PANA OLY approach vs CANIKON approach to mirrorless

Started Jun 9, 2012 | Discussions thread
Forum ProPosts: 26,068
Re: Differences in AF system behaviour
In reply to boggis the cat, Jun 11, 2012

boggis the cat wrote:

bobn2 wrote:

I wouldn't take either that article or that author as gospel. Nonetheless, I never said that PDAF was fully or even exclusively open loop. However, the technique has open loop capability and that allows the bulk of the focus adjustment to be made without a continuous feedback loop. If you read the AF patents (which are probably a more accurate indicator of the algorithms than WilbaW's experiments) and look at the AF protocols that is exactly how it works, with the major adjustments being made open loop. CDAF does not even get an indicator of which direction to move focus until it has moved focus.

Easily seen by repeated re-focus on the same target. PDAF usually doesn't shift the lens focus at all, CDAF always does a dance as it must re-check.

Which I think gives us a clue as to why we have the hybrid systems on the 650D and N1 - both of these are very video orientated, and in video that little dance, however rapid, is distracting. The end-of-focus dance is probably not so problematic, because it's a fine focus adjust not a coarse one. For most still purposes, CDAF is just fine, and really all that's needed is that initial 'which direction' clue - which is I think what Canon has done. What I don't understand is why this insistence on TTL CDAF, with a non TTL rangefinder, as on Leica still cameras, that initial clue could be given without need for messing with the sensor.

(A very fast dance on the E-M5 with the 12-50, though.)

No argument there. CDAF and closed loop are more accurate, which is why every PDAF has closed loop part or the cycle.

With the lenses I have this is usually a two-step process. Sometimes the system appears to shift the focus within the broad target area after the first coarse step, however, and so may take more steps. This can be annoying if it suddenly decides that a more distant or closer object is a "better" target (worse if it then gets "obsessed" with the new target and won't re-acquire the desired object).

CDAF always seems to start in the same direction (toward near focus, I think, but I haven't verified that -- and it could be body specific, obviously) so will reverse quickly if required then rapidly get to the correct focus. This is for the native lenses on the E-M5. CDAF on the E-5 and other standard FT bodies seems to be much more prone to over-shoot and having to jump back and forth.

In theory, CDAF performance depends on processing capability and fast linear AF element movement whereas PDAF depends on the hardware used and the ability to shift lens elements accurate distances very rapidly. (If I understand the processes correctly.)

It's the fast mechanical movement of the lens elements which is the key for effective CDAF. That means making them small and light or the motors very powerful. Its an area where a smaller format has a definite advantage (as it does also for PDAF due to the smaller f-numbers small sensor systems tend to have)

Reply   Reply with quote   Complain
Post (hide subjects)Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow