spain 18db streetfootage

I don't know how I missed this post but... wow. That is some SUPER clean footage. Literally looks cleaner at 18db than most cameras I've seen at 0db. Even though this footage is ... well garbage, it shows you that the image coming off that sensor is just amazing. This sensor is as if there was no such thing as noise to begin with. It's as if the problem of noise doesn't exists.

Simply lovely.
 
Anyone have any ideas as too how their pulling this off. Is it codec, the sensor, some kind of noise reuction? maybe a combo of all. Just curious if any of you technical folks out there might have an idea of how they can get this quality at only 13,000
 
maranfilms said:
Anyone have any ideas as too how their pulling this off. Is it codec, the sensor, some kind of noise reuction? maybe a combo of all. Just curious if any of you technical folks out there might have an idea of how they can get this quality at only 13,000
Click to expand...

it's the fact they are using a very large sensor, except unlike the dslrs and such, its not 3000x2800 pixels or whatever, its native 1920x1080. so the photosites (pixels) are HUGEE, allowing each pixel to take in much more light, aka put out much more voltage. higher voltage, harder for very small environmental and circuitry noise to poke its head through. Has nothing to do with any codec, and there is no noise reduction in these cameras. the sensor is just obliterating the 0.005 volts of circuitry noise into nothing by outputting 20 jiggawatts with half a stop of light hitting it
 
Fohdeesha said:
it's the fact they are using a very large sensor, except unlike the dslrs and such, its not 3000x2800 pixels or whatever, its native 1920x1080. so the photosites (pixels) are HUGEE, allowing each pixel to take in much more light, aka put out much more voltage. higher voltage, harder for very small environmental and circuitry noise to poke its head through. Has nothing to do with any codec, and there is no noise reduction in these cameras. the sensor is just obliterating the 0.005 volts of circuitry noise into nothing by outputting 20 jiggawatts with half a stop of light hitting it
Click to expand...

I hate to break it to you but thats completely WRONG haha. The sensor is a 3.5k sensor (about the same as Arri Alexa) The photosites are not huge, they are very small. The 1920x1080 image is due to internal debayering to get the highest image quality, that is NOT the resolution of the sensor, its much higher. The sensor is spesificly desgined to produce beautiful images with low noise level. I don't know the details yet of the sensor (they didn't wanna talk about it much at the CineAlta meeting) but I'm sure I could get you spec info from "Sony Joe" soon.

The 3.5k sensor is what makes this camera a mini Alexa, its basically like shooting with RED at 4k and debayering to get a clean and perfect 1080p image. Only thing is with RED you have to debayer in post to get your final image, with the F3/Alexa the camera does this internally and give you the end result right away.

It's still a mystery on how Sony did this with the new chip, its not the photosite size, because they are just as tiny as the dslrs. Its the sensor, it has nothing to do with codec. It could also be some internal noise reduction but until I ask, we wont know... so another question to ask Sony!
 
My understanding is that the sensor is higher resolution than 1920x1080 so that it can have full raster resolution despite the debayer filter. But still, a larger sensor allows for larger photo sites. That's what the Alexa does too.

edit: GP beat me to it.
 
Fohdeesha said:
it's the fact they are using a very large sensor, except unlike the dslrs and such, its not 3000x2800 pixels or whatever, its native 1920x1080. so the photosites (pixels) are HUGEE, allowing each pixel to take in much more light, aka put out much more voltage. higher voltage, harder for very small environmental and circuitry noise to poke its head through. Has nothing to do with any codec, and there is no noise reduction in these cameras. the sensor is just obliterating the 0.005 volts of circuitry noise into nothing by outputting 20 jiggawatts with half a stop of light hitting it
Click to expand...

Over sampling number of photosites to number of output-pixels (which is typically 1080x1920) is necessary to do a good interpolation of color out of a Bayer sensor. Having only 1920x1080 photosites would not be enough to get 4:2:2 or 4:4:4 color sampling. So to get good color HD output you need to over sample photosites relative to output-pixels, just enough to get proper interpolation of color, but not too much so you can avoid severe line skipping and pixel binning (which is the problem with DSLRs).

As an example, Arri's Alev III sensor in the Alexa has a real resolution of 3392 x 2200 pixels (7,5 million). Sony F3 3.5K sensor has something similar to this. This 3.6x increase in the numer of photosites relative to output-pixels seems to be the sweet point for a Bayer s35mm sensor when you want to get full color sampling in HD video (1080 x 1920).
 
Giuseppe Pugliese said:
It's still a mystery on how Sony did this with the new chip, its not the photosite size, because they are just as tiny as the dslrs. Its the sensor, it has nothing to do with codec. It could also be some internal noise reduction but until I ask, we wont know... so another question to ask Sony!
Click to expand...

Arri's Alev III sensor in the Alexa has a real resolution of 3392 x 2200 pixels (7,5 million). Sony F3 3.5K sensor has something similar to this. This 3.6x increase in the numer of photosites relative to output-pixels seems to be the sweet point for a Bayer s35mm sensor when you want to get RGB color sampling in HD video (1080 x 1920).

DSLRs on the other hand have typically more than twice the number of photosites than Alexa and F3 for a similar area. GH2 has 18 million photosites, GH1 14 million, Canon 7D 19 million (all of them packed in area very close to s35mm). Canon 5D has the photosites packed in a Full-Frame sensor which is significant larger area than s35mm, but because they are 22 million pixels, their absolute size is still significantly smaller in Canon 5D than in Alexa or F3.

Sony Japan say their s35mm Exmor CMOS in the F3 has pixels which are 4x the size of typical DSLR CMOS ones. Approximately 12 microns sized versus as little as 4 microns on APS-C DSLRs.

some pixel pitch in microns (approximate width of one pixel):

Canon 7D: 4.3 microns
Nikon D300S: 5.5 microns
Nikon D3X: 5.9 microns
Sony A900: 5.9 microns
Canon 5DII and Canon 1DsMKIII: 6.4 microns
Nikon D3S and Nikon D700: 8.5 microns

RED One & RED M-X sensor: 5.4 microns

Sony PMW F3: 12 microns

So it seems to me that pixel size -and its direct effect of increasing signal to noise ratio- still plays an important role in the kind of noise-free image quality that you can get from the Alexa and the F3 sensor.
 
12 micron size photosites can be the reason why, but I don't think thats the whole picture. Noise can still be introduced in the A/D conversion, I'm just glad they did such an amazing job.

The mystery to me is not how they get the clean image off the sensor at 0dB but more so how the heck they did it at 18dB. The sensor is already rated at iso800 at 0dB... so the engineering behind getting such a clean image off of 18dB is just mind blowing. The photosites surely are not picking up that much light, there is electronic gain being added at that point which means they are doing an excellent job of handling their voltages.
 
you guys sure? I remember seeing a bunch of press releases saying that the photosites were much larger due to not downsampling so much. when I said native 1920x1080, I meant native after debayer, so yes more than that many photosites, but still not downsampling from 4k.

hmmm

lexicon said:
Sony Japan say their s35mm Exmor CMOS in the F3 has pixels which are 4x the size of typical DSLR CMOS ones. Approximately 12 microns sized versus as little as 4 microns on APS-C DSLRs.
Click to expand...


yeah there we go. On the scale of surface area to output voltage, that's a lot of more jiggawatts


Giuseppe Pugliese said:
12 micron size photosites can be the reason why, but I don't think thats the whole picture. Noise can still be introduced in the A/D conversion, I'm just glad they did such an amazing job.
Click to expand...

I think it's mostly the whole picture. Noise is ALWAYS introduced in the a/d stage, and more importantly the signal path to the a/d converter. let's say any average camera has around 0.5 volts of noise picked on the way to the a/d and then after, when smaller photosites are only putting out a volt, that half a volt of noise is very noticeable. When huge photosites are putting out 4 volts, the 0.5 becomes much, much smaller in ratio. hence it's absurd SNR of 64db. All those numbers are completely made up and arbitrary, but you know what I mean? It's like when you run line level audio cables you almost never worry about noise, but when running phono cables form a turntables very very low voltage output, suddenly the tiny amount of ground and environmental noise is very noticeable.
 
Last edited:
After some perusing around the internets, it seems the f3 is using column-parallel ADC's

http://www.freepatentsonline.com/6137432.html

besides the obvious lower power consumption (less voltage going to the a/d, less crosstalk and noise possible), it almost sounds like they are running balanced, using reference pixel data to compare to its first pixel input, and canceling any difference. Sounds like balanced audio to me almost, except for imaging sensors! Or perhaps only the clock input is redundant, I have some reading to do. Now, what to sell to pick an f3 up....
 
Fohdeesha said:
you guys sure? I remember seeing a bunch of press releases saying that the photosites were much larger due to not downsampling so much. when I said native 1920x1080, I meant native after debayer, so yes more than that many photosites, but still not downsampling from 4k.

hmmm
Click to expand...

The 3.43k sensor resolution of the F3 comes from Sony's Bill Drummond via an interview by Philip Johnston in this thread:

http://www.dvxuser.com/V6/showthread.php?229329-Answers-to-your-questions-on-the-new-Sony-F3

But there are more fundamental reasons why a ~3.5k resolution makes plain technical sense. As you know, a bayer filter sensor cannot do true 4:4:4. A bayer filter sensor only has one third of the required information and two thirds are missing. The missing color information is determined by interpolation.

Arri, and Sony, have come to a similar conclusion, that a sensor's ~3.5K pixel count is ideal balance between optimum image sharpness, sensitivity and latitude when downscaled to1920 X 1080 and 2K DI workflows. Its not a 100% RGB resolution, but it gets close. Increasing resolution further would mean smaller photosites with diminishing returns in RGB resolution.

Fohdeesha said:
After some perusing around the internets, it seems the f3 is using column-parallel ADC's

http://www.freepatentsonline.com/6137432.html

besides the obvious lower power consumption (less voltage going to the a/d, less crosstalk and noise possible), it almost sounds like they are running balanced, using reference pixel data to compare to its first pixel input, and canceling any difference. Sounds like balanced audio to me almost, except for imaging sensors! Or perhaps only the clock input is redundant, I have some reading to do. Now, what to sell to pick an f3 up....
Click to expand...

These colomn-based ADC is interesting stuff, and are actually used in all modern EXMOR sensors. We are talking about thousands of ADC converters on the sensor, one for each column. The sensor also does correlated double sampling (CDS), basically samples a pixel more than once to reduce noise. The same CDS technology can also be applied to increase dynamic range (probably the technique Red is using for their HDRx).
 
It is a combination of all of these, pixel size, parallel adc etc.... The latest still sony still sensor in the Nikon D7000 and pentax K5 is delivering exceptional low light and DR with a 16 megapixel apsc sensor. The DR is in the 14 stop range, much higher by 2 stop than the Canons. They call it the iso less sensor, you can shoot at base iso and change the image with curve in post. You can see point black photos that can reveal good images from the shadows, it is astonishing. Sony has been working a lot on its exmor technology and more and more you are seing the result of it. Now this is a chip which has twice the number of pixel than the F3 one!!!
 
You must log in or register to reply here.
Back
Top