开云体育

The NanoVNA firmware (as I understand it) assumes that the SMA Short, Open, and Load standards are ideal, with the single exception of defining the Open to have a C0 50 femtofarads.

I was wondering -- is this a sufficient definition for these standards, given the NanoVNA's frequency limit of 900 MHz? Or should we also be including values for C1-C3, L0-L3, Offset Loss and Offset Zo?

So I wrote some Matlab code to calculate a standard's Reflection Coefficient (Gamma) based on Keysight's "Full" model (using C0-C3, L0-L3, Offset Loss, Offset Delay and Offset Zo), and compared this to a simplified model in which all of these values were set to 0 except for C0 and Offset Delay.

I've attached a table of results.

My conclusion is that the simplified model (C1-C3, L0-L3, and Loss are set to 0, and Zo is set to 50) is perfectly adequate for use with the NanoVNA.

But my math could be wrong. Do my results seem correct?

(A detailed explanation of what I've done is in the top post of this blog: )

Thanks,

- Jeff, k6jca

P.S. I did NOT zero-out Offset Delay in my simplified model (I believe (but have not verified) that this delay IS assumed to be zero in the NanoVNA firmware). If the actual Offset Delays for the Open and Short standards are defined by their supplier to be non-zero, then the net effect of zeroing out these delays is to simply move the reference plane of each -- i.e. rotate the Open and Short's reflection coefficients around the Smith Chart circle. However, note that zeroing out the two delays could become an issue if the Short's delay is NOT equal the Open's delay -- you would then lose this difference in delays (should this delta be large enough to be important)).

On Sat, 28 Sep 2019 at 16:26, Jeff Anderson <jca1955@...> wrote:

The NanoVNA firmware (as I understand it) assumes that the SMA Short,
Open, and Load standards are ideal, with the single exception of defining
the Open to have a C0 50 femtofarads.

The definition of calibration kits in the NanoVNA seems woefully

inadequate. I don’t know what irritates me the most

* Screen too small
* No proper support for calibration kits.

I was wondering -- is this a sufficient definition for these standards,

given the NanoVNA's frequency limit of 900 MHz? Or should we also be
including values for C1-C3, L0-L3, Offset Loss and Offset Zo?

I don’t think your simple model is really suitable for the following
reasons

1) The variation of C with homemade standards is likely to exceed that of
commercial standards - this is from experience measuring them.

2) The inductance of shorts is likely to be more with homemade standards
than commercial ones - again this is based on experience measuring them.

3) People may well want to make measurements in a 75 ohm system.

4) it is possible to improve upon the accuracy of loads at low frequencies
by using a DC resistance measurement.

5) In the case of a female N, a simple standard can be made by just leaving
the connector open. This will create a higher impedance transmission line
than 50 ohms as the centre conductor sits in a cylindrical section with a
greater diameter than when its mated.

6) The loss of homemade standards is likely to be greater than commercial
ones from Keysight - again this is based on actual measurements I have
performed.

So I wrote some Matlab code to calculate a standard's Reflection

Coefficient (Gamma) based on Keysight's "Full" model (using C0-C3, L0-L3,
Offset Loss, Offset Delay and Offset Zo), and compared this to a simplified
model in which all of these values were set to 0 except for C0 and Offset
Delay.

I've attached a table of results.

My conclusion is that the simplified model (C1-C3, L0-L3, and Loss are set
to 0, and Zo is set to 50) is perfectly adequate for use with the NanoVNA.

I disagree with that, for the reasons stated above.

I have seen something written by Dr. Joel Dunsmore saying that depending on
the accuracy requirements, one should consider the variation of capacitance
with frequency above about 200 MHz. However, I see your notes that the
phase variation up to 1500 MHz is smaller than the uncertainty in the
calibration standards. I can’t square that circle.

But my math could be wrong. Do my results seem correct?

I have not checked your maths, but based on my experience using commercial
calibration kits from Keysight, I would accept what you are saying is
probably reasonable up to 900 MHz. I have here

85032B - 6 GHz N
85038A - 7.5 GHz 7-16
85939B - 3 GHz, 75 ohm F
85050B - 18 GHz APC7
85052B - 26.5 GHz 3.5 mm
85054B - 18 GHz N

plus an Ecal and any I have forgotten!

If you want an even simpler model for commercial calibration kits you could
do without C0, *as long as the offset delay of the open is suitably
increased to compensate for this*. The problem would be people would not
know how much to increase the delay, although it can be worked out. it
would be difficult for each calibration kit.

(A detailed explanation of what I've done is in the top post of this blog:

)

Thanks,

- Jeff, k6jca

P.S. However, note that zeroing out the two delays could become an issue
if the Short's delay is NOT equal the Open's delay

The short delay should *always* be longer than the open delay, which makes
a mockery of all these web pages that go to great lengths to tell you to
make the delay of the short as little as possible. People dreaming about
zero offset shorts don’t understand the subtleties of calibration kit
design. In the calibration kits my company produces, the delay of the short
is chosen for optimal performance, which is *never* zero. If you look at
the Keysight kits you will find that *most* of the kits have a longer delay
on the short than the open. You will see this more with the high-end kits,
such as the 85052B & 85054B kits I mentioned earlier.

Dave

--
Dr. David Kirkby,
Kirkby Microwave Ltd,
drkirkby@...

Telephone 01621-680100./ +44 1621 680100

Registered in England & Wales, company number 08914892.
Registered office:
Stokes Hall Lodge, Burnham Rd, Althorne, Chelmsford, Essex, CM3 6DT, United
Kingdom

On Sat, Sep 28, 2019 at 03:49 PM, Dr. David Kirkby from Kirkby Microwave Ltd wrote:

I don’t think your simple model is really suitable for the following
reasons

1) The variation of C with homemade standards is likely to exceed that of
commercial standards - this is from experience measuring them.

2) The inductance of shorts is likely to be more with homemade standards
than commercial ones - again this is based on experience measuring them.

3) People may well want to make measurements in a 75 ohm system.

4) it is possible to improve upon the accuracy of loads at low frequencies
by using a DC resistance measurement.

5) In the case of a female N, a simple standard can be made by just leaving
the connector open. This will create a higher impedance transmission line
than 50 ohms as the centre conductor sits in a cylindrical section with a
greater diameter than when its mated.

6) The loss of homemade standards is likely to be greater than commercial
ones from Keysight - again this is based on actual measurements I have
performed.

Dave, thanks for taking the time to reply. I appreciate your comments, and I agree with you on all these points -- but I wasn't really concerning myself with homemade standards, which I assume are almost always uncharacterized.

Instead (and I should have made this clearer), I was wondering what the impact was of the Capacitance and Inductance terms of characterized Open and Short standards. Ditto for their Delay, Loss and Offset Zo terms.

For the HP Open and Short standards I looked at, Delay has the largest impact on Gamma (which was the reason I never set this term to 0 in my calculations), followed by the Open's C0 term. The other terms have an effect, but that effect is much smaller than the effect of Delay or C0.

As for homemade standards, probably the best one can do to characterize them is verify that the load is as close to 50 ohms as possible (using a 4-terminal ohms measurement) and determine the Delays of the Short and Open using an *already-calibrated* VNA (although I did come across a web page where the author actually derived C0-C3 (and perhaps L0-L3? I don't recall any longer).

I see your notes that the
phase variation up to 1500 MHz is smaller than the uncertainty in the
calibration standards. I can’t square that circle.

I was quite surprised to see this result, which is one reason why I originally posted the results here with my question wondering if there were a math error.

Running through the numbers again, they look good to me. Here's an example of the effect of the capacitance terms of an 85032F Male-N Open at 900 MHz. Note that C0-C3 are spec'd as follows (all values in Farads):

C0 = 89.939e-15;
C1 = 2536.8e-27;
C2 = -264.99e-36;
C3 = 13.4e-45;

First, calculating the Open's Gamma using C0-C4. The equations are:

divisor = (2*pi*f * (C0 + C1*f + C2*f^2 + C3*f^3))
Zopen = -j / divisor
GammaOpen = ((Zopen/50) - 1) / ((Zopen/50) + 1)

If the frequency (f) is 900MHz, then GammaOpen equals 0.9962 - j0.0877 (i.e. magnitude of 1, angle of -5.0292 degrees).

Let's now calculate Gamma for the C0-only model. The equations now are:

divisor = (2*pi*f * C0)
Zopen = -j / divisor
GammaOpen = ((Zopen/50) - 1) / ((Zopen/50) + 1)

Gamma now is 0.9964 - j0.08461 (i.e. magnitude of 1, angle of -4.8538 degrees).

From this I draw a couple of conclusions:

1. At 900 MHz this standard's C0 has a significant impact on the angle of Gamma (about -4.9 degrees).

2. Adding in the additional C1-C3 terms only changes the Open's angle of Gamma by about 0.18 degrees, which is significantly smaller than this Standard's spec'd "Deviation from Nominal Phase" of +/- 0.65 degrees (from DC to 3 GHz).

(By the way -- should any one else like to verify the results, the equations and C0-C3 terms are above.)

Thanks again for your comments and insights,

- Jeff, k6jca

Jeff Anderson

On Sat, Sep 28, 2019 at 08:26 Jeff Anderson wrote:

The NanoVNA firmware (as I understand it) assumes that the SMA Short, Open, and Load standards are ideal, with the single exception of defining the >Open to have a C0 50 femtofarads.
My conclusion is that the simplified model (C1-C3, L0-L3, and Loss are set to 0, and Zo is set to 50) is perfectly adequate for use with the NanoVNA.
I did NOT zero-out Offset Delay in my simplified model .

Jeff, if I understand what you wrote, you would like the NanoVNA to have the ability to include custom standard characterization by inputting C0 and Offset Delay for both the open and short, while assuming the reference load is exactly 50 ohms. Per your calculations, including higher order terms only results in a trivial (for most uses) correction.


On Sat, Sep 28, 2019 at 03:49 PM, Dr. David Kirkby from Kirkby Microwave Ltd wrote:

I don’t think your simple model is really suitable for the following
reasons

1) The variation of C with homemade standards is likely to exceed that of
commercial standards - this is from experience measuring them.

2) The inductance of shorts is likely to be more with homemade standards
than commercial ones - again this is based on experience measuring them.

3) People may well want to make measurements in a 75 ohm system.

4) it is possible to improve upon the accuracy of loads at low frequencies
by using a DC resistance measurement.

5) In the case of a female N, a simple standard can be made by just leaving
the connector open. This will create a higher impedance transmission line
than 50 ohms as the centre conductor sits in a cylindrical section with a
greater diameter than when its mated.

6) The loss of homemade standards is likely to be greater than commercial
ones from Keysight - again this is based on actual measurements I have
performed.

Dr. Kirkby, if I understand your post, and objections to Dave Anderson's "Very Simple Characterization Model", You are against Mr. Anderson's idea because it doesn't account for the possibility that inductance, capacitance and loss terms are likely to be significant in home-made standards and a desire to have an option to change the reference resistance to account for actual load resistance, or to use the VNA at something other than 50 ohms, but you agree with the idea that Offset Delay should be an input option.

Dr. Kirkby, I too like the idea of being able to reference the NanoVNA with 75 ohms or measureing a home-made load with a 4 terminal system and getting a more accurate result. But I'm not sure if your other objections are valid (1) (2) (6), because most home-made loads won't be characterized anyway. The operator will take the answer he gets, publish it in QST or RadCom, and move on.

I can see the merit of Mr. Anderson's idea when using characterized standards. This moves the NanoVNA a little bit farther away from MFJ toward Rohde and Schwarz, while still keeping things simple. Of course this brings up the operations of data entry and storage; which starts taking away from the original concept of turn it on, do a simple 1-2-3 cal, and use it, so I think that option should be 1st choice in the software.

Mr. Anderson and Dr. Kirkby, if I have misstated or made a bad assumption of your posts I apoligize, and please correct me; this is a new area of electronics for me and I'm sure I can (and will) be educated by the group.

Jim McEwen, KA6TPR

On Sun, 29 Sep 2019 at 02:46, Jeff Anderson <jca1955@...> wrote:

On Sat, Sep 28, 2019 at 03:49 PM, Dr. David Kirkby from Kirkby Microwave
Ltd wrote:

I don’t think your simple model is really suitable for the following
reasons

Dave, thanks for taking the time to reply. I appreciate your comments,
and I agree with you on all these points -- but I wasn't really concerning
myself with homemade standards, which I assume are almost always
uncharacterised.

Generally speaking I would agree with you, but some hams will make their
own standards and get them measured by someone with a decent VNA - either
one they have access to at work, or one they know someone else who can do
this.

There are some Anritsu open/short/load devices which are designed for the
Anritsu sitemaster, which at least in some formats is designed for *scalar*
network analysers. Anritsu don't publish coefficients, but I have computed
the coefficients for people so they can use them with VNAs. I can’t recall
if C1, C2 and C3 have a greater effect than on the Keysight kits.

*I believe that your simplified model of the Keysight one, using just C0,
is a good compromise between complexity and accuracy. Congratulations on
that work. *

If the NanoVNA software only allowed entry of C0, anyone could *very easily
change the value entered into the NanoVNA to improve the accuracy even
more. *I can think of several ways to do this, all of which would give some
improvement for little extra work. One method would be even better, but is
more tricky to do.

*1) VERY EASY *
* Determine C at DC, which will be set by just C0 from the calibration kit
data.
* Determine C at 900 MHz using the full model with C0, C1, C2 and C3.
* Enter into the NanoVNA a value of C0 which is the mean of the two
capacitance at DC and 900 MHz.

So if for example C at DC was 80 fF, and C at 900 MHz was 81 fF, one would
enter into the NanoVNA a value for C0 of 80.5.

*2) AGAIN VERY EASY*
* Determine the mean frequency of the NanoVNA, which would be 900/2=450
MHz.
* Determine C at 450 MHz.
* Enter into the NanoVNA the value of C0 corresponding to the value of
capacitance.

So if for example one calculated C was 123 fF at 450 MHz, one would enter
123 for C0.

*3) AGAIN VERY EASY *
* Determine the capacitance at the frequency of interest. In my case that
would be 144 MHz, so I could calculate the capacitance at 144 MHz.
* Enter into the NanoVNA the value at C0 corresponding to the capacitance
at 144 MHz.

*4) A BIT HARDER*
* Find a value of capacitance which results In minimising the maximum
error. This would be a bit more tricky to find, but hardly that difficult,
especially if someone wrote a small program to calculate this value.
* Enter into the NanoVNA the above value.

So I can think of a few improvements which would allow your simplified
model to be used, by slightly adjusting the value entered into the NanoVNA,
with the minimum of effort. If you get a maximum phase error of 0.18
degrees with this 85032F, you could certainly reduce that error - perhaps
down to 0.09 degrees.

For the HP Open and Short standards I looked at, Delay has the largest
impact on Gamma (which was the reason I never set this term to 0 in my
calculations), followed by the Open's C0 term. The other terms have an
effect, but that effect is much smaller than the effect of Delay or C0.

yes, it would be. In fact I believe that you could probably dispense with
C0 if the offset delay was suitably adjusted, to compensate for its loss, *but
it would complicate matters for the user significantly, so not something I
would advise. *

As for homemade standards, probably the best one can do to characterize
them is verify that the load is as close to 50 ohms as possible (using a
4-terminal ohms measurement) and determine the Delays of the Short and Open
using an *already-calibrated* VNA (although I did come across a web page
where the author actually derived C0-C3 (and perhaps L0-L3? I don't recall
any longer).

Kurt Poulsen has I believe done some work characterising homemade
calibration standards using the T-checker, which allows one to see if a
calibration is good or not. Kurt would have to clarify the scope of his
work, but I think it allowed coefficients to be determined. However, Kurt
uses an even more complicated model than the Keysight one, with for example
shunt capacitance values across the load and delays on the load. Those pa

I see your notes that the
phase variation up to 1500 MHz is smaller than the uncertainty in the

m effect of the capacitance terms of an 85032F Male-N Open at 900 MHz.
Mnitude of 1, angle of -5.0292 degrees).

Let's now calculate Gamma for the C0-only model. The equations now are:

divisor = (2*pi*f * C0)
Zopen = -j / divisor
GammaOpen = ((Zopen/50) - 1) / ((Zopen/50) + 1)

Gamma now is 0.9964 - j0.08461 (i.e. magnitude of 1, angle of -4.8538
degrees).

From this I draw a couple of conclusions:

1. At 900 MHz this standard's C0 has a significant impact on the angle of
Gamma (about -4.9 degrees).

Yes, it would.

2. Adding in the additional C1-C3 terms only changes the Open's angle of

Gamma by about 0.18 degrees, which is significantly smaller than this
Standard's spec'd "Deviation from Nominal Phase" of +/- 0.65 degrees (from
DC to 3 GHz).

The 85032F is not one of Keysight’s best kits. In fact if you look at this
document for the National Physical Laboratory (NPL)



you will read NPL found the phase error of the opens to be well out of
specification! Although Agilent updated the coefficients, the design of
the 85032F is not great.

The 85054B & 85054D N kits have a specification of +/- 1.5 degrees to 18
GHz. There is no specification to anything lower, but I would expect the
85054B and 85054D to *significantly* outperform a 85032F as the 85054B and
85054D use precision slotless connectors. However, both the 85054B and the
economy version the 85054D, are both *very* expensive on the used market. I
paid over $4000 for my used 85954B, and I think that was pretty cheap.

It would be interesting to know what errors you get with your simplified
model for the 85050B 18 GHz APC7 calibration kit. That kit is often
available cheaply on the used market, yet has a good specification on phase
at +/- 0.3 degrees to 2 GHz, and it good at 18 GHz too.


(By the way -- should any one else like to verify the results, the

equations and C0-C3 terms are above.)

Thanks again for your comments and insights,

- Jeff, k6jca

G8WRB

--

Dr. David Kirkby,
Kirkby Microwave Ltd,
drkirkby@...

Telephone 01621-680100./ +44 1621 680100

Registered in England & Wales, company number 08914892.
Registered office:
Stokes Hall Lodge, Burnham Rd, Althorne, Chelmsford, Essex, CM3 6DT, United
Kingdom

On Sun, 29 Sep 2019 at 11:43, Dr. David Kirkby <
drkirkby@...> wrote:

Mmost always uncharacterised.



In fact I believe that you could probably dispense with C0 if the offset
delay was suitably adjusted, to compensate for its loss, *but it would
complicate matters for the user significantly, so not something I would
advise. *

I mean if the delay of the open standard was adjusted for the missing value
of C0. But I don’t think doing so would be a good idea. It is almost
guaranteed for people to put an inaccurate value of offset delay.

One thing I would add if implementation of the full model including C0 to
C3 is not much harder than just implementing the simple model with C0, one
might as well do the full model for completeness. It really depends on the
effort required to implement the full model, especially as I believe some
non-Keysight calibration kits would benefit for the other terms.

Dave.
--
Dr. David Kirkby,
Kirkby Microwave Ltd,
drkirkby@...

Telephone 01621-680100./ +44 1621 680100

Registered in England & Wales, company number 08914892.
Registered office:
Stokes Hall Lodge, Burnham Rd, Althorne, Chelmsford, Essex, CM3 6DT, United
Kingdom

Hi Jeff

My humble opinion is like your indications that L and C coefficient are not relevant for the calibration kit delivered with the NanoVNA and in particular as we have no idea if the kit are the same for all deliveries.

However the NanoVNA has already a build in correction for the open in the form of 50fF which is pretty much correct for the CH0 Female SMA left open, so to use the supplied open standard is wrong and is adding further delay.
Basicly the NanoVNA is for me the "engine" and for using other calibration kits the way is to use the NanoVNA-saver where you can enter delays and L and C coefficient IF YOU HAVE THEM and that is not the case for the majority of NanoVNA users for whatever homemade kit they want to use. One must remember to subtract the 50fF from the Open as the NanoVNA is internally pre-compensated by 50fF equal to a one way delay of 2.5ps. This can be verified by standalone calibrating the NanoVNA, using no open adaptor and run a phase s11 track with 1degree/division and then enable the Scale/Electrical delay to twice the one way delay, as we are dealing with a reflection, so 2.5x2=5ps, and the phase trace is horizontal as proof. This is true for frequencies up to 300MHz and above for a fresh calibration, else it is drifting over time above 300MHz du to temperature changes. Remember to set the Electrical delay back to 0ps ?
Until there is a full blown calibration kit definition embedded in the NanoVNA this is the way forward to use the NanoVNA-saver. By a full blown calibration kit definition I mean that also 6/12 term error correction implemented and again my opinion is that would be an overkill for the majority of NanoVNA users. It is far better to focus on how and with simple means to find the needed delays for a homemade kit or e.g. a BNC kit bought from SDR kits where all these data are supplied with the kit.

David is giving a comment the a short always has a longer delay than open, and that can be misunderstood. That is not the caser for the supplied kit for the NanoVNA. I have made a comment on this on this reflector as it is anticipated to be 0ps by design but it has a very small negative value. I did measure the supplied kit based on calibration by my HP 3.5mm kit on another VNA and I will repeat and publish the result for those values to be entered in NanoVNA saver. It would be nice if David had done that instead of lecturing about the way he seem everything.

Long live a pragmatic approach

Kind regard

Kurt





-----Oprindelig meddelelse-----
Fra: [email protected] <[email protected]> P? vegne af Jeff Anderson
Sendt: 29. september 2019 03:47
Til: [email protected]
Emne: Re: [nanovna-users] Cal-Kit Standards' Definitions



On Sat, Sep 28, 2019 at 03:49 PM, Dr. David Kirkby from Kirkby Microwave Ltd wrote:

I don’t think your simple model is really suitable for the following

reasons

1) The variation of C with homemade standards is likely to exceed that

of commercial standards - this is from experience measuring them.

2) The inductance of shorts is likely to be more with homemade

standards than commercial ones - again this is based on experience measuring them.

3) People may well want to make measurements in a 75 ohm system.

4) it is possible to improve upon the accuracy of loads at low

frequencies by using a DC resistance measurement.

5) In the case of a female N, a simple standard can be made by just

leaving the connector open. This will create a higher impedance

transmission line than 50 ohms as the centre conductor sits in a

cylindrical section with a greater diameter than when its mated.

6) The loss of homemade standards is likely to be greater than

commercial ones from Keysight - again this is based on actual

measurements I have performed.

Dave, thanks for taking the time to reply. I appreciate your comments, and I agree with you on all these points -- but I wasn't really concerning myself with homemade standards, which I assume are almost always uncharacterized.



Instead (and I should have made this clearer), I was wondering what the impact was of the Capacitance and Inductance terms of characterized Open and Short standards. Ditto for their Delay, Loss and Offset Zo terms.



For the HP Open and Short standards I looked at, Delay has the largest impact on Gamma (which was the reason I never set this term to 0 in my calculations), followed by the Open's C0 term. The other terms have an effect, but that effect is much smaller than the effect of Delay or C0.



As for homemade standards, probably the best one can do to characterize them is verify that the load is as close to 50 ohms as possible (using a 4-terminal ohms measurement) and determine the Delays of the Short and Open using an *already-calibrated* VNA (although I did come across a web page where the author actually derived C0-C3 (and perhaps L0-L3? I don't recall any longer).

I see your notes that the

phase variation up to 1500 MHz is smaller than the uncertainty in the

calibration standards. I can’t square that circle.

I was quite surprised to see this result, which is one reason why I originally posted the results here with my question wondering if there were a math error.



Running through the numbers again, they look good to me. Here's an example of the effect of the capacitance terms of an 85032F Male-N Open at 900 MHz. Note that C0-C3 are spec'd as follows (all values in Farads):



C0 = 89.939e-15;

C1 = 2536.8e-27;

C2 = -264.99e-36;

C3 = 13.4e-45;



First, calculating the Open's Gamma using C0-C4. The equations are:



divisor = (2*pi*f * (C0 + C1*f + C2*f^2 + C3*f^3))

Zopen = -j / divisor

GammaOpen = ((Zopen/50) - 1) / ((Zopen/50) + 1)



If the frequency (f) is 900MHz, then GammaOpen equals 0.9962 - j0.0877 (i.e. magnitude of 1, angle of -5.0292 degrees).



Let's now calculate Gamma for the C0-only model. The equations now are:



divisor = (2*pi*f * C0)

Zopen = -j / divisor

GammaOpen = ((Zopen/50) - 1) / ((Zopen/50) + 1)



Gamma now is 0.9964 - j0.08461 (i.e. magnitude of 1, angle of -4.8538 degrees).



From this I draw a couple of conclusions:



1. At 900 MHz this standard's C0 has a significant impact on the angle of Gamma (about -4.9 degrees).



2. Adding in the additional C1-C3 terms only changes the Open's angle of Gamma by about 0.18 degrees, which is significantly smaller than this Standard's spec'd "Deviation from Nominal Phase" of +/- 0.65 degrees (from DC to 3 GHz).



(By the way -- should any one else like to verify the results, the equations and C0-C3 terms are above.)



Thanks again for your comments and insights,



- Jeff, k6jca

On Sun, 29 Sep 2019 at 12:16, Kurt Poulsen <kurt@...> wrote:

Hi Jeff

However the NanoVNA has already a build in correction for the open in the

form of 50fF which is pretty much correct for the CH0 Female SMA left open,
so to use the supplied open standard is wrong and is adding further delay.

That is acceptable if your DUT has a male SMA connector, but it is not if
you wish to use any other sort of connector.

Basicly the NanoVNA is for me the "engine" and for using other calibration
kits the way is to use the NanoVNA-saver

That is your intended use Kurt, but it is certainly not how I intend using
the NanoVNA. For my use, which is to use it as an entirely self-contained
unit, being able to define calibration kits in the NanoVNA is important,
whereas using the NanoVNA with external software is less so.

where you can enter delays and L and C coefficient IF YOU HAVE THEM and

that is not the case for the majority of NanoVNA users for whatever
homemade kit they want to use.

You can estimate that based on physical measurements as you well know.

One must remember to subtract the 50fF from the Open as the NanoVNA is

internally pre-compensated by 50fF equal to a one way delay of 2.5ps.

Or better still remove that 50 fF, and let the user define calibration
kits. Let the NanoVNA default to using the calibration kit supplied, but
allow the user to enter a kit(s) with defined parameters.

David is giving a comment the a short always has a longer delay than open,

and that can be misunderstood. That is not the caser for the supplied kit
for the NanoVNA.

I was talking about the vast majority of professional calibration kits.
Clearly Jeff is thinking about the higher quality kits. When I mentioned
that supporting C1, C2, C3, L0, L1, L2 & L3 would be better for homemade
kits, he pointed out that was not his main aim.

I have made a comment on this on this reflector as it is anticipated to be

0ps by design but it has a very small negative value. I did measure the
supplied kit based on calibration by my HP 3.5mm kit on another VNA and I
will repeat and publish the result for those values to be entered in
NanoVNA saver. It would be nice if David had done that instead of lecturing
about the way he seem everything.

Initially people were saying that the NanoVNA assumed ideal standards. Now
I believe it’s a 50 fF open and probably an ideal short. As you say, the
short supplied has a small negative delay.


Long live a pragmatic approach


The correct pragmatic approach depends very much on ones intended use,
which will be either

* *Use the NanoVNA standalone*, requiring support for calibration kits
internally *in* *firmware*. Jeff has convinced me that a delay for the
opens and shorts, as well as C0, is adequate for the HP calibration kits. I
don’t think that’s the case for homemade kits though.

* *Use the NanoVNA as data collection device*, and perform other functions
with external software.

That’s two very different uses, requiring a totally different approach

Kind regard

Kurt

Dave
--
Dr. David Kirkby,
Kirkby Microwave Ltd,
drkirkby@...

Telephone 01621-680100./ +44 1621 680100

Registered in England & Wales, company number 08914892.
Registered office:
Stokes Hall Lodge, Burnham Rd, Althorne, Chelmsford, Essex, CM3 6DT, United
Kingdom

On Sat, Sep 28, 2019 at 11:28 PM, Starsekr wrote:

Jeff, if I understand what you wrote, you would like the NanoVNA to have the
ability to include custom standard characterization by inputting C0 and Offset
Delay for both the open and short, while assuming the reference load is
exactly 50 ohms. Per your calculations, including higher order terms only
results in a trivial (for most uses) correction.

Hi Jim,

You are essentially correct, but I was thinking more of Rune's Nanovna-Saver application (which allows the user to specify C0-C3, L0-L3, and delays for the two) rather than the NanoVNA, itself. (I should have been more explicit).

Anyway, I was wondering how important it was to enter the various C and L terms for the standards, or if, for expediency, most of them could be left as zero. And, at least for the commercial standards I looked at, it would seem that only C0 and Delay have a significant impact.

Best regards,

- Jeff, k6jca

On Sun, Sep 29, 2019 at 03:43 AM, Dr. David Kirkby from Kirkby Microwave Ltd wrote:

The 85032F is not one of Keysight’s best kits.

Hi Dave,

You are probably right. I chose this kit because, of the 50-ohm cal-kits listed by Keysight as supported for the 8753D, it seemed to have the worst higher-order capacitance terms, thus a good choice for testing my theory.

It would be interesting to know what errors you get with your simplified
model for the 85050B 18 GHz APC7 calibration kit. That kit is often
available cheaply on the used market, yet has a good specification on phase
at +/- 0.3 degrees to 2 GHz, and it good at 18 GHz too.

Ask and you shall receive!

First, Keysight's values defining the 85050B open:

C0 = 90.4799e-15;
C1 = 763.303e-27;
C2 = -63.8176-36;
C3 = 6.4337e-45;
Delay_Open = 0;
Loss_Open = 0;
Offset_Zo_Open = 50;

Now, the results of my calculations...

1. Open using C0 - C3 : rho = 1.0000, theta = -4.9381 degrees

2. Open using only C0: rho = 1.0000, theta = -4.8830 degrees
(Note: the angular delta between the two thetas is 0.055 degrees)

3. Open using "full" keysight model: Results same as '1', above. (To be expected because Loss is 0 and Zo is 50, thus the "full" equations ought to collapse down to my "simple" (i.e. C0-C3 only) model).

Best regards,

- Jeff, k6jca

On Sun, Sep 29, 2019 at 04:16 AM, Kurt Poulsen wrote:

Hi Kurt,

Thanks for the reply.

My humble opinion is like your indications that L and C coefficient are not
relevant for the calibration kit delivered with the NanoVNA and in particular
as we have no idea if the kit are the same for all deliveries.

Agreed.

Until there is a full blown calibration kit definition embedded in the NanoVNA
this is the way forward to use the NanoVNA-saver.

Agreed.

again my opinion is that would be an overkill for the majority of NanoVNA
users. It is far better to focus on how and with simple means to find the
needed delays for a homemade kit or e.g. a BNC kit bought from SDR kits where
all these data are supplied with the kit.

Agreed. Describing how to characterize a homemade BNC kit would be a great idea.

I did measure the
supplied kit based on calibration by my HP 3.5mm kit on another VNA and I will
repeat and publish the result for those values to be entered in NanoVNA saver.

Kurt, that would be excellent! I'm looking forward to seeing these values

Best regards,

Jeff, k6jca

Forgot to mention...the 85050B’s calculation were for 1500 MHz (corrected this -- I'd originally typed 900 MHz)

- Jeff

On Sun, 29 Sep 2019 at 07:28, Starsekr via Groups.Io <Starsekr=
[email protected]> wrote:

On Sat, Sep 28, 2019 at 03:49 PM, Dr. David Kirkby from Kirkby Microwave
Ltd wrote:

I don’t think your simple model is really suitable for the following
reasons

1) The variation of C with homemade standards is likely to exceed that of
commercial standards - this is from experience measuring them.

2) The inductance of shorts is likely to be more with homemade standards
than commercial ones - again this is based on experience measuring them.

3) People may well want to make measurements in a 75 ohm system.

4) it is possible to improve upon the accuracy of loads at low

frequencies

by using a DC resistance measurement.

5) In the case of a female N, a simple standard can be made by just

leaving

the connector open. This will create a higher impedance transmission line
than 50 ohms as the centre conductor sits in a cylindrical section with a
greater diameter than when its mated.

6) The loss of homemade standards is likely to be greater than commercial
ones from Keysight - again this is based on actual measurements I have
performed.

Dr. Kirkby, if I understand your post, and objections to Dave Anderson's
"Very Simple Characterization Model", You are against Mr. Anderson's idea
because it doesn't account for the possibility that inductance, capacitance
and loss terms are likely to be significant in home-made standards and a
desire to have an option to change the reference resistance to account for
actual load resistance, or to use the VNA at something other than 50 ohms,
but you agree with the idea that Offset Delay should be an input option.

Yes, essentially

Dr. Kirkby, I too like the idea of being able to reference the NanoVNA with

75 ohms or measureing a home-made load with a 4 terminal system and getting
a more accurate result. But I'm not sure if your other objections are
valid (1) (2) (6), because most home-made loads won't be characterized
anyway. The operator will take the answer he gets, publish it in QST or
RadCom, and move on.

I believe implementing the full model could be beneficial for homemade kits
in *some* circumstances, such as

* Have the ability to measure homemade standards at work. I would suspect
that a fair few NanoVNA users work in the RF field.

* Know someone with a VNA able to measure them

* *Possibly* compute the properties using a software package like openEMS



* *Possibly* compute the approximately properties, then tweak them to
produce the best calibration possible by using the T-checker.



Jeff has convinced me that for the HP kits, C0 is sufficient.

I think we can all accept the possibility of making slight tweaks of the
load based on 4-wire resistance measurements, are the possibility of
working in 75 ohms.

*ONE OTHER THING I HAD FORGOTTEN ABOUT IS THE NEED TO BE ABLE TO ENTER THE
DELAY OF A THRU FOR A 2-PORT CALIBRATION *

Of course this brings up the operations of data entry and storage; which
starts taking away from the original concept of turn it on, do a simple
1-2-3 cal, and use it, so I think that option should be 1st choice in the
software.

If the firmware could

a) Define a number of calibration kits
b) Default to the most used one

then once the VNA is configured once, the rest would be a simple 1-2-3.

There seems a good argument for the VNA defaulting to the parameters of the
supplied kit (50 fF on the open, some small negative delay on the short).
But I would like to override that, as I will never use the supplied kit, as
it’s impossible to avoid rotating the male pin in the female.

Jim McEwen, KA6TPR


Dave

--

Dr. David Kirkby,
Kirkby Microwave Ltd,
drkirkby@...

Telephone 01621-680100./ +44 1621 680100

Registered in England & Wales, company number 08914892.
Registered office:
Stokes Hall Lodge, Burnham Rd, Althorne, Chelmsford, Essex, CM3 6DT, United
Kingdom

On Sun, 29 Sep 2019 at 17:40, Jeff Anderson <jca1955@...> wrote:

On Sun, Sep 29, 2019 at 03:43 AM, Dr. David Kirkby from Kirkby Microwave
Ltd wrote:

The 85032F is not one of Keysight’s best kits.

Hi Dave,

You are probably right. I chose this kit because, of the 50-ohm cal-kits
listed by Keysight as supported for the 8753D, it seemed to have the worst
higher-order capacitance terms, thus a good choice for testing my theory.

I don’t know if you have an 85032F, but if you do, you will find the male
and female opens have the same coefficients in your 8753D. But Agilent
later revised the coefficients, with different ones for the male & female.

It would be interesting to know what errors you get with your simplified
model for the 85050B 18 GHz APC7 calibration kit.

Ask and you shall receive!

First, Keysight's values defining the 85050B open:

C0 = 90.4799e-15;
C1 = 763.303e-27;
C2 = -63.8176-36;
C3 = 6.4337e-45;
Delay_Open = 0;
Loss_Open = 0;
Offset_Zo_Open = 50;

Now, the results of my calculations...

1. Open using C0 - C3 : rho = 1.0000, theta = -4.9381 degrees

2. Open using only C0: rho = 1.0000, theta = -4.8830 degrees
(Note: the angular delta between the two thetas is 0.055 degrees)

Yes, the difference of 0.055 degrees at 1500 MHZ is small compared to the
0.3 degree uncertainty of the standard

I accept that C0, Z0 and delay are sufficient for any of the Keysight kits
you have tried, and I expect any others too.

Having support in firmware for this would be really good for anybody, like
myself, who is keen to use the NanoVNA standalone for use outside on
antennas, not in conjunction with a PC.

Dave.

--

Dr. David Kirkby,
Kirkby Microwave Ltd,
drkirkby@...

Telephone 01621-680100./ +44 1621 680100

Registered in England & Wales, company number 08914892.
Registered office:
Stokes Hall Lodge, Burnham Rd, Althorne, Chelmsford, Essex, CM3 6DT, United
Kingdom

Hi Jeff
I have given a report I made on the last day of 2017 a brush-up which you can download from
to design a homemade male BNC calibration kit.pdf
It is a full blown how to and if someone else seems it is overwhelming the trick is to calibrate to the rear of the adaptor and use 0ps for short, the fringe C for the open simulated in my report and the tuned shunt C for the load. That is a good starting point. Then check the calibration with a semirigid cable or a BNC test cable of length 0.5 meter until the S11 dB trace run without oscillation. Then you measure S11 in a super way an to "hell" with the measurement plane is not the defined calibration plane for the BNC adaptor :) If the NanoVNA could provide a negative Electrical delay it could be fixed.
To all !!! Please note and respect the reservations I have made at the end of the report
Kind regards
Kurt

-----Oprindelig meddelelse-----
Fra: [email protected] <[email protected]> P? vegne af Jeff Anderson
Sendt: 29. september 2019 19:04
Til: [email protected]
Emne: Re: [nanovna-users] Cal-Kit Standards' Definitions

Hi David
One detail if I may. It is not possible to use the R&S T-Check software it requires the NanoVNA are able to do a full 10/12term error correction.. Sorry
Kind regards
Kurt

-----Oprindelig meddelelse-----
Fra: [email protected] <[email protected]> P? vegne af Dr. David Kirkby from Kirkby Microwave Ltd
Sendt: 29. september 2019 20:00
Til: [email protected]
Emne: Re: [nanovna-users] Cal-Kit Standards' Definitions

On Sun, 29 Sep 2019 at 07:28, Starsekr via Groups.Io <Starsekr= [email protected]> wrote:

On Sat, Sep 28, 2019 at 03:49 PM, Dr. David Kirkby from Kirkby
Microwave Ltd wrote:

I don’t think your simple model is really suitable for the following
reasons

1) The variation of C with homemade standards is likely to exceed
that of commercial standards - this is from experience measuring them.

2) The inductance of shorts is likely to be more with homemade
standards than commercial ones - again this is based on experience measuring them.

3) People may well want to make measurements in a 75 ohm system.

4) it is possible to improve upon the accuracy of loads at low

frequencies

by using a DC resistance measurement.

5) In the case of a female N, a simple standard can be made by just

leaving

the connector open. This will create a higher impedance transmission
line than 50 ohms as the centre conductor sits in a cylindrical
section with a greater diameter than when its mated.

6) The loss of homemade standards is likely to be greater than
commercial ones from Keysight - again this is based on actual
measurements I have performed.

Dr. Kirkby, if I understand your post, and objections to Dave
Anderson's "Very Simple Characterization Model", You are against Mr.
Anderson's idea because it doesn't account for the possibility that
inductance, capacitance and loss terms are likely to be significant in
home-made standards and a desire to have an option to change the
reference resistance to account for actual load resistance, or to use
the VNA at something other than 50 ohms, but you agree with the idea that Offset Delay should be an input option.

Yes, essentially

Dr. Kirkby, I too like the idea of being able to reference the NanoVNA with

75 ohms or measureing a home-made load with a 4 terminal system and
getting a more accurate result. But I'm not sure if your other
objections are valid (1) (2) (6), because most home-made loads won't
be characterized anyway. The operator will take the answer he gets,
publish it in QST or RadCom, and move on.

I believe implementing the full model could be beneficial for homemade kits in *some* circumstances, such as

* Have the ability to measure homemade standards at work. I would suspect that a fair few NanoVNA users work in the RF field.

* Know someone with a VNA able to measure them

* *Possibly* compute the properties using a software package like openEMS



* *Possibly* compute the approximately properties, then tweak them to produce the best calibration possible by using the T-checker.



Jeff has convinced me that for the HP kits, C0 is sufficient.

I think we can all accept the possibility of making slight tweaks of the load based on 4-wire resistance measurements, are the possibility of working in 75 ohms.

*ONE OTHER THING I HAD FORGOTTEN ABOUT IS THE NEED TO BE ABLE TO ENTER THE DELAY OF A THRU FOR A 2-PORT CALIBRATION *

Of course this brings up the operations of data entry and storage;
which starts taking away from the original concept of turn it on, do a
simple
1-2-3 cal, and use it, so I think that option should be 1st choice in
the software.

If the firmware could

a) Define a number of calibration kits
b) Default to the most used one

then once the VNA is configured once, the rest would be a simple 1-2-3.

There seems a good argument for the VNA defaulting to the parameters of the supplied kit (50 fF on the open, some small negative delay on the short).
But I would like to override that, as I will never use the supplied kit, as it’s impossible to avoid rotating the male pin in the female.

Jim McEwen, KA6TPR


Dave

--

Dr. David Kirkby,
Kirkby Microwave Ltd,
drkirkby@...

Telephone 01621-680100./ +44 1621 680100

Registered in England & Wales, company number 08914892.
Registered office:
Stokes Hall Lodge, Burnham Rd, Althorne, Chelmsford, Essex, CM3 6DT, United Kingdom

On Sun, 29 Sep 2019 at 21:40, Kurt Poulsen <kurt@...> wrote:

Hi Jeff
I have given a report I made on the last day of 2017 a brush-up which you
can download from
to design a homemade male BNC calibration
kit.pdf

I had a problem with the link - initially I thought there was no file
present, but later I see my iPhone had only put the first part in a link.

You don’t surprise me with your comment about the Huber & Suhner load. I
think their products are overrated.

I recall testing a large number of female-female N connectors and found
Huber and Suhner had just about the worst electrical performance of any. I
was going to phone them to complain, but on checking the data sheet I could
see it met the specifications. What I did find however is that the quality
of their machining, which was done in China ???, was the best of any of
the adapters.

From an electrical perspective, a really cheap adapter someone sent me free
was the best. However, the threads looked as though they had been cut with
an axe, rather than rolled or cut with a die. ????

I suspect that there are tradeoffs between mechanical strength and
electrical performance of adapters. If you grip the PTFE and inner
conductor well, it makes it mechanically strong, but gives poor electrical
performance.

Dave.

On Sun, 29 Sep 2019 at 22:00, Kurt Poulsen <kurt@...> wrote:

Hi David
One detail if I may. It is not possible to use the R&S T-Check software it
requires the NanoVNA are able to do a full 10/12term error correction..
Sorry
Kind regards
Kurt

I was unaware that it could not do that, but it doesn’t change the fact you
can use T-checker or airlines on another VNA, or perhaps the external
software for the NanoVNA.

I am sure I will learn more about this little device when I finally get it
boxed up with decent switches, USB connector and RF connectors. Otherwise I
know that I will damage it. ????

Dave.

--

Dr. David Kirkby,
Kirkby Microwave Ltd,
drkirkby@...

Telephone 01621-680100./ +44 1621 680100

Registered in England & Wales, company number 08914892.
Registered office:
Stokes Hall Lodge, Burnham Rd, Althorne, Chelmsford, Essex, CM3 6DT, United
Kingdom

Mr. Poulsen, very nice pages and work. Great primer explaining reference planes and delays. I think I'm going to have to read it a few times to absorb everything. I hope someone will modify the calibration instructions in the group files section to include your suggestion and reasoning for not using the open load during calibration.

Jim KA6TPR

Kurt.

Thanks for these documents, I updated the link as this will work




It helped me to understand some calibration problems w.r.t. reference planes

Erik.