DSLRs and Unity Gain - How to Properly Set Your ISO

In the time of film, sensitivity of the emulsion increased in a near- linear fashion according to the film's "ISO" for short exposures.  But longer exposures died quickly in a phenomenon known as "reciprocity failure."   Here, film speed was somewhat of a secondary concern to the film's staying power.   Truly, it didn't matter how "fast" the film was if it was going to lose its speed after a short amount of time anyway.

But for normal exposures, practically speaking, film ISO was a trade-off between the emulsion's sensitivity and the point where the film "grain" became objectionable - you needed larger emulsion crystals to lift sensitivity to light, which made high ISO films more grainy.

So, utilizing high ISO film was not always practical because it was inherently noisy. This was also true of any film if it was "push-processed" - overdeveloped in the darkroom to mimic a higher speed film.  But shooting film was even more troublesome because the single, long  exposures were a prime target for airplane trails, satellites streaks, and cosmic ray hits.

In the film camera, such as a the Nikon F2 that I still fondly cherish to this day, physically setting the ISO on the camera did not magically cause the film to change its "speed" like today's DSLRs.   Instead, the ISO dial reminds the camera's internal metering electronics as to what type of film your camera is using so as to properly adjust its automatic, programmed mode setting, fondly known by "real" photographers of my acquaintance as "dummy mode."   

So, why talk about film here?   Well, the ISO setting is NOT entirely different with a digital camera. 

---

This is a brutally confusing topic.   Here's why.  

If you read ANY website over photography it will go into all sorts of dialogue about sensors being ISO-less and about dynamic range and about digitization of pixel values at the ADC's bit-rate.     But in astrophotography, none of this matters.   

I'm going to give you the rule first and then we'll discuss why.

The rule:  

Optimum ISO = 


Follow along...

The best SNR you can achieve in a sub-frame will be the square root of the full-well capacity of the sensor.   This represents the brightest "signal" divided by shot noise arising from that signal.    For example, in a Nikon D810A, 

​

Like film, it does not "speed up" the exposure in the sense that you are collecting data faster.  Rather, it is simply responsible for determining how the camera will read-out its pixel charges into the Analog-to-Digital Converter (ADC).  While you may think that the imaging chip is much more sensitive at the upper limit of ISOs on your Canon 5D or Nikon D5, this is simply far from the truth, and brainlessly setting it to the highest setting possible will not be to your benefit.

There is an ISO setting that is proper for you, where you will maximize the collection of your on-camera data.   This article explores that.   But first, let's talk about how the DSLR works.

How a DSLR Represents Data...

We must be careful when we call DSLRs "digital."   In a sense, this is incorrect.   When the chip records light from the night sky, the silicon stores the photons as electron "charges" within the photosensitive areas of the chip, which we know as "pixels."   But at this point, the amount of current in each pixel is strictly "analog."   In fact, digitization of the charges doesn't happen until after the next step of amplification.  This level of amplification (known as the camera "gain") in a DSLR is established by the user upon setting of the exposure's ISO.   

What is likely not apparent is HOW the amplifier works.   When charges are accumulated



SNR before amplification will be the same after amplification. 



For example, if a pixel is 1/4 full of charges, then setting the ISO to anything more than an amplification of 4x (times) will fill up the pixel. 


"Under" Optimized ISOs

With CCDs and CMOS chips, as found in today’s DSLRs, the electronics attempt to simulate ISO changes by increasing the amount of "gain" at each level. The idea is that if you have, for example, 1000 converted photons in a single pixel at ISO 100, then you can double the gain and achieve 2000 levels of Analog-to-Digital Units (ADU) at ISO 200 (ADUs are the output upon readout). So, upon read-out, the camera has twice the value in each pixel. If a pixel starts near the saturation point (full-well capacity) of the pixel, then adding too much gain will put values over the threshold, meaning you burn-out the data in those pixels. This means that overexposure comes faster.

So, from the camera, it appears you’ve doubled the sensitivity.  But this is merely an appearance.

DSLRs, upon read-out, have a number of potential levels of illumination, which may, or may not, represent all the photons that were recorded. This number is typically 4096 possible levels of illumination, for 12-bit DSLRs, or 16,384 levels for newer 14-bit DSLRs. How the converted photons, or pixel charges, are converted depends on how you set your ISO (or gain).

So, in a 12-bit camera at ISO 100, the camera will likely set its gain factor where it takes several converted photons in order to represent one ADU out of the camera. It does this in order to protect against overexposure on typical shots. So, if the gain factor is 5 at ISO 100, a 1000 level pixel is assigned to the 200th level upon output (5 photons per ADU). At ISO 200, or gain 2.5, it would be at the 400th level. At ISO 400, or gain 1.25, it would be at the 800th level….and so on. Thus, lower ISOs surrender photons captured in order to protect the top-end from overexposure.  For the astrophotographer, light photons captured in those elusive shadow regions just got deleted from existence, never to be seen again, sadly.

Thus, at gain factors greater than 1 -  "under" optimized ISOs on the camera - we lose data because the actual amount of photons getting recorded are wasted by this "rounding" of pixel charges, something known as "quantization error". So, up to the point where a camera has a "1 to 1" representation of converted photons to ADUs, there is decreased "sensitivity," to use the term somewhat incorrectly. To say it more accurately, the data loss in this conversion from analog to digital units causes a loss in dynamic range.   Areas in the shadows that should have had a gradient of illumination values will now be under-represented.  This "1 to 1" level is something known as "unity gain," and it varies according to what camera you have - normally somewhere between ISO 800 and 1600 on 12-bit DSLRs and 50% of that on 14-bit DSLRs. It is affected by the camera’s total "full-well capacity" (the amount of charges a pixel can hold before saturation) and the size of the AD converter, which, as I said, is typically either 12-bits or 14-bits in a DSLR.  

So, for typical photography, the manufacturer has designed the camera with a range of ISO values to maximize the types of photography you can do with your DSLR, from fast-motion, low light situations, all the way to bright-light landscape shots.  And they must accomplish this using a level of technology that puts a cap on the overall dynamic range of possible illumination values.  Truly, DSLR makers never complete knew that our application, astrophotography, required a whole different kind of technological specifications!

So, you might think that the highest ISOs on your DSLR would be your best bet and, perhaps, allow you to accumulate your data "faster"?  Well, no.


"Over" Adjusted ISOs

At ISOs where the gain is set to less than 1 (higher than unity gain), what happens is that the charges in a pixel are assigned to ever increasing ADUs, normally skipping ADU numbers. For example, if unity gain is at ISO 800, then going to ISO1600, or increasing gain by a factor of 0.5, will assign 1000 charge pixels to the 2000th ADU level. In the same trend, a pixel with 1001 charges will get assigned to the 2002nd ADU level. This means you will have no representation at the 2001st level.  This might not seem like much, but when you increase to ISO 3200 (or 4 "stops), then you will have 3 missing units between numbers, and so on.  When you do this, especially on a 12-bit DSLR with 4096 possible AD units, you will be greatly shrinking your camera’s full well depth. In other words, you have full photon representation in the conversion (no quantization error), but more and more pixels will become increasingly oversaturated, which does NOT take long on bright stars.   

Reaching saturation faster is not a virtue.   Doing so is merely wasting photon capacity at the top-end, clipping your stars. Therefore, in ISOs above unity gain, there is no increase in "sensitivity" since photon representation was long ago maximized.

So, in reality, increasing your ISO is not gaining you any more photons than you originally available.  There is no magical switch on the camera that allows you to accumulate photons "faster."   Photons are what photons are.  There will be no actual increase in the NUMBER of total photons recorded by the chip itself.  Applying "gain" is just a way of performing some mathematics (through electrical currents) to the pixel to achieve the perception of image brightness.  Remember, most people do not post-process their images, so they want their images to be ready for viewing upon image readout.  For the astro-imager, we just need solid data for post-processing.   


Putting It Into Practice

If photon maximization and great dynamic range were the only virtues to properly optimizing your ISO then that would be quite enough.  But there are certainly other noteworthy aspects worth considering, all of which help to improve upon the performance of the DSLR as an imaging platform.  

On the pros/cons list of DSLRs, particularly in comparison to their bigger Astro-CCD cousins, DSLRs surrender a lot in the way of thermal noise.  While today's DSLRs are improving in that regard, the lack of internal cooling options makes long-exposure astroimaging much more difficult than with a dedicated astro-camera.   And, in fact, setting a camera's gain to higher than necessary simply compounds that issue. 

High ISOs naturally produce extra levels of thermal noise because of the camera amplifier - the part that sets the level of charge based upon the gain - is working harder to do its job.  It gets hotter because higher ISOs necessitate more AD conversions (more electricity is required) and thermal noise is the byproduct. So, at a certain exposure length and/or ambient temperature, you will end up with a better image (greater S/N ratio) with an ISO that best balances photon representation with the level of thermal noise.  That level will be at unity gain.

But perhaps the bigger issue, an issue which I seldom hear addressed, is what happens in smaller bit-rate cameras to our star images when we begin to lengthen our sub-exposure times?

To see what I mean, take a long sub-exposure with a DSLR and open the RAW file in your image processing program.  Of course, this will represent the unstretched data, exactly how the data was read-out of the camera.   Do you notice the stars of this unstretched image?  More than likely, even in a 14-bit DSLR, the bright stars are easily seen, as well as many of the fainter ones.  What this means is that the cores of those stars have already maximized the full-well capacity of the chip and have oversaturated to the point where there is significant data loss on the high-end.   Whereas there is no data loss in terms of image details, there will be a huge loss in star color.   

This is data you cannot retrieve.  It is gone.   It is complicated by the fact that now you have larger stars in the image than you might have intended and stars that have lost their natural softness.   And how are you going merge the color of the star's outer halo with the already maxed-out central core?  

And to make matters worse, this makes image processing more difficult because you now have to be extremely careful in how you stretch your data, otherwise, you lose the rest of the color of your stars.  Many techniques have now been developed which might involve star masking or their entire removal during processing, but I have found these developments to be reactionary to the fundamental problem, that namely, even with a 14-bit DSLRs we must be very careful to first find the optimal ISO and then be careful not to go too long with our sub-exposure lengths.  12-bit DSLRs simply do not offer the necessary bit-rate.

As a response to the difficulty in finding this balance, software such as Pixinsight offers nice HDR compression routines to help preserve the top-end, as well as the ability to do the aforementioned star-masking.   However, I see many people now using such techniques with data acquired with dedicated 16-bit (65536 ADU levels) Astro-CCD cameras and I am left to wonder why?  Proper logarithmic stretches (good curves) are all one needs to combat this lack of headroom.  

But the lesson is as follows...care must be taken to protect your star images, because once that data is clipped it is gone forever, as is the star color that that data represents.   Of course, a higher than necessary ISO just makes that point of star loss come much faster, all the while you increase noise in your image with absolutely no gain in additional information. 

In truth, setting the actual ISO for long exposures depends on many factors, and it's often not a matter of simply finding out what unity gain is for your camera and keeping it there.   For many DSLRs at any given exposure length, you will likely find that even at unity gain you will need to shorten exposure times in order to prevent star over-saturation.    A target such as Rosette Nebula, where you have a bright NGC cluster of relatively bright stars in the center, will peak out the stars VERY quickly at unity gain with 12-bit cameras and, in my experience, you cannot maximize your sub-exposure lengths to achieve best signal to noise efficiency.

I used that same target recently for some images (see here) and found my 5 minute sub-exposures to be quite objectionable with regard to star saturation.  I used a Canon 60Da set at ISO 1600 using a pair of f/7-ish 6" apochromatic refractors for these images.   In retrospect, ISO 1600 was too much for the 60Da, as I learned afterward that unity gain is much lower, perhaps as low as ISO 250 to 400.  This means I was over-exposing at 5 minute sub-exposures.  However, it was a cold night so I had hoped to take advantage to go longer with the sub-exposures.   More investigation into the optimal ISO and sub-exposure length is definitely needed for the Canon 60Da and other 14-bit cameras, however it goes to show my point...brainlessly setting your ISO too high can cause problems you do not anticipate!

Even so, imagine had the target been the Pleiades? 

Finding the sweet spot regarding ISO will be quite individual to the gear being used and the target being acquired.  But one thing is for certain...you will never gain improvements in the image beyond the unity gain ISO of your camera.  

​