|
It is currently 07 May 2024, 20:49
|
View unsolved topics | View unanswered posts
|
|
|
|
Author |
Message |
bguzman
|
Post subject: Re: (WIP) Laserdisc software image decoder from raw signal Posted: 13 Jul 2014, 21:47 |
True fan |
|
|
Joined: 17 Aug 2013, 04:49 Posts: 398 Location: California USA Has thanked: 1 time Been thanked: 2 times
|
happycube wrote: Dusted this off about a week ago and made some progress - I implemented a Hilbert transform for FM demodulation in my Python version, then removed all scalar code and got slightly better and about 70% faster results than the C version. The nice thing about working in Python is that one can iterate a lot faster, and I was able to fine tune things a bit.
One of my tests is decoding a still frame several times and comparing the PSNR of the results. Interestingly, the current code when decoding frame 40735 on side B (a test black frame) of THX Wow! gets the exact same results for each captured frame once the frame transition bit is cropped - because of the intrinsic signal noise and motor speed variation, the TBC has to be working perfectly for that to happen. The same comparison with a frame of VE gets 49dB (between one frame and the average) and the Pioneer ref disk only got 46, probably because the V2800's spindle is less accurate with the smaller disc.
White SNR is about 40db with current settings, which is less impressive but OK given the lack of NR that most players do have.
At this point the comb filtering needs the most work - on some scenes/frames, it looks fine, but going from color to color is rough.
(test pics will be edited into this post soon)
_________________ Main - CLD-95, LX-1000U W/AC-3 -> Faroudja NR Series -> Display Backup - LX-900, CLD-D703
|
|
|
|
|
msgohan
|
Post subject: Re: (WIP) Laserdisc software image decoder from raw signal Posted: 16 Jul 2014, 15:51 |
Serious fan |
|
|
Joined: 15 Jan 2013, 22:44 Posts: 136 Location: Canada Has thanked: 5 times Been thanked: 1 time
|
happycube wrote: White SNR is about 40db with current settings, which is less impressive but OK given the lack of NR that most players do have. It took me a little while to parse this (it doesn't help that the "NR" in "SNR" stands for something different than "NR" alone). I think what you're saying is that the SNR from your decoder is okay when you consider that we're comparing a raw ratio -vs- players' reported ratios which have been massaged with noise reduction? Following with interest as always.
|
|
|
|
|
Guest
|
Post subject: Re: (WIP) Laserdisc software image decoder from raw signal Posted: 18 Jul 2014, 09:22 |
|
msgohan wrote: comparing a raw ratio -vs- players' reported ratios which have been massaged with noise reduction? It's interesting how you put that. That's why I personally prefer a player from the 80s. Most players that started with the internal comb filter to svideo thing, just don't seem so great to me. I honestly don't trust any player with digital CLV still frame features, and increased SNR. A lot of times SNR is increased with DNR and other noise filters, which doesn't always represent what's on the LD IMO. Also can't help but notice a distracting slimness of color with capture shots from HLD-X0 and other supposedly great players; I get way better color through a line doubler than most shots I've seen from the HLD-X0 on the net. Great work happycube! I wish I was smart enough to learn python and all that stuff you're dealing with.
|
|
|
|
|
happycube
|
Post subject: Re: (WIP) Laserdisc software image decoder from raw signal Posted: 18 Jul 2014, 17:20 |
Absolute fan |
|
|
Joined: 18 Apr 2012, 18:02 Posts: 1614 Location: United States Has thanked: 71 times Been thanked: 88 times
|
Vexatious: That's part of what makes LD so interesting to me, it isn't just a "plug in via HDMI and if you're smart adjust TV settings" thing like a Blu-Ray player. There are many ways to get good results And just start playing with things, find a problem that interests you and take a spin at it, even if you don't get far at first. I think what I've been getting recently is close to what's on the disk, but could be better. There's a resolution/noise tradeoff I'm trying to work out. I'm also looking into probability and probabilistic programming - and I'm thinking it might be useful for comb filtering. (Anyone got suggestions for LD's and DVD's based off the exact same digital master? I'm going to take a look at Titanic, which wasn't anamorphic in it's first release.)
_________________ Happycube Labs: Where the past is being re-made, today. [meep!]
|
|
|
|
|
rein-o
|
Post subject: Re: (WIP) Laserdisc software image decoder from raw signal Posted: 18 Jul 2014, 19:35 |
Jedi Master |
|
|
Joined: 03 May 2004, 19:05 Posts: 8116 Location: Dullaware Has thanked: 1227 times Been thanked: 851 times
|
vexatious wrote: msgohan wrote: comparing a raw ratio -vs- players' reported ratios which have been massaged with noise reduction? It's interesting how you put that. That's why I personally prefer a player from the 80s. Most players that started with the internal comb filter to svideo thing, just don't seem so great to me. I honestly don't trust any player with digital CLV still frame features, and increased SNR. A lot of times SNR is increased with DNR and other noise filters, which doesn't always represent what's on the LD IMO. Also can't help but notice a distracting slimness of color with capture shots from HLD-X0 and other supposedly great players; I get way better color through a line doubler than most shots I've seen from the HLD-X0 on the net. Great work happycube! I wish I was smart enough to learn python and all that stuff you're dealing with. then you would want to look into an LD-S1, that has NO S-video, only thing it has on the front is a sharpness knob but that's more like the knob on the older TVs to change sharpness etc. new players give great images but a lot of these older ones give great images also ________________________________ the rein-o official signature:
|
|
|
|
|
happycube
|
Post subject: Re: (WIP) Laserdisc software image decoder from raw signal Posted: 24 Aug 2014, 19:40 |
Absolute fan |
|
|
Joined: 18 Apr 2012, 18:02 Posts: 1614 Location: United States Has thanked: 71 times Been thanked: 88 times
|
Did a bit more since the last update: - Added noise reduction in an analog-like style, following the description in a Pioneer patent. Signal/noise ratio varies by level - I can get 45-48dB now. It also allowed me to increase resolution at the same time. - Put in dropout detection (they were much more visible with the newer FM decoder, which I debugged a bit more in the process) - disks with dropouts/speckles can still cause picture errors, though. - The comb filter is more or less a proper non-adaptive 3-line 2D filter now. Not great, but pretty good with real footage. - I wrote a simpler and slightly more effective TBC in Python, but it's dog slow even by my standards. I finally updated my sample video: https://www.youtube.com/watch?v=IEpt5nQ ... e=youtu.be - this is an old Pioneer Japan pressing, so it gives the dropout compensation quite a workout --- Other musings: Pioneer actually got a patent in 1986/7 for an all-digital video signal decoding path - everything from FM decoding to comb filtering (US 4,736,238). (I might be able to use the related FM decoding bits, patent #4656432) It looks like memory et al made it too expensive at the time to implement, sadly. It's quite possibly prior art for CubeTek's videotape decoding work. edit/addon: The FM decoding scheme, probably used because doing arctan's was too much in the 80's, was definitely lacking - I got it going last night and it's much worse than the arctan method I (and it turns out, a lot of 80's VTR digitization work) use. Unless there's something I'm missing, Pioneer probably didn't use it because it wasn't very good, and commercially it was a bit late by the time it wasn't too hard to do right. *sighs* There's also a ton of prior art about handling VTR data in a computer/digitally that I was able to pull down at the local uni library.
Last edited by happycube on 26 Aug 2014, 16:29, edited 3 times in total.
_________________ Happycube Labs: Where the past is being re-made, today. [meep!]
|
|
|
|
|
Guest
|
Post subject: Re: (WIP) Laserdisc software image decoder from raw signal Posted: 25 Aug 2014, 01:19 |
|
Dang that rally car reminds me of the 78 Saab 99 sedan I had. Got stolen and crushed and I've been sad since.
|
|
|
|
|
Guest
|
Post subject: Re: (WIP) Laserdisc software image decoder from raw signal Posted: 02 Sep 2014, 09:44 |
|
This is great news!
I am following this project with great interest. Are you still using the same capture device but with new FM decoder? How much more quality do you think you would get with more expensive hardware?
|
|
|
|
|
acuozzo
|
Post subject: Re: (WIP) Laserdisc software image decoder from raw signal Posted: 18 Sep 2014, 22:22 |
True fan |
|
|
Joined: 08 Jan 2013, 18:13 Posts: 320 Location: United States Has thanked: 0 time Been thanked: 5 times
|
happycube wrote: USB3 56mhz/10+bit capture device Such as? You might be able to crowdfund this over at AMPS.
_________________ Fill my eyes with that DiscoVision!
|
|
|
|
|
Guest
|
Post subject: Re: (WIP) Laserdisc software image decoder from raw signal Posted: 22 Sep 2014, 20:03 |
|
|
|
|
|
happycube
|
Post subject: Re: (WIP) Laserdisc software image decoder from raw signal Posted: 23 Sep 2014, 03:02 |
Absolute fan |
|
|
Joined: 18 Apr 2012, 18:02 Posts: 1614 Location: United States Has thanked: 71 times Been thanked: 88 times
|
It'd need to be a straight ADC (and about 10% the cost!), but that's close to the idea. I can probably get where I want to be with current capture quality, but better quality raw data can't hurt. I'm more interested in a NXP ARM microcontroller that has USB 2 high speed and a 80mhz 12-bit ADC on a reference board (lpclink 2) for $20. If the DMA can support ADC->USB streaming well enough, that and an analog board could get me a nice ~32mhz/8bit gain adjusted signal, and would be easy enough for someone else to put together as well. (The microcontroller could probably handle digital audio decoding natively, but doesn't have nearly enough power to do video demodulation and time base correction) Did more on the TBC code, it now works directly with audio, and seems quite accurate although it can still be fooled by rotters. Right now I'm looking at the comb filter code again, hopefully I'll get around to making it 3D one of these weeks. A CX audio decoder would be nice too... I still want to do AC3, but without the error correction specifications it'd be very hard to do if it scrambles the data like regular digital audio. I should've asked disclord the right questions about getting in touch with the right people...
_________________ Happycube Labs: Where the past is being re-made, today. [meep!]
|
|
|
|
|
happycube
|
Post subject: Re: (WIP) Laserdisc software image decoder from raw signal Posted: 09 Oct 2014, 06:37 |
Absolute fan |
|
|
Joined: 18 Apr 2012, 18:02 Posts: 1614 Location: United States Has thanked: 71 times Been thanked: 88 times
|
Making some headway on 3D comb filtering - here's a 10 second bit of S&W: https://www.youtube.com/watch?v=uZ_ZlEP6HKwThe code itself is still rather crude - it currently only looks at three frames, lines, or pixels if all else fails, and compares the forward and back samples to each other to decide if there's motion/gradients/etc. The end result is workable but not teriffic, and the chroma filtering is resulting in dot crawl (and the lack of it results in worse artifacts) Now to make some headway on CX decoding... just not tonight (yawn)
_________________ Happycube Labs: Where the past is being re-made, today. [meep!]
|
|
|
|
|
|
|
|
|
|
|
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot post attachments in this forum
|
|