5 ms·

Always wished pixels were hexagonal. Yeah the math would be a bit harder and vertical lines would be fuzzier at low resolution, but still.... https://www.quo

Always wished pixels were hexagonal. Yeah the math would be a bit harder and vertical lines would be fuzzier at low resolution, but still.... https://www.quora.com/Why-are-computer-pixels-square https://www.quora.com/Why-are-computer-pixels-square

tzs 9y ago
I wonder how hard hexagonal pixel layout would have been to do on a CRT? Two ways come to mind. If the basic scan line is kept horizontal, then a small vertical modulation on each scan line could result in a hexagonal layout if you timed it right. Alternatively, if the grid is tilted (so that the basic scan line is diagonal instead of horizontal), then you just have to offset the odd scan lines by half a pixel from the even scan lines. I think we are just lucky that we ended up with rectangular screens. Imagine if they had went with circular screens. The scan might then easily have been a spiral starting in the center. That's fairly easy to do. It's just a matter of driving the horizontal and vertical deflectors with sine waves with the right phase difference, with a saw tooth amplitude. From a television point of view, I don't think it really matters which of these you use, as long as the cameras and the displays use the same scan pattern. From a computer point of view, though, it would have been a lot more painful if CRTs used a spiral scan. For most graphics applications we'd still need to manipulate rectangular areas, and that would be quite annoying in a coordinate system based on a spiral scan.
- robbrown451 9y ago
  Right. At least a hex grid is still essentially a 2-dimensional array. In most ways it would be identical to a rectangular grid, but the algorithm for sampling surrounding pixels would be slightly different because of the spacial arrangements.
- lightedman 9y ago
  Well, the problem there is that CRTs used a triangular-patterned phosphor screen, excepting Trinitrons which had their phosphors lined up in a row. How would you handle the missing spot for a phosphor in a hexagonal arrangement? I guess with current tech, we could make ultra-fine phosphor groups that we'd not be able to see. an 8K CRT at 32" would be quite nice .
  - robbrown451 9y ago
    Well to be honest, I'm thinking more of how pixels as data are stored in images rather than implemented in hardware. But still, if you follow the link, its something I posted on Quora 5 years ago, and shows an image I pulled from somewhere of a LED or OLED or whatever device where the hex arrangement is already there. It makes a lot of sense. In hardware, though, it makes the most sense for R, G and B components to not share a "center point" for each pixel, instead having their own center point that is spacially offset.
  - Stratoscope 9y ago
    There was no correlation between the phosphor dots on a typical color CRT and "pixels" as we think of them. It wasn't like the way we use an LCD or OLED display at all. A color CRT had no native resolution: display pixels were not locked onto specific phosphor dots. Consider all the analog adjustments a CRT offered: you could tweak the overall height and width of the displayed image and nudge it up or down and left or right. A high end CRT would have additional controls to adjust the shape of the image to correct for pincushion or barrel distortion. You could also drive the CRT with different display resolutions. Obviously the phosphor dots didn't move around when you did this. Even on a Trinitron display there was no connection between logical pixels and the aperture grill spacing. A good analogy for today's displays would be an LCD/OLED display that you can't run in native resolution, and can't even discover what its native resolution might be: the pixels you generate in software are not directed to specific physical points of light on the screen. A monochrome CRT came much closer to having something that today we would recognize as "pixels", because there was no shadow mask or phosphor dots. Whatever problems might have stood in the way of using a hexagonal pixel layout on a color CRT, the phosphor dot or stripe layout wasn't among them. Edit/meta: I really wish people would not downvote comments like lightedman's parent comment, which may have been wrong on the facts but provided an opportunity for me to jump in with some hopefully interesting information that not everyone may have known about CRT technology. Some of the best conversations I've had have been where I've had a misconception about something and someone was kind enough to set me straight on it. Yes, yes, I know, we're not supposed to complain about downvotes. So if my complaint bothers you, here's my offer: downvote this comment and upvote lightedman's parent comment, which received some downvotes that I think were undeserved. Fair deal?
- mjevans 9y ago
  Analog circuits aren't my area of expertise. However it seems likely that the complexity of a stable and identical spiral scan circuit is either extremely difficult or possibly improbably expensive. Compare the above to what we have in reality: it seems that there were some vacuum tube tricks that made a grid system much easier. https://en.wikipedia.org/wiki/NTSC#Technical_details https://en.wikipedia.org/wiki/NTSC#Technical_details
  - mimhoff 9y ago
    A spiral scan circuit would just be something like: x = r cos (theta) y = r sin (theta) when you vary r so it gives you a spiral instead of a circle. So it shouldn't be too hard to generate the signals required. Or there could be a different arrangement of the beam deflectors that didn't give you independent control of X and y.
    - nwatson 9y ago
      But getting equal time on each pixel as you spiral out would be crucial to trigger the right phosphorescent equal luminance from your light-emitting material. Theta would not be linear with time.
      - Jyaif 9y ago
        The intensity of the beam could be modulated to counter this.
    - gugagore 9y ago
      There are sync pulses on every line of analog video, as well as to color burst in most analog video standards. You couldn't just sweep theta and r continuously for a whole frame.
      - Shikadi 9y ago
        Why not? The sync pulses are because the scan is discontinuous, you could just have one plus color burst every spiral
taeric 9y ago
There is a section in the manual for METAFONT that explains why a certain shaped pen is actually superior to a simple circle. I've always been curious if this was explored more for actual screen rastorization.
userbinator 9y ago
and vertical lines would be fuzzier at low resolution Horizontal ones too. Non-square pixels have been around for a long time in digital camera LCDs and more recently the https://en.wikipedia.org/wiki/PenTile_matrix_family https://en.wikipedia.org/wiki/PenTile_matrix_family but while they're fine for photos and other gradient-like images, text and lineart have a noticeable "grain" on them.
- dogma1138 9y ago
  Most lcd screens today use non-square pixels, OLED displays use non-equal sizes for different colors and o don't see anyone complaining about sharpness.
jacobolus 9y ago
At this point, displays have fine enough resolution and GPUs are fast enough, that if you want to make hexagon-pixel images you can go right ahead and it won’t make too much difference (except you’ll reduce moiré artifacts), as long as you’re willing to implement your own code to rasterize the image to a square grid for display. It would be neat if camera sensors and displays would switch to hexagonal grids of pixels – considering most images get resampled right before display now anyway, it should be all upside (except for a bit of extra implementation hassle). A hexagonal grid is nice for several reasons: it can easily handle refinement to 3 or 4 subpixels per pixel, while a square grid needs 4 subpixels to keep its proper grid; it is much more isotropic than a square grid (straight lines at a variety of angles look much better); it is notably more efficient at covering the plane; hexagonal filters have a much nicer 2-dimensional frequency response; dithering works quite a bit better on a hexagonal grid; etc. There are some printers that use a hexagonal grid, and some hexagon-pixel cameras used for stuff like medical imaging or astronomy.

tzs 9y ago

I wonder how hard hexagonal pixel layout would have been to do on a CRT? Two ways come to mind. If the basic scan line is kept horizontal, then a small vertical modulation on each scan line could result in a hexagonal layout if you timed it right. Alternatively, if the grid is tilted (so that the basic scan line is diagonal instead of horizontal), then you just have to offset the odd scan lines by half a pixel from the even scan lines. I think we are just lucky that we ended up with rectangular screens. Imagine if they had went with circular screens. The scan might then easily have been a spiral starting in the center. That's fairly easy to do. It's just a matter of driving the horizontal and vertical deflectors with sine waves with the right phase difference, with a saw tooth amplitude. From a television point of view, I don't think it really matters which of these you use, as long as the cameras and the displays use the same scan pattern. From a computer point of view, though, it would have been a lot more painful if CRTs used a spiral scan. For most graphics applications we'd still need to manipulate rectangular areas, and that would be quite annoying in a coordinate system based on a spiral scan.

robbrown451 9y ago
Right. At least a hex grid is still essentially a 2-dimensional array. In most ways it would be identical to a rectangular grid, but the algorithm for sampling surrounding pixels would be slightly different because of the spacial arrangements.
lightedman 9y ago
Well, the problem there is that CRTs used a triangular-patterned phosphor screen, excepting Trinitrons which had their phosphors lined up in a row. How would you handle the missing spot for a phosphor in a hexagonal arrangement? I guess with current tech, we could make ultra-fine phosphor groups that we'd not be able to see. an 8K CRT at 32" would be quite nice .
- robbrown451 9y ago
  Well to be honest, I'm thinking more of how pixels as data are stored in images rather than implemented in hardware. But still, if you follow the link, its something I posted on Quora 5 years ago, and shows an image I pulled from somewhere of a LED or OLED or whatever device where the hex arrangement is already there. It makes a lot of sense. In hardware, though, it makes the most sense for R, G and B components to not share a "center point" for each pixel, instead having their own center point that is spacially offset.
- Stratoscope 9y ago
  There was no correlation between the phosphor dots on a typical color CRT and "pixels" as we think of them. It wasn't like the way we use an LCD or OLED display at all. A color CRT had no native resolution: display pixels were not locked onto specific phosphor dots. Consider all the analog adjustments a CRT offered: you could tweak the overall height and width of the displayed image and nudge it up or down and left or right. A high end CRT would have additional controls to adjust the shape of the image to correct for pincushion or barrel distortion. You could also drive the CRT with different display resolutions. Obviously the phosphor dots didn't move around when you did this. Even on a Trinitron display there was no connection between logical pixels and the aperture grill spacing. A good analogy for today's displays would be an LCD/OLED display that you can't run in native resolution, and can't even discover what its native resolution might be: the pixels you generate in software are not directed to specific physical points of light on the screen. A monochrome CRT came much closer to having something that today we would recognize as "pixels", because there was no shadow mask or phosphor dots. Whatever problems might have stood in the way of using a hexagonal pixel layout on a color CRT, the phosphor dot or stripe layout wasn't among them. Edit/meta: I really wish people would not downvote comments like lightedman's parent comment, which may have been wrong on the facts but provided an opportunity for me to jump in with some hopefully interesting information that not everyone may have known about CRT technology. Some of the best conversations I've had have been where I've had a misconception about something and someone was kind enough to set me straight on it. Yes, yes, I know, we're not supposed to complain about downvotes. So if my complaint bothers you, here's my offer: downvote this comment and upvote lightedman's parent comment, which received some downvotes that I think were undeserved. Fair deal?
mjevans 9y ago
Analog circuits aren't my area of expertise. However it seems likely that the complexity of a stable and identical spiral scan circuit is either extremely difficult or possibly improbably expensive. Compare the above to what we have in reality: it seems that there were some vacuum tube tricks that made a grid system much easier. https://en.wikipedia.org/wiki/NTSC#Technical_details https://en.wikipedia.org/wiki/NTSC#Technical_details
- mimhoff 9y ago
  A spiral scan circuit would just be something like: x = r cos (theta) y = r sin (theta) when you vary r so it gives you a spiral instead of a circle. So it shouldn't be too hard to generate the signals required. Or there could be a different arrangement of the beam deflectors that didn't give you independent control of X and y.
  - nwatson 9y ago
    But getting equal time on each pixel as you spiral out would be crucial to trigger the right phosphorescent equal luminance from your light-emitting material. Theta would not be linear with time.
    - Jyaif 9y ago
      The intensity of the beam could be modulated to counter this.
  - gugagore 9y ago
    There are sync pulses on every line of analog video, as well as to color burst in most analog video standards. You couldn't just sweep theta and r continuously for a whole frame.
    - Shikadi 9y ago
      Why not? The sync pulses are because the scan is discontinuous, you could just have one plus color burst every spiral

taeric 9y ago

There is a section in the manual for METAFONT that explains why a certain shaped pen is actually superior to a simple circle. I've always been curious if this was explored more for actual screen rastorization.

userbinator 9y ago

and vertical lines would be fuzzier at low resolution Horizontal ones too. Non-square pixels have been around for a long time in digital camera LCDs and more recently the https://en.wikipedia.org/wiki/PenTile_matrix_family https://en.wikipedia.org/wiki/PenTile_matrix_family but while they're fine for photos and other gradient-like images, text and lineart have a noticeable "grain" on them.

dogma1138 9y ago
Most lcd screens today use non-square pixels, OLED displays use non-equal sizes for different colors and o don't see anyone complaining about sharpness.

jacobolus 9y ago

At this point, displays have fine enough resolution and GPUs are fast enough, that if you want to make hexagon-pixel images you can go right ahead and it won’t make too much difference (except you’ll reduce moiré artifacts), as long as you’re willing to implement your own code to rasterize the image to a square grid for display. It would be neat if camera sensors and displays would switch to hexagonal grids of pixels – considering most images get resampled right before display now anyway, it should be all upside (except for a bit of extra implementation hassle). A hexagonal grid is nice for several reasons: it can easily handle refinement to 3 or 4 subpixels per pixel, while a square grid needs 4 subpixels to keep its proper grid; it is much more isotropic than a square grid (straight lines at a variety of angles look much better); it is notably more efficient at covering the plane; hexagonal filters have a much nicer 2-dimensional frequency response; dithering works quite a bit better on a hexagonal grid; etc. There are some printers that use a hexagonal grid, and some hexagon-pixel cameras used for stuff like medical imaging or astronomy.