Higher 320x240 FPS?

[Lead Head]

In normal form, the S3 only lets you do 60FPS 320x240, vs. 640x480 30FPS, now 320x240 @ 30FPS is 4x less video information then 640x480 30FPS, so couldn't theoretically 120FPS be achievable, this would be very use full for some slow motion type of work. Or is it a sensor limitation?

[GrAnd]

Do not forget that the limitation in FPS is not (mostly) by data amount processing, but due to sensor limitation.

[vit40]

Sensor in S3 is ICX624, and it has 3 readout modes:

- full frame readout mode at  4.28 fps
- 4/8 line readout mode: 532 horizontal lines at 30 fps
- 4/16 line readout mode: 266 horizontal lines at 60 fps

[kwf]

Sensor in S3 is ICX624, and it has 3 readout modes:

- full frame readout mode at  4.28 fps
- 4/8 line readout mode: 532 horizontal lines at 30 fps
- 4/16 line readout mode: 266 horizontal lines at 60 fps

So the best thing which might be possible, is to increase the recorded horizontal resolution a bit. Probably not worth it, and far from easy. So lets wait for a Canon camera with the IMX017-sensor (http://www.dpreview.com/news/0702/07021801sonyhighspeedcmos.asp) 

Edit:
On the other side, this sensor is suspected to be used in the new Casio Pro EX-F1 (http://www.dpreview.com/news/0801/08010601casiof1.asp) which can achieve up to 1200fps with 336x96, while the datasheet claims that only 300fps is possible with a 1/5 read-out. So maybe it is possible to achieve higher fps, by reading even less lines. Although this might require serious modification, maybe even hardware, only possible for camera manufacturers.

I didn't understand the 1/5, 4/10, 4/20 etc. readout modes of the sony sensors. But i suspect that they really just read every X-th line in the video read-out anyway and they only add two horizonal lines when the datasheet says it. So reading the sensor at 1/10th of the vertical resolution only makes use of 1/5th of the sensor (for the icx629 e.g. because they add two horizontal lines). Can that be right?

[vit40]

Well, it would be possible to have something like 640x240, interpolated to VGA (640x480), at 60 fps. That's exactly what A610 and A630 have, but at 30 fps

4/8 readout mode, for example, means that 4 horizontal lines from every group of 8 lines are used, so 1/2 of sensor surface (= 1/2 of all pixels) is used in video mode. From pixels used, two neighbor pixels of the same color are added in vertical direction during readout (in sensor), and more resulting neighbor pixels are added by software in horizontal direction, finally resulting with one "video pixel" of something like VGA resolution at 30 fps

[kwf]

Well, it would be possible to have something like 640x240, interpolated to VGA (640x480), at 60 fps. That's exactly what A610 and A630 have, but at 30 fps

4/8 readout mode, for example, means that 4 horizontal lines from every group of 8 lines are used, so 1/2 of sensor surface (= 1/2 of all pixels) is used in video mode. From pixels used, two neighbor pixels of the same color are added in vertical direction during readout (in sensor), and more resulting neighbor pixels are added by software in horizontal direction, finally resulting with one "video pixel" of something like VGA resolution at 30 fps

Are you sure? For the A630 this would be the ICX476 chip? Datasheet says it can output VGA at 30fps.
The datasheet for the sensor of the a570is is very similar, maybe it does the same for VGA video.

It is unclear to me how many pixel are added, as all these sensors perform horizonal and vertical addition operations, if your interpretation is correct, maybe 4/10 means a block of 4 by 10 pixels is added and outputet, so all pixels are used. But the resolution of one image would be 776x232 pixel (for vga at 30fps). Somehow this does not make sense with the last column, claiming 466 lines, but that is maybe the meaning of fotenote 2. 

[vit40]

Yes ... in A630, sensor is capable of producing VGA video, but it seems it has been intentionally downgraded, as it has jaggies like A610 video, at least on samples from various sites. A620 and A640 have true VGA (or almost), S2 and S3 also etc ...

Anyway, according to spreadsheets of some Sony (CCD) sensors, only a part of pixels is used for video, as readout process is shown on diagrams. It's different story on CMOS sensors, where all pixels could be used in various combinations

[Chibs]

Sensor in S3 is ICX624, and it has 3 readout modes:

- full frame readout mode at  4.28 fps
- 4/8 line readout mode: 532 horizontal lines at 30 fps
- 4/16 line readout mode: 266 horizontal lines at 60 fps

Does the A630 sensor allow for 266 horizontal lines at 60 fps as well?

[kwf]

Yes ... in A630, sensor is capable of producing VGA video, but it seems it has been intentionally downgraded, as it has jaggies like A610 video, at least on samples from various sites. A620 and A640 have true VGA (or almost), S2 and S3 also etc ...

Anyway, according to spreadsheets of some Sony (CCD) sensors, only a part of pixels is used for video, as readout process is shown on diagrams. It's different story on CMOS sensors, where all pixels could be used in various combinations

Can you point me to a source where the readout procesc is shown in diagrams? I only know the 2-page datasheets of the individual sensors, where i could not find these diagrams.

[vit40]

http://www.riddle.ru/?page=articles/ccd

For example, download pdf for icx452 (1/1.8", 5 Mpix) and check page 11

ICX476 which is, I suppose, in A630 (1/1.8", 8 Mpix), has 494 lines at 30 fps (4/10 line readout) and 247 lines at 60 fps (4/20 line readout). So, true VGA mode should be possible at 30 fps, but for some reason, A630 doesn't have the same video quality as A620 and A640

[a710is]

does the a710 have the ICX629 sensor? it says it can capture full frames at 3.33 fps so what about higher framerate for continuous shooting? is that possible?

[kwf]

does the a710 have the ICX629 sensor? it says it can capture full frames at 3.33 fps so what about higher framerate for continuous shooting? is that possible?

My guess would be that the digic-III cannot encode the JPEGs fast enough. But digic-III in DSLR is capable of encoding the JPEG pictures faster than that, probably the digic-III aren't the same in every camera (clock frequency, memory bandwith etc.).

Somehow i doubt that the limitation of the powershots and ixus cameras is only artificial, because even the high price P&S cameras don't reach the maximum frame rate of the sensor (which would be possible with 65Mpix/s like the 40D manages), but maybe a moderate fps increase wouldn't be impossible:

A570IS: 1.7 fps @ 7.1M = 12 MP/s
A720IS: 1.3 fps @ 8.0M = 10.4 MP/s
S5 IS: 1.5fps @ 8M = 12 MP/s
SX100: 1.3fps @ 8M = 10.4 MP/s
A650IS: 1.2 fps @ 12.1M = 14.5 MP/s
G9: 1.5fps @ 12.1M = 18.15 MP/s

[vit40]

Well, obviously, there are some differences between cameras with Digic II. Starting with different voltage - lower models on 2 AA batteries, higher on 4 batteries, SLR-s on Li-ion etc ... I don't believe that it's possible to raise frame rate in photo or video mode

When using CHDK+RAW, frame rate in photo mode is limited to less than 1 fps anyway (in case of my A620), because writing speed to SD card is 8-9 MB/s and size of raw + medium size jpeg is about 10 MB

But I'm still hoping that it's possible to change sound format in video mode to 16 bit 22 kHz or something like that, now when I can use optical zoom and manual focus with Ewavr's latest builds

[kwf]

Btw. for those who are interested, i found a datasheet of an older sony sensor, explaining readout modes in detail:
www.eureca.de/pdf/optoelectronic/sony/ICX282AQ.PDF

So it might even be possible to achieve higher frame rates for the center proportion of the frame reading in an AF-mode (if this readout mode is still supplied). But i guess this would need some serious hacking which is probably not worth the effort.

[a710is]

sorry if i'm missing out on some detail but I just thought of something: if the sensors are capturing full lines that get resized maybe it's possible to get a wide screen format. instead of 640x480, 1280x480 would be nice for example which would then have its vertical resolution doubled in camera or on the PC
or uninterpolated 640x466 or 1280x 466 for example in the case of the A710 sensor
or maybe get the 1280x466(480) video and then double its vertical resolution in camera or in software thus obtaining pretty high quality 1280x932 video. and of course another option would be to resize from 1280x466 to 1280x720 which would basically get close to 720p HD video although a bit flattened. also 466x1.5 = 699 and 480x1.5=720 ( x1.5 =inserting a line between every 2 lines that would be their average).

so I'm saying that if we at least could have more horisontal resolution, it would be nice to use it.