DxOMark suggest Four Thirds sensor technology has made a big leap in performance

[Four Thirds User editorial team]

Respected experts at DxOMark.com say the Panasonic Lumix DMC-GH1 Micro Four Thirds camera heralds a major jump in LiveMOS sensor performance. We examine why this could be great news for Four Thirds as a whole.

More...

[AndyElliott]

I have long been trying to fathom why dxo consistently rate the dynamic range of fourthirds sensors as quite a bit lower than their APS-C counterparts whereas dpreview, if anything, rate them higher. I think the reason must be noise thresholds. DPReview may be purely looking at the ability to distinguish levels of grey on their custom 'wedge', whereas dxo may factor in that if the noise levels are above a certain level then the sensor cannot distinguish reliably between levels.
This effect would make the perceived shadow DR look worse whereas the highlight DR should look similar between the methodologies employed (as the S/N ratio is worse in shadow areas).

This would seem to make sense in the GH1 case - the SNR is similar, so it doesn't adversely affect the DR comparison. This also suggests much better shadow noise control at low ISOs than we have seen of late (something the old Kodak sensors were very good at in the ISO 100-200 range). The low-ISO shadow noise is visible on the E-3, a new model that doesn't show this as much would be very welcome.

Given the rumours of better, newer generation of sensors on the cards for the new models, we might even see the day in the next few months where the FT cameras are better for noise than the current Canon and Nikon competitors! And I don't think that is pie-in-the-sky either!

Andy

[Ian]

It had occured to me that the GH1 sensor, which is basically a limited edition product as the GH1 is not a big volume seller, could be a back-lit sensor?

Ian

[AndyElliott]

Originally Posted by Ian

It had occured to me that the GH1 sensor, which is basically a limited edition product as the GH1 is not a big volume seller, could be a back-lit sensor?

Ian

That would surprise me - it's a pretty massive change to manufacturing technique to do that just for one sensor - even if it is a limited edition model. Also, so far as I can gather, the benefits of backlit sensors are down to the increased photo-sensitive area vs the circuitry - the latter being more or less fixed irrespective of pixel size. Backlit sensors 'hide' the circuitry behind the pixel photosite so it doesn't 'rob' the pixel of photosensitive area. The benefits look only significant for the small sensors in compacts - the benefits for FT and APS-C are far less, still less for 35mm frame.

Andy

[f22]

Hi Ian,

Is the GH1 sensor different to the G1 sensor? I am not so clued up on the micro four thirds panasonic kit...

I noticed Thom Hogan has been hinting towards a fuji sensor for an Olympus camera. Do you think Olympus would go this route for the E5, or stick with Panasonic?

cheers
f22

Originally Posted by Ian

It had occured to me that the GH1 sensor, which is basically a limited edition product as the GH1 is not a big volume seller, could be a back-lit sensor?

Ian

[AndyElliott]

Originally Posted by f22

Hi Ian,

Is the GH1 sensor different to the G1 sensor? I am not so clued up on the micro four thirds panasonic kit...

I noticed Thom Hogan has been hinting towards a fuji sensor for an Olympus camera. Do you think Olympus would go this route for the E5, or stick with Panasonic?

cheers
f22

The GH1 sensor can support multiple aspect ratios - so is different, it covers more of the image circle and isn't rectangular! (Actually over 14MP in total, 12MP are covered for the 4:3 aspect ratio)

I suspect Oly will stick with Panasonic for the E-5 - no bad thing, considering how good the GH1 is!

Andy

[f22]

Hi Andy,

Thanks for the reply, I was just looking at the DP review diagram here:

http://www.dpreview.com/reviews/panasonicdmcgh1/

As you say, the sensor is a different size allowing for multi-aspect ratio, but I wonder if the technology in the actual photosites is different from the G1, or its the same but put onto a larger wafer.

I notice on the DPreview summary page they give both the G1 and the GH1 the same pixel pitch of 5 MP/cm² pixel density.....so maybe the photosites are the same, it's just scaled up on the GH1 onto a larger wafer.

There again, I did notice another reference on the web from an interview with Panasonic. Here the guy points towards the difference being in the imaging pipe-line, rather than the sensor:

"GH1 is used in a sensor multiaspecto prepared for high-definition video in the whole process and transmit the signal is digital, so the interference and noise are lower. The GF1, by contrast, uses the Lumix DMC-G1 and its four-channel analog transmission."

source: http://translate.google.com/translat...istory_state0=

I don't know....to much for me

Cheers
f22

Originally Posted by AndyElliott

The GH1 sensor can support multiple aspect ratios - so is different, it covers more of the image circle and isn't rectangular! (Actually over 14MP in total, 12MP are covered for the 4:3 aspect ratio)

I suspect Oly will stick with Panasonic for the E-5 - no bad thing, considering how good the GH1 is!

Andy

[AndyElliott]

f22

"GH1 is used in a sensor multiaspecto prepared for high-definition video in the whole process and transmit the signal is digital, so the interference and noise are lower. The GF1, by contrast, uses the Lumix DMC-G1 and its four-channel analog transmission."

That comment is a bit odd and goes against my assumption of how the sensor to process comms work. That implies the final stage of the processing 'on chip' is a D-to-A stage, connecting the chip to the processor via 4 channels (R, G1, G2, B) in an analogue fashion (varying voltage), the processor then having a further A-to-D stage to turn it back into digital information. This would explain why the E-P2 and the GF-1, with nominally the same sensor, perform differently in noise - the TruePic V would appear to be better at the last A-to-D step.

However, this is all bonkers really. The ideal would be to not have the D-to-A followed by the A-to-D step at all - noise is bound to be introduced. This appears to be what the GH-1 is doing, namely providing the 12-bits-per-pixel as a digital stream of information to the processor. I do hope that Olympus employ the GH-1 method on their new cameras this year. The advantage of decoupling the digital information from sensor to processor via having the D-to-A followed by the A-to-D does allow you to process in a different number of bits per pixel on the sensor and processor side, as well as being able to accept analogue data from a CCD-type sensor too, but surely introduces significant noise.

Andy

Ian: can you comment on what I've said, perhaps asking around to see if that is correct?

[f22]

[QUOTE=AndyElliott;50837]

"GH1 is used in a sensor multiaspecto prepared for high-definition video in the whole process and transmit the signal is digital, so the interference and noise are lower. The GF1, by contrast, uses the Lumix DMC-G1 and its four-channel analog transmission."

That comment is a bit odd and goes against my assumption of how the sensor to process comms work. That implies the final stage of the processing 'on chip' is a D-to-A stage, connecting the chip to the processor via 4 channels (R, G1, G2, B) in an analogue fashion (varying voltage), the processor then having a further A-to-D stage to turn it back into digital information.

However, this is all bonkers really. The ideal would be to not have the D-to-A followed by the A-to-D step at all - noise is bound to be introduced. [QUOTE=AndyElliott;50837]

Andy, agreed - more stages to introduce noise and distortion. Maybe this is why the G1 shows more noise in tests than the GH1. It is perhaps more down to the difference in processing pipeline than superior photosite/sensor technology.


I do hope that Olympus employ the GH-1 method on their new cameras this year.

I hope so too, the improvement in results are pretty clear with the GH-1. I notice further down in the article the interviewer asks why they did not employ the GH-1 method to the later GF-1, and higher cost was quoted as the reason. However if they used this approach in more models the unit cost should come down.


The advantage of decoupling the digital information from sensor to processor via having the D-to-A followed by the A-to-D does allow you to process in a different number of bits per pixel on the sensor and processor side, as well as being able to accept analogue data from a CCD-type sensor too, but surely introduces significant noise.

Yep, looking at the noise comparison on Dpreview between the G-1 and GH-1 the GH-1 has the edge.


Thanks,
f22

[dingenus]

Andy, as far as I know is a Sensor an analoge divice. There is no D-A--A-D conversion, only A-D.
But you can made that in different ways. You can amplify the analoge signals before conversion or after that.
I think that this is the difference between the G1 and the GH1.
If you do analoge amplifying etcetera before conversion, that would adding noise and, maybe, interfere between channels.
If you do the conversion (nearly) on chip and the amplifying after conversion, and do that well, there would be a cleaner signal.
Both ways are usable and there are naturally more possible variationen.

Regards, Dingenus