Time for a new sensor design? - techies chime in

[Zach Nelson]

Some of my background is in theatrical lighting so I'm down with dichroics and the RGB processing approach of color cameras, ie. the bayer pattern. I'm wondering why in 2008 we still "waste" the surface area of our sensors. It seems to me that we should have a better method for measuring light than the CCD/CMOS technology from the 60's.

What if:

Instead of appropriating a diode to measuring a single RGB value we developed a "smarter" diode that could measure light intensity (as an electrical signal or resistance) for a given wavelength being sampled?

In other words, in a single chip arrangement you're assigning a diode (pixel) to measure a single RGB value and then in the case of the bayer pattern you process the result and create the final image. In the case of a 3-chip camera the entire surface of a chip is dedicated to a single color but it's 3 times as expensive and consumes space.

How cool would it be to have a circuit that could sample a diode's response to a given wavelength (measure intensity) but the wavelength was variable?

If you're familiar with the newer DLP projectors they have more than the simple RGB dichroic filters, they have intermediate wavelengths to help reproduce a higher gamut of colors. Inkjet printers work the same way - 6 colors instead of 3.

So in theory you could sample the entire surface of the sensor for 650nm (red), 510nm (green), and 475nm (blue) .. but also any wavelength stop along the spectrum from 400nm to 650nm. Maybe an RGB/CMY would produce a higher-fidelity image, who knows. If it takes .01ms to sample the surface for a wavelength then you're in good shape.

Maybe it's just a pipedream or maybe in 20 years we'll see something like this in use. I'm hardly a physicist so I don't know how diodes could be made wavelength-aware and specifically make the wavelength sensitivity controlled via a voltage.

You know, if diodes and circuits were fast enough - incredibly fast - they could, in theory, measure the zero-crossing of light as the wave energy hits the sensor and therefore calculate the wavelength and intensity. In this case you have a sensor that can measure all wavelengths and report the intensity at each simultaneously. But again this is just me dreaming...

 

[Cail Young]

You know, if diodes and circuits were fast enough - incredibly fast - they could, in theory, measure the zero-crossing of light as the wave energy hits the sensor and therefore calculate the wavelength and intensity. In this case you have a sensor that can measure all wavelengths and report the intensity at each simultaneously. But again this is just me dreaming...

Also it's beyond photoelectric sensors of any current design due to the quantum physics behind photons; whilst light can have a measurable wavelength it isn't strictly speaking a pure energy wave with 'zero crossings', it's individual photons.

 

[jbeale]

Multispectral imaging is done now by various satellites, but as far as I know, it's done the "old fashioned" way with selectable absorption filters. You can make a spectrometer with a diffraction grating and scan it to form an image across specific wavelengths. The kind of tunable quantum color sensitivity you're talking about has not been done, to my knowledge.

It might be possible to do this in some manner, although I do not think it would be quite so simple as a fast diode. Maybe a synthetic molecule could be adjusted to absorb light in adjustable bands. I think that's something like how the eye works, using the pigment molecules photopsin I, II and III for red, green, and blue color vision.

 

[Dylan Reeve]

I did wonder a while ago about something like a reverse DLP arrangement.

Don't know enough about it though.

 

[Zach Nelson]

I did wonder a while ago about something like a reverse DLP arrangement.

Don't know enough about it though.

I thought about the same thing. You could use a single chip camera but rotate a disc of dichroics (RGB) in front. So you know which filter is in front of the sensor and you read the chip output and count it as "red" and likewise for green and blue.. hell you could even sample the chip for RGB/CMY if that would produce a better final image (I suspect it would).

Only problem with that is now you have a rotating disc in your camera which would be prone to mechnical problems/burning out.. etc. You'd also effectively cut your frame rate because several physical frames (3 for RGB or 6 for RGBCMY) would be used to form 1 logical frame. Aside from that, I think you'd have great output.

 

[Justin K Phillips]

It'd be alright for static scenes, but for anything with motion you'll get rainbow trails around moving objects.

 

[michael zaletel]

There are some excellent concepts presented in the following U.S. Patent Application filed on March 4, 2008. Perhaps some of this applies to the discussion in that many related patents are cited and there may already be someone that has come up with a solution.

Link to Patent Application Number 7,339,216 Vertical color filter sensor group array with full-resolution top layer and lower-resolution lower layer

Arrays of VCF sensor groups, such as some of those described in above-cited U.S. patent application Ser. Nos. 10/738,484 and 10/355,723, have important advantages over arrays of single-layer sensors (e.g., conventional arrays having a Bayer pattern, and arrays that implement conventional color-filter-mosaic technologies), including the advantages of measuring luminance at every pixel sensor location and measuring color in a way that does not induce color aliasing artifacts. The additional advantage of measuring color at every pixel sensor location to achieve good chroma resolution is less important and can be sacrificed by implementing one layer (e.g., the green, red, or blue layer) of an array of VCF sensor groups to be readable with (to "have") full resolution and at least one other layer of the array to have lower resolution (e.g., as described in above-cited U.S. patent application Ser. Nos. 10/738,484, and 10/355,723).