Lorin Thwaits

The behavior of CMOS sensors is not well understood amongst videographers, and in this post I hope to explain one of its major benefits: wide dynamic range.

The CCD sensor found in most camcorders today naturally has a limited amount of dynamic range.  So to compensate, the iris, or aperture, setting modifies to a degree the amount of light that reaches the sensor.  This adjustment of aperture also affects depth of field.  In addition to this, lower-end CCD cameras usually tweak the shutter speed in order to make up for such a broad range.  And higher end cameras often employ one or two neutral density filters to cut some of the incoming light in outdoor scenes.  This makes the levels more in line with what the sensor can actually handle.  The neutral density filters used in the higher end cameras can still allow nice depth of field by allowing a wider aperture for brightly lit scenes.  But since the HDR-HC1 uses a CMOS sensor instead of CCD, these tricks are not necessary.  The CMOS sensor allows for an outrageous amount of dynamic range, its effectiveness dependant mostly on how precise the analog light levels get converted into digital bits.  Let's get a closer look at what's happening by experimenting with the exposure control found on the HDR-HC1. This is a really handy adjustment, shown here circled in red:

When enabled, you have 24 different "notches" that can be chosen in order to alter the aperture and processing being performed on the image.  I recorded a segment at a fixed shutter speed while cycling through these 24 options, and then played it back with Sony's "Data Code" setting on to see the results.  With this readout you can determine with certainty what aperture and gain settings were chosen by the system.  Notches 1 through 5 and 14 through 24 showed changing values for the aperture and gain.  But in the middle range from notch 6 through 13 the settings always list the same parameters: f4 and 0dB gain.  How can this possibly be?  It is quite obvious that when moving through this range of settings that the picture is changing in intensity, so something has to be affecting the exposure.  But what exactly is changing to make this happen?  There is an explanation.  Unlike CCDs that may be converting their intensity values to at most 9 or 10 bits of digital data, the CMOS sensor can give meaningful data back across a wider range of light intensity.  In the HDR-HC1 it gives back 14 bits of data, which is a range over 50 times wider than offered by a low-end CCD sensor.  The image processing section of the camera then trims the extra values out of range to obtain a usable picture, which offers additional variation to the exposure control.

That 14 bit range comes out of the CMOS sensor from its on-chip analog to digital converter, or ADC.  It is then processed by Sony's Enhanced Image Processor (EIP) to provide exposure control, narrowing in on ranges of bits to provide a kind of "sliding window" of intensity from the broad range of possible values.  The window does not always cover the same number of bits.  I expect it ranges from 5 bits wide in dim scenes recorded with 18db of gain to about 12 bits wide for really brightly-lit scenes, so as to not wash out the detail in the shadows.

First examining bright sunlit scenes, the values coming out of the sensor are kept quite high, near the uppermost values that the sensor can possibly report.  But as well significant detail in the shadow is available.  In HC1 footage it appears that around 12 bits are used to represent the red, green, and blue brightness values under that kind of lighting.  This gives the broad dynamic range many people rave about in daylight HC1 clips.

When the scene is darker, the camera still provides a linear response, but with lower values coming in it is forced into using fewer bits altogether.  When the response of the sensor drops really low, even lower than what HDV's 8 bits would normally record, gain is applied by multiplying the signal to bring it up to fill the 8 bits.  This is the same kind of gain that we've been accustomed to using all along in CCD cameras, and is represented as an amount in decibels (dB).  For each 6dB of gain the signal gets doubled, being shifted up by one bit.  This effectively throws out half of the possible values, and adds noise to the image.  In low light conditions if you have to use 18dB of gain it then means the signal has been doubled three times, which is the same as being multiplied by 8, or that three bits of data are lost.  In this circumstance the bottom three bits hold no meaningful values, effectively resulting in a signal with only 5 bits of variation.  This then gives a picture with lots of patchy grain.

Here is a chart with the aperture and gain settings used in the camera, and a third column of what I expect could be the sliding bit ranges used by the camera out of the available 14 bits.  That third column is total speculation on my part, based on the video appearing to have about 12 bits of range when the image is really bright, shifting gradually down to around 8 bits of range with indoor lighting, and finally about 5 bits in the dimmest of settings:

It is possible that on each of the color channels the bit ranges are set differently.  Or even possible that the output from the EIP stage is always only 7 bits, even though the MPEG compression in HDV calculates things based on 8 bits.  So the bit ranges are known only by Sony at the moment.  All I know for sure is that some table of sliding ranges like this exists in the camera to manage the broad dynamic range coming out of the CMOS sensor.

That's all for this installment.  My next video-oriented post will be about how to capture Sony's HDV signal as an MPEG-2 program stream, all with 100% freeware tools.

Whoa... On a totally unrelated note, someone posted an Excel document gone HTML that included a style to set all TDs to have white-space:nowrap, and it totally screwed up the display on the main gwb page. I threw in another style tag in this post to hopefully compensate! If the screen looks normal then it worked. Please don't put any HTML-ified Microsoft Office documents in as posts here!!!

Feedback