[mercurial]

I originally posted this here but decided to start a new thread here to get a bigger audience. I did a search on the forum and didn't see any other threads that seemed to match.


I recently had an ISF calibrator out to get my HLN617W tuned up. I'd been seeing issues with what I'd called "banding" in the picture (what I now know to call "solarization"). Basically the gradients in the picture (especially the bright parts), often become a block of one color. He immediately said something looked very wrong. When he attached a test pattern generator and put up a gray ramp, some of the bars were pastels of different colors (blue, pink, green) and instead of getting steadily brighter, there were breaks (it would go from dark to lighter to lighter to light to dark to lighter to lighter to light to dark instead of dark to lighter to lighter to lighter to light).


He believes that there is an issue with the video processing on this unit and it needs to be serviced. His exact comments in email where:

Quote:

The RGB gains and cuts won't fix the color issues in your gray scale. If the gains and cuts were the problem, a gray scale pattern would start out the wrong color, and generally remain that way from one gray step to the next. In your case, you have one "gray" bar that's pink, and the next one is green, and the next one is blue, and that's because of this bizarre digital error that's in the signal processing chain in the set.

Some folks on the tweak thread think it is just a calibration issue but I trust the ISF tech since he has a very good rep. He seems to think some color information is getting lost in the processing.


Has anyone else encountered this type of issue? Is it known? Was it fixed by Samsung? What repair did they do?


Thanks for any help.

 

[GSB]

Hello mercurial


I have owned a Samsung HLN437W for 11 months and noticed this issue right from the beginning. I’m glad to see an ISF calibrator acknowledge the problem. Now we need to find out if this problem exists in only a few sets, or in all of them.


I suspected the component output of my DVD player was at fault, so I hooked up my PC, which has an ATI RADEON 9500 with DVI output. I set the HLN to correctly display a 1:1 pixel ratio on DVI.


To test the set, I generated some gray ramp test patterns in Photoshop. The patterns are 720x1280 pixels (native resolution of the set), have perfect gradient (from RGB 0,0,0 to FF,FF,FF), and are saved as BMP files to avoid compression artifacts.


On my DVI LCD monitor, the patterns display with perfectly smooth gradient, but on the HLN, they display exactly as you have described (lots of pastel stripes). I used the same DVI cable for both monitors.


To investigate further, I generated pure red, green and blue ramp patterns. The LCD displays them perfectly, while the DLP reveals fairly severe banding in all three colors.


For each of the 3 colors, the banding appears in different places along the ramp, hence, when those colors are superimposed as in a gray ramp, the bands show up quite colorfully. Adjusting Brightness/Contrast in the user menu causes the position of the bands to shift around. Very bizarre!


Similar errors can be observed when using an Analog-to-Digital converter in a noisy (badly designed) circuit. However, in this case, a DVI connection completely eliminates the need for A-to-D converters.


What else could be the cause? Another conversion has to take place in the DMD controller chipset. Each shade of RGB data is converted to a PWM (Pulse-Width-Modulated) signal in order to tilt the mirrors at the correct rate (and through the correct segment of the spinning color wheel). Maybe that’s where the trouble lies.


Gary

 

[GSB]

The quote below was found through the DLP site. It contains a nice explanation of the PWM (modulated) signal used by the DMD chip in a DLP display. The transitions in a gray ramp with 1024 shades are indiscernible to the human eye. But my set displays a choppy, multi-colored ramp which is certainly not a smooth, "highly detailed grayscale image" as is claimed.


QUOTE:

"The Digital Micromirror Device at the core of DLP TV technology can modulate light much more quickly than other display ingredients.


A DMD panel's micromirrors are mounted on tiny hinges that enable them to tilt either toward the light source in a DLP projection system (ON) or away from it (OFF)-creating a light or dark pixel on the projection surface.


The bit-streamed image code entering the semiconductor directs each mirror to switch on and off up to several thousand times per second. When a mirror is switched on more frequently than off, it reflects a light gray pixel; a mirror that's switched off more frequently reflects a darker gray pixel.


In this way, the mirrors in a DLP projection system can reflect pixels in up to 1,024 shades of gray to convert the video or graphic signal entering the DMD into a highly detailed grayscale image".


Gary

 

[augmental]

I have noticed the banding as well using the gray ramp test w/ Avia, but this is on the HLP4663W. I am having an ISF calibrator coming out next week...we will see what happens I guess.

 

[mercurial]

 Here is a link to my post with pictures of the solarization on the tweaks thread.

 

[kenhdtv]

merc, how long have you had your TV, has it always had this problem, does it occur with every input source? Also, what are your User Menu settings (contr, bright, color) and what are your Service Menu settings (delay, gamma, RGB gains, RGB offsets)?


Ken

 

[JohnPM]

Any processing in the set that uses 8-bit resolution will cause obvious problems with banding in ramps (which is nearly all the processing in sets from a couple of years ago and still much processing today), and anything that treats the colours differently will lead to the multi-coloured grey ramp effect. Prime candidate for this is the set's colour controls, especially any colour temperature adjustments, though contrast adjustments will also affect banding. Gamma corrections are another problem area, though these are often 10-bit. However, on the 5085W as an example the service menu gamma setting for green is different to that for red and blue, presumably to compensate for characteristics related to the extra green segment on the 7-segment wheel. For analog inputs missing codes and nonlinearities in the input digitisers contribute to banding, on top of that introduced by the various picture controls within the video decoder, even before the signal gets to the set's own processing. Best candidate for minimising artefacts is usually to feed in RGB data from PC over DVI, as this path will often have less processing that the various other video inputs.

 

[mercurial]

Quote:

Originally posted by kenhdtv merc, how long have you had your TV, has it always had this problem, does it occur with every input source? Also, what are your User Menu settings (contr, bright, color) and what are your Service Menu settings (delay, gamma, RGB gains, RGB offsets)? Ken

I've had it since January. I don't *remember* it being like this from the start but I don't know if it is something that was always there and I started noticing it a few months ago after the "wow" factor wore off. The bulb has a little over 2000 hours on it. It does occur with every input source. I'll try to post my settings later. I've been trying to find the defaults for the 309 firmware to see if anything is out of whack but no one has responded to my request yet.

 

[mercurial]

Quote:

Originally posted by JohnPM Best candidate for minimising artefacts is usually to feed in RGB data from PC over DVI, as this path will often have less processing that the various other video inputs.

Well, the I posted were from an HR10-250 (HD TiVo) attached via DVI. When the ISF tech put up the gray ramp, his generator was attached via DVI and producing a native 720p signal so it should have had the LEAST defects of any possible picture. (I wish I'd gotten a picture of that but I think the room would have been too bright.)

 

[mercurial]

Here are my SM settings (per-input settings are for DVI). If anything has been changed, the original value is in parens:


DDP1010

POS-Y 15

POS-X 143

LAMP TOS/PHIL

DELAY 222

V-FLIP NORMAL

H-FLIP NORMAL

Gamma 0 (4)

SLR OFF


DNIe

R-GAIN 120 (122)*

G-GAIN 120

B-GAIN 111

R-OFFSET 126

G-OFFSET 127

B-OFFSET 132

PATT_SEL 0

S_BR(DDP) 246 (248)

S_CT(DDP) 100

SCA_MAX_Y 48

SCA_MIN_Y 16

Y_TH_HPF 0

Y_TH_EDGE 4

NR_SEL 2

CE_UPPER 220

CE_CUTOFF 160

ALPHAL 75

D_ALPHAL 127

TH_CORING 5

DET_GAIN 8

R_INT 64

R_H_CONT 128

R_V_CONT 128

B_RATIO 200

W_RATIO 200

W_CCT_FAC 300

SCALE_ALP 40

R_LIMIT 150

NOISE_TH2 8


*When I restored the default values after trying someone elses tweaks, I must have missed this one but I don't think that is the issue.


3D_COMB

DYCOR 1

DYGAIN 10

DCCOR 2

DCGAIN 10

CORING OFF


OPTION

AUTO POWER OFF

WB CONTROL ON

COLOR ON

WBDEFSET 0FF

LAMP LIFE 2518

MUTE_TIME 700 MSEC

DNIe DEMO ON

FAN_NUM 3

V_SLICE 50


USER MODE SETTINGS

MODE DYNAMIC

CONTRAST 100

BRIGHTNESS 50

SHARPNESS 50

COLOR 65

COLOR TONE NORMAL


Firmware T-B3K6101-309 (Sep 29 2003 10:47:58)


Has anyone looked at the screen shots I posted and can you comment on them?

 

[umr]

mercurial,


I would look at this thread. It might help you fix your problem. I suspect your effective gamma is too high, your A/D conversion may be out of calibration and your gray scale is tracking very poorly.

http://archive2.avsforum.com/avs-vb/...67#post2825967

 

[ddisplay]

Quote:

Originally posted by GSB The quote below was found through the DLP site. It contains a nice explanation of the PWM (modulated) signal used by the DMD chip in a DLP display. The transitions in a gray ramp with 1024 shades are indiscernible to the human eye. But my set displays a choppy, multi-colored ramp which is certainly not a smooth, "highly detailed grayscale image" as is claimed. QUOTE: "The Digital Micromirror Device at the core of DLP TV technology can modulate light much more quickly than other display ingredients. A DMD panel's micromirrors are mounted on tiny hinges that enable them to tilt either toward the light source in a DLP projection system (ON) or away from it (OFF)-creating a light or dark pixel on the projection surface. The bit-streamed image code entering the semiconductor directs each mirror to switch on and off up to several thousand times per second. When a mirror is switched on more frequently than off, it reflects a light gray pixel; a mirror that's switched off more frequently reflects a darker gray pixel. In this way, the mirrors in a DLP projection system can reflect pixels in up to 1,024 shades of gray to convert the video or graphic signal entering the DMD into a highly detailed grayscale image". Gary

That is interesting but I don't see how TI's numbers add up. I don't know TI's latest numbers on the mirror switch speed, but the paper below (from the TI DLP) web site is fairly recent (year 2003). I would like to know the latest numbers but I haven't seen them published (i would welcome being shown some better numbers). Also, these numbers are for an XGA (1024x768) rather than a 1280x720P device but I would not think they are that radically different.

http://www.dlp.com/dlp_technology/im..._copyright.pdf


It says that it takes about 100 microseconds to load all the mirrors. That it takes about 6 microseconds to load 1/16th of the mirrors (100/16=~6)since the DLP supports loading the array in 16 blocks. The Mirror switching time is about 18 microseconds so that it takes about 24 microseconds (18+6) to load and switch a mirror. (He also talks about a new mode they are working on that apparently will shut the mirrors off earlier but there are no details on this mode, it it not clear if this mode would save just the 6 microsecond, the 18 microseconds switching time or both).


This all sounds pretty fast until you consider what these numbers mean.


A 5X color wheel would have 5x180 (60Hz of R, G, and B) = 900 Fields per second. Doing the simple math for 1024 levels would give:


(1/900Hz)/1024 = 1.085 Microseconds which means that 24 microseconds it is about 22 times too slow.


So obviously they are not doing all 1024 levels at 900Hz.


Even at 8-bits per pixel (256 levels):


(1/900Hz)/256= 4.34 microseconds.


So obviously they are not doing this either.


I think part of the answer is that they are obviously not doing all the bit planes on all the 900 fields per second. This probably explains why people see the Rainbow Effect more in darker scenes.


Even the load time of 100 microseconds for a full field seems problematical at 900 Fields per second.


(1/900Hz)/100 microseconds = 11 bit plane loads per field or about 176 (11 x 16) of the partial loads but if the mirror flip time is 18us there are:


(1/900Hz)/18 microseconds = 61.7 possible mirror flips per field


This would give only 11 equally weighted levels per field at very best case 61.7 possible mirror "flips" per field. That is no where need what it needed to support 256 gray scale levels and way low for supporting "gamma correction" (correcting for the eye versus linear switching response).


I think the above math explains why DLP is doing "dithering" and why the Rainbow Effect is more visible in dark scenes. I think it also gives a clue as to why the DLP has such a problem with a smooth gray scale ramp.


I would be happy to have anyone improve/fix my numbers above, particularly with more up to date information. I have a background in designing computer graphics hardware so I think my computations above are OK except for not having data on the current DLP devices.

 

[umr]

I have seen many DLP's. None had posturization as bad as mercurial. I am no DLP fan, but this problem should not be anywhere near this severe.

 

[mercurial]

Quote:

Originally posted by umr I have seen many DLP's. None had posturization as bad as mercurial. I am no DLP fan, but this problem should not be anywhere near this severe.

Not that I need external validation, but THANK YOU. Now if I can just get a Samsung tech out.

 

[ddisplay]

Quote:

Originally posted by umr I have seen many DLP's. None had posturization as bad as mercurial. I am no DLP fan, but this problem should not be anywhere near this severe.

I would agree. In spite of my post about the TI fuzzy math, the DLP should be no where as bad as the pictures that were posted. The DLP does seem to have problemss with totally smooth gray scales, but not to the extent of the pictures mercurial posted.

 

[GSB]

mercurial,


After seeing your pictures, I also agree that your problem is something other than the grayscale flaws we are discussing here - or perhaps a very extreme case of this.


ddisplay,


Your calculations look pretty good to me. However, it would not be necessary for the DMD to generate 1024 shades of gray for each color field (900 per second) – only for each NTSC field (60Hz x 3 colors = 180 per second). But, according to the article you referred to, the DMD would still not be fast enough. However, the same article says, "For digital cinema (using three DMDs, one for each primary color), the fast switching enables over 14-bits of grayscale per color, producing images that meet or exceed the quality of film."


How can that be? Even at the slower speed of film, 24 frames per second, and without having to split into 3 colors, the calculations STILL do not agree. We must be missing something.


You're right about Gamma correction being another factor. That makes grayscale processing much more complicated. Since the curve is non-linear, you would need an extremely high-bit converter to obtain a smooth looking grayscale, particularly in the darkest part of the curve. But how do they get it right in my DVI LCD?


Gary

 

[ddisplay]

Quote:

Originally posted by GSB ddisplay, Your calculations look pretty good to me. However, it would not be necessary for the DMD to generate 1024 shades of gray for each color field (900 per second) – only for each NTSC field (60Hz x 3 colors = 180 per second).

It probably was not clear, but I was trying to say that selectively updating only some planes is probably why that the Rainbow effects seems to be more visible in darker scenes (my my observations and other reports). I would guess that they are supporting the long/slow and bright larger bits on every field but gradually support the lesser significant bits fewer times. Then in darker scenes when only the mid and ls-bits of the intensity are only on, the field rate gets gets lower as ones eyes adjust to the darker scenes the Rainbow effect becomes more visible


Quote:

But, according to the article you referred to, the DMD would still not be fast enough. However, the same article says, "For digital cinema (using three DMDs, one for each primary color), the fast switching enables over 14-bits of grayscale per color, producing images that meet or exceed the quality of film." How can that be? Even at the slower speed of film, 24 frames per second, and without having to split into 3 colors, the calculations STILL do not agree. We must be missing something.

It doesn't add up to me either. Even at 24 Frames per second with 3 separate devices the equation is:


(1/24)/24 microseconds = 1736 or less than 1/9th of 2**14=16384. Maybe the mirror is faster for that application but I can't imagine that it is 9 times faster.

Quote:

You're right about Gamma correction being another factor. That makes gray scale processing much more complicated. Since the curve is non-linear, you would need an extremely high-bit converter to obtain a smooth looking gray scale, particularly in the darkest part of the curve. But how do they get it right in my DVI LCD? Gary

As you may know, the idea is that you take 8-bits of input data and covert it to at least 10-bits with "gamma correction." CRT's naturally have a gamma/non-linear effect (caused mostly by the electron gun). Since the DLP mirrors switch the light directly to the eye, they are inherently linear. If you map 8-bits of data with a typical CRT gamma onto 8-bits you effectively get less than 6-bits worth of useful gray levels. Thus you want to map the 8-bits to 10- or 12-bits.


The LC material in the LCD panels probably has some inherent non-linearity in its response. I think this helps them not need as much gamma correction as say the DLP.


If after gamma correction they don't have enough bits, then the blacks and/or whites get crushed (loose detail).

 

[GSB]

ddisplay, very interesting.


Maybe LCD's inherently slow pixel response rate (and persistence) also helps smooth the grayscale a little more than DLP's extremely fast mirror transition times.


I'd love to know if any DLP is capable of truly smooth grayscale. The HD2+ DMD can evidently sustain faster mirror-switching speeds. Used with the new 7-segment color wheel, the HD2+ reportedly benefits from 10-bit processing as well. That certainly helps reduce the need for dithering in the dark scenes, but it should improve the overall grayscale performance too.


The next, highly anticipated table-top model from Samsung, the 74 series, will have the HD2+ but will revert to a 6-segment color-wheel. That would not be a bad thing at all, if the 10-bit processing remained intact. But that remains to be seen!


Gary

 

[ddisplay]

Quote:

Originally posted by GSB ddisplay, very interesting. Maybe LCD's inherently slow pixel response rate (and persistence) also helps smooth the grayscale a little more than DLP's extremely fast mirror transition times. I'd love to know if any DLP is capable of truly smooth grayscale. The HD2+ DMD can evidently sustain faster mirror-switching speeds. Used with the new 7-segment color wheel, the HD2+ reportedly benefits from 10-bit processing as well. That certainly helps reduce the need for dithering in the dark scenes, but it should improve the overall grayscale performance too. The next, highly anticipated table-top model from Samsung, the 74 series, will have the HD2+ but will revert to a 6-segment color-wheel. That would not be a bad thing at all, if the 10-bit processing remained intact. But that remains to be seen! Gary

I don't know for sure but the numbers still don't add up. They could be doing a bunch of 10-bit processing but then dithering and the like to get "10 bits" of gray scale. But of course dithering is trading spatial resolution for color depth and has obvious drawbacks.


By the way, even assuming a "base" 3 fields at 60Hz or 180 fields per second (R, G and B shown once per 60Hz) and a maximum mirror flip speed of 24 Microseconds:


(1/180)/24 microseconds = 231.5 or still about 1/4th of 10-bits = 1024.


This is best case and assuming supporting the higher significant bits at 900 fields per second does not take away from this. Now it may be that they have gotten the mirror to switch faster but TI has not (that I have seen) reported this.

 

[zefres]

I can't see the link to the pictures of this problem. Can you send me the link? I have the HLP4663 and think I am having the same issue. I can get the grayscale pretty damn close, but I see the pastel colors in there no matter which gain I change.