Dynamic Range
Dynamic range of a pixel in a sensor (CCD or CMOS) is the difference (in for instance f-stops (1)) between the exposure needed to get an acceptable signal in the darkest areas and the exposure needed to reach high light clipping.
Pixel Size
A pixel beeing exposed with light is like a flat container in the rain. As long as the height of the container is the same, it doesn't matter how big or small the suface is - it will be filled with water at the same time. However a container with a big surface will hold more water than a container with a small surface.
Same thing with pixels. Here it is the effective silicon area that counts, not the pixel area (3). Effective silicon area = pixel area x fill factor (4). A pixel with large silicon area can hold more electrons than a pixel with small silicon area, but both pixels will be filled with electrons at the same exposure level! There is a common misconception that larger pixels do not fill up with electrons so quickly and that should be the reason for a larger dynamic range. "There is more time to capture the dark pixels before the bright ones start to overflow." If that was the case, the exposure time for larger pixels should be longer, meaning lower ISO for larger pixels. Actually it is the other way around. Since larger pixels have more information in the shadows, it is possible to shorten the exposure time - getting more headroom for high-lights without sacrificing shadow details.
Counting electrons
An example: We have two sensors, one with the pixel pitch (2) of 10 µ and one with 3,2 µ. The fill factor is 50 % for both sensors. The effective silicon area for the first pixel is 10 x 10 x 50 % = 50 sq µ. The same calculation for the second pixel is 3,2 x 3,2 x 50 % = 5 sq µ.?Letīs make an assumption. The small pixel can hold a maximum of 64 000 electrons (full exposure, bright high-lights). The big pixel will then be able to hold 640 000 electrons at the same exposure level. Now let's look at different exposure levels and count electrons in the different shades of an average 7 f-stop luminance range (5) image.
Full exposure, bright white, 255, 255, 255 64 000 640 000
One f-stop down to a bright tone 32 000 320 000
Two f-stops down to a light gray 16 000 160 000
Three f-stops down to almost mid gray 8 000 80 000
Four f-stops down to a slightly darker gray 4 000 40 000
Five f-stops down 2 000 20 000
Six f-stops down 1 000 10 000
Seven f-stops down to almost black 500 5 000
So far we have been shooting with the cameras lowest available ISO setting (100). Now let?s change from ISO 100 to ISO 400. That means two f-stops less exposure.
Eight f-stops down 250 2 500
Nine f-stops down 125 1 250
Finally we change to ISO 1600 with another two f-stops less exposure.
Ten f-stops down 62 620
Eleven f-stops down 31 310
Not a clue!
One important thing to consider: In the exposure moment, when the sensor in your camera (CCD or CMOS) is beeing exposed to light with the shutter open for 1/125 sec and the aperture stopped down to f8, the sensor doesn't have a clue what ISO or white balance setting you have chosen. These settings comes into account when the electronic circuits in your camera are converting image signals from number of electrons to digital bits and creating jpeg or tiff versions of the image. At this moment lots of valuable image information is dumped. The RAW format however is a lot more causious with the image information.
Introducing noise
Noise is uncontrolled electrons in the pixels. The noise level differ from pixel to pixel. There are different kind of noise, but they all interfere with the image signal.
Let?s make a new assumption: Noise electrons differ between 4 and 35 for different pixels. And let?s look at the eleven f-stop down level of exposure. The small pixel holds 31 electrons from the exposure. The big one holds 310. Let?s add the noise pixels. Now the number of electrons will differ between 35 and 70 for the same exposure with the small pixels. The big pixels will count between 314 and 345. With small pixels the difference is 100 % and with big pixels only about 10 %. Big pixels will give you cleaner, smoother - less noisy - shadow areas.
This is also the reason why noise is more apparent in the shadows. The higher up you go in exposure, the less impact noise will have. If it is 100 % for the darkest shadow area with the small pixel, it will be less than 1 % in the brightest highlight (between 64 004 and 64 035).
Reducing noise
If you know how many noise electrons each individual pixel will generate, you can subtract that amout from the total number of electrons. This is a trick Canon is performing on pixel level with their SLR cameras in order to successfully reduce noise in the shadows. Right after the exposure information is collected from the sensor, they make a second collection without exposure. This time it is only the noise electrons that is beeing collected. Now they subtract the noise information from the full information (exposure and noise) in each individual pixel.
There is a lot of research going on about noise reduction and we will see other solutions from different companies in the near future. Reducing noise in shadow areas, resulting in better details, cleaner colors and higher ISO speeds, is an important goal for camera manufacturers.
Calculate camera DR
At the time of exposure the sensor doesn't care what ISO is set on the camera. Knowing this, it is quite simple to determine the total dynamic range of the sensor. Pick a neutral surface as a target, for instance a gray or white patch on QPcard 101. Put the card in a neutral light like daylight. If possible make a custom white balance setting. Set the camera on Manual exposure mode.
Now set the camera on the lowest ISO, in my case (EOS 5D) to ISO 100. Measure the exposure preferably using spot metering. Take a shot using time and aperture setting for a mid gray exposure (Zone 5, or zero on the exposure scale of the camera). We are now going to make a top exposure in order to find the sensor's top level just before clipping high lights. Open the exposure with 2 2/3 f-stops and make an exposure, open another 1/3 stop and make the next exposure. Open another 1/3 stop for the last exposure.?Next step is to find the lowest possible exposure that still gives us some detail in the shadows. Set ISO on highest ISO setting (1600 for 5D). Measure exposure on the same surface in the same light. Take a shot using time and aperture setting for a mid gray exposure. Stop down exposure with 2 2/3 f-stops and take a shot. Stop down another 1/3 stop for next shot, and 1/3 again for last shot.
Open all pictures in Photoshop and crop them to 200 by 200 pixels. Use Filter --> Blur --> Average to get an average value of the pictures. Measure L of the L*a*b values with the eyedropper tool. With custom white balance setting a and b should be 0 and 0.The two mid gray pictures should be the same for ISO 100 and ISO 1600, around L 55. Measure L for the white frames from ISO 100 and for the black frames from ISO 1600. Find the frames with an L-value that is not 100?99 for white and 0-2 for black. Note the exposure setting for the two frames and calculate the difference in f-stops between the two frames.?In my case the difference between 1/6 f5.6 and 1/3200 f8 is 10 f-stops. The total dynamic range for my camera is 10 f-stops. Also note that from mid gray to white there are only 3 f-stops for ISO 100. Between gray and black there are 7 f-stops. Lots of footroom but no headroom at ISO 100.
Conclusion
If you look at noise, dynamic range, head- and footroom (exposure lattitude), high ISO - using a camera with big pixels is a great advantage. Most cameras are likely to cut highlights at low ISO, and cut dark shadows at high ISO. There is footroom, but no headroom at low ISO and vice versa.?Using a mid ISO, like 400 in the example above, and shooting with RAW format is a great advantage. Shooting with jpeg (and tiff) will cut most of the head- and footroom in the image (not in the sensor). With RAW you can extend the dynamic range in your images (not in the sensor) and thus save bright high-lights in over exposed and shadow details in under exposed images.
NOTE!
This is a schematic view over how pixels and sensors in digital cameras work. This article is published for basic understanding and not as a scientific document. We hope that understanding the basic physics of sensor and digital camera technology will give you a visual insight that might come in handy when you analyze your digital images and when you are facing a difficult subject with your camera.
Glossary
1) One f-stop: Double or half of the amount of light entering the camera. The difference between two "full" shutter speeds or aperture settings. Changing the shutterspeed with one f-stop is like going from 1/125 to 1/250 sec, or from 1/30 to 1/15 sec. Changing the aperture setting with one f-stop is like going from f8 to f11, or from f4 to f2.8.
2) Pixel pitch is the sensor length devided with the number of pixels along the length. A Canon 5D sensor is 36 x 24 mm and has 4368 x 2512 pixels. 36 / 4368 is 8,2 microns (µ).
3) Pixel area for the same sensor is 8,2 x 8,2 µ = 67 sq µ.
4) Fill factor is the part of the pixel covered with light sensitive silicon devided with the full area of the pixel x 100. If a pixel is 10 x 10 µ and is covered with 6 x 6 µ silicon, the fill factor is 6 x 6 / 10 x 10 x 100 = 36%.
5) Luminance range is the lightness difference (in for instance f-stops) between the darkest shadow area and the lightest high-light area in a subject. If you pin point the darkest and the lightest areas of a subject and measure the exposure levels with a spot meter, you will get the luminance range in f-stops. Seven f-stops is a normal luminace range for an average subject with dark areas in the shade and white areas in the light. For high contast scenes the luminance range can be eight or nine f-stops.
Download interesting Powerpoint "CCD Basics, A Technology Overview"
For those of you who are interested in a more scientific report about pixel size