The Subtleties of Video Quality
Jan 1, 2002 12:00 PM, By Richard J. Welk
Most computer users expect extremely precise, highly refined monitor images. Some require it. Continuous advances in supporting technologies have allowed the computer industry to develop progressively more sophisticated video boards and very high-resolution cathode ray tube (CRT) monitors. Those developments are both meeting and driving user expectations. And demand for even higher monitor resolutions can be expected in the future.
This article will address the need for accurate signal throughput to high-resolution monitors and related video display systems. First, we'll examine monitor operations and color video functions and explain why accurate signal reproduction throughout a video system is necessary to produce the results expected at the monitor. The next section illustrates essential components of a video signal and demonstrates inadequacies in methods currently used to calculate a monitor's fundamental frequency. The maximum fundamental frequency of a system is actually much higher than suggested by these commonly used calculations — as much as 35% higher. The final section presents actual results of our lab experiments, which show that video equipment with bandwidth curves that substantially deviate from unity gain (higher or lower) through the claimed valid range will output distorted signals. Furthermore, comparing the effectiveness of two pieces of equipment requires more than knowing their -3dB fall-off points: Flatness of response is equally important, if not more so.
BASIC MONITOR OPERATIONS
MOST TELEVISIONS, ENTERTAINMENT video projectors and even LCD monitors operate under low-resolution standards that are perfectly acceptable for home applications. However, as monitor prices fall and the mass of entertainment and educational applications skyrockets, consumers and professionals expect increasingly precise, high-resolution images on their desktop PCs, monitors and dedicated projection systems. This demand for progressively higher resolutions in CRT monitors and high-end video projectors is driving the bandwidth requirements of peripheral video equipment higher and higher.
System designers and equipment specifiers should recognize that equipment installed in front of the matrix in the signal chain must be judged by the same criteria as those that follow the matrix. A matrix can output signals only as good as those supplied at the input. It is common practice to select video equipment based on the published frequency requirements of a system and the manufacturer's claimed -3dB bandwidth fall-off point. But knowing that spec, as we'll see, is not enough.
Images on a computer monitor's screen are comprised of groups of individually illuminated picture elements or pixels — very small dots of varying light intensity that are organized in rows and columns. The dots of light are produced when an electron gun energizes phosphors within each pixel.
The electron gun scans horizontally, sequentially illuminating each pixel. When it reaches the end of a horizontal line, the computer stops producing pulses until the gun has returned to begin the next line. This is called horizontal retrace. When it completes scanning the entire screen, the computer stops sending pulses until the electron gun is ready to begin again at the top of the screen. This is called the vertical retrace.
Horizontal and vertical retrace signals require different finite periods of time that are specified by the video chip or, more commonly in a graphics system, a video board on a computer's CPU. Note that retrace times are significant in determining frequency requirements.
Retrace signals are called sync signals. In component color monitors, sync signals can be sent from the video board separately as horizontal sync (H) and vertical sync (V) as in 5-component RGBHV systems. They can also be combined as a single, composite sync signal (S) as in 4-component RGBS systems. When the composite sync signal is combined with the G signal (sync on green), the resulting 3-component set can be designated RGB or RGsB.
Component Color Monitors
A monitor produces colors by illuminating pixels consisting of a triad of dots: one red, one green and one blue. Individual color intensity levels are sent simultaneously through separate lines to three electron guns (R, G and B), each of which energizes only its targeted dot within each pixel. The shape of each dot in a triad and the screen layout pattern varies among manufacturers, but most use three overlapping circular dots in a triangular layout (see Figure 1).
In early color monitors, each dot was simply on or off. These monitors could only support eight colors: black, red, green, blue, cyan, magenta, brown or light gray — all possible combinations of R, G and B, on or off. Some monitors began supporting high and low pixel intensities, thereby doubling the possible colors to 16. Today, most consumer monitors can display over 16,000,000 colors. Extremely high-resolution monitors — like some used for medical and astronomical applications — can display nearly 300 trillion colors!
These virtually infinite numbers of colors are produced in a process similar to mixing paint on a palette: A bit of red with some green added to a lot of blue produces a very specific shade of blue. The computer translates input to digital color data, which the video board converts to precise voltage levels between 0.0 volts (least intense) and 0.7 volts (most intense). These voltages are then sent to each electron gun. Color intensity is a function of pulse amplitude. Various combinations of voltage levels sent to each dot define the color value (hue) and intensity (brightness) of each pixel.
The highest-resolution images are presented on monitors capable of accurately displaying the widest range of luminance intensities at each dot on a screen with an extremely large number of pixels. When every piece of equipment installed between a video board and monitor consistently outputs accurate voltage levels for each color component, a correctly adjusted high-quality monitor will display what a user perceives as correct colors.
SUBJECTIVITY VS. OBJECTIVITY
Users intuitively process their perception of pixel values and intensities within the context of the images displayed. Based on that perception, they subjectively judge the resolution of the video as either adequate or substandard. Sometimes the precision of the voltage levels output to a monitor is relatively unimportant. Low-cost monitors are often incapable of displaying accurate images, and even high-quality monitors can be improperly aligned or badly adjusted by the user. Some consumers simply don't care about minor color variations, and others are actually incapable of discerning any differences. In fact, the palettes of today's commonly used video boards can define many more colors than most people can recognize.
Conversely, resolution standards for monitors and dedicated video projectors are high for scientific, military, communications, medical and industrial applications; and system installations for these applications are of growing importance to the contracting industry. Competition among computer and monitor manufacturers to meet the needs of these applications is a primary reason for the higher and higher bandwidth specifications in all video-processing equipment.
VIDEO SIGNAL FREQUENCIES
A monitor's resolution is basically stated as the horizontal pixel count by the vertical pixel count. But a complete definition of the resolution also requires a specification of the horizontal and vertical scan rates. Our lab computer was configured at 1280 by 1024 with a vertical scan rate of 75 Hz and a horizontal scan rate of 80,000 Hz. This means the entire screen of pixels is illuminated 75 times per second, and 80,000 horizontal lines of pixels are illuminated every second. These numbers are interdependent in that, including the off-time for vertical and horizontal retraces, in order to illuminate all 1,310,720 pixels 75 times per second, 80,000 horizontal lines must be illuminated each second (more about this later).
Ideal Pulse Streams
Ideal pulse streams from a computer video board producing three different patterns of a single color at full intensity across one horizontal line of pixels are illustrated in Figure 2. (Note that black on a monitor is a lack of pixel illumination: R=0, G=0, B=0.) A substantial number of off/on combinations can be produced ranging from all pixels off to all pixels on across an entire line. The first example (one-on, one-off) represents the most rapid transition required of the monitor. To fully energize the blue dot in the odd-numbered pixels and energize no dots in the even-numbered pixels, an ideal video board would produce a stream of alternating DC pulses as shown on the left. Alternately energizing the blue dot in two pixels and none in the next two across the entire line results in the pulse stream and pixel pattern shown in the middle. A blue dot in all-pixels-full-on command would result in the continuous stream of equal pulse, on the right in Figure 2.
Pulse Streams and Harmonics
All periodic wave shapes — including a square or ideal pulse stream — can be described by an infinite series of sine functions. To explain what this means and why it is important, it's necessary to get into a little math.
The Fourier series representation of a pulse is composed of an infinite series of sine waves that would approximate a stream of pulses. The equation for this series is:
where A is the pulse amplitude, w is frequency in Radians
(w = 2pf), and t is time (seeFigure 3)
The first sine component is the fundamental frequency and the rest are harmonics. The ⅓ sin 3wt component is the third harmonic, etc. Figure 4 illustrates the effect of harmonics in the production of a pulse stream.
As the number of included harmonics increases, the closer the resulting wave approximates a square or ideal pulse stream. Equipment specifiers generally try to ensure that at least the third harmonic of the maximum fundamental frequency will be maintained throughout a system. Equipment installed to pass a video pulse stream to a monitor or dedicated projection system should have a bandwidth wide enough to pass as many harmonics as possible in order to support the highest resolution required by the application.
This is more than a mathematical concept. In application, if the stream of voltage pulses produced by a video board are input to a system that cannot pass the third harmonic, then only the fundamental frequency would be output to the monitor, and video resolution would be compromised.
Fundamental Frequency Formulas
The above point is partially illustrated by Figure 5. If a computer card sends something close to the square, ideal pulse shown, but the system cannot pass the third or higher harmonics, then only the fundamental frequency — the sine wave — will be passed. Moreover, Figure 5 shows that a full fundamental cycle will cover two pixels. This leads us to the calculation of the maximum fundamental frequency for a monitor, namely the pixel rate divided by two.
There are two commonly used methods to calculate maximum fundamental frequency. Method one:
w • h • fvert/2 = maximum
where w is pixel width, h is pixel height and fvert is the vertical retrace.
In the case of our lab system, this would be 1280 times 1024 times 75, divided by 2-49,152,000 Hz (49 MHz).
Method two for calculating maximum fundamental frequency is:
w • fhoriz/2 = maximum
where w is pixel width, and fhoriz is the horizontal retrace.
Again, for our lab system, this would be 1280 times 80,000, divided by 2-51,200,000 Hz (51 MHz).
Neither method is entirely adequate because neither accounts for both the horizontal and vertical retrace times during which no pixels are being illuminated. The second method is better because it does recognize that there are no pixels illuminated during the vertical retrace. But it still doesn't account for the horizontal retrace time when none are being illuminated.
In our lab system, one horizontal scan line occurs in 1/80,000 second or 12.5ms. Looking at a horizontal scan line on our oscilloscope showed that the horizontal retrace time was 3ms, leaving only 9.5ms to illuminate 1280 pixels. Therefore, the actual fundamental frequency was:
1280/9.5 • 10-6 • 2 = 67,368,000 Hz
This 67MHz result is more than 30% higher than the frequencies calculated by the commonly used methods. Our experiments verify the accuracy of this calculation.
ALTHOUGH SOME computer users don't mind minor color variations, others require precise, highly refined monitor images. Extremely high imaging standards are the norm in the fields of medicine, astronomy, military operations, business communications, industrial quality control and other applications. These needs can be met only when every piece of equipment installed before a monitor consistently reproduces the original signal output from the video board. Specifically, the third and higher harmonics must be accurately reproduced by any equipment installed either before or after the matrix, or else the precision the of matrix itself becomes irrelevant.
To determine the maximum frequency requirements of all equipment to be installed in a system, add at least 35% to account for the off-time of the vertical and horizontal retraces:
1.35 •(w • h • fvert/2) = maximum
Or, to ensure that any third harmonic present will be passed, a good approximation would be to multiply the above result by 3. This can also be directly calculated as:
2 • w • h • fvert = third
Finally, knowing the -3dB bandpass of a system is not sufficient to judge the output of video equipment. Our experiments show that the bandpass curve should be as flat as possible. The equipment with the high bandpass curve distorted both the fundamental and harmonic content of the high frequencies, which affected the leading pixels of all patterns.
The use of excessive peaking to achieve a wider bandwidth does not improve the ability of a system to faithfully reproduce a signal — it makes it worse.
Richard Welk is one of the founders of XN Technologies. He has been involved in all phases of AutoPatch design and development since the company started. Welk joined XN while working on a postgraduate degree in Computer Sciences at Eastern Washington University. He met the president of XN while conducting research under a NASA grant to develop launch support transputer hardware and software. Welk obtained his EE degree at Marquette University.
Experiments in Bandwidth Linearity
TESTING SWITICHING-MATRIX EFFICIENCY: We configured our computer to 1280 by 1024 at 75 Hz vertically and 80 kHz horizontally. These tests can be reproduced using any Windows 95+ PC, an oscilloscope (preferably 300 MHz or higher) and a 5-component (RGBHV) breakout cable.
- Exit all programs to keep as few icons onscreen as possible.
- Set your monitor to the test resolution (e.g., 1280×1024 at 75 Hz).
- Go to Screen Properties, select a basic background, then select Edit Pattern and create one of the patterns in this article. Click Done.
- Select Appearance and open Item: Desktop Color. Select one of the primary colors (we used pure blue) to get a full amplitude pulse. Click Apply then OK to close the window. The monitor background signal will now be available at the breakout cable. If you selected blue for the desktop appearance, the signal will be on the B line of the cable.
When measuring output from the computer, be sure to terminate with a 75 ohm load if you connect the output line directly to the scope. If you use a high impedance probe to sample the line output while it is connected to the monitor, a termination will already be present.
To measure the output from a piece of video processing equipment, connect the B line from the computer to that equipment and connect the B ouptut line from that device to the oscilloscope, or connect that line to the monitor if you are using a probe to access the signal.
EQUIPMENT USED IN OUR LAB EXPERIMENTS:
Dell XPS P90 with Dell monitor D1726HS and a Matrox Millenium color graphics card
HP16702A Logic State Analyzer (oscilloscope function)
HP8753ES Network Analyzer with 50 ohm - 75 ohm and 75 ohm - 50 ohm low-loss pads (for bandpass measurements)
FIRST EXPERIMENT: One-On, One-Off Pattern
We set up a background color scheme that contained only blue with every other pixel either full-on or full-off (refer to Figure 2, left). All icons were removed so that only the background was displayed. We then used an oscilloscope to view the blue signal produced by the computer. Figure A shows the signal captured by the scope. As you can see, the computer's video board did not produce a square wave at this frequency. This was the maximum resolution of the video board, and it was unable to produce the higher harmonics necessary to form a better square wave. The frequency of the signal, as predicted, was about 67 MHz.
The signal was then routed through a matrix switcher with the bandpass curve shown in Figure B — we'll call this the Brand-A matrix. The matrix output as captured by the oscilloscope is shown in Figure C.
We then passed the same signal through a matrix switcher from another manufacturer to compare. This Brand-X equipment had the characteristic bandpass curve seen in Figure D. The difference is striking, and indicative of some equipment that is currently produced. Note that the fundamental frequency is predicted to be approximately 3 dB higher than the input signal level, and the third harmonic is predicted to be close to the same as that of the input signal. These data are verified by the oscilloscope capture of the video output in Figure E.
Although the third harmonic is present, the fundamental is so high that the contribution of the third harmonic is negligible. According to either of the commonly used methods of determining bandwidth requirements (described under Video Signal Frequencies in the main article), a piece of equipment capable of passing 150 MHz would be adequate to pass the third harmonic of the incoming signal. The Brand-X matrix would have been qualified by those calculations since its -3dB fall-off point is greater than 200 MHz. Obviously, just knowing the -3 dB fall-off point would not have been enough information to predict the distorting effect of this piece of equipment.
SECOND EXPERIMENT: Two-On, Two-Off
The distortion illustrated above is not exclusive to the one-pixel-on, one-pixel-off pattern. We proceeded to set up a background color scheme that contained only blue with a two-pixel-on, two-pixel-off pattern (see Figure 2, middle). As before, all icons were removed so only the background was displayed. An oscilloscope was used to view the blue signal produced by the computer and captured the signal shown in Figure F. As anticipated, the fundamental frequency for this pattern is half the one-on, one-off pattern. Note that the video board can generate more of the harmonics at this lower frequency. Figure G shows the signal captured by the scope after it has passed through the Brand-A matrix.
For this pattern, the Brand-X matrix showed a bandwidth curve in which the fundamental more closely equaled the input signal, but the third harmonic was over 5 dB higher than that of the input signal. Figure H compares oscilloscope captures to illustrate the effect of this distortion.
THIRD EXPERIMENT: All-On
Finally, we set up a background pattern with all pixels on (see Figure 2, right). Figure J compares the oscilloscope captures of the signal generated by the video board when routed through the Brand-A matrix and routed through the Brand-X equipment.
While all three signals settle into a steady-state DC level, the Brand-X equipment with the highly peaked bandpass curve has a high spike at the beginning. This illustrates that even with more typical pixel patterns, illumination of the leading pixels is consistently distorted by the highly peaked bandwidth response. This consistent distortion is clearly illustrated in Figure K, which overlays the Brand-X equipment's two-pixel pattern output over the Brand-X all-on output.