MTF Revealed – Part II: JVC DILA (RS35, RS20, RS1) vs DC4 DLP

December 22, 2009

In Part I of this series we examined the Samsung SP-A900B projector. This projector uses the larger .95″ DC4 DMD, it has excellent optics across the field and it is relatively free of chromatic aberrations and geometrical distortions. Since it is a single chip DLP design, there is no reduction in sharpness due to convergence issues. The Samsung was chosen for the first article with the thought that it would probably set the sharpness bar for all displays to come. With that in mind, we present Part II of this series and examine three successive models of the JVC RS series LCOS projectors and compare them to the DLP results we found in Article I. All of these JVC projectors utilize the smaller .7″ LCOS chips and are three panel designs and so conventional wisdom would preclude them from reaching the stratospheric MTF numbers achieved in the part I Samsung article. Will conventional wisdom prevail, or do we have some surprises in store for the reader? As you will see as you read on there is a little of both.

Sharpness on JVC DILA based projectors has historically been described as being relatively soft compared to other projector technologies. This has been improved over the years, culminating in the RS35 which is generally regarded as being the sharpest of the RS series projectors. While most people agree that the RS35 represents an improvement in sharpness over previous RS series models, in absolute terms, the various reviews of the RS35 have been all over the map as far as describing the sharpness of the RS35 compared to other projectors. Some reviews describe it as being very sharp, while others say that it’s only average in this area. Some people note that it resolves all of the available detail, but they still curiously find the image to be soft. Putting these comments together as a whole, there seems to be a wide disparity of opinions and most people tend to struggle when describing the sharpness differences of LCOS compared to other display technologies. This project was undertaken in the spirit of attempting to find some objective criteria to base these sharpness comparisons and to hopefully shed some new light on the sharpness discussion. In the process of researching this article, there were many surprises encountered along the way. It became clear, early on, that this area is much more complex than many people think. Hopefully after reading this article, readers will come away with a new understanding of some of the elements that determine image sharpness, and they may have a new appreciation of the subject.

The reader should also read Part I of this series, as it provides a discussion on how inter-pixel contrast is measured, and how MTF in turn is calculated. As was mentioned in Part I of this article, the reader should not make the assumption that all projectors will share the same exact numbers. As is the case with any projector metric, the results will vary from unit to unit. In some cases the sharpness differences may also vary significantly across the field due to convergence and focus irregularities, so these results may not even apply equally across the entire image. This however, doesn’t mitigate the usefulness of the data presented here. As we will see, there is still some useful information that can be gleaned from this project. As was also mentioned in article 1, what is being measured is inter-pixel contrast which to my knowledge is the first time this has been done independently, outside of a manufacturers lab.

Enter the JVCs

Note: None of the units used in this article were review samples. As such they might be representative of the typical unit that a user may receive rather than a cherry picked gem.

For all of the results used in this article, the convergence was optimized at the location being measured and each unit was carefully focused. Often this required refocusing as some of the adjustments have the potential to defocus the lens. The following is the subjective impression from each of these displays.

JVC RS35 – The RS35 measured for this article was found to have outstanding convergence vertically and good convergence horizontally. There was a slight red misconvergence in the upper right-hand section of the image, but this didn’t affect any of the readings as the probe position used was always in the middle of the image. The pixel grid was very well delineated and very noticeable. This was the sharpest LCOS projector that I’ve yet seen and the screen door effect (SDE) made it look very much like a DLP.

JVC RS20 – The RS20 used in this article had excellent vertical convergence – better than other RS20s that I’ve seen first-hand. Horizontal convergence was good but not as outstanding as in the vertical direction.

JVC RS1 – The RS1 was the first of the high contrast projectors from JVC. When it was first released there were many comments that the lens could have been a little better in that chromatic aberrations were visible and the focus of some of the colors was off which leads to color bleeding on the edge of some images. The RS1 used for this review has many of these issues in addition to less than optimal convergence which varied across the field as did the sharpness in general.

Vertical Sharpness

MTF is often measured with vertical line pairs and so it’s not by chance that vertical MTF was the first sharpness measurements that were performed for this article. Vertical sharpness was determined by measuring inter-pixel contrast for a series of test patterns using vertical on/off columns (line pairs) that varied the line width (and spacing) by 1, 2, 3, 4, 5, 6 and 7 pixels. We should remind the users that the cut-off (Nyquist Limit) frequency represents the smallest line pair pattern which is achieved by setting one column of pixels on (white) and the adjacent column of pixels off (black). All three projectors were measured with neutral lens shift and a throw distance of roughly 15 feet (mid throw for the JVCs and at shortest throw for the Samsung) and displaying an image 8′ wide.

As was done in Part I of this series, a line scan camera was used to capture the relative intensities of the white (on) and black (off) pixels. A comparison of line scans from white line pairs between the Samsung A900B (.95″ DC4 DLP) and the JVC RS35 (3 x .7″ LCOS) is shown below.

one pixel on/off line scans - Samsung left, RS35 right

Fig 1 – One pixel on, one pixel off line scans (white) – Samsung left, RS35 right

These line scans show a shocking result in that the RS35 has a slight inter-pixel contrast advantage over the Samsung. As was explained in Part I, the peak in the line scan images above represents “on” white pixels and the trough between pixels represents an “off” black pixel. MTF for this specific line pair pattern of one pixel on and one pixel off (@ nyquist limit) is calculated using the formula:

MTF = (Imax – Imin) / (Imax + Imin)

As we can see from the line scans above, the peaks are higher on the RS35 than on the Samsung A900B and the valleys are also deeper. The distance between the peak and the valley is traditionally called, “The Depth of Modulation” and this directly shows a higher MTF (if only slight) for the JVC than the Samsung.

As a side note, one can also see a slight bulge near the bottom of each trough on the right side of the JVC. This is something that was confirmed with visual inspection and will be discussed later can also be seen on photos taken with single pixel line pairs later in this article.

The process of measuring the peak and troughs and calculating MTF was repeated for many different line pair sizes using all of the projectors in this comparison and the data was plotted and is shown below:

jvc_vertical_mtf

Fig. 2 – JVC MTF Plots

At the Nyquist Limit, the RS1 comes in at 66% MTF, the RS20 (which had excellent vertical convergence) at an MTF of 82% and the RS35 at a mind-blowing 91%. These results are truly astonishing, because everyone knows that DLP is sharp and a 3-chip LCOS (with small, .7″ chips at that) can not achieve the same level of performance as DLP. Right? With this thought in mind, I was initially stunned at the result and assumed that there must be a measurement or instrumentation problem. After spending much time remeasuring and reverifying the same results, checking the linearity and accuracy of the CCD sensor, I found these results to be repeatable, consistent and accurate. After looking into this in more detail, including visually inspecting the pixel grid while using MTF test patterns, Windows desktop images and video content, I slowly came to the realization that these numbers were correct but there were other factors at play that help to explain these surprising results. Some of these factors will be described in detail and step by step in this article.

Incidentally, as we see from the chart above, the three curves quickly converge to close to the same MTF numbers at right around 0.2 lp/mm which corresponds to line pairs that are two pixels wide and spaced two pixels apart (50% of the Nyquist Limit) displayed on an 8′ wide screen. At 0.2 lp/mm, the results between JVC projectors are very close to one another and also very close to the same number delivered by the Samsung in Part I on this article. This tells us that most of the MTF differences between all of these projectors happens at only the highest frequency representing single pixel test patterns.

The MTF measurements above were done with white which utilizes all 3 panels simultaneously on the JVC projectors. If we examine all three panels individually, we get the data shown below.

jvc_vertcal_mtf_rgb1

Table 1 – Individual RGB MTF (JVCs)

These results are even more surprising, because they represent MTF measurements from one panel only, and it is therefore more of an apples-to-apples comparison with a single chip DLP. Here we see even higher numbers than we were seeing with white and higher numbers compared to the DC4 equipped Samsung. Clearly, something very interesting is afoot….

The first thing that was done after measuring these results is to examine the pixel on/off test pattern carefully to see if the data holds up after visual inspection. The images are shown below:

1_pix_all_proj

Photo 1 – 4 line pairs each from RS1, RS20, RS35 and Samsung A900B (left to right respectively)

These photos are unretouched, color photos taken from a Canon 50D camera off of an image displayed on a Stewart Studiotek 130 screen. As can be seen from the photos, the RS35 does indeed look quite a bit sharper than the RS1 and RS20 and it also shows less color irregularities. Looking at the Samsung, we see better pixel definition, but as we mentioned in Article I, the definition of MTF looks at only the light intensity differences between on and off pixels. The Samsung pixels are brighter, but since the exposure of these photos varied and the black pixels are so dark, there isn’t much that can be inferred from the inter-pixel contrast from these photos.

One interesting thing to note from these photos however, is that the width of the DLP pixels seem to be a constant – the ”on” pixels and the “off” pixels are roughly the same width. Looking at the RS35 pixels we see a different situation, where the “on” pixels appear thinner than the ”off” pixels. Pixel spot size is a well known trait in CRTs that often was a limiting factor in the resolution of CRT displays. We will discuss the reasons for the irregularity of the pixel size later in this article, but for now let’s just point out the possibility that a non-uniform pixel size may have some attributes that can affect inter-pixel contrast.

Incidentally, the line scan photos in Fig 1. also shows a visible “ghosting” on the right edge of each vertical line on the RS35, as was mentioned earlier, this also shows up in the Line scan photo as a “shoulder” at the right edge of each “on” pixel. The reason for this irregularity is unclear, but as we will see later, the RS35 and LCOS in general has some surprises in store when it comes to edge transitions.

It’s also apparent from the photo that even though the thickness of the on pixels has changed in the photo above, the distance between pixels has remained the same.

The next step that was done is to check real images and compare this with the conventional wisdom assumption that DLP is in fact sharper than LCOS. The screen shots below show similar images from a Windows desktop for each projector.

RS1:

RS1_desktop

RS1 Desktop Image (Click to Enlarge)

RS20:

RS20 Desktop Image (Click to Enlarge)

RS35:
RS35 Desktop Image (Click to Enlarge)

RS35 Desktop Image (Click to Enlarge)

A900B:

Samsung A900B Desktop Image (Click to Enlarge)

As can be seen, the image from the Samsung display does in fact look sharper. The JVCs all seem to have a darker look to the text, almost as though the text is in Bold. So clearly, something is still amiss – the line scan camera results show that the JVC RS35 seems to have slightly higher inter-pixel contrast and yet the Samsung images look sharper. So it’s time to continue to dig deeper and the next place that we’ll take a look is with horizontal sharpness.

Horizontal Sharpness

Horizontal sharpness is equally important and it must be weighed as part of the overall sharpness equation. With this in mind we found the 3 JVCs to have the following MTF with horizontal line pairs at the nyquist limit (one pixel on and one pixel off):

jvc_horizontal_mtf

Table 2 – Horizontal MTF (JVCs)

Here we see strikingly similar results between all of the JVC projectors and in particular, the RS35 is merely mortal. One possible explanation for the lower MTF in the horizontal direction is poor inter-pixel contrast caused by misconvergence. By examining the MTF of the individual colors we can get better insight into this question and see if in fact convergence is an issue. This was done on the RS35 with the following results:

rs35_horizontal_rgb

Table 3 – RS35 Horizontal MTF (RGB)

This is an interesting result in that it shows that if we were to completely eliminate convergence errors, the RS35 will still not achieve the same level of MTF that we saw in the vertical direction with white. Although the green MTF is okay (but still down 5% from the vertical green MTF), the other colors are not nearly as impressive. It’s an open question as to why the horizontal MTF measured considerably lower than the vertical MTF. It may be due to the lens design, the alignment of the liquid crystal layer or some other property of the DILA panels. In the vertical direction we have seen MTF improvements in each succeeding model of projector, but those improvements are not seen in the horizontal direction. This fact and the fact that all three generations of RS projectors are so similar in this respect, seems to suggest that there may be some inherent limitation in the technology such as the alignment of the liquid crystal layer. Unfortunately there is no way to prove this conjecture and only JVC has the answer for this. One thing is clear though, horizontal sharpness is equally as important as vertical sharpness and now we have a measurable reason why our LCOS projectors are not as sharp as the vertical MTF numbers would initially imply. There are also other factors at play that will be discussed below.

Variance of White

In Part I of this series, we noted that reflective technologies like DLP and LCOS preserve their white levels very well down to a few pixels, but we noticed that LCOS dropped more in brightness with single pixels compared to DLP. This trait was further investigated in this article with the following results:

variance_of_white

Table 4 – Brightness of line pairs of various pixel widths

Here we measured the white level from a full field white pattern and also with successively smaller line patterns down to single pixels. The value of the full field white (entire screen was set to 100% white) was used to normalize the data (i.e. full field = 1.00) and the results displayed in the table above. As can be seen, each of the projectors above maintain their white levels very well down to a few pixels. The RS35 has a single pixel white level that is 79% of full field white, the RS20 at 62% of full field white and the RS1 has dropped over 50% in brightness with this single pixel pattern. In the earlier Part I article, we found the Samsung to be 94% of it’s full field value. It’s interesting to note that the RS35 is considerably better than the other two projectors in this respect and as we will see later, this same effect may help to partially explain the higher MTF of the RS35 compared to the other JVC projectors.

The significant differences in brightness at the single pixel level between DLP and these DILA projectors indicates that there is some as yet unexplained force force at play. Earlier when discussing the vertical MTF above, we noted a trait of LCOS where the pixel size is not constant and the off pixels were slightly larger than the on pixels. Here we are noting single white pixels with significantly reduced light intensity. Both of these traits may be manifestations of a property that is unique to liquid crystal displays, which is fringing of the electric field between adjacent pixels. To understand how this works a person must understand that each pixel in a LCOS display has a thin layer of liquid crystal molecules above it. The orientation of the liquid crystal determines the polarity of the light passing through it which is how LCOS displays modulate the brightness of each pixel. The orientation of the liquid crystal molecules is driven by a bias voltage contained in drive circuitry underneath each pixel. This arrangment is simlar to a bunch of adjacent parallel plate capacitors whose electric fields will interact with each other and influence the state of the capacitor next to it. In the case of adjacent LCOS pixels, one being off and the other on, the electric fields of both will interact and the response of the liquid crystal will not necessarily respect the boundaries of the pixel area. As an example, the liquid crystal molecules on the edge of an “on” pixel may be influenced by surrounding ”off” pixels which can cause the phenomenon that we saw earlier where “on” pixels were thinner than “off” pixels. Since they are thinner, they also won’t be as bright and we see that phenomenon in this discussion as well.

Interestingly enough, despite being thin and not as bright, the inter-pixel contrast on the JVC projectors with single alternating pixels was measured and found to be excellent. This can only be attributed to the fact that the dark off-state pixels have also become darker due to less “washout” from the adjacent white pixels.

(As an unrelated side note, all of these projectors show a slight increase in brightness with windowed white patterns compared to full field white. This effect was also noted when I measured low-APL intra-image contrast in this thread at AVS Forum. The difference is small, but it’s an interesting fact that seems to hold with all LCOS projectors including those using SXRD panels.)

Electric field fringing – the smoking gun?

At this point, a skeptical reader may be thinking that the results that we’ve just discussed from both the variance of white and the changing geometries between on and off state pixels may be due to E-field fringing, but maybe not. Perhaps it’s an optical phenomenon such as chromatic aberration or some other unexplained phenomenon. There is additional confirmation of this theory that comes in the form of the information from the line scan camera used to determine inter-pixel contrast. All of the line scan camera plots show something interesting happening on the edge pixels of a white cluster of pixels next to a dark cluster of pixels. The edge white pixels are heavily influenced by the dark pixels and end up being both thinner and less bright than they should be and they are clearly not as properly formed as the center white pixels.

To see what I mean, let’s first take a look at what a line scan image looks like from a DLP (Samsung) with a 5 pixel on and 5 pixel off pattern:

DLP Line Scan (Samsung A900B) - 5 pixels on and 5 pixels off

Fig. 3 – DLP Line Scan (Samsung A900B) – 5 pixels on and 5 pixels off

From this scan we can clearly see the peak in light intensity from the 5 on-state pixels which look sort of like “fingers” in the image above. The large trough between the 5 peaks represents the 5 off-state pixels. The fingers themselves are caused by the delineation of the pixel grid (screen door effect). The key thing to see from this image is that the light intensity of each of the 5 fingers is very similar and only the first (left) finger is slightly diminished compared to the others.

Next let’s take a look at a similar scan of 6 on and off state pixels from an RS20.

Line Scan - JVC RS20 6 on (white) pixels and 5 off (black) pixels

Fig. 5 – JVC RS20 line scan – 6 on (white) pixels and 5 off (black) pixels

Here we can see that center 4 on (white) pixels are well defined, but the outer edge pixels have lost their definition and are both dimmer and dimensionally thinner than their inner counterparts. Visually inspecting these patterns also shows chromatic effects at the edges which is something that we’ll address later. There are two key things to take away from this image: 1) The inter-pixel contrast (MTF) of the 6 pixel line pairs is not affected by what has happened to the edge pixels. This is because the calculation we are using for MTF = (Imax – Imin) / (Imax + Imin) relies solely on the peak brightness of the 6 pixel line pairs and the dip in brightness of the 6 pixel trough (black pixels). 2) Even though MTF is not affected by what has happened to the edge pixels, in real terms there has been a loss of definition and detail in the image (and also apparently an introduction of color artifacts). The abrupt edge that we saw with the Samsung has been smoothed which creates a smoother and more fluid looking image. The other thing to note is how this phenomenon ties directly into what we observed earlier with the variance of white and also the pixel geometry change that we saw with single alternating pixels.

This phenomenon was observed in all line scan images of the LCOS projectors and is an inherent trait of the technology. It happens in the vertical and horizontal direction and in various forms for the RS1, RS20 and RS35. As another example, let’s take a look at a 5 pixel vertical line scan from a JVC RS1:

JVC RS1 Line Scan - 5 pixels on and 5 pixels off

Fig. 6 – JVC RS1 Line Scan – 5 pixels on and 5 pixels off

As we can see from this line scan the edge pixels on the RS1 have significantly lost their shape and light intensity. When we compare the line scans from the RS1 and RS20 we see a noticeable improvement in the RS20 compared to the RS1. This seems to be a clear indication that there is more going on with sharpness than simply optical and lens performance. With every release of a new RS series, there have been comments by JVC about panel changes that yield improved on/off contrast performance and interestingly enough we have also seen sharpness improvements as well (at least in the vertical direction). There have also been rumors of improvements to the RS35 panel design and from the earlier MTF data we saw that the RS35 has higher vertical MTF than the other JVCs, so let’s take a look at the line scans of the JVC RS35.

The JVC RS35 – A different animal?

When we take a look at vertical line scans from the JVC RS35 we see something very interesting and unique to the RS35. Let’s take a look:

Line Scan from JVC RS35 - 5 pixels on and 5 pixels off

Fig. 7 – JVC RS35 Line Scan – 5 pixels on and 5 pixels off

Looking closely at this line scan, the reader should notice that the first pixel that follows a black to white transition is very well delineated. It has almost its full width (although you can still see that it is slightly thinner than the center pixels if the thickness of the “fingers” is measured). Most importantly, this edge pixel has been able to maintain it’s full light intensity. Looking at the white to black edge pixel however, we continue to see the same effect that we saw earlier with the RS1 and RS20. It is thin and it’s light intensity has dropped significantly.

This phenomon with vertical edge pixels was seen in all line scans on the RS35 and it seems to point to a device improvement on the RS35 panel design. Optical/lens design changes alone don’t seem that they would explain this behavior because it is not symmetric and this lack of symmetry was something else that we noted at the beginning of this article when discussing the “shoulder” on the right side of vertical single pixel on and off test patterns. This is speculation of course and only JVC will know the answer to this, but something is clearly different on the RS35 compared to the earlier RS1 and RS20 projectors. The panel change that improves the edge pixels that we are seeing here would also help to explain why white levels on single pixel test patterns on the RS35 are brighter than earlier models. It also goes a long way towards explaining the improved vertical MTF numbers (with single pixel on/off patterns) and also the visible improvements in sharpness that we noted earlier.

If we take this speculation one step further, there is a possible explanation for this improvement that can be found in literature. Unfortunately, publications that describe improvements to panel designs are becoming increasingly rare. In 2004 however, JVC and Aurora released a ground breaking white paper that described the technique and benefits of adding a digital drive to LCOS displays. Not long after that paper was released, JVC moved to a digital backplane for all of their panel designs and this is the design that is in use today for all generations of RS series projectors. The digital backplane paper can be read here: sid-06_04.

As described in the white paper, rather than drive a pixel with a constant voltage during a frame time as was done with earlier analog designs, the digital panel design allows the frame time to be subdivided into smaller intervals, each with a unique drive voltage, PWM (pulse-width modulation) techniques could then be employed as a drive technique rather than a constant bias. As described in this paper, one key benefit of this design is that early sub frame timing can use a high initial voltage which can be used to overdrive the liquid crystal molecules and force them to respond more quickly. Taking that technology a little further, it seems like it could be possible to also set the drive voltage higher in order to reduce the effects of e-field fringing that can happen when a white pixel is adjacent to a dark pixel. Since the e-field fringing is deterministic (the state of the surrounding pixels is something that could be known), it seems plausible that a compensation circuit could be developed which would use this information to counteract the fringing effects by increasing the bias voltage to the pixel. As I mentioned, this is all complete speculation, but the fact that the behavior is not symmetric seems to directly point to the panel as the only possible explanation rather than to other factors such as the lens and optical design.

When examining horizontal lines of pixels we see the same diminution effect on all edge pixels regardles of white to black or black to white transitions. It’s unknown why only the vertical edge pixels that follow a black to white transition seem to be improved on the RS35 and why this improvement wasn’t incorporated in other edge transitions. If the improvement was brought on by a design change to the panel drive technology as we’ve speculated, it’s possible that adding it along both axes and with both transition types (black to white and white to black) would increase the complexity and density of the drive electronics beyond what is currently possible. If this were the case, then we may see similar improvements on the horizontal axis and with both transitions in the future as new IC process improvements allow for greater density and complexity due to Moore’s Law.

Visible Effects of E-Field Fringing

Next let’s take a look at real images and see if the results that are clearly evident in the line scan data can be seen in actual images.

The image below shows a vertical line made from 5 white pixels on an RS20 with excellent vertical convergence. In this photo, we can see color artifacts in the edge pixels that may be due to convergence or possibly related to the problem we saw in the line scans above. Both edge pixels seem a little darker than the center pixels, particularly the right most pixel.

rs20_5_v_pix_white

Photo 6 – 5 white vertical pixels (RS20)

Next let’s take a look at the same pattern with an RS35. Interestingly enough we see none of the edge color arifacts, and we see the right most pixel being darker and thinner than the other pixels while the left most edge pixel seems as bright and as wide as the other pixels which is confirmation of what we saw with the Line Scan Images.

Incidentally, it’s worth pointing out that obtaining good quality photographs is much more difficult than obtaining good quality line scans. Focus is critical and it’s much easier to focus using the Line Scan camera than by eye. Since the camera is about a foot in front of the screen, the projector must be refocused by eye onto the screen. Then the DSLR camera must be focused carefully. The camera optics and screen surface also play a role. It’s for all of these reasons that I think the line scans represent the most precise measurement technique. What is obvious from the line scans is more difficult to see with photographs and nearly impossible to see with the naked eye.

Photo 7 - 4 white pixels (RS35)

Photo 7 – 4 white vertical pixels (RS35)

Now that we’ve discussed field fringing and seen some interesting geometry and light intensity changes with LCOS, let’s go back to the desktop images that looked sharp on the LCOS projectors. If you will recall, text on all of the LCOS projectors looked almost as if they were in Bold font. There is an interesting explanation for this that we can see below. Unfortunately neither of these photos came out that well, but if you look carefully at the white to black transitions, you can clearly see a large reduction in the size of the white pixels. Black to white transitions don’t seem to have this same limitation, although on the RS20 you can see the same cyan tint that we discussed earlier with 5 pixel vertical line pairs.

rs20_text

Photo 8 – RS20 Desktop Text

By comparison, the RS35 does not seem to have any of the vertical color artifacts that we saw with the RS20 image above, although horizontal color artifacts above and below the text are visible. Black to white transitions are abrupt, although white to black transitions continue to “eat” away a large part of the white pixel area. Both of the transitions can probably be seen best in the single pixel that appears between the ‘o’ and the ‘u’.

rs35_text

Photo 9 – RS35 Desktop Text

Looking at the A900B photograph (a better photograph by the way), we see none of the color artifacts and the white to black and black to white transitions are well defined and the pixel geometries are preserved.

a900b_text

Photo 10 – A900B Desktop Text

In summary, it’s very clear from these images that there is more to sharpness than the light intensity of black and white pixels that comprise an image. These photos show the same diminution of the bright to dark edge pixels that we saw earlier. This diminution distorts the geometry of the pixel and is at least partly responsible for the darker (Bold) effect to the text that we commented on earlier. This same effect may also be responsible for some of the color artifacts that we’ve noted earlier. With edge pixels that follow a black to white transition, we see that the edge is well defined and there is little to no diminution of the geometry of the pixel. Another thing that we notice from these photos is the color fringing in the horizontal direction on the RS35 which we expect to see as we’ve already described how the MTF is lower in the horizontal direction than the vertical.

Perception of SDE (Screen Door Effect)

In Article I, we discussed how the perception of the pixel grid can give the impression of sharpness beyond the resolution of the display because the eye is seeing finer detail (the pixel grid) than the pixel itself. The line scans above provide a great opportunity to objectively measure the pixel grid and assign a metric that we can use for comparisons. By looking at the height of the fingers, we can easily see for example that the grid is less pronounced on the RS1 and RS20 and it is significantly more pronounced on the RS35 and Samsung A900B. Measurements of the brightness differences from the pixel grid itself should translate directly into how easily it would be for a person to perceive the grid. By comparing the brightness of the base of the “fingers” to the tip of the fingers we can calculate the depth of the light drop off caused by the grid and this was calculated and placed in the table below:

sde

Table 5 – Brightness of grid vs Pixel brightness

As we can see from this table, the pixel grid on the JVCs seems to have grown more pronounced with each new product. The RS35 has a significantly more pronounced pixel grid and the brightness difference between the grid and the pixel is very similar between the RS35 and the Samsung. Interestingly enough, the “improvements” that we are seeing here happen with interior pixels where the adjacent pixels all have close to the same light intensity and E-field fringing would be less of an issue. It’s hard to say if the increased degree of SDE is due to a change in the pixel structure or the aligment or formulation in the liquid crystal later. It’s an interesting observation nonetheless and one that effects the perception of sharpness and places the RS35 into a similar league as the DC4 DMD equipped Samsung (at least in this respect).

Putting Sharpness into Perspective

This article has taken a close look at sharpness at the pixel level, but it’s important to realize that with video images viewed as a whole, the sharpness differences will be very subtle. Most of the video that we encounter is downsampled and compressed and devoid of much of the high frequency single pixel content that we see with a windows desktop and so the hard edges that we examined closely are not as prevalent. What was readily apparent from a close examination of a Windows desktop is much more difficult to spot with video content. As a case in point, the images below were taken from a scene from Chronicles of Narnia: Prince Caspian. This scene seems to contain fine detail from the lion and girl’s hair, so the thought is that we might be able to see some differences in display technologies with his shot. Zooming into the area by the girl’s face we see these images:

narnia_source1

Chronicles of Narnia: Prince Caspian (Click to Enlarge)

rs1_narnia1

RS1 Closeup (Click to Enlarge)

rs20_narnia1

RS20 Closeup (Click to Enlarge)

rs35_narnia1

RS35 Closeup (Click to Enlarge)

a900b_narnia1

A900B Closeup (Click to Enlarge)

As we can see from these photos, with video content, it is very difficult to detect any differences related to resolution. If you look closely at the dark line from the girls closed eyelid there may be some slight differences (the RS1 looks darker and less defined).

(note: the color differences should all be ignored. Some of the color differences may be due to the camera and the RS35 and A900B had not yet been color calibrated)

Summary – Putting it all together DILA vs DLP

As we’ve dug deeply into the sharpness differences between DILA and DLP we’ve found some surprising results. The first is that inter-pixel contrast (MTF) with 3 DILA panels can in some circumstances exceed that in 1-chip DLP, even though DILA uses smaller .7″ vs .95″ chips and the DILA uses 3 panels rather than 1. This alone is a shocking result and flys in the face of conventional wisdom, but despite our initial disbelief and skepticism, it does seem to hold up under close examination. The second thing that we’ve found is that inter-pixel contrast alone does not completely determine sharpness (perceived or real), especially when clusters of pixels are being compared. With classical MTF measurements, all that matters is the peak intensity of the cluster (line pair) of white pixels and the intensity of the dark “off” pixels. With real images such as the Windows desktop text images that we saw, the behavior of clusters of pixels is important and the diminution of any edge pixels will soften an image. In this area we have seen DLP excel because its pixels are well delineated and invariant, while LCOS (DILA) is more fluid. The third thing that we’ve noticed with the DILA projectors is sharpness directionality, with vertical sharpness being significantly better than horizontal sharpness. DLP on the other hand seems to be uniform as far as sharpness in both directions and the geometry and light intensity of the pixels seems to be relatively invariant.

We’ve also noted that with LCOS displays, the panel technology itself can have a pronounced effect on the sharpness of the image, this despite the fact that the consumer focus is universally on convergence and lens quality. On a related note, much has been made of the high pixel fill ratio of LCOS panels, but as we have seen, e-field fringing and interactions between adjacent pixels can change the geometry of the pixel and so the fill ratio itself has less meaning than many people think. The fill ratio will still play a role in large areas where the pixel luminance is more uniform (blue sky for example).

We’ve seen some pronounced improvements in sharpness in some areas with each release of a new JVC model. This has culminated in the astonishingly high, vertical, inter-pixel contrast measurements that we are seeing with the JVC RS35. Some DLP enthusiasts have attributed almost mythical image quality benefits of high MTF and we are now seeing JVC DILA models reach and slightly surpass these levels. This even though some other attributes of image sharpness are not as good as their DLP counterparts. Some of these improvements seem to point to differences in the panel design, although the convergence and optics have also likely improved. Interestingly enough, we’ve also seen indications that the JVC RS35 seems to be able to compensate for some types of e-field fringing. These are preliminary results to be sure, but if they hold up, this is an exciting area and we are sure to see further improvements along these lines with future products. We also hope to measure other JVC RS35s soon to see if the MTF and fringing effects are consistent. These will be posted as follow-up articles.

Fundamentally, this project has shown that explaining and even attemping to describe sharpness differences between LCOS and DLP displays is a much more complicated subject than most people may be aware. It’s no wonder then that subjective descriptions run the gamut. Both technologies have their strengths and weaknesses and now we have a better appreciation of what exactly comprises those strengths and weaknesses. Fans of both technologies will probably find things in this article that they like as both technologies seem to have performed exceptionally well. This project has been fascinating to work on and in it’s own way has been very rewarding. If you’ve read through this article in detail, we hope that it has shed some new light on your understanding on this topic and Videovantage is proud to have provided some of the first independent glimpses into this important aspect of display technology behavior.

As this article has primarily focused on a sharpness comparison of JVC DILA and DLP projectors, we have only peripherally touched the JVC RS35. In upcoming articles we will provide more MTF information on the RS35, in particular we will show how sharpness is affected by the various iris settings, modes, lens shift options and throw options. We also hope to glean some additional data on the unique panel behavior that we described earlier in this article. If time permits we also hope to write a standard review on the RS35 and provide color and contrast information as well as subjective viewing impressions.

3 Responses to “ MTF Revealed – Part II: JVC DILA (RS35, RS20, RS1) vs DC4 DLP ”

  1. [...] Note:  Part II of this series of articles on MTF compares DLP with 3 successive generations of DILA projectors and can be found Here. [...]

  2. Tony359 on November 13, 2010 at 5:59 am

    Hi

    I love scientific approach, well done.
    But: DLP are single chip projectors, DILA are three chips. I would think that the issue you see at the side pixels are just due to slight alignment issues of the chips.
    I work in cinema industry and the same DILA ‘issue’ are seen on 3-DLP projectors. I think it’s virtually impossible to align them perfeclty, like a single DLP device.

    I would analyze scientifically a picture of 5 pixels on using a colour at a time, not white. I would expect the result similar to DLP.

    What do you think?

    Thanks

  3. admin on March 4, 2011 at 11:34 am

    Good comment! As you point out, panel alignment is a definite issue with any 3 panel projector. This topic was heavily investigated in the Part III MTF article, where line scans of each R,G and B panel was examined independently. This process allows us to examine fringing with each panel and also to examine panel alignment very precisely. I recommend that you read the article, but the end conclusion found that fringing issues presented in Part I and Part II were found to be inherent in the technology itself rather than caused by panel misalignment.

Leave a Reply