Kirkby calibration kit alternatives?

[rfspezi]

Don't forget that your cal kit SHORT and OPEN is at the far end of either a male-male SMA adaptor or a female SMA adaptor. The typical delay in the F-F adaptor is about 42ps and the M-M is probably about 58ps.

So the moment you put your cal kit OPEN back on the VNA after a calibration the open end of your OPEN will be either 42ps or 58ps away from your reference plane. So the smith chart on the VNA will show a 0-6GHz line going maybe halfway around the smith chart. It's effectively measuring the SMA F-F barrel.

My guess is that you have tried to do a 1 port calibration and told the VNA the wrong gender for the cal kit. So the 42ps and 58ps are muddled. Try again for the 42ps female OPEN (and don't forget the VNA refers to this as OPEN (M) ) and it might look like this:

Thank you for the explanation!
So I definately used the wrong sex and need to think at the reference plane.
If i understood right, i should be able to verify a correctly performed calibration by e.g. connecting the SHORT-reference and applying a port extension that exaclty matches the one specified in the calibration data for convergence?

Concerning the loads the calibration file says:

Code: [Select]

** Male load. **
Minimum frequency:  0 MHz
Maximum frequency:  7000 MHz
Offset Zo: 50.0 ohms
Offset loss: 1000 Mohm/s
Offset delay: 0  ps

** Female load. **
Minimum frequency:  0 MHz
Maximum frequency:  7000 MHz
Offset Zo: 50.0 ohms
Offset loss: 3000 Mohm/s
Offset delay: 0  ps
Shouldn't the resistor values match the truely measured ones? (although measured with DC)
I guess the correct way would be to get better 50 Ohm references, right?

[drkirkby]

Hi,
I'm Dr. David Kirkby, the director of Kirkby Microwave.

Since a few people asked, I can confirm that the attenuator supplied in the SMA kits
http://www.kirkbymicrowave.co.uk/support/Kits/85033/
is measured with an Agilent 85052B 3.5 mm cal kit, not the SMA kit supplied. There are 5 measurements made on the SMA attenuator:

• S11
• S21
• S12 (in theory the same as S21)
• S22
• Reflection coefficient at the male port, with the female port unterminated

S11 and S22 measurements can't be expected to be very accurate at low frequencies, as the return loss of the attenuator is very good. A high return loss means a high measurement uncertainty. As the frequency rises, so the return loss of the attenuator deteriorates, and the S11 and S22 measurements are more meaningful. But the S21 measurement is useful at all frequencies.

The measurement of a reflection coefficient at the male port with the female port unterminated gives a return loss that will be met in many practical circumstances.

On the N kits
http://www.kirkbymicrowave.co.uk/sales/85054/
we also measure the reflection coefficient at the female port, with the male port unterminated. But we don't do that on the SMA kits as the nut on the male SMA wobbles around, changing the capacitance to the inner conductor. Since such a measurement is potentially unstable, we don't do it.

There's no doubt the attenuator is not a perfect test of a VNA, but it has in the past highlighted problems when people have entered the coefficients incorrectly.

There's a review of our kit on Amazon UK by Dr. Robert Watson at Bath University. He describes how he used airlines, and a Beatty Standard and compared the kit with various other kits. See

https://www.amazon.co.uk/Kirkby-Microwave-85033-calibration-verification/dp/B01BAWR70M/ref=sr_1_1?ie=UTF8&qid=1467301985&sr=8-1&keywords=85033+sma+calibration+kit/

I have no connection with Robert Watson, other than he bought a couple of kits, and another UK university bought one when they were collaborating with him. I never studied at Bath University, and have never met him, but he has agreed if anyone has any questions, he may be contacted. You can find his details on the Bath University website.

As for the comments by someone about the DC resistance of the loads, we are less concerned about their DC resistance than their RF performance. The loads are tested to have a return loss of at least 32 dB from 50 MHz to 7 GHz in a standard kit. (They are usually much better). We have an HP 3457A 6.5 digit multimeter (currently at Keysight for calibration), so could easily measure the DC resistance using 4-wire Kelvin connections. However, I feel that's irrelevant, as it does not indicate the RF performance at all, because the real part of the resistance is not independent of frequency. You will never see a specification for the DC resistance (other than 50 or 75 ohms) in any cal kit from Agilent. Of course, if the DC resistance is well away from 50 ohms, the return loss at low frequencies would be poor, and we would reject the load, irrespective of what its performance is like at microwave frequencies.

On some loads, the return loss at 50 MHz (the lowest frequency we measure) will be worst than at 7 GHz. Whilst this is the exception, rather than the rule, it does indicate that the DC resistance is not so important.

We do now provide measured S-parameters on the opens, shorts and loads so if one has a VNA that allows the S-parameters to be imported, that data could be used.

Someone asked why when they loaded the floppy disk into an 8753, it indicated a maximum frequency of 3 GHz. That is because we find it quicker to create the files on our 8753ES, which is a 3 GHz unit. The actual coefficients are valid to 7 GHz, and a 6 GHz 8753 will calibrate up to 6 GHz. I have never seen one, but there's some information in one of the data sheets from Agilent that suggest there were versions of the 8753 up to 8.5 GHz. I guess these were "specials". If you did have such a unit, a standard SMA kit would not calibrate it beyond 7 GHz.

We can do SMA kits up to 12 GHz now, but they are more expensive, as very few loads have suitable performance at 12 GHz.

I strongly suggest if anyone does have any queries about how to use our kits, it is better to ask us, than ask on a forum like this. I see people asking if the SMA_M_M file is the right one to use in a particular circumstance.

Oh, and one more thing. The Agilent N9912A FieldFox is a very poor VNA. If you ask Agilent, they will tell you it is only designed for basic measurement. It can not even measure the phase of S21 - only the magnitude. When I first played with an N9912A I thought there was a fault on it, as the dynamic range is only about 50 dB. I'm told that is normal. The N9912A really is a brain-dead instrument.

Dr. David Kirkby,
Kirkby Microwave.

[drkirkby]

Some sellers from China on ebay sell kits with PNA trademark.

They wrote in description:
"Accurate electrical models are needed of the opens and short circuits. A number of eBay sellers sell low-cost calibration kits, but these are effectively useless, as they don't have the mathematical models of the opens and shorts necessary to load the constants into a VNA, and we doubt they have optimal phase difference between open and short. This kit is different, in that it is designed, rather than just assembled. With this calibration kit, there is a support page with information on what coefficients to use for Agilent/HP,R/S,ADVANTEST,ANRITSU...".

Note those words are a direct copy from the Kirkby Microwave auctions!  They did not copy the link to our website! The Chinese have just copied what we wrote. I did think of complaining to eBay, but decided not to bother.

Dr. David Kirkby.

[drkirkby]

I could need some advice on what i am doing wrong with my Kirkby cal-kit.
(Please appologise stupid questions, since VNA measurements are totally new to me )

My setup consists of:
-) HP 8753E VNA (6 GHz)
-) Kirkby 85033 SMA calkit
-) 50 cm RG316D koax

I followed the instructions on the Kirkby 8753 page here: http://www.kirkbymicrowave.co.uk/support/VNAs/HP-Agilent-Keysight/8753/

On the floppy disk there are 3 files (SMA, SMA_M_M, SMA_F_F).
I selected "SMA" and loaded it with "RECALL STATE".
Is that the correct file to choose?

After that, the frequency range is automatically changed to 300 kHz ... 3 GHz.
Shouldn't it be set to the full range of 30 kHz - 6 GHz of my VNA?

If you need advice, you should contact us!

The reason the span is set from 300 kHz to 3 GHz on your 6 GHz 8753ES is that we actually create the files to go on the floppy disk using an 3 GHz 8753ES, which works from 300 kHz to 3 GHz. This is less tedious, and less error prone than trying to use any Agilent software, like the Cal Kit Manager 2.1 software. The coefficients are valid to 7 GHz though. If you load the SMA kit, and look at the frequency range of any of the standards, you will find the minimum frequency is 0 GHz and the maximum 7 GHz.

Someone else has already pointed out some other errors you made, so I will not repeat them, but I thought I'd explain why you see a span of 300 kHz to 3 GHz. You will just need to set the start and stop frequencies to what you want.

Dr. David Kirkby.

[TheSteve]

It is great to have you here David.

I have been playing with more VNA's. I have something that is full 2 port now to 6 GHz. With a full 2 port OSLT cal my Mini Circuits VAT10+ looks quite reasonable when compared to the datasheet. Much better then the results my N9912A was providing.

[sixtimesseven]

Sorry for hacking this thread, but what do you guys think about the possible cal kit options for a HP3577A with S-Parameter test set (100khz-200Mhz, 50Ohms)?

The Kirkby Kit with 7GHz seems to be overkill for me at the moment

Thanks

[TheSteve]

DIY is totally possible but he will still need someone to measure them for him.

[drkirkby]

I could need some advice on what i am doing wrong with my Kirkby cal-kit.
(Please appologise stupid questions, since VNA measurements are totally new to me

If you need advice on the use of our kit, please contact us, but I will answer them best I can here.

On the floppy disk there are 3 files (SMA, SMA_M_M, SMA_F_F).
I selected "SMA" and loaded it with "RECALL STATE".
Is that the correct file to choose?

After that, the frequency range is automatically changed to 300 kHz ... 3 GHz.
Shouldn't it be set to the full range of 30 kHz - 6 GHz of my VNA?

I put the answer why your VNA is set from 300 kHz to 3 GHz in our FAQ.

https://www.kirkbymicrowave.co.uk/Support/FAQ/Quick-questions-about-vector-network-analyzers/#Frequency-is-3-GHz-on-6-GHz-VNA

The short answer is we write the coefficients using a 3 GHz 8753ES. The calibration standards are not measured on the 3 GHz VNA, but using the 3 GHz 8753ES it is just a convenient way for us to write the floppy disks. The standards are measured on  a 20 GHz VNA (8720D) with a 26.5 GHz calibration kit (85052B).

As for the correct file to load (SMA, SMA_F_F or SMA_M_M), it does not matter for 1-port calibrations. For 2-port calibrations it does, as the different files each have a different delay on the thru standard. An explanation is at

https://www.kirkbymicrowave.co.uk/Support/FAQ/What-file-should-I-load-from-the-floppy-disk-into-the-VNA/

P.S.:
Another thing that irritates me is that the 50 Ohm references are 49,5 and 51,1 Ohm instead of "50,0".
I would expect to get better quality references for that price?

A low frequency (DC)  resistance of 49.5 ohms, the VSWR is 50/49/5=1.0101 (return loss 46.0 dB), which is no worst than the return loss of the loads in the Keysight 85052B  (US$ 12,956  in July 2018) 3.5 mm calibration kit.  A DC resistance 49.5 ohms would not be out of specification in that kit.

Now 51.1 ohms is a VSWR of  51.1/50 = 1.022 (return loss 39.3 dB), which is within the specification of our kit (> 32 dB return loss to 7 GHz), but not the 85052B, but then our kits do not cost almost $13,000. (For the 8, 10 and 12 GHz version of the SMA kit, the return loss of the loads will be at least 32 dB at 8, 10 or 12 GHz respectively)

What we care about is the RF performance of the kit. If the loads were made from wirewound resistors, as standard resistors usually are, the DC resistance could be made very close to 50 ohms, but the RF performance would be appalling.

You might note there's no specification for the DC resistance of RF terminations in any of the Keysight calibration kits, nor in the loads from companies like Minicircuits. It really is irrelevant.

We have a HP 3457A 6.5 digit multi-meter here, which we had calibrated by Keysight only a month or two ago. It supports 4-wire resistance measurements. We could easily measure the DC resistance of the loads, but tests performed in the past indicate there's no correlation between DC resistance and microwave performance. So worrying about DC resistance is pointless. Loads are screened based on their RF performance, not the DC resistance.

Dr. David Kirkby
Kirkby Microwave Ltd.

[G0HZU]

What we care about is the RF performance of the kit. If the loads were made from wirewound resistors, as standard resistors usually are, the DC resistance could be made very close to 50 ohms, but the RF performance would be appalling.

You might note there's no specification for the DC resistance of RF terminations in any of the Keysight calibration kits, nor in the loads from companies like Minicircuits. It really is irrelevant.

Hmm... I think you are trying to justify your use of a low cost wideband SMA 50R 'termination' here rather than the type of precision 50R load found in an expensive commercial RF cal kit. I don't 'test' HP/Agilent/Keysight cal kits very often but usually the performance/agreement between our kits at work (at LF) is excellent and this implies the LF resistance must be very close to 50R.

When I made my homebrew SMA cal kit many years ago I tried to get as close to 50R on a DMM (at DC) as possible so spent some time selecting the best resistors to use. This cal kit always gives extremely good agreement at LF with a proper cal kit from Agilent so I think all the works Agilent/Keysight kits (and my homebrew kit) are very close to 50R at LF (eg at test frequencies <10MHz). I'd like to think they are all within about 0.1 to 0.2R of 50R at LF but I've never tried this test to prove it. I also have a Suhner (18GHz) SMA load here that I selected from a batch to be the closest to 50R (on a DMM) I could find. I sometimes use this instead of my homebrew SMA load.At work I think we have about a dozen 85033E 3.5mm kits and various 13-26GHz Ecal modules and a couple of 85032F N kits. However, I haven't tested all of them for LF performance. I have checked a few of them though...

So a cal kit load that measures 51.1R at LF is an 'unusual case' in my experience.

The Ecal kits will be corrected in a lookup table to be very close to 50R at LF although the only Ecal module I've tested extensively is my own one here at home. It usually agrees within a few tens of milliohms of 50R at LF compared to my homebrew cal kit and the hand selected Suhner 18GHz SMA load. The Suhner load and my homebrew SMA load have been tested in the past against an 85033E cal kit and I got similar agreement. So I generally take it for granted that a commercial cal kit will be very close to 50R at LF even though this isn't quoted in the datasheet.

[G0HZU]

Here's an old plot of my Suhner 18GHz SMA load showing the LF performance. This load was hand selected from a large batch of identical Suhner 18GHz loads to be the closest to 50R and I just measured it again this evening (after many years) and it is 49.99R at DC on a Keithley 2000 4W meter.

This plot was taken at work (a few years ago) using an 85033E cal kit and the VNA was a 6GHz 8753ES. It shows very close agreement with 50R all the way up to 200MHz. So this implies the 8753ES/85033E can measure a real 50R quite accurately.

On my VNA here at home (using my N4431B-60006 Ecal to calibrate the VNA)  the same Suhner 18GHz load measures 49.94R at LF. So pretty close to 50R again!
 I also dug out another Suhner 18GHz load and it measured 49.87R on the VNA (with the same Ecal calibration) across about 2-10MHz. On the 4W Keithley this second Suhner load measured 49.89R at DC. So good agreement again I think.

[G0HZU]

FWIW at work today I tried measuring a number of 50R loads from our Agilent N and 3.5mm cal kits. I used a decent Agilent 4 wire bench DMM (calibrated at Agilent in April this year) and the 'worst' load I found in the group test measured 49.97 ohm at DC. The rest were slightly better than this but all very close to 50 ohms as expected.
I also tried calibrating an E5071 VNA with one kit and measuring the resistance of the other Agilent cal kit loads at low frequencies. They all measured within a few tens of milliohms of 50.00 ohm at a few MHz and some were within just a few milliohms of the reference load used for the VNA calibration. Some of our Agilent cal kits must be >15years old so this is a good result I think.

[hendorog]

Here's an old plot of my Suhner 18GHz SMA load showing the LF performance. This load was hand selected from a large batch of identical Suhner 18GHz loads to be the closest to 50R and I just measured it again this evening (after many years) and it is 49.99R at DC on a Keithley 2000 4W meter.

This plot was taken at work (a few years ago) using an 85033E cal kit and the VNA was a 6GHz 8753ES. It shows very close agreement with 50R all the way up to 200MHz. So this implies the 8753ES/85033E can measure a real 50R quite accurately.

On my VNA here at home (using my N4431B-60006 Ecal to calibrate the VNA)  the same Suhner 18GHz load measures 49.94R at LF. So pretty close to 50R again!
 I also dug out another Suhner 18GHz load and it measured 49.87R on the VNA (with the same Ecal calibration) across about 2-10MHz. On the 4W Keithley this second Suhner load measured 49.89R at DC. So good agreement again I think.

This is what I did too when I got my 8753 up and running so I'm glad to hear it makes some sense.

In addition I entered the measured resistance into the VNA cal standard table, instead of just 'Load'.
IIRC that helped the VNA measure the other loads more precisely.

Interestingly the SDR Kits guys use this DC measurement to characterise their loads in their low cost cal kits - you get a little label in the box with the result hand written on it.

Of course this doesn't necessarily mean that it will still be 50 ohms at 6GHz, which is I think what Dr Kirkby was saying.

[G0HZU]

I think you also have to correct for the time delay in the SMA load I'm not sure how muddy things get if you were to try and do some precision work with a full 2 port VNA with an uncorrected 51.1R SMA load as the calibration reference. I wouldn't want to use it like that even if it would be OK for many casual tasks. I'm not sure I want to waste the time working out how much it will degrade certain types of measurement. I'd rather source a decent load and keep it simple

[Mechatrommer]

Of course this doesn't necessarily mean that it will still be 50 ohms at 6GHz, which is I think what Dr Kirkby was saying.

what i suspect he's saying is, albeit the inaccuracy of the load at DC and to some extend at RF, the error should be close to insignificant (for normal Joe's purposes).. but to know exactly, one has to make a comparison test requiring both the inaccurate load and super accurate $$$ load in hand, calibrate VNA with both and make few practical measurements to see how the measurements will be far off. i cant do the test since i cant justifiably afford $2,000 and beyond calibration kit let alone $13,000. so i have to anticipate more ripply measurement than it should. my 2cnts.

[Bud]

You do not have to do that elaborate test, and it will not be conclusive anyways because the best commercial calibration loads still have errors. All you need is to know your calibration load's Retun Loss within the frequency range of interest, then you can reference existing charts / tables that show +/- error of measurements for a given Return Loss figure. You can pull these charts from the Internet.
Now the problem is to know your cal load RL, that you need to measure against a decent calibration kit on a VNA or from the manufacturer data.

[hendorog]

I think you also have to correct for the time delay in the SMA load I'm not sure how muddy things get if you were to try and do some precision work with a full 2 port VNA with an uncorrected 51.1R SMA load as the calibration reference. I wouldn't want to use it like that even if it would be OK for many casual tasks. I'm not sure I want to waste the time working out how much it will degrade certain types of measurement. I'd rather source a decent load and keep it simple

Yes - but consider the inverse case, which is what I was trying to do:

Can you make your measurements _more_ accurate by entering the exact resistance of the load?

I believe that the assumption that a load is exactly 50 ohms is a bit of an historical relic.

So taking the argument one step further, why not do away with the model altogether? Since a better model of the load is an S-param measurement of the load, then we could just use that instead and calibrate on the PC.

[G0HZU]

Can you make your measurements _more_ accurate by entering the exact resistance of the load?

I guess so, assuming the VNA supports this feature. However, I'm not sure how it handles both the resistance and the Zo of the delay in the standard. It just seems to me to be a bad idea to have to resort to this. Pretty much all of my VNA experience is with official cal kits or my own homemade ones and they are probably all within 0.05R of 50R at LF so I don't have much experience of fudging the 50R standard.

I chimed in because DrKirkby seemed to imply (to his customer) that the DC/LF resistance is irrelevant and can be ignored in the cal kit file. But I think the real/imaginary parts of the load at lower frequencies needs to be very close to a pure/accurate 50R resistor if you want to exploit the potential of a decent VNA when measuring impedance or if making a decent 2 port model at these lower frequencies. In my experience. a good VNA and a decent cal kit can deliver really good accuracy for impedances in the 5R to 2000R range at low frequencies. Much better than the uncertainty stats given for a default VNA with fairly grim uncertainty factors for the various contributors to overall uncertainty.

An uncorrected 51.1R resistor would give an additional error of over 2% for this when measuring typical LCR parts at low frequencies. I think the error will be worse for some types of measurement. Also, even a decent VNA will typically struggle a bit when measuring the Q of inductors on the HF bands but in my experience the results can be quite good with a good VNA and cal kit as long as the Q is in the order of 50-100 and the inductance is in a typical/sensible range. I'd expect an uncorrected 51.1R load to magnify the uncertainty of a Q measurement quite a bit.

Up at microwave frequencies the impact is going to be less but down at a few MHz to maybe 100MHz I think a VNA can deliver really good performance as long as the 50R load has a decent reflection coefficient down at these frequencies.

Bud: Now the problem is to know your cal load RL, that you need to measure against a decent calibration kit on a VNA or from the manufacturer data.

Agreed, but based on the physics of the component I'd expect the return loss of a decent 50R cal kit load to be  in the ballpark of 70dB at a few MHz. Even a cheapo homebrew load made from selected chip resistors will be close to this at a few MHz if selected to have a load resistance very close to 50.00R at LF. Obviously, this would have to be verified as you say, but I don't think it's rocket science to make a reasonable 50R SMA load.

A couple of years ago I posted up a measurement of my homebrew SMA load. It was quite a few years old by then and quite worn but it still managed 40dB return loss by 3GHz and the LF resistance was within maybe 20 milliohms of 50R. This was made using a pair of hand selected 100R chip resistors mounted on a decent quality SMA PCB end launcher. I chose resistors and an end launcher type that gave very good results across this range of LF through 3GHz.

The plot below of my homebrew load was taken with an ENA VNA calibrated by a decent 13GHz Ecal module. It isn't quite this good today because it has been used a lot in the last two years and is showing increasing signs of wear.

[G0HZU]

In terms of a typical application for that old homebrew load I used it as part of my homebrew cal kit to design and verify a precision (SMA connector based) resistive splitter a few years ago. This was needed to help me service an ancient HP8405A vector voltmeter and verify the performance.

I used the cal kit to check out various SMD chip resistors (when piggybacked) to make up the resistors in the splitter. I wanted to match the performance of a decent commercial splitter up to the 1GHz range of the HP8405A. I needed ultra low port VSWR and sub 1degree phase balance and very good amplitude tracking on both arms of the splitter. To get this I needed resistances in each arm that were very close to 50.00R across LF to as close as 1GHz as possible. So I needed a decent cal kit for this.

I used my old/lowly HP8714B VNA and the homebrew cal kit to do all this because that is all I had back then. When I got the 4 port ENA VNA and the 4 port Ecal module I remeasured it formally as a 3 port device (to 3GHz!) and you can see the results below.

It was very impressive indeed for port VSWR, phase tracking and amplitude tracking across LF to 1GHz as you can see in the plots below. It degrades a bit by 3GHz but is still about as good as a typical commercial precision splitter up to 1GHz

[hendorog]

In terms of a typical application for that old homebrew load I used it as part of my homebrew cal kit to design and verify a precision (SMA connector based) resistive splitter a few years ago. This was needed to help me service an ancient HP8405A vector voltmeter and verify the performance.

I used the cal kit to check out various SMD chip resistors (when piggybacked) to make up the resistors in the splitter. I wanted to match the performance of a decent commercial splitter up to the 1GHz range of the HP8405A. I needed ultra low port VSWR and sub 1degree phase balance and very good amplitude tracking on both arms of the splitter. To get this I needed resistances in each arm that were very close to 50.00R across LF to as close as 1GHz as possible. So I needed a decent cal kit for this.

I used my old/lowly HP8714B VNA and the homebrew cal kit to do all this because that is all I had back then. When I got the 4 port ENA VNA and the 4 port Ecal module I remeasured it formally as a 3 port device (to 3GHz!) and you can see the results below.

It was very impressive indeed for port VSWR, phase tracking and amplitude tracking across LF to 1GHz as you can see in the plots below. It degrades a bit by 3GHz but is still about as good as a typical commercial precision splitter up to 1GHz

That is impressive - I made a splitter as well but as I recall it is not even close to that performance.

As a matter of interest, what does your eCal measure itself as? I am curious how close that is performance wise to a physical cal kit.

[G0HZU]

That is impressive - I made a splitter as well but as I recall it is not even close to that performance.

Thanks. A fair bit of design effort went into it and it would have been far more sensible to just buy a splitter 

I used the homebrew cal kit, the HP8714B VNA and a suitable test fixture to measure the best resistor type to help me reverse engineer a decent model of the chosen chip resistor. I then designed and simulated a (0.02" Rogers 4003C) circuit board using Genesys and Sonnet EM with my resistor models to try and get the lowest port VSWR across at least 1GHz. I used a precision PCB mill with decent end mill tools to make the PCB I had a good result with the very first PCB design and I think there was an element of luck in the phase matching. I didn't expect it to be that good and I only needed a few scrapes with a scalpel in the right places on the PCB to get the phase that flat. This was taken with the SMA connectors done up with a torque spanner as the phase shifts with connector tightness in the GHz region as you would expect.

I'm not sure it's possible for me to directly measure the load in the Ecal because it is all controlled by the VNA via USB. I think the Ecal is at its very best above a few hundred MHz. By that I mean probably better than a typical 85033E mechanical kit. Below this (especially down at a few MHz) I think the mechanical kits (and my homebrew kit) are marginally better for both 1 port and 2 port measurements. But that is based on my experience of using all three.

When I designed my SMA cal kit I went to similar (nerdy) lengths to verify I had the delays and corrections in my user cal kit as close as possible. I developed various passive 2 port test fixtures here that can easily show even slight errors in the cal kit's user file when the VNA test fixture s2p data is post processed on a computer. This work really highlighted to me why the 85033E cal kits and the Ecal modules are so expensive and I'm impressed how consistent all our 85033E cal kits are at work.

[Andrey_irk]

G0HZU,
If you are concerned about the RL accuracy when measuring low RL values then you should look at the residual directivity value. These are in order of 40-50dB for the most of commercial kits. Keysight, for instance, provide these values for every VNA calibrated with every their kit or ECal (as long as I remmember).
51.1 Ohms gives you these 40dB of RL  at low frequencies and the high frequency values will depend on the parasitics (which are usually much more difficult to make small). And it won't help you if you get a 49.999Ohm resistor as the HF performance won't be affected.
If you measure say 20dB RL DUT with with a VNA calibrated to have 40dB of directivity - it will give you an error of about 0.8-0.9dB, which is pretty good.
At work I have an SMP cal kit, which has a Load standard with (according to specs) 50+-0.25Ohms, but only 25dB RL beyond 4 GHz. And I'm not aware of any commercial kits with better specs for this type of connectors. So, with our 20dB RL device this kit gives us an error of 4-7dB (!).

P.S. I'm not a guru at this, so hope Dr. Kirkby will correct me if I'm wrong.
P.P.S. Is it possible to invite Joel Dunsmore here?

[G0HZU]

I think you are missing the point I'm trying to make. DrKirkby said the resistance at LF is irrelevant.
Suppose I wanted to test some 100R resistors across temperature to see how accurate and consistent the resistance was over maybe 0-200MHz.
If I take DrKirkby's advice then it won't matter which Kirkby cal kit I use to measure the complex impedance as long as they all have the same 39dB return loss. So for each test at each temperature I could select a Kirkby cal kit at random (from a box of several) to use and not care if the real part of the calibration load was 48.9R or 51.1R between the kits.

But I'd see errors of > +/-2% for the real part of the impedance even for this simple test at a fixed temperature. I could correct the resistance in the user file for each cal kit  to minimise the error but DrKirkby implies all this is irrelevant.

[Bud]

Can't you enter the reference impedance value in the VNA settings and call it a day. Not all VNAs may have that setting changeable though.

[G0HZU]

I guess so but I don't know if a VNA would interpret this as a 51.1R resistor with a 50R Zo delay or as a 51.1R resistor with a 51.1R delay.

So it might cause subtle problems when being used up at higher frequencies. I had a rummage at work to try and find an SMA 50R termination as poor as 51.1R at LF but couldn't find one. The other option is to use a homebrew SMA load up to a couple of GHz that is selected to give close to 50R and then use the wideband 51.1R load up at higher frequencies if critical measurements are being done in either case.

[Bud]

VNAs should not assume things, unless it is a Chinese one where anything is possible. I'd think the math certainly has delay variable in it but any delays should be specified by the user explicitly.