|The Panasonic Lumix DC-GH6 has a ‘dual gain’ sensor, but not in the sense you’re likely to have encountered before.|
The Panasonic GH6 features a mode called Dynamic Range Boost which the company says is based on a sensor that features ‘dual output gain.’ This mode is always engaged in stills shooting mode, making it the first time we’ve seen dual output gain on a stills camera.
Given the widespread adoptions of the similar-sounding ‘dual gain’ sensor design, we thought this would be a good time to look at these systems and what they do.
Why use these approaches at all?
To understand what these systems are trying to achieve, it’s best to go right back and consider why gain/hardware amplification is applied in digital photography at all.
Pixels (or, more correctly, photodiodes) release electrons in response to being hit by light, in a process called the photoelectric effect. In conventional sensors, the pixels themselves always produce the same output voltage in response to a given amount of light.
Analog to digital conversion
Technically speaking, most camera sensors are analog devices: the signal they output is a continuous range of voltages. These voltages are converted into a series of digital values by a device called an Analog to Digital converter.
Chip designers measure the output of their pixels and choose ADCs with a bit-depth that can encode the sensor’s signal with sufficient precision (exceeding this would just mean generating a lot of excess data describing noise). The ADC is designed so that it delivers its maximum value (eg 16383 for a 14-bit ADC) in response to the maximum usable output from the pixels at base ISO.
However, this means that if you apply 2x amplification to the signal coming off the sensor, in the expectation of the sensor getting less light, the brightest areas of the image may end up producing signals too large for the ADC to correctly encode: meaning they still get recorded as the maximum value. The brightest lights in a low-light scene are often amplified to the point of clipping.
In very low light conditions, there are very few photons available, so the voltages generated by the sensor are very small. To deal with this, most sensors apply increasing amounts of amplification, which has two benefits:
Firstly, it means that the electronic signal coming off the sensor is a good match for the range of signals that the analogue-to-digital converters (ADC) is tuned to encode: this means that you encode the sensor’s output in plenty of detail. They tend to be linear, so the darkest few stops of light are encoded with only a handful of values.
Secondly, it means the signal coming off the sensor is much bigger than any electronic read noise that might get added between the amplification stage and it reaching the ADC, which minimizes the impact of this particular type of noise in the final image.
The disadvantage of applying amplification is that it pushes some of the data that the sensor might have captured beyond the range that the ADC can cope with.
In most older sensors, a programmable gain amplifier was used to apply increasing amounts of amplification to the signal as the camera’s ISO setting increased. The closer the amplifier was to the pixel, the better the sensor performed, because it amplified the signal before too much read noise could get added in.
Base ISO would typically be capable of capturing the widest dynamic range, then each increased step of amplification would push the brightest captured information beyond the range of the ADC (losing a stop of highlight information) but then reduce the impact of downstream read noise, helping to extend the DR in the shadows.
It was an effective way of making the most of those quiet signals.
Dual conversion gain sensors
|The 1″-type sensors made by Aptina in Nikon’s early 1-series cameras were the first we encountered to include the dual conversion gain design.|
Up until now, most cameras with ‘dual gain’ sensors in the photo sphere have been ones using a technology developed by chipmaker Aptina and later licensed by Sony Semiconductor. This uses a pixel design with two readout modes: one that includes a capacitor in the path, to provide extra electron storage capacity for bright, high DR conditions and another that disengages this capacitor. Taking the capacitor out of the readout path delivers less dynamic range but increases the conversion gain: boosting the signal for low light conditions.
This higher level of gain occurring right inside each pixel’s circuit means the gain is applied very early in the process, before much read noise has crept in, which helps deliver a high ISO improvement over conventional designs.
Notably, though, you need to make the decision whether to run the pixel in its high or low gain state: you can’t have both. Panasonic models, including the S1H and GH5S let you explicitly choose which mode the sensor operated in, via a menu option called ‘Dual Native ISO,’ but most cameras simply switch modes at an ISO setting chosen by the camera maker. It’s not always publicized but dual conversion gain sensors are used in most modern cameras from Fujifilm, Nikon, Ricoh, Olympus, Leica and Sony.
In future we’ll try to refer to this approach as ‘switchable Dual Gain,’ not in the sense that it’s a setting you can switch on and off, but because it’s a system that needs to switch between one of two states.
Dual output gain
The technology Panasonic is using is different, and has more in common with the approach taken in sensors used by cinema camera company Arri. Canon also takes a comparable approach in some of its Cinema EOS cameras.
This takes two outputs from the sensor and applies different levels of amplification to each: a high gain output with improved noise characteristics in the shadows and a low gain output that retains the highlight information that would otherwise by amplified to the point of clipping.
|DR Boost with matched exposures|
|DR Boost off
ISO 2000, F3.2, 1/125
|DR Boost on
ISO 2000, F3.2, 1/125
These two streams are combined, using the high gain data for the shadow regions and the low gain data for the highlights. The results are combined in a 16-bit space, so that the shadow regions aren’t just crushed back into the low-precision tail-end of the Raw file.
We’d expect the dual output approach to produce slightly less clean shadows, compared with the switchable dual gain design, simply because the amplification is occurring further down the readout path. But, conversely, we’d expect dual output gain to be able to retain highlights (and hence, wider total dynamic range) more effectively than the switchable dual gain approach does.
When we engage Dynamic Range Boost on the GH6, we gain an extra 1EV of highlight DR, as seen above. In some scenes, this ability to capture brighter detail will be useful. But in darker conditions, it gives us the option of raising exposure by 1EV, compared with the DR Boost off setting:
|DR Boost with ‘On’ given 1EV more exposure|
|DR Boost off
ISO 2000, F4.5, 1/125
|DR Boost on
ISO 2000, F3.2, 1/125
With an additional stop of light given to the DR Boost On footage, both files would clip in response to the same real-world tone, but everything else in the DR Boost On version would be one stop cleaner, giving you more flexibility when it comes to grading the mid-tones and shadows.
Both these technologies only really come into play at higher ISOs. With the existing, switchable Dual Gain designs, there’s typically a fixed ISO at which the camera changes from using the low conversion gain mode to using the higher one.
In a Dual Output Gain system, you can’t use both readout paths until you’re at a high enough ISO to need the amplified shadow regions. On the GH6, Panasonic says the Dual Output Gain becomes active at ISO 800 and above.
In both cases the ISO rating at which Log modes switch over will be higher. This is because Log modes are designed to incorporate more highlight information, so their base ISO states are given a higher rating, reflecting the lower exposures they’re meant to be given. So, for instance, the GH6 says that ‘Dynamic Range Boost’ is available from ISO 800 in stills and standard color mode, and ISO 2000 in Log video: this is simply because these thresholds are both are Base ISO + 3 stops.