E-5 ISO settings

Started Jan 2, 2014 | Discussions thread
Flat view
Richard Turton Regular Member • Posts: 177
E-5 ISO settings

There has been a lot of speculation in the past about how the E-5 handles the different ISO settings and whether the intermediate values are noisier than the whole stop values. I have not seen much in the way of actual measurements other than those published by DXOMark, and restated by Sensorgen in a different form. I have recently measured the performance of my Olympus E-5 by analysing test exposures, using the RawDigger program (http://www.rawdigger.com/) to obtain the mean and standard deviation for the Red, Green, and Blue channels. Some of the results are presented here.

This is a plot of the read noise (standard deviation) of the E-5 over the ISO range of 100 to 3200, obtained by analysing dark frame exposures. In the absence of light, the output of the R, G, and B channels are all the same. The plot appears to show that the E-5 has only three "hardware" ISO settings (200, 400, & 800), presumably implemented by an analog amplification of the sensor output before it is digitized by the A/D (analog to digital) converter. The other sixteen ISO settings are apparently achieved through software scaling. The camera labels ISO's of 3200 and above as extended, but it appears that the extended range actually begins at ISO 1600 (for 1 EV steps).

The effect of the hardware gain change is to increase the available dynamic range slightly at the higher ISO's, as shown in this plot of dynamic range vs ISO.

The dynamic range was calculated by taking the base 2 log of the maximum A/D converter output divided by the measured read noise, both in A/D counts. The results agree amazingly well with those published by DXOMark for the E-5, measured using much more sophisticated techniques. But the DXO data does not capture the discontinuities in the curve caused by the changes in analog gain because they do not measure at all the intermediate ISO values.

A 12-bit A/D converter has a digital output range of 0 to 4095, but the E-5 subtracts a black level of 64 from the raw data that reduces the nominal output range to 0 to 4031. The black level subtraction must be turned off in Raw Digger to obtain the dark frame noise measurements. The actual full scale output of my camera varies from 4027 at low ISO to 3776 (0.1 EV lower) at ISO 6400. The accuracy of the dark frame measurements deteriorates above ISO 1600 due to clipping of the noise distribution at zero. That is why results above ISO3200 are not shown.

This is a plot of Signal to Noise Ratio for the green channel measured over the full ISO range using an old Kodak grey card. There are no visible discontinuities in the SNR curve because read noise is much less than photon shot noise at these exposure levels.

The maximum theoretical SNR of an ideal sensor is equal to the square root of the so-called Full Well Capacity in electrons (http://www.clarkvision.com/articles/digital.signal.to.noise/index.html). I have calculated the FWC of the E-5 sensor as approximately 14500, corresponding to a maximum possible SNR of 6.9 EV. Sensorgen.info gives a FWC of 18000 electrons for the E-5, corresponding to a maximum SNR of 7.1 EV.

From the same measurements it is possible to determine that unity gain, where the digital output equals the number of electrons captured by each pixel, occurs at about ISO 720. This explains why the analog gain remains constant above ISO800.

The results of these tests are interesting, but the implications for practical photography are not very significant. First, there is no benefit to using the intermediate ISO values (250, 320,...) even for JPEG's, because you lose dynamic range (less than 1 EV) without improving the signal to noise ratio.

Second, provided that you are shooting raw, there is no benefit from using ISO's above 800 because it just clips the highlights without improving the signal to noise ratio. You might just as well underexpose at ISO 800 to keep all the raw data and adjust the exposure in post processing, instead of allowing the camera to rescale the data and push any highlight information that may be present off the top of the curve.

Third, and this was already well known, the actual nominal ISO of the E-5 sensor is 200, and ISO's below 200 sacrifice dynamic range to improve noise by overexposing.

These observations are not new because they apply to some degree to most DSLR's, but I wanted to find out the specific details for the E-5, especially how it handles high ISO values.

Olympus E-5
If you believe there are incorrect tags, please send us this post using our feedback form.
Flat view
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow