VNA for cable characterization

[joeqsmith]

For the case where the load is a short, Zl = 0.   Zero times anything = 0.  The square root of 0 = 0.  So for a short, Zo = 0.   

Considering Zl = infinity where infinity times any non zero = infinity.  The square root of infinity = infinity.   So for an open, Zo = infinity.   

I would think we could measure the cables actual impedance using these two extreme cases but rather than Zl, we would use the impedance at the lowest test frequency. 

Showing both extremes with the section of TEFLEX cable, with the coffee bag ties, using the impedance at 300kHz and the first crossing point. 

[joeqsmith]

Using the a 25 ohm terminator at the end of the TEFLEX cable rather than the 50 mentioned in both videos, we can now see the actual crossing point.  We measure 50.69.   Much closer to the TDR measurement using the Agilent (white) and LiteVNA (green). 

[G0HZU]

By doing a bit of mathematics with complex numbers, it's possible to estimate Zo by calculating Zo = sqrt(Zoc*Zsc) at each test frequency. That's how I did it using the nanovnaH and also with the physical model of sucoform 141 cable. This method is ancient and dates back over 100 years.

However, when measuring real cables it's best to use a really short cable if you want to do this for VHF or UHF frequencies as it isn't possible to fit a perfect short and open at the far end of the cable. Because of this, there will be a blip or glitch in the results at the quarter-wave frequency and also at higher multiples.

A typical 50 ohm coax cable that is about 3 feet long will give a glitch up around 55MHz for example.

[G0HZU]

It ought to be possible to predict the performance of the Tektronix cable if it uses (double screened) RG223 cable. The insertion loss of a 1m section of genuine RG223 is about 0.22dB at 250MHz according to the datasheet.

This just over 3 feet long, so if you factor in the connector losses the insertion loss of the cable is probably going to be about 0.25dB at 250MHz. It might be a bit higher than this but it will probably be less than 0.3dB.

A similar cable made with good quality RG58 will have a loss about 0.05dB greater than this. Some of the loss is due to the inferior screening and some will be due to the stranded centre conductor. Lower cost versions of RG58 (eg for Ham or CB radios) will have a bit more loss. A cheapo cable that is quite flexible will have poor screening and increased metal losses. It might show 0.35dB loss or even 0.45dB loss at 250MHz for example.

Good quality connectors will be needed, and even then there will be some risk of variations in output level if testing something old like a 1970s scope because of the connector wear on the BNC ports of the scope.

The harmonic distortion from the SG503 will also contribute to uncertainty in the output level (especially odd order harmonics). For example, a third harmonic at -40dBc can cause up to about +/- 0.09dB uncertainty in the output level. This is about +/- 1% in terms of Vrms.

The accuracy spec of the SG503 isn't that great either, especially above 100MHz where it is spec'd at 3%. So it probably isn't worth getting too serious about this stuff.

[joeqsmith]

Getting back to your questions,

One thing is the step and impulse diagrams in the time domain. As far as I understand it, those should be convertible into each other by means of integration and differentiation. ...

Beyond not having enough data points, looking at my software I have been using a trapezoidal integral rather than Simpon's which would cause some small errors.  I doubt you would see a difference but I have gone ahead and changed it. 

In the graphical views, there is the possibility to deactivate auto scaling for the X and Y axis via the context menus, which I find quite useful for looking at zoomed portions of the graphs. While this works perfectly in the time domain graph in the advanced tab, for the graphs in the main tab (like Impd Rectagular and Reflection Coeff), the context menu is missing the check marks in front of the auto scale entries. They also do not appear when switching auto scaling on and off for the vertical axis. At the same time, horizontal auto scaling is reset to auto for every sweep. So, for looking at a horizontally zoomed portion of the graph (e.g. 0 to 500M out of 0 to 4G), one needs to disable the sweep first, or all manually entered values will get sweeped away.

I have a note in my software about this.  Originally both the X&Y autoscaled.  After I made the software public, someone was measuring attenuators and was wanting to disable the vertical autoscale.   I had added that for them but not the horizontal.  I have gone ahead and added that while I was looking.   Let me know if there is anything else you want me to look at.  If not, I'll go ahead and release this.

***
New release is available with the mentioned changes.   

[Pinörkel]

Getting back to your questions,
Beyond not have enough data points, looking at my software I have been using a trapezoidal integral rather than Simpon's which would cause some small errors.  I doubt you would see a difference but I have gone ahead and changed it.

Ah, so the step and impulse responses are not both calculated independently from the S parameters, but one is calculated from the S parameters and the other one derived by numerical integration. As far as I remember it, trapezoidal really shines for periodic integrands and odd numbers of function evaluations and Simpson is better on even number of function evaluations.

I have a note in my software about this.  Originally both the X&Y autoscaled.  After I made the software public, someone was measuring attenuators and was wanting to disable the vertical autoscale.  I had added that for them but not the horizontal.  I have gone ahead and added that while I was looking.   Let me know if there is anything else you want me to look at.  If not, I'll go ahead and release this.

Thank you very much for addressing this. This will be really helpful. I better do not start looking for GUI issues. If I do, I will find a lot of odds and ends (its one of my occupational diseases, related to my computer graphics and data processing origins), and it is not your duty to play repair monkey for someone else. You are incredibly helpful without that already.

Edit: I just tested the new release and turning off the horizontal auto scale works in the main tab. :-) However, the check marks on the menu entries in the axis context menus are still not updated. They initially show no check mark, when the auto scaling is on and do not change, even when switching back and forth multiple times. Also, the time domain graph now defaults to horizontal auto scaling being switched off with some strange initial values. However, here the auto scale status is correctly reflected by the check marks.

[joeqsmith]

Correct, the step is derived from the impulse by performing the integral, then its just Z=Z0(1+S11)/(1-S11), where Z0 is the normalized value (50 ohms).

https://unacademy.com/content/gate/study-material/chemical-engineering/integration-by-trapezoidal-and-simpsons-rule/
 
Yes, the checkmarks will not work with the main XY graph like they would with all other standard graphs.  You basically have a toggle mode as a work around.  Thats the nature of the beast.  I've had a few people ask about scaling the entire program to support different screen sizes.  Part of the problem with this is how that front graph works.  Way back when I wrote this thing, Labview had no support for Smith charts and such.  While they now have some support for Smith which will autoscale, I don't like how slow it is.  So for now, you are stuck with something I put together 20 or so years ago.

I could have very well changed some of the starting defaults.  I'm pretty bad about doing that.  Some of the defaults I have saved to the defaults file, but not all.   

[G0HZU]

For the case where the load is a short, Zl = 0.   Zero times anything = 0.  The square root of 0 = 0.  So for a short, Zo = 0.   

Considering Zl = infinity where infinity times any non zero = infinity.  The square root of infinity = infinity.   So for an open, Zo = infinity.   

I would think we could measure the cables actual impedance using these two extreme cases but rather than Zl, we would use the impedance at the lowest test frequency. 

Showing both extremes with the section of TEFLEX cable, with the coffee bag ties, using the impedance at 300kHz and the first crossing point.

I'm afraid that you aren't going to get much joy if you combine data at different test frequencies because of skin effect for one thing. Skin effect will result in different conductor resistances at each frequency. Also, it's risky to trust the quality of an open (or a short) up at the quarter-wave frequency up at VHF. So combining data at 300kHz and 45MHz isn't going to give useful results.

It's much easier and better to measure the cable twice with a VNA set up for an S11 measurement. Once with a decent short at the far end and once with an open at the far end. It's best to add a cal kit open and short in each case as this will give very close to the same electrical length in both cases.

Then at each frequency point compute mag(Zo), re(Zo) and Im(Zo) using Zo = sqrt(Zoc*Zsc)

This should give valid data up to frequencies that are just below the quarter-wave frequency. Close to the quarter-wave frequency, the data will have errors because it isn't possible to have a perfect open or short at the far end of the cable. So the equation will give poor results here. However, it is possible to add a delay to one set of results and this delay might be a picosecond or so. This can account for any delay offset  between the open and short. This can mute the glitch in the results at the quarter-wave frequency and give a smoother Zo response here.

[joeqsmith]

For the case where the load is a short, Zl = 0.   Zero times anything = 0.  The square root of 0 = 0.  So for a short, Zo = 0.   

Considering Zl = infinity where infinity times any non zero = infinity.  The square root of infinity = infinity.   So for an open, Zo = infinity.   

I would think we could measure the cables actual impedance using these two extreme cases but rather than Zl, we would use the impedance at the lowest test frequency. 

Showing both extremes with the section of TEFLEX cable, with the coffee bag ties, using the impedance at 300kHz and the first crossing point.

I'm afraid that you aren't going to get much joy if you combine data at different test frequencies because of skin effect for one thing. Skin effect will result in different conductor resistances at each frequency.

Time domain combines several test frequencies.  Plus we introduce other concepts like gating. 

Also, it's risky to trust the quality of an open (or a short) up at the quarter-wave frequency up at VHF. So combining data at 300kHz and 45MHz isn't going to give useful results.

It's much easier and better to measure the cable twice with a VNA set up for an S11 measurement. Once with a decent short at the far end and once with an open at the far end. It's best to add a cal kit open and short in each case as this will give very close to the same electrical length in both cases.

Then at each frequency point compute mag(Zo), re(Zo) and Im(Zo) using Zo = sqrt(Zoc*Zsc)

This should give valid data up to frequencies that are just below the quarter-wave frequency. Close to the quarter-wave frequency, the data will have errors because it isn't possible to have a perfect open or short at the far end of the cable. So the equation will give poor results here. However, it is possible to add a delay to one set of results and this delay might be a picosecond or so. This can account for any delay offset  between the open and short. This can mute the glitch in the results at the quarter-wave frequency and give a smoother Zo response here.

The point of that little exorcise was mention what happens with the formulas presented in the two videos at the endpoints.  Of course, you could never achieve these conditions as obviously nothing is perfect,  open/short or anything else.  Even thinking the standards would have the same delay is flawed.  While using the impedance at 300k does get us much closer than using infinite and 0, its obviously not something that would be used.

[joeqsmith]

Also, the time domain graph now defaults to horizontal auto scaling being switched off with some strange initial values. However, here the auto scale status is correctly reflected by the check marks.

I have gone ahead a set that back to auto and changed the default start and stop frequencies to the range of the LiteVNA.   

***
Member RHB introduced me to ego pimping and padding.  I put both to use.      Around this post, you will find more details on the time domain:

https://www.eevblog.com/forum/rf-microwave/nanovna-custom-software/msg2682576/#msg2682576

[gf]

Getting back to your questions,

One thing is the step and impulse diagrams in the time domain. As far as I understand it, those should be convertible into each other by means of integration and differentiation. ...

Beyond not having enough data points, looking at my software I have been using a trapezoidal integral rather than Simpon's which would cause some small errors.  I doubt you would see a difference but I have gone ahead and changed it. 

You can of course zero-pad in the frequency domain (before doing the IFFT) to interpolate in the time domain. With increasing resolution in the time domain, the integration method becomes less and less important.

EDIT: That results not only in a higher resulution in the time domain, but also in a proper sin(x)/x interpolation, while e.g. trapezoidal integration calculates the integral as if there was linear interpolation between the points, and Simpson is not equivalent to sin(x)/x either.

[Pinörkel]

@G0HZU and @joeqsmith: Thank you very much for all your incredibly interesting and helpful input on this topic. At the moment, I do not understand all aspects of your posts, but I will try to reproduce some of the measurement ideas, once I understand how and I have the materials to do so. Non-50Ω-loads are kind of difficult to get here, so I will try to build some DIY ones. Those will not be of good quality, but maybe they will be good enough to learn something. Unfortunately, I am still waiting for my ordered experimentation parts, which seem to have gone lost during shipping.

Apart from that, I tried to get something useful out of playing with the liteVNA calibration standard characterization values, I found online, which I linked earlier. Regarding this, I have a question: Assuming the liteVNA standards have parameters different from the ideal standards, and one would precisely measure and enter those parameters in Solver64 and then perform a SOLT calibration. What would be the expected effect on a reflection coefficient measurement with just the calibration load connected, like the one I made in this post(green curve)? How would that possibly differ from a result obtained with faulty calibration standard parameters?

[joeqsmith]

Assuming the liteVNA standards have parameters different from the ideal standards, and one would precisely measure and enter those parameters in Solver64 and then perform a SOLT calibration. What would be the expected effect on a reflection coefficient measurement with just the calibration load connected, like the one I made in this post(green curve)? How would that possibly differ from a result obtained with faulty calibration standard parameters?

Never, ever connect a SMA to a 3.5 mm calibration standard, VNA cable, or some other expensive 3.5 mm device. Use the jack savers.

https://www.microwaves101.com/encyclopedias/how-to-not-trash-a-calibration-kit

If you use a bad standard and/or wrong coefficients to cal the VNA and you measure that same part, I am expecting very good return loss.  The VNA doesn't know the difference.   If you wanted to measure actual return loss of an unknown device, you need to cal the VNA with a known standard.   

***
https://www.eevblog.com/forum/rf-microwave/nanovna-custom-software/msg5206677/#msg5206677

[Pinörkel]

If you use a bad standard and/or wrong coefficients to cal the VNA and you measure that same part, I am expecting very good return loss. The VNA doesn't know the difference.

That is approximately what I would have expected, based on your good explanations in this and other threads. However, I got a kind of unexpected result, when entering the liteVNA calibration standard characterization values, I found online. I know that those could as well not be closer to the real parameters of my liteVNA calibration set than the ideal parameters Solver64 assumes by default. The effect I get is the following:
I calibrated my warmed-up liteVNA in Solver64 using the supplied calibration set multiple times and did so with the ideal calibration values active and the set of values, I found online for the liteVNA calibration set active. In both cases, the results in the smith chart and the reflection coefficient measurement were practically indistinguishable, when conducting the measurements while the ideal calibration values are active in Solver64. I consequently got the impression that the calibration does not depend on the calibration standard values that are active in Solver64 during the calibration (I may be wrong there). Interestingly, if I switch to the set of non-ideal calibration values for the measurements, the reflection coefficient looks, like it is getting linearized to a target value of -40dB (see image). The better(less noise) the calibration is in terms of the smith chart when checking with the short, open, and load conditions, the closer the curve gets to -40dB. Now, regardless of whether this measurement is nonsense, I would like to understand the reason that is causing this. Is there any obvious explanation for the observed effect?

[joeqsmith]

Post a screen shot of your settings.  Not the report. I want to see exactly what you are doing. 

I consequently got the impression that the calibration does not depend on the calibration standard values that are active in Solver64 during the calibration (I may be wrong there).

This is correct.  During calibration, Solver64 disables calibration and collects the raw data for each standard.  You can average but raw meaning no coefficients are applied.   The raw data is then combined with the selected cal term coefficients to form the error terms.  These error terms are then applied to each subsequent measurement.

[joeqsmith]

Measure S11 of the male load. The phase data will be meaningless, but the amplitude should show a very high return loss. Exactly how high depends on your VNA, but 50 dB or better (|S11| < -50 dB) should be observed on any laboratory VNA. This does not mean the return loss of the load is 50 dB! Irrespective of how good or bad the load is, if you measure the device that has just been used to calibrate the VNA, it should show almost perfect results - even if you used a wire-wound resistor!

From:  https://www.kirkbymicrowave.co.uk/Support/FAQ/How-do-I-verify-the-calibration-kit-is-working-properly/

[Pinörkel]

I just identified the value I entered, which triggered the observed -40dB linearization behavior. I thought, I had to enter the precise resistance of the load in the Z0 setting of the load (which is of course wrong, since it is an impedance, not a resistance). So, I put 51,06Ω there, which then triggered the described behavior. Just entering e.g. 53 as the Z0 parameter for the load in the Custom Ideal Standards set can trigger the same effect. The more the value differs from the other Z0 values for the open, short and thru, e.g. 50Ω, the worse the return loss will get.

[joeqsmith]

I take it that you don't expect this.  Maybe you could explain what you are thinking.

***
Check out the following recent thread discussing using a terminator as a load standard. 
https://www.eevblog.com/forum/rf-microwave/mini-circuits-karn-50-18-as-the-reference-termination-for-a-vswr-bridge/

[gf]

I just identified the value I entered, which triggered the observed -40dB linearization behavior. I thought, I had to enter the precise resistance of the load in the Z0 setting of the load (which is of course wrong, since it is an impedance, not a resistance). So, I put 51,06Ω there, which then triggered the described behavior. Just entering e.g. 53 as the Z0 parameter for the load in the Custom Ideal Standards set can trigger the same effect. The more the value differs from the other Z0 values for the open, short and thru, e.g. 50Ω, the worse the return loss will get.

I'm not sure I understand what you are after.

1-port SOL calibration basically works as follows: You have three standards with known S11(f). The VNA's calibration procedure mesures their response, and you tell the VNA the known ground truth S11(f) values of your standards. Then the VNA can calculate the error correction terms for subsequent real measurements. It does not matter if  the load standard is exactly 50 ohms or not -- you just need to know its actual S11(f). The same applies to the short and open standard as well. They do not need to be ideal (you cannot realize ideal standards anyway), but you only need to know their actual S11(f).

If you tell the VNA that your standards are ideal (although they are not), then you have to live with the resulting error. This may be good enough at low frequencies if the load standard is very close to 50 ohms, but at UHF or GHz frequencies every millimeter of length begins to matter, and parasitics play more and more a role.

Instead of dealing with tabulated S11(f) of the standards directly, usually a parametric model with a low number of parameters is fitted to the measured S11(f) of the standards in order to reduce dimensionality, which also aids noise reduction and enables re-calculation of S11(f) for arbitrary frequencies from the model parameters. In this case you have to tell the VNA the (known) model parameters of each standard instead of S11(f).

EDIT: If you calibrate the VNA using the actual model parameters of the standards and then measure the load standard (with calibration applied), then the VNA is expected to display the actuall S11 of the load standard.

OTOH, if you tell the VNA that your standards are ideal (although they are not) and then measure the load standard (with calibration applied), the the VNA is expected to display S11=0, even if the true S11 is different. Btw, note that you'll never see S11=0 in practice, due to noise and uncorrectable residual errors. IMO you can interpret your green curve as S11=0 + some residual errors + noise.

[Pinörkel]

@gf: Thank you for your good explanations. This is approximately what I already figured out based on the input I got in this thread.

I'm not sure I understand what you are after.

It seems like I am not very good at explaining my intentions behind my little experiments in this matter. So, I'll try again: For using the liteVNA, I have only the supplied calibration standards, which seem to be of bad quality, according to the most likely well founded opinion of several people here. Buying a good set of calibration standards would be nonsense in this specific case, since those cost a small fortune and my application case and usage frequency would in no way justify spending that much money. Additionally, I am after measuring BNC cables, which are known to not have very good HF capabilities. I also do not have the possibility to measure my crappy calibration standards on a good and calibrated VNA to overcome this issue. Then I found someone online who did exactly that and measured the liteVNA calibrations standards with a good VNA. Said person also provided the measured offsets and error correction coefficients for free. So I thought it could be a good idea to use those values as my calibration parameters, since I have the same type standards. I further thought, that most of those parameters (except the load resistance) should result from the physical geometry of the standards and should consequently be very similar in my actual calibration set. At least more similar, than just assuming ideal parameters. So I tried to match the provided values with the input fields of the Solver64 software, made an error while doing so, and got this, at least for me, unexpected effect of linearization, which I could not explain. However, I was curious what exactly caused this.

At the moment, I have the theory that a deviating value of the calibration constant for Z0 of the load somehow causes the frequency invariant noise floor to be amplified proportional to the deviation, which then eats up the real S11 curve and results in kind a linear curve as the noise level rises all over the spectrum.

[gf]

S11 = (Z_L-Z0)/(Z_L+Z0), where Z0=50 ohms.

The expected value of S11 depends on Z_L alone. Consequently, if your load's Z_L would be (say) 51 ohms (independent of frequency), then its S11 would be (51-50)/(51+50)= 0.009901 or about -40dB (also independent of frequency). Only the deviation from the expected value (Z_L-Z0)/(Z_L+Z0) is caused by noise and residual errors which cannot be calibrated out.

This was just an example for an purely ohmic Z_L. Eventually, Z_L(f) can be frequency-dependent and complex-valued and is determined by the model parameters you enter for the load.

EDIT: When the calibration is applied, it attempts to adjust the raw readings so that the VNA displays S11(f) = (Z_L(f)-Z0)/(Z_L(f)+Z0) when you measure the load, where Z_L(f) is calculated from the model parameters you entered. You cannot figure out if the true value of your load deviates from Z_L. The calibration procedure relies on the model parameters you enter being consistent with the calibration standards you are using.

[joeqsmith]

For checking calibration quality, I read that I could maybe have a look at how to do a T-check. For this I would most likely only need an SMA all female t-piece.

Solver 64 uses the same coefficients HP uses.  My attempts to make a T-check have not worked out.  You don't want a T with an added stub and connector.  I think I have documented some of that in the NanoVNA custom software thread.   

Ah, you mean the PCB-style T-check fixtures? I just read from someone that a T-check could be performed with really simple materials and you do not even need a high quality load. However, I do not know whether that is true.

Showing FR-4 insertable T-Check next to the last coaxial one I made.   You could try and put something together from OTS parts like shown but I don't think you would find adapters to get the phase matched.   A non-instertable setup like I had demonstrated with the LiteVNA and transfer relay would avoid that problem.  The coaxial one was tested on the PNA using a set of Agilent standards for a reference.

For the math side of things, I suggest downloading the old HP application notes and start reading.  There is a treasure trove of information out there and it is all free.  Just be aware that these are not the gospel and I have caught errors in them.  I would have thought that book you purchased would have dove into the math as well. 

***
Showing homemade T-Check with transfer relay and original NanoVNA.   This would have been FR4 standards with ideal model. 
https://youtu.be/GJNMnq8eD0E?t=2433

The V2Plus4 with a mechanical transfer relay and FR4 T-Check.  Both cases, it looks like I ran it to 50MHz was all and results were not great.  When shown with the Agilent, we are using the proper coefficients.  I would imagine if I derived the coefficients for the V2Plus4 standards, then measured the T-Check, it would have greatly improved.   
https://youtu.be/XaYBpPCo1qk?t=2173

[joeqsmith]

Just to add to GF's post, or further confuse you...   Ignoring the math, if you tell the VNA an imperfect load is perfect, the VNA will show it as perfect.  If you tell the VNA it is not perfect, it will show it as imperfect.  Even is the load is perfect and you tell the VNA it is imperfect, the VNA will show it as imperfect.   You can really see this with high reflection standards (short and open) as rather than a dot, you will see the rotation.   

If you cal a perfect standard, and tell the VNA it is perfect, then measure an unknown, you will get the actual value.   

If you cal an imperfect standard and you tell the VNA it is perfect, now if you measure a part that is perfect, it will show it has errors. 

If you were to cal using say a 1Mohm resistor for your 50 ohm load cal standard but you know the coefficients for it and enter them into the VNA,  what will happen when you try and measure a 50 ohm load?? 

***

I think if I wanted to try and improve the T-Check on the LiteVNA 64, we would need to:

Start with a non-insertable setup so we can use a standard T an load for the T-Check. 
Use a known good setup to characterize the LiteVNA's standards.  This would mean using a high quality set of adapters to connect our low quality standards to.
We would also want to validate the T-Check's performance with that same setup.
Then it's just a matter of using the calculated coefficients for the low quality standards, running a full 2-port calibration (with the transfer relay...) after locking down the cables (tape).  Then measuring the T-Check.   Looks like it was pretty poor even at 500MHz in that one video.   

As far as transferring the coefficients to others, hard to say.  The ones I received are not marked like the ones you show in that paper, making me suspicious.  Even if they did look the same, how tight are they controlled? 

***
40&42 showing the standards that were supplied with the latest LiteVNA 64 revision 64-0.3.1. 

44 showing the standards supplied with the latest and previous standards supplied with the LiteVNA 64 and also the standards supplied with the the V2Plus4.  The two sets of standards were modified with end caps that allow me to lock the center conductors in place with a wrench.   

The DCR for the most current load is 51.057 ohms.   The previous measures 50.890.   I think to try the T-Check, I would want to select something better to use with the T.   

[Pinörkel]

I finally got my first delivery of experimentation material, which is some Belden H155 PE cable and fitting BNC connectors from Telegärtner. The Telegärtner connectors were the only non-no-name connectors I could get and I had used this German brand before on my RG58 cables. To get the connectors on, I first had to design a simple 3d-printed coax stripper matching H155 cable and the datasheet of the Telegärtner connectors. It worked out very well on the first try. Why H155? I wanted to try out a low loss cable stiffer than RG58 with non-foam dielectric for better impedance stability. In addition to that, there are some return loss measurements for my specific application case and H155 cable available that I could compare my measurements against.

So, I expected a cable with a much better return loss than my previous RG58 and a better impedance stability with an average impedance value somewhere close above but not necessarily at 50Ω. The following shows, what I got in terms of reflection coefficient and time domain step response, in comparison to my previous RG58 cable:

RG58, H155 and H155 (cable reversed) time domain step response:


From this I get that the impedance is indeed more stable, but still about 2Ω along the cable. Consequently, I would expect effects of the reflections at the cable end (BNC to SMA Adapter + SMA 50Ω load) to be more dominant than on the RG58 cable. Unfortunately, the impedance of my DIY H155 cable is further away from 50Ω than my RG58 cable.

RG58 reflection coefficient:


H155 reflection coefficient:


The main thing, I get from this is that there seems to be a strong resonance phenomenon on the H155 cable. Maybe this is a result from the less pronounced local impedance changes causing less distributed reflections, the lower theoretical cable loss and the cable acting as kind of a resonance cavity at certain frequencies. My H155 cable has a length of 36", so it is electrically short and the reflections caused by the connectors and adapters at both ends could bounce around and resonate. On the RG58 cable this effect(which is also kind of visible) may be much less prominent due to many additional reflections along the cable and the worse theoretical return loss. To my surprise the general shape of the frequency dependent reflection coefficient is very similar on the two cables, except the H155 performing worse at frequencies below 500 MHz. I am still searching for the much better return loss to show up somehow.

[joeqsmith]

HP used to make a similar BNC connector to this Amphenol part. 
https://www.digikey.com/en/products/detail/amphenol-rf/112332/3088368

Left Amphenol 74868 UG-88B/U  (assembled for photo, not actual install)
https://www.amphenolrf.com/031-2.html

Right is old HP 1250-0256 with section of RG400.