-3 db gain at DVinfo.net

Shekar Dattatri · February 10th, 2005, 07:47 PM

I'm going out for a test shoot with my new XL-2 (what joy!) to a nearby bird sanctuary. While discussing this with a friend yesterday he recommended that I shoot all video at -3 db whenever possible as it would give richer blacks and reduce video noise as well. Any comments? Thanks - Shekar

Chris Hurd · February 10th, 2005, 08:41 PM

Shooting at -3dB gain will also help to compress your depth-of-field somewhat; if you really like that shallow depth-of-field look where the foreground is sharp and the background is softer, the -3dB gain setting will help a bit in that regard.

Shekar Dattatri · February 10th, 2005, 09:04 PM

Thanks, Chris, but will it in fact reduce video noise and give richer blacks? Is it good practice to shoot -3 db whenever possible as my friend suggests? Or is there a flip side to it (artefacts etc.). I understand the point about depth of field and that is certainly an added advantage. Thanks!

Rob Lohman · February 11th, 2005, 04:41 AM

The flip side is that you need more light, so shooting -3 may not
always be possible. I always shot at -3 db when possible with the
XL1S though. It indeed reduces noise. It tones down the analog
signals coming from the CCD's which in effect reduce the noise
levels.

Shekar Dattatri · February 11th, 2005, 08:03 AM

Thanks Rob, I just shot my own test and am looking forward to evaluating the shots and comparing 0 db vs -3 db gain.

A. J. deLange · February 11th, 2005, 03:37 PM

Thing is that -3 dB doesn't appreciably reduce noise. At -3dB gain I measured SNR of 49.1 dB and at 0 dB gain 49.5 (the difference is so small that it's probably not statistically significant). Going to plus 3 dB costs a bit of SNR: 47.8 dB and then, and here's the surprise, +6 dB gives the best SNR of all a whopping 55.3. Going to +12 then degrades SNR to 45.4 - the worst of all the settings and as to be expected. I'm not really comfortable with the 6 dB result but did repeat the measurement several times. If it's right then +6 is the number to use for best picture quality and peoples' experiences should come in confirming that. Details of my measurement technique can be found in my post of about a week ago. Just so there is no confusion: we're talking about the XL2 here.

A. J. deLange · February 12th, 2005, 11:30 AM

Bothered by the SNR measurements I got last week I took some more measurements today. Two things were bothering me (besides the strange +6 dB) result and these were that noise is a mix of chip noise which enters the sytem before gamma correction, and preamp, amp and quantizing noise which may come in before or after depending on where gamma correction is done (i.e. in the digital, as I suspect, or analog domains). The second was that the way in which I measured SNR (variance between pixels with lens cap on) would pick up variations in biases across the face of the chips as noise. In this set of measurements I captured pairs of frames, extracted luminance, differenced the pair members and took the variance of the differences. Biases, if any, are canceled by this method. The SNRs, relative to luminance of 255 obtained with this method with the lens cap on at respectively -3,0,+3, +6 and +12 dB gain are 57.5, 52.3, 49.1,50.8 and 48.1 dB. Note that these new data still show +6 to have better SNR than +3. At least the new data show that the best SNR is obtained at the lowest gain as I think we all felt intuitively would be the case.

Now these data show the signal to noise in the darkest parts of the picture which are not the same as for other parts because of the nonlinear gamma correction. So I took a second set of measurements in which I captured pairs of frames of a normally exposed scene, converted to luminance, differenced and measured the variance of the differences. These numbers are, to me, truely representative of the picture quality (with respect to noise) we get with the XL2 because they directly represent the variation from frame to frame which is how we see noise (like tiny little bugs crawling around in our video). Computing SNR for these data again relative to 255 counts gave values of 41.1, 40.0, 39.2, 38.6 and 36.6 dB SNR at -3, 0, 3, 6 and 12 dB gain. At least these numbers are monotonic with gain. The bottom line conclusion is more or less the same with respect to -3 dB: it's SNR is not appreciably better than the SNR at 0 overall (but is appreciably better in the deepest shadows). At lest the hypothesis that +6 is the best setting is off the table!

Matthew C. Abourezk · February 12th, 2005, 11:56 AM

WOW A.J.

Thanks for the detailed research information... very interesting. Also nice to have statistical information to prove what we had thought to be the situation.

Good work and thanks for sharing your findings.

Matt

Shekar Dattatri · February 13th, 2005, 05:54 AM

A.J. Thanks for that really comprehensive answer to my original question! After your first reply I ran my own tests and although I had no way to measure SNR, judging purely by eye, I found both -3 dB and O dB to be the best, but I was surprised at how good + 6 dB was. I'll certainly feel more confident about using this setting in low light conditions. I also tried + 18 dB with and without NR (at high, middle and low) and found that while NR did reduce noise, it introduced some other artefacts that I did not quite like. If I were in a situation that demanded that I use the +18 setting I would probably not have NR on. However I'm treating this test as preliminary and want to run more detailed tests on anything above +6 dB with and without NR.

Greg Boston · February 13th, 2005, 07:10 AM

A. J. I would also like to thank you very much for contributing these readings. One thing I noted however was the lack of testing the +18db setting. I am particularly intererested in that one because one thing that pushed me to buy the XL-2 was the night footage shot in Vienna, Austria. The final shot in that footage was taken at +18 according to the shooter. I thought it looked pretty good.

Can you re-run your tests at the +18 setting? We might convince Chris to make this a page 2 article on the main site sort of like the lense comparisons he did awhile back.

Your humble XL-2 co-mod,

=gb=

A. J. deLange · February 13th, 2005, 09:20 AM

Greg,

The reason I didn't do +18 was that I was working with 30p where if you dial in 18 the camera sets 12. I did 60i this morning and the results for a normal scene are, for respectively -3, 0, +3, +6, +12, and +18 dB gain, SNR's of 41.73, 40.72, 39.8, 39.4, 38.1 and 35.6 dB. These show the same pattern as the 30p numbers: only a couple of dB degradation until gain settings go above 6. Note that the 60i numbers are a little better than the 30p numbers.

Dark SNRs are 51.1,50.6,50.2, 49.7, 55.0 and 45.4 with the fact that the dark SNR is best at a + 12 dB gain setting being the surprise in this case.

All measurements were made with the camera in the 16:9 aspect ratio.

Greg Boston · February 14th, 2005, 01:11 AM

Thanks A.J.! I do appreciate your taking the time to grab that extra data. BTW, did you ever see the footage I referred to in my previous post? Some great stuff.

-gb-

Tom Hardwick · February 14th, 2005, 05:08 AM

It's interesting that Sony saw fit to have a -3dB setting on the TRV900 and the PD100, but dropped this on subsequent models. There's no such setting on the PDX10, the VX/PD or the new FX1, pointing to the suggestion that there's no other gain than a half-stop aperture increase.

tom.

February 10th, 2005, 07:47 PM	#1
Shekar Dattatri New Boot Join Date: Feb 2005 Location: Chennai, India Posts: 24	-3 db gain I'm going out for a test shoot with my new XL-2 (what joy!) to a nearby bird sanctuary. While discussing this with a friend yesterday he recommended that I shoot all video at -3 db whenever possible as it would give richer blacks and reduce video noise as well. Any comments? Thanks - Shekar

February 10th, 2005, 08:41 PM	#2
Chris Hurd Obstreperous Rex Join Date: Jan 2001 Location: San Marcos, TX Posts: 27,368 Images: 513	Shooting at -3dB gain will also help to compress your depth-of-field somewhat; if you really like that shallow depth-of-field look where the foreground is sharp and the background is softer, the -3dB gain setting will help a bit in that regard. __________________ CH Search DV Info Net \| 20 years of DVi \| ...Tuesday is Soylent Green Day!

February 10th, 2005, 09:04 PM	#3
Shekar Dattatri New Boot Join Date: Feb 2005 Location: Chennai, India Posts: 24	-3db gain Thanks, Chris, but will it in fact reduce video noise and give richer blacks? Is it good practice to shoot -3 db whenever possible as my friend suggests? Or is there a flip side to it (artefacts etc.). I understand the point about depth of field and that is certainly an added advantage. Thanks!

February 11th, 2005, 04:41 AM	#4
Rob Lohman RED Code Chef Join Date: Oct 2001 Location: Holland Posts: 12,514	The flip side is that you need more light, so shooting -3 may not always be possible. I always shot at -3 db when possible with the XL1S though. It indeed reduces noise. It tones down the analog signals coming from the CCD's which in effect reduce the noise levels. __________________ Rob Lohman, visuar@iname.com DV Info Wrangler & RED Code Chef Join the DV Challenge \| Lady X Search DVinfo.net for quick answers \| Buy from the best: DVinfo.net sponsors

February 11th, 2005, 08:03 AM	#5
Shekar Dattatri New Boot Join Date: Feb 2005 Location: Chennai, India Posts: 24	-3 db gain Thanks Rob, I just shot my own test and am looking forward to evaluating the shots and comparing 0 db vs -3 db gain.

February 11th, 2005, 03:37 PM	#6
A. J. deLange Major Player Join Date: Jun 2004 Location: McLean, VA United States Posts: 749	Thing is that -3 dB doesn't appreciably reduce noise. At -3dB gain I measured SNR of 49.1 dB and at 0 dB gain 49.5 (the difference is so small that it's probably not statistically significant). Going to plus 3 dB costs a bit of SNR: 47.8 dB and then, and here's the surprise, +6 dB gives the best SNR of all a whopping 55.3. Going to +12 then degrades SNR to 45.4 - the worst of all the settings and as to be expected. I'm not really comfortable with the 6 dB result but did repeat the measurement several times. If it's right then +6 is the number to use for best picture quality and peoples' experiences should come in confirming that. Details of my measurement technique can be found in my post of about a week ago. Just so there is no confusion: we're talking about the XL2 here.

February 12th, 2005, 11:30 AM	#7
A. J. deLange Major Player Join Date: Jun 2004 Location: McLean, VA United States Posts: 749	Bothered by the SNR measurements I got last week I took some more measurements today. Two things were bothering me (besides the strange +6 dB) result and these were that noise is a mix of chip noise which enters the sytem before gamma correction, and preamp, amp and quantizing noise which may come in before or after depending on where gamma correction is done (i.e. in the digital, as I suspect, or analog domains). The second was that the way in which I measured SNR (variance between pixels with lens cap on) would pick up variations in biases across the face of the chips as noise. In this set of measurements I captured pairs of frames, extracted luminance, differenced the pair members and took the variance of the differences. Biases, if any, are canceled by this method. The SNRs, relative to luminance of 255 obtained with this method with the lens cap on at respectively -3,0,+3, +6 and +12 dB gain are 57.5, 52.3, 49.1,50.8 and 48.1 dB. Note that these new data still show +6 to have better SNR than +3. At least the new data show that the best SNR is obtained at the lowest gain as I think we all felt intuitively would be the case. Now these data show the signal to noise in the darkest parts of the picture which are not the same as for other parts because of the nonlinear gamma correction. So I took a second set of measurements in which I captured pairs of frames of a normally exposed scene, converted to luminance, differenced and measured the variance of the differences. These numbers are, to me, truely representative of the picture quality (with respect to noise) we get with the XL2 because they directly represent the variation from frame to frame which is how we see noise (like tiny little bugs crawling around in our video). Computing SNR for these data again relative to 255 counts gave values of 41.1, 40.0, 39.2, 38.6 and 36.6 dB SNR at -3, 0, 3, 6 and 12 dB gain. At least these numbers are monotonic with gain. The bottom line conclusion is more or less the same with respect to -3 dB: it's SNR is not appreciably better than the SNR at 0 overall (but is appreciably better in the deepest shadows). At lest the hypothesis that +6 is the best setting is off the table!

February 12th, 2005, 11:56 AM	#8
Matthew C. Abourezk Regular Crew Join Date: Jan 2005 Location: Easton, CT Posts: 28	WOW A.J. Thanks for the detailed research information... very interesting. Also nice to have statistical information to prove what we had thought to be the situation. Good work and thanks for sharing your findings. Matt

February 13th, 2005, 05:54 AM	#9
Shekar Dattatri New Boot Join Date: Feb 2005 Location: Chennai, India Posts: 24	-3 db gain A.J. Thanks for that really comprehensive answer to my original question! After your first reply I ran my own tests and although I had no way to measure SNR, judging purely by eye, I found both -3 dB and O dB to be the best, but I was surprised at how good + 6 dB was. I'll certainly feel more confident about using this setting in low light conditions. I also tried + 18 dB with and without NR (at high, middle and low) and found that while NR did reduce noise, it introduced some other artefacts that I did not quite like. If I were in a situation that demanded that I use the +18 setting I would probably not have NR on. However I'm treating this test as preliminary and want to run more detailed tests on anything above +6 dB with and without NR.

February 13th, 2005, 07:10 AM	#10
Greg Boston Wrangler Join Date: Oct 2003 Location: DFW area, TX Posts: 6,117 Images: 1	A. J. I would also like to thank you very much for contributing these readings. One thing I noted however was the lack of testing the +18db setting. I am particularly intererested in that one because one thing that pushed me to buy the XL-2 was the night footage shot in Vienna, Austria. The final shot in that footage was taken at +18 according to the shooter. I thought it looked pretty good. Can you re-run your tests at the +18 setting? We might convince Chris to make this a page 2 article on the main site sort of like the lense comparisons he did awhile back. Your humble XL-2 co-mod, =gb=

February 13th, 2005, 09:20 AM	#11
A. J. deLange Major Player Join Date: Jun 2004 Location: McLean, VA United States Posts: 749	Greg, The reason I didn't do +18 was that I was working with 30p where if you dial in 18 the camera sets 12. I did 60i this morning and the results for a normal scene are, for respectively -3, 0, +3, +6, +12, and +18 dB gain, SNR's of 41.73, 40.72, 39.8, 39.4, 38.1 and 35.6 dB. These show the same pattern as the 30p numbers: only a couple of dB degradation until gain settings go above 6. Note that the 60i numbers are a little better than the 30p numbers. Dark SNRs are 51.1,50.6,50.2, 49.7, 55.0 and 45.4 with the fact that the dark SNR is best at a + 12 dB gain setting being the surprise in this case. All measurements were made with the camera in the 16:9 aspect ratio.

February 14th, 2005, 01:11 AM	#12
Greg Boston Wrangler Join Date: Oct 2003 Location: DFW area, TX Posts: 6,117 Images: 1	Thanks A.J.! I do appreciate your taking the time to grab that extra data. BTW, did you ever see the footage I referred to in my previous post? Some great stuff. -gb-

February 14th, 2005, 05:08 AM	#13
Tom Hardwick Inner Circle Join Date: Mar 2003 Location: Billericay, England UK Posts: 4,711	It's interesting that Sony saw fit to have a -3dB setting on the TRV900 and the PD100, but dropped this on subsequent models. There's no such setting on the PDX10, the VX/PD or the new FX1, pointing to the suggestion that there's no other gain than a half-stop aperture increase. tom.