This has been commented on by online experts stating that Hugen, the designer of this board, posted notes about his efforts to keep the grounds for tx and rx port circuits isolated to some extent.
Opinion by some is that the modification I performed above which electrically bonds the two connectors through a brass bar of about 60mm length is likely to significantly degrade performance.
Let us look at currents that might be induced in the outside surface of the shield in a typical test setup. Unavoidable ambient RF fields will induce a voltage in the loop circuit formed by the outer surface of the shield.
In the case of the unmodified nanovna-H, the extraneous shield currents flow approximately around the green path, traveling though the ground planes of tx and rx port circuitry before being joined aproximately in the middle of the instruments.
In the case of the modified nanovna-H, the extraneous shield currents flow aproximately around the magenta path, with far less of it flowing in through the ground planes of tx and rx port circuitry and creating ground conductor noise voltage.
It is difficult to understand the claim that the green current path should be better, and that the magenta current path degrades performance. Rather one might question the wisdom of isolating the tx and rx port grounds to drive these currents further into the box.
Of course I defer to credible measurement that shows that the modification does degrade performance.
]]>This article presents a theoretical prediction based o A model of current distribution in copper clad steel conductors at RF of the matched line loss (MLL) at 1.8MHz.
The assumption is a 1.024mm steel cored conductor with 30.7µm copper cladding.
Above is a plot of the predicted current magnitude and phase distribution in the conductor.
From that, we can calculate the expected MLL to be 0.00405dB/m, 66% higher than we would expect of a solid copper conductor of the same diameter.
Of course credible measurements are of greater value than simply theoretical predictions, and there may be good reasons why real product varies from theoretical.
John Oppenheimer, KN5L, has made measurements of two different samples of cable marked JSC 1318 and reported MLL @ 1.8MHz of 0.0049dB/m in both cases.
To the point of theoretical vs actual, note that this product is advertised as 30% IACS cladding which would suggest MLL should theoretically be 0.00241dB/m @ 1.8MHz, a long way short of the measured 0.0049dB/m. The measured DC resistance strongly hints a likely reason for serious departure from theoretical.
The expected loss of 1.024mm (#18) 21% IACS cladding is similar to a solid copper conductor at 7MHz and above, and poorer below that frequency.
Note that the results here cannot simply be applied to the more popular stranded CCS types of line, they are significantly worse MLL due to the thinner copper that accompanies stranding.
The burning question is, what do you actually get when you buy these windowed ladder lines?
Above is a diagram of the so-called “cable balun”.
My evaluation essentially showed that it was not effective in an example practical scenario where one might want to use a balun, and that of itself, it was not likely to significantly reduce common mode current in most scenarios.
Radcom Mar 2020 published a letter in “The last word” from the author defending the device citing a NEC model of one scenario, curiously though without explanation, a different topology to the diagram above from the original article. Note also that it is a structure in free space with no discussion of how that is relevant to real world antennas near ground.
The author states it is an NEC4 model, and I have run it in NEC-4.2 but I doubt that it will give significantly different results in NEC-2.
I have made two changes to the model:
Here is the retyped NEC deck.
CM Typed from Radcom March 2020 pages 97-98. K3MT Letter CM RSGB CABLE BALUN STUDY CM AUTHOR: K3MT@JOKALYMPRESS.COM CM 40 M DIPOLE IN FREE SPACEH CM 20m LENGTH 5MM DIAMETER CM CM THE DIPOLE CM THE BALUN CM THE HALF WAVE FEEDLINE CM NO GROUND - FREE SPACE CM CM VOLTAGE SOURCE 2V AT CENTER OF DIPOLE CM 6700 kHz CM RADIATION PATTERN CE GW 1 48 -10.2 0 10.0 0 0 10.0 0.005 GW 2 52 0 0 10.0 10.5 0 10.0 0.005 GW 3 25 0 0 10.0 0 5.4 10.0 0.005 GW 4 25 0 0 9.5 0 5.4 9.5 0.005 GW 5 5 0 5.4 10.0 0 5.4 9.5 0.005 GW 6 100 0 0 9.5 0 0 -10.0 0.005 GE 0 GN -1 EK EX 0 1 48 0 2.0 0 0 FR 0 0 0 0 6.700 0 EN
Above is a plot of the current distribution of the model given by K3MT to demonstrate the balun performance.
To discover whether the common mode current is well controlled by K3MT’s balun, lets extend the vertical conductor (modelling the feed line common mode current path).
Above is a plot of the current distribution of the revised model, it is clear that common mode current on the vertical conductor has increased significantly.
Above is a slice of the antenna pattern which shows significant distortion due to the asymmetry of the antenna system.
If the common mode current is sensitive to vertical conductor length, let’s try a little shorter conductor.
Above is a plot of the current distribution of the revised model, common mode current is significantly lower.
Comparing figures 4 and 6 which show a marked variation in common mode current caused by changing the linear length of conductor from feed point to lowest extremity from 3λ/4 to λ/2 hints that the common mode current may be simply responsive to that length and the nature of termination (ie open end or grounded), and that K3MT’s folded section doesn’t of itself control common mode current.
K3MT’s balun in his demonstration scenario has not isolated the nominal radiator (the flat top dipole) from the influence of the common mode feed line conductor.
The model results are applicable to the structure in free space and not directly extensible to an antenna system near ground, much less a grounded transmitter.
]]>The common assumption is that as frequency is reduced, so is loss, and at low frequencies loss is roughly proportional to square root of frequency.
That model is for homogenous conductors with well developed skin effect and is not applicable to the CCS line under discussion.
Above is a plot for various cladding depth on a 1.024mm (#18) 30% IACS (67µm cladding) CCS conductor at 1.8MHz where skin depth δ is 49µm. MLL is minimum around cladding depth 100µm or 2δ.
It is often stated that a conductor with outer layer of at least 3δ gives copper like performance, but in fact 2δ cladding is sufficient in this case (and slightly lower than a full copper conductor).
W551 specs state that it uses 30% IACS CCS which in turn is specified to have a 67µm cladding. So, it should be suitable for frequencies where the skin depth in copper is less than 33.5µm which is 3.8MHz. The curve is fairly low gradient at that point, and you could take that operation above 3.5MHz was quite fine.
At 1.8MHz we can see from the graph that loss is very slightly worse, insignificant really, and so whilst not optimal, it is still quite good.
Above is a plot of the current distribution on a 1.024mm (#18) CCS conductor with 17.8µm cladding at 25MHz where δ is 13µm. The W553 product with 19 strands (#30 / 0.25mm??) of 30% IACS CCS probably behaves similarly to this. At 30MHz MLL is close to copper and the current distribution explains why, the current density in the copper cladding is high falling to about 30% at the inner extremity of the copper. An alternative view is that at 1.8MHz, it is probably towards three times the MLL as the W551 line.
It is very sensitive to any jiggling of the cable or connector, causing a reset of the nanovna which almost always means lost work.
Having tried a number of different cables that have worked reliably on other devices, I initially thought there was little difference.
I did have a good response to jetting plug and jack with IPA, but the effects are shortlived.
This brings me to consider whether the connector is degrading making debris that makes for unreliable contact, or whether this is too little spring pressure in the plug.
Above is a view into the supplied USB-C plug. The pic has been taken with care to line up the die parting marks at back and front of the connector, so the view is in line with the connector axes.
First it is obvious that only the bare necessity of pins are populated so an indicator of CHEAP. Functionally though and therefore more importantly, the contacts on the lower side of this connector do not project as far towards the connector midline as the upper contacts. I make the distance from the lower contacts to the midline to be four times that of the upper contacts.
Above is an attempt to measure the distance from the midline to the top of the contacts in the lower side of the jack. Given that the specification thickness of the tongue in the plug is 0.7mm or 0.35mm either side of the midline, contact pressure from these contacts will be low and possibly break with wiggling of the plug.
Cheap Chinese junk.
5 new USB cables were tested and in ALL cases, the connection is much more unreliable with the connector plugged in one way than the other which is a strong hint that the jack is faulty. Visual inspection of the jack does cast some doubt, and despite that it is a difficult repair I intend replacing the jack if / when an order from China arrives, it is a country with serious third world health problems right now and that is disrupting deliveries.
Some users apparently lacking the skills to replace the jack have simply cut the end of the cable and permanently attached the wires to the board. I don’t scoff at that at all, it might wind up being the best solution.
Everything about the nanovna-H is a squeeze to keep the price below US$50 without much consideration to performance.
Discussion about the upcoming nanovna V2 is likewise centred on selling price which appears to be the prime design criteria, it shapes every aspect of the solution.
It also speaks to the target market, hams, who they think will buy anything that is cheap. Who am I to argue with people who have sold thousands of these things?
]]>
The details of the model are a little sketchy, I was interested in how it modeled the phase of the layer currents, or if you like the implied velocity of propagation of the EM wave in the conductor.
Again the model is of a copper clad steel conductor, but tweaked a little to fit the apparent limit to the number of layers modeled in Simsmith, it is 1mm diameter, 500 layers (1µm per layer).
Above is the model with cladding thickness set to 20% or 100µm.
Skin depth δ in copper @ 1.8MHz is 48.43µm, close to 10% of radius in this model.
Above is a comparison of the two models.
Both models show that although the loss decreases with increasing cladding thickness initially, there is a turning point where it then increases for a while.
That characteristic is due to phase change in the current at increasing depth, reaching a point where the current is in the opposite phase to the surface current and so whilst it contributes to loss flowing in the imperfect conductor, it decreases the net current flow. Increased loss and reduced net current means increased equivalent internal resistance.
The turning point is different though, which questions whether one or both models are wrong.
If we take my own model, it might surprise some readers that the loss is less when the cladding is around 2δ than 3δ, it does not fit with the traditional explanation of skin effect which ignores the phase change with depth.
This article explores using a Bird 43 directional wattmeter to measure line loss in a similar scenario. We will use 6m of Belden 8359 (RG58A/U) @ 3.6MHz.
A short digression, what is the specification Matched Line Loss (MLL) at 3.6MHz? Using TLLC we get 0.171dB, that is our expectation.
(Bird 2004) gives the following advice.
Line loss using open circuit calibration: The high directivity of elements can be exploited in line loss measurements, because of the equality of forward and reflected power with the load connector open or short circuited. In this state the forward and reflected waves have equal power, so that φ = 100% and ρ = ∞.
Open circuit testing is preferred to short circuit, because a high quality open circuit is easier to create than a high quality short. To measure insertion loss, use a high quality open circuit to check forward and reverse power equality, then connect an open-circuited, unknown line to the wattmeter. The measured φ is the attenuation for two passes along the line (down and back). The attenuation can then be compared with published data for line type and length (remember to halve Ndb or double the line length to account for the measurement technique).
This also contains the hoary old chestnut that a good OC termination is hard to achieve, but this author’s experience of measurement with modern VNAs is not consistent with Bird’s assertion.
So lets do a theoretical simulation of the Bird 43 applied to this problem.
Lets say we connect a source to the line section with a short circuit (SC) termination, and that the Bird 43 reads Pfwd=90W, and we read Pref=78W, we can calculate return loss \(RL=10 \cdot log_{10}\frac{P_{fwd}}{P_{ref}}=0.65dB\), so RL/2=0.65/2=0.325dB.
Hmmm, that is nearly double the expected 0.171dB, time to chuck it?
No not yet, let’s do an open circuit sweep.
Now we connect a source to the line section with a open circuit (OC) termination, and that the Bird 43 reads Pfwd=75W, and we read Pref=72W, we can calculate return loss \(RL=10 \cdot log_{10}\frac{P_{fwd}}{P_{ref}}=0.177dB\), so RL/2=0.177/2=0.088dB.
Hmmm, that is less than half the expected 0.171dB, a quarter of the SC measurement, what is going on?
The problem is that the Bird 43 is calibrated using a 50+j0Ω load, and its reference impedance is now 50+j0Ω meaning that directional power measurements and therefore RL is wrt 50+j0Ω, and since Zo is more like 51.42-j1.33Ω there is error in the two results that should theoretically agree with each other.
A good approximation when the departure of actual Zo from the reference impedance is small is that \(MLL\approx\frac{RL_{sc}+RL_{oc}}4dB\).
So \(MLL\approx\frac{0.65+0.177}4=0.206dB\), not a lot worse than spec at 0.171dB, and given measurement uncertainties, we could not say if fails spec.
This is discussed in more detail at On Witt’s calculation of Matched Line Loss from Return Loss .
A problem in applying the Bird 43 to this example is that we cannot read the meter to sufficient resolution or accuracy, though it might be quite suitable for a higher loss section.
Nevertheless, the measurements do expose the fact that the measured Return Loss wrt 50+j0Ω may be quite different for a SC section to a OC section, and it is not explained by the traditional mumbo jumbo about the quality of an OC termination. It IS true that an OC has associated with it some fringe capacitance which for a N type OC calibration piece amounts to some tens of fF (femtoFarads are thousands of a pF) so work out the effect at HF, and even VHF is very small. A N type plug with nothing connection is a little worse, but inconsequential at HF. For this scenario 50fF fringing capacitance is roughly equivalent to the cable appearing to be 0.5mm longer than you measured.
Other instruments mentioned earlier may be capable of higher resolution and accuracy than the Bird 43 and may be more practical for low loss scenarios.
This article explores using a nanovna to measure line loss in a similar scenario. We will use 6m of Belden 8359 (RG58A/U) @ 3.6MHz.
The same technique could be used with a quality antenna analyser.
A short digression, what is the specification Matched Line Loss (MLL) at 3.6MHz? Using TLLC we get 0.171dB, that is our expectation.
A common method proposed is to measure Return Loss (RL) of a section with load end RL=0dB and halve it. Many experts advise that the section should be terminated in a short circuit (S) because short circuits are more reliable than open circuits. So let’s get cracking.
Above is measured |s11| using a nanovna with recent OSL calibration from 1-30MHz. |s11| @ 3.6MHz is by eye -0.651dB, RL=-|S11|, so RL/2=0.651/2=0.325dB.
Hmmm, that is nearly double the expected 0.171dB, time to chuck it?
No not yet, let’s do an open circuit sweep.
Above is measured |s11| using a nanovna with recent OSL calibration from 1-30MHz. |s11| @ 3.6MHz is by eye -0.138dB, RL=-|S11|, so RL/2=0.138/2=0.069dB.
Hmmm, that is less than half the expected 0.171dB, a quarter of the SC measurement, what is going on?
The problem is that the nanovna was calibrated using a 50+j0Ω load, and its reference impedance is now 50+j0Ω meaning that s11 and therefore RL is wrt 50+j0Ω, and since Zo is more like 51.42-j1.33Ω there is error in the two results that should theoretically agree with each other.
A good approximation when the departure of actual Zo from the reference impedance is small is that \(MLL\approx\frac{RL_{sc}+RL_{oc}}4dB\).
So \(MLL\approx\frac{0.651+0.138}4=0.197dB\), not a lot worse than spec at 0.171dB, and given measurement uncertainties, we could not say if fails spec.
This is discussed in more detail at On Witt’s calculation of Matched Line Loss from Return Loss .
]]>You might expect that using a directional wattmeter has exactly the same problems because as many online experts advise, at the end of the day they are just a voltmeter.
They are wrong, a Bird 43 might use a half wave detector driving a d’Arsonval meter and you might regard that to be a voltmeter, but the RF signal it measures is a combination of samples of forward and reflected waves wrt to its calibration impedance (usually 50+j0Ω) and we will see that makes a difference.
Where a directional wattmeter is calibrated for a purely real impedance (ie X=0), then the relationship \(P=P_{fwd}-P_{ref}\) holds true (On the concept of that P=Pfwd-Prev).
Lets take an example to explore the theoretical answer. We will use 10m of Belden 8359 (RG58A/U) @ 3.6MHz.
Lets model the scenario in TLLC. We will select the “Use Lint” switch for a better model of this specific cable at 3.6MHz and take the “Long” output.
Above is the input form.
Parameters | |
Transmission Line | Belden 8259 (RG58A/U) |
Code | B8259 |
Data source | Belden |
Frequency | 3.600 MHz |
Length | 10.000 m |
Zload | 50.00+j0.00 Ω |
Yload | 0.020000+j0.000000 S |
Results | |
Zo | 51.42-j1.33 Ω |
Length | 67.357 °, 0.187104 λ, 10.000000 m, 5.197e+4 ps |
VF | 0.642 |
Line Loss (matched) | 0.284 dB |
Line Loss | 0.282 dB |
Efficiency | 93.71 % |
Zin | 5.330e+1-j1.281e+0 Ω |
Yin | 1.875e-2+j4.506e-4 S |
VSWR(50)in | 1.07 |
R, L, G, C | 3.247376e-1, 2.670566e-7, 4.548358e-6, 1.010800e-10 |
Γ, ρ∠θ, RL, VSWR, MismatchLoss (source end) | 1.795e-2+j9.163e-4, 0.018∠2.9°, 34.909 dB, 1.04, 0.001 dB |
Γ, ρ∠θ, RL, VSWR, MismatchLoss (load end) | -1.418e-2+j1.293e-2, 0.019∠137.6°, 34.341 dB, 1.04, 0.002 dB |
Vout/Vin | 3.714e-1-j8.606e-1, 9.373e-1∠-66.7° |
Iout/Iin | 3.739e-1-j9.270e-1, 9.995e-1∠-68.0° |
S11, S21 | 3.212e-2-j1.200e-2, 3.730e-1-j8.927e-1 |
Y11, Y21 | 9.562e-4-j8.077e-3, -8.316e-4+j2.103e-2 |
NEC NT | NT t s t s 9.562e-4 -8.077e-3 -8.316e-4 2.103e-2 9.562e-4 -8.077e-3 ‘B8259, 10.000 m, 3.600 MHz |
k1, k2 | 1.487e-5, 2.744e-10 |
C1, C2 | 4.701e-1, 2.744e-1 |
Mhf1, Mhf2 | 4.531e-1, 8.362e-3 |
dB/m @1MHz: cond, diel | 0.014866, 0.000274 |
γ | 3.275e-3+j1.176e-1 |
Loss model source data frequency range | 1.000 MHz – 1000.000 MHz |
Correlation coefficient (r) | 0.999924 |
We can calculate the expected (or theoretical) readings on Bird 43 wattmeters or the like placed at both ends of the line.
Source: \(P_{fwd}=80.00000,P_{ref}=0.02583,\)
\(P_{source}=P_{fwd}-P_{ref}=80.00000-0.02583=79.97417\);
Load: \(P_{fwd}=74.97381,P_{ref}=0.02759,\)
\(P_{load}=P_{fwd}-P_{ref}=74.97381-0.02759=74.94622\);
Overall: \(Loss=10\cdot log_{10}(\frac{P_{load}}{P_{source}})=0.28200dB\).
Readers will note that the reflected power values are very small and in practice it would be a challenge to read forward and reflected power to this resolution. For this type of test scenario with a nominal load, it should usually be the case the reflected power is very small and mostly the result of imperfection of the nominal load that the difference between theoretical Zo and nominal Zo.
So, in this case if we ignored reflected power in the calcs we would obtain \(Loss=0.28180dB\), but practical measurement with great care of forward power would make it difficult to achieve tenth dB accuracy for low loss cable sections. So, with care, one might measure this section to be 0.3dB and that would be sufficient to indicate whether it was likely to be faulty when compared to the MLL specification (0.284dB).
Note that measurement using a poor load means you are measuring loss under standing waves and it may be significantly higher than MLL.
The objective in this type of test is to use a load that produces insignificant reflected power, and then do the full calculation of forward minus reverse if the reverse power observed is significant, otherwise calculation based on forward power at both ends will be sufficient.
Measuring short low loss sections risks measurement uncertainty dominating the result.
]]>
The subject raises some immediate concerns:
Lets take an example to explore the theoretical answer. We will use 10m of Belden 8359 (RG58A/U) @ 3.6MHz.
Lets model the scenario in TLLC. We will select the “Use Lint” switch for a better model of this specific cable at 3.6MHz and take the “Long” output.
Above is the input form.
Parameters | |
Transmission Line | Belden 8259 (RG58A/U) |
Code | B8259 |
Data source | Belden |
Frequency | 3.600 MHz |
Length | 10.000 m |
Zload | 50.00+j0.00 Ω |
Yload | 0.020000+j0.000000 S |
Results | |
Zo | 51.42-j1.33 Ω |
Length | 67.357 °, 0.187104 λ, 10.000000 m, 5.197e+4 ps |
VF | 0.642 |
Line Loss (matched) | 0.284 dB |
Line Loss | 0.282 dB |
Efficiency | 93.71 % |
Zin | 5.330e+1-j1.281e+0 Ω |
Yin | 1.875e-2+j4.506e-4 S |
VSWR(50)in | 1.07 |
R, L, G, C | 3.247376e-1, 2.670566e-7, 4.548358e-6, 1.010800e-10 |
Γ, ρ∠θ, RL, VSWR, MismatchLoss (source end) | 1.795e-2+j9.163e-4, 0.018∠2.9°, 34.909 dB, 1.04, 0.001 dB |
Γ, ρ∠θ, RL, VSWR, MismatchLoss (load end) | -1.418e-2+j1.293e-2, 0.019∠137.6°, 34.341 dB, 1.04, 0.002 dB |
Vout/Vin | 3.714e-1-j8.606e-1, 9.373e-1∠-66.7° |
Iout/Iin | 3.739e-1-j9.270e-1, 9.995e-1∠-68.0° |
S11, S21 | 3.212e-2-j1.200e-2, 3.730e-1-j8.927e-1 |
Y11, Y21 | 9.562e-4-j8.077e-3, -8.316e-4+j2.103e-2 |
NEC NT | NT t s t s 9.562e-4 -8.077e-3 -8.316e-4 2.103e-2 9.562e-4 -8.077e-3 ‘B8259, 10.000 m, 3.600 MHz |
k1, k2 | 1.487e-5, 2.744e-10 |
C1, C2 | 4.701e-1, 2.744e-1 |
Mhf1, Mhf2 | 4.531e-1, 8.362e-3 |
dB/m @1MHz: cond, diel | 0.014866, 0.000274 |
γ | 3.275e-3+j1.176e-1 |
Loss model source data frequency range | 1.000 MHz – 1000.000 MHz |
Correlation coefficient (r) | 0.999924 |
Now the ratio of the magnitudes of correct voltage readings from output to input is given by the figure |Vout/Vin|=0.9373=-0.562dB so loss appears to be 0.562dB.
I say appears to be because the load impedance for Vout is 50+j0Ω, and for Vin is 53.30-j1.281Ω. There is some impedance transformation because Zo is not 50+j0Ω, but 51.42-j1.33Ω so there is a very slight standing wave (VSWR=1.04 at the load end).
So, the spec tells us MLL=0.284dB and we measured 0.562dB under mismatch, albeit slight and a configuration chosen to reveal the problem.
If a physical measurement is taken to test a sample of cable, there are further uncertainties (errors) that might compromise the measurement.
You might ask why the published specification is not subject to the problems discussed here. The manufacturers choose a longer sample and they smooth the measurement curve.
The theory behind this is well over 100 years old, it comes from a self educated mathematician, Oliver Heaviside who built on the work of James Clerk Maxwell.
]]>