCD, this number becomes irrelevant, as using EM-CCDs at high gain virtually eliminates read noise from the SNR equation (see why here), so we’ve autofilled this variable for any EM-CCD entries.
  2. Quantum efficiency and wavelength. An often overlooked spec, quantum efficiency is the percentage of photons striking the sensor that are converted to electrons and therefore measured. A low QE means more exposure time is required to reach similar SNR levels and therefore more photobleaching and/or toxicity can occur. QE is sensor and wavelength dependent, thus we need your camera’s QE at your fluorescent emission wavelength of interest to make sure that all the other pre-programmed cameras are, well, on the same wavelength. For example, if mCherry is your preferred fluorescent label you would enter the QE of your camera at 610 um which is the peak of mCherry emission.
  3. Pixel Size. If all cameras had identical pixel sizes, then camera comparison would be easier. Unfortunately, pixel size is a key factor in SNR calculations because larger pixels collect more photons (at the expense of resolution). But in this virtual world we can bypass these optical issues and simply include an optical mag factor to normalize all pixel areas. We could have chosen to match to any pixel size and the results would have been the same, but since EM-CCDs have been the standard for low-light imaging, we’ve normalized to the 16 mm2 pixel of 512×512 EM-CCD cameras. The effect of normalized pixel area means that as you enter various pixel sizes, the actual graph does not change. That is not a bug; it is as planned.
  4. Binning. Traditionally, analog binning in CCDs and EM-CCDs is used to increase SNR while sacrificing spatial resolution by electronically creating a larger pixel. With sCMOS, binning is only digital, so read noise increases by the bin factor. When comparing EM-CCDs and sCMOS, a common mistake is to apply a 2×2 bin to the 6.5 μm2 sCMOS pixel to approach the 16 μm2 pixel size of an EM-CCD. Seems like a pretty logical and simple solution. The problem is that the total area of the EM-CCD pixel is still 51% larger than the “13 μm” binned sCMOS pixel and the effective read noise of the binned sCMOS pixel has increased by a factor of two. Such a comparison is like expecting a small bucket with a slow leak to collect more rain that a perfect bucket two times bigger. Not so logical after all.
  5. Sensor Type. EM-CCD, CCD (including EM-CCDs in CCD mode), or sCMOS? This question is relevant because different technologies have unique considerations. The EM in an EM-CCD provides electron multiplication of the signal but also adds noise, termed excess noise factor, equivalent to 1.4x. CCDs and sCMOS do not and therefore this factor defaults to 1x for these cameras. Also binning in sCMOS is digital and this, as discussed above, alters how binning affects read noise.
  6. Relative vs. Absolute SNR. SNR plots are usually presented as absolute with log scale of intensity (or photons) as the x-axis. But this limits the ability to see what’s happening at various light levels because of the nature of the log plot. To explore these low-light regions, we use relative SNR (rSNR) plots, which simply makes all SNRs relative to a perfect camera (0 noise and 100% QE). The tutorial allows quick switching between these two graphs.
  7. Input photon number. This is the million dollar question. How do you relate the x-axis of the SNR plots to your sample? In other words, how many photons do you have? With a few simple, readily available numbers from your camera and your images, we can walk you through a ballpark calculation of this in <2 minutes—find out how here.
  8. Background photons. If only every biological sample provided a crisp, clear, bright signal on top of a black background. But we know this is not the case. Often the signal exists in the murky haze of background photons which add noise to SNR calculations. Even just a few photons of background can mean the difference between an EM-CCD or sCMOS having the best SNR. You can use the same calculation for how many photons do I have to estimate your samples background photon level. Just measure the intensity in a region that is near to, but does not directly include, your labeled target.

Want to see the math behind the SNR calculator? Click here.

Inquiry about Life Science Cameras

Contact us

 

Scroll to Top