3 ms·

At a previous job, we had informal "teardown Tuesdays" where someone would bring in a device, and we'd take it apart over lunch to see how it worked. Sometimes

by ohazi 5y ago

At a previous job, we had informal "teardown Tuesdays" where someone would bring in a device, and we'd take it apart over lunch to see how it worked. Sometimes it was a rare/expensive/new device and we had to promise to at least attempt to get it back together afterward, but this was the exception. Everybody should try this... It's incredibly fun and informative. Two particularly good ones that I remember were the VCR and the SLR camera. I brought in the camera -- I think I paid ~$25 for it on eBay, and I still have the shutter mechanism somewhere. Makes for a good fidget toy. This exercise really made me appreciate just how far you can get with good design and clever engineering. Even though I knew roughly how the camera worked and what its advantages were, we inevitably came across some assemblies that we couldn't immediately identify. Eventually someone would work out that it was part of the metering or autofocus system, and everyone had fun shining light through it to try and gain an intuitive understanding of how it worked. And you can absolutely do this! Most of those assemblies were clever because they were so devilishly simple. The autofocus, for example, redirects a small patch of light from a particular location onto a separate mirror or prism that can be placed out of the way. Some optics expand the light onto a 1D CCD or a pair of detectors, and then some rudimentary circuitry can be used to drive the lens motors and close the feedback loop. The modern techniques might be conceptually similar, but it's so cool to see the problem solved without a massive sensor and a DSP that's arguably a supercomputer -- Just some discrete components. It feels very: (1) I can't believe this actually works, (2) I can't believe they managed to assemble this tiny thing in a cost-effective way, and (3) you're trying to tell me this Rube-Goldberg machine was relatively reliable too!? I know our modern monolithic slabs of glass and aluminum and potted silicon and millions of lines of code can probably beat these old devices on almost any quality metric that anyone actually cares about, but there's something charming and magical about the way we used to make things.

toomanybeersies 5y ago
Phase detection autofocus (what you're describing) is still used in cameras today, and hasn't changed significantly since the 90's. DSLRs are essentially just a film camera, but with a CMOS sensor replacing film. One of the reasons why they still exist, despite the obvious inconvenience of not having a big screen to look at, is because until recently it wasn't possible to build a mirrorless camera with phase detection autofocus. I have a Canon EOS 55 from the late 90's, and in many situations it has better autofocus performance than my Fujifilm mirrorless from 2017.
- ohazi 5y ago
  Yeah, I remember when the first CMOS sensors with phase detection elements started showing up. This is what I meant by "conceptually similar" although I probably should have clarified, since we've had massive sensors without these elements for longer. I guess contrast-based autofocus is what really needs the DSP, since you need to do edge detection / FFTs on more patches of image for the technique to work well, as well as PSF analysis if you want to guess the lens defocus direction. With phase it's just a few cheap 1D correlations, and you get accurate lens direction information for free, which explains why it feels so much faster. Unfortunately we didn't have access to an electron microscope, so a modern mirrorless camera would have made for a rather less interesting teardown.
  - sudosysgen 5y ago
    Contrast based autofocus just doesn't compete either. New, higher end mirrorless cameras still use phase detection just like the DSLRs of old. Sony still uses old-style sensors but mounted directly on the chip, while Canon uses pixels that only accept light from some directions to calculate phase information. Both of these tactics end up with very similar results. Though it is augmented with DSP.
- sudosysgen 5y ago
  Mirrorless cameras with phase detection have actually existed for a long time - actually since 2010 - before they became practical with EVFs and better batteries. Something that really slowed it down is that Canon filed a patent for Dual Pixel AF which was the first on sensor pdaf system to outperform previous systems, and then proceeded to not make any serious mirrorless camera using it. All the worse is that DPAF was a fairly minor invention that other companies didn't do first mostly because they didn't have the manufacturing capabilities yet - basically using a microlens to split the light instead of a divider to obscure it. So Sony had to find ways to make it work without infringing on the patents and without artifacts.
  - toomanybeersies 5y ago
    As a disclaimer: I haven't bought a camera (or looked at new models) since 2018 > Mirrorless cameras with phase detection have actually existed for a long time - actually since 2010 I guess that depends on your timescale. The first SLR camera was 62 years ago, with the Nikon F (1959-1973, 14 years span). The first [SLR] camera with phase detection autofocus, the Canon EOS 650 (incidentally my first SLR [1]) was released in 34 years ago, in 1987. It wasn't until 2002 (Canon EOS-1Ds, 19 years ago) that there was a DSLR that could capture images on par with film. Even if Canon deliberately suppressed mirrorless PDAF (which I don't dispute), phase detection autofocus has been around much longer than digital cameras, and at least until recently DSLRs have had better autofocus performance than mirrorless. To be clear, I'm not cargo culting DSLRs here. My point is that beyond the change from film to CMOS sensor, DSLRs are not significantly different from the film cameras they replaced. [1] I mentioned to a coworker that I was thinking picking up photography; the next day she gave me a bunch of film photography gear that had been sitting in her attic for the past 20+ years. Unfortunately, the EOS 650 ended up getting stolen.

by ohazi 5y ago

toomanybeersies 5y ago

Phase detection autofocus (what you're describing) is still used in cameras today, and hasn't changed significantly since the 90's. DSLRs are essentially just a film camera, but with a CMOS sensor replacing film. One of the reasons why they still exist, despite the obvious inconvenience of not having a big screen to look at, is because until recently it wasn't possible to build a mirrorless camera with phase detection autofocus. I have a Canon EOS 55 from the late 90's, and in many situations it has better autofocus performance than my Fujifilm mirrorless from 2017.

ohazi 5y ago
Yeah, I remember when the first CMOS sensors with phase detection elements started showing up. This is what I meant by "conceptually similar" although I probably should have clarified, since we've had massive sensors without these elements for longer. I guess contrast-based autofocus is what really needs the DSP, since you need to do edge detection / FFTs on more patches of image for the technique to work well, as well as PSF analysis if you want to guess the lens defocus direction. With phase it's just a few cheap 1D correlations, and you get accurate lens direction information for free, which explains why it feels so much faster. Unfortunately we didn't have access to an electron microscope, so a modern mirrorless camera would have made for a rather less interesting teardown.
- sudosysgen 5y ago
  Contrast based autofocus just doesn't compete either. New, higher end mirrorless cameras still use phase detection just like the DSLRs of old. Sony still uses old-style sensors but mounted directly on the chip, while Canon uses pixels that only accept light from some directions to calculate phase information. Both of these tactics end up with very similar results. Though it is augmented with DSP.
sudosysgen 5y ago
Mirrorless cameras with phase detection have actually existed for a long time - actually since 2010 - before they became practical with EVFs and better batteries. Something that really slowed it down is that Canon filed a patent for Dual Pixel AF which was the first on sensor pdaf system to outperform previous systems, and then proceeded to not make any serious mirrorless camera using it. All the worse is that DPAF was a fairly minor invention that other companies didn't do first mostly because they didn't have the manufacturing capabilities yet - basically using a microlens to split the light instead of a divider to obscure it. So Sony had to find ways to make it work without infringing on the patents and without artifacts.
- toomanybeersies 5y ago
  As a disclaimer: I haven't bought a camera (or looked at new models) since 2018 > Mirrorless cameras with phase detection have actually existed for a long time - actually since 2010 I guess that depends on your timescale. The first SLR camera was 62 years ago, with the Nikon F (1959-1973, 14 years span). The first [SLR] camera with phase detection autofocus, the Canon EOS 650 (incidentally my first SLR [1]) was released in 34 years ago, in 1987. It wasn't until 2002 (Canon EOS-1Ds, 19 years ago) that there was a DSLR that could capture images on par with film. Even if Canon deliberately suppressed mirrorless PDAF (which I don't dispute), phase detection autofocus has been around much longer than digital cameras, and at least until recently DSLRs have had better autofocus performance than mirrorless. To be clear, I'm not cargo culting DSLRs here. My point is that beyond the change from film to CMOS sensor, DSLRs are not significantly different from the film cameras they replaced. [1] I mentioned to a coworker that I was thinking picking up photography; the next day she gave me a bunch of film photography gear that had been sitting in her attic for the past 20+ years. Unfortunately, the EOS 650 ended up getting stolen.