Testing 75 ohm signals with 50 ohm gear

[rx8pilot]

I had started a thread about how to terminate a 75Ohm transmission line into my scope and maintain a reasonable response out to at least 6Ghz. My first look was matching pads but they are not a common find at that bandwidth.
https://www.eevblog.com/forum/testgear/75ohm-to-50ohm-at-high-frequenies-6ghz-how/msg1166793/#msg1166793
I ended up solving the immediate challenge by purchasing a 6Ghz active probe. That allowed me to examine signal structure and timing on the serial signals of interest - it is not modulated RF. This is a good start for some of what needs to be measured, but certainly not the end.

This led me to the another question (getting further and further from my level of knowledge) - in TV production, the system is dominated by 75 ohm transmission lines as far as signal distribution is concerned. We now have serial data at 12Gbps on a coaxial 75ohm cable terminated to BNC. Clearly, this puts quite a demand on the cable and the BNC connectors, especially when any distance is needed. Curious how manufacturers are measuring return loss and insertion loss on cables, connectors, receivers, etc out to 18Ghz. Keep in mind - I am just dipping my toes into high-speed signals and have never actually measured return-loss on anything. What little I do know has come entirely from books and the internet - nothing practical.

Is a typical VNA going to work when testing 75 ohm return loss (something like an Agilent 8720ES)? Are there 75ohm cal kits for 20Ghz? At these frequencies, my guess is that it is not at all trivial. What equipment, fixtures, and most importantly - skills are needed to test these cables, connectors, and various devices they are connected to for response and return loss? I have only book knowledge of VNA's, therefore only guessing this is even the right tool for the measurement task.

As I said, this is new territory  - so speak slowly 

[rfeecs]

Here's a 75 ohm cal kit to 12GHz:
https://www.maurymw.com/Precision/BNC75_Cal_Kits.php

An alternative, if you have say SMA connectors on your device, you can calibrate at 50 ohms and then convert mathematically to 75 ohms.  After all, a return loss measurement is just an impedance measurement.

[Someone]

I had started a thread about how to terminate a 75Ohm transmission line....
As I said, this is new territory  - so speak slowly 

The answer is in that older thread already. You can build an "ideal" test fixture with perfect transmission lines at 75ohm and then de-embed the cables and fixture when you measure the DUT. Its easier when the VNA automates this for you but its still possible to do manually if you can't afford the modern tools, and design of the test fixtures allows you to balance the important characteristics for your application.

[rx8pilot]

The answer is in that older thread already.

Well....I....am.....clearly....slooooow. ;-)

You can build an "ideal" test fixture with perfect transmission lines at 75ohm and then de-embed the cables and fixture when you measure the DUT. Its easier when the VNA automates this for you but its still possible to do manually if you can't afford the modern tools, and design of the test fixtures allows you to balance the important characteristics for your application.

I guess with the VNA and the (presumably) fresh new skills and knowledge of the instrument and the related procedures - rolling my own test fixture would be possible. Part of what I am hoping to accomplish is what I need to budget for in terms of equipment purchases and the education I need to dive into. Since I am currently in an overload condition trying to learn many new things - is a DIY approach going to be worth the time? While I am not on a specific schedule or deadline, the faster the better in general.

Is it totally naive to think that measuring return loss and insertion losses (initial need) will only require a modest learning curve? Will doing this at 18+Ghz be an uphill battle or a relatively simple affair? In terms of de-embedding various fixtures - are the newer VNA's along with the much fancier software 2x faster? 20x faster? Or are the older models just not going to practically allow de-embedding a complex fixture? I have heard and read what seemed like very scary long and challenging efforts needed for de-embedding - not sure if those stories have merit or if the authors were graduates of the volt-nut academy looking for over-the-top accuracy. It may seem like a bargain to get an old model until I realize that I have to beat the information out of it. It would also be very nice to have a lot of room to grow if it is not too ridiculous a price jump.

I am prepared for a steep learning curve, but don't know what to expect and if it is not obvious yet - don't know what questions to ask to best stay out of trouble. What do I really need here in terms of knowledge and gear? In addition to my own designs, I will be measuring a lot of cables, connectors, and receivers to best understand where they will lose control of the signals. Any suggestions for books, white papers, etc on the topic of high bit rate signal transmission testing would be good.

[G0HZU]

I have heard and read what seemed like very scary long and challenging efforts needed for de-embedding - not sure if those stories have merit or if the authors were graduates of the volt-nut academy looking for over-the-top accuracy.

I assume you are referring to me from your other thread here.

Here's a passive x10 probe (-20dB) I made using components selected after careful de-embedding to evaluate which chip resistors will give the flattest response in the probe up at many GHz. This probe will have minimal impact on VSWR when probing low impedance transmission lines.

This probe cost me some time when evaluating the chip resistors on a VNA but cost me nothing in parts because I made it from salvaged parts. But the response to 8.5GHz looks pretty good to me...

BTW, it is a bit offensive to be associated with a typical volt nut. Up at many GHz most measurements are a battle against various contributors to overall measurement uncertainty and it's often difficult to measure stuff within 20%. Pay poor attention to detail and you can easily end up with huge/ridiculous errors. A casual choice of SMD resistors could easily give a 10dB slope in the probe response up towards 8.5GHz. I had to measure/model the parasitics of every resistor package to get the response this flat and choose resistor values that behaved well over this huge bandwidth.

[G0HZU]

What do I really need here in terms of knowledge and gear?

If you are running a business, my advice would be to buy a decent RF simulation suite and learn how to use it. It won't be cheap but this is the best tool to have in any RF workplace. There are freebie alternatives like RFSIM99 but this can only take you so far.

You can practise de-embedding and mess about with virtual VNAs and various circuits on the simulator. Use downloaded s1p, s2p, s4p models of real components from Minicircuits to play with on the simulator. eg accurate models of filters, amplifiers, transformers, couplers, and see how cheaper cables cause problems up in the GHz region compared to the exotic ones. Maybe start with RFSIM99 and play with a few s1p and s2p models and see how much you can learn. Then buy a proper simulator tool?

[G0HZU]

The other thing I'll add is that you have recently bought a high speed probe so I assume you will be looking at PCB design for high speed data. I don't have much experience of ultra high speed data design but I have got some experience of the design process for JESD204B. eg I've worked on circuit design for a complex board that has a control interface to a couple of very fast DACs that runs at up to 6Gbps. Generally, a lot of the work is done via simulation and we also work closely with the PCB manufacturer to ensure that the printed transmission lines used in the interface are the same length and behave as expected. Both the PCB design software (Mentor) and the PCB manufacturer have some input to this and a VNA often isn't used at any time. However, we have used a VNA with TDR on some external high speed connectors on some other designs. But a lot of this stuff can be simulated with fairly low technical risk. The risks increase if the board gets busy and the traces have to bend or meander quite a bit.

[rx8pilot]

I assume you are referring to me from your other thread here.

Not specifically - and no offense intended in any way. For the past 5-6 months, I have been trying to gain a 'managers understanding' of high-speed digital signal integrity and a pinch of RF to see if it is worth steering the business in that direction. I have watched a lot of video, read a lot of app notes, white papers, and websites. Your contribution in the previous thread was hardly the most intimidating. A lot of it is over my head at the moment so, naturally, it looks like black magic from my perspective. I am old enough and experienced enough to know that it is not really magic, but rather a hard learned and detailed engineering discipline. What I poorly communicated is that I cannot tell the difference between real life and academic knowledge. The 'volt-nut' term was a poor choice, but I hope you can understand what I am hoping to communicate. The decision I am faced with right now is more of a management decision of whether or not to pursue opportunities that will require investment in myself, other people, and equipment to achieve. If that decision is a 'yes' - I then have to know enough and buy enough to build the first few devices (relatively simple) to gain enough confidence to hire someone with much more skill that can be dedicated to the tasks that I will never have the ability to do on my own. When (if) we get that far, I need to be able to effectively communicate in detail and be realistic with the new engineer instead of just hoping they know what they are doing.

It is very unlikely that I will ever be an amazing RF or high-speed designer. My goal is to be decent enough to be able to understand the magnitude of various details and build an infrastructure that people smarter than I can effectively design and build new things. At that point, I will move on to the next big challenge. I have done this a dozen times in my career where I dove into totally unknown territory, figured out as much as I could, hired people to take it to the next level and moved on to build another team.

If you are running a business, my advice would be to buy a decent RF simulation suite and learn how to use it. It won't be cheap but this is the best tool to have in any RF workplace.

What are the top packages for software like this? What price range do they fall in? I remember initially avoiding SPICE thinking it was faster and easier to just build the circuit and practically test it. Hard lesson to learn. The time spent learning SPICE has cut enormous amounts of time off of my various design efforts. For my first few efforts, I am planning to layout essentially app-note and proven circuits to learn how to measure them. There are some demo boards, and reference signal generators that should help me learn how to measure without measuring my own unknown mistakes as well. Hopefully, the expense of the software can be deferred to a point where we get a few things on the market that have known designs. Does it make sense to learn to measure first before simulating so that when the sim is complete - I can compare it with reality.


On the stated subject - Does a VNA care what impedance your DUT is or does it simply show you impedance/frequency? That additional information would seemingly be the reason to use a VNA over an SNA or SA/TG for return loss - the user gains a broader picture of why return loss is poor.

[rx8pilot]

The other thing I'll add is that you have recently bought a high speed probe so I assume you will be looking at PCB design for high speed data. I don't have much experience of ultra high speed data design but I have got some experience of the design process for JESD204B. eg I've worked on circuit design for a complex board that has a control interface to a couple of very fast DACs that runs at up to 6Gbps. Generally, a lot of the work is done via simulation and we also work closely with the PCB manufacturer to ensure that the printed transmission lines used in the interface are the same length and behave as expected. Both the PCB design software (Mentor) and the PCB manufacturer have some input to this and a VNA often isn't used at any time. However, we have used a VNA with TDR on some external high speed connectors on some other designs. But a lot of this stuff can be simulated with fairly low technical risk. The risks increase if the board gets busy and the traces have to bend or meander quite a bit.

Yes - high-speed PCB design is part of the challenge for sure. I am very much hoping that the PCB challenges can be met entirely in a software domain. As the industry is pushing for 12Gbps signals, every detail has become big and important. Having a lot of those signal aggregate into a small PCB, driven into an FPGA, and back out - there is little forgiveness. I worked on a project that used 3Gpbs serial I/O (not as the circuit PCB designer) and saw what it took to ensure signal integrity at that speed - not trivial. Now the trend is 4x faster and I am on the bottom of the learning curve to make it even more fun.

Thank you for taking the time to help and contribute. Looking forward to the day where I wake up and it no longer looks like magic, but rather engineering.

[G0HZU]

What are the top packages for software like this? What price range do they fall in?

I think this question could easily set off a heated discussion if this was on a dedicated RF design forum

Just like scopes and other tools there are quite a few choices. Usually there is a love/hate aspect somewhere for some vs others. 12 or so years ago I would have said Eagleware Genesys was the best (for me) for various reasons. But since Agilent/Keysight took over it went through a poor phase lasting several years. The recent versions are much improved. It's expensive these days and I've lost touch with prices. I think you could easily spend many, many thousands even for the basic version. Get lots of options and you are heading towards upmarket saloon car prices. The same applies to all of the tools from the other vendors below.

Genesys is still my favourite. It's like an old comfy pair of slippers for me because I've been using this SW right back to its roots nearly 30 years ago in the dark days of DOS

http://www.keysight.com/en/pc-1297125/genesys-rf-and-microwave-design-software?cc=US

The other popular one is AWR Microwave Office and I've used this a lot too. Very powerful, very versatile, very expensive but not very intuitive to use (for some at least). At work I haven't upgraded to the latest version (for several versions now) because it does what I need already. The UI may have improved and I know some of my colleagues use the latest EM tools in the latest version with great success.

http://www.awrcorp.com/products/ni-awr-design-environment/microwave-office

The link above seems to be their main page since NI took over. Looks a bit bland in terms of marketing... have I missed their proper/current website?

Sonnet EM is a brilliant EM Simulation tool for analysing PCB based filter designs, transmission lines and other structures on a PCB even with many layers. They do a free Lite version which is quite useful as they have upped the limits on the free version. However, if you don't do much RF design yet you probably won't like using Sonnet for a while. But it is my favourite EM simulation tool. It has helped me many times where other EM simulators have failed to model a problem. I've been using Sonnet for about 14 years at work.

http://www.sonnetsoftware.com/

There are several other big ones like Keysight ADS and the range of ANSYS products. But we don't have these at work and I've never even tested a trial version. But they are very expensive and very powerful.

Keysight ADS is probably the daddy for most people...
http://www.keysight.com/en/pc-1297113/advanced-design-system-ads?cc=US&lc=eng

ANSYS has a very powerful EM simulator amongst all its simulation tools. It has loads of other simulation packages too.
http://www.ansys.com/?gclid=Cj0KEQjwk-jGBRCbxoPLld_bp-IBEiQAgJaftY5zY9DlelESDo5pBF2lDyk1vsrRbw3pRazOTTHm_3UaAtwR8P8HAQ

One thing I will warn everyone about is to not try any 'shortcuts' to getting any of these big name SW packages if you are running a business. At least one of the big names above will come after any company that does this. I won't name it for obvious reasons but I know of this stuff happening and they do try to detect (via the internet?) if you have ever run the SW on or offline.

You can get a free 30 day trial for many of the above and I think there are budget suites available too. But you need deep pockets even if running a business. 

RFSIM99 is obviously very dated now and hasn't been revised for many years but it is free and quite a few people still use it

It might be good enough for your needs especially if used with Sonnet Lite. I don't know if Sonnet Lite can export s parameter data but if it can these two would be quite powerful when used together and they are both free? Note that RFSIM99 is just a linear simulator that operates in the frequency domain. It isn't like using SPICE. It's at its best if you want to look at transmission line models or filter models along with lumped components. You can't do any fancy transient analysis with it but it does come with some filter synthesis tools if you want to design lumped RF filters and simulate them.

[G0HZU]

On the stated subject - Does a VNA care what impedance your DUT is or does it simply show you impedance/frequency? That additional information would seemingly be the reason to use a VNA over an SNA or SA/TG for return loss - the user gains a broader picture of why return loss is poor.

The advantage of using a typical full 2 port VNA that has been properly SOLT calibrated with a decent cal kit is that you can measure any 2 legged linear component and produce a small signal (linear, steady state) model of it to use in the frequency domain. This model will be a huge data file, with a name like device.s2p.

Once you have a valid model that has been properly de-embedded (i.e. the model applies at the connection pins of the device and not the test fixture connections) you can import this device.s2p file into a linear simulator and it will look like a two legged 'box' and you can use it as any other component. You don't have to connect it to 50 ohm ports, you can place it anywhere.

For example, when I made that resistive 75R to 50R match pad I measured it with the 50R VNA as a two port device and exported the model data into a simulator and drove it in the simulator with a 75R port on the input side and a 50R port was placed on the outside in the simulator. So the insertion loss and port matching was then valid for the 75 to 50R matching pad. This may seem too good to be true but you don't have to connect 2 port models (taken with a 50R VNA) to 50R ports in the simulator as long as you de-embed right up to the pins of the device when taking the model with the VNA. Or you can de-embed in the simulator. I often do it this way.

Another example is a high impedance crystal filter. eg something with an impedance of several hundred ohms. You can measure it as a two port device with a 50R VNA, export the 2 port model to a simulator (as a component with two legs) and then set the simulator ports to the correct impedance for the filter. The simulated filter model will then give a frequency response pretty much as expected/shown in its datasheet with good match and low loss.

[G0HZU]

Another example closer to home would be wrt RF cables. You can measure a coax cable as a two port device, export it as an s2p file and use it in a simulator. You can use the model several times in the simulator if you like. You could also take a model of an RF adaptor or a coaxial attenuator or matching pad or a filter in the same way.  Then have hours of fun connecting them together in the RF simulator to predict the cascaded VSWR and loss and group delay etc. Even RFSIM99 can do this.

But you need a decent (full) n port VNA and cal kit and fixtures to do this stuff properly especially up at many GHz.
A full 2 port VNA can measure S11, S21 S12 and S22 sequentially in the same setup without having to change connections and it can correct for its own system errors with the SOLT calibration.

[G0HZU]

It's late and I'm off to bed but see below for an image of an old Genesys simulation I did of a Yaesu crystal filter at about 8MHz. This filter has an impedance of several hundred ohms. Definitely not 50 ohms!

You can see three two legged grey boxes in the schematic. These are VNA data files.

The top one SP3 is the VNA model file taken of a direct (therefore badly mismatched) connection to the filter with a 50R 2 port VNA and the simulator port impedance (ports 1 and 2) has then been altered to peak up the response for best match. This is somewhere around 416 ohms as you can see in the image. The response looks good at this port impedance! If I had left the impedance at 50R in the simulator it would have looked very ripply and lossy because it is so mismatched like this.

The next one down is SP1 and this is exactly the same model file as SP3. But in this case I've connected to regular 50R ports via a matching network and I've designed/tweaked up the matching network to match 50R to 416R at 8.215MHz to get nearly the same response (it's actually better) as the one above but from 50R ports.

The last grey box is SP2. This is measured data (of a straight 50R VNA measurement) from the real filter built on a PCB and now matched to 50R  using lumped matching components with values as per the simulated matching network. So I'm now measuring the real filter with the VNA using real matching parts and you can see that the real hardware matches VERY closely with the circuit simulation. You can also see that both SP1 and SP2 match slightly better than SP3 because I've optimised the capacitive tap matching network in the simulator for best match and each matching circuit has a different inductance. One of the advantages of doing it this way is you can experiment with lumped matching for as long as you like and optimise it for a sweet spot in terms of matching.

The point of showing you this is that the same principle applies at any frequency although the care required goes up a LOT in terms of getting the calibration and de-embedding correct. But usually, the fixture is a regular one so the de-embedding is already characterised over many GHz BW and saved in advance to be used over and over with lots of future circuits. You don't have to stick with 50R connections in the simulator once you export a 2 port model from the VNA

[rx8pilot]

@G0HZU That was interesting going through that list.

Curious if there are modular options for designs that are for high-speed digital. I have no (apparent) need for filter and attenuator structures. The cost of full blown RF simulation is enormous and I feel like it may be like using Photoshop to crop family photos. Am I in for a huge surprise? Are circuit and PCB designers dealing with high-speed memory and other warp-speed high bit rate systems use these simulation systems for signal integrity analysis?

I did find an interesting test note from Samtech about how they tested their high-speed BNC's. Can't wait to explain to my wife that I plan to buy a 20Ghz VNA.

http://suddendocs.samtec.com/testreports/rfc-report-sma-bnc7t_th1-and-bh1_web.pdf