Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: [802.3EEESG] 10BASE-T question



Joseph-

At 07:54 PM 3/28/2007 , Joseph Chou wrote:
Hi Pat,

Here is my two pennies.

I am guessing that the reason to lower down the output voltage of 10BaseT is to reduce its power consumption so that higher speed phy can be downgrade to 10BaseT.

That would be the reason

However, even though we modify the standard to allow lower output voltage for 10BaseT, we probably will end up a 10BaseT phy which has comparable power consumption of 100BaseT.

No, the idea would be that with revised specs AND new designs based on contemporary supply voltages the power comsumption would be (a) lower than 100BASE-Tx and (b) the power consumption of the 10 Meg would be significantly lower when it was in IDL that when it was transmitting data.

 It will lose the advantage of speed change. The benefit of changing the spec could turn out to have a new lower power 10BaseT when it drives longest CAT 3 cable thus only 10Mbps can be negotiated successfully.

No. Pat's proposal was to drop Cat3 compatibility and design the new one around Cat5 cable. Cable that was worse than Cat3 (AT&T DIW) was the design point for 10BASE-T. Cat5 is significantly better than Cat3 or DIW in every way and there is very little true Cat3 left these days. Nobody has installed in new installs for years. In particular, 10BASE-T has enough drive to drive about 180 meters of Cat5 cable. If we made no other change than to cut the drive level back to that required for 100 meters we should be able to save quite a bit of power. They are other tricks we could do for additional power saving once we have the design open. The rules would be that it has to be backward compatible with existing 10BASE-T over Cat5 at up to 100 meters.


By the same token, so far no one considered adding a Power Back Off mode on 1000BaseT and 100BaseT for shorter cable length because the saving of power may be very marginal.

Again, this would be a 2 way win:
        - Lower drive power because of the significantly lower attenuation
        - The very low duty cycle and power of the 10 Meg IDL when there is no traffic
And by signalling between packets at a lower frequency we will be operating in a lower attenuation region of the spectrum which should produce lower power requirements.

I am afraid that the incentive of changing 10BaseT spec is not as great as devising an electrical idle mode so that all phy modes (10, 100, 1000, 10G) can be switched to it.

This has sufficient promise to be worth investigation.
It is however, non-trivial.
The requirements, as I see them, would be:
        - Maintain link integrity state information
        - Not interfere with PoE probe pulse
        - Support the code transfer requirements of Auto-Negotiation
        - Meet the bandwidth requirements for keeping the DSP parameters current in the higher speed PHYs
        - Probably needs to have some sort of baud width compatibility with current PHYs
        - Be enough lower power to be worth doing
Additional requirements that there would be a strong push to add some other "features":
        - Rapid link integrity response time, sufficiently improved to support fast switch-over to a redundant link.
        - Reliable and speed consistent mechanism for far-end fault detection
        - Combining the 2 items above


By the way, in order to meet the template of Fig 14-9 the transmitter normally needs to pre-emphasize the waveform for fat bit (20ns or 2.5MHz carrier). I don't have simulation result at hand so I am not sure for the same amount of pre-emphasis (preset in IC design) used for CAT 3 model test it still fits the template when we use CAT 5 cable model. We may need to consider the attenuation difference between 2.5MHz and 5MHz for both cable models.

We certainly would.
ISO/IEC 11801- 1995 gives the following values for cable attenuation (dB/100m):
Freq    Cat3    Cat5
 1 Mz    2.6     2.1
 4 Mz    5.6     4.3
10 Mz    9.8     6.6

Best regards,

        Geoff Thompson


Best Regards,

-Joseph Chou

-----Original Message-----
From: Pat Thaler [mailto:pthaler@BROADCOM.COM]
Sent: Wednesday, March 28, 2007 6:37 PM
To: STDS-802-3-EEE@listserv.ieee.org
Subject: Re: [802.3EEESG] 10BASE-T question

Mike,

I think that some adjustment to the 10BASE-T transmit voltage would be entirely appropriate.

The 10BASE-T output voltage spec (IEEE 802.3-2005 14.3.1.2.1) currently requires that the driver produce a peak differential voltage of 2.2 to 2.8 V into a 100 Ohm resistive load - a very normal output voltage when the standard was written in the late 80's, but pretty high nearly 20 years later. This voltage allowed 10BASE-T to coexist in bundled Cat 3 cable with analog phone ringers. The transient when an analog phone ringer goes off-line in that situation could produce over 250 mV.

That high output voltage is not necessary over Cat 5 or better cable.

The simple change would be to add a differential output voltage spec for operation over Cat 5 or better cable. In that case, remove the minimum voltage spec for peak differential voltage into a 100 Ohm resistive load. One still would keep the maximum voltage spec of 2.8 V or perhaps substitute a lower maximum. Change the requirement for the Figure 14-9 output voltage template to be the signal produced at the end of a worst-case Cat 5 cable instead of at the end of the (Cat 3) twisted-pair model.

This should be fully backwards compatible with existing 10BASE-T compliant PHYs over Cat 5 cable. The newly specified transmitters will produce a signal over Cat 5 cable that is within the range of signal that the original 10BASE-T produces over the Cat 3 cable channel it specified. That template provides a minimum eye opening of 550 mV. If I plugged the numbers into my calculator correctly, the attenuation difference between Cat 5 and Cat 3 cable at 10 MHz is more than 4 dB so this should allow the transmit voltage to drop by that. It should be very little work to do this change.

A more aggressive change that would require real work would be to determine what receive voltage could be tolerated by today's receivers which probably can tolerate a smaller eye-opening especially if they are a 1000BASE-T receiver operating in a slowed down mode. But in that case, one would either need to only use the lower eye-opening when stepped down by EEE or add negotiation for low voltage 10BASE-T to auto-neg because it wouldn't ensure backwards compatiblity with classic 10BASE-T receivers.

I think the fully-backwards compatible change would be pretty easy to justify. To summarize, for operation over the channels specified by 100BASE-TX, 1000BASE-T and 10GBASE-T, delete the spec for minimum voltage into a 100 Ohm load and change the test condition for the Figure 14-9 voltage template to be over a worst case 100BASE-TX channel.

Regards,
Pat


At 01:46 PM 3/28/2007 , Mike Bennett wrote:
>Folks,
>
>For those of you who were able to attend the March meeting, you may
>recall we had a discussion on 10BASE-T (in the context of having a low
>energy state mode) and what we might change to specify this, which
>included possibly changing the output voltage.  Concern was raised that

>the work required to specify a new output voltage for 10BASE-T would
>far outweigh the benefit.  Additionally, there was a question regarding

>the use of 100BASE-TX instead of doing anything with 10BASE-T.  Would
>someone please explain just how much work it would be to change
>10BASE-T and what the benefit would be compared to using 10BASE-T with
>the originally specified voltage or 100BASE-TX for a low energy (aka
"0BASE-T" or "sleep") state?
>
>Thanks,
>
>Mike