Ethernet "unbalanced" links

[Harvs]

Unbalanced may not be the right term here.

I'm looking to use a interesting slip ring that has a single 1G fibre and a 10M rated copper pair.

I'm hoping there might be some off the shelf reasonably standard solution for putting ethernet across this unusual combination of links (i.e. 1G in one direction and 10M in the other.)  My googling hasn't turned anything up, but maybe that's just because I don't know the right search term, or maybe it just isn't a thing?

Any ideas without resorting to a custom switch?

[nightfire]

Well, the speed usually is no problem- you can also put 10 Mbit/s over fibre, if you want...

Could you describe your application a bit?
Do I really think that you want to have at the same logical interface at the system 1 Gbit/s datastream down, and 10 Mbit/s up? (or vice versa)
From manufacturers like Leoni there are so called "Hybrid cables", that incorporate copper and fibre in one piece of cabling.

[ogden]

Before we discuss "how" - why don't you tell "why"? What's the point of fibre+TP_copper bidirectional link? You may consider bidirectional single fiber transceiver modules and drop TP completely. First in the search hits: https://edgeoptic.com/products/sfp/bidi-1-25g-sfp-20-ad/

[ve7xen]

The word you are probably looking for is 'asymmetric' a la A(symmetric)DSL.

Ethernet, as in 802.3, is symmetric. If you want asymmetric bandwidth, you need something that is not Ethernet. The closest thing in 'standard' networking technology would be VDSL, which can get you up to 300/50 or so over a single pair. This would probably run over your '10m' pair. You can get off the shelf VDSL2+ bridges, often sold as Ethernet extenders. I'm not aware of anything POTS-based that can operate at 1G in one direction.

If you have a single fibre strand, you can run a BiDi optic and just do 1G or 10G Ethernet over it.

Another option would be wireless.

But I agree, more details on how you have ended up at these constraints and what problem you are solving would probably lead to better answers.

[Harvs]

We're looking at a specialised large ID slip ring in order to pass low speed control up to the payload, and "high speed" sensor data back.

Moog make such a slip ring off the shelf, where the normal application is for MRI machines (so the ID is large enough to fit a human through, but that's not what I'm doing.)  The slip rings have a unidirectional fibre rated for 1.25Gbps and a twist pair rated for 10Mbps.  They'll also happily make any combination you want, noting that the current off the shelf item is a number of months lead time and a small number of tens of thousands of dollars each. Changes to the off the shelf design require covering all NRE costs and lead time, so yeah...

We currently use RF and used a few different methods, but there are a number of issues such as the inability to run 1588 PTP at decent resolution.

Essentially there's more than one solution on the table for a whole bunch of requirements that I won't discuss. So I was just looking at whether throwing some money at the problem could make it go away in a solid "off the shelf" solution.

[ogden]

Why don't you use high speed connection for slow speed control? Using plain 1GBps bidirectional Ethernet over two or one fiber will be least complex solution one can ever dream of. Anything else will increase complexity and inevitably reduce reliability. Oh, and 1588 PTP needs *symmetrical* bidirectional Ethernet connection.

[ve7xen]

We're looking at a specialised large ID slip ring in order to pass low speed control up to the payload, and "high speed" sensor data back.

Moog make such a slip ring off the shelf, where the normal application is for MRI machines (so the ID is large enough to fit a human through, but that's not what I'm doing.)  The slip rings have a unidirectional fibre rated for 1.25Gbps and a twist pair rated for 10Mbps.  They'll also happily make any combination you want, noting that the current off the shelf item is a number of months lead time and a small number of tens of thousands of dollars each. Changes to the off the shelf design require covering all NRE costs and lead time, so yeah...

Fibre itself isn't directional, so I'd assume you can run 1G Ethernet both ways on the same fibre strand either with a standard (or not so standard but still off the shelf) BiDi module. Singlemode ones are widely available and cheap, sold as the standard 1000base-BX. Multimode ones are out there but not standardized, so slightly less available and less cheap. Given that I assume the distances involved are on the order of single metres, the fibre type probably doesn't matter much and either will likely work in practice with the singlemode optics. Seems like a straightforward solution, though if it's intended for unidirectional use you might need to get some confirmation from the vendor that there aren't circulators or some other weird optics in the path.

[Harvs]

Fibre itself isn't directional, so I'd assume you can run 1G Ethernet both ways on the same fibre strand either with a standard (or not so standard but still off the shelf) BiDi module.

Why would you assume that when Moog rate it as a unidirectional fiber interface?

It's a very expensive slip-ring with an internal hole large enough to fit a human through. So they're somehow coupling an input through a head and onto a track that ends up as a fibre at the other end.  I would imagine this will include some sort of tracking lens like in an optical drive and at least one optical gain stage to compensate for losses in the track.  I don't think these are anything like a normal piece of fibre. 

Small slip rings, particularly those that do not have a through hole are a completely different beast.  With those it's a precision light pipe that you can put anything you want through.

[Harvs]

Why don't you use high speed connection for slow speed control? Using plain 1GBps bidirectional Ethernet over two or one fiber will be least complex solution one can ever dream of. Anything else will increase complexity and inevitably reduce reliability.

Because the manufacture states that it is a unidirectional fibre interface.

Oh, and 1588 PTP needs *symmetrical* bidirectional Ethernet connection.

PTP requires IP packets with path delays compensated for in the compensation field of the packet, or the path delays be symmetrical so they can be measured and compensated for by the node.  If the network appliance on each of the legs can do timestamping properly like any good PTP switch then the differences between having different network speeds on each side of the interface will be insignificant compared to other errors in a typical PTP system (YMMV, I'm not talking white rabbit, just bog standard PTP.)


Anyway, looks like my question was answered.  It would be a custom thing so this slip ring doesn't sound all that appealing.

[ogden]

Why don't you use high speed connection for slow speed control? Using plain 1GBps bidirectional Ethernet over two or one fiber will be least complex solution one can ever dream of. Anything else will increase complexity and inevitably reduce reliability.

Because the manufacture states that it is a unidirectional fibre interface.

Unidirectional fibre interface means it's not even Ethernet nor IP. If there is no IP then there's no 1588 PTP. Oh, and please show single pointer to assymetric 1G/10M 1588 PTP  implementation that works.

[ve7xen]

Why would you assume that when Moog rate it as a unidirectional fiber interface?

Because without very intentionally doing so, all of the optics in a typical fibre or free-space-optics system are symmetric. Why would they put a one-way mirror or circulator in what is intended as a one-way data channel? It's just going to add loss to the system, when if you want a one-way channel, the simple solution is to only put one transmitter on it. It doesn't make a lot of sense for it to be otherwise, so I would assume this is an instance of writing the spec either to an RFP or to what they expect the market expects/understands, not to what the capabilities are. This is, of course, assuming that the bare fibre is what is exposed to you, and there isn't an OEO conversion happening to get through the ring - but if that were the case it seems like there's no reason for using optical at all and doing an electrical interface all the way through would have made more sense. In any case, if it's a completely optical system it is very likely to be bidirectional, but obviously you need to confirm this before buying a dozen of these expensive gadgets.

Edit: This datasheet is lacking in detail but looks like more or less what you're talking about and says 'bidirectional channel support' and most of their other literature says 'all of the advantages of fibre', and suggests that multiplexing and bidirectional links work, and suggests it can carry any signal (which implies no OEO), so it is almost certain to me that there is nothing here that would prevent 1000base-BX from working over this medium. But yeah, before spending my $50k, I'd be asking them or if I had one I'd buy the $50 SFPs and just try it.

[Harvs]

Yep so those 4 options they have for the FORJ (noting they're all separate configurations):
- Single Ch 2.5 Gbps* R->S
- 4 Ch 2.5 Gbps* R->S
- 4 Ch R->S with 1.25 Gbps* S->R path
- Bidirectional Gigabit Ethernet (1000Base - SX)

Only the first option was off the shelf, and it was only 1.25Gbps on the proposal provided.

They also wrote to us specifically to confirm that it was unidirectional, and which direction we were after.

So not sounding so great.

[magic]

Probably can't do it with Ethernet but can do it with IP over two unidirectional Ethernet links.
I think copper NICs can be configured manually to run at specific speed, without bidirectional communications and autonegotiation.
Dunno if the same can be done with fiber NICs.
Dunno about PTP.