Above, the antenna is a 4 element Yagi with Gamma match.

The feed line is about 12m of RG213 coax (sheathed with 13mm HDPE irrigation tube for bird resistance), and for measurement purposes its electrical length was determined by placing a short circuit on the load end and measuring input Z, adjusting the length of feed line ‘subtracted’ to get a load end reactance close to 0Ω. When the equivalent length of the short circuit part was subtracted, the adjusted length was 12.23m.

Above is the VSWR plot at the load end of the feed line. The antenna is tuned a little low, VSWR=1.62 @ 144.2MHz. VSWR is a little higher than expected. The antenna is 51 years old, but refurbished 13 years ago in 2007, and warrants a review based on the observed VSWR. I have another identical antenna which I will rework and swap in if needed.

Above is the R,X plot at the load end of the feed line. Z=72+j19Ω @ 144.2MHz.

Under these conditions, loss in the coax is 1.04dB.

Gain of the antenna system is about 8.8dB, so with 2W of input power, EIRP is 15W.

Measurements were made with a Rigexpert AA-600 driven from Antscope (1) v4.2.57 which allows real time measurement with feed line de-embedded.

]]>This article explores a simple series match to improve the load seen by the transmitter.

In the Simsmith model above, the estimated feed point impedance is imported into element L, then a series section of lossless 50Ω line to represent the coax in the common mode choke (balun), then a series section of lossless 75Ω to perform the impedance transformation, then a section of 50Ω lossless line to make up the required length to the transmitter.

In this case, the first step was to find the frequency at which the VSWR(75) looking into the balun was 1.5.

Above, Simsmith is setup to display VSWR(75) at element G and T2 &T3 are made zero length so we are observing VSWR(75) looking into the balun (T1).

Above, element G is set to measure VSWR(50) and length of T2 adjusted for VSWR=1.

Above, the last step is to then build out T3 to the required length.

Note that the example here is used as a vehicle to explain a matching scheme, and the actual line lengths relate to that specific scenario and are not simply extensible to a different antenna. A solution for another antenna starts with measurement of that antenna.

I have modelled all these as lossless sections as I have issues with the loss models in Simsmith. For example, the use of RG6 CCS cable might not reconcile well with Simsmith’s loss model (RG6/U with CCS centre conductor at HF). Nevertheless, the lossless models explain the concepts.

In practice, rather than treating the dipole as a fixed length, you would adjust its length to achieve VSWR(75) looking into the balun at the desired operating frequency. Then you would insert sufficient 75Ω line to measure VSWR(50)=1 looking into that line section. Then connect sufficient 50Ω coax to reach the transmitter.

The astute reader will realise that there are longer length of the 75Ω line that will work, indeed one of these longer length might be conveniently used and T3 dispensed with (ie the 75Ω line terminated at the transmitter jack).

]]>A correspondent recently shared an AIM 4170 scan file of his 40m half wave dipole antenna system taken from the transmitter end of the coax and maintaining the common mode current path by bonding the shield of the coax connector to normal connection point on the transmitter.

Above is his graphic of the measurement looking into around 23m of RG58 feed line.

It shows the VSWR curve is quite classic in shape, the frequency of minimum VSWR is a little low, and the minimum VSWR is 1.478 which is quite within expectations of such an antenna.

We can make some further inferences from this chart… but to inform the process, lots consider the half wave dipole.

Let’s look at an NEC model of a 40m half wave dipole at a height of a quarter wavelength above average ground.

Above is a plot of the VSWR, for this scenario it is a minimum of 1.62 at 7.02MHz.

Above is the impedance plot, focus on the R, X plot as it is simpler to characterise. Around resonance of the dipole (ie where X=0), R increases very slowly with frequency and X increases relatively rapidly with frequency. At minimum VSWR, X is almost zero, and feed point impedance is almost purely resistive, approximately VSWR times 50, so Z is about 80+j0Ω ohms in this case.

Above is a Smith chart view of impedance about resonance and minimum VSWR.

Note that the shape near minimum VSWR (the distance to the chart centre) is much like part of the letter C but rotated clockwise a few degrees. The actual value of R is scenario dependent, quite sensitive to height above ground and the ground parameters, and commonly between about 55 and 95Ω.

So lets return to the first plot of the real world dipole system.

In the knowledge of the feed point characteristics of a typical half wave dipole, we can infer feed point impedance from the measured VSWR=1.478. In fact the feed line has loss, so depending on that loss, feed point VSWR might be more like 1.6 and the feed point impedance at frequency of minimum VSWR is close to 1.6*50+j0=80+j0Ω.

Viewing the scan data as a Smith chart is quite revealing.

Above is the Smith chart view from the AIM software with the cursor set at the frequency of minimum VSWR. The blue arrow is my addition to show the vector that is Γ or s11. You may note that the value of s11 is given in the lower right corner of the graphic as -0.1038+j0.1630=0.1933∠-57.521. Anyone with basic understanding of complex numbers will recognise that the conversion to polar form is wrong, it should be 0.1933∠122.5° which is the blue vector added to the chart above. (This is quite possibly the result of a amateur programmer using a form of the ATAN function that returns results in the range ±90°… kids in the kitchen!)

This angle of s11 is important as it implies that there is an electrical length of (360-122.5)/2=118.9° lossless line to the first point where X is zero and R is around 80Ω. The impedance is approximately repeated every half wavelength of line, so again at 118.9+180=298.9° and so on.

So an informed guess is that since there is around 23m of feed line, the actual length is closer to 298.9° or 23.4m… it is still an approximation as it has not taken into account line loss or the small residual feed point X at minimum VSWR.

So the next step would be to use the ‘refer to antenna’ function in the AIM software to apply a lossless line transformation of 23.4m (VF=0.66) and adjust that length until the Smith chart plot is C shaped with a small clockwise rotation in the region of minimum VSWR. I cannot show that to you because one of the many limitations of the AIM software is it has to remeasure the DUT to apply the transformation… DUH!!!

Nevertheless, lets extract the datapoints by hand from the scan file (another of the many software deficiencies) and plug them into Simsmith and de-embed some lossless line.

Above, the magenta plot is the measured data, and the blue plot is the measurement with 23.6m of lossless VF=0.66 line ‘subtracted’ and represents what should be seen in the AIM software when the length is tweaked for the characteristic C shape with a few degrees of clockwise rotation.

With this transformation, the other AIM plots should be now adjusted to the feed point (though an approximation only because feed line loss was not factored in). But like everything with AIM, double check everything.

As mentioned, the AIM software does not allow factoring in the line loss when de-embedding the line.

Above is a plot derived from the scan file of R, X and VSWR looking into the line.

In the above plot, the de-embedding of RG58 (including its loss) is performed in a spreadsheet to obtain the feed point R, X, VSWR and loss of the line section.

Note that:

- minimum VSWR at the load is at the same frequency as at the line input, but marginally higher as a result of line loss; and
- the frequency of minimum line loss coincides with minimum VSWR.

The first optimisation target for this antenna system should be to adjust the dipole length so that minimum input VSWR occurs at the desired operating frequency.

]]>Above, the latest repair. A new battery socket to replace the original that crumbled apart… sub-standard plastic from all appearances. This was from a reputable supplier, so it is probably a genuine JST part rather than some cheap Chinese knock off.

The blue wire is part of a mod to invoke the bootloader on power up, R5 was also changed to something small, 1k IIRC.

]]>

Above, the 10dB attenuator is semi permanently attached to Port 2 principally to improve the Return Loss (or impedance match) of Port 2, a parameter that becomes quite important when testing some types of networks than depend on proper termination (eg many filters). I should remind readers that the improvement in Port 2 Return Loss comes at a cost, the dynamic range of Port 2 is reduced by 10dB.

Several correspondents have suggested the there is also benefit in permanently attaching such an attenuator to Port 1. These suggestions were traced to the assertions of one online expert.

The nanoVNA-H v3.3 samples the signal supplied to the measurement bridge, albeit prior to a voltage divider and we might assume that enters into calculation of the value of s11. I have not examined the source code to see if it includes the effect of that voltage divider and its variable load, but I suspect not as the voltage divider action is dispensed with in v3.4 and they both use the same firmware. I suspect normal VNA correction algorithms are relied upon to deal with non ideal hardware design.

While adding an external attenuator to Port 1 does nothing to address shortcomings in the implementation of the measurement bridge, it does degrade the dynamic range for s11 and s21 measurement.

I think it wrong to think that a network such as a filter will not measure properly if the source is not Thevenin impedance matched. The VNA is using a measurement bridge and its calibration / correction processes to come to a steady state value that depends entirely on the (complex) ratio V/I at the measurement plane, and that ratio is determined by things downstream from the measurement plane and NOT by things upstream of the measurement plane. This might not be accepted by disciples of Walt Maxwell’s re-re-reflection proposition.

So, whilst a network such as a filter might not exhibit a correct response in an ordinary sweep generator – filter – detector configurations (such as a spectrum analyser with tracking generator) unless both generator and detector are matched to the design impedances of the filter, in the case of a VNA measurement it is important that the filter is terminated in a ‘matched’ load, the measurement / calculation of s11 and s21 is corrected by the calibration and measurement processes of the VNA.

Above is an extract from the schematic for v3.3. A first approximation for the source impedance seen by the measurement bridge is the Si5351A output impedance (50Ω) in parallel with with R16+R17 (497Ω) for a combined 45.43Ω, in series with R13 (150Ω) for 195.43Ω, in parallel with R14 (56Ω) for combined 43.53Ω.

Measuring the output power @ 10MHz and applying a second termination results in a drop of 3.3dB on the power meter indication. We would expect 3.52dB for a 50Ω source and 3.32dB for a 45.53Ω source, validating the approximate circuit analysis.

Above is an extract from the schematic for v3.4. A first approximation for the source impedance seen by the measurement bridge is the Si5351A output impedance (50Ω) in series with R13 (0) in parallel with with R16+R17 (497Ω) for a combined 45.43Ω, in parallel with R14 (56Ω) for combined 25.08Ω.

The ‘improvements’ to v3.4 move further from an idealised 50Ω Zs for the measurement bridge…. but the instrument works with the same firmware as v3.3 hardware.

I don’t have a v3.4 to test, but for a double termination test, we would expect 3.52dB for a 50Ω source and 2.5dB for a 25.08Ω source. Happy to quote a valid experiment here if data is offered.

I wrote at Return Loss Bridge – some important details of the dependence of an ordinary Return Loss Bridge and its calibration / measurement process on near ideal Thevenin source impedance equaling the calibration or reference impedance. In that case, failure to supply a correct Thevenin source cannot be solved by putting an attenuator on the ‘unknown’ port.

]]>I make the index finger nail width exactly the same as the round part of the SMA nut which is 7.6mm. That is a very tiny hand… or the image is a composite fraudulently not to scale.

The small screen size is one of the most criticized ‘features’ of the nanoVNA.

My nanoVNA-H v3.3 screen escutcheon width is 60mm (2.4″), diagonal 74.7mm (2.94″), the active area of the screen is a little less. There are models with bigger screens, but this pic is not one of them, it is a fraud.

My hand is not particularly large, but it does make the nanoVNA-H v3.3 look a lot smaller than the first pic. (The second pic is not a mash up, it is not retouched or edited.)

So Chinese!

]]>Eager owners are trying to apply them to solve lots of problems, often without sufficient knowledge or experience to properly inform the measurements.

An example that has a appeared a few times on online forums in the last weeks is measuring the matched line loss (MLL) of a section of RG6 coax… to inform a decision to discard it or keep it.

The common approach is to use a measurement of |s11| and to calculate Return Loss and infer the MLL.

For discussion, lets consider an example of 30′ of Belden 1694A RG6 solved in Simsmith. We should note that unlike most RG6 in the market today, this uses a solid copper centre conductor.

Some authors insist that the half return loss method is to be performed using a short circuit test section. Bird does this in their Bird 43 manual.

Above is a plot of calculated |s11| (-ReturnLoss) from 1 to 20MHz for the test section. The three plots are of |s11| wrt 50Ω, 75Ω and frequency dependent actual Zo (as calculated for the model). The cursor shows that the actual |s11| is -0.37474dB (ReturnLoss=0.37474dB). Using the half return loss method MLL=ReturnLoss/2=0.37474=0.187dB/m.

Now to the other two traces.

|s11|(50)=-0.4241dB, quite different from the correct value 0.37474dB (the red curve).

|s11|(75)=-0.4521dB, quite different from the correct value 0.37474dB (the red curve).

Note that at other frequencies the error is different.

Above is a plot of calculated |s11| (-ReturnLoss) from 1 to 20MHz for the test section. The three plots are of |s11| wrt 50Ω, 75Ω and frequency dependent actual Zo (as calculated for the model). The cursor shows that the actual |s11| is -0.37474dB (ReturnLoss=0.37474dB). Using the half return loss method MLL=ReturnLoss/2=0.37474=0.187dB/m.

Now to the other two traces.

|s11|(50)=-0.2636dB, quite different from the correct value 0.37474dB (the red curve).

|s11|(75)=-0.2922dB, quite different from the correct value 0.37474dB (the red curve).

Note that at other frequencies the error is different.

If we average the |s11|(50) measurements for short and open short we get (-0.4241+-0.2636)/2=-0.3439 for MLL=0.1719dB, well below the correct value of 0.187dB.

If we average the |s11|(75) measurements for short and open short we get (-0.4521+-0.2922)/2=-0.3722 for MLL=0.186dB, very close to the correct value of 0.187dB.

The method of inferring Matched Line Loss from measured Return Loss of an open circuit or short section line section is soundly based but often fails in the execution. The Return Loss (or -|s11|) used as the basis MUST be with reference to the actual Zo of the line, though a good approximation can be obtained by averaging the Return Loss measurements for S/C and O/C section when Zref is close to Zo.

The knowledge and experience important to exploitation of the nanoVNA does not come in the box.

]]>check the attenuation loss in some old & weathered RG-6 (75 ohm) cables for the TV signal frequencies. Excuse the term

attenuation loss, lets assume the poster is asking for matched line loss (MLL).

The assembled experts are offering solutions to transform the ports to 75Ω and make a measurement, deducting the loss of the transformation (minimum loss pads were suggested).

There is a very simple solution that should be quite practical for the scenario described. Let’s work through two examples using 35.5m of unbranded quad shield RG6 with CCS centre conductor (of unknown quality) for the DUT.

The cable is connected to the nanoVNA Port 1 (Ch0 in nanoVNAspeak), and the far end is open circuit. A swept impedance measurement is made around a frequency where the cable is an integral number of quarter waves electrically and the R value at resonance noted.

Above is the impedance sweep, and R at resonance is 468.76Ω.

Lets now calculate MLL using Calculate transmission line Matched Line Loss from Rin of o/c or s/c resonant section . We need to know the length and Zo to fairly good accuracy. We will assume that at 28MHz, Zo is very close to 75Ω.

Above, the calculation returns 0.042dB/m, or 4.2dB/100m. That is a little higher than copper cable datasheet loss, possibly a result of insufficient copper cladding for the frequency.

Above is the impedance sweep, and R at resonance is 23.07Ω.

Lets now calculate MLL using Calculate transmission line Matched Line Loss from Rin of o/c or s/c resonant section . We need to know the length and Zo to fairly good accuracy. We will assume that at 145.5MHz, Zo is very close to 75Ω.

Above, the calculation returns 0.078dB/m, or 7.8dB/100m. That is very close to datasheet loss.

Sometimes, when you have a hammer in your hand, every problem you see looks like a nail. The method described here is a different application of the VNA (and could be done with a one port analyser or other instrument that can measure Z), but a bit of lateral thinking is needed to look beyond the obvious.

The method does not allow arbitrary choice of frequency, the cable must be resonant at the measurement frequency. That is not a great restriction when used for the purpose of checking that the cable is close to specification.

The method is simple, and does not involve uncertainty of tranformation elements etc.

]]>Users of some ATUs may have noticed particular sensitivity to hands on the capacitor adjustment knobs. It is a common problem with cheap implementations of the T match as the capacitor rotor is usually at high RF voltage and if that shaft is extended to the adjustment knob, under certain circumstances tuning becomes very sensitive to hands on the knobs.

In some of these implementations, if the users hand touches the metal grub screw in the knob, or the metal panel bushing behind the knob they may get a significant RF burn.

Let’s use the MFJ-949E as a discussion example. It is a T match, and the metal capacitor shafts in the knobs and panel bushings carry RF voltages.

So why is this only sometimes a problem?

The RF voltage across the coil, and impressed on the capacitor shafts can be extremely high when using loads with small resistance and large negative reactance, more so on the lower bands.

Let’s explore a load of 10-j500Ω @ 3.6MHz, it is in the range of what might be experienced at the base of a shortened vertical, but it might also exist at certain points on a multi band antenna feed line such as a G5RV.

Above is a simulation in W9CF’s T Tuner simulator using values typical of the MFJ-949E. The selected inductance is the smallest L tap that will match this load, the values of the capacitors are that needed to match the load. Note the value of the input C is smallish (29.1pF) and calculated tuner loss is 49% and efficiency is 51%. Small input C causes high internal voltages.

If we have knowledge of the relationship between knob calibration and capacitance and inductance, we can make a pretty good estimate of efficiency using T Match efficiency estimator.

Plugging the relevant values into T Match efficiency estimator we get…

It is an approximation, but it is very close to that from the T match simulator.

The efficiency in this case should be very concerning, it means that for 100W average input power, the internal dissipation in T match elements (mostly in the inductor) is 35W… and it will not withstand that dissipation without heating and softening the styrene coil supports permanently damaging the coil.

The calibration curve of the capacitor and inductor knobs can be obtained by measurement.

Above is my measurement of the total circuit capacitance of the capacitors in the MFJ-949E vs the knob calibration. The linear model is C=228-20*setting(pF).

My measured inductance values are above.

Above is a Simsmith model of the ATU with 10-j500Ω load and 300W input. Capacitor Q is assumed to be 2000 and inductor Q is based on measurement.

Note that the power dissipated in the inductor is 140W, way above what it could withstand continuously.

The voltage across the load is 1944Vrms, the voltage across the capacitors is C1: 3750Vrms, 5303Vpk and C2: 1793Vrms, 2536Vpk, voltage across L1 (and C2 shaft) is 3736Vrms, 5284Vpk. The ATU would not withstand these voltages and would not handle 300W input on the example load, and would be unlikely to handle more than about 30W input continuously to avoid melting the coil supports.

Operation on extreme loads causes extreme voltages and high internal dissipation.

The sensitivity to touch is a hint of a more sinister problem that extreme voltage exist inside the ATU and that may drive damaging coil losses.

The efficiency of a T match on the lower bands can be well estimated using T Match efficiency estimator.

]]>Common advice given by online ham experts include:

- it just cannot be done, the best (only) point to measure an antenna is at the feed point;
- it can be done, but only with an integral number of half waves of feed line;
- use the port extension facility in your software;
- use software package x;
- do an OSL cal with the feed line being part of the fixture.

That is simply wrong.

This usually assumes that the impedance at the load end repeats every electrical half wave. This is true for lossless lines, but there is error in applying that assumption to real world lines and the error may be significant. If a range of frequencies is measured with a fixed length line, then further error is introduced as the line is not exactly a half wave (or multiple) at all frequencies.

It is classic ham pseudo science.

Some software includes a facility to apply a constant time offset to measurements (at the calibration Zo). This addresses the multiple frequency problem described for the last method, but it does not address the loss or more grossly, a different Zo.

Some software packages do de-embed transmission line using a loss model for the specified line. Trust these only to the extent that you understand the accept that the loss model is appropriate to your application.

Most packages I have evaluated use a transmission line model that I would prefer to not use for many purposes, so I tend to not turn to those packages.

I do use Rigexpert Antscope 4.2.57 which has a simple add/subtract line facility which I used with line data derived from TLLC, it is quick, interactive, convenient and effective. It is the only off the shelf software package that I use for this purpose.

I do use custom spreadsheets, and PERL, Python, and iPython scripts for de-embedding.

This does work, and it does properly place the reference plane at the desired measurement terminals.

It may not be convenient or even possible in some scenarios, but can be an excellent solution.

If you are able to save the calibration files and restore them at a later time, new measurements can be compared with historical archives and it becomes more practical as the calibration is a once only operation.

Unfortunately, some VNAs and analysers invalidate old version calibration files apparently with disregard for their possible value… so don’t count on this method beyond the current measurement set.

The following examples show different techniques and an explanation of why the transformation was useful.

In these examples a scan of a UHF mobile vertical was made using a nanoVNA looking into 4.025m of Belden 8259 coax and saved as a s1p file.

This is a case where it is not practical to measure directly at the feed point as the feed point is hidden between the vehicle roof and roof lining. Nevertheless and understanding of the feed point characteristics is useful, especially more complex antennas as such as this ‘gain antenna’.

Above is a Smith chart of the actual measurement data looking into the feed line.

Above is the measurement data with the transmission line de-embedded using custom Python scripting and graphics to transformed the saved raw .sp1 data.

Antscope provides a similar facility with its add/subtract cable feature.

Above is the actual measurement data rendered in Antscope v4.2.57.

Above is the Smith chart plot with the known transmission line ‘subtracted’ using its interactive facility. This used the same transmission line data as in the Python scripts, but the internal algorithms are unknown.

Above is the transformed s1p file from the Python scripts rendered in VNWA, it does not have a facility to de-embed transmission line of this type (its port extension feature does not model line loss effects).

These graphs are based on measurements made with a Rigexpert AA-600, and de-embedded in spreadsheet calculations. In this case, it was simply convenient physically to measure looking into a short length of feed line.

Above is a set of measurements looking into 3.2m of Belden 8262 coax.

Above, the coax de-embedded to give the feed point Z which give a clearer understanding of the antenna behavior.

These are measurements of a 7MHz vertical using a MFJ202B noise bridge through 25m of LMR400 coax. In this case, the feed point was covered in snow and direct measurement was not convenient. Note the instrument (which needs a receiver), it takes quite a long time to make a measurement set like this ‘by hand’ and it was much more convenient to do this from indoors and de-embed the feed line effects.

Above, the actual measurement data. Minimum VSWR is just below the band, but why?

Above is the measurement with the coax de-embedded to give a good estimate of the feed point impedance. This perspective shows that R is changing relatively slowly with frequency, and X is just too negative, it is classic behavior of a near quarter wave vertical and in this case, the vertical needs to be a little longer or include some series inductive loading.

There are methods to de-embed feed line, they each bring their own uncertainty and have their own practicality, repeatability, etc.

]]>