| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
|
Oisin – thank you for raising the question. I have understood, and perhaps been assuming, that fast link recovery was to deal with the a drop in SNR. In many systems, a retrain will improve the SNR. Many systems (2.5G/5G/10GBASE-T)
use a fast retrain concept to allow a more rapid startup based on the fact that echo cancellers, equalizers and timing phase may reconverge to different solutions if given the chance to retrain. I am unclear whether 40msec is feasible for a system of this
baud, and whether the result has effect is dependent on how the system itself is designed. For example, in the 2.5G/5G/10GBASE-T systems, these are all Tomlinson-harahima precoded, and the retrain allows a free re-setting of the THP coefficients – which enables
quite different equalizer outputs. Other PR-related systems benefit from similar retrains, and, DFE systems do too, for example, freeing the slicer timing phase in a DFE-based system with a retrain enables convergence outside of a local minimum error state.
These types of parameters are more constrained in what they can track while the link is up than others are. All of this said, a 40msec fast retrain for a PHY at only 10Mbps is pretty quick, and I’m concerned that my first reaction is that it would probably end up at a worse state than the PHY started out in by tracking the environment. As Mehmet
said on the call – it is very much tied in with the PHY design. I am concerned that adding this as a requirement will lead us to PHY architectures and designs which are inherently less robust. I agree that the best design state is that the PHY adjusts rapidly enough to account for the environment. I don’t yet know whether this is true in the environment.
I think this leads me to a similar point as you – the real requirement is link reliability under an environmental condition. Some more definition or examples of the proposed environmental case might be useful. -george From: Cuanachain, Oisin [mailto:Oisin.Ocuanachain@xxxxxxxxxx]
Hi all, 'Fast Link Recovery' was discussed at the ad hoc meeting but I'm not sure I understand what this proposed objective is intended to achieve ?
My own background is in PHY design (10/100/1000BaseT) but I have very little prior experience of Ethernet used in industrial environments.
There was some discussion of the objective but I am not sure I understand what is driving the proposal.
Specifically, does the proposal (i) hope to use the 'Fast Recovery' to remedy packet reception problems (within 40ms) to prevent message loss (here I use 'message' to mean the information which, in the
industrial protocol, is repeated multiple times in successive Ethernet packets to make the communication more robust)
Or (ii) is it simply that industrial end-users are aware that retraining is behaviour that PHYs can exhibit when packets are being lost and the proposed
objective hopes to ensure that should such a retrain occur it would have completed within 40ms ? (In this scenario, from the application point of
view, the retrain manifests itself as the 'outage' depicted on slide 8 of
http://www.ieee802.org/3/10SPE/public/adhoc/brandt_101016_10SPE_01a_adhoc.pdf) If the intent is (i) above, then I don’t think this will achieve the desired effect. If some property of the channel changes sufficiently to cause a
drop in SNR and a retrain, then in general I expect the new link that will be established will ultimately have the same SNR as if the retrain had not been initiated. ie. the PHY should be continuously adapting its receiver at a fast enough rate to keep up with channel variation over time. I appreciate that this may not be the case with existing PHYs which may not have been designed with industrial environments in mind, but for a new standard they
should be in which case I would argue that the requirement to keep up with channel variations is already implicitly covered in the existing objective: 'Support 10 Mb/s operation in industrial environments' If (ii) is the intent then would it not simply be better to disable retraining when using PHYs in industrial applications ? (Equivalent to setting the retrain
duration to 0ms). Perhaps someone can clarify ? Thanks, Oisín. |