One of the most important components in any modern receive system is
the bandwidth limiting component. Most often, this is a bandpass
filter found in an IF (Intermediate Frequency)
stage of the receive system. This component is perhaps the
component that most dramatically determines the performance of the
system in the presence of other signals and thermal noise.
It should be no secret that the majority of transmitted power in an analog video signal is contained within several hundred KHz of the video carrier. In fact, the other components that contain significant power are the chroma (color) carrier (with, at most, 15% of total power) and the sound (typically with 5% or less of the total power.)
The fact that these other components are so much weaker makes them less prone to interfere with an adjacent channel signal. It also helps one understand why, on a cable system, it is important to make sure adjacent cable channels are well matched in terms of their signal strengths.
There are other implications as well: Even though the video components that are farther removed from the carrier are quite weak (as compared to the carrier itself) they carry things like color, sound, and the fine detail (i.e. resolution) of the picture. The "weakness" of these components also makes them the first to suffer degradation when signal levels are low. It also should be no surprise that these same spectral components are among those that are the most easily affected by interference from other sources.
It only makes sense that, the wider your filter, the more energy you will intercept: This "energy" could be thermal noise and maybe even other (interfering) signals. It also makes sense that if your filter is too narrow you will remove important portions of the signal you are receiving: For example, too narrow a bandpass filter on a video receiver may remove things like sound and/or color and cause inordinate degradation of video quality. Good receiver design implies that the filter bandwidth is a good match for the bandwidth of the received signal: There is no reason to have a filter in your receiver that is much wider than the signal you are trying to receive.
On Amateur Television repeater - particularly on 70cm - good filtering is even more important: The repeater site is often shared with other transmitters that operate on "nearby" frequencies (this might even include the repeater's transmitter!) Even if not on-site, other 70cm transmitters in the repeater's coverage area may present the ATV receive system with strong signals as well.
Modern consumer-grade TV receivers employ SAW (Surface AcousticWave) filters in the IF to set the desired 6 MHz system bandwidth. While these filters provide excellent bandpass response characteristics, they are limited in their bandstop rejection - that is, the extent to which signals outside the filter's response curve are attenuated. For SAW filters, this is typically 30db or so.
For broadcast and cable TV usage this filtering is good enough to provide adequate performance, even on a cable-TV system where there are adjacent channel signals present. The fact that modern demodulators uses synchronous demodulators plus the fact these other signals have carefully controlled signal levels - plus the various attributes of an analog TV signal (see sidebar) reduce the amount of energy that these adjacent signals put into various parts of the baseband of the received signal. On a shared resource (such as the 70cm amateur band) one is no longer in a "controlled" spectrum environment and one may have the ATV signal amongst other (possibly much stronger) FM (and other) signals. In these instances the 30db of stopband attenuation offered by the SAW filter in the demodulator may simply not be enough to keep those off-frequency carriers from degrading the received video: Providing additional "stopband" attenuation will help reduce the effects of these other carriers.
There are also instances where one might wish to remove portions of the IF passband that fall within the video carrier itself. One might do so to allow reception of weak signals that were otherwise not visible. These "notches" may also be used to reduce QRM from narrowband signals that might fall in the video passband.
It should be made perfectly clear at this point that any additional filtering that reduces the AM video bandwidth (this includes punching "holes" in the passband) to something less than the standard 6 MHz (i.e. removal of certain spectral components) will likely cause visible degradation of the video. It is important to consider these effects when you decide if the gains outweigh the disadvantages when implementing these filters.
Because this (the WB7FID repeater's) receiver is to be located at the repeater's receive site atop a mountain with a lot of RF, and because this receiver is the source of video for a repeater (that is, a lot of people will be seeing the output of this one receiver) it would make sense to make this receiver as good as we can make it and for this reason it is worth going to a bit of trouble to improve its performance. Since this receive system is modular (that is, it consists of a downconverter that is separate from the demodulator) it is relatively easy to insert additional signal processing modules into the IF line.
Why add filtering to the IF line instead of at the antenna? The simple answer is that filters that operate at RF frequencies (i.e. 70 cm) are much more difficult to construct than ones constructed to operate at lower frequencies - such as the IF. This does not mean that we can forgo all front-end filtering: There is no substitute for a good bandpass filter on the input of the receiver to protect it against strong adjacent signals, image response, and lightning. We must also make sure that the RF and first IF stages have sufficient dynamic range to handle all signals within the filter bandwidth.
There are several ways that an IF filter may be implemented: The easiest way is to implement the IF filter is to put it in the IF line. (That makes sense, doesn't it...) Since the television IF frequency range spans from 41 to 47 MHz (with the video carrier being at 45.75 MHz - spectrally inverted, since the local oscillator operates 45.75 Mhz above the receive signal) this means that the filter components that you use will have to operate at those frequencies.
Another way to do the same filtering is to downconvert
the 41-47 MHz frequency span to a lower range of frequencies where
of filter elements is less complicated and components tolerances are
critical: Component Qs may be much lower for the same effect, and
capacitors and coils are less lossy, allowing filter responses to be
that may be impractical (or, at the very least, awkward) at higher
After the filtering, an identical conversion (except that it is
is done to restore the signal to its original frequency.
This module performs its function by a combination of both schemes: Some filtering is done at the 45 MHz IF prior to downconversion. This "prefiltering" adds at least 20db of stopband performace to the existing SAW IF filter. A conversion is also done to a lower frequency where the "heavy duty" filtering is done at the lower (converted) IF frequency.
Description of the Up/Down converter section of the Filter Module:
[Refer to block diagram]
The 45 MHz video IF signal (from the Downconverter) enters the filter module through an electronic attenuator, a Mini-Circuits device: This is essentially a double-balanced mixer, but optimized for use as a low-distortion current-controlled attenuator. The AGC control for this device comes from the TV demodulator module. The AGC circuit helps preserve the dynamic range of the IF chain.
Following the attenuator is a MAV-11 MMIC. This provides about 13 db of signal gain as well as a source impedance for the 8 MHz L/C filter. This filter was extricated from an old VCR (one that was made before SAW filters were common) and has been modified to have an 8 MHz bandwidth. (See the spectrum analyzer plot of this filter.) Note that this bandwidth is wider than the 6 MHz bandwidth of the SAW filter built into the demodulator - and this was done for a reason: Cascading two SAW filters had a nasty tendency to clip either the vestigial sideband of the video, or the sound. Having the first filter being wider than the second alleviated this problem. It should be also be noted that this L/C filter presents a 25-30 db insertion loss - a figure that is comparable to SAW filters - thus requiring the additional amplification.
Following the bandpass filter is a MAR-6 MMIC. This amplifier provides a termination impedance for the filter as well as providing about 20 db of low-noise amplification to overcome the insertion loss of the bandpass filter. Following this amplifier is a Mini-circuits JMS-11X doubly-balanced mixer. The local oscillator (a 36 MHz computer-type oscillator module) mixes the 41-47 MHz IF frequency bandpass down to 5-11 MHz. Following the mixer is yet another MAR-6 MMIC amplifier to provide mixer termination as well as to provide a bit of overall gain.
Note that the output of the mixer (and the amplifier) contains both
the difference frequencies (5-11 MHz) as well as the sum frequencies
MHz) and it is necessary to get rid of the latter: Failure to do
so will result in another signal (also containing video) floating
that may (or may not) be passed by the various filter sections that
be added. Worse yet, this "other" signal, should it get through
filter sections, will have been affected differently (in terms of group
delay and amplitude) and corrupt the video. A relatively simple 5
element L/C lowpass filter easily attenuates this "image" (as well as
local oscillator bleedthrough) adequately.
At this point, one has a version of the video converted to 5-11 MHz and further filtering may be applied as desired.
Once the filtering is done, it is time to convert the 5-11 MHz back to the 45 Mhz IF. The 5-11 MHz signals (after having been filtered - or not...) are amplified by a MAR-6 MMIC (to overcome insertion loss incurred by the filtering) and applied to another JMS-11X mixer. Because the 5-11 MHz frequency range is well below the rated frequency range of the mixer, the IF input of the mixer is used rather than the more "traditional" RF port. This mixer, using the same 36 MHz local oscillator used to perform the downconversion, translates the 5-11 MHz IF back to the 41-47 MHz IF that the demodulator can use. Following the mixer is a MAR-3 MMIC amplifier which provides further amplification as well as termination of the mixer's output.
It is worth noting that the upconversion also results in the generation of two undesired signals: A 25-31 MHz image as well as bleedthrough of the 36 MHz local oscillator signal. As it turns out, the filtering in the demodulator itself (e.g. the pre-existing SAW filter) is more than adequate to prevent the 25-31 MHz image from causing any measurable degradation of the demodulated video. The 36 MHz bleedthrough, however, is a different story: Even though the balance of the mixer keeps this signal about 30 db below the level present on the local oscillator input pin, after amplification, the result can be a very strong signal compared to the video. To combat this, a small trimmer capacitor and trimmer potentiometer are "wrapped around" the mixer, from the local oscillator input pin to the "output" pin (which is really the RF [input] pin, as we are actually putting our input on the IF [output] pin of the mixer.) Careful adjustement of these two components allow an additional 25-30 db of nulling of the 36 MHz signal (without affecting the video, as a filter might do) putting this bleedthrough well below the level that might "bother" the demodulator.
Why use 5-11 MHz? Why not lower? The choice of 5-11 MHz was made partly because a 36 MHz "computer type" crystal oscillator is readily available. Another factor to consider is that putting the local oscillator "closer" to the IF (thereby moving the converted IF "lower" in frequency) would have placed the "upconverted" image (the 25-31 MHz signal mentioned above) even closer to the original 41-47 MHz passband, potentially making it a source of signal degradation.
Using the "new" IF:
Now that we have this "new" IF, what do we do with it? First, let's calculate where our critical spectral components will be:
Remember: The 45 MHz video IF is spectrally inverted because a high-side local oscillator is used on the 70cm downconverter (and on TVs as well) and since a low-side local oscillator is used to get it down to 5-11 MHz, it is still inverted. This explains why the "higher" video baseband frequencies are lower in the IF passbands.
Let us further analyze what we know about the video signal:
The "Notch" filter:
Referring to the spectrum analyzer plot (Note: Vertical is 5
and the horizontal is 1.5 MHz per division,) you can see why this
is so-named. Approximately 1.5 MHz of the active video spectrum
been attenuated by at least 20db. Does this "trash" the
The answer is: Not really. Looking carefully, one can see
the "critical" video components are left intact: The sound is
there, but the "lower" edge of the chroma passband has been removed
that the IF of the video signal is inverted, so the "lower" video
are to the left [i.e. higher] on the plot.) The notch extends
to approximately the 1.7 MHz point in the luminance bandwidth.
What does happen to the video, then? Surprisingly little, at first glance. The two side-by-side images demonstrate the effects. (Note: Click on the image to get a full-sized view to be able to see the details.) Careful comparison of the two images will reveal that there are some subtle differences. The "original" image (on the left) is somewhat sharper than the "notch filtered" image on the right. This is a natural result of the fact that we are eliminating the very components that make up the fine details in the image. The color is also slightly "smeared" due to the fact that we have eliminated a portion of its bandwidth as well.
Additionally, there some artifacts that may be seen to the right of sharper lines that are a result of the group delay characteristics of the notch filter itself (and possibly some "ringing" in some part of the circuit.) It is certainly true that with very careful design and attention to detail, these "group delay" effects may be reduced or even eliminated, but (as those of you that are familiar with design of video equipment know) the "delay equalization" portion of the filter will likely be more complicated than the filter itself. Perhaps it will be done another day...
Interestingly, if you look carefully you will notice that the
image has a bit less noise than the "original" image. Why?
As it turns out, much what is perceived of as noise in the video is
energy in the higher-frequency portion of the luminance
Since this filter effectively eliminates much of the high frequency
of the video, this noise significantly reduced.
This leads to another important point: For a given signal strength, the signal-to-noise ratio of a received signal is directly related to the receiver's bandwidth. The wider the bandwidth, the more noise one is likely to pick up. In the case of the "original" video signal, we can assume that we have the usual 6 MHz video bandwidth. In the case of the "notched" signal, we eliminated 1.5 MHz of that 6 MHz yielding a bandwidth of only 4.5 MHz total. Now, this is kind of misleading, as a television only uses, at most, about 5.5 MHz of that (the luminance and chroma go only up as far as 4.2 MHz, and the IF filter cuts off a bit of the lower vestigial sideband) so in reality, we end up with about 4 MHz with the "notch" in place. The elimination of this 1.5 Mhz results in a signal/noise ratio improvement of about 1.4 db from the bandwidth reduction alone. If there were other signals in that "notch" (such as FM repeater inputs and outputs) then the improvement in the signal/noise ratio could be far greater. On a "noisy P2" picture, there is a rather obvious reduction in the visual noise when this filter is switched in.
The 1 MHz filter:
Referring to the spectrum analyzer plot (the same parameters apply as for the previous plots) one can see that this is a 1.5 MHz wide bandpass filter. Why is it called a 1 MHz filter then? While the filter is 1.5 MHz wide, 500 KHz of it falls below the video carrier while the other 1 MHz of it is above the carrier and because of this, it will pass luminance information up to 1 MHz. Notice that it effectively removes all color and sound as well.
Why would we want to do this? This goes back to that video
stuff, again: Assuming that we have about 5.5 MHz of video
on a "normal" signal, if we reduce it to 1.5 MHz, we get a theoretical
signal/noise ratio improvement of about 5.6 db! This is
the equivalent of increasing transmitter power from 10 watts to about
What does this gain us? A 6 db improvement can turn a signal that is so weak that all you see is sync bars into a signal where a large, full screen ID may be just visible. If there are interfering signals within that portion of the passband that we are removing, the effective improvement can be far greater!
Doing this filtering exacts a price: The picture demonstrates that all color is lost. From the analyzer plot one can infer that the sound is gone, too (but more on this later...) Additionally, even more of the fine detail has been lost. Looking closely at the picture, one can also see the effects on the video caused by the lack of equalization to compensate for the variation in the group delay across the filter's bandwidth. As in the "notch" filter, these effects manifest themselves as "echoes" to the right of some of the picture elements.
Prior to constructing this filter module various tests were done at the original 45 MHz IF to determine the feasibility of constructing filters at that frequency. While it is perfectly possible to build such filters, extreme care must be taken to assure high filter Q and frequency stability. As it turns out, building simple notch filters at 45 Mhz is quite simple, but building steep-sided bandpass (or bandstop) filters that have a "flat" bandpass (or bandstop) over a given bandwidth is not. The "steeper" the filters sides are, the higher Q that is required and getting a suitably high Q from a filter element at this frequency may require that it be physically large.
In the 5-11 MHz frequency range, even though higher values of
and capacitance are used, getting the sorts of high Q's required at the
higher frequencies isn't as important: Being at a lower frequency
implies that the "steepness factor" is much less severe. For
if you wanted to build a bandpass filter that went from 3db to 20db
in a 500 KHz span, this would require a "steepness" factor of 15 if the
filter were built at 7.5 MHz. Building the same filter at, say,
MHz, would require a "steepness" factor of 87 - a factor of 5.8 more
One additional problem is that the higher the Q of a particular filter
element, the more it tends to "ring" when it is hit with energy.
Having the filter operate at lower frequencies (thereby reducing the
Q) reduces this effect.
The 300 KHz Filter and the "Sound-pass" filter:
The last filter section could be considered a "DX Mode" filter. This filter has a bandwidth of approximately 500 KHz - 200 KHz of it is below the video carrier and about 300 KHz of it is above, resulting in an effective luminance bandwidth of about 300 KHz. Assuming a "video" bandwidth of 5.5 MHz for a "normal" demodulator, this represents an improvement of about 10.4 db. This can, quite literally, turn a undetectable sub-P0 video signal into one where a large, full-screen ID may be copied!
As is the case of the other filters, one must sacrifice several things: Chroma (color) is out of the question, the sound is gone, and the resolution of the video is very poor. But, as can be seen from the picture, it is very useable if the picture contains some very large (i.e. low-bandwidth) components such as a large ID.
Because the filter topology is different, and due to the fact that the filter is nearly symmetrical (a result of it's bandpass being reasonably well centered on the video frequency) this filter exhibits no obvious group delay problems. Even though the image is a bit "soft" there is little or no evidence of "ringing" on fast-risetime video components.
The spectrum analyzer plot not only shows the "300 KHz" filter (toward the right of the plot) but a second bandpass filter response is shown (on the left of the plot) that is designed to pass the aural (sound) carrier. This is actually a separate filter (about 300 KHz wide) that may be switched in and out as desired. This filter is designed to be used in conjunction with the "1 MHz" filter to permit sound reception with it as well.
You may ask: "Doesn't turning on the sound filter 'noise-up' the picture?" The answer is: Not much. For the most part, the frequencies around the sound carrier are already removed by the demodulator to prevent the presence of the sound carrier from "fuzzing up" the picture. What does get through can be easily removed with a low pass filter on the video output of the demodulator.
While this filter does, indeed, allow the possibility of the aural to get through, the signal may be so weak that the aural carrier is already too weak to copy. This is actually quite likely, as most AM TV transmitters use between 1% and 10% (the amount can vary) of their total power to send the audio. Nevertheless, it is nice to have the option available.
I.F. Filtering versus Baseband Filtering:
You may be asking yourself, "Self, this is a lot of trouble to go through to filter out parts of the video. Wouldn't it be a lot easier to do this after it has been demodulated?"
This is a good question to ask, so we'll look at it for a moment. In an ideal demodulator, we simply recover the information in the sidebands. End of story. The demodulator doesn't really care whether there is one sideband, or two (ignoring the amplitude of the demodulator's output for the moment) or just a portion of a sideband (as is the case for standard AM video.)
In the old days, envelope detectors were used for video detection. This is essentially an RF detector where the output of the detector is in direct proportion with the RF voltage going in: A stronger signal results in more voltage out and a weaker signal results in less voltage. To work properly, this type of detector requires that the video carrier be intact.
Nowdays, the vast majority of analog televisions use a synchronous detector. This is a big improvement over the envelope detector for several reasons: An envelope detector needs a good video carrier to work properly. One problem is that, as the entire video signal gets weak, the video carrier gets weaker (and noisier) as well. With the video carrier getting noisy, it adds to the noise that is already in the video from the rest of the signal being weak, making the video appear noisier than it really is. A synchronous detector, on the other hand, takes the received video carrier and reconstructs it. It can do this because, as it turns out, the video carrier can be received even though it may be very weak because the circuit that recovers it has a narrower bandwidth than the video detector itself. This is the same sort of situation where you know that there is a video signal on frequency, but you cannot see it on your TV, even though you can hear it on your FM receiver just fine. This happens because your FM rig has only about 15 KHz bandwidth as compared to the 6000 KHz (or thereabouts) bandwidth of the video receiver.
With the video carrier "regenerated" within the detector, we now have a "pristine" copy of the carrier that we may use to demodulate the video. As it turns out, synchronous demodulators produce pictures that are 6db less-noisy for a given signal than an envelope detector. This would be equivalent to the transmitter increasing its power by a factor of 4!
There is a separate circuit called the AGC (Automatic Gain Control) that is also involved: This circuit looks at the amount of RF going into the detector and if it the signal is greater than a preset threshold, it reduces the gain of the RF amplifiers (these amplifiers may be in the RF front end and/or in the IF stages) and if the RF level is lower than threshold, it will increase the the gain: The whole purpose of this circuit being to keep the RF level constant. (I'll mention briefly that there are also some circuits built into the AGC to look only at certain parts of the video so that the AGC doesn't go up and down with the brightness of the picture.) If this AGC weren't present, then a weak signal would have a vastly different brightness than a strong signal.
Now, what do these things have to do with filtering? As it turns out, if the demodulator were perfect, we could simply put, say, a 300 KHz lowpass filter on the video output and get reasonable results. If you do that, however, you might note that the demodulator is still seeing the entire 6 MHz of IF passband. The rest of this IF could be filled with noise or other signals. The presence of any other signals in the video output can cause picture degradation (even if you would seem to be filtering them out) because of nonlinearities in the demodulator. If these other signals are strong you may simply overload the demodulator.
What about the AGC then? As it turns out, if you are using the video lowpass filter, the demodulator's AGC will faithfully do its job: If there is a strong signal in the passband, the AGC may reduce the gain of the receiver- possibly to the point where the weak signal you are trying to copy is simply gone! If the filter is in the IF, on the other hand, then that interfering signal will never even get to the demodulator: Problem solved. A similar thing goes for noise, as well. A well-designed video receiver will be doing at least a little bit of AGC action just on the noise of an empty channel (note that I said "a little" bit of AGC. Any more than that will compromise your receive system's dynamic range.) If you have the full 6 MHz of bandwidth, then your AGC is causing the gain of the receive system to be throttled back just a bit on the entire 6 MHz of noise. Now, if you have the "300 KHz" filter (described above) then the AGC is not acting on the noise of the entire 6 MHz and is now running at higher gain.
Let's look at it another way: If you consider that your ATV receive system has a noise figure of, say, 5 db (this isn't really too bad - this would be typical if you had a GaAsFET downconverter and a couble of db of coax loss in front of it) then the minimum discernible signal that you are going to see on your TV is going to be in the area of 2 microvolts or so. This signal is weak enough that all you'll probably see are sync bars and not anything that you'll be able to recognize.
Now if we were to drop the IF bandwidth of the TV from the normal 6 MHz to 500 KHz (as in the case of the "300 KHz" filter above) then our sensitivity actually increases. It will not take only about 0.7 microvolts (or less) to see the same sort of signal that you could "see" with 2 microvolts and the normal 6 MHz IF. The effects of this filtering can clearly be seen in the split image: The left half of the image shows a signal just barely at the threshold of the demodulator maintaining sync on the picture while the right half shows the same signal at the same signal strength, but with the "300 KHz" filter switched in.
A (possibly) even more dramatic demonstration may been with the other picture: In this case the signal was not even detectable (i.e. no sync bars visible at all!) in the full 6 MHz bandwidth. Even though it is very noisy, the large ID is still legible (albeit with some difficulty.)
The "filter module" idea originally started out as an "I wonder how it well it would work if..." concept and has developed into what it is now through a bit of number-crunching and experimentation.. While we have yet to put this portion of the repeater "on the air" the workbench "simulations" look very encouraging.
There is little doubt that this can be improved upon: A bit of effort to provide group-delay equalization in the "Notch" and "1 MHz" filters would likely reduce the filters' artifacts. A bit of high-frequency pre-emphasis could be applied to the video to make the 1 MHz and 300 Khz video "appear" to be sharper. It may be worth looking into an "ultra narrow" mode (about 50-100 Khz of video bandwidth) - just to see how narrow it could be and still be useful.
All of the test pictures on this page were caputured using actual ATV gear. The transmitter is a PC electronics TXA5 (Rev. C) and the receive system is that used for the WB7FID ATV repeater.
The video images were generated via computer using an ATI All-In-Wonder 128 and were captured using a Matrox Millenium video card.
Even at "best" the images look a bit "noisy." This is due to several factors. Unfortunately, my TXA5 was a bit "sick" when I captured the images, so its linearity is a bit poor (which explains a bit of the "grunge" and herringbone in the picture.) There is also a problem with the Matrox video capture card: There seems to be a "black level" problem on the video decoder portion, making all captured pictures look a bit dark and "muddy."
Finally, there is the problem that the picture was generated with a computer and it was captured on one. The ATI card does a pretty good job in its generation of NTSC video in terms of anti-aliasing. The Matrox, being quite a bit older, doesn't do quite as good a job at capturing: Unfortunately, when the two are put together, the artifacts from the NTSC generation AND the those from the capture tend to multiply, making the end result look worse than the sum of the two.
There is one final thing to consider: When video is noisy, the "noise dots" tend to move around randomly. This allows the eye and brain to "average" a lot of noise out of the picture and because of this, the pictures shown actually look better than they appear as frozen images. To test this yourself, record some noisy video on your VCR and then freeze-frame it and you'll see what I mean!
What does all of this mean? It means: "No, the video does not look as bad as these still images would lead you to believe.
In our application, these filters are going to be remotely switchable: Each filter section is selected via PIN diodes so there are no relays to wear out. One idea that has been tossed around is that, when the repeater is "idle" it will default to the "300 KHz" filter. Upon detection of sync, it may "automatically" select which filter it will use, based on the received signal strength (after calibration, of course!) but any filter setting could be overridden remotely should we wish to make our own decisions.
This page is an evolving document. More details on this (and related) methods will be posted here in the near future.
Please check back occasionally for updates!
Go to the Utah
ATV Home Page...