TPS2378: TPS2378EVM-105 fails to negotiate 802.3at with 2 of 5 tested 802.3at switches - is there a fix?

Murray, We, at TI, take PoE interoperability issues very seriously and we will get to the bottom of this situation.  We are integral participants in the Ethernet Alliance PoE logo program (for more information on that, please refer to our video training here) and will figure this out.

As you may already know, PoE certification testing is done at a system level (not IC) so we need to get more information from you in order to move forward.  Would you please provide some additional details?

• When you say our eval kit fails, we which one did you purchase?  Was it the TPS2378EVM-105?
• Also, when you say it fails, you mean to say that it does not achieve Type 2 power levels with the TP-Link DGS-1008P 802.3at switch, correct?
• Have you run either system (yours or ours or the TP-Link switch) through SIFOs testing to prove out compliant operation?

It is highly unlikely that the issue resides in the IC...as we have shipped millions of these ICs without reported interoperability issues so we will need to look into this closely.  As such, I am assigning this thread to our PD Application Engineer and he will follow up once you provide more details on the questions asked.

Hi Thomas

We understand that certification is for a complete system, do you have a list of products that use the TPS2378 and are certified?    Is the D-Link ( not TP-Link) DGS-1008P a known bad actor?  The UBNT GP-C500-120G is another one.  

Sorry - we don't have the time / Budget  to test the D-Link DGS-1008P as you request.    I do ask anyone for a list of PoE switches that fail to get to T2P with the TPS2378.

We have been using the NCP1093 since 2016 without issues.    We switched to the TPS2378 and now have customer returns.

Recall that the TPS2378EVM-105 was purchased for overnight delivery after a day of experimenting with the TPS2378 on our boards.    We were extremely surprised when it could not get to T2P.

"Failure" in this case is described as:

A) Very slow to get to 802.3af
B) never got to T2P
C) As a result of failure to get to T2P the PoE switches shut down at power levels in excess of 802.3af.

A search on google for D-Link DGS-1008P problems with 802.3at did not show relevant results.

As a innovator of PoE - my company has shipped millions of PSE ports since 2011.

Best Regards
Murray

Correction - D-Link DGS-1008P

Murray,

I sent you a connect request and I can share some of our test reports.

If you post a schematic we can help review it. 

Can you explain further what you mean by, " never got to T2P"? 

Can you please show the VDD_VSS and RTN_VSS waveforms so we can see the detection / classification?

Also, in your original post, you said that the device "never completes phase 2." Can you explain this further? The PSE is responsible for sending two class fingers for Class 4 IEEE802.3.at handshake. 

If this post answers your question, please indicate so by marking this thread as resolved. Thank you.

Regards, 

Michael P.

Applications Engineer

Texas Instruments 

Hi Michael,

I think you have the schematic of the TPS2378EVM-105 already ?    At least your product should work - right?  We bought your eval board so we could communicate with you effectively.

The T2P led does not turn on.   Sorry if calling that "never got to T2P" is imprecise.

I agree - the PSE needs to send the AT handshake, and the TPS2378 needs to respond.  It doesn't respond in a way that works.

Do you have access to a D-Link DGS-1008P, or the UBNT device ?  They are on Amazon.   Setting up a scope in our lab to capture the current and voltage waveforms  to capture and debug this for you is our of our budget and time available.

Meanwhile - my questions are not answered

A) list of PoE switches that are incompatbile with the TPS2378??
B) List of devices that use the TPS2378 and passed certification??

thanks!

Murray

Murray, 

Please accept my connect request and I can send you documentation of the EVM  IEEE802.3.at certification. 

Please send VDD_VSS, VDD_RTN and ISS waveforms so we can see what is going on exactly. Since you have the PSE and the EVM together, this will lead to the fastest resolution time. 

A). There are hundreds of PSE's on the market, it is not feasible to test all of them. That is why we test compliance for the EVM, which we have documentation of. 

B.) We do not post our customer devices that use our parts. As Thomas said, IEEE802.3at compliance is a system level certification. It is up to our customers to acquire this certification. However, our EVM is compliant and I can send you the documentation if you accept my connect request. 

If this post answers your question, please indicate so by marking this thread as resolved. Thank you.

Regards, 

Michael P.

Applications Engineer

Texas Instruments 

HI Michael

Sorry my message was not clear.  We don't have the budget or time to set up to capture VDD_VSS, VDD_RTN and ISS waveforms for your chip and a popular PoE switch that we don't make.

We do have a NCP1093 device that we can use to replace the TPS2378.

thanks

Hello Murray,

We are interested in this project and are trying to learn more. 

One thing we know about the TPS2378EVM-107 is that it does not meet MPS on its own, and therefore needs some load. Would you be able to try adding some load (100ohm for example) to VDD and RTN?

Additionally, when you set up your tests, was J5 open, or closed?

If this post answers your question, please indicate so by marking this thread as resolved. Thank you.

Regards, 

Michael P.

Applications Engineer

Texas Instruments 

Hello Michael,

MPS is a function that occurs hundreds of ms after T2P.     So no load is required for T2P to occur - indeed, with CDB the load is OFF until long after T2P.

careful attention to this detail means that adding a load or a jumper has no relevance to the problem we have, however, for fun, we tried it.   No change.

A 100 ohm load is not needed - MPS in most PoE products is between 350 and 500 mW.   a 100 ohm load causes the PoE port to shut down since without T2P, the port is limited to 13 watts.

Best Regards
Murray Freeman
Ex-Applications Engineer ( Cypress, Benchmarq, ADI, Mostek)

Murray, 

Thanks for your kind reply and for trying our suggestion. 

During your tests, can you please comment on the jumper J5?

Thank you!

If this post answers your question, please indicate so by marking this thread as resolved. Thank you.

Regards, 

Michael P.

Applications Engineer

Texas Instruments 

Jumper 5 is in,

Please see the attached photos with J5 in place and a 100 ohm load, its not good.

Obviously J5 cannot help with T2P.  So it still fails to turn on the T2P led.   This is exactly as expected, since during T2P negotiation, the output load is shut off ( J5 is irrelevant ).

But the worse news is that the TPS2378 goes into a destruct mode with a 100 ohm load.  With J5 in place, the TPS2378 still enables the output to the 100 ohm load, but the switch is in 802.3af mode since T2P failed, so after a few hundred ms at 25 watts of load, the PoE switch shuts down.   When it powers up again, it power up in a current limited state - so the 100 ohm load stablizes at 5 volts, 50 mA, therefore 41 volts @ 50 mA or 2 watts is burning in the TPS2378 and that is really bad.

So I don't think asking a customer to put a 100 ohm load on a TPS2378 in 802.3af mode is a good debug strategy.   The load must not be greater than about 300 mA.    You might want to review the debugging steps...

So my original question remains - is there a list of 802.3at PSE  that do not work with the TPS2378.      The new info is that the CDB pin will stay active even when the TPS2378 is subjected to 2 watts of heat for a long time, so designers must use the T2P output to insure that no load greater than Class 3 be presented to the TPS2378 at any time

Murray

This is the TPS2378 with no load, an 802.3at switch and failure to get T2P

 Here the device has a 100 ohm load and still fails.

Hello Murray,

Thank you so much for all of your kind responses. I was asking about J5 because sometimes the TPS2378 fails the in-rush phase and needs the additional FET. The 100ohm load would certainly clear the MPS signal, which is a known issue in this EVM. I was wanting to check these two things first. 

Thank you again for trying our suggestion and your quick responses. We will continue to look into this. 

If this post answers your question, please indicate so by marking this thread as resolved. Thank you.

Regards, 

Michael P.

Applications Engineer

Texas Instruments 

Hello Murray,

Thank you again for trying a few of our tests to check our EVM. By adding load, you have proven that it is not MPS. 

As we have continued to look into this issue, we will be unable to go any further until we see a VDD_VSS and input current waveform. If you could please provide these waveforms, then we can continue to move this issue forward. Thanks so much!

If this post answers your question, please indicate so by marking this thread as resolved. Thank you.

Regards, 

Michael P.

Applications Engineer

Texas Instruments 

Hi

I guess we have reached an impasse.

We don't have the budget or time to investigate for TI the reason why it does not work with various PoE PSE devices,  like the D-Link 1008P or the UBNT. 

If TI wants to open purchase order for us to do this testing, we can help TI.    Note that current testing when the scope ground is at +48 volts requires a Class II isolated power supply - kids be careful if you try this at home - a Class 1 supply with minus grounded will demonstrate violently that there is no effective current limit on the +48v rail in any 802.3af/at PSE.

Fortunately, the NCP1093 works great since 2017, so we will revert to that.

Full disclosure would be a list of PSE that do not work with the TPS2378, so that users can make a informed decision about your part, and if failure to negotiate T2P is significant or not.  In this post, that hopefully google will index, we hope that at least 2 devices will be associated with the TPS2378 failure.

thanks!!

Murray

Murray,

I certainly understand your decision to move forward with the competitive solution given where you are in your design cycle. These interoperability debug issues are not fun for any of us in the industry and is precisely why TI has worked with other industry leaders (such as Cisco, HPE, ADI...) to create a program designed to more quickly resolve them and avoid them all together!  If you are not familiar with that program yet, I would like to suggest you review our FAQ on that topic here and then follow the links in that FAQ to learn even more.

With that said, and coming back to your design in particular, TI remains confident that if we could see a VDD_VSS and input current waveform this issue could be resolved.  The TPS2378 is an industry high runner, released in 2012, and this is the first interoperability issue that has been reported and not resolved.  Further, it is definitely not clear yet which side of the cable is really "at fault" here.  For sure, to say that any load utilizing the TPS2378 with these two switches would demonstrate the same interoperability issue is also not correct.  As you know, interoperability is not as simple as an IC-to-IC "test."  Rather, it is a "system to system" test.  The TPS2378 has multiple designs available on ti.com that pass SIFOs and UNH-IOL test suites.

Should you change your mind and become open to sharing waveforms, we would be happy to pick back up on this thread.  Otherwise, I will be closing it now.  Best of luck.