Global Sources
EE Times-Asia
Stay in touch with EE Times Asia
EE Times-Asia > Amplifiers/Converters

Analog video signal requirements: Similarities and differences (Part 3)

Posted: 02 Aug 2007 ?? ?Print Version ?Bookmark and Share

Keywords:digital filtering? analog filtering? analog video signals? analog video transmission?

By Randy Stephens
Texas Instruments Inc.

Analog video has been around for decades and is still in use today. The original and most common video standards include NTSC and PAL. Other modern consumer analog video transmission systems include S-Video, Component Video, professional G'B'R' video and computer R'G'B' systems. This article will explore some of these analog video signal requirements and how they are similar, yet different, from each other and how to simplify the analog I/O design in these video systems.

Analog filtering
Why is filtering even used in analog video? As stated initially, it is common to convert the analog video signal into the digital domain or visa-versa. For displays and receiver boxes such as DVD-recorders, this implies the use of an ADC. For STBs and DVD players, this implies the use of a DAC. Both ADCs and DACs have images dictated by the sampling frequency. These images can "fold" down into the baseband and cause picture quality problems.

Even if a DAC or ADC has digital filtering, analog filtering external to the converter is still needed. Images will continue to exist in these systems unless filtering is done. The reason is simple: the converters are still sampling and as such, will have images.

To meet SMPTE filtering requirements, the entire system should strive to meet the standard, not the filter by itself. Many DACs already have digital filtering and interpolation filters. This by itself helps considerably. Coupled with a respectable analog filter and the SMPTE requirements can be approached.

To maintain good picture quality, these data converter images must be filtered out (see Figure 2). This is where placing an analog filter into the signal path becomes very important. Using a filter to eliminate DAC images (also a reconstruction filter) is important in a system that uses a DAC. However, is a filter required for an input device such as a display? Absolutely!

Figure 2: Standard definition video DAC images are reduced by filtering.

Because a display can connect to virtually any source, there is a possibility that the source may not have filtering, or very poor filtering. Moreover, if there is any EMI interference, the ADC anti-aliasing filter will minimize any visual problems. As a side benefit, an anti-aliasing filter will also reduce the general noise floor of the signal by reducing bandwidth.

Design factors
Eliminating both DAC and ADC images will improve visual quality. However, what filter type and how many poles should be used? What corner frequency, flatness and group delay is best for the video signal? If 10 engineers are asked this question, there probably will be 10 different answers.

As stated previously, the corner frequency for each video signal can be determined relatively easily. It is desirable to have the flattest pass-band as possible with the most attenuation near the data converter sampling frequency. Using only this requirement, an elliptic or Chebyshev type of filter comes to mind. If the only concern is amplitude flatness and attenuation, then these filters would be idealallowing any system to meet SMPTE filter characteristics. But, group delay must NOT be forgotten because the SMPTE standard also includes group delay limits and a system should strive to meet BOTH of these elements, not just one.

Group delay is defined by the change in phase (radians/second) divided by the change in frequency. The flatter the group delay, the more linear the phase change is over frequency. In the time domain, this is very important for pulse responses. The analog video transmission system can be primarily considered a time-base system. Imagine a video display changing from black, to white, and back to black for every pixel. This means the video signal voltage will go from 0-700mV as fast as possible for one pixel, and then back to 0mV for the next pixel. If the group delay variation across the frequency band changes considerably, overshooting and ringing will occur. Elliptic and Chebyshev filters will have this ringing response due to an excessive variation in group delay. On a display this can appear as ghosting or fuzzy edgeswhich is not good, even though the attenuation is very good.

As such, a good balance must be made between amplitude flatness, corner frequency, attenuation and group delay variation to achieve an acceptable video filterwhich is why there are so many different views on this subject. The general consensus is the Butterworth filter is a respectable compromise for consumer video. It has a maximally flat amplitude response, a reasonable rate of attenuation and respectable group delay. The Butterworth filter is not ideal, but usually is good enough for the system.

Examples of filters
The new THS73x3 series of integrated filter/amplifiers from Texas Instruments utilizes a modified fifth-order Butterworth filter. It has been modified by slightly reducing the Q, or peaking factor, to minimize group delay variations. The drawback is the flatness is not as ideal as a true Butterworth but the attenuation is nearly identical.

Five poles were chosen rather than four or six poles because odd-order filters have a true real-pole rather than all complex-poles that even-order filters realize. While the real-pole may be considered irrelevant by some people, real world experience has proven that a real-pole can be beneficial in active filter systems, especially when implemented by the Sallen-Key architecture. The Sallen-Key system has a high-frequency path through the system and passes high frequencies relatively easy above the amplifiers bandwidth limitation. A real-pole in the system shunts any high frequency signals to ground above and beyond the amplifier's bandwidth limitation. Thus, it helps make sure the filter remains a filter at very high frequencies.

In an effort to show the effects of group delay and amplitude flatness, another filter was simulated using the Filter Pro Program. A five pole, 0.5dB Chebyshev filter with a corner frequency (-0.5dB down) of 10MHz was simulated. Additionally, a fifth-order modified Butterworth filter with a corner frequency (-3dB down) of 8.5MHz was also simulated. Figure 3 shows the amplitude responses of each filter. The Chebyshev filter has the 0.5-ripple expected, but the "flatness" is out to 10MHz, greatly exceeding the Buttwerworth's flatness. Additionally, the attenuation rate is much higher with the Chebyshev achieving over 56dB of attenuation at the critical 27MHz point. The Butterworth "only" achieves 46dB of attenuation at 27MHz. In reality, this is generally plenty enough for a video system.

Figure 4 shows the phase and group delay responses of the filters. The Chebyshev filter has considerably more variations in the group delay compared to the Butterworth filter, especially at the corner frequency. This can also be seen in the phase responses. Keep in mind that for most systems, the absolute value of the group delay is essentially irrelevant. It is the variations in group delay that are more important.

Figure 5: The pulse response on each filter system.

Figure 5 shows the impact of a pulse response on each filter system. The pulse has a transition time of 37ns, which is in theory what a 27MHz DAC step could provide. The modified Butterworth filter, with a much smaller group delay variation, has a much improved response. The overshoot is almost the same, but the Chebyshev filter continues to have a ringing response for a considerable amount of time.

Figure 6 shows a close up view of the same pulse response. Many video systems try to keep variations less than 1 IRE, or about 7mV. The minor lines shown in the figure are 10mV. As such, the Chebyshev response will have at least 1 IRE variation up to about 480ns after the pulse was applied, as compared to the modified Butterworth "settling" at about 220ns. This can cause a negative effect such as ghosting or fuzziness.

When using active filters, keep in mind the higher the Q of the filter, the higher the amplifier bandwidth needs to be. Using Bessel or Butterworth filters, or even high order versions, keep the Q of each stage relatively low. A filter designed with Elliptic or Chebyshev responses have much higher Q values and can demand a much higher bandwidth amplifier to implement reasonably. Otherwise, the impact of the amplifier on the filter will alter the desired responses. Granted these can be designed out by modifying component values, but amplifier-to-amplifier variations start coming into play much more significantly than before.

Passive vs. active filters
Passive filters are commonly found in systems today because they can be fairly inexpensive. However, drawbacks include PCB area, extra component inventory, more assembly time, pass-band signal loss, electromagnetic influences on inductors and tolerances. Inductors and capacitors with 10 percent variations are common, especially for low-cost components. However, these tolerances can have a major impact on the filter response due to individual component-to-component variations and the fact that several poles are involved.

Monte-Carlo analysis is a useful tool to see the impact on performance for passive filters. Simulations show that there will be considerable variations in corner frequency, flatness, attenuation and peaking when using 10 percent tolerance components.

Using active filters can improve the shortcomings of passive filters. In a semiconductor process such as BiCom-3 used in the THS73x3 devices, element-to-element matching is typically very tight. It is not uncommon to see less than one percent variation from resistor-to-resistor and capacitor-to-capacitor. Keep in mind that there will be significant variations in absolute component values. It is not uncommon to see 10 percent or more, depending on the component and the type of component. This will impact the corner frequency and attenuation characteristics of the filter.

However, in an integrated active filter design such as the unity gain Sallen-Key filter used in the THS73x3, the variations in flatness and peaking can be very tightly controlled. Sensitivity analysis of a unity gain Sallen-Key filter (not be replicated here due to space constraints) shows that as long as resistor-to-resistor and capacitor-to-capacitor matching is very tight along with unity gain, essentially the only variations will be with corner frequency with no variations in Q. A variation in Q leads to significant group delay variations which is undesirable. As long as high-quality capacitors and resistors are used, and assuming the amplifiers natural bandwidth is much higher than the corner frequency of the filter, the active filter can have much better controlled characteristics than a passive filter. Additionally, an active filter typically uses much less PCB area and with only one component to procure, inventory is reduced significantly.

Another important aspect about multi-pole passive filters is that their corner frequency cannot be changed easily without becoming extravagant and costly. An active filter designed with selectable filters is very easy to implement. This may not be the most attractive feature in a CVBS and S-Video system as the filter frequency does not need to be changed. However, for a component video system or G'B'R' system, changing the filter frequency can be very beneficial due to the fact that it can be SD, ED, or HD (720p/1,080i) or 1,080p HD.

This is especially important for receiver systems that accept components Y'P'BP'R or G'B'R'. For example, a fixed 35MHz passive filter was used to allow all component signals into a display. However, what happens if a 480i or 576i SD component signal was applied to the input? A common DAC sampling frequency is 27MHz for these type of signals. If the DAC has no reconstruction filters, the images appearing on both sides of the 27MHz fundamental will come directly through the display's passive 35MHz filter. The result is no attenuation of the images and the display will most likely appear very poor.

This is also a possibility for the ED 480p/576p signal. These types of signals generally have a sampling frequency of 54MHz and the video bandwidth is 12MHz. As such, the second Nyquist zone image will start to appear at 42MHz. If the passive filter is at 35MHz or higher, there will be very little attenuation of this image, which again can lead to poor image quality.

This is where a selectable filter becomes very important, for both the DAC side utilizing the THS7303 and the ADC side utilizing the THS7353. These integrated filters/amplifiers incorporate a selectable fifth-order modified Butterworth filter which can be set for 9MHz for SD signals, 16MHz for ED signals, 35MHz for HD 720p/1,080i signals or >150MHz bypass mode for very fast signals such as 1,080p. Figure 7 illustrates this point:

Figure 7: Fixed filter vs selectable filter benefits with Y'P'BP'R signals.

For added flexibility, each channel of the THS73x3 is individually controlled. With this feature, someone can select 35MHz for the Luma channel, and 16MHz for the color-difference channels, which is acceptable based on the analog signal bandwidth requirements. One drawback is that the delay associated with the different filters will vary by the same frequency scaling. This could result in timing issues if not dealt with through digital processing.

High-end systems also benefit where phase shift and group delay are very important parameters. Here a 16MHz filter could be used for SD signals, ensuring a very smooth and flat response throughout the SD spectrum with essentially no overshoot in the time domain pulse response. This is also applicable with 35MHz filters for ED signals or bypass mode for HD signals.

Lastly, passive filtering will have significant variations in impedance over frequency. The can cause interaction issues with both DACs and ADCs. Additionally, this could lead to ringing issues if the source resistance or terminating resistance is outside of the 75 requirement. The THS73x3 active filter/amplifiers mitigate this problem. Their input impedance can be greater than 1Mohm, while their output impedance is less than 1ohm at 10MHz. This can eliminate issues with ADC kick-back issues or decoder input clamping issues.

For Part 1, please click here.

For Part 2, please click here.

About the author
Randy Stephens is a member group technical staff at Texas Instruments Inc.

Article Comments - Analog video signal requirements: Si...
*? You can enter [0] more charecters.
*Verify code:


Visit Asia Webinars to learn about the latest in technology and get practical design tips.

Back to Top