| jwb |
So, S/PDIF sucks bad. RCA-terminated coax sucks really, really bad. I decided to use something better for my digital interconnect.
I started with El Cheapo, my JVC DVD/CD transport. Under the cover, there is an integrated stereo DAC with a 3-wire serial input. The data format is I2S. To move the I2S out of the box, I decided to use a quad LVDS transmitter (National DS90C031 in a SO-16) on the transport side and a quad LVDS reciever (National DS90C032 in the same package) on the DAC side. These parts are really small, but that's part of their advantage, and it could be worse: the 3.3V parts come in TSSOP. You can get them from Digi-Key which is a triple bonus in my book.
For the physical interface, I got some 3M surface-mount mini-d ribbon (MDR) connectors (part 10226-1210VE). Allied stocks these. This is a nice cable tech: 13 shielded twinax pairs in one cable, and the cables are widely available from computer stores. You could also use the more widely available 20-pin MDR connector, but Allied only stocks the through-hole version of that part. If you need to make your own cables, you could use twinax or twisted pair for runs up to a few meters.
After I got these parts and some tiny PCBs made up, everything else was cake. The LVDS parts need decoupling, and I went with the triple decoupling scheme described in National's data sheet. The LVDS tx/rx chips look pretty dinky next to a 1206 cap :) Also the reciever required a quartet of 100Ω terminators.
So far, no listening. The digital signal emerges at the other end of the receiver, which is good, but I need to modify the DAC to use the serial input instead of the AES receiver. But I'm pretty geeked about this interconnect, and I believe it will work better. It almost could not be worse than S/PDIF over coax.
There is a lot of room for variation here. SCSI VHDCI or HD80 connectors are suitable and widely available, albiet with 36 unused cable pairs. I would have used SCSI connectors if I had some laying about. Certainly you could also just use a group of twinax, or STP ethernet cable, or DVI.
Have fun, and to hell with S/PDIF. |
|
|
| HBarske |
| LVDS transceivers are fairly used in the Mark Levinson D/A-converters. They carry the I²S-signals from the digital front end to the converter boards. |
|
|
| jwb |
Well that's a slight validation of this application, I suppose. Obviously LVDS has been used for many high speed applications, but I hadn't seen it in any audio equipment before.
Do you know of any audio (or video?) systems using LVDS to transmit serial digital audio between chassis? |
|
|
| ftorres |
What about positive ECL (PECL) for this purpose ? Has anybody already tried this ? Are LVDS faster ?
I've attached the I2S (enhanced) input section of the Assemblage DAC3.1 It uses MC10H350 for receivers. Transmitters on the other side must be MC10H352, if memory serves. |
|
|
| Jocko Homo |
LVDS and ECL......I love it. Now we are getting somewhere.
To hell with SPDIF.
Jocko |
|
|
| jwb |
| I dunno ftorres, your picture looks pretty complicated. PECL requires both-ends termination and uses more power than LVDS. I made the LVDS link with only 2 ICs and 4 passives. |
|
|
| HDTVman |
LVDS is used a lot for data transfer from one chassis to another in HDTV broadcast systems. 19.39 megbits/s Is that fast enough for ya? We use ECL and PCEL too. It just depends on what your doing.
Now just how did I know that? :joker:
Later
Bruce :geezer: |
|
|
| ftorres |
| Not as complicated as it looks :) Just 'extra' components : input transformers, impedance matching resistors, pullup resistors, and so on. IMHO, the same goes for LVDS. If impedance matching is achieved (even with your ribbon cable - typically around 100R), you will have IMHO much better results than with no matching. Transformers are nice here, but you can try without... It's up to you. Whatever the technology behind, you must take care of the propagation between transmitters and receivers, and try to avoid any reflections caused by improper matching. Jocko ? |
|
|
| jwb |
| With LVDS, you don't need the pullup resistors. That is a PECL-specific thing. Of course you terminate your cable with a 100Ω resistor. In fact LVDS doesn't work *at all* without the resistor at the end. But still, it is extremely simple: on tx->cable->resistor->rx. Pulse transformers not needed. |
|
|
| Brian Brown |
Hi,
I'm new to this forum, and this seemed to be a good place to jump in since it's a subject near and dear to me.
I was planning on using the same approach with LVDS transmitters and receivers, although I haven't actually done it yet.
My thoughts had been to use a CAT5 flat cable (four twisted pairs) with an eight pin modular connector. It's reasonably cheap, easy to put together, and can easily be fitted on the transmitter circuit board in a small space with a square hole to plug in the cable on the back panel of most players.
One concern that I've had with this approach is how much jitter exists in the player to begin with. You don't want to simply transmit and capture the timing of a clock that's jittery to begin with.
My plan with the CAT5 cable is to use a precision oscillator *at the receiving end*, transmit it over an LVDS pair to the disk player to replace the player's original clock source, and then have it reflected back by a transmitter back to the receiver along with the data. The data would be latched into a buffer using the reflected clock, and read out of the buffer with the clock directly from the oscillator. This way both units are fully synchronous with each other, but with the clock source closest to where the data timing is most critical.
Anyway, that was my plan on how to approach this. It'll be at least a couple of months before I can actually turn boards to give it a try. I'd be interested to hear other peoples' thoughts and experiences regarding this topic.
Regards,
Brian.
P.S.
I think there's a lot to be said for using an ASRC on the receiving end. IMHO these really do a great job, even if S/PDIF is used. |
|
|
| alex douglass |
| This is an interesting post. Would you mind sharing the details of your PCBs (circuit layout & schematics)? |
|
|
| alex douglass |
| Sorry, my post was for jwb. I'm interested in the LVDS stuff. |
|
|
| hifiZen |
:up: Lookin good... I was going to use LVDS with a different connector for exactly this same purpose, but jwb, I think you've chosen a better one... Should we create our own little "standard" for high grade DIY digital interconnect?
Then we could exchange digital module designs, PCB layouts etc based on this scheme...
If there's enough interest, we should add this to the wiki. I'd be happy to collect the info and do the first draft write-up. |
|
|
| artnyos |
There is already a standard, SPDIF. :nod:This interface can be made to sound excellent if you know what the hell you are doing.:wiz: The issues of impedence matching, pulse transformers, cable length, connector reflections, common mode noise rejection, optimum rise times, and hysterisis in receivers will not dissappear by jumping to LVDS. :tons: You think impedance matching for coax is hard, wait till you match twisted pair which has a differential AND common mode characteristic impedance.:wrench:I think you guys had better understand the limitations and strengths of SPDIF before you throw even more complicated "solutions" at the problem.:smash:
Like Mark Twain said:
"Everyone complains about the weather but nobody does anything about it."
Sorry to rain on your parade.:umbrella:
Art:wave: |
|
|
| Jocko Homo |
While all that may be true, what do you do about jitter in the reciever?
Yeah, yeah....I can hear it now......multiple PLLs with VCXOs.
Jocko |
|
|
| Ren Hoek |
| Hey... we could put on a show right here in the barn! |
|
|
| Elso Kwak |
| quote: | Originally posted by Jocko Homo
While all that may be true, what do you do about jitter in the reciever?
Yeah, yeah....I can hear it now......multiple PLLs with VCXOs.
Jocko |
Hi Jocko,
No, no, just a 100MHz Asynchronous Reclocker before the DAC.
Does wonders to the sound; "metastability" problem solved!:idea: :) |
|
|
| Brian Brown |
artnyos:
I agree with you that S/PDIF can be made to work well, especially if it is reclocked on the receiving end.
Yes, proper design guidelines must be followed for LVDS, but I don't consider these insurmountable.
(For those interested,
LVDS Design Guide
is an excellent source of information.)
My main motivation for an alternate interface is for a multichannel audio: direct digital from a DVD-A, DTS, or SACD source. S/PDIF doesn't support this without resorting to some form of compression.
1394 or USB could be used as an interface, but the overhead for these is much more painful than a proper interface that transfers discrete signals in real-time, rather than in packets.
As far as options for getting audio clock and data signals from one box to another, I've considered 422, PECL, LVDS, and even plain single-ended TTL (for a very short distance).
My conclusion was the same as jwb, that LVDS offered the best combination of performance and ease of implementation.
I like hifiZen's idea of trying to come up with a standard for this.
There could be some degree of flexibility for variations in the clocking and data formats.
I'll have to think a bit more about MDR vs RJ-45 modular connectors. MDRs are a more expensive solution, but they do offer more pairs. In order to limit the number of pairs needed for multichannel, I was considering an EPLD to switch between different data sources in the disk player (2 channel IIS, 6 channel IIS, 2 channel DSD, 6 channel DSD) and then pack the selected source together in a TDM type stream. The use of an MDR connector would possibly allow the data lines to remain separated, but this would also require at least one additional quad LVDS transmitter/receiver pair (almost the same cost as the EPLDs).
Anyway, I'd appreciate feedback.
Brian.:cubist: |
|
|
| jwb |
I like your ideas Brian. I'll try to post my schematics and layouts here when I get some time. This weekend, I suspect.
I don't think RJ-45 termination is going to work for a standard. As you correctly point out, that only gives you 4 pairs. You'd need five pairs for 6-channel audio and two clocks. Actually, if you know the word length, you can derive the word clock from the bit clock, so I guess four pairs would work. I'll have to think about it.
For that matter, you could multiplex more than two channels on one serial stream, through obvious methods. You would need more logic, though.
Another disadvantage of twisted pair over twinax is distance. You could easily run twinax across or between rooms. I don't know why that would be desirable, but there you go. |
|
|
| hifiZen |
| quote: | | Hey... we could put on a show right here in the barn! |
Say, is that you on the left there, Ren? You must be wearing your best duds for the hoe-down! I hope you call a good square dance. See you there... |
|
|
| rfbrw |
Brian,
If you choose to serialize and deserialize multichannel audio I am sure at least some of the benefits of doing away with the AES rx will be lost. Given that even in the extreme you are unlikely to have more than 8 individual channels, 16 data pairs does not seem unmanagable and as you intend to use a FPGA, even the low-end devices have LVDS I/O.
ray. |
|
|
| hifiZen |
| quote: | | My main motivation for an alternate interface is for a multichannel audio: direct digital from a DVD-A, DTS, or SACD source. S/PDIF doesn't support this without resorting to some form of compression. |
Mine as well. People constantly post about wanting DTS decoders etc, when you can buy an inexpensive DVD player that has DTS decode built right in. MLP decode will become more common soon too, so why not just mod your player so you can tap the I2S busses and pipe that straight out to your external digital DIY stuff?
In my view, MDR is the way to go over RJ-45. When spending a few hundred bucks on a decent digital project, I'm happy to spend an extra few dollars for the superior interconnect. Don't forget that you may need to send other info such as preemphasis status or other flags, user data etc, in addition to the I2S data and clocks. So, there is even more reason to move away from RJ-45. Now, as for all this auxiliary data, perhaps we should specify something like I2C as the standard method of carrying "other" data. The benefit of going with I2C is that it is scalable without affecting the physical specifications of the interface. It can go unused for those who just want the pure audio data. At first I hesitated to add an interface like this in with the audio data lines. But, the impact on jitter performance actually should be basically nil anyway, provided the master clock originates at the DAC end. If the master clock is the source, what then should we do with the auxiliary data? Relegate it to an entirely separate cable?
Anyway, for a bit more inspiration, check out the I2S-e specs put forth by Ultra Analog, and as seen in Sonic Frontiers equipment. |
|
|
| ftorres |
RJ45 connectors are not the best regarding impedance. We've tested some of them at work, both for EMC and signal integrity, and some TDR (yes Jocko) measurements have shown that the pin arrangement leads to nice impedance discontinuities. IMHO, best (quality/price) connectors for our purpose are the type SonicFrontier uses : Sub-D shells with 50R coaxial inserts. They come in several flavours, from 2 to 5 places for coax, plus some room for "normal" pins. The drawback is that you have to make your own cable, but I use some lengthes of miniature 50R coaxs tied together with a heat shrinkable tube, and it works fine.
Just my $0.02 |
|
|
| hifiZen |
Hm. Interesting. I would certainly expect the 50ohm coax connectors to perform best... actually you might do even better with separate SMA or SMB etc. connections for the main clock if you want to be really anal about impedance. The I2S-e connector is indeed very nice, but maybe a little limited for our multi-channel applications?
Did you do any measurements on these MDR connectors by any chance?
Thanks, |
|
|
| ftorres |
Sorry, no measurements yet for MDR connectors. :(
Concening the Sub-D shells, I've attached a picture of various ones along with coax inserts. These ones are made by FCI (Framatome Connectors International), but other suppliers like 3M, Amphenol, Amp, etc are manufacturing them too. Up to 8 insert places are available on the same connector. They're cheaper than the same number of SMA/SMC/SMC/MCX connectors, and far easier to plug :). Ok, their bandwidth is smaller (1GHz), but large enough in this application.
Edit : I know that Harting makes some miniature multipoint coax connectors, but they're apparently very hard to source. I'll try to find the web link back. |
|
|
| hifiZen |
| Ooooh, nice! Cheaper than separate coax, but still pretty expen$ive, no?... ;) |
|
|
| Jocko Homo |
Hmmmm, looks like someone has been taking an old H-P spectrum analyzer apart. Guess that means that they probably are expensive. And that they do work.
Jocko |
|
|
| Brian Brown |
| quote: | Originally posted by ftorres
RJ45 connectors are not the best regarding impedance. We've tested some of them at work, both for EMC and signal integrity, and some TDR (yes Jocko) measurements have shown that the pin arrangement leads to nice impedance discontinuities. IMHO, best (quality/price) connectors for our purpose are the type SonicFrontier uses : Sub-D shells with 50R coaxial inserts. They come in several flavours, from 2 to 5 places for coax, plus some room for "normal" pins. The drawback is that you have to make your own cable, but I use some lengthes of miniature 50R coaxs tied together with a heat shrinkable tube, and it works fine.
Just my $0.02 |
Did you try CAT5 rated RJ-45 connectors? I believe these have better symmetry in the pin arrangements than regular RJ-45, and are intended to be impedance matched with CAT5 cable.
I realize that most generic RJ-45 jacks split the signals into two rows (one row with longer pins), which screws up the quality of common mode rejection.
I've considered the Sub-D's with the 50R inserts, but the cost has made me leary of these. They're also a bit of a pain to wire. Also, if unbalanced coax is to be the medium, then LVDS isn't the best type of interface. I'd probably go with a 422 driver, possibly with transformers.
I don't want to come across as being too hung up on cost. I'll happily spend money to try out new parts and turn a board to play with them. (NPC wanted $190 for samples of their SM5816AF that I intend on using as part of this interface. This is no problem for me because it was clearly the best option for me to be able to process DSD signals at this time.) I just don't want to commit to a higher cost solution than is necessary, especially if it's to be a 'standard' that I'll have to pay for again and again. Of course I don't want to limit the performance potential either. It's a matter of figuring out the cheapest and easiest way of acheiving the desired goals.
The key word there is 'goals'. It's usually harder to figure out what precisely you want to do, than it is to do it.
Having said that, I think that if we want to try and come with some sort of DIY Audio Digital Interface Standard (DIYADIS?), it would be a good idea to temporarily turn the discussion away from the specifics of implementation, to try and come up with a set of performance and feature requirements, along with an outline of the general architecture.
I'll try to come up with an outline of my thoughts for requirements and topology in a later post.
Brian. :cubist: |
|
|
| jwb |
This is the reason I chose the MDR cable. It is a good cable tech, designed for the signalling technology, that you can buy right off the shelf. Standard unshielded twisted pair is right out, because it is unshielded. You will have a much harder time finding shielded network cable, because nobody uses it, and the jacks and plugs are more rare as well.
The only problem I can see with the MDR is it is no longer used for computer monitors, and therefore it might become difficult to find.:xfingers: |
|
|
| alvaius |
Art, do you ever have a positive post? I hate to ask that, but everywhere I see your posts, they are always negative.
SPDIF is a low speed interconnect. It was designed for low speed, it is low speed, period. The whole design methodology for SPDIF was a cheap and somewhat robust. Low jitter, if it was a consideration, was certainly not evident.
LVDS was designed for low noise ultra-high speed short range (5-10 meters), links. LVDS parts are speced up to 1 gigabit per second. It is not uncommon to run multiple links in parallel.
Will there be reflections in LVDS? Of course there will. However, LVDS edge speeds are so fast, that the "susceptibility period", i.e. the period during the transition where a reflection could cause a bit flip is so small, that you would need to really screw up the implementation to run into a problem. And unless you really screwed up the termination, the chances of the reflection causing a bit flip once the level is solid, is just about nill. Also keep in mind, that a reasonable level of hysteresis is built into the chips to avoid just such problems.
100 ps (typ) of pk-pk jitter is a spec for some current parts with no PLL to clean it up... |
|
|
|
