I have reviewed five different 3D TVs. All five of them had noticeable problems with crosstalk.
Crosstalk is a word that has been used for decades in regard to consumer electronics. It is used to label a legitimate signal ending up in the wrong place. In audio, it typically refers to the leakage of signal from left channel to right, and from right channel to left.
Andrew Woods from the Centre for Marine Science & Technology* at the Curtin University of Technology, WA, recently delivered a paper on the subject of crosstalk in 3D displays in Japan (right click to download PDF: Understanding Crosstalk in Stereoscopic Displays). In this Andrew outlines pretty much everything you’d want to know about crosstalk, including identifying the main causes. Of these, the two that seem me to be most likely to have significant effects in the home environment are:
- Timing: if there is a time mismatch between the operation of the LCD LC* shutters in the eye wear and the display of the picture, then material from the wrong eye views will be seen. Timing might be a straight misalignment between glasses and screen, or due to the switching time of the glasses from opacity to transparency and back again.
- Opacity: how well the glasses actually block the other view when they are at full opacity.
In the paper, Andrew points out an exciting way to combat crosstalk: pre-cancellation. If the amount of crosstalk is known, then the system could feed negative crosstalk into the signals which would cancel the leakage. For example, if you know that you can see a black dot from the left eye image over a white background in the right eye image at the 5% level, then you could make that dot in the right eye image whiter than white (possible in the video world) by a sufficient amount to cancel out the black leakage. That’s probably oversimplifying, but you get the idea.
A lot of work would be required for this, along with enormously complicated cancellation maps (covering an enormous range of colour-on-colour and -intensity-on-intensity possibilities, and perhaps requiring different treatment for different parts of the screen).
I’ve just started fiddling with this stuff. I created my own 3D stills (using simple geometric shapes) to show that I could do it, and to determine how the left-right shifts of objects between the views is perceived. (This might sound hard, but it’s just a matter of creating side-by-side stills, and then playing them on the 3DTV using — in my case — the network photo display capabilities of a Beyonwiz PVR.)
I’ve also started trying to work out how to use some of the suggested test patterns to quantify crosstalk. At a guess (Samsung didn’t supply a working remote control with the TV I’m using, so it’s all bit of a struggle at the moment), I figure black images from one eye show through on white backgrounds of the other eye at something like the 10% level, and white images break through over black backgrounds at the same level.
That’s a worst case. Using this TV I was watching some richly colourful animation in which no crosstalk at all was discernible for seconds at a time.
Now consider audio crosstalk – the leakage between left and right stereo channels. These days 60dB or more as regarded as a bare minimum for uncompressed digital signals (all leakage would occur in the analogue stages). In the original home stereo format — the vinyl LP — midrange separation of ~20-25dB was considered fine, and there was never really any trouble with that.
Now, 25dB is about 5%. So should we consider 5% okay for 3D video crosstalk?
Although both stereo audio and stereoscopic 3D use left and right signals to give the false impressions that images (aural or visual) are coming from places which they aren’t truly coming from, they are still entirely different. Stereo audio is primarily about where the signal appears to be between the left and right points. If a bit of leakage is occurring (consistently across the frequency spectrum), then that particular piece of sound will appear to be coming from a place a little closer to the centre than it should be. (If the leakage varies according to frequency, then the image will be smeared horizontally, which may explain the attraction of LP for some listeners, since this smearing can sound attractive, even though it is highly inaccurate).
3D vision is not about mixing the left and right signals. You don’t get an average from the leakage, you get a distracting ghost.
My guess is that 3D displays will have to get the figure to well below 1% (-40dB) before we can pronounce them to be of really high quality.
* Thanks to Andrew for the corrections. I’m really going to have to be careful about the second. As he has previously reminded me, the shutter glasses are not LCDs – liquid crystal displays. They use liquid crystals as shutters, not display panels.