TPS23756: 802.3at compliance for a Class 4 device

Neil,

Sorry to hear you are having trouble and we will be happy to assist. I am going to assign this thread to PoE Strategic Application Engineer Darwin Fernandez. He is familiar with solutions on both sides of the cable and will be able to suggest some tests and waveform captures to help debug the situation on Monday. May I ask the end equipment the TPS23756 is designed into? Is it a product under development/testing or already released?

The answer to your direct question is, yes, the TPS23756 does support hardware and software (LLDP) negotiation. With that said, the fact that you suspect the failing behavior is only present when in LLDP mode is a significant clue.

Thank you for designing with PoE products from Texas Instruments and we will get back to Monday,

The end equipment is a released medical product that we have used with a number of HP and Cisco PoE+ switches. The Cisco Catalyst 2960-X Series is the first customer switch we've had an issue with. That is why we are trying to track down the exact cause.

I did try a test on Friday, and I turned off LLDP transmit/receive on our Cisco SG300-52MP switch. Our device continued to function just fine on that switch after un-plugging and re-plugging-in the device. So, LLDP or not, our device seems to negotiate correctly on the SG300 series switches.

Thanks for your help,

Neil

I am not sure if this answers your question, but I did turn off LLDP transmit/receive on our Cisco SG300; and, our device continued to boot and run as normal.

Hi Neil,

Are you able to try to disable it on the Catalyst switch as well to see if it'll force hardware classification? I took a look at the Catalyst datasheet and couldn't get much out of it relating to PoE.

Darwin,

Thank you for your response. I have tried disabling LLDP on the Catalyst 2960 switch as well - unfortunately, it did not make our product start working. In fact, when using Cisco's "show lldp neighbors" console command, I don't see our device at all (indicating that perhaps it doesn't negotiate using LLDP at all?).

This is the best documentation for PoE on the Catalyst 2960-X Series I have found:

https://www.cisco.com/c/en/us/td/docs/switches/lan/catalyst2960x/software/15-0_2_EX/int_hw_components/configuration_guide/b_int_152ex_2960-x_cg/b_int_152ex_2960-x_cg_chapter_0111.html

After further testing this week, some observations:

1. Per Cisco, the 2960-X Series switch should auto-negotiate PoE+ with our device. There is no special switch or setting that has to be configured. However, there is a max of (any) 12 ports that can be used for PoE+ - mostly due to power budget. The switch will not allow more than 12 ports for PoE+. We're well within that limit when testing at 1-2 devices on the entire switch.
2. The only way to "switch on PoE+ for a port" (the terminology in Cisco documentation we initially found) is to configure the port as "static", as opposed to "auto". This assigns/reserves up to 30000mW to that particular port. Cisco tech support said (and the documentation at the link above confirms) that this effectively turns off power negotiation and straight provides up to 30000mW to the port. From the docs: "The switch allocates the port configured maximum wattage, and the amount is never adjusted through the IEEE class or by CDP messages from the powered device." Despite setting the "static" configuration of 30000mW on our test port, our device based on the TI chipset would not boot, even after un-plugging and re-plugging it in.
3. We have two PoE+ devices, and they use two different Ethernet PoE chipsets. The Analog Devices PoE chipset in one of the devices appears to follow the Cisco "rules" that you power up on 15400mW and then negotiate for more power later. The TI PoE chipset doesn't appear to follow this rule, despite the switch showing that 15400mW is being provided to the port, by default. It would appear that the switch knows that PoE is required, or it would disable all power on the port. However, even under the "static" setting, our TI-based product will not boot on the 2960-X Series switches. From the Cisco docs: "The initial allocation for Class 0, Class 3, and Class 4 powered devices is 15.4 W. When a device starts up and uses CDP or LLDP to send a request for more than 15.4 W, it can be allocated up to the maximum of 30 W." I am at a loss here, as it sounds like maybe our TI-based device is not powered-up enough to negotiate the second part of that requirement?
4. I am not sure whether our LLDP negotiation on the TI-based device includes the requested/required TLV information in the messages: From the Cisco docs: "With PoE+, powered devices use IEEE 802.3at and LLDP power with media dependent interface (MDI) type, length, and value descriptions (TLVs), Power-via-MDI TLVs, for negotiating power up to 30 W. Cisco pre-standard devices and Cisco IEEE powered devices can use CDP or the IEEE 802.3at power-via-MDI power negotiation mechanism to request power levels up to 30 W." Does the TPS23756 provide these correct TLV attributes in its negotiation? We do have the proper resistor set for a Class 4 device on the TPS23756.
5. The TI-based device does not even power up enough to get the MAC address of the device on the Cisco switch. It is obvious from the switch logs that a static 30000mW can and is being provided to the port, and no power negotiation is expected or required.
6. The Cisco SG300 switch has a maximum of 31500mW power per port, whereas the 2960-X Series switch only has a maximum of 30000mW power per port. It doesn't seem likely that we're anywhere near that limit on startup. In fact, from looking at the power logs on the Cisco switch, our peak draw is at about 5200mW. This is just from logs, so there may be a spike in there that I am not seeing. Still, it is a notable difference between two Cisco switches - one that works (SG300), and one that does not (2960-X).

At this point, Cisco cannot determine whether there's an incompatibility with the switch. The tech support person suspects incompatibility with the TI chipset, but there is no way to confirm without our TI-based device even beginning to boot.

I want to stress that this device with the TI TPS23756 works as expected on other Cisco 802.3at switches and other manufacturers' 802.3at switches. The 2960-X Series just seems to be incompatible for some unknown reason.

Thanks for any help,

Neil

Hi Neil,

Are you able to take a waveform of VDD-VSS when you plug in the PSE? If there is interoperability issues between either the PD or PSE, we might be able to narrow it down by looking at this waveform.

What the dialogue is saying above is that per the IEEE802.3at spec, before going to 25W, the PD's load must wait ~80ms before going to higher power 25W so that the PSE can finish charging the PD's bulk cap.Does the PD's load output 25W right away after startup? Or is there some delay?

Unfortunately, I do not currently have a way to take a waveform at this time. As far as I understand, we should not be spiking that high (25W), immediately. From logging, I don't see our draw go above 5.2W during startup; but, as I said, I could have missed a spike by just observing the logs.

If I could get the TPS23756-based device to come up enough to register its MAC address with the switch, I could do more power debugging on the switch side - the Catalyst switches have a pretty good set of ILPOWER debug tools available. Unfortunately, the SG300 series does not have the same debug capability.

This is the part that confuses me:  the TPS23756 chipset just seems to refuse to come up. The switch sees it as a PoE-powered port and can/does supply power. The TPS23756 just never responds. Our board's diagnostic LEDs are very dim, indicating that not enough power is available to boot.

Hi Neil,

Can you attach your 756 schematic so I can do a quick review? Thanks!

Darwin,

Thank you for your offer of help. Is there some place that I can share the schematic, without it being posted publicly?

Thanks,

Neil

Hi Neil,

Yes I have sent a friend request and you can send it to me on private message offline. Thanks!

Regards,

Darwin