Sunday, 8 July 2007

Sony DSLR Speculations

Update (20/8/2007): Sony have released the details of the IMX021 12.47MP CMOS sensor which will presumably be in the new Sony DSLR. The main features include parallel ADC and 10fps.

My regular camera is a Konica Minolta Dynax 7D, I've had it for 2 1/2 years so far and it's given me stellar service. I was able to use the lenses I bought for the Dynax 7 (and AF 9000) to which I've added one reduced circle lens, the Tamron 17-50mm/2.8 XR di II. For (Konica) Minolta's first DSLR (ignoring the RD-175 and RD-3000, and even the video back for the 9000) of the modern era, it was a great camera, first with in-body anti-shake, more knobs and dials you could poke a stick at, and a nice hefty feel to the body.

However, times have moved on and it's time to upgrade. I think that 6MP is more than enough for 99% of the shots I take, but it's that 1% where higher resolution could come in handy that moves me to want 10MP or more. The AF was a step backwards compared with the Dynax 7, and digital flash has always had the problem of Off-The-Film (OTF) metering incompatibility leading to worse flash performance than in the film days.

So, with Sony having taken over the camera division of Konica Minolta and released the Alpha-100 DSLR last year, many Minolta AF mount (alpha mount) shooters are eagerly awaiting Sony's next move. From all indications, the next camera to be released will be the one on the right, referred to by Sony as an Advanced Amateur Model (hereby to be referred to as the AAM). No details have been released, nothing about the size of the sensor, the resolution, or any other features apart from the inclusion of Super Steady Shot (SSS). So I will dare to make some predictions, and wishes, about what this camera will bring. This is based on the technology developments which have occurred since the 7D was released in 1994 and what should be fitting for the spiritual successor of the 7/700/7000 in the Minolta line.

The heart of a camera is the sensor, whether it be film, a CCD or CMOS. Sony have jumped on the CMOS bandwagon, trying it out on the large sensor R1 previously and stating that it would be transferring its development and production efforts in this direction. CCD technology does have a few theoretical advantages over CMOS in the areas of fill factor, full-well capacity and fixed pattern noise but Canon has shown that practically, CMOS implementations can match or exceed CCD performance in the consumer space. However, some advantages of CMOS are too great to ignore hence Sony's interest in transitioning from CCD. CMOS sensors allow greater integration of image processing functionality into the sensor itself, and it is this functionality which Sony may well want to use to bring DSLRs into the mainstream.

Whenever I hand my 7D to someone else to use, I invariably get asked how do the turn the back screen on in order to frame the shot. I have to explain that they have to look into the viewfinder which leads to much confusion. Integration of Live-View into a large sensor is much more easily achieved in a CMOS chip. Instead of having to clock all the charges out of a CCD in order to rapidly display only a subset, in a CMOS sensor, region of interest (ROI) addressing is easily achieved by random access to each pixel.

Sony recently announced the IMX017CQE 1/1.8" 6MP sensor. What was interesting about this chip is the implementation on a consumer-type device of parallel or column analogue to digital conversion (ADC). Each column of the device has a dedicated readout circuit, hence the charge in each pixel can be measured much more rapidly than in conventional devices. This technology is not new, but so far has been confined to research or low volume high-end devices. It would not be inconceivable that Sony will implement parallel ADC on the new CMOS sensor for the AAM. This is assuming that the technology can overcome some potential hurdles: Cost, more circuitry means larger silicon, lower yield; fixed pattern noise, calibration of each ADC will be essential especially with changes in temperature. With so many ADCs, each one needs to only read out pixels less quickly, hence read-out noise can be much lower. This is due to lower intrinsic noise at lower frequency of operation and also from multi-sampling, where the number of electrons in each pixel is samples many times and averaged, leading to a reduction in the random noise by a factor of the square root of the number of samples. It should be quite possible to reduce the equivalent read-out noise to single electron figures by this method and still retain high frame rates. This would result in very high signal to noise and good low light performance. Video also becomes a viable option for DSLRs, though whether it would be a useful feature is to be seen.

Recently, Kodak announced a "new" Bayer pattern array which mixed clear or "white" pixels with green, red and blue pixels primarily to increase the sensitivity of a sensor. This was touted as a revolutionary development, but in fact is merely one permutation out of many possible Bayer patterns, many of which have been proposed and tried before. For maximum sensitivity, a monochrome sensor would be optimum, however there would be no chroma information recoverable. At the other extreme, the traditional Bayer RGBG pattern gives reasonably good chroma resolution. In between these two points, a continuum of possibilities balance sensitivity versus chroma resolution. One pattern implemented has been the use of complementary filters, Cyan (Green and Blue) Magenta (Red and Blue) Yellow (Green and Blue). An alternative would be to use White (R+G+B) Cyan and Yellow in a CWYW pattern. This would be equivalent to a RGB pattern in terms of chroma resolution, but would have an average photo-electron absorption of 5/6ths compared to 2/6=1/3 of an RGB array, or 2 1/2 times greater sensitivity. In terms of colour accuracy, the CWYW pattern may be more sensitive to noise so this would have to addressed in the de-mosaicing process.

Flash performance on DSLRs has been problematic. This stems from the fact that digital sensors have specular reflection characteristics. OTF metering relies on diffuse reflection from the surface of the film in order to calculate flash exposure. However, the light reflected from the surface of the sensor (of the order of a percent or less due to anti-reflective coatings on the IR filter/anti-aliasing stack) bounces like as from a mirror, which means that the usual flash metering sensors cannot intercept the rays. There are several ways this could be overcome in an all-electronic flash metering system embedded in the sensor. Most sensors incorporate overflow drains, pixels which exceed full well capacity dump their excess electrons into these drains and not into neighbouring pixels. By suitable high speed monitoring of these drains, overexposure could be detected and the flash output cut-off. The overflow regions could be segmented allowing a degree of selective exposure. More sophisticated would be a layered structure, the top surface pixels would be the conventional picture taking elements but below would be larger light sensitive regions which would capture a small fraction of the photons falling onto the sensor. These metering regions could also be used for on-chip continuous metering, useful for mirrorless operation.

With the use of Live-View, contrast AF becomes feasible. This could be used in conjunction with phase-difference AF as a Minolta patent shows. Phase-AF would provide coarse AF after which the mirror flips up and then contrast AF fine tunes the focus. I can imagine several permutations being available depending on the shooting situation. Sony may also implement an electronic shutter (like Kodak's CMOS Global Shutter). This would also mean far higher flash-sync speeds.

Another Minolta patent may show a future development, though it is not likely in the near future. This being the integration of in-lens image stabilisation with the current in-body SSS. This would combine the advantages of both systems, long lens performance of the in-lens system, and universal performance (especially wide aperture short focal lengths) of the in-body system.

The above are the more revolutionary aspects of the camera I envisage Sony releasing shortly. More mundane features I would like are: interchangeable viewfinders screens, type M/ML in particular; intervalometer; focus distance display; custom metering programs; a real mirror lock-up (MLU) as well as the 2s option; 1/2 pixel shift using SSS; Bluetooth GPS compatibility; good ergonomics.

My specifications/wishlist for the Sony AAM:

  • 24mm x 16 mm (approx.) 12MP CMOS Bayer Array (White,Cyan,White,Yellow this is pure speculation)

  • 5.5 micrometer pixel with 50,000 e- FWC

  • Parallel ADC readout, 10e- equivalent noise in normal mode, 5e- in "slowscan" mode

  • Pentaprism 95% coverage viewfinder, x0.85 magnification. Interchangeable screens including Type M/ML (for manual focussing/gridlines)

  • 9 AF sensors (distributed wider than on the 7D). Centre X and + type, outer + type

  • Imaging AE in pentaprism, colour sensitive

  • Live-View, real-time histogram and metering

  • 4 fps with conventional mirror flip up, 10fps with live-view

  • 30s to 1/8000th shutter (mechanical) or 1/16000 (electronic)

  • On chip flash metering system. Flash sync to 1/16000

  • Wireless flash control, compatible with 5600HS/3600HS

  • Red focus assist lamp

  • Super Steady Shot with improved low frequency drift

  • Focus distance display on rear LCD

  • Dual CF and SD card slots

  • Compressed RAW format/DNG

  • Bluetooth connectivity for GPS, file tagging, camera control

For David Kilpatrick's take on things, look here.