Display Artifacts and Image Quality
This is Part III of a four-part series that compares and analyzes the relative strengths and weaknesses among CRT, LCD, plasma, and DLP display technologies. In
Artifacts
In Parts I and II we measured the photometry and colorimetry of each of four test displays representing the four technologies we’re considering, CRT, LCD, plasma, and DLP. If all of the pixels in a display performed exactly as our testing reported, there wouldn’t be anything more to discuss, because images are composed of pixels. The whole would just be the exact sum of its parts. Unfortunately, things aren’t quite that simple: A considerable amount of processing is required between the image or signal source and the actual display device. All of this processing affects and modifies what each pixel ends up displaying, often adversely affecting image quality and accuracy and frequently introducing artifacts into the image.
An artifact is simply any distinct, visible feature that doesn’t belong in an image. How the different display technologies respond to these issues will be the subject of this article. While artifacts frequently occur when an image or signal is encoded or compressed for storage or transmission, we will be focusing primarily on artifacts that result from the displays themselves. Compression artifacts are a vast subject in their own right that we will be touching upon lightly.
First we’ll consider the question of analog versus digital for intensities, pixels, and signals. Then we’ll examine signal quality and accuracy, followed by discussions on native resolution and image rescaling, image noise, motion artifacts, and signal processing. The discussions will deal primarily with broad issues that affect all of the display technologies together. We’ll discuss many specialized artifacts that apply to only one particular display technology under the detailed individual technology assessments in Part IV.
There are three fundamental issues for any display–how it generates pixels, how it produces intensities, and how it processes input signals. Each one of these can exist in either the analog or digital domains. The terms analog and digital encompass a wide variety of contexts. Generally analog refers to anything that can take on any continuously variable value, while digital refers to anything that can only take on a discrete set of enumerated values. Analog is smooth but imprecise, and digital has jumps but is perfectly precise. Within each section, we will further explain and define these meanings. While most people assume that digital is somehow automatically better, that really isn’t the case. Each method has its own particular advantages and disadvantages: Either one can produce excellent image and picture quality and may be superior to the other. The details of the implementation determine the quality of the end result and the artifacts produced. We’ll explore these issues next.
Analog versus Digital Intensity
The intensity (brightness or luminance) of a display pixel can be controlled by either an analog or a digital process. Each display technology must do this in a particular way. While any display can accept both analog and digital signals, the signal must be converted to the proper native mode by the time it gets to the actual display device.
CRTs and LCDs produce their intensity scale through an analog voltage control on the device. The range of brightness produced is both perfectly smooth and infinitesimally graduated. Some signal processing may be necessary to obtain the desired gamma or gray scale, as discussed in Part II. If analog circuitry does all of this, then the display retains its pure analog nature, meaning it’s free of intensity artifacts (see below) but is susceptible to degradation in signal and image quality.
Plasma and DLP displays have only digital ON and OFF pixel states, so they must produce their intensity-scale digitally, by rapidly switching between the two states and varying the percentage of time spent in each state. If the switching frequency is high enough, the eye responds to the time-averaged brightness of the pixel. For example, for 25% brightness, a quarter of the time is spent on and the other three-quarters is spent off. In principle, it’s possible to produce an infinitesimally graduated intensity scale just like that of the analog devices above. In practice, the switching frequency is fixed, and the states are digitally controlled, so only a discrete set of intensity levels can be produced. As a result, the intensity scale is no longer smooth but rather increases as a series of steps.
The resulting jumps in intensity produce what is known as a quantization error because all intensities are forced up or down to the nearest available digital value. If the steps are very fine, the eye will not notice the lack of intermediate values and smoothness. If they’re not fine, the artificial intensity and color contours (examples of artifacts) may be visible in the image, particularly at lower intensities (see below). Since most signals are now digitally generated (including almost all analog signals), there is already a quantization error built-in to most images. We’ll discuss this later. At low intensities, the “on” time is so brief that it may be visually perceived as image noise. (See “Image Noise” and “Temporal Dithering”). In most consumer devices, the goal is to produce 256 intensity levels (8-bits per primary color or 24-bits total). Signal processing generally reduces the total number of levels actually available, so in reality, less than 256 levels are provided. This introduces additional gray-scale artifacts that we’ll discuss in “Signal Processing.”
The pixel structure of the display can also be either analog or digital. This is determined by either the inherent nature of the display technology or the implementation chosen by the manufacturer.
DLP and plasma displays have an intrinsic discrete pixel matrix format that’s fixed at the time of manufacture and cannot be changed. Each DLP pixel is made up of a micromirror and each plasma pixel is made up of three gas cells containing red, green, and blue phosphors. This type of discrete image structure is referred to as a digital pixel (or discrete or fixed pixel). Each digital pixel has a unique digital address on the display. This approach has many competing advantages and disadvantages. Digital pixels are generally organized in a regular matrix of horizontal rows and vertical columns. As a result, the screen is tiled in a repeating structure of squares or rectangles with uniform internal brightness.
This structure can become quite noticeable in some images, as an effect called pixelation. Gaps between the pixels will make pixelation even more apparent, resulting in a screen door effect. The name comes from its similarity to the appearance of mesh screen on a storm door. The gaps are noticeable in plasma and LCDs, but almost invisible for DLP (and LCoS). The higher the display resolution, the less apparent pixelation becomes. At 1920×1080, it’s generally invisible at typical viewing distances, but it can be a factor at lower resolutions, such as 1280×720.
If you have a digital image with a resolution identical to the display’s, the image can be reproduced exactly and appear perfectly sharp. If the image has a different resolution, it must be digitally rescaled to match the pixel structure of the display, which introduces errors and artifacts into the image (see “Image Rescaling,” below). If the image is analog and not digital, it must be digitized for digital displays, which introduces a different set of errors and artifacts. (See “Digitizing Analog Signals.”)
CRTs as pure analog devices don’t have any preordained pixel structure built in, so they’re free to support a wide range of image formats. The phosphor dots or stripes in direct-view color CRT monitors do introduce a grain into the image and can produce wispy moiré interference patterns (similar in appearance to what you see when looking through two layers of semi-transparent thin fabric), but they don’t impart any particular pixel structure. CRT projectors use CRTs with smooth phosphor coatings so that they don’t produce moiré artifacts. Technically, CRTs are pixel-free, but this lack of structure is still referred to as analog pixels (or continuous pixels) because they can be thought of as formed and stretched however necessary. For example, if you have a CRT monitor connected to your computer, you can select from a wide variety of resolutions that will be accurately reproduced without any pixel or rescaling artifacts.
The above is not the entire story, because virtually all CRTs are operated in a mode where images are made up of a specified number of continuous horizontal lines, called a raster. The raster is actually a property of the image structure rather than of the CRT itself, which in principle will accept whatever line structure is delivered to it. So images on a CRT are actually analog horizontally but digital vertically because of the image line structure. When the CRT’s beam size is properly adjusted, the raster is virtually invisible, but that doesn’t always happen, especially with CRTs that support multiple resolutions.
LCDs are also analog devices and don’t have any intrinsic digital pixel structure (because they’re filled with a uniform liquid crystal layer), but they are generally manufactured with a pixel matrix structure like the DLP and plasma technologies. All of the flat panel technologies behave identically in this regard. But LCDs are a hybrid technology because they have digital pixels but require an analog signal internally for the panel. The digital structure takes precedence, and signal processing for LCDs is digital.
The signal is the manner in which an image is delivered to a display and also the object of all of the processing that takes place. Most displays now accept both analog and digital signals. In principle, displays with digital pixels prefer digital signals, and displays with analog pixels prefer analog signals. Note that the signals are doubly digital or analog: A digital signal has digital pixels with digital intensities, and an analog signal has analog pixels with analog intensities. Converting between the two signal types always involves some image degradation and artifacts. An analog signal is transmitted as a voltage–typically ranging between 0 and 0.7 volts–while a digital signal is transmitted as a number ranging between 0 and 255.
Analog Signals
An analog signal has the advantage of not possessing any built-in structure, which makes it artifact free. It can contain a lot more subtle image information than a digital signal, which is constrained to a discrete pixel structure and a discrete set of allowed intensity levels. On the other hand, analog signals are, by their very nature, easily susceptible to degradation and interference. Just as with audio systems, there are significant differences in performance between components, and even subtle differences can have visual consequences. High-quality analog signal processing is now more difficult and expensive to implement than digital signal processing (even though the analog electronics are much simpler, or less complex, than the digital electronics), which is why we’ll continue to see more and more of the latter.
Analog signals typically come in the form of RGB (separate red, green, and blue primary color channels) for computers (via DB15 VGA or BNC connectors) and component YPBPR (separate luminance and encoded blue and red channels) for video (via BNC or RCA connectors). Since analog signals are easily degraded, it’s important to use high-quality signal sources (graphics boards for computers; cameras, DVDs, VCRs and set-top boxes for video). Often the biggest differentiator in performance is the quality of the analog electronics within the display itself, because of the considerable amount of processing electronics required there. We’ll discuss this under “Signal Quality,” below. High quality cables are also important, particularly for runs longer than a few meters. Long runs of more than 25 meters typically require repeaters with equalization and signal processing in order to compensate for the transmission losses. With these aids, distances of 100 meters or more are possible with little overall signal degradation.
Digital Signals
A digital signal has the advantage of absolute precision when it is generated, transmitted, or processed. Still, limitations and artifacts can result from digital’s discrete nature. But there are many advantages to using digital signals. If both the signal source and display work with digital signals, it’s usually better to use a digital rather than an analog connection. There are two reasons for picking analog instead over digital: (1) Digital often produces stronger and more visible artifacts, and (2) analog often provides a larger set of controls for adjusting the image. We’ll discuss both of these issues in detail later.
Digital signals come in the form of DVI (Digital Visual Interface) for computers and DVI with HDCP (High Definition Copy Protection) for video. (A number of other digital signal formats are in use, such as FireWire (also known as IEEE 1394). SDI and HD-SDI are production serial and high-definition serial digital interfaces that aren’t yet found on many displays or projectors.) DVI has red, green, and blue channels with 8-bits of intensity information per channel. There’s a crucial difference in the signal levels between the different types: Computer signals cover the complete 8-bit 0 to 255 range, with 0 for black and 255 for peak intensity, while 8-bit video signals operate over a somewhat smaller range, with 16 for black and 235 for peak white, with the remaining levels reserved to accommodate signals that overshoot this nominal range and for synchronization.
For accurate image reproduction, the display must provide (automatic or manual) adjustments to accommodate these range differences. (We strongly recommend a manual black-level control, see Parts I and II.) Another issue with DVI signals is that the transmission distance is typically limited to relatively short distances of a few meters (up to 5 meters in some cases), but the distance depends on the resolution and the quality of the signal transmitters, receivers, and cable. Beyond that, some form of repeater is required. After the maximum recommended distance, the image will start to degrade, first in the form of intermittent digital noise artifacts appearing in the image and then by a total loss of signal. Much greater distances are possible: Silicon Image, one of the principal developers of DVI, has demonstrated DVI transmissions of 20 meters at high UXGA 1600×1200 resolutions using chipsets specially designed for long cable runs. Even longer runs of 100 meters or more are possible using fiber-optic DVI cables.
HDMI (High Definition Multimedia Interface) is the next generation of digital interconnect and beginning to appear in the marketplace. It’s backwards compatible with DVI. From our perspective, it has two major advantages–support for up to 12-bits of intensity (4,096 levels) and typical transmission distances of up to 15 meters. (These can be extended by the same methods used with DVI.) This will significantly reduce many of the digital artifacts we discuss here. Other advantages for HDMI include a smaller connector and the inclusion of audio and control signals together with the video signal, saving you from running a separate cable.
One issue that occasionally affects DVI (but not HDMI) is the use of components that aren’t fully compliant with the official DVI specification. This problem arises because official compliance testing is not required for DVI components (it is for HDMI). When a compliance problem arises, some DVI signal sources (such as computer graphics boards or DVD players) may not work properly with some DVI receivers (displays). This leads to image artifacts–on-screen noise. The problem is more likely to appear at higher resolutions and longer transmission distances. Historically, only a small number of combinations have resulted in a problem, and it’s less common now.
Both analog and digital signals are subject to various types of errors, so they are both unable to deliver perfect image quality and accuracy for different reasons. Video and photographic images always begin in the analog world and almost always need to be converted into digital signals for transmission or storage. Computer-generated images displayed on DLPs or plasmas with digital connections are the only images that can remain purely digital end-to-end. Most computer and video hardware still connects via analog signals, so a reverse digital-to-analog conversion is necessary. Most displays are still CRT- or LCD-based and therefore actually require an analog signal internally at the display device. Also, these displays often include some form of internal digital signal processing, so almost all images spend time in both the analog and digital domains and need to be converted between them at least once, often more than once. We’ll examine these issues here and in the “Signal Processing” section later.
Digital Granularity
Though digital signals and intensities are absolutely precise, they always introduce some brightness artifacts for each pixel in an image. This results from the quantization error that comes from restricting the intensity and brightness values to a set of digital levels, instead of allowing any value in between the minimum and maximum, as analog signals do. The intensity steps apply individually to each of the red, green, and blue primary color channels, so they also introduce quantization errors into hue and saturation for all resulting color mixtures. The greater the number of digital levels, the smaller the error. If the steps are very fine, the eye will not notice the jumps and lack of smoothness. If they’re not, the Granularity of the digital steps will introduce false visual intensity and color contours into an image. They’re most noticeable in fine graduations of either intensity or color within an image.
The real question is: How fine do the steps need to be for these granularity effects not to be visible? As we discussed in Part II, the eye responds to brightness ratios. (Technically the correct term here is luminance ratios, but we’ll informally use the two terms interchangeably.) When comparing two values, it’s their brightness ratio that determines what your eye sees. So how small a variation in the brightness ratio can the eye detect? Over a wide range of brightnesses, it turns out to be roughly 1 percent of the brightness level being viewed.
If the change in brightness between two digital intensity steps is more than about 1 percent, then the brightness granularity can be detected by the eye. The number of intensity steps available is determined by the number of bits used to specify the intensity levels. For example, 8-bits allows 256 levels–used in most computer and video signals. A greater number of bits provides a finer intensity scale. Sometimes more bits are used for internal signal processing within a display (see “Signal Processing”). HDMI will eventually allow up to 12-bit signals for 4,096 levels, but for now, most computer and video content is restricted to 8-bits and 256 levels.
In Figure 1 we’ve plotted the percentage change in brightness per digital step from an 8-bit intensity scale (with 256 steps) up through a 14-bit scale (with 16,384 steps). They’re all plotted as a percentage of maximum brightness on the horizontal scale. We’ll discuss the meaning of the 7-bit graph later. (The plotted values depend on the value of gamma, so we’ve used the standard value of 2.20, see Part II.) We’ve again used a logarithmic graph, because present the results more clearly. You don’t need any advanced math to understand logarithmic graphs–just pay attention to the scale values. The horizontal axis goes from 0.1 percent to 100 percent of the maximum screen brightness–a dynamic range of 1000:1 in brightness. The vertical scale goes from 0.01 percent to 100 percent. A 1-percent increase, the eye’s detection threshold, is exactly in the middle of the vertical scale. Values below this cannot be detected by the eye and values greater than this can be detected and may produce visible false intensity and color contours in an image.
8-bit Granularity
From the graph, you can see that almost the entire 8-bit intensity scale is above the 1 percent detection threshold. Also, as brightnesses get lower, the granularity dramatically increases by more than a factor of 20 over the range shown in the graph. At 10-percent maximum brightness, the change in brightness between adjacent steps is 2.5 percent, and at 1 percent of maximum brightness the change is 7 percent between adjacent steps. Plenty of image content extends down to 1 percent of maximum brightness (level 32), so granularity and contouring will definitely be visible in an 8-bit intensity scale.
The graph extends down to 0.1 percent of maximum brightness, which represents a dynamic range of 1000:1. CRTs have a dynamic range 10 times more than that, and the best flat panels are now approaching 3000:1 and steadily improving. At 0.01 percent maximum brightness (not shown in the graph but corresponding to a dynamic range of 10,000:1), only the 14-bit scale is below the 1-percent threshold. But at very low light levels, the detection threshold for the eye will increase. So, in order to take full advantage of the larger dynamic ranges for the current generation of displays, it will be necessary to extend the intensity scale by at least two more bits over the current 8-bits. Hardware capable of handling 10- and 12-bit signals is starting to make an appearance and will significantly reduce granularity effects in the long-term; for now almost all image content is based on 8-bits.
A detection threshold is probably much too stringent a condition. We should instead be looking for something more along the lines of a noticeable threshold–about a few percent, depending on the particulars of the image content and the ambient lighting conditions. For example, contouring will be much more noticeable in a dark image when viewed in a very dark room. With any version of DisplayMate for Windows, you can determine the threshold values yourself, using the “Color Explore and Match” test pattern screen. This has three side-by-side panels that can be set to any 8-bit intensity or color value. First, set the slider for the middle panel to the intensity or color value you want to study. Then adjust the sliders for the other two panels to values just above and below the central panel where you start to detect an intensity or color difference. That will be your eye’s threshold for that intensity or color.
The DisplayMate Multimedia Edition has additional test patterns to help determine gray-scale sensitivity. To perform these tests, you’ll need to use an analog CRT display, because all of the other technologies generate their intensities digitally, which interferes with the experiment. (The CRT generates an analog signal from the 8-bit digital signal. The conversion is generally very accurate and shouldn’t interfere with the test.) Turn down the room brightness as low as possible, or at least down to the same value as your home cinema, and make sure your computer is set to 24-bit color (sometimes 24-bit is identified as 32-bit color), which uses 8-bit intensities.
Granularity Due to Signal Processing
The granularity effect is actually greater than the above discussion indicates, because with an 8-bit digital input signal, it’s unlikely that a digital display will be able to produce a true 8-bit intensity scale. The reason: Any digital processing causes some reordering of the perfectly ordered set of 256 intensity levels. A shift of just 1 intensity level will degrade the signal to 7-bit granularity (see Figure 1); a shift of 3 levels will degrade the signal to 6 bits; and a shift of 2 levels falls between 6 and 7 bits. One way to reduce this effect is for the display to process the 8-bit signal internally with more than 256 levels; 10-bits with 1024 levels is used in many displays. It’s a form of digital interpolation. This results in smaller processing irregularities (which result in artifacts). If the display device can actually produce 1024 intensity levels, it will deliver overall performance relatively close to 8 bits. If the display device is limited to 8 bits (often the case), the overall performance will be closer to 7 bits. If straight 8-bit signal processing is used throughout, the overall performance will typically be closer to 6 bits.
16-bit PC Video Modes
Many personal computers are operated in a 16-bit color mode, in which the red and blue channels each have 5 bits or 32 levels of intensity, and the green channel has 6-bits or 64 levels of intensity. (Note that this adds up to 16 bits.) The granularity is so large that it doesn’t noticeably degrade in displays designed for 8-bit intensity levels, so you get true 5 and 6 bit performance, which is not too far away from the effective 6 to 7-bit performance discussed above. The green channel is given an extra bit (or twice as many intensity levels) because it contributes roughly two-thirds of the eye’s photopic response (Part II). Unfortunately, that introduces a significant green caste (an artifact) into all 16-bit images, because the finer intensity scale allows green to increase half a step sooner than either red or blue.
DisplayMate for Windows has a 6-bit 64-level test pattern to demonstrate this effect. If color and gray-scale accuracy is important make sure that your system is operating in a 24-bit color mode (8 bits per primary color). (Note that many computer systems now list a 32-bit color mode, which is actually just a 24-bit mode with a dummy 8-bit pad that significantly increases speed performance because each pixel takes exactly 4 bytes of memory. It’s a speed-for-memory tradeoff. If you have a choice, always pick 32-bits over 24-bits unless you don’t have enough video memory to support it.)
A video analog signal behaves exactly like an audio analog signal. Its primary strength, which is its ability to carry very minute and subtle details, is also its primary weakness: It’s susceptible to very minute and subtle variations and degradations in performance that can easily affect on-screen image quality. While there are an almost endless set of possibilities as to how an analog signal can degrade, the most common problem is a loss of sharpness and fine image detail due to shortcomings in high-frequency response, or video bandwidth. It can originate with the signal source like a DVD player or a computer graphics board or in the display’s own signal processing electronics. The higher the screen resolution the greater the video bandwidth that’s needed. If there’s insufficient bandwidth, fine image detail will appear dim and blurry. It takes very high-quality engineering and components to deliver excellent analog high-frequency response and different units vary in performance significantly.
DisplayMate for Windows Multimedia Edition includes a proprietary DisplayMate test pattern that accurately and numerically measures video bandwidth on-screen using only your eyes (no instrumentation is needed). It works by adjusting an on-screen slider so that a checkerboard pattern of high- and low-frequency components appear equal in brightness. The measurement is on a scale of 0 to 100, with 100 being perfect. Scores greater than 100 are also possible, but this is a tell-tale sign of overpeaking–where the high-frequency response is set too high in order to artificially enhance sharpness in poor-quality hardware.
Another issue with analog electronics is that the red, green, and blue primary color channels all need to perform identically in every respect. In particular, if the video bandwidth for the RGB channels is not identical, high-frequency fine-image detail will be produced with an unbalanced set of RGB primaries. This will introduce color errors, including a different white-point chromaticity for fine detail. DisplayMate for Windows Multimedia Edition includes a very sensitive test for this effect.
Transient Response is another performance criterion related to video bandwidth. It describes how quickly and accurately the analog signal changes when going from one pixel to the next, and therefore has a major impact on image sharpness. In principle, the signal amplitude needs to change quickly (the rise-time or fall-time) and also change smoothly and accurately between the signal levels for adjacent pixels. It’s quite a challenge to make this happen, especially at high resolutions, and a fair amount of circuit tweaking is often necessary. Peaking circuits are used to improve the rise time and frequency response, and filtering circuits are used to reduce undesirable signal irregularities. Each also produces undesirable side effects: Peaking produces signal overshoots that add overbright or overdark Edge Artifacts to fine image detail and Ringing Artifacts that add one or more oscillations in brightness after fine image detail. Filtering dampens and slows down signal transitions. Both elements are needed, and they work together in an opposing compromise.
Another important issue is that the amplifier gains and the gammas for each primary color channel must be identical. If they aren’t, the gray-scale will exhibit variations in color with intensity. This and many other signal-quality aspects can be examined with the extensive suite of test patterns provided within all DisplayMate products.
When an analog signal is fed to a display that uses digital pixels, the signal needs to be digitized. Since almost all analog signals are actually generated from digital sources (computer or MPEG) it’s necessary for the display to accurately identify the underlying digital pixels within the analog signal and then accurately measure the signal level at the center of each one. Otherwise, the transition region between pixels will be digitized. The result is highly variable and will appear as on-screen noise.
First, it’s very difficult to accurately locate the beginning and end of each pixel, because they generally don’t have well-defined edges. If this isn’t done with 100% accuracy, there will be visible on-screen interference in the form of moiré patterns and digital noise, which arise because the image pixels are not all being properly identified and mapped to the appropriate display pixels. As a result, the digital image actually becomes a bit fuzzy (much like an analog CRT). In addition to the difficulty in identifying the start and end of each pixel, signal quality problems such as rise-time, overshoot, and ringing make it difficult to measure the intensity of each pixel accurately. Digital clock jitter and drift over time make this problem even harder. Other factors include the accuracy of the signal-level-measurement hardware and the number of digital levels generated. This is called the bit-depth, and is normally between 8 and 10 bits, but it can effectively be 7-bits or less if the measurement errors are large.
Most computer displays have an automatic “Auto” function (button or menu selection) that attempts to do this pixel identification automatically. “Auto” generally does a pretty good job, but seldom gets it exactly right, so most flat-panel monitors and projectors with analog connections actually operate with a bit of fuzziness and noise–a fact that will probably shock most flat panel users. (Rescaling further magnifies these artifacts.) Manual adjustment of a pixel “tracking” or “phase” Control is necessary in order to get it exactly right and produce a perfectly sharp image. (Some displays are simply incapable of a perfect pixel lock.)
All DisplayMate for Windows products include special proprietary “Pixel Tracking and Phase” test patterns that accurately adjust and test the setting. An Auto pattern helps the Auto function get as accurate a setting as possible, and a more sensitive Manual pattern helps you adjust it as perfectly as possible. Video displays generally don’t provide any manual adjustments for pixel tracking or phase, relying exclusively on internal circuitry. That often leads to the artifacts discussed above, but they’re not as objectionable or easy to see, because video images are photographic in nature and therefore relatively smooth to begin with.
Converting Digital Signals to Analog
Very few remaining sources produce true native analog video signals, so most analog video signals are actually generated from digital sources. One analog example is vidicon cameras, which produce a native analog signal, but these are now used only in specialized applications. (Most cameras are now digital CCD or CMOS based.) VCRs and older camcorders record analog signals, but they generally use digital signal processing internally, so they’re actually just another example of digital hybrids. It’s much easier to convert digital signals into analog than the opposite, as discussed above. Very high-quality analog signals can be produced by standard chipsets, but cost considerations may result in the use of lower-performance components with reduced signal quality and the same video bandwidth, overshoot, and ringing effects that were noted above.
Encoded and Compressed Signals
Both analog and digital signals are often encoded or compressed to reduce the requirements for transmitting or storing the images. Computer RGB analog signals are one of the few that don’t use any encoding or compression. DVI and HDMI are encoded for transmission but the digital signals are reconstituted exactly by the receiver. None of these signals generates any artifacts itself, though they’ll transmit artifacts contained in the source material. NTSC and PAL video are examples of analog encoding and compression; MPEG and the new Windows Media Video are examples of digital compression. The compression is lossy, so some image content is lost, which results in compression artifacts. The various compression algorithms produce varying artifacts at different compression levels. (They compete primarily on this issue.) In general, the greater the compression the greater the loss and the more apparent the image artifacts, but results vary with the algorithm. Compression produces various types of noise and intensity-scale contouring artifacts. These are very similar in visual appearance to the various display technology artifacts that we’ve considered above. They interact together, often magnifying the overall appearance of artifacts. Often the compression artifacts are visually indistinguishable from similar display artifacts.
Almost all signal sources are now digitally generated, so they have an underlying pixel format and resolution, even when delivered as an analog signal. The format is always expressed in the form of total horizontal pixels by total vertical pixels, HxV. For example, computer based signal sources typically range from 640×480 up through 2048×1536. MPEG video sources, including DVDs and digital television, are 720×480 for NTSC systems and 720×576 for PAL systems. High Definition Television signals have resolutions of 1280×720 and 1920×1080. (Though these resolutions are not technically required by the ATSC standard, all current High Definition displays support them).
Native Resolution
All non-CRT based displays have a fixed digital pixel format, referred to as the native resolution of the display. For non-CRT computer displays–now primarily LCDs–the most common resolution formats are 1024×768, 1400×1050, and 1600×1200 for 4:3 aspect ratio screens; 1280×1024 for 5:4 aspect ratio; and 1280×768 for 5:3 aspect ratio. For Enhanced or High Definition Television displays, the most common resolution formats are 852×480, 1024×1024 (plasmas with non-square pixels), 1280×720, 1365×768, 1386×788 (Sony projection LCDs), and 1920×1080–all with wide 16:9 aspect ratio screens.
When the pixel format of the image or signal matches that of the display, there is a one-to-one correspondence, so in principle it’s possible for the display to produce an image that’s a perfect match to the source image. For computer images, which typically have fine text and graphics, this correspondence is essential to maintain critical image detail. Fortunately, computer graphics boards can be configured to support almost any resolution format required by a display, though manufacturers generally support only the most popular ones. Specialized utilities, like PowerStrip, allow most graphics boards to be programmed for almost any resolution that the hardware is capable of supporting. For video, on the other hand, the agreement between signal source and display native resolution is unfortunately not good. If you compare the resolutions listed above, there are only two out of six matches: 1280×720 and 1920×1080.
Image Rescaling
When the pixel format of a signal or image doesn’t exactly match the display’s own native pixel format then the image needs to be rescaled up or down to match that format. Otherwise the image will overfill or underfill the screen. To accomplish the rescaling for a display with digital pixels, the image pixels must be digitally interpolated and then resampled to the display’s native pixel format. For example, consider rescaling a 720×480 DVD image into 1280×720 one line at a time, the simplest case. (Sophisticated algorithms will use several lines.) No matter how you try to do it, there will be periodic irregularities (that show up as visual artifacts) in mapping 720 pixels into 1280 pixels, and they will be quite noticeable whenever there is fine image detail. Note that there is always a loss of sharpness with rescaling, because image pixels are combined by varying amounts. Rescaling also produces increased pixelation artifacts.
Because rescaling produces severe artifacts, the image must then be filtered in order to help reduce the irregularities. Filters try to smooth over the irregularities while maintaining as much sharpness as possible. They work by adjusting the intensity values of neighboring pixels. Many displays provide a choice of several different filters so you can pick the best compromise based on the image content and structure and your visual preferences. Advanced filtering techniques are a very important component of high-quality rescaling. Many different rescaling and filtering algorithms are in use, so there are huge differences in the quality of rescaled images among displays. While rescaling has improved dramatically over the last 10 years, fundamental mathematical limitations restrict the accuracy and quality of a rescaled digital image, so a rescaled image will always be lower in quality than the original. This is true whether the signal is up-converted to a higher resolution, or down-converted to a lower resolution.
Since most video images are photographic in nature, they are already relatively fuzzy and smooth, so the rescaling artifacts are not as apparent as in computer graphics images. Given the resolution format mismatches above, rescaling is the rule rather than the exception in video. In many cases, both the signal source and the display are capable of rescaling the image. The most important point is to avoid double rescaling. This happens when the signal source cannot rescale directly to the native mode of the display. The display is then forced to perform a second rescaling, which should be avoided because of the additional artifacts and image degradation. For example, if you have a DVD player that can rescale the image up to 1280×720, turn this feature off unless you have a 1280×720 display, as found in many current DLP displays and projectors. If you have two or more components that can rescale the image, make sure that you try each one separately to determine which does the best job.
CRTs, as analog devices, don’t have a native resolution, so image rescaling occurs automatically once the display’s analog size and position controls are adjusted. The image is simply stretched or compressed as needed. It’s an entirely analog process, so there aren’t any artifacts (except when the raster is visible). As a result, CRT images generally look better than digitally rescaled images on digital displays. In situations where the raster is visible (because the CRT beam size is too small) it’s possible to increase the number of lines by digitally rescaling the image, but that will introduce some of the rescaling artifacts we discussed above. (Horizontal rescaling artifacts are not as severe, because a CRT lacks a discrete digital pixel structure.)
DisplayMate includes a large selection of test patterns that can be used to examine the accuracy of the image rescaling and the resulting artifacts. The Multimedia Edition’s “Moiré Montage” test pattern is the ultimate test of rescaling artifacts. It includes 12 different dither patterns that visually demonstrate how the pixels are redistributed and merged together after rescaling. Other test patterns, like resolution, focus and fine text can be used to visually evaluate the effects of rescaling.
Noise is any seemingly random variation in image content that doesn’t belong (static or time varying). Many possible sources can introduce noise. One that’s always present is the image or signal source itself. In video, noise can arise from the camera sensor, grain in the film, a poor quality transfer, or encoding and compression artifacts. Composite video signals also produce various cross-color interference patterns, a form of image noise. Computer generated images, on the other hand, generally have exceptionally low noise in analog signal modes and zero noise in digital signal modes. If noise is visible from an analog computer signal source, it’s generally the result of analog electrical interference from other analog or digital signals inside the computer.
In analog displays like the CRT, noise tends to automatically cancel out through temporal and spatial averaging because of its random nature. This makes the noise much less visually apparent. On the other hand, most digital displays will magnify the appearance of analog signal noise because they sample the signal, so the noise component becomes pixelated. Also, the noise component in an analog signal generally averages out to zero at each point in the image within a few frames’ time. But digital sampling generally won’t produce a series of digital values that average out to zero, increasing the visibility of the noise. These effects were visually apparent in all of the flat panel displays we tested. Digital rescaling further magnifies the noise. These noise artifacts are a major reason why poor signal quality images (such as standard definition broadcast NTSC video) typically look better on analog CRTs rather than on digital pixel displays. (Enhanced digital signal processing and filtering methods are closing the gap.) Once an analog-to-digital conversion takes place, no additional noise propagation can occur, though the digital signal can become distorted through signal processing.
Temporal Dithering
A source of image noise for displays that use a digital intensity control, such as plasmas and DLPs, is Pulse Width Modulation (PWM), which generates the gray scale by rapidly switching the pixels between on and off states and varying the percentage of time spent in each state. The eye responds to the time-averaged brightness of the pixel. This process is a form of temporal dithering. The switching frequency is fairly high, typically up to 60,000 clock cycles per second (60 kHz). That’s generally too fast for the eye to notice, but at low intensities a pixel will spend most of its time in the off state, so the occasional switch to the on state becomes noticeable as a form of switching noise. With this method it’s possible to produce between 8- and 10-bit digital intensity scales, with 256 to 1024 levels, at standard video frame rates. In some implementations, the PWM cycle time extends down to 24 frames per second (fps)–lower than the 60 fps rate of the video signal. This increases the number of intensity levels that can be generated but increases the visibility of the PWM noise patterns and may also introduce some visible flicker into the image.
Spatial Dithering
In some cases a digital display may not be able to generate all of the required (or desired) intensity levels. For example, in an 8-bit digital intensity scale, there may be duplicates or gaps in some of the 256 levels produced by the display device. (This is sometimes an issue over only a portion of the gray scale as the result of an irregular transfer characteristic for the display device– typical for LCD and LCoS technologies). To overcome this limitation, the display electronics is configured to automatically adjust the intensities of neighboring pixels so that their combined intensities average out to the desired values. This procedure is called spatial dithering and it increases the gray-scale resolution. Image sharpness is reduced somewhat in return for a smoother gray-scale and reduced intensity contouring. Spatial dithering is used in most plasma and DLP displays and in some LCDs, but never in CRTs, because they already produce a perfectly smooth intensity scale. The technique typically provides an additional 2-bits of intensity resolution.
The averaging process may involve only adjacent pixels, or it can encompass larger groupings of nearby pixels to produce a finer intensity scale. The dithering process introduces a form of spatial noise into the image. The dithering algorithm may involve a fixed pixel pattern, which is often more noticeable because it tends to produce repeating pixel patterns on the screen, or error diffusion. This generates a seemingly random pattern and a finer intensity scale, but also introduces more random noise into the image. In some cases displays will provide a menu option to choose among several dithering algorithms.
In some cases a perfect end-to-end digital signal train may produce an image so “clean” that all of the artifacts that we discuss here will become quite noticeable. In that case using a high-quality analog signal connection may instead produce a better looking picture, because the analog noise will soften and obscure many of the artifacts. In general, it’s a good idea to compare both methods to see which one works best. Since that varies with the source material, it’s best to permanently maintain connections for both types of signal.
Sharpness and Noise Controls
Although they are often misused, sharpness and noise controls serve a very important function in improving image and picture quality on all displays. For analog signals, the sharpness control can help to restore high frequencies (which produce fine image detail) that have been lost somewhere along the signal path. This can arise from mediocre performance in the signal source or the display electronics. Also, long cable runs and poor-quality cables attenuate the high frequencies more than the low frequencies. The Sharpness Control increases or decreases the amplitude of the high frequencies (similar to the peaking circuits discussed under Signal Quality). The trick is not to overdo it, which introduces spurious edge artifacts and magnifies signal noise. DisplayMate for Windows Multimedia Edition includes a Video Bandwidth Index test pattern that allows the Sharpness Control to be set very accurately.
In many cases the analog signal is intentionally optimized to look best on medium- or low- performance displays (that make up most of the market), which generally don’t have very good sharpness. This is done by applying excessive high-frequency peaking and edge enhancement. It may be implemented in the signal source hardware (computer graphics board or DVD player) or through processing of the image content by the content creator (or both). Unfortunately, high-quality displays with excellent sharpness will then show peaking and edge artifacts. The solution is to turn down the sharpness control from its default setting, which will improve, but not eliminate the artifacts.
CRTs benefit from a specialized sharpness control, called an aperture control. Even if the CRT display has perfect electronics, image sharpness is reduced because the CRT’s beam has a significant diameter (aperture) that smears the image detail as it’s being drawn on the screen. The aperture control helps correct for this effect.
Digital signals should not, in principle, need a sharpness control, since that function should have been taken care of by the creator of the image content. The problem is that the same digital output is used to generate the analog output signal, and that signal is often designed to be overpeaked so that the image looks sharper on lower performance displays. If that’s the case, the digital signals will also be overpeaked and also show artifacts. If you find that to be the case, then lower the sharpness control to reduce the artifacts. Turning up a digital signal sharpness control is something that shouldn’t be done, because much of the original signal information has been lost in digitizing and processing the signal, so even if it’s necessary, the results won’t be very good.
Turning down the sharpness control serves to reduce image noise. Many displays include a separate noise control that’s generally implemented through digital signal processing. Its primary function is to reduce specialized forms of noise, such as compression artifacts, that show up as image noise.
A motion artifact is any image irregularity that results from an image’s changing over time. This includes camera panning and the motion of objects within the image. Some motion artifacts originate with the signal format; others arise from the manner in which the internal electronics updates the pixels in the display. Still others are due to inherent properties of the specific display technology.
Interlacing
The most famous motion artifact is due to interlacing, where the odd and even lines in an image are transmitted 1/60 of a second apart in NTSC systems and a 1/50 of a second in PAL systems. This results in some flicker and also a structured blur (called combing) when there is motion, because the image is made up of components that were taken at different times. The solution is to use signal-processing electronics to combine the odd and even sub-images into a single image that’s shown all at once and has double the scan rate. This is called deinterlacing (formerly line doubling). Special algorithms try to compensate for motion shifts within the image between the even and odd fields. Many displays include their own deinterlace processing, but these motion artifacts actually have nothing to do with the display technology itself and are due entirely to the nature of the signal and the signal processing.
Image quality on some displays may (surprisingly) decrease when they’re fed a non-interlaced (progressive scan) signal because the video bandwidth requirements are double those of an interlaced signal. Properly digitizing a progressive analog signal (including pixel tracking) is also twice as hard. So try it both ways: If the display’s deinterlacer is comparable in quality to that in your DVD player or other signal source then there is a reasonable chance that you’ll get better results with an interlaced signal input to the display.
3:2 Pulldown and Judder
This motion artifact arises from the fact that movies are shot at 24 frames-per-second (fps) but video displays in NTSC systems operate at 60 fps (or 60 fields-per-second for interlaced displays). The current solution is to show half the movie frames three times each and the other half two times each, which is called 3:2 pulldown. This exactly translates 24 fps into 60 fps. The problem with this solution is that it produces irregular on-screen motion, because alternate movie frames are shown for significantly different periods of time. This artifact is called Judder. (PAL displays run at 50 fps and use 2:2 pulldown, but this speeds everything up by 4 percent–also an undesirable artifact.) The proper solution is to run the displays at 72 fps and show each frame 3 times (3:3 pulldown). Surprisingly, very few video displays, projectors, or DVD players can currently operate in this mode. (Some video processors can generate 3:3 pulldown signals.)
Most CRT computer displays and projectors can run at any frame (refresh) rate from 50 to 150 Hz or more, including the industry standard rates of 60, 72, 75 and 85 Hz. Almost all flat panel displays (computer or video) are designed to run internally at a fixed refresh rate of 60 Hz, regardless of the external frame or refresh rate of the signal provided. Many computer users mistakenly drive their flat panel monitors and projectors at 75 or 85 Hz because they’ve heard that it cuts down on flicker, but this works only for CRT displays. This just generates motion judder due to the same frame rate conversion artifacts discussed above. Worse, it can needlessly decrease analog signal quality because of the video bandwidth and pixel tracking effects noted in the Signal Quality section.
In order to go from 60 fps to 72 fps, all of the display’s internal clock rates need to increase by exactly 20 percent for NTSC systems (and 44 percent for PAL systems). The video bandwidth of the signal-processing electronics also needs to go up by the same amount. This is a significant performance increase, but it’s easily within the capabilities of current display technology. More video displays, projectors, and DVD players in the near future should be able to do 72 Hz and 3:3 pulldown processing.
Many display technologies with digital pixels produce some form of artifact when motion occurs in the image. One culprit is differences in how the pixels are updated from one frame to the next by the device electronics. In some digital displays, the pixels are not all updated at the same time, but rather in cycles and/or zones. For example, the top and bottom halves of the screen may be updated separately, or the odd and even lines may be updated separately. The latter is a variation on interlacing, except that the input signal to the display is progressive rather than interlaced. You’re more likely to see these artifacts in older or lower-performance plasma and LCD displays. These motion artifacts are most visible when the image contains fine detail. DisplayMate Multimedia Edition’s “Moiré Montage” with 12 different dither patterns is the ultimate test for these motion artifacts.
Motion artifacts also arise when any form of temporal dithering extends longer than the video frame time, such as in some implementations of Pulse Width Modulation, because the motion interrupts long switching cycles and thereby introduces some gray-scale contouring errors around portions of the image that are moving or changing.
These photographs displayed on a CRT (top) and an LCD (bottom) were made using DisplayMate Multimedia with Motion Edition, which moves DisplayMate test patterns in varying directions and speeds in order to look for all sorts of motion artifacts.
LCD Motion Artifacts
LCDs have their own special issues of motion smear because of the slower response time of the liquid crystal. CRTs are virtually immune from motion artifacts because they are analog devices, and the phosphors have fast response times. Compare the photographs of a moving DisplayMate test pattern in Figure 2 for a CRT (top) and an LCD (bottom). These photographs were made using DisplayMate Multimedia with Motion Edition, which moves DisplayMate test patterns in varying directions and speeds in order to look for all sorts of motion artifacts. In the photographs, the image was moving diagonally, from the lower-right to the upper-left. The photographs were made with a traditional film camera, because digital cameras can introduce their own artifacts while capturing moving images on a display. Both displays were fed identical signals. Since the LCD is brighter than the CRT, the grays show up differently in the photographs. (This doesn’t affect the comparison.) Note that these photographs are not of the Sony CRT and NEC LCD that we measured in Parts I and II.
The CRT photo looks virtually identical to a stationary image, because of the fast phosphor response times and lack of motion artifacts. But on the LCD photo, five images of the black text are visible–the most recent plus four old decaying ghost images–a result of the slow liquid crystal response time. Each image is offset from the previous one by 16.7 ms (milliseconds), or 60 Hz. The oldest visible ghost image is between 67 and 83 ms old. (The uncertainty arises because we don’t know exactly at what point within the frame cycle the shutter was activated.)
The manufacturer’s published response time specifications are for black to peak intensity transitions (rise-time), peak intensity to black transitions (fall-time) or the sum of the two (generally the total response time). However, most transitions in normal viewing are between different gray-levels rather than between the peak intensity and black extremes. The gray-to-gray (or black-to-gray or white-to-gray) transitions generally take longer because the electrical forces changing the liquid crystal alignment are much weaker (so their response times are likely to be much larger). That’s one reason the ghosts in Figure 2 last as long as they do. Note that switching to either black or peak intensity is much faster than to gray, because those transitions show much less ghosting or lag in the photograph.
Special signal processing can speed up the gray transitions (by briefly overdriving them) but that tends to produce flicker artifacts. This panel’s specified response time is 25 ms. The NEC LCD4000 that we measured in Parts I and II has a manufacturer’s specified response time of 23 ms, but its motion artifacts were much less noticeable than the panel photographed though quoted response times are similar. This emphasizes that the response time specifications can be ambiguous or misleading. (Unfortunately, we didn’t take any photographs of the motion artifacts for the displays analyzed in this series.)
A fair amount of signal processing must occur between the display’s signal input connectors (whether they’re analog or digital) and the display device. Any adjustment, correction, or change in the image requires some sort of signal processing. For example, each of the user, service, factory, and calibration controls modifies the signal in some way and therefore requires signal processing. Getting the display to produce the desired gamma and gray scale (see Part II) generally requires a considerable amount of processing.
Analog Processing
For analog signals, it’s relatively simple (in principle) to implement the necessary signal processing, because most of the controls merely involve a change in amplitude (gain) or bias (DC or black-level offset), but it takes very high-quality engineering and components to do it well. The most complex analog circuitry is involved in producing the desired gray scale and gamma, because it has to correct the non-linear variations in the display’s transfer characteristic. (Most CRTs generally don’t try to change their native gamma, and digital displays that accept analog signals will implement the gamma correction digitally after the analog-to-digital conversion.) The biggest problem is that analog circuits are imprecise, and that leads to the artifacts discussed under Analog Signal Quality.
Digital Processing
The electronics for implementing digital signal processing are considerably more complex than analog signal processing. This is the main reason most displays currently offer fewer controls with digital inputs than with analog inputs. For example, for the standard 8-bit 0-255 intensity scale that has 256 levels, merely increasing the black-level by 10 levels requires that the number 10 be digitally added to the intensity value for each pixel received by the display. The first problem is that this needs to be implemented in special image-processing hardware. The second problem is that we automatically lose 10 intensity levels, so there are now only 246 active levels on the 0-255 scale. A third problem arises if we wish to recover the now illegal values of 256 to 265 from the addition and restore them to the 0-255 scale. To do this, it’s necessary to effectively multiply all of the signal intensity values by a factor of 255/265=0.962. (This can be done with fixed-point arithmetic or a look-up table.) This is actually a rescaling of the intensity values and is equivalent to lowering the analog video gain using a contrast control. A fourth problem arises because the intensity scale has now acquired some irregularities (artifacts); the former 256 steps are now remapped and squeezed into 246 steps, resulting in some visible 7-bit intensity-scale banding and false contouring.
This same type of processing is required for each control pair that’s implemented digitally–brightness; contrast; red, green and blue drive; and bias. More complex processing is required to implement color tint and saturation controls. Finally, additional processing is required to implement the desired gray scale and gamma (generally in the