Friday, 16 December 2011

Electronic First Curtain

One of the nifty features of the A77 (and other cameras like the NEX-5n, A65 and NEX-7) is the option of using Electronic First Curtain (EFC) shutter mode. This eliminates the need to use a mechanical shutter to cover the sensor prior to exposure. In a live view camera, usually the sensor needs to be covered so that the pixels can be reset to zero before the exposure. EFC can accomplish this without requiring the sensor be covered. This has been available in some previous consumer DSLRs but the A77 is the first to really capitalise on its advantages due to the SLT design.

To see how EFC works, we should first understand how a normal shutter works. The purpose of the shutter is to allow light to fall onto the sensor for a specified period of time. In a traditional (non-liveview) camera, the sensor starts off covered by the first curtain, typically a set of interleaved metal blades. As the sensor is not exposed to light in this mode, it is a simple matter to reset it to accept the exposure and record the image. At the beginning of the exposure, the first curtain begins to slide down, exposing one edge of the sensor. As it uncovers the sensor, light falls on the pixels and creates charge. The edge of the first curtain continues to draw across the sensor until the whole sensor is uncovered. After the first exposed pixels have received the proper exposure, the second curtain now starts to move in the same direction as the first curtain to cover the sensor. The charges accumulated in each of the pixels are readout and converted into a digital image. Both first and second curtain then return to their original starting positions keeping the sensor covered, ready for another exposure to be taken.

For even exposure, the first and second curtains should traverse the sensor at the same speed. The delay between the motion of the first and second curtain determines the exposure time. On a professional camera, it take roughly 1/250th of second for the curtains to travel across the height of the sensor. This is called the flash-sync speed as this is the fastest shutter speed you can use for even flash exposure. The whole sensor must be uncovered at the point at which the flash is fired. At higher shutter speeds than flash sync, the region of the sensor which is exposed at any one moment is a slit between the edges of the first and second curtains. Faster sync speeds are difficult as this would require accelerating and decelerating the mechanical blades of the curtains beyond what is easily accomplished given the engineering constraints. Here's a link to a video explaining this.

In a liveview camera, the sensor starts off exposed to light. When the shutter is depressed, normally the first curtain rises from its stowed position to cover the sensor, the sensor is then reset (all the charges in the pixels are drained), and then the sequence as described above is triggered except that at the end, only the second curtain retracts so that the sensor is again exposed for liveview.

EFC eliminates the need for the first curtain to rise to cover the sensor before pixel reset. The pixels are design so that they can be reset without having to be in darkness. When the shutter is depressed and exposure initiated, rows of pixels of the sensor are reset beginning from the top. This line of reset pixels travels down the height of the sensor, just as a first curtain traversing the sensor reveals successive rows. As soon as a pixel is reset, it starts accumulating photocharge. After the proper exposure time, the second (mechanical) curtain then begins its journey across the sensor to shut off light and the image can be read out.The second curtain now returns to the top exposing the sensor for liveview.

The key point is that the start of exposure for any pixel can be electronically and independently controlled, and only the end of the exposure is controlled by a mechanically moving shutter blade. There are particular advantages to this, especially in a pellicle mirror (SLT) camera. Eliminating the moving mirror improves sharpness by avoiding mirror slap, but the first curtain would then be the dominant source of vibration. EFC eliminates this source of unsharpness as well. The sound of the shutter is quieter as there are less moving parts. And the time between the shutter button being depressed and the start of exposure can be shortened as well.

However, there are further potential benefits which could be exploited in future models. Flash sync speed could be effectively halved. EFC could start row initialisation from both sides of the sensor simultaneously, meeting in the middle, at which the whole sensor is exposed and the flash could be triggered. Then both mechanical curtains could move so that they also meet at the middle. Since they have to only travel half as far, the time it takes to cover the sensor is also half leading to a shorter flash sync time. The exposures of the top and bottom halves would have to be matched quite accurately (making sure the traversal speeds were the same), and the overlap region in the centre of the frame would also have to be designed to give even exposure.

A more extensive use of EFC is to do per-pixel exposure control. If each pixel can be reset independently, then the time it accumulates photocharge can be also be controlled individually. Let x be the distance from the top edge of the sensor (where the second curtain starts to cover the sensor) and v be the second curtsin traversal velocity. If at time t_0 the second curtain starts to move, the pixel will be covered at time t_x=t_0 + x/v. To obtain an exposure time of t_e, the pixel needs to be reset at time t_x-t_e. This choice of exposure can be done independently for all pixels.

In a liveview camera, since the camera has a realtime high resolution exposure map of the scene, it can calculate suitable per-pixel exposure times in order to compress the dynamic range of a scene, for instance using a short exposure time to capture the sky and a longer exposure for areas in shadow. This will require specific pixel reset capabilities of the sensor, current ones may only be able to reset pixels on a row by row basis. However, even row by row exposure compensation could give you an in-camera neutral grad filter (albeit only for landscape orientation).