MCF52259 + 8041NL PHY fails at 100Mb

[accutime]

Hello

We are using MQX3.7 and Codewarrior along with the MCF52259 and the Micrel 8041NL PHY

The schematics from the tower board have be checked against our schematics and they are identical.

The only difference is that we have a POE power supply on our board.

Our board works 100% on 10Mbit over 100Metres of CAT5E cable but fails to work when the PHY is set to 100Mbit with any length of cable over 15 Metres.

The board is very tight for space and the POE components are very close to the Magnetics.

The noise on the 3.3V rail with the POE supply  is <50mV

The board functions 100% at 100Mb over 100 metres when the board is powered locally . ie 3.3V is applied and the POE is disabled

Has anyone had experience using POE with this Processor and Phy.

Thanks in advance

[TomE]

POE has a minimum supply voltage of 44V, so your POE circuitry must have some fairly serious magnetics in it.

I'd suspect magnetic coupling between the POE switching regulator and the Ethernet signals.

You may be getting direct coupling from the POE switching inductor/transformer and the Ethernet magnetics.

You might have a high current loop (track layout) in the POE that is coupling to the Ethernet magnetics., or possibly coupling into the tracks from the PHY chip to the ethernet plug.

The "15 metre limit" may either be due to the voltage drop on the cable changing the POE switching characteristics, or increasing the signal cable impedance.

Try a long cable (so you're getting data errors, but supply the POE from a bench supply and vary the voltage to see if it is voltage level problem. Different voltages may make the switcher run at different frequencies, some of which might interfere more than others.

Look for switching noise on the Ethernet cable with an oscilloscope.

Place magnetic shielding material (ferrite, aluminium) between the POE and other parts of the board and see if that makes the problem better or worse.

Good luck.

Tom

[accutime]

Hi - Thanks for the feedback

Note: The POE is running at  a fixed frequency set by a resistor  (tested 250KHz , 280KHz , 300KHz).

The PHY is approx 10mm from the POE switch and Inductor (could be too close ?)

I took the magnetics and RJ45 off board (Magjack - wiring length  100mm)  - 100Mb is fine for 50M but not for 100Metre

I then cut all tracks (TX+,TX-,RX+,Rx-) from the PHY and connect the MagJack directly to the PHY  (wiring  length 60mm but away from Magnetics ) - 100Mb is fine for 50M but not for 100Metre

I power the board from a 3.3V linear power supply (with no POE supply) . 100Mb is fine for 50M but not for 100Metre

i power the board from a  3.3V linear power supply (with POE supply supplying a load) . 100Mb fails a 50M !

Exactly how does the 100Mb function compared to the 10Mb. When 100Mb fails the Link LED fails continously (3 - 4Hz)

Thanks for the feedback

Paul H

[TomE]

So with the POE running and the on-board magnetics it fails at 15M.

Otherwise it fails at 50M in all cases.

Test other devices over the same 100M cable. Does a PC run at 100M over that cable?

> connect the MagJack directly to the PHY

How about all the grounding and little capacitors you're meant to have to the centre-taps of the magnetics?

Do you have the "Ethernet ground' connected to a Logic, Power or Shield ground, and is there any difference in these?

Get as many examples of the proper connection from the PHY to the magnetics as you can.

Do you have the correct turns-ratio transformers to match the PHY? Some are 1:1 while others have different ratios. This is important. Make sure any series and parallel resistors on the pairs are the correct values.

> Exactly how does the 100Mb function compared to the 10Mb.

It is almost exactly 10 times harder to get working over that distance! 100M has to use far higher frequencies than 10M so it is more sensitive to everything.

Have you put an oscilloscope on the "cable" (on the pins on the last connector to the Ethernet cable) and looked at the signals? Compare "good signals" (from a working PC) and "bad signals". Compare the noise, undershoots and overshoots and the voltage levels.

Find a hardware engineer who's good at this if you can't.

Tom

[accutime]

Hi Tom

The schematics ie the caps , resistors , transformer ratios etc etc are all correct - as per the Tower schematics

BUT ....

I went back to the MCF52259 Tower system and SER Board and that does NOT run 100Mb over the 100M CAT5E cable.

The Tower system does run 100Mb over 50M CAT5E cable

I have tested this with two PCs and a laptop

This is without POE circuitry and with no POE Power supply in line !

On the tower system the LINK Led shows the same 3Hz - 4Hz when a 100Meter cable is used

So the problem in inherent in the Freescale design

I will try and contact Freescale

Thanks for the help and any further questions

Paul H

[TomE]

> The schematics ie the caps , resistors , transformer ratios etc etc

> are all correct - as per the Tower schematics

> BUT ....

> I went back to the MCF52259 Tower system and SER Board and that

> does NOT run 100Mb over the 100M CAT5E cable.

Which means there may be a design, or possibly a manufacturing fault with that board.

You should check the specifications of all of the magnetics (on the serial board) against what the PHY manufacturer says to use in their data sheets and app notes.

If you find a problem this may get you working a lot faster than the alternatives.

I assume you're testing this with a POE-equipped Ethernet switch of some sort. Which pins are the power on? Is it 802.3af Standards A (power on data pairs) or B (power on spare pairs)? How are you picking the power off, as I don't think it can be done with the Pulse J1011F01PNL?

The Pulse J1011F01PNL has 75 ohm resistors ON ALL PAIRS to a common internal capacitor-coupled ground. I'd say it looks to be incompatible with Power Over Ethernet. In fact I'm surprised you didn't get a plume of smoke the first time you plugged it in. Look at the internal wiring of the J1011F01PNL here (or search for it if this link doesn't work):

http://productfinder.pulseeng.com/products/datasheets/J402.pdf;jsessionid=0EFFD341DA67B7FC9F7E7D22F1...

If they haven't been damaged already, I'd try the Tower and out board on a non-POE hub to see if there's any difference.

I think you need the Pulse J1006, except it dead-shorts pins 4, 5, 7 and 8 together and wouldn't work with 802.3af B, and there's no power pickup for 802.3af A either.

Tom

[accutime]

Hi Tom

Thanks for the feedback but I think you may have missed the point.

Forget the POE and just look at the Ethernet Spec that states that Ethernet will function at 100Mbit over 100M (330 ft)

Freescale have designed and manufactured the Tower Systems Evaluation boards and released the schematics (MCF52259 + 8041 PHY) which ourselves and other people have used

The Freescale solutiion does not function at 100Mbit and over 100 meter CAT5E cable

If the Freescale solution does not function correctly its a good bet that mine will not function correctly.

I tried the Kinetis K60 Tower System this afternoon and that had the same problem as the MCF52259

Anyone else having problems with long ethernet cables at 100Mbit

Kind Regards

[JimDon]

Just wondering, exactly what spec, under what conditions, calims that 100MB will run over 100M on cat5.

From WIki:

Each network segment can have a maximum distance of 100 metres (328 ft). In its typical configuration, 100BASE-TX uses one pair of twisted wires in each direction, providing 100 Mbit/s of throughput in each direction (full-duplex). See IEEE 802.3 for more details. 

[accutime]

Hi JimDon

I am sure that the 802.3 specification states that 10Mbits and 100Mbit  Ethernet  specifications have a maximum length / segment of 100 meters.

The PHY can be configured for 10Mbit Half or Full Duplex or 100Mbit Half or Full Duplex or auto negotiate

Surely the statement from Wiki just confirms that fact !

[JimDon]

Well, you kept saying "over 100M" which perhaps I mis-understood to mean "more than 100M", when you meant "over a length of 100M"

Plus at one point you stated:

The board functions 100% at 100Mb over 100 metres when the board is powered locally . ie 3.3V is applied and the POE is disabled.

Sorry.

[TomE]

Slightly strange request. Try REVERSING your 100 metre cable. Swap the ends over. That will swap the pairs being used for Receive and Transmit. The switches and PCs you're using might have "auto pair swap" and might be working with the opposite pairs.

That "3 to 4 Hz LED flashing" - does that involve the software, or is the PHY doing all that by itself?

Does the "LED Flashing" require the PHY to be initialised by software, or does it do it all by itself even when the CPU is held reset? If it does work when the board is reset, does it work any better? Holding the board reset could reduce any noise from the CPU that might be interfering with the PHY.

Can you read the PHY registers back? There might be some useful error indication in there, or some status bits that tell you what it doesn't like.

I've read up on Power Over Ethernet. If the device isn't properly configured for POE then the power source should remain turned off. You've tested this without the power supply you're using connected, presumably a "mid-span" unit.

Tom

[accutime]

I put a service request from Freescale and received the follwoing answer

I reviewed with the experts and they told me that the tower wasn’t characterized
with a 100 meters cable. Also that is related with the PHY, as I stated before.
You need to contact Micrel in order to get a recommendation for a PHY that can
support the 100 meters distance. It is definitely possible, the cost might
increate on the PHY side but it should be possible.

[JWW]

All:

The Freescale SR is correct.  On the Tower card we used best design principles, but on the TWR-SER card the focus was on cost so the ethernet PHY was selected for that as the first criteria as we had numerous customers requesting low cost implementations, who were NOT stressng the max side of the IEEE spec.  For customers that were interested in a more reliable e-Phy we put out the TWR-SER2 which demonstrates two port ethernet but more importantly it uses National's ephy which is "very" reliable and has excellent signal recovery at the end of 100M cable.  With their solutions it is possible to go even farther and the PHY includes a variety of additonal diagnostic tools built in.

Each PHY vendor has focused on different aspects of the industry. The Natinal phy can be 5-10x more expensive than Micrel style phys...etc...

For what it is worth... 100M is really difficult for almost all the phy vendors and selecting a good set of magnetics etc really is challenging.  I've looked at a variety of scope outputs at the end of a 100M cable and the signal looks almost flat.  That is why a 100Mbit phy uses a completely different encoding scheme as compared to 10Mbit mode.

I would enter an SR with some of the phy vendors and see if they have any quick guidance on how to extend the range of their designs.

Hope this helps.

-JWW

[accutime · Accepted solution]

Thank you for all for your help in this matter

Kind Regards

Paul H