75ohm to 50ohm at high frequenies - 6Ghz How?

[Howardlong]

Although that BNC minimum loss pad is the neatest solution.

Another jury rig option is a 75 ohm RF termination on microstrip and use a FET probe rated to 6GHz or higher probing across the resistor. In fact, even a lo-Z 500 ohm x10 probe would work with minimal effect, use a 91 ohm termination resistor instead.

The biggest problem is finding BNC connectors that'll work reasonably well to 6GHz, making that mnimum loss pad much more palatable.

[rx8pilot]

The cable and BNC manufacturers are making tighter and tighter tolerances to keep up with the bandwidth requirements of TV/Film where the connector has outlived its original intention.

I am trying to get the Keysight N2752A 6Ghz probe which will allow analysis of the whole signal path instead of simply the 75 ohm output. Ideally, I would have two of them, but one will give me the measurements and confidence needed to put the new designs out in the wild. Baby steps and simple things first, then I can get another one in time. If I am successful in getting the probe - it eliminates the need for the impedance matching.

This is my first dive into high-speed digital - made possible by the fancy Keysight 'Test to Impress' win. Pretty excited to be in this place for sure. First up is a few general purpose I/O devices where I can get my feet wet with a modest challenge. The next year is designing FPGA hardware to manipulate the data I bring in. Fun.

[G0HZU]

I'm probably missing a point or 2 here, but what about the N2882A? 75 to 50, BNC, good to 8GHz...

http://www.keysight.com/en/pd-1651984-pn-N2882A/75-to-50-bnc-adapter?cc=US&lc=eng

Good find! And they seem to cost little over $100 each so that seems like a no-brainer to me to solve the problem.

But the RF performance of that Agilent adaptor is very poor up at 6GHz? eg very poor VSWR. Worse than 2:1.

If I had to do this I'd experiment with a direct transition from (decent) 75R semi rigid cable 'slotting' into a tiny PCB and experiment with different (pi) 75R to 50R attenuator values using 0603 resistors. I'd probably start with 11dB as this gives three very friendly resistor values for use up at 6GHz. I'd exit the PCB with a decent SMA end launcher connector. I'd also experiment with a simple 25R series resistor although I'd expect to be able to beat this with the pi pad method.

It would probably take me several attempts but I'd hope/expect to get better VSWR performance up at 3-6GHz when compared to the Agilent adaptor that uses a BNC connector at the 75R input.

I'd also consider the tapered line approach using microstrip but this will only give a good match above a certain frequency. eg 1500MHz. Fine if you only want to look across 3-6GHz but not really a wideband solution if you want to maintain performance down to DC.

[Howardlong]

54006a lo-Z 6GHz probes appear on ebay quite frequently at reasonable prices, make sure you get the original BNC/SMA adapter, in line DC blocking cap and walking sticks with spare resistors.

[rx8pilot]

The listing says they will charge 7.5% sales tax which seems weird because it is coming from Malaysia and going to California. Not sure if that is right or not and they have not yet responded to the question. The buy-it-now price includes shipping while the offers do not - not sure if they will simply upcharge the shipping to make up for my lower offer. The shipping is not stated in the counter-offer. Anyway...should have an answer soon.

I find it almost funny that I am trying to buy a probe for $2k+ and feeling like it is a bargain because it is normally $5.8k. My previous scope - a used Tek TDS754C was only about $750.

[rx8pilot]

But the RF performance of that Agilent adaptor is very poor up at 6GHz? eg very poor VSWR. Worse than 2:1.

If I had to do this I'd experiment with a direct transition from (decent) 75R semi rigid cable 'slotting' into a tiny PCB and experiment with different (pi) 75R to 50R attenuator values using 0603 resistors. I'd probably start with 11dB as this gives three very friendly resistor values for use up at 6GHz. I'd exit the PCB with a decent SMA end launcher connector. I'd also experiment with a simple 25R series resistor although I'd expect to be able to beat this with the pi pad method.

It would probably take me several attempts but I'd hope/expect to get better VSWR performance up at 3-6GHz when compared to the Agilent adaptor that uses a BNC connector at the 75R input.

I'd also consider the tapered line approach using microstrip but this will only give a good match above a certain frequency. eg 1500MHz. Fine if you only want to look across 3-6GHz but not really a wideband solution if you want to maintain performance down to DC.

Building my own solution would be a fun personal project for sure, but I am limited in options since I don't have an SA/TG or a VNA to properly characterize the finished device. The primary goal is to ensure signal integrity and match it to the published specification. I certainly cannot guess my way through the design without any testing. It is, however, a good excuse to get more test equipment in due time.

For now, I can only look at verified commercial solutions.

[G0HZU]

The primary goal is to ensure signal integrity and match it to the published specification.

But are you going to get that with the Agilent BNC adaptor? Looking at the datasheet the VSWR at the 75R port is worse than 2:1 across a fair bit of 1GHz thru 6GHz. 2:1 could mean 150R or 37.5R resistive or anywhere else on the 2:1 circle
Looking at the (mismatch induced?) ripple in the S21 plot and the shape of the S11 plot (75R port) I'd suspect that the input impedance is going around a loop shape on the smith chart that breaks the 2:1 VSWR circle fairly regularly across the band.

A 2:1 VSWR could mean an input impedance that looks like 150R or 37.5R resistive or some other impedance on the 2:1 circle on the smith chart. I guess it depends if that matters or not for your application?

[rx8pilot]

But are you going to get that with the Agilent BNC adaptor?

Nope - I would clearly have to do way better than that to get any useful information. The limitations you already pointed out ensure it does not help. Some of the other commercial options end up being about $1k after adaptations which is more than I expected.

BREAKING NEWS: Keysight took the offer and the N2752A 6Ghz probe is on the way. VERY excited about that!

[G0HZU]

FWIW here's a quick and dirty simulation of what I think the insides of that Agilent adaptor will contain. The simulation shows thru loss and input match at the 75R port.

This model took a few seconds to produce based on my experience of stuff like this. But you can see that the model and the official plots (on the right) agree even at this simple level.

There are only a few components in the model. To refine it would mean creating a very complex model but the quick and dirty model shows that this Agilent adaptor is a crude and unsuitable device if you expect it to give good performance up at 6GHz (no wonder it is so cheap ).

[rx8pilot]

What a mess, nuts. Thanks for the quick look.

With T&M - it is very rare to truly find a bargain. Low price, as you said, generally has a reason. The list price of the probe I just purchased is completely wild at $5,800. Since I can count on its performance (up to 7Ghz) and got it for a little over $2k - it kinda feels like a bargain. I only have to sell a handful of the resulting circuits to cover that cost and Keysight knows that. I can only sell them if it is very reliable and meets the published specs too.

Looking forward to putting all the learning efforts to understand high-speed signal integrity to use - even if only on a small scale at the moment. I am still quite deficient in T&M for what I am after - but time and determination will fix that.

[TheSteve]

Please do post some sample scope shots once you receive it and connect it to the DUT.

[G0HZU]

The other thing the model predicts is the group delay for that BNC 75R to 50R adaptor and it shows a ripple as high as 37ps peak to peak. See the plot below:

Obviously, this is just a model but it would be interesting to see what the official spec is for the group delay response. For a lot of users that Agilent adaptor will be fine even up at many GHz but only you can decide if the performance is good enough for your needs. It isn't a terrible device to be avoided at all costs, it just isn't a precision component up in the GHz region. If you can afford to splash out big money on probes you could still consider the adaptor for casual measurements or for use at lower frequencies?

[rx8pilot]

If you can afford to splash out big money on probes you could still consider the adaptor for casual measurements or for use at lower frequencies?

I financed a probe that was 60+% off list for 6 months  Not much of a cash splash, lol. I just recently paid off the various power related probes that I needed - they were also expensive, but allow me to design better circuits and therefore pay for themselves after I release a new product. Each T&M tool I purchase has to have a specific path to paying for itself within a modest timeframe.

I am considering some lower cost scopes and probes for the more common daily tasks that could risk the expensive stuff. It is a slow crawl at the moment. I better get something new in my catalog to make sure I can pay for this thing.

[G0HZU]

All my RF probes here are either homemade or salvaged from a skip and repaired.

For a bit of fun I had a go at making and testing a very crude pad to match 75R to 50R using cheapo SMD resistors.

I know how my SMD resistors here behave up at several GHz so I knew this pad would not be that great. If I had access to more choice in terms of SMD resistor package (like I have at work) I could do much better.

But see below for a plot of the response of a pad that cost about a penny in parts. OK, it's on a low loss test fixture made with a decent PCB but I have de embedded the result to just show the response of the resistor section. This is a real pad measured on a VNA calibrated with an Ecal module.

You can see that the RF performance is a bit better behaved. I'd probably embed 75R semi rigid into a real design and use a decent 75R N connector at the input and then use a decent 50R connector at the output. With better SMD resistors and some fiddling I think I could get the VSWR at 6GHz much better than this

The performance begins to unravel just above 5GHz in the plots below but it manages to keep the VSWR below 1.4:1 even with this early/crude attempt.

[nctnico]

What kind of connectors are on your impedance matching circuit? I have a feeling the BNC's aren't helping on Keysight's product.

[rx8pilot]

That's not bad for a such a quick effort. This is my primary reason for wanting a  VNA - literally to just do experiments for the sole purpose of learning and having fun. Thanks for sharing EXCELLENT!

Sent from my horrible mobile....

[G0HZU]

What kind of connectors are on your impedance matching circuit? I have a feeling the BNC's aren't helping on Keysight's product.

It's (kind of) 'connector-less' because I've de embedded the (50R SMA) connectors and the PCB microstrip used in the test fixture. So my reference plane is right at the input to the resistors. If I was to make a real version of this I would go straight in via 75R semi rigid and avoid the BNC connection. I think there are a few things that are less than ideal about the Agilent pad and I included this in my attempt to model it. I don't think it is just the BNC connector type although that obviously doesn't help.

[G0HZU]

If you are interested, here is an old plot of a precision 6dB splitter I made a couple of years ago. It was designed to be 'very' accurate up to 1GHz to allow me to test my old HP vector voltmeter  but it ended up working a lot better than this. I tried quite a few resistor types and once I found the best, I then did a select on test to find the most accurate resistors before building it. I also used Sonnet EM on a PC to model the PCB layout to optimise the VSWR across the full BW.

But the plots below include the connectors and it was all cal'd measured using a 4 port VNA and a 4 port Ecal module. Obviously, I only needed 3 of the 4 ports and the plots below are of the s parameter data exported from the VNA. The phase matching and VSWR and amplitude match are all very good. As good as some of the 'real' offerings from the big boys

If I was to try and make a precision 6GHz 75R to 50R match pad I would try similar tricks to get it as good as possible. With care you can make some decent accessories like this quite cheaply

[1audio]

Getting past 3 GHZ at 75 ohms is really hard. Keysight doesn't have a VNA for 75 Ohms above 3 GHz. I just looked. I have an 8753E 75 Ohm which was the best Agilent had at the time. The internal hardware could go to 6 GHz but no support etc. was available and its limited to 3 GHz. I don't know if you can get a test kit today good for much above 3 GHz @ 75 Ohms. Its all uncharted territory. If you can't get a termination for 75 Ohms good for much past 3 GHz its difficult to know if anything else is OK. I have a 75 Ohm air line but its terminated in 75 Ohm GR 874 connectors so 6 GHz is not likely.

And be sure to keep the 75 Ohm N connectors etc well away from the 50 Ohm N connectors. A casual mistake can be devastating if its on the front of the VNA.

The active probe may be your best option.

[Someone]

Getting past 3 GHZ at 75 ohms is really hard. Keysight doesn't have a VNA for 75 Ohms above 3 GHz. I just looked. I have an 8753E 75 Ohm which was the best Agilent had at the time. The internal hardware could go to 6 GHz but no support etc. was available and its limited to 3 GHz. I don't know if you can get a test kit today good for much above 3 GHz @ 75 Ohms. Its all uncharted territory. If you can't get a termination for 75 Ohms good for much past 3 GHz its difficult to know if anything else is OK.

Its not uncharted territory:

You should be able to use an Agilent 11852B (with N to BNC adapters)

There is a special order option for it: 11852B Option H12, Minimum loss attenuator tested to 12 GHz

So with a calibration (de-embedding http://literature.cdn.keysight.com/litweb/pdf/5989-5765EN.pdf) of the setup you can use 75Ohm systems (or other impedances with your own adaptors) out beyond 3GHz. How do you think people measure waveguides at their arbitrary impedances?

[G0HZU]

No test equipment is perfect, including a vector network analyzer (VNA). However, using a powerful
tool box of special operations enables the measurement errors to be dramatically minimized. These
operations improve the laboratory measurement data turning it into an excellent representation of
the device under test (DUT). While calibration is a pre-measurement step which minimizes errors, the
most important operation to reveal information about just the DUT is a post-measurement process
called de-embedding.

Traditionally, de-embedding has been used only by experienced users.

By understanding the principles and how to use the new generation of built in de-embedding features of
VNAs, this powerful technique can be leveraged by all users and the quality of information extracted
from all measurements dramatically improved. This practical guide to de-embedding will enable all
users to take advantage of this important feature.

I've taken this quote from the link in Someone's post above. These techniques are vital for the stuff I do at work because I have to do wideband design up to many GHz. I have to be able to examine/model various components or structures with a VNA and I can only do this using advanced de embedding techniques.

When I upgraded to my 4 port VNA just over a year ago for home use I spent nearly 2 months playing with the Ecal module and perfecting the de-embedding for several test fixtures I have here. That's virtually all I did with it in those first few weeks because this stuff is so important if you want to get meaningful/accurate results from your test setup. 'Having the gear' is only the first step.

Get this stuff right and combine it with a decent RF simulator and it enhances your capabilities way beyond the typical/casual VNA user

[1audio]

Thanks for the link. Lots to learn. The extended characterization moves the cost from $67 to $753. Clearly a lot of time invested in the measurements and number crunching. I expect there is some selection up front as well.

[rx8pilot]

Some great info for sure.

'Having the gear' is only the first step.

Yes, indeed. I have spent the last year just learning how to approach the high-speed circuits I have been hoping to design. I suspect that it will continue to be an uphill learning curve as I dig into the practical realities. There are so many test and measurement traps that can send the engineer on a useless tail chasing adventure. It does seem that there is always one missing probe, adapter, or cable that costs a ton of money though.

That is why this forum is so useful and fun - there is always someone that has ben stuck in the same trap and willing to talk about it. So much to learn, so little time......