I think the 750 PCI-E is basically undocumented and has no Linux driver. The PCI-e successor to the card I use probably has a raw mode, but the datasheet/register specs are kept under NDA lock and key so I haven't dug into it. Would be nice especially if it allows higher capture speeds.

USB versions are usually very different, so I could see someone whipping together SDR mode for that - any idea where I can get more info on it? (The regular rtl-sdr in SDR mode has enough bandwidth for audio capturing - if it's in direct/tunerless mode - but nowhere near enough for video)

Here is the guys github page. https://github.com/roysjosh/tvw-sdr It seems pretty undeveloped actually, now that I look at it.

Interesting, but the only speed reference I see is 8mhz@12bit, and I need at least 27mhz/8bit+ (maybe 13.5 I/Q)

Too early to say. I'm personally very optimistic about the LD-4300D, as noted elsewhere, the timebase looks very stable on it. However, aside from gross tracking errors, sampling the RF signal directly after the laser head removes a great deals of the player from the equation: It may not really matter that much which player you have.

If you already have a player: I would suggest starting there. Modifying the player to add an RF output is probably going to be fairly trivial compared to everything else.

Only the early diode laser players have a direct full-range RF out on the back, and I don't have any of them. AC3 RF is only usable for audio - disclord said it was supposed to be full range, but in practice they took it off the pre-filtered audio RF signal.

The LD-V2000 has the IO port and may still have the RF (I checked the S1 service manual - it has an IO port with no RF), and early runs at least have the red laser, the pickup PN is the same as the 1010's - but it was made through 1990, so who knows... It might be actually be the best player for this if it's in good shape.

Later players have a standardized diagnostic test strip with RF, which is relatively easy to hack up a cable for. I use a CLD-V2800 myself, they survive 'less than perfect' shipping, and the test strip points out to the right side of the case, so I mangled the side so I could get the cable through there.

Which brings this around - the V2800 is basically a repacked S104 which is about as low-end as you can get As long as the player tracks well (which is an argument for later players, which are known for even tracking dodgy Discovisions pretty well) and you have a clean RF signal you can get good results.

I think the quality bottleneck now is the RF capture itself. I was able to get 0.5-0.7 DB more this morning by adjusting the capture offset, and I think a better capture system would probably get 2-3 more DB.

This might sound a bit strange but I don´t own and have never owned anything laserdisc related. I am interested in this coming from tapebased media, hoping I can use this to capture RF from rotting tapes before they completely die and in the future restore the sometimes lost signal. I have a friend who is a real LD enthusiast so we will work on this together. But he understandably does not want to tamper with his nice working players so we need something to experiment on. Will have a look out for the later players that hopefully has better tracking.

Do later players from all brands have this standardised diagnostic test strip?

What benefit would 2-3 extra dB give us in noticeable picture quality do you think? When I get the system up and running I could try and capture some frames with a digital oscilloscope. Have you been able to do something similar to compare?

Best regards

Just Pioneers have the test strip I mentioned. And I don't have a 'scope with deep enough memory to do a frame capture, but if you have one I don't see why it wouldn't work.

Luminance noise coring increases final quality by several dB over the raw decoding. ffmpeg's own processing performs the chroma NR. On a good disk the end results are currently quite good.

A higher depth capture would help CAV disks in particular, since I have to set the capture level according to the end of the disk, leaving the beginning with less RF dynamic range. In addition it may allow for sharper captures with less ghosting and ringing, but I won't be sure about that until I actually do it.

I'm interested in looking at magnetic media as well, I just haven't gotten around to tapping a VHS player yet. Note that capturing sound RF requires a second synchronized RF capture of either the hi-fi or linear heads.

Just a question about comb filtering. I hear everybody talking about motion adaptive comb filtering where there is a threshold for movement and if that threshold is exceeded then the comb filter switches to 2d mode and uses information from surrounding areas in the frame to eliminate artifacts. If the threshold isn't exceeded the filter uses information from a different frame to eliminate artifacts. Wouldn't it be possible now that we aren't limited by 90's electronics and real-time decoding that a motion adaptive comb filter could also be Motion Compensated? Maybe that's not possible in the realm of comb filtering but I know that it works very well with denoising filters and allows the whole frame to be filtered in 3d instead of 2d. Anyway, I know nothing about comb filters but it occurred to me that it could be possible.

Yeah - right now I'm using a low pass filter on luma, which doesn't work all of the time. Eventually I (or someone) should write a better adaptive method. Right now my 'major' project is working on digital audio decoding, which I'm being rather slow at - but I think I have the lowest level decoding taken care of now.

Thanks to some tips from robwhar, I wired up the player without using the VGA amp - the results are not quite as good, but quite acceptable.

I recaptured VE with it and sliced out 8MB raw bits with the frames megapixie pointed out in another thread:

http://we.tl/jk0XbSs2Yf
http://we.tl/ktlj7Pt6Fc

To run ld-decode against these raws:

./ld-decoder.py -s 2 -c -.12 [file] | ./ntsc | ./comb -d 3 - > tmp.rgb
convert -size 744x480 -depth 16 rgb:tmp.rgb -resize 1280x960\! test.png

Decoded images are here:
http://imgur.com/a/JhAGJ

WOW!!! Final result has a very good quality!

Happycube,

Pretty good !
The luma noise results are better than any other player I've measured, at the low-end by more than 4db ! Are you doing any luma noise reduction ?
The chroma noise results are very good (on par with the X-9 ).
Frequency response is a weak spot, starts off strong, but rolls off fast. Still better than the PAL players. The frequency ramp on S&W doesn't have any ringing (which I see on every other player): I suspect the other players are boosting the frequency response somewhere.

The ballerina is giving your 3D comb filter the same problems it seems to give all the others. The mixture of 3:2 pulldown and very similar luma values is a definitely very confusing.

Mega.

Thanks!

Yup, I've got a noise coring system in comb.c - it rolls off as high frequency changes increase, so it doesn't clamp frequency response the way a regular noise coring system does it. I can put up some non-noise reduction pics soon, or you could run it for yourself by adding "-n 0" to the beginning of the comb options list. (Get the latest source first though, I checked in some changes to optimize for the slightly different amp-less captures)

edit: The noise coring was inspired by the 1D analog systems, which I think are used in the later FM demodulation chips.

There's one other unique thing I do - I (try to?) compensate for disk velocity changes on signal levels in the TBC. That helps reduce vertical noise.

Early on I found VE to be one of the noisier disks - I think the AC3 signal gets into the picture.

The frequency response/ringing tradeoff is interesting. I'm still trying to find the right balance. If you change the ld-decode parameter to "-s 3" you get more frequency response but ringing.

edit: There's another comb filtering problem with scene changes, so I'll be looking at that tonight/this week if I'm up for it. Hopefully it hasn't been inflating the SNR too much.

I have the video buffers and capture card. Need to get some wire, caps and a few other bits. None of my computers has a PCI slot, so I'm looking to get a small format computer with a PCI slot (thinking about a small Dell workstation).

Cool Roberts RF tap pictures in the google doc will apply to the D925. The 4300D may need the amp circuits, if it's like the 8000 at all.

I'm going to be looking at the USB2-based LPCLink in the next few weeks - the AirSpy SDR device uses the same LPC4370 chip and it looks like I can do a 10bit 28mhz capture off it easily enough, now that GPLd firmware exists to do the USB streaming.

The smaller-format Dell desktops have half-height PCI slots and won't be able to fit the card - but there might be others with the same chip that are low-profile. The workstations might be better - I'd take a peek at Y!J for T3500's, you can get the equivalent of a 1st Gen i7 in Xeon clothes (the computer equivalent of a CLD-79? ) for a reasonable price that way. Core 2 Quad-based Xeon boxes might be even cheaper, but make sure it comes with enough RAM, FB-DIMMs can get expensive.

Happycube,

Thanks for the tip on the T3500.

The Airspy is interesting, but I only see people testing on the forum at 10MSps... Can you point me at where people are doing better?

Native peak-to-peak is quite low, but that might suit the RF source signal.

Re: FB Dimms. As the proud owner of a 2008 Mac Pro I'm already feeling that pain.

No problem - I don't know what the used PC market is like in Japan, but there's always Y! at least.

The Airspy is capturing and transmitting 20 million 12-bit samples/second, 16-bits at a time - 40MB/sec. It looks like some people can only get ~36-38MB/sec (9.xMs/s) The LPC4370 itself supports up to 80mhz captures, and with repacking (10 bit, maybe 11?) I can fit 8fsc/28mhz into that 36MB. It looks like the main core (200mhz M4) is unused most of the time, so it should all work.

I'm actually going to use the $20US LPCLink 2 with a modified version of the firmware (which is GPL'd) and the same chip. It's going to require voltage compensation and protection circuitry (the AirSpy would need it as well), so it'll be a few, and possibly several, weeks until I can do any captures.

Should have new comb filter code/examples up this weekend. There's some cross-color on the guy's face in the ballerina scene to fix first, but the dot crawl is reduced to the dancer's fingers.

Happycube,

Sounds interesting. I'm wondering if you might do better with 60Msps at 4bits per sample or even 80Msps at 3bps with some software AGC ?

Good luck on the Ballerina scene: that's pretty much worst case for a 3D comb filter, optical flow (or falling back to 2D) is probably the logical conclusion to that arms-race.

I'm heading off to Akiharbara to get some supplies.

I thought this was interesting: http://www.cypress.com/?rID=99916 - A $50 USB 3 interface board that could handle up to 100 MSps at up to 32 bits wide - just need to find an ADC that could be easily interfaced with it.

I don't know how much processing it could do on 80Msps... it's a 200mhz core afterall. And that board caught my eye too, I just haven't figured out what to wire to it yet...

As for the comb filter - the new code I was working on didn't work, but I was able to take the idea and it's a bit better now, but I still need to adjust detection so both the zone plate and the ballerina scene look good. Right now it's a lot like the X9 in C.Wide mode. A lot of the issue is that VE's an intrinsically noisy disk for whatever reason...

(Been working on something non-LD-related today, mostly)