Current status of ld-decode / vhs-decode (true "backup" of RF signals)

[dellsam34]

So I have a few creased tapes that my deck refuses to play and I searched google and found this: https://itstillworks.com/how-to-correct-wrinkled-vhs-tapes-10141.html

Does this sound like a reliable method for fixing creased tapes or should I use something else? I really do not want to resort to splicing but I will if I have to.

The worst advice ever, Never touch a tape with your fingers, all the oils and dirt from the finger print will be scrapped off by the video heads and clog them.

[fsquared]

So I have a few creased tapes that my deck refuses to play and I searched google and found this: https://itstillworks.com/how-to-correct-wrinkled-vhs-tapes-10141.html

Does this sound like a reliable method for fixing creased tapes or should I use something else? I really do not want to resort to splicing but I will if I have to.

The worst advice ever, Never touch a tape with your fingers, all the oils and dirt from the finger print will be scrapped off by the video heads and clog them.

I never touch my tapes either. I thought the article only mentioned rotating the reels but maybe I read it wrong.

[dellsam34]

Articles like that have no purpose other than online traffic for ad revenue, the people who write those articles are very tech unsavvy and they target a population that is usually dumber than them.

[Titan_91]

Made some progress today, turns out I was looking at the wrong spot, and it was the final chroma band-pass filter that was too narrow. Thought I had checked that earlier but apparently not. Anyhow this makes colour look much nicer:


[Attachment 51451 - Click to enlarge]

Also added a --cxadc3 option for 35 mhz cxadc, and --othertrack to swap which track has phase rotation (try this if you get no color).

That is awesome!! Great job and thanks a lot for continuing to work on this. It seems the thread has gathered a small following. The dot crawl that was mentioned is I think some minor luma and chroma crosstalk. However, I don't see this between the colors themselves. Is this a result of PAL's better color encoding or your chroma decoding code after you upconverted the signal to 4.43MHz? I also see scanlines in your screenshot. Is this a filter you added manually or does vhs-decode do this when decoding a single field upscaled to 576 pixels?

[oln]

I'm not entirely sure what's the cause of the checkerboarding, it may be some interference between strong chroma signal and the luma FM either during recording or playback. I will have to do a normal S-Video capture to check if it's there, but I've seen this dotty effect on strong colours on other tapes, so it may be something inherent to the format. It's not the normal luma/chroma crosstalk you get from a composite capture at least, that's higher frequency. Chroma and luma go through separate path right from the start after filtering for the relevant frequency, so there won't be any 4.43 mhz stuff on the luma channel.

The scanline effect is just result of slightly different brightness/chroma levels between the two fields, it's not something I've added intentionally. The VCR repair book I got states that having the white/black point frequencies slightly higher on one track compared to the other field was used in LP mode to reduce crosstalk, but I'm not sure if that's what this is as it's not an LP tape.

[radomarkov]

I see you already censorship--ed my messages, maybe because I say the truth and today nobody want to hear the truth.
First problem you have is impedance matching between PB RF control point of VCR (high impedance) and CX input (50 ohm low impedance). From impedance mismatching can become a lot of problems that cannot be cleared in software way. Like low level (gain) even if the RF PB point is ALWAYS after videohead amplifier, reflection (cannot be cleared by software), different Amplitude-frequency characteristic that can drive to strange results on video and many more. Somebody's approach here "I connected it without care to impedance and it work" only confirm my spokes that here is a kingdom of deep amateurs. Because if you raise level that is decreased due to impedance mismatching you also rise noise that is anyway high in home VCR.
Second problem is that you missed some steps in decoding and "isolating" luminance from chrominance. At this point capturing from composite will be better than software decoded picture. And this will last till some professional is involved and give you the right advice. But the problem is that real professionals hate deep amateurs and will give advice when all amateurs disappear from this thread, even amateurs with good intentions "to help".
I bet till next January we will not see software decoded picture , better than captured from component VCR output.

Sorry to say that, but truth is crude.

[Titan_91]

I'm not entirely sure what's the cause of the checkerboarding, it may be some interference between strong chroma signal and the luma FM either during recording or playback. I will have to do a normal S-Video capture to check if it's there, but I've seen this dotty effect on strong colours on other tapes, so it may be something inherent to the format. It's not the normal luma/chroma crosstalk you get from a composite capture at least, that's higher frequency. Chroma and luma go through separate path right from the start after filtering for the relevant frequency, so there won't be any 4.43 mhz stuff on the luma channel.

The scanline effect is just result of slightly different brightness/chroma levels between the two fields, it's not something I've added intentionally. The VCR repair book I got states that having the white/black point frequencies slightly higher on one track compared to the other field was used in LP mode to reduce crosstalk, but I'm not sure if that's what this is as it's not an LP tape.

Interesting stuff. So ideally luma and chroma will never mix, and after upsampling the existing ld-decode code will decode them separately in the same way after separating a PAL composite video source?

[Titan_91]

Fsquared you should be able to find a PCI supported computer pretty easily. The other specs shouldn't matter if you're running something lightweight like Linux Mint. Something as old as 2010 should work as well as long as you're using SATA (shouldn't be an issue anyway as it's 2020, lol).

[dellsam34]

But the problem is that real professionals hate deep amateurs and will give advice when all amateurs disappear from this thread, even amateurs with good intentions "to help".
I bet till next January we will not see software decoded picture , better than captured from component VCR output.

Sorry to say that, but truth is crude.

I don't think professionals are interested in such project, They are already making and selling crappy capture devices to the average clueless, Designing and making a product that requires a surgical operation to a VCR rather than just a product that connects to its video ports is not a business model. I'm thankful that someone is at least trying so I don't see a reason for bashing them. If you can't help then posts like yours serves no purpose.

[Brad]

Is it just me, or did post #156 by Laserdisc appear from nowhere? I swear it wasn't there before when I read the posts above and below it by Titan_91 & fsquared. Does the forum have a delay for new users' posts to be approved & show up if they contain links, as this one did?

---

Made some progress today, turns out I was looking at the wrong spot, and it was the final chroma band-pass filter that was too narrow. Thought I had checked that earlier but apparently not. Anyhow this makes colour look much nicer:


[Attachment 51451 - Click to enlarge]

Also added a --cxadc3 option for 35 mhz cxadc, and --othertrack to swap which track has phase rotation (try this if you get no color).

Awesome!!

So ideally luma and chroma will never mix

Technically, a portion of the VHS & Beta spectrum always overlaps Y & C. S-VHS & Super Beta fixed this little design flaw. U-matic and Video8 were truly separate from their inception.

In standard VHS and VHS-C, the sync tips cause the modulator frequency to drop to 3.4 MHz and the 100% white level causes the modulator frequency to increase to 4.4 MHz. Maximum deviation of the modulated signal
equals 1 MHz.

Two limitations occur due to the frequency and deviation limits selected. The lower frequencies of the luminance signal mix slightly with the down-converted chroma signal. In addition, the frequency range limits the resolution to 240 lines. The NTSC signal is capable of 330 lines of resolution.

To improve picture quality (resolution) further, the enhanced formats were developed. Super VHS and Super VHS-C VCRs have higher resolution capabilities by raising the overall frequency of the luminance signal. Thus, the FMed signal is placed farther away from the chroma to reduce luminance and chrominance mixing.

---

the problem is that real professionals hate deep amateurs and will give advice when all amateurs disappear from this thread, even amateurs with good intentions "to help".

WTF? Hate?

I thought your original complaint was people wanting to profit off the work of the open source community without contributing anything. Now that's been replaced by "if you don't already know what to do, give up, you can never learn; the people who might share their knowledge by teaching you actually hate you for your ignorance."

I bet till next January we will not see software decoded picture , better than captured from component VCR output.

And the non-existent projects of these "professionals"? How are they coming along?

[Zcooger]

Hello,
here's my GNURadio software prototype for VHS HiFi stereo decoder:



For those who want to try it I upload here basic version which lacks some secondary features like 2:1 dynamic range expander and additional noise reduction.
- https://drive.google.com/open?id=1K8mq7R_QMjhNmA7V4KxE0c2aArTvQdb2

Here's a HiFi Stereo sample to play with (35.8MHz PAL):
- https://drive.google.com/open?id=1HuDEXcGxctyVmNVqiQ4KKz9mfu8g5DNf

Longest VHS-Decode sample in color - have a taste:
- https://youtube.com/watch?v=evHMMreIUQE

[dellsam34]

Pretty stable frames, looking good apart from that nasty dot crawl. Would you be able to record a fresh tape from a good digital source and capture it just to see how good this method can be?

[Zcooger]

Enjoy; signal generator:
- https://drive.google.com/open?id=1iqShC4MvunVILQjSUE8HlXhuUGTTxIHX

Prerecorded tape:
- https://drive.google.com/open?id=1WuO6sJno6RaLB5A7HMP8eBL3wOkFl20s

Dot crawl appearance may come from out of-spec signal or VCRs internal comb filter used for downconverting the signal (misaligned or old components like capacitors). It is visible on luma signal.

Edit: I meant the opposite of comb filter (process of decombing).

[dellsam34]

I though the comb filter comes after the RF signal where the luma and chroma are combined by the VCR's video processing circuit for composite output. They should be separate coming out of the tape, Am I not understanding something here?

[outhud9]

Longest VHS-Decode sample in color - have a taste:
- https://youtube.com/watch?v=evHMMreIUQE

Very nice. It would be interesting to see the same tape captured through VCR component output to a normal capture card for comparison.

[Titan_91]

Hello,
here's my GNURadio software prototype for VHS HiFi stereo decoder:


[Attachment 51461 - Click to enlarge]
[Attachment 51462 - Click to enlarge]

For those who want to try it I upload here basic version which lacks some secondary features like 2:1 dynamic range expander and additional noise reduction.
- https://drive.google.com/open?id=1K8mq7R_QMjhNmA7V4KxE0c2aArTvQdb2

Here's a HiFi Stereo sample to play with (35.8MHz PAL):
- https://drive.google.com/open?id=1HuDEXcGxctyVmNVqiQ4KKz9mfu8g5DNf

Longest VHS-Decode sample in color - have a taste:
- https://youtube.com/watch?v=evHMMreIUQE

That's great! Could this be modified to get a rough luma demodulator that saves raw samples to a wav file similar to this? I've messed with GNU Radio before and may actually try this myself. Of course it would only be good for testing with AviSynth's RawSource filter or some other rasteriser software. I'd like to get a general idea of what the output could look like before NTSC support is eventually added to vhs-decode.

[Titan_91]

Wow, that Kids Next Door credits sequence looks amazing. How in the world can the colors look this sharp with the luma signal being just 39.93 cycles per line?? I don't see even a hint of smearing as is typical for VHS. Is the chroma signal higher than 629kHz for PAL?

[outhud9]

I don't see even a hint of smearing as is typical for VHS.

I think it shows at the bottom left corner of the pink shape. You can see the horizontal line along the bottom of the pink shape is sharper than the vertical line up the left side of it.

It shows in the text also. See the letter m in 'music' near the bottom. The edges on the horizontal line at the top of the letter are much sharper than the vertical ones on the 'legs' of the m.

[Zcooger]

So here's regular capture without color correction and TBC: https://youtu.be/nE48jybPApc
And the most recent denoised piece from this tape captured with calibrated capture card that demonstrates Panasonic DMR-ES15 TBC: https://youtu.be/mlagfhrW_0M

Sharpened pictures are obviously sharp but they aren't detailed. A lot of detail is lost during such long processing chains where 3 or 4 devices are involved.
This old camcorder video is interesting - take a look:
https://youtu.be/TcdhJsssdcs

[Titan_91]

I think it shows at the bottom left corner of the pink shape. You can see the horizontal line along the bottom of the pink shape is sharper than the vertical line up the left side of it.

It shows in the text also. See the letter m in 'music' near the bottom. The edges on the horizontal line at the top of the letter are much sharper than the vertical ones on the 'legs' of the m.

I was mainly talking about chroma but do see a bit in the red background, yes.

[Titan_91]

I'm getting this error in GNU Radio Companion:

1:1:FATAL:PARSER:ERR_DOCUMENT_EMPTY: Start tag expected, '<' not found
>>> Failure

[Zcooger]

Hey, which version of GR do you use? Mine is 3.8.0.0, ensure that you use the same. It may be a problem of software library which still holds old version of the software. Use the instructions: https://wiki.gnuradio.org/index.php/InstallingGR#Ubuntu_PPA_Installation
Or simply copy the command:

sudo add-apt-repository ppa:gnuradio/gnuradio-releases && sudo apt-get update && sudo apt install gnuradio

Another problem is that you are trying to start the graph using mine settings for input and output. Change the directory in "File Source" block and similar for "Wav File Sink".

[Titan_91]

Ah, I was on 3.7.11 using the built-in Ubuntu PPA. I added the PPA you listed, updated the apt cache, updated the package to 3.9.0.0, installed python3-distutils, and it works. I tried to edit your flow graph but because I don't know what I'm doing I broke it, generates a lot of Python errors now. Could you edit it to output a basic WAV file of the demodulated luma carrier at 3.9MHz? If it's not too much trouble. I'm just curious to see a rough idea of some NTSC captures if I have a raw file of the scanlines. In particular I have the Arthur cartoon recorded at both sample rates I could compare.

[oln]

Is the chroma signal higher than 629kHz for PAL?

No it's the same frequency on tape, it's the upconverted frequency that differs. The luma signal is a bit higher though (but has the same 1mhz deviation.) - see vaporeon800's post

This is from the same tape captured from a Panasonic NV-HS1000 through S-Video, with TBC on, and set to passive(edit mode) to turn off noise reduction.

[Attachment 51469 - Click to enlarge]
You can see the checkerboarding here as well, presume it's due to the slight chroma/luma overlap, I guess it may look slightly different on NTSC since the luma is at a slightly different frequency range and the color is encoded differently.

I though the comb filter comes after the RF signal where the luma and chroma are combined by the VCR's video processing circuit for composite output. They should be separate coming out of the tape, Am I not understanding something here?

There at least two separate places comb filtering is done, at least in somewhat modern VCRs.

There is one during playback, it's (for VHS) is a simple filter that adds a delayed chroma signal back into the chroma signal to reduce chroma crosstalk (for PAL/NTSC, not SECAM). Luma and chroma aren't combined before the very end after all the standard decoding is done. May have linked it before, but this manual describes a basic decoding chain. More fancy VCRs may have more complex stuff and other filters for noise reduction. It's possible this is disabled in EDIT mode on some VCRs, but I don't know for sure, also possible that early VCRs did not have this.

There's also a comb filter (a very basic/old VCR may only use a simple notch filter) at the composite input of a VCR. If you are recording from a composite source it will go through some sort of Y/C filter to separate Y and C for tape though, how complicated depends on the VCR. Same thing with a analog-to-digital video chip.

[lordsmurf]

I've been asked about this thread a few times, and recently asked to weigh in with my thoughts. So, here goes...

It's an interesting project, and I've followed it for quite a while now, even before this thread was made. I figured progress would be slow, if progress was made at all. I've been around digital video for a long time, and have seen much vaporware, lots of broken promises. To me, the most interesting aspect of this project is the potential for true software TBC (not just software-controlled hardware).

Recent advancements in the project now have a "usable" image. Great, congrats! But some folks are getting a wee too excited. As impressive as the project is, from all samples I've seen thus far, the quality is currently about on par with a Chinese Easycrap/EZcrap USB card, using noisy composite, plugged into a cheap consumer VHS VCR. Hence why I used the term "usable" in quotes, because it's pretty lousy. So while it is indeed an impressive feat, it won't dethrone the standard hardware workflow anytime soon (JVC/Panasonic S-VHS VCR with line TBC, external framesync TBC, known-quality capture card).

Understand I'm not criticizing, however also not being a blind cheerleader. Just observing. And that's where this is right now.

But point is ... it's an image! Congrats on the progress! Keep going.

[dellsam34]

Indeed, With a good S-VHS VCR like JVC and good capture device like the BE75 SDI capturing from S-Video the results are not bad, I have to admit though with the VHS-decode method the frame is very stable, The quality is not there yet but signal timing is dead on.

[Zcooger]

Yeah, fine-tuning for the quality is the milestone for this project.
All I was always looking for is complete control over the signal and ability to get all frames that would be dropped by standard capture cards. The ability to freeze the magnetic media degradation sounds wonderful.
For these who didn't see the frequency response (you can play with filter values in formats.py file):

[Attachment 51492 - Click to enlarge]

[Attachment 51493 - Click to enlarge]

[Attachment 51494 - Click to enlarge]

[Attachment 51495 - Click to enlarge]

These examples represent cases for bad cabling, bad filter alignment or the difference between LP and SP modes.

[Titan_91]

I saw the Cartoon Network sample played back from a real VCR which didn't have as many white horizontal dropout stripes. Do you know if high end VCRs did some kind of digital dropout compensation?

[oln]

vhs-decode currently has no droput correction so any dropout will show up as noise/stripes.

VCRs usually have analog dropout (luma) compensation - on newer ones it's mostly done by replacing the video signal by a signal delayed by 1 line by a CCD chip (old VCRs used glass delay lines) when a dropout is detected, as part of the analog luma noise reduction circuit. A few high-end VCRs have digital dropout compensation but I don't think it's very common. The technical manual for the JVC-HR S9500 here seems to suggest the 3D digital board is involved in DOC. My S8500 doesn't act like the DOC is digital though, maybe it's only for NTSC or maybe it's only for the 9xxx models. The digital board in the newer JVCs - like 9700 here seems to have changed and does not have a connection to the dropout detector.

The AG1980P may have it in it's fancy digital circuit, but I don't know, I think the huge broadcast models with TBC may have it as well. Haven't seen anything about it in other prosumer panasonic service manuals.

For 8mm video, I suspect the newer sony camcorders also have some sort of digital DOC (or whole decoder), as they seem to be able to repeat a line infinetly on large dropouts.

There is an interesting comparison between the DOC in various VCRs here.

[dellsam34]

Was wondering if someone could get one of these devices and reverse engineer them for VHS-decode, Nothing better than re-inventing the wheel:
https://forum.videohelp.com/attachment.php?attachmentid=47487&d=1544778291