HDMI is always a pain to debug, so I am looking into measuring cable performance somehow.
-
@cccpresser hm, this would result in a single-ended measurement, wouldn't it?
probably more relevant to somehow test the differential pair itself (sounds like a usecase for a balun, but at those speeds... hmm)@manawyrm The grounds on those SMAs are not connected to each other. So one diffpair end up as signal + shield/return on one SMA connector.
The idea is that the VNA effectively measures the whole diff-pair this way. Not sure if that is clever or stupid.
I just figured that I also will need some kind of 50R HDMI-Plug for calibration... hmmm...
The alternative plan is to build a more complicated device with a chip that outputs a Eye-Open-Diagram. But I want something simple now.
-
@manawyrm The grounds on those SMAs are not connected to each other. So one diffpair end up as signal + shield/return on one SMA connector.
The idea is that the VNA effectively measures the whole diff-pair this way. Not sure if that is clever or stupid.
I just figured that I also will need some kind of 50R HDMI-Plug for calibration... hmmm...
The alternative plan is to build a more complicated device with a chip that outputs a Eye-Open-Diagram. But I want something simple now.
@cccpresser ah, hm, that would require a 4port VNA to get proper measurements, then, right?
-
@cccpresser ah, hm, that would require a 4port VNA to get proper measurements, then, right?
@manawyrm Only if I am interested in cross-talk between diff-pairs.
With a 'normal' dual port VNA I could connect each vna-port to one SMA on each end of the HDMI cable.
The circuit is completed with the shield/return.
The other SMA connectors most likely need termination to not influence the measurement. -
@manawyrm Only if I am interested in cross-talk between diff-pairs.
With a 'normal' dual port VNA I could connect each vna-port to one SMA on each end of the HDMI cable.
The circuit is completed with the shield/return.
The other SMA connectors most likely need termination to not influence the measurement.@cccpresser I'm kinda expecting this to be an important piece of the measurement, though. At least with CAT7/10GbE copper cabling, NEXT/FEXT crosstalk between the pairs is usually what fails to meet specs, not the individual line itself.
But yeah, for individual measurements, that should work, i guess
-
@cccpresser I'm kinda expecting this to be an important piece of the measurement, though. At least with CAT7/10GbE copper cabling, NEXT/FEXT crosstalk between the pairs is usually what fails to meet specs, not the individual line itself.
But yeah, for individual measurements, that should work, i guess
@manawyrm Thanks, that is exactly the insight I am looking for.
Crosstalk would kinda be possible to measure with this as well, terminating one pair, and measuring another diffpair.
Thinking more about that, it would probably take hours to make a decent measurement of one cable using the tools I currently have available.
So perhaps I need to make a board with the https://www.ti.com/lit/ds/symlink/ds280df810.pdf or a similar device that has a build in eye-monitor.
-
@manawyrm Thanks, that is exactly the insight I am looking for.
Crosstalk would kinda be possible to measure with this as well, terminating one pair, and measuring another diffpair.
Thinking more about that, it would probably take hours to make a decent measurement of one cable using the tools I currently have available.
So perhaps I need to make a board with the https://www.ti.com/lit/ds/symlink/ds280df810.pdf or a similar device that has a build in eye-monitor.
@cccpresser and having an open source design for such a thing would also enable you to build testers for many other cables like USB-C, SFP28, etc., which seems very useful to me? I think that's worthwhile -- such testers are usually unobtainably expensive.
-
@cccpresser and having an open source design for such a thing would also enable you to build testers for many other cables like USB-C, SFP28, etc., which seems very useful to me? I think that's worthwhile -- such testers are usually unobtainably expensive.
@manawyrm Yeah, that has been on my todo-list for a few years already. Just not with any priority - also because it seems to be non-trivial. I don't have the equipment to do validation measurements either.
TI has a few chips that can do those EyeOpenMonitor plots. At least for HDMI there are cheaper options with easier packages.
This could also be a 'real' product, I can easily imagine a company spending 1k€ on a device that lets the user judge the quality of HDMI wiring.
-
@manawyrm Thanks, that is exactly the insight I am looking for.
Crosstalk would kinda be possible to measure with this as well, terminating one pair, and measuring another diffpair.
Thinking more about that, it would probably take hours to make a decent measurement of one cable using the tools I currently have available.
So perhaps I need to make a board with the https://www.ti.com/lit/ds/symlink/ds280df810.pdf or a similar device that has a build in eye-monitor.
@cccpresser @manawyrm I designed a board around the "big brother" of this chip: https://git.paranoidlabs.org/jackie/sfp28-retimer-mux
I haven't finished bringup yet but the chip is answering my I2C requests (I'm currently procrastinating implementing the absolutely cursed configuration protocol/format).
I specifically designed this to measure some cursed SFP28/56 contraptions. -
HDMI is always a pain to debug, so I am looking into measuring cable performance somehow.
Has anyone build a HDMI to Coax adapter board that I can use together with a VNA? I am thinking of having two of them, one on each side of the HDMI-Cable.
Is this a stupid idea anyway? What are your thoughts?
@cccpresser We had build something like this for Micro-SD-Cards some time ago. We have measured S11 and S12 for some of the signals for our products and also some commercially available stuff. Worked well, with good repeatability.
But SD-Cards are (keeping the modern PCIe-ones aside) single ended.
-
HDMI is always a pain to debug, so I am looking into measuring cable performance somehow.
Has anyone build a HDMI to Coax adapter board that I can use together with a VNA? I am thinking of having two of them, one on each side of the HDMI-Cable.
Is this a stupid idea anyway? What are your thoughts?
@cccpresser That’s definitely an off-the-shelf product. Looking for “HDMI fixture”, it seems leo bodnar sells one for a reasonable price.
-
@cccpresser That’s definitely an off-the-shelf product. Looking for “HDMI fixture”, it seems leo bodnar sells one for a reasonable price.
@jaseg @cccpresser yes but the usual way to do it is one sma per conductor, not per pair, for a 4-port VNA / differential BERT.
The pairs are not designed for 50 ohm zdiff so doing it your way will likely result in big reflections at the connector
-
@jaseg @cccpresser yes but the usual way to do it is one sma per conductor, not per pair, for a 4-port VNA / differential BERT.
The pairs are not designed for 50 ohm zdiff so doing it your way will likely result in big reflections at the connector
@azonenberg @cccpresser yeah, you definitely want to do a four port measurement. The only way around that if you only had a 2 port VNA would be to bodge something with some baluns, but good ones covering that frequency range can be very pricey and you’d still have to cal them out of your measurements.
-
@azonenberg @cccpresser yeah, you definitely want to do a four port measurement. The only way around that if you only had a 2 port VNA would be to bodge something with some baluns, but good ones covering that frequency range can be very pricey and you’d still have to cal them out of your measurements.
@jaseg @cccpresser yeah.
Realistically, for this kind of testing I would want to use something like my ML4039 BERT (or an open source equivalent made by a less awful company) with multiple differential TX/RX lanes
-
@jaseg @cccpresser yeah.
Realistically, for this kind of testing I would want to use something like my ML4039 BERT (or an open source equivalent made by a less awful company) with multiple differential TX/RX lanes
@jaseg @cccpresser building one around an artix u+ FPGA or similar has been on my TODO for a while, it would work with any cable standard given appropriate fixturing
-
@manawyrm Thanks, that is exactly the insight I am looking for.
Crosstalk would kinda be possible to measure with this as well, terminating one pair, and measuring another diffpair.
Thinking more about that, it would probably take hours to make a decent measurement of one cable using the tools I currently have available.
So perhaps I need to make a board with the https://www.ti.com/lit/ds/symlink/ds280df810.pdf or a similar device that has a build in eye-monitor.
@cccpresser @manawyrm AMD/Xilinx IIRC has a built in eye scan feature on their high speed transceivers. Most off the shelf FPGA boards with HDMI inputs or outputs have redrivers or retimers in front of the FPGA though.
-
@jaseg @cccpresser building one around an artix u+ FPGA or similar has been on my TODO for a while, it would work with any cable standard given appropriate fixturing
@azonenberg @jaseg I have never used a BERT. For my common use-case of measuring (or estimating) the signal-integrity/margin of cables, what would be the better approach?
BERT or EOM?What I want is a measurement that tells me that the how good a given signal chain (cable, perhaps switches, retimers) is. So I can exchange components, measure again and remove/replace bad components.
-
@cccpresser @manawyrm AMD/Xilinx IIRC has a built in eye scan feature on their high speed transceivers. Most off the shelf FPGA boards with HDMI inputs or outputs have redrivers or retimers in front of the FPGA though.
@jaseg @manawyrm The Appnote from AMD mentions the KC705 which is available at work...
I have not found a HDMI-FMC card without drivers, so I guess that I need to make one. FPGA + FMC would be kinda cool since it allows different connectors like USB, SFP, ... with a rather simple and cheap adaptor.
I guess the main issue is finding the time and motivation to build any kind of device.
-
@azonenberg @jaseg I have never used a BERT. For my common use-case of measuring (or estimating) the signal-integrity/margin of cables, what would be the better approach?
BERT or EOM?What I want is a measurement that tells me that the how good a given signal chain (cable, perhaps switches, retimers) is. So I can exchange components, measure again and remove/replace bad components.
@cccpresser @jaseg The BERT can do eye opening measurements, the ML4039 is actually built around a 100GbE retimer chip very similar to that TI part you linked above.
I have a YT vid from a while back demonstrating the ML4039 used with a very early version of ngscopeclient to give you an idea.
-
R relay@relay.infosec.exchange shared this topic
