HMC40 and low light performance

[droliff]

Hi all,

I recently purchased an HMC40 and am mostly quite pleased with it thus far.
I did a lot of research before investing and did hear that the 40 does not perform real well in low light conditions.
Now that I've had the camera out on a few gigs, i can see that the 'low light' performance is a bit of an issue in some cases (available light in some venues).
Is the poor performance in low light due to the cameras 1/4" capture chip size?
Is the HMC150, which has 1/3" chips, a better performer in less than ideal light conditions?

Thanks for any input.

Dennis

Dennis Roliff Photography

 

[GrahamH]

Yes and yes.

http://www.dvxuser.com/V6/showthread.php?t=182865

Sensitivity -- okay, so here's the bad news. We knew it wasn't going to be nearly as sensitive as the HMC150; when you cram 4x as many pixels onto a smaller chip, you can expect a drop in sensitivity. Mathematically, I would have guessed about 3 stops' less sensitive. The pixels are 4x as many, meaning 1/4 the size, meaning 1/4 as sensitive (2 stops), but also the chips are smaller (1/4" instead of 1/3") making the pixels even smaller, so... yeah, 3 stops. When I observed the two cameras side by side, yeah, looked about 3 stops different. When I shot a grayscale chart, lighting for best response, with the HMC40 at f/2.8, the HMC150 needed to be at f/8 to deliver the same brightness level. So... yep. 3 stops. The HMC150 is about 3 stops faster than the HMC40. Which, when rating the HMC150 at 500 ISO, that puts the HMC40 at about 64 ISO. Not exactly a speed king. Plenty fine for outdoors, but indoors you're going to either need lots of light, or lots of gain. Fortunately the HMC40 provides bucketloads of gain, up to +34dB! That's nearly four stops' worth of gain. That gets you your sensitivity back, at the expense of noise.

 

[v3rlon]

The short answer: yes to both questions.

Think of the sensor like a chess board with each sqare representing one pixel. For each square, a given amount of light is available. Just to keep it simple, we'll call that one light unit.

So each of 64 squares on our standard sized chess board is getting one unit of light. If we used a smaller chess board, then each sqaure would get less light because the area of the sqare in part multiplied by the length of time the square is exposed determines how much light you get.

Also, if you take a board that is the same size as our original board, and double the number of squares on it, then each square will get less light because they will be smaller than the sqaures on the original. Again, thinking of the chess board, the only way to add squares without making the board bigger is to make the squares smaller. The HMC40 has high resolution sensors (lots more squares than needed for 1080p so it can take 10MP stills) that are small (1/4") so its lowlight performance is lackluster.

So, for a given technology, smaller sensors tend to have worse lowlight performance.

The only caveat there is "for a given technology." In the same way that they can make computer chips use less and less power, they can also make them use less and less light as they find ways to tweak the process. So, cameras 5 years from now will not compare evenly to cameras of today. They will reach a point where laws of physics get in the way, but we still have some room to work with yet.

Does that all make sense?

 

[Robert M Wright]

Obviously, the 150 is going to outperform the 40 in dim lighting, but the difference is mitigated a bit by cleaner gain with the 40. The HMC40 is the first video camera I've ever handled that I'm comfortable cranking the gain up to 12dB.

 

[Bic Parker]

Hi all,

I recently purchased an HMC40 and am mostly quite pleased with it thus far.
I did a lot of research before investing and did hear that the 40 does not perform real well in low light conditions.
Now that I've had the camera out on a few gigs, i can see that the 'low light' performance is a bit of an issue in some cases (available light in some venues).
Is the poor performance in low light due to the cameras 1/4" capture chip size?
Is the HMC150, which has 1/3" chips, a better performer in less than ideal light conditions?

Thanks for any input.

Dennis

Dennis Roliff Photography

The size of the chip is irrelevant. But the size of the pixels sort of has a bit to do with the light sensitivity, but only compared with similar chips. The actual answer is a bit more involved.

So, first thing, comparing a CMOS chip (in the HMC40) to a CCD chip (in the HMC150) is not a very good comparison because of how the electronics are designed. Essentially, because of the onboard electronics in the CMOS chip, you are going to inherently have more electronic noise, which will, in turn reduce the signal to noise ratio and reduce the overall light sensitivity compared to a CCD chip with similar pixel dimensions. A CCD can be more sensitive than a CMOS and even the comparative pixel dimensions may be irrelevant within a range.

The pixel size is important for many reasons, but in the case of light sensitivity, it is a factor in the well depth of a pixel (i.e., how many photons can a pixel potentially read at a given point in time). A larger pixel can have a potentially larger well depth (emphasis on "potentially").

But there are other factors go hand in hand to define the overall light sensitivity. The combined factors are the chip's characteristics in pixel well depth, signal to noise ratio, dark current, and quantum efficiency. These factors together will give an accurate picture of how well the chip (and ultimately, the camera) does in low light.

I did the math in another post, estimating the pixel size (for image scale reasons, not light sensitivity) for the HMC40 sensors, since I couldn't find any published information (I put the math here again at the bottom of this post). It basically works out that each pixel is about 3.5 microns across (that may be the average dimension, since they could be rectangular instead of square). That is not just small, that is tiny for imaging chips (ultra sensitive chips may have pixels as large as 25-40 microns, by comparison).

Now 3.5 microns, though, isn't completely bad since each color has a pixel. So think of 3.5x3 = 10.5 microns, which means that you have a moderate light gathering potential. It is probably somewhat less than the light gathering potential of a 10.5 micron pixel, though, probably closer to a 9.5 micron pixel.

The kicker, though, that brings this back is the CMOS architecture. This essentially means that dark current is higher (that is basically the inherent electronic noise in the pixel when the cap is on), the quantum efficiency is lower (this is a percentage of photons detected to those received), and the signal to noise ratio is lower than an identical pixel on a CCD chip.

When you kick up the gain on a camera, you are going to be boosting the overall data from the chip, which is going to include noise, which why noise appears. Some of the more sophisticated designs these days, mathematically eliminate some of the noise (actually it is a statistical elimination from the tails of the histogram curves, usually) with the increase in gain, which is what I suspect is going on here with the HMC40 (which is really cool, by the way).

Here is the estimated calculation, by the way, of the pixel size in the HMC40:

Each sensor has ~3.05 million pixels on a ~1/4.1" square (I love converting metric to english and back!). So 1,747 (the square root of 3.05 million rounded up for conservative measure) divided into 6.1722 mm gives you ~0.0035mm (3.5 microns), which sounds about right for a chip that size with that many pixels (interestingly enough, this also gives a ballpark explanation of the well depth of the pixels, which goes to in part, not including dark current and quantum efficiency, its light sensitivity).

 

[GrahamH]

The size of the chip is irrelevant.

.... I did the math in another post, estimating the pixel size ... It basically works out that each pixel is about 3.5 microns across ... that is not just small, that is tiny.

.... 3.5 microns ... sounds about right for a chip that size

Interesting post - thanks for that. I have to say, though, that your second two points above contradict your first.

 

[Barry_Green]

Not really. The individual pixel size is far more important than the chip size. The chip size does restrict how big each pixel can be, of course; the more real estate you have to work with (for a given pixel count) the more surface area can be allocated to each pixel.

I think Bic's summation is pretty excellent, with the only real issue being the math on the chip size. Because a 1/4" chip is nowhere near 1/4" in size, in any dimension, and it isn't anywhere near to being 1/4" square. We don't know the exact dimensions of the HMC40's chip, but it's more likely to be about 3.7mm x 2.1mm (for 16x9). I think the HMC40's chip is actually 3:2 because of its still capabilities, but I don't know what the dimensions would be for that aspect ratio.

 

[GrahamH]

Not really. The individual pixel size is far more important than the chip size. The chip size does restrict how big each pixel can be, of course

Fine. I don't want to get into an argument about choice of words. Chip size being 'less important' sure, but I don't agree it's "irrelevant".

 

[BobDiaz]

This is an interesting side test. I took a video shot at 1920x1080/30p and reduced it down to 640x480. In this test, 9 pixels combine to make 1 pixel. In theory, this should help to reduce the impact of noise in the video.

No question that it helps, but it's subjective as to how much more you can push things. Some may feel that +18 dB should be the upper limit, others may find that +24 dB still holds together very well, and a few may be willing to push to +30 dB.

In order to avoid any extra compression noise getting in the way, I compressed this video at a very high bit rate, thus the large size (352MB). You can download the fie and judge for yourself what the upper limit should be when converting to SD resolution.

You can download the video here and tell me what you think:

http://vimeo.com/9991641


Bob Diaz

 

[Barry_Green]

Fine. I don't want to get into an argument about choice of words. Chip size being 'less important' sure, but I don't agree it's "irrelevant".

Well, I guess the point is -- if someone says "my camera has an S35 sensor, so it has better low-light performance than your 1/3" camera" -- that isn't necessarily true. So that's what I think he was getting at. You can have a 1/3" camera that's way better in low light than a "full frame" camera, depending on how they build it.

I guess in other words, it's not a determining factor. But, all other things being equal, the larger chip does provide the *potential* for better low light performance, depending on how the engineers utilize the additional space.

 

[Bic Parker]

Interesting post - thanks for that. I have to say, though, that your second two points above contradict your first.

I'm not certain as to what you mean, I think I was pretty clear, however you are taking that last line a bit out of its original context. I simply meant that for the form factors, 3.5 microns seems consistent with what I expected.

However you want to take it, the point here is that chip size means nothing other than providing a space for pixels and their support electronics. It is the pixels, the chip design, and their support electronics.... the combination of pixel size, quantum efficiency, dark current, and signal to noise ratio (which can be affected by several factors, too) is most significant when you are talking about low light sensitivity. Pixel size can also get quite significant to the picture quality in specific bandwidths when you get to tiny pixels (i.e., colors like red/green/blue) because at that point, you may be operating against some optical limits for a given wavelength.

So the chip size perhaps gives a bit of a guideline, but really means little, especially with comparisons. It is worth noting, too, that Panasonic's 3MOS design is a lower noise/lower power CMOS design (I think they are using fewer circuit elements per pixel) than prior CMOS designs of theirs. It is too bad that Panasonic doesn't publish more of their chip specs. I think the information would be enlightening and not really reveal any trade secrets (btw, this sort of data is readily available for Sony and Kodak chips).

Barry G said, "...with the only real issue being the math on the chip size. Because a 1/4" chip is nowhere near 1/4" in size, in any dimension, and it isn't anywhere near to being 1/4" square. We don't know the exact dimensions of the HMC40's chip, but it's more likely to be about 3.7mm x 2.1mm (for 16x9). I think the HMC40's chip is actually 3:2 because of its still capabilities, but I don't know what the dimensions would be for that aspect ratio.

Thanks for pointing that out Barry.... I'm more used to looking at imaging sensors for astronomical imaging systems and in that case, the sensor measurements are what they say they are! LOL... Having said that, I just did a bit of research on camera imaging sensors by Panasonic and others and found that 2.5 microns or less is probably pretty close to what they actually use, and possibly less, since there is a bit of a border around the effective collection area (for support design and connections). They are also likely not square (though some imaging sensors, such as those from Kodak, do have square pixels). In any case, we are talking about a pretty small pixel area.

 

[GrahamH]

We all seem to agree that larger pixel size contributes to light sensitivity, all else equal.

So if there's presently a bunch of empty space on the HMC40's imaging chip, such that if the engineers had wanted to they could have maintained the present number of pixels, but made them bigger, then I'll accept that the 1/4" size is not important.

 

[Robert M Wright]

From the HMC40 manual:

Effective pixels (still image)
Approx. 2.32 million pixels x 3 (4:3),
Approx. 2.65 million pixels x 3 (3:2),
Approx. 2.51 million pixels x 3 (16:9)

It's pretty obvious that the dimensions of the chip are not 16:9, but that shouldn't throw Bic's calculations off by much. (It's not realistic to expect camcorder manufacturers to design imaging chips around putting them in camcorders. They sell way more imaging chips in still cameras, and a 3:2 chip is close enough that it just isn't likely to be worth the additional R&D and production costs to make precisely 16:9 dimension chips specifically for camcorders.)

 

[Bic Parker]

I think probably the most important thing I noticed about the chip(s) (and we talked about this in another thread) is that it appears to be really well matched for the HMC40's optics, in terms of image scale, which goes a lot towards not just optimizing low light performance against the optical train, but the overall quality of the pictures this camera can deliver in a variety of situations.

Robert, I think you are right about how Panasonic supplies its chips across its products. Sony and Kodak do the same thing and, in fact, sell them to other manufacturers for a wide variety of camera applications (Panny apparently does not, since they don't list their chips on their industrial product listings). The 3.05 million pixels I used, btw, is roughly the raw number of pixels on each chip that I got from the HMC40 spec sheet.

 

[Bic Parker]

We all seem to agree that larger pixel size contributes to light sensitivity, all else equal.

So if there's presently a bunch of empty space on the HMC40's imaging chip, such that if the engineers had wanted to they could have maintained the present number of pixels, but made them bigger, then I'll accept that the 1/4" size is not important.

Well, whatever works for you then. It is not so much what you or I accept, but how these doodads are made. If it helps, go look at chip specs for Kodak and Sony chips. You will find that with a given chip size, the pixel size will vary by several microns, depending upon the chip design. This is for several reasons, but it all boils down to matching a chip with the optics. Imaging chip makers such as Sony, Panasonic, and Kodak will make a range of models that may have the same form factor size, but different pixel specs to accommodate the requirements of the camera they are installed in. So, for instance, one 22mm chip will work well in a larger aperture/long focal length optical train (larger pixel), while another will work well in a smaller aperture/shorter focal length train (smaller pixel).

Larger pixel chips, btw, are generally more expensive to make than smaller pixel chips (the rejection rates go way up in creating larger pixels over a chip that have a flat performance across the chip). Another factor in chip costs are how many errors are allowed in each chip. A Class 0 chip, with virtually no errors, is extraordinarily expensive and is generally only found in research level equipment. Class 1 to Class 3 chips have increasingly higher numbers of defects in them (identified as point, cluster, and column defects). These chips are what are often found in consumer and professional cameras. Usually these errors get mapped out by binning and/or oversampling, so they don't show up in the final images (except for direct bitmapped images such as RAW, FITS, TIFF, et al).

I just see so many comments relating chip size and light sensitivity and this comparison is just wrong. Even more so in the context of the OP, who was asking about a comparison with the HMC150's larger chip, which was a different type of chip (CCD, not CMOS).

Imaging chip designs can vary widely, but the most common parameters of useful comparison focus on the pixel and the underlying electronic performance.

 

[v3rlon]

Well, he was right that the 150 has better low light performance, and part of that is related to a larger chip (which has larger pixels on it). Its just that the answer gets more complicated when you start digging. The CMOS versus CCD is important, as is the cleaner gain of the HMC40.

Visible red light caps out around 700nm or .7 microns. I think any pixel size below that will create problems without some breakthrough work-around. That would be effective pixel size or the light gathering portion...not the support electronics. Even at that size, you'll have to deal with signal to noise nightmares.

But, silicon real estate is what drives cost, and the smaller the chip, the less likely something bad will happen to it. On a 12" silicon wafer, you can only fit so many 35mm full frame chips on it. The only way to get more is to make the sensors smaller. THe manufacturing equipment won't handle bigger wafers, and upgrading the fab (manufacturing plant) costs billions, not millions. Now, the air in the fab is kept clean to some specification, but there will always be SOME particles out there. The bigger then sensor, the more likely it will be hit by one of these particles creating a defect to make the chip worthless (its a bigger target, after all). However, that does not reduce the manufacturing cost of the wafer.

Lets go back to a chess board. You have 64 sqares on the board that you can sell for $20 each. If you could sell EVERY square, you'd make $1280. You can't though. Defects and other issues mean you can on average only sell 40 square and make $800. Some boards have more and some have less, but 40 is the average. The chess board costs you $500 to make regardless of whether you get 64 sqares or none, so it behooves you to get as many squares as possible to work.

If you invent a new board with half size squares, you can get 256 on a board and sell them for $15 each. If you get 160 of them to work (yield), you make $2400 and still only pay about $500 to make the board.

So boatloads of money are being poured into ways to make the chips smaller all the time. Compter processors, memory, CMOS, CCD, it doesn't matter. Smaller is cheaper to make in the long run.

And chip size IS important with regard to other things, like depth of focus. All full frame sensors and 35mm film have the same DOF regardless of the number of pixels. Similarly, all DX sensors (like in Nikon cameras) have the same DOF regardless of pixel count, but have a longer DOF than full frame sensors. All DX sensors also effectively multiply the lens focal length value by 1.5 (crop factor) since still photographers seem to be stuck in the mode of discussing zoom as "100mm" instead of "2x" even though a 100mm lens on anything but a full frame sensor (or film size other than 35mm) is going to produce a different level of magnification.

just in case you weren't confused...

It may not be rocket science, but it can definitely give you a reason to hit the books every now and again.

 

[Cranky]

Does one big pixel produce a better (whatever "better" is, I think the major criteria being less noisy) than several small pixels grouped together having size of one big pixel? Because while the difference in resolution between, say the HMC40 and the HV20 is in favor of the HMC40, it is not THAT huge. The HV has about 3 Mpix with 1/2.7-inch sensor, the HMC has 9 Mpix on a 1/4-inch sensor, and the difference in resolution is only about 50 lines. Is it worth to have so many pixels, especially in this price class?

Also, is there a formula to calculate comparative light sensitivity of a single-chip system vs. multi-chip? Does a 3-chip system deliver more light, or is it actually less light because of losses in the prizm?

 

[Barry_Green]

Does one big pixel produce a better (whatever "better" is, I think the major criteria being less noisy) than several small pixels grouped together having size of one big pixel?

Yes, because the inter-pixel dead space (the "registers") aren't there with the one-big-pixel model.

 

[BobDiaz]

The size of each pixel does play a part in the light sensitivity. In general, bigger pixels = more light gathered. I like to use the example of going out in the rain with a tea cup and a big bucket. Both will collect rain drops, but the larger bucket will collect more rain drops.

Now we know that the HMC150 has a larger chip and the pixel count is lower than the pixel count for the HMC40. No surprise that the HMC150 has bigger pixels, thus appears to be about 3 F-Stops more sensitive to light than the HMC40. IF we ignore the other factors and put 1920x1080 1/3" CMOS chips in the HMC150, I'd expect the modified camera to be about 2 F-Stops less sensitive to light.

In as much as we'd love to have the pixels almost touching, we do need to have some dead space around the pixels for support circuits. This forces us to have smaller pixels. Now the use of micro-lenses does help to focus or direct light that would land in the dead areas toward the active pixels. Thus, take the exact same sensor with and without micro-lenses and the chip with the micro-lenses performs better in lower light.

The choice of how the chip is made plays a role in how sensitive it is to light. In the past, with all other things being equal, CMOS was always less sensitive to light than CCD. However, changes to the technology have reduced the difference and today I'm not sure how much the difference really is today. Still, the way the chip is made, even CCD vs. CCD, can play a role in light sensitivity.

The chips are manufactured in extreme clean rooms. Dust and any dirt landing on a single chip could render the chip useless. To make this easy to understand, assume we have a work area of silicon that's 10" by 10". You can put as many or as few chips in this area. We will also assume that 25 dust particles will land on this area when we make the chip.

If I try to make the chip 10" by 10", it will be rendered 100% useless by all 25 dust particles landing on it. Make the chips 2" x 2" and we'd get 25 chips (5 x 5), but most of the chips would be rendered useless by the dust. A few would survive, because some chips might have 2 or more particles on them. Reduce the chip size to 1" x 1" and we get 100 (10 x 10) chips. Even if every damaged chip had only one dust particle, of the 100, 25 would be useless and the other 75 will be OK. Now we're starting to see a higher yield. Drop the size to 1/2" x 1/2" and we'll see a 20 x 20 layout with 400 chips. Thus, 400 - 25 (worse case dead chips) = 375 good chips.

The best test is to compare the cameras in low light rather than try to calculate it on paper. However, how we set the cameras can alter the results. For example, select Cine Like V or Black Press and the HMC40 is about 1/2 a stop less sensitive to light.


Bob Diaz

 

[hazydave]

The size of the chip is irrelevant. But the size of the pixels sort of has a bit to do with the light sensitivity, but only compared with similar chips. The actual answer is a bit more involved.

Actually, the size of the sensor is a good indicator. The real answer is that the size of the pixel is one key element. Larger sensor, larger pixels. Higher definition, smaller pixels. This is why all HD cameras are worse low-light performs than their SD counterparts. Also why DSLRs are such stellar low-light cameras, in video mode.

So, first thing, comparing a CMOS chip (in the HMC40) to a CCD chip (in the HMC150) is not a very good comparison because of how the electronics are designed. Essentially, because of the onboard electronics in the CMOS chip, you are going to inherently have more electronic noise, which will, in turn reduce the signal to noise ratio and reduce the overall light sensitivity compared to a CCD chip with similar pixel dimensions. A CCD can be more sensitive than a CMOS and even the comparative pixel dimensions may be irrelevant within a range.

Most of those things were once true, but that's also not the whole story. The problem is that noise isn't a single event, its noise integrated over time. In a CMOS chip, you have a distributed grid of pixels, but also a distributed grid of ADCs (analog to digital converters). So sampled pixels don't stay analog very long.

On a CCD, rather, you have something akin to an electronic bucket brigade -- you take the sample, then pass each pixel from one to the next, and finally to a very small number of ADCs. There's far more potential for noise aggregation, even if the noise started out at lower levels. On the other hand, the CCDs has fewer support electronics in place, so there's more potential room for the actual photodiode area, though this is usually dealt with on CMOS chips by using microlenses to concentrate light on the photodiode in each cell.

This phenomena has been seen in the evolution of Digital SLRs. The old CCD models delivered nice, clean images up to ISO400 or so, with 12-bits per pixel in RAW mode. Today, all modern DSLRs use CMOS sensors, deliver 14-bits per pixel in RAW more (or perhaps more on the higher end), and deliver amazingly solid images at ISO3200 or ISO6400. CCDs have been a mature technology, CMOS sensors have been improving by leaps and bounds over the last decade.

Another issue is fixed pattern noise... this is dark current noise (each photodiode will have a slightly different "dark" level), variations in initial gain (a transistor is coupled directly to a photodiode, in order to amplifiy the level immediately), variations in ADC, etc. In a modern CMOS designs, they actually employ a method called correlated double sampling. Basically, for each image sampled, there's a sample of the image array taken immediately after its reset. That contains the visible image of the non-random variations in the imager right at that moment (these things can change with temperature and voltage), and this sample is subtracted from the exposed sample, nulling out the fixed pattern noise. The FPN might be lower in a CCD, and mitigated by using fewer ADCs, but you can't zero it out in the same way. This is another reason modern CMOS sensors are quieter.

The pixel size is important for many reasons, but in the case of light sensitivity, it is a factor in the well depth of a pixel (i.e., how many photons can a pixel potentially read at a given point in time). A larger pixel can have a potentially larger well depth (emphasis on "potentially").

But there are other factors go hand in hand to define the overall light sensitivity. The combined factors are the chip's characteristics in pixel well depth, signal to noise ratio, dark current, and quantum efficiency. These factors together will give an accurate picture of how well the chip (and ultimately, the camera) does in low light.

And even fairly basic things, like how large is the pixel, really. There is space between pixels, some designs use microlenses to eliminate the effect of space, some control pixel shape or interleave rows with offset, etc.

I did the math in another post, estimating the pixel size (for image scale reasons, not light sensitivity) for the HMC40 sensors, since I couldn't find any published information (I put the math here again at the bottom of this post). It basically works out that each pixel is about 3.5 microns across (that may be the average dimension, since they could be rectangular instead of square). That is not just small, that is tiny for imaging chips (ultra sensitive chips may have pixels as large as 25-40 microns, by comparison).

Hmmm... the Canon 5D Mark II, with a sensor larger than that of the Red Camera, has a 6.4µm pixel pitch. Ok, it's 21 Mpixels native, they're bucketed when doing video. Still, if we're talking about 3.5µm square for the HMC40, x3 of course, that actually doens't sound quite so horrible.

Does this work out? A 1/4" sensor at 4:3 will have an area of about 9.72mm^2. On the HMC40, this contains about 3,050,000 pixels. Thus, each has a maximum area of 3.19µm^2. I think you started with the wrong sensor size.

The kicker, though, that brings this back is the CMOS architecture. This essentially means that dark current is higher (that is basically the inherent electronic noise in the pixel when the cap is on), the quantum efficiency is lower (this is a percentage of photons detected to those received), and the signal to noise ratio is lower than an identical pixel on a CCD chip.

Except, CMOS chips are actually quieter these days, for the reasons stated... particularly their ability to cancel out the FPN noise issues. This is why the gain can be cranked up much more on modern CMOS chips than on CCDs, without undue noise.

When you kick up the gain on a camera, you are going to be boosting the overall data from the chip, which is going to include noise, which why noise appears.

Noise is a low-level random signal on top of the analog data captured in each sensor. The gain is nothing more or less than a variable gain amplifier sitting directly in front of each ADC. When you crank up the gain, you amplify the analog signal more. What you'd like to have happen is that the level of noise is so low it's below the threshold of the ADC entirely... it has no effect on the output. As you crank the gain, the noise will be amplified along with the signal, eventually rising high enough to sometimes change the output of the ADC, versus the non-noise situation.

Here is the estimated calculation, by the way, of the pixel size in the HMC40:

Each sensor has ~3.05 million pixels on a ~1/4.1" square (I love converting metric to english and back!).

You're off here a little, simply because a 1/4.1" chip isn't 1/4.1" square. These standard measures go way back to the days of the vidicon tube. A 1/4.1" chip is any rectangle rectangle (usuaully 4:3, as it is in the case of the HMC40, but it doesn't have to be) that exactly fits within a 1/4.1" diameter circle. That would have been the circle of the vidicon tube, back in the day. Unless you have a square sensor, the actual diagonal is a bit less than the specified size of the sensor. It's a stupid convention, but it's been the standard for generations.