"The IMX224LQR supports a DOL (digital overlap)-type WDR function. This function uses a method that outputs the data for three frames with different storage times line by line instead of frame by frame, enabling improvement of picture quality especially under low illumination compared to the existing multiple exposure WDR function." The sensors feature a nice set of interfaces: LVDS, 4-lane MIPI CSI-2 with 594 Mbps/lane, and CMOS parallel.
Sony shows a higher resolution image shot at 0.05 lux (10x of the stated minimum) with F1.4 lens at 30fps and internal gain of 60dB (click to expand to 799x597 pix image):
Another recent announcement is system LSI for entry-level IP security cameras, the CXD4145GG, supporting ISP and H.264 encoding at 30fps up to 3MP, or 15fps for 5MP:
72 dB is a gain of 4000 and 60dB is a gain of 1000. Although this is not clear what is the conversion gain and the full well capacitance for unity gain, probably the FWC will be around 10000 electrons or in this order. If a gain of 4000 is applied, the resulting FWC will be of only 2.5 electrons, which produce a SNR maximum of sqrt(2.5)=1.6. This is assuming that the shot noise is the only noise source, which is not true.
ReplyDeleteIn consequence I think that this is very good pixel, but no as good as seems. The obtained image quality is of a very bad quality. Only the average effect when the image is represented at a lower resolution, produce the visual impression that the image quality is reasonable.
I guess they have a variable-gain amplifier. However, I wonder how they solve the column fixed partern noise problem for such a high gain amplifier.
DeleteI guess they have variable gain amplifiers. However, I wonder how they can calibrate the column FPN for such a high gain amplifier.
DeleteIt's YOUR problem. This has been resolved for most of the sensor makers now.
DeleteI think that the gain is mainly digital. Hasn't too much sense to use an analogue gain as high as 60 or 72dB.
DeleteIn addition, the analog gain cannot be controlled as fine as the digital one. This will produce artifacts as VFPN gain and non linearity.
Thank you Rafa, I totally agree.
DeleteBTW, could anyone of you please tell me why gain-adaptive amplifiers (whose gains are adapted to the input signal strength, so the readout noise is shot-noise shaped to achieve higher DR) are not so widely used as variable-(fixed)gain amplifiers (ISO), which is only good for all-dark or all-bright scenes?
Probably speed
DeleteActually, gain-adaptive amplifiers have the same speed with high gain ISO mode. Only a small extra time is deditated to the comparison. Someones told me that is the contrast issue when the dark parts look brighter. However, I am not very convinced.
DeleteBut the problem isn't if the gain amplifier can be implemented or not. The fundamental problem is that, even with ideal amplifiers (noise free) the resulting maximum signal to noise ratio when a huge gain is applied is very small. Then each pixel has almost nothing of information. And only a lot of close pixel has the enough information to represent an useful signal. In summary, apply a ultra high gain is almost equivalent to reduce significantly the spatial resolution. In this sense, is preferable to implement binning as was traditionally performed in CCD sensores that reduce the resolution increasing the sensitivity in the same amount.
ReplyDeleteIn consequence. This is more marketing that a real improvement, although I say another time that this is a fantastic sensor.
Well, if you want to obtain a low-light image with a given resolution, you need to amplify. Something is better than nothing!
Delete