Analog video recording principles
 

First off, thanks to everyone who have been answering my questions about analog video recording formats. I re-read the answers and digged into a couple of books, but I still have questions!

I would appreciate if someone fills the blanks in my understanding of video signal modulation, storage, bandwidth, resolution, etc taking into account that I am not an electronics engineer and don't plan to become one, just want to have a little clearer understanding of how things work and why one format is better than another.

I've been reading some introductory texts on the topics of analog radio, television and color-under recording systems. I'll go with what I've learned (or I think I've learned), and please correct me where I am wrong.

First, how many samples per second we need to send, this is a rather simple idea: take frame size times frames per second, say 525 lines total, 4/3 aspect ratio with square pixels will give 525 * 4/3 = 700 samples per line, time 30 fps: 525*700*30 = 11 million samples per second.

Each wave period has two half periods, and I thought that each of them could be used to describe brightness of a pixel, so 11 million / 2 = 5.5 MHz. At this point I had a naive idea of how things work, I thought that this wave can be manipulated (a.k.a. modulated) so that amplitude of the each of its half-periods could be increased or decreased, this would give us 5.5 million unique samples. Based on this, I could not understand why do we need a RANGE of frequencies, why we cannot use ONE frequency modulated this way.

https://i.ibb.co/y8T1Cnh/naive-modulation-idea.png

Apparently, this is not how things work in real life :) Instead, I need 5.5 million of DIFFERENT FREQUENCIES, each frequency representing a pixel on the screen. To help me understand this, I was reading a book on analog radio, so the example was, like, real music has range of frequencies, say from 100 to 10 KHz, and want to send all of them. So, if I modulate the carrier with this signal, I will get a BAND with two sidebands, corresponding to the carrier minus signal and carrier plus signal. Ok, I get this.

OTOH, I am not sure that sending audio over the air is exactly the same as sending samples on a screen… But if this analogy is correct, then each of the samples on screen is represented by its own frequency, just like each tone in music is represented with its own frequency. Is this correct so far?

So, the position of a sample is represented with frequency, the intensity can be represented with amplitude. This is how AM radio works, and this is how OTA TV works - luminance use amplitude modulation (NTSC/PAL color uses AM, SECAM color uses FM, audio uses FM).

Amplitude modulation is easy to represent with a picture and to understand. I've read arguments for FM modulation, so it seems that all signals on tape are FM modulated: luminance, chrominance and audio, correct?

Next, broadcast video is converted for recording on tape. Because of various technical difficulties, AM signal cannot be used for quality video recording, so first the luminance is separated from the chroma, then the luminance bandwidth is reduced, then the luminance signal is then used to FM modulate a carrier before it is recorded on the video tape.

It is harder for me to make a mental picture how FM modulation works. The first part of the modulation is the same with AM, two frequencies are added together, producing two sidebands. When recording on tape, carrier is chosen high enough for the upper sideband to basically be removed automatically just because such high frequencies cannot be recorded on tape. So this issue solves itself. Lower sideband is used for luminance signal, I get it so far.

I suppose everything else is FM-modulated as well including chroma? Including NTSC and PAL chroma?

https://i.ibb.co/CwKK9zJ/downconversion.png

But the sideband represents the POSITION of sample onscreen, right? To represent the brightness, the signal modulates the carrier frequency, not the carrier amplitude. This is what DEVIATION is about, it is represents brightness. The higher the deviation, the larger the difference between black and white, or in other words, better contrast and lower S/N.

What I don't get is how brightness corresponds to the sample position (or in case of audio, how loudness corresponds to audio frequency)??? If the carrier is modulated by frequency, it is moved up and down, but the sideband that describes the position of a sample, is relative to the carrier, how does it work???

The book on radio that I am reading, re-iterates: the "rhythm" (well, frequency) of the carrier deviation corresponds to the signal frequency, while the amount of the deviation corresponds to the brightness (loudness). But I still don't get how the two parts: the sideband, that describes the POSITION, and the deviation that describes BRIGHTNESS are reconciled together. I can that with AM the carrier does not change, so whatever the distance from the carrier is my frequency (audio tone or sample position). But in case of the FM the carrier constantly changes, how do I know what actual audio tone or sample position it corresponds to??? Please, explain!

Now looking at this picture, which is one of the series of graphs about VHS/SVHS/Hi-8, I have several questions.

https://i.ibb.co/c3LmRcq/VHS-bands.png

First, what is "220% peak white limit", orange question mark. It is my translation from non-English text, but it corresponds to Panasonic info, which says about SVHS: "A 210% increase in the peak white level enhances the picture quality even more, with excellent delineation of image borders and highly faithful reproduction of detail." - does this mean that SVHS has 210% increase in the peak white level compared to VHS, and the above picture that describes VHS is incorrect? Also, "white level" is brightness, and brightness is defined by deviation of carrier frequency, not by amplitude, but the picture shows the amplitude. Is it a mistake, or what does it mean?

Second, red question mark: what is this frequency, from which the sideband is measured? It is not black level, not white level, not the middle. Is it 70% level of the deviation band, because the average value of a pure sine wave is 0.7 of max value?

Third, green question mark: chroma. It has both sidebands. Is it FM or AM modulated? If it is FM modulated, it should have its own deviation? Is chroma bandwidth of all color-under systems the same? If not, why it is not specified? The increase in chroma bandwidth means the increase in chroma resolution? But apparently this is not what they did, they only ever increased luma resolution. Why increasing chroma carrier is touted as a big deal, improving picture? Isn't the only thing that matter is that chrominance and luminance do not collide?

P.S. Please excuse me for using chroma and chrominance interchangeably.

"Instead, I need 5.5 million of DIFFERENT FREQUENCIES, each frequency representing a pixel on the screen."

Stop right there! Each spot on the screen does not have its own frequency.
You need to understand Fourier transforms, which calculate the frequencies produced by a sequence of dark and light spots.

So, a particular frequency (5.5 MHz in your example) produces a continuous sequence of alternating dark and light spots. A frequency of 2.75 MHz would produce a continuous sequence of two light spots followed by two dark spots, and so on. (In analog, since there are not discrete pixels, this would just be a coarser pattern of wider light and dark areas.)

So, each frequency in the signal affects ALL the spots on the screen, but in patterns of different coarseness and strength (contrast). The overall spectrum of a signal represents how much of each frequency is present such that when all these corresponding time patterns are combined, they form the sequence of light, dark and gray spots to make a scanned picture.

If you can first understand this, then we can discuss your further questions, and maybe see why some of them make sense and some don't.

Ah! Ok, I get this now. This is the relationship between frequency and resolution. Higher frequency creates a finer pattern of bright and dark samples, which means more detail. And a complex image is made of a combination of different patterns. I think I get it now without learning too deep about Fourier transforms. BTW, this video is awesome! I think I understand the concept now.

So, there is a "cloud" of frequencies around the carrier representing a complex pattern. This cloud changes every 1/60 of a second for interlaced video. Well, actually it changes constantly, as analog video is just one big string of samples. I guess my attempts to imagine how it would look like in static are pointless, because frequency by definition presumes something happening over time.

I am not sure this helps me understand the relationship between the pattern onscreen and some of its parts being brighter or darker.

Ok, so you now have the idea that you need to keep in mind whether you are discussing time domain or frequency domain.

So:
"What I don't get is how brightness corresponds to the sample position (or in case of audio, how loudness corresponds to audio frequency)??? If the carrier is modulated by frequency, it is moved up and down, but the sideband that describes the position of a sample, is relative to the carrier, how does it work???"

You should now understand that loudness does not correspond to audio frequency, but to audio amplitude. And that a particular video sideband does not describe the position of one sample, but a repeating pattern of light and dark.

Regarding the 220% peak white limit:

Before the Freqency Modulation, the luma signal is "pre-emphasized." Thiis means the high frequencies are amplified more than the lower frequencies. Since the high frequencies are involve in producing sharp edges, this means that (in the case of a black to white transition) instead of going from the normal black level to the normal white level, the pre-emphasized luma signal continues to rise above the normal white level for a while and then comes back down to normal white. During playback, the demodulated luma signal is subject to a corresponding de-emphasis that restores the transition to its normal black-to-white transition. Because de-emphasis is reducing the high frequency components of the signal, it also reduces any high frequency noise generated by the tape. A similar technique is used in FM radio broadcasting, and even in recording LP records. Dolby noise reduction and other noise reduction methods go further and use a variable pre-emphasis and de-emphasis to adjust the amount of it depending on whether the audio is soft or loud.

Here is a web site that shows a little-known (these days) video recording system that used variable pre-emphasis, and compares it to VHS's fixed pre-emphasis (Figure 1):
http://www.digiommel.fi/Improving%20...%20Quality.pdf

Well, ok, a sideband describes a pattern of light and dark.

Thank you for explaining the pre-emphasis thing. Yes, I know how Dolby works, this helps!

But what about the deviation It defines brightness. How does it correspond to the pattern? Does it define the brightness of the whole pattern (frame) or for each sample? I suppose it should somehow act on separate samples to make certain parts darker or brighter.

Regarding 220%, here is a Sony picture. Here the luminance bandwidth goes from the carrier to the left, not from like 0.7% of the deviation.

https://i.ibb.co/xznCbHM/hi8-video-system.png

"Second, red question mark: what is this frequency, from which the sideband is measured?"

The video signal has 40 "IRE units" of sync and 100 units of black to white, so a mid gray sits at 90 units, which is about 68% of the total deviation from sync tips to white. So, that frequency corresponds to middle gray.

Look at the Sony diagram again - 7.0 MHz is the FM frequency for mid gray, and the bandwidth of the signal is shown measured down from that. If you think about this for a while, you will see that there is more bandwidth for high frequency detail when that detail is riding on a higher brightness background, and some what less when it's riding on a dark background.

Now suppose the image has a full black-to-white fine repeating pattern in some area. In that case, the average background in that area is mid gray by definition, since it is the average of alternating black spots and white spots. This will produce a sideband frequency at the lower limit shown in the diagram. It would be impractical to draw a diagram showing every case of detail amplitude and background gray level, so systems are compared using the kind of diagram you have posted.

Chroma is quadrature AM modulated. In fact, it is the same as the chroma signal in the original NTSC signal, just shifted in frequency. It could not be FM modulated unless it was first split into two separate color components with each modulated on a separate FM carrier.

"there is more bandwidth for high frequency detail when that detail is riding on a higher brightness background" - indeed! I see it!

The VHS pre-emphasis picture shows 14 dB gain, which is about 500%, not sure where 220% came from.

"Now suppose the image has a full black-to-white fine repeating pattern in some area. In that case, the average background in that area is mid gray by definition, since it is the average of alternating black spots and white spots. This will produce a sideband frequency at the lower limit shown in the diagram. It would be impractical to draw a diagram showing every case of detail amplitude and background gray level, so systems are compared using the kind of diagram you have posted." - this I did not quite get. What is "a sideband frequency at the lower limit", does it have numerically low frequency closer to the left, or does it mean that the difference (I don't want to use deviation here) between this frequency and the carrier will be narrower, so the actual frequency will be closer to the carrier? I am looking at the part to the left of the carrier. And I still haven't reconciled in my mind how the same pattern can be reproduced with different brightness.

Say, I have a pattern, that is, a complex frequency. If I want to make it brighter, I increase the carrier frequency, but then my whole pattern moves to the right, so a frequency that was, say 2 MHz, becomes 2.5 MHz, which means it is a different pattern!

Looks like I am still missing something. I am going to shut up for a while and let you finish :)

"The VHS pre-emphasis picture shows 14 dB gain, which is about 500%, not sure where 220% came from."
1) The pre-emphasis is in the frequency domain. The picture sledom will include a full amplitude single frequency, so a 100% black to white excursion of one frequency is uncommon. More commonly, this is emphasizing the high frequencies that are part of making a step edge between black and white.
2) If the pre-emphasis does produce a waveform that goes over 220%, the luma signal is limited to 220% in the time domain, this preventing overload of the FM demodulator circuit. This means that if the original signal was a very high frequency stripe pattern of high amplitude, it would be clipped to a lower amplitude before FM modulation and therefore appear as a lower amplitude when demodulated. This is a good trade-off for better-defined, less noisy edges.

"Say, I have a pattern, that is, a complex frequency. If I want to make it brighter, I increase the carrier frequency, but then my whole pattern moves to the right, so a frequency that was, say 2 MHz, becomes 2.5 MHz, which means it is a different pattern!"

Exactly right. It is a different frequency spectrum.
This should not bother your understanding, as in the luma input time domain the high frequency ripple pattern is riding on an average background level; in the frequency domain, the luma baseband input consists of the high frequency plus some DC (zero frequency) energy representing the average background brightness.

After FM modulation, in the frequency domain, the sideband frequency, determined by the high frequency pattern frequency, is "riding on" an average carrier frequency determined by the average brightness of the background. In other words, if the average brightness is different, the average carrier frequency is different, and all the associated sidebands move to keep their same frequency distance from the average.

You should also understand that the diagrams show the frequency bands that are available to carry the FM signal, not the actual FM spectrum that is being sent in the available band in a particular case. The flat top of the FM range just indicates that the FM carrier and sidebands will be recorded if they fit in that range. If the sidenbands go too low in frequency, they will be cut off and their pattern will not appear in the output image. That's what happens in the extreme example where a very high frequency luma pattern is riding on a dark background. In this case, the average carrier frequency is lower, the sideband shifts lower by the same amount as the average, and it gets cutoff by the low end of the system response curve.

This I do understand. What I do not understand how these sidebands will be represented onscreen. The distance from the average is the same, but the absolute frequency is different. Unless a TV knows how to represent the picture in relation to the changing average I just don't get how the picture will look the same, but brigher, because absolute frequencies are different. And as you said before, frequencies in the sideband represent the pattern, then the pattern will change with a change in brightness. I guess I am missing some important info about obtaining the average frequency for any moment in time.

It's the mathematics of what the TIME waveform of the FM signal looks like, and how it is composed of a combination of pure sine waves that individually have a constant frequency and amplitude (the spectrum).

With the extreme case where the luma consists of a high-frequency alternating stripe pattern having peaks and troughs and a certain average level, the TIME waveform of the FM signal shows the FM carrier frequency changing rapidly and repeatedly from low frequency (wide cycles) to high frequency (narrow cycles) representing the troughs and peaks of the luma waveform. Fourier analysis shows that this TIME waveform of the FM signal can be obtained by adding a sine wave with the (constant) average frequency to sine waves of the sideband frequencies. The plot of how much of the average frequency sine wave is there and how much of the sideband sine wave is there is the frequency domain plot of the signal.

Note: because upper sidebands are cut off by the recording/playback bandwidth, the recovered FM signal will have both frequency and amplitude variation; but the FM demodulator ignores the amplitude variations.

Um... Not sure I understood this. Can you explain in layman terms? Say, I have a certain pattern, it is described by the sideband, right? Then I have the same pattern, but brighter, this means that it should look like the previous pattern, but moved to the right? But the frequencies, numerically, will change, they will all move to the right. How the display device will know that it is the same pattern unless it displays the image in relation to the [ever changing] middle frequency?

Thinking about it now, maybe this can be likened to a more simpler case of analog tape recorder. If tape plays faster than normal (so the average frequency is higher) I still hear music, just all notes are shifted, but the intervals between them remain the same. Not sure though, how video patterns correlate to music tones - is it a good enough analogy?

Let me see if I can make a few drawings. This will take a while. Unfortunately, I have reached my limit for posting images to this site, so have to host them elswhere and link here.

Meanwhile, when you ask about "patterns," please specify if you mean a pattern in the video image, in the video time signal, or in the FM spectrum.

Thanks, I appreciate your help! When I say pattern I primarily mean video image.

It's the instantaneous time domain changes that determine the pattern on screen. You can have more than 3 different video frequencies in effect on a segment of a line of video and the phase relationship of those video frequencies determine where on the line they add to and or subtract from each other and the nature and position of the pattern they make.
The side bands don't exactly describe the just the signal. If you modulate one sine wave with another regardless of the modulation scheme you get the main sidebands and an infinite series of harmonics at multiples of the carrier. You also get additional harmonics from non-linearities in all non-ideal (read that real world) modulators. (Ever been close to a radio tower and gotten the station on multiple places on the dial? Those are harmonics.) So there can be things in a spectrum analyzer view of a sideband that are more incidental energy than desired ideal signal.

The passband of a signal chain tells you what frequencies will be let through, but not whether in actual video they're intended or unintended to be in the picture. Without the precise phase timing duration and the rate of rise and fall of the different video frequency elements you could get a variety of patterns and images from the same sideband pattern you see from a line or frame duration sideband spectrum sample.

Does this help?

https://live.staticflickr.com/65535/...9eae4d09_b.jpg

Um, sort of. I think I understood that the whole spectrum would move, but I did not understand how a TV set would know that the frequencies that moved to a different location represent the same image. I suppose, the TV set "measures" it relative to the mid frequency and it can detect this mid frequency automatically, just like a wow & flutter tester can, if you have two sidebands, just measure the full difference from left sideband to right sideband and divide by half. But the right sideband is cut off, so I am not sure how this works.

https://i.ibb.co/yRSvcqM/bands.png

Anyway, I think I know enough to be able to read the charts and to make some conclusions, like you can have the most detail at the brightness close to the max, and without color it would be even higher.

BTW, I've heard that analog TV signal used to have a "flag" that indicated a program as color or B&W. Was it widely used? If a VCR was connected to HF antenna cable, it was able to receive this flag. Did VCRs act upon it? Did they omit recording chrominance? Is this flag sent over composite/SVideo?

It is not the TV that interprets the FM signal. The VCR has an FM decoder that reconstitutes the baseband video signal you see in the top row of drawings.

The time domain plot of the FM signal during the pattern shows the carrier frquency during the pattern varying directly according to the instantaneous baseband voltage in the top row diagram.

So, just picking some numbers out of the air thatcould apply to the HI-8 system:

If the baseband video pattern average is mid gray voltage, meaning the sine wave varies equally above and below mid gray, the instantaneous FM time signal might vary in frequency from 6.8 MHz to 7.2 MHz. The instantaneous FM frequency corresponds exactly to the instantaneous video baseband voltage. The FM demodulator converts this instantaneous frequency back to the exactly corresponding video baseband voltage, forming the signal that is sent to the TV. The FM spectrum is only telling you that you can make this varying frequency FM time signal by adding togehter a certain collection of sine waves that each have their own constant unvarying frequency and amplitude.

This is the genius of Fourier, that he showed this is possible - to decompose a varying waveform into a sum of unvarying waveforms. In this case, exact replication of the FM signal requires a center frequency and two sidebands. (As mentioned before, the tape system loses the upper sideband, so the recovered M signal then has amplitude modulation in addition to frequency modulation - but the FM decoder ignores the amplitude modulation.)

Now suppose you have in the baseband video signal the same pattern with the same peak-to-peak amplitude, but its positive and negative peak voltages are both shifted lower by the same amount. All instantaneous FM time signal frequencies are now shifted lower by the same amount compared to the mid-gray case (for example, 6.5 to 6.9 MHz), because the video baseband voltages that control the frequency have been shifted by a particular amount. The FM demodulator sees these shifted frequencies and produces a baseband video with shifted voltages that match the original. The spectrum plot just shows that all the frequencies of the first case are replaced with shifted frequencies in the second case, and the amount of shift is the same for all frequencies.

"Anyway, I think I know enough to be able to read the charts and to make some conclusions, like you can have the most detail at the brightness close to the max, and without color it would be even higher."

Correct.

"BTW, I've heard that analog TV signal used to have a "flag" that indicated a program as color or B&W. Was it widely used? "

https://en.wikipedia.org/wiki/Colorburst

Color burst MUST be present in an analog color signal or the color cannot be recovered. An analog TV would detect the presence of burst to turn on the color circuits, thus preventing colored noise in monochrome programs.

Eventually, sloppy practice (especially in analog cable systems) left the color burst on for all programs.

I believe VCRs also detected the burst so that color could be turned off for monochrome programs, but I don't know for sure or if it varied in different products.

"Is this flag sent over composite/SVideo?

Yes, both, as it is necessary for the analog color circuits to work.

In VCRs it is carried in the color-under signal just like the color subcarrier for the image content, since it is just a sample of the reference chrominance subcarrier. It is used in the VCR's special fast acting correction circuitry to take out variations due to mechanical tolerances and provide a stable chrominance signal for output to the TV.

Back in the tube era they would change the horizontal and vertical scanning rates slightly for color TV. I believe one of the things that started to reduce that practice was VCR timecode being used in broadcast automation....The time code was based on frame count and it probably would be hard to handle it accurately every time if it changed between color and monochrome.

I suspect color network logo watermarks in the programming contributed to the constant burst practice, since if the CEO tells you the watermark has to be in color and not change no matter what the program is then the burst has to stay on.
IIRC in the 70s didn't PBS do a thing where they made their bursts ultra precise so NIST could use them as a lab calibration reference?..If that was the case they may have had to keep burst on during monochrome shows to make that work, and may have been doing constant burst before cable.

It is possible to regenerate colour from a NTSC or PAL signal that lacks its colour burst. It was called chromalock and was not reliable. Worse in PAL because it has no information aboujt the polarity of the PAL Switch which had to be set manually.

Why do this at all? Because sometimes the burst would be removed to prevent colour being received. I forget why this was sometimes done.

NTSC and Timecode. AAAARRRRGGGHHH!!! I'm looking at the old NTSC papers in the IRE journal from 1954 which explain the reasons. The H and V frequencies were reduced by 0.1% to minimise patterning that could arise between the sound carrier and colour subcarrier. This avoided retuning the sound circuits on monochrome sets. It seemed harmless at the time. Some years later timecode was invented. The lack of a sensible relationship between frame rate and time of day has caused massive amounts of hassle.

The relationship between subcarrier and H had to be fixed (at an odd multiple of H/2) to minimise visibility of subcarrier on monochrome sets.

PAL is more complex (quarter line offset + half frame offset) which doesn't have a problem against the sound carrier. Hence we have no timecode problems in 50Hz countries.

ISTR reading this was done in Israel in the 70s because they thought color TV was an unfair luxury....All it did was create a market for systems that would try to achieve color sync without the color burst.

You've got a close idea, but missing some points.

The monochrome standard had wide tolerance for scan frequencies; the FCC broadcast rules required the vertical and horizontal to be locked to each other for a precise interlaced line count, but the master rate could vary; in fact, the rate could be tied to the station's 60 cycle power and still be within tolerance, as the power companies maintained average correct frequency so synchronous motor electric clocks would tell the right time. In practice, stations adopted crystal references for the scanning frequencies, which were 60.00 Hz and 15750 Hz. An additional requirement was placed on the carrier frequencies such that the undeviated (silence) frequency of the audio carrier was 4.5 MHz above the video carrier with a rather tight tolerance. This was to allow TVs with separate audio IFs to fine tune the audio and video correctly simultaneously.
When color came along, there was concern with ~ 920 kHz video beat beween the color subcarrier and the undeviated audio being visible in monochrome sets. Because of the precise scan and audio frequencies used in monochrome, you could choose the color subcarrier to interleave (be less visible) with the video, but then it would be worst case for the 920 kHz; or vice versa. Mathematically you could not optimize both. The decision was made to change the scan rates and keep the 4.5 MHz audio spacing exactly as before, to prevent sound problems in legacy monochrome sets. So, the scan rates were changed by the ratio 1000/1001. Now the color frame rate ran slow compared to a 60 Hz wall clock, necessitating the invention of time code to make the conversion from frame count to seconds, minutes, and hours for program duration.

Details continued:

The color burst had to be present for color programs, of course, but there's a gray area for monochrome. If the monochome broadcast was at the old monochrome scan rates (as often happened in the early days) then the color burst had to be omitted because it would become a color broadcast at illegal scan rates for color. Early on, there could be a lot of hot switching between different sources.

If the local station was equipped for color and always operated at color scan rates, then I suppose they could leave the burst on and argue that they were sending a color broadcast of a monochrome image, but I believe they always followed the practice of turning off the burst for monochrome programs even if they used the color scan rate throughout their plant. This allowed color sets to turn off the color circuits on monochrome programs and eliminate any color noise as well as cross-color (high frequency monochrome info being interpreted as color "twinkles").

Things really got messed up in analog cable TV systems. Regulations required that a clean color burst and sync be inserted into all incoming signals before they went out on the cable. Headend equipment, as far as I know, typically did not include the capability to turn the outgoing burst on and off, so that monochrome programs were sent to the home with the burst on.

Regarding the use of burst for a high precision frequency reference, this started prior to the use of frame synchronizers at local stations. The three major networks each installed a master atomic clock, and the local stations would lock to the network, so that the scan and burst frequencies had accuracy approaching that of the National Bureau of standards. You could actually sync the video from one network with the video of another except for a constant offset in time delay. NBS then published a paper on using the color burst as a tight-tolerance frequency reference. The networks' purpose had been to prevent vertical flipping due to hot-switching sources, but that also created the precise reference frequency.

After the development and installation of frame synchronizers in local studios, the locals went back to their own crystal references, so the use of the burst for precise frequency reference was lost.

Today, digital broadcasts are all locked to GPS time, even more precise than the original network atomic clocks. This allows precision offset of stations' carrier frequencies and phase for minimum co-channel interference, and even the implemention of single-frequency networks where multiple transmitters cooperate to cover a given service area.

A couple more notes:

1) The change of scan rates by the ratio 1000/1001 was still within the looser original tolerance for monochrome. Changing the audio carrier frequency offset from 4.500 Mhz to (4.500)(1001/1000) would have been outside the original audio tolerance.

2) By the time that analog cable began messing up the burst on/off issue, all sources, color or monochrome, were running at color scan rates.

"and even the implemention of single-frequency networks where multiple transmitters cooperate to cover a given service area." - ATSC 3.0 ?

Yes, ATSC 3.0, but also possible in a few situations with ATSC 1.0, though much more limited in application and difficult to design.

Were there other color-under formats except for umatic, betamax, vhs and 8-mm? Like, the early Philips, or Grundig or some others? For example, the wikipedia article about the Philips format does not specify how the signal was recorded.

As far as I know, all domestic and industrial analogue videotape formats used colour under. That includes unusual machines like the Sony CV5600 (colour version of CV2100).

Possible exceptions are the IVC and Ampex 1" industrial formats that predate C format. In some case these had enough bandwidth to record colour directly. A special timebase corrector was needed to recover the colour. In the 1970s I had access to an unusual VR7003H (for high band) variant of the Ampex 1" machines. This certainly had the bandwidth but I never had a suitable TBC. You got flashes of colour on playback but certainly not useable.