Capacitive Coupling

Working on a design that utilizes a W5500 which is capacitively coupled (no magnetics) to an on-board switch (KSZ8995XA). Micrel/Microchip has an App Note that explains how to terminate the RX/TX pairs on the switch side of the dc-blocking caps. Trying to figure out the appropriate termination on the W5500 side.

This concept has been brought up in two other posts that I could find (linked below). In both posts it was stated that Wiznet has not tested a transformerless PHY to PHY connection. Assuming that is still true and there is not a known good configuration, I am hoping to get a little more information about the W5500 PHY’s transmit drivers. Are they current drive of voltage drive transmitters? Or current drive for 100Base-TX and voltage drive for 10Base-T? If current drive, what is the current strength? Do the receiver and transmitter have internal bias or all external bias?

Link to Micrel app note:

Links to previous forum topics on the subject:


I can not speak for WIZnet as a company, but in my opinion the interest in manufacturer testing and solution certification depends on the commercial viability. I guess if they did not certify this solution yet simply means they are not convinced there’s a market for it. And this interest of course depends on the volume of the devices they can sell for this transformer-less solutions. If you want and need this solution to be formally certified, you should go to WIZnet’s sales department rather than to its support forums.

If you look at the W5500 reference schematics, you will find that its TX connects to AVDD without DC coupling, thus it means that there’s no internal DC bias for TX lines. Looking to the RX lines, in particular to circuit diagram at the bottom of the web page, you will see that they are connected to ground through capacitor, thus it means that RX lines are DC biased.

I suspect this information will not be available publicly. The Microchip’s guide you refer to has relatively good explanation of options you have, and the modifiers are just (1) DC biased/unbiased and (2) resistor ratios. For first you have an answer in W5500 reference schematic, and second you have well explained way to test yourself and see related voltages at both devices’ pin pairs.

Thanks Eugeny … I appreciate the thoughtful response. I agree that reaching out to a support forum is not as ideal as having direct technical contact at Wiznet (or a manufactures rep that has direct contacts or a major disti like Arrow or Avnet with dedicated FAE resources). But it definitely seems to be a sign of the times that more and more technical support is being pushed to forums as a first line of defense. I am not looking for a formally certified solution from Wiznet … just hoping to get some guidance on the W5500 transceivers so that I can figure out what is preventing my design from functioning.

On the technical front, your intuition on the RX lines being internally biased and the TX lines being external biased matches my read of the W5500 reference schematics. So the default component stuffing options of my prototypes follow that line of thinking … On the Ethernet switch side of the DC-blocking caps, I followed the Micrel app note for termination/biasing and on the W5500 side of the DC-blocking caps I followed the mag-jack version of the W5500 reference design (minus center tap connections). Relevant circuits from my schematic attached.

Thus far though, I have not been able to establish a link between the switch and the W5500. In initial debugging, I am seeing a few things that make my wonder if I guessed correctly on the W5500’s biasing needs. At 100M speeds, the switch is showing signs that it is receiving data from the W5500 (pulsing its link/activity indicator), but the W5500 is not attempting to assert its link indictor. In looking at the receive waveform at the W5500, there is very good correlation to a working Arduino Ethernet Shield 2. Nearly identical common mode voltages (~2V) and voltage swings. Attached are scope shots of the W5500 RX pins for both my board and the Enet Shield for comparison.

On the TX side, when looking at the waveforms from a differential perspective, there is good correlation between my board and the Arduino Ethernet 2 shield. But when looking at the single ended components of the diff pair, I am seeing some differences between my board and the shield that I can’t explain. The DC offset voltage of my board is in the neighborhood of 1.2V where it is closer to 2.7V on Arduino. So not sure if that means the current drive of the W5500 is much higher than I expected (40mA range), it is not externally biased like I thought, or the driver is not purely CML. The other thing (from a single ended perspective) is the MLT3 zero is not centered between the MTL3 1 and -1 … it is closer to the -1 side. So again not sure if there are different source vs sink currents or if the driver technology is different than I had assumed.

At 10M speeds, the device that seems happy flips. The W5500 will assert its link and activity indicators but the switch doesn’t.

W5500_RX_Comparison.pdf (99.7 KB)

I can not make similar tests as you did. Ensure you do not have R140 installed as W5500’s RX is already DC biased. Looking to scope pics I do not see in much differences in min/max/peal-to-peak voltage levels. Can you point where differences are?

I do not understand. Does it work properly at 10M?

Another question - does W5500 itself (its digital core, not only PHY) functioning properly all the time? I see you used SMD oscillator (you do not give part number for it) and recall there may have been issues with some of such oscillators.

Edit: none sources state this, but consider using C0G/NP0 caps as decouplers between networking devices instead of X7R.

Hi tims.

By default, the W5500 must be LINK UP before the TX signal is output.

It is judged that LINK UP does not come from the current situation.

Did you try measuring the LINK PULSE signal that the W5500 output?

Generally, PHY to PHY is designed to be designed like Micrel’s data.

Many customers using the W5500 know that Micrel’s circuit uses PHY to PHY.

But we can not guarantee. It is because we have not been verified.

And Does Digital work? For example, SPI communications.

Thank you.


1 Like

Hi Scott and Eugeny,

Thank you both for the additional feedback. I believe I found the root cause of my link issue today. The proper External Reference Resistor was not placed on the board. Not sure if the CM received miss-marked parts from the distributor, or the line operator loaded the wrong reel of parts, or the pick and place machine was programmed incorrectly, but those things happen occasionally. With the correct 12.4k resistor in place, basic Ethernet functionality is now working. Still have more detailed analysis to do on the link, but confirmed basic DHCP/ping/telnet server operation at 100Mb and 10Mb along with both full and half duplex. Fingers crossed I am over the hump now.

Circling back to the points you both raised for completeness:

  • The digital side of the W5500 worked solidly all along. Reading the chip revision register, PHY Configuration Register, etc was always reliable.
  • The crystal I used is Abracon PN abm8g-25.000mhz-18-d2y which has very similar specs to the Fronter Electronics FTX25.000M12SM3S used on the Arduino Ethernet Shield 2. The notable difference is the 18pF vs 12pF load capacitance, hence the difference in tuning cap values between my board and the shield.
  • Correct, R140 is not installed (by default). Those pads were placed in the design as a contingency in case the W5500 RX wasn’t internally biased like I had inferred from its reference schematics.

Thanks again!


I’m trying to use a w5500 to interface with a GigaBlox Nano - Micro GigaBit Switch 4 Port Network Switch

from what i can see the Gigblox Nano is push pull and has the needed isolation caps built in. however it is push pull is transformerless operation in this case still possible ? using this topology

@Taylor_Duchinski did you have any success getting this working? I have run into issues attempting capacitive coupling with exactly that switch and the W5500, so far with no success!

1 Like

@andrewteasdale hi, i was still in the planning phase when i pushed out that comment. As of now i have the board built up to this schematic and am reading a response from the w5500 that would indicate non duplex communication.

i believe that the BotBlox does not support this

looking again i may be biasing the wrong channel ? I’m going to give switch the biasing a try than post another update.

Thanks so much for replying and giving an update on this - I did not get quite as far as you did. I put together a test design and could not get the connection from the W5500 to Gigablox Nano to work. Nor did I manage to identify why. My bias circuit looks like this - although I believe C18/20/35/36 to be redundant as they are present on the GigaBlox Nano, as you have also identified.

My connection to the GigaBlox Nano is as follows:

I ended up connecting the W5500 to the GigaBlox Nano via 2 x RJ45 magjacks and a short RJ45 patch cable [not shown in the photo]. For various reasons, this made sense on this design with the time constraints we had but suffices to say I am keen to understand why it did not work as intended.

I would guess that the biasing on your design is indeed on the wrong channel and should be on the TxP/TxN pair. This assumption is based on my working W5500 to RJ45 bias circuit:

Perhaps interestingly, I have found that plugging a W5500 into a GigaBlox switch results in a constant red port LED - except when there is activity. My reading of this is that it indicates a port fault, despite the fact everything seems to be working. This is true both of my design and the W5500-EVB-Pico board.

If you learn anything more about whether this is possible I’d really appreciate your input.