Originally posted by filmamigo Lower ISOs are provided by attenuating the output in some manner.
A true lower ISO would require that either there is a lower input, or that the sensor has a capability of detecting lower signals than it currently does.
Here is how I understand digital cameras, please tell me where I go wrong: It seems to me that the current base ISO is the 'full' sensitivity of the sensor, ie, it records what it detects in its full single-pixel dynamic range (in case of an 8-bit sensor (for the sake of simplicity), the camera settings will be such that the brightest parts of the scene will register at near-max values (256). For an 8-bit sensor, the lowest values are ~1, and the dynamic range is basically given by this: the contrast between the highest and lowest value.
Now if we increase the ISO, we collect less light/lower values by changing the camera settings, and adjust by amplifying the signal. In doing so, we also amplify noise. Besides, if we use the full sensor range (0-256), we have 256-intensity steps per colour, but if we collect less light (lets say, for example, 128 steps) and amplify by a factor 2 to get back to the same values, we have less information, because we basically only have the even numbers from 0-256. We will have much less nuance in the colours. I don't entirely believe that this explanation is 100% correct, so if someone knows, please tell me.
If we want to do an attenuation step, we have a similar problem. For ISO 50, if our base or native iso of the sensor is 100, we'd have to set the camera up to massively overexpose and then de-amplify all sensor values. We run into a few problems: the sensor clips at maximum value of 256, so everything above is lost. So after de-amplification, if it was for example by a factor 2, there will be no brightnesses in the raw values above 128. Basically we lose the right part of the histogram. For the other values, we have rounding problems: for a deamplification of factor 2, each 2 histogram bins in the original is collapsed into a single bin post deamplification. So I think we should get very poor / no highlight recovery at all, and a hit in dynamic range.