UltraWideBand antenna design write-up - pedantry required

[hagster]

Hi All,

I have recently been writing up some blog posts on the selection, design and performance of an UWB antenna designed for use with SDR radios.

https://hexandflex.wordpress.com/2017/12/03/making-an-ultrawideband-antenna-pt1/

https://hexandflex.wordpress.com/2017/12/06/making-an-ultra-wideband-antenna-part-2/

Part 3 is in the works and will provide data on the Radiation Pattern, Gain, VSWR etc. Is there anything else that you would particularly like to see?

I am keen for the information in these posts to be both accurate and easy to understand(at least for someone with a basic of knowledge about antennas). As such I would be keen for any comments or corrections, so I can update and improve this write-up. Don't spare my feelings

[edit] Part 3 here https://hexandflex.wordpress.com/2017/12/11/making-an-ultra-wideband-antenna-part-3-performance/

[Edit 2] I have placed a small number for sale on Tindie. https://www.tindie.com/products/11151/

[jgalak]

What's the frequency range on this design?

[hagster]

What's the frequency range on this design?

Its got gain from about 1.7GHz, but is usable as an omni down to about ~600MHz where it starts to fall off a cliff.

Some Figures for examples

Gain
600MHz Gain = -3dBi
1GHz Gain = 0dBi
1.7GHz = 6dBi
6GHz = 10dBi

SWR is better than 2:5:1 from 1.6GHz to at least 5.4GHz(limit of the VNA I used)
SWR better than 4:1 from ~600MHz

[Kire Pûdsje]

Do not know your timeline. I would be able to comment, but not for the next 2 to 3 weeks. No major mistakes were made concidering the audience. I could add some remarks and explanations here and there. If it fits your timeline, and I do not react, by then, please send me a reminder, as I will no doubt have other distractions.

FYI, I have a little bit of experience with Vivaldi's (understatement). I was one of the lead engineers of the Embrace system http://www.astron.nl/dailyimage/main.php?date=20090803.
BTW, the latest version of this system uses ordinary FR4 as a feedboard and has O.o.m. 30 K Trec @ room temperature (0.4 dB NF).

Edit:
If you have access to "Microstrip Lines and Slotlines", 2nd ed by K.C. Gupta, et al, in chapter 5.6 it discusses all kinds of slotline transitions. I could prepare you a drawing showing the evolution of the transition with the radial stub and cavity.

[hagster]

Kire, that would be brilliant. I dont really have a timeline, so Im not in a rush.

That array looks amazing. I bet it was fun putting that together..

I have just been looking at one of the EMBRACE papers on the array design. Very interesting. I'm curious how the antenna in your design works down to such low frequencies with such a small apperture. I had thought that the apperture should be at least lamda/2.

I will see if I can borrow a copy of that book. Like many technical books the price on amazon is a bit much for a casual purchase.

[Kire Pûdsje]

Please remember that the Embrace design should be placed in an array environment. Stand-alone its performance is not great.
See http://astron.nl/dailyimage/main.php?date=20150903 for some results outside of the array environment. This was a modelling exercise, to see if the modelling was done correctly. Based on this modelling, we trusted the array simulation.

In an array environment, as a first order approximate, the height of the element should be at least lambda/2 at the lowest frequency. The pitch should be lambda/2 at the highest frequency.
At low frequencies, the coupling to the other elements really help to extend the performance. Do not know if the terms active and passive  impedance make sense, but it boils down to the fact, that in an array you are only interested in the impedance when all elements are excited. The array had a return loss of 10 dB.

By the way, have you seen my PM?

[hagster]

Ah yes, i forgot about the spacing limitation at  upper frequency.

Im going to have to do some thinking about how the s11 improves when its in an array.

In the long run i have a half plan to build some wide band splitters allow my antennas to be combined into an array and i am also thinking about making and antenna version with a built in LNA(possibly phantom powered).

[hagster]

Part 3 now online

https://hexandflex.wordpress.com/2017/12/11/making-an-ultra-wideband-antenna-part-3-performance/

[Kire Pûdsje]

You have a gain plot. The label is realized gain. Do not know if you mean the gain realized by you, or the realized gain in the IEEE sense

2.321 realized gain. The gain of an antenna reduced by the losses due to the mismatch of the antenna input
impedance to a specified impedance.

I think in your case both definitions coincide.
For gain, you are allowed to remove the mismatch loss.
Another interesting graph is the effective area. which for this kind of antenna should flatten out. A_eff = lambda^2/(4*pi) * G. The low end should compensate.

Finally, to include an LNA directly at the antenna, there is no need to match it to 50 ohms. The design goal, is to keep the noise match within reason, finding a good compromise between the Gamma_ant, Gamma_opt and R_n. Please note, that when looking purely at the reflection of the antenna, its mismatch loss does not translate to increase in noise figure.

[hagster]

Hi Kire. As you noticed the gain quoted is the 'Realized Gain' and does indeed include all losses. I think that this is the figure that he majority of people will intuitively expect. Actually, I suspect that the error bars on my measurements are probably bigger that the difference between gain and realized gain.

I will probably stick with 50ohms for the an LNA board. I think that you are correct that a improvements could technically be made by diverging from this limitation, but it would incur additional design and measurement effort. Additionally, if I change the impedence of he antenna, I would need to repeat all he measurements again. Hopefully I can just pick a good LNA and follow the app note.

[Kire Pûdsje]

For purely the antenna, realized gain tells you in general how the antenna would behave in transmit, I agree.
For receive performance, at the low end, it would make a lot of difference, depending on which definition you use. Since mismatch loss is not a dissipative loss, it should not be as much of a contributor for the system performance G/T or A/T (besides the Rn effect).

Some hints
Be aware that your chosen antenna topology behaves like an E-field probe for low frequencies, still being able to pick up signals all the way to almost DC. If you use a discrete transistor design, the antenna performance drops at lower frequencies, but in general transistor gain increases tremendous at lower frequencies.
Also one of the blades is connected to the LNA input. A say 10k bleed resistor (only 2K increase in noise temperature) across the input might also prevent building up static on the antenna, possibly destroying the input cap in the end.

[cdev]

I would be interested in seeing it!

What are your thoughts on DIY baluns for UWB in the range below? Ive tried making a number of different kinds of transitions. Ive had the best luck with "traditional" (very small) transformers. (Mini-Circuits)

Also, do you know of any DIY or low cost SMD 90 degree hybrids ideally that are broadband most of what I fool around with that would benefit from extra arms - maybe  1500-2300 Mhz ?

Ive got a bunch of antennas I want to try out.

Thank you!!!





 

Do not know your timeline. I would be able to comment, but not for the next 2 to 3 weeks. No major mistakes were made concidering the audience. I could add some remarks and explanations here and there. If it fits your timeline, and I do not react, by then, please send me a reminder, as I will no doubt have other distractions.

FYI, I have a little bit of experience with Vivaldi's (understatement). I was one of the lead engineers of the Embrace system http://www.astron.nl/dailyimage/main.php?date=20090803.
BTW, the latest version of this system uses ordinary FR4 as a feedboard and has O.o.m. 30 K Trec @ room temperature (0.4 dB NF).

Edit:
If you have access to "Microstrip Lines and Slotlines", 2nd ed by K.C. Gupta, et al, in chapter 5.6 it discusses all kinds of slotline transitions. I could prepare you a drawing showing the evolution of the transition with the radial stub and cavity.

[Kire Pûdsje]

What are your thoughts on DIY baluns for UWB in the range below? Ive tried making a number of different kinds of transitions. Ive had the best luck with "traditional" (very small) transformers. (Mini-Circuits)

Also, do you know of any DIY or low cost SMD 90 degree hybrids ideally that are broadband most of what I fool around with that would benefit from extra arms - maybe  1500-2300 Mhz ?

Sorry, not being a native English speaker I do not completely get what you mean about the extra arms?
As for the 90 deg hybrid. It all depends on your requirements, a diy solution would be a double branched line coupler. One approach would be to take the filter approach. Another would be to start with e.g. the ones from https://www.microwaves101.com/encyclopedias/branchline-couplers#double and then start tweaking in a simulator (Qucs?). Do not forget to model the T-junctions.
Another approach would be to use two tandem couplers of 8.34 dB. But as said, it all depends on your requirements.

[hagster]

Another interesting graph is the effective area. which for this kind of antenna should flatten out. A_eff = lambda^2/(4*pi) * G. The low end should compensate.

Yes this is an interesting graph. The average Effective Area is about 45cm^2 which seems intuitively about right for a 9cm tall antenna. I also used this figure to see what the gain would look like if the average A_eff was constant.

[Kire Pûdsje]

The linear scale makes it always look much worse than actual. I am generally used too seeing dB m^2 (dB square meter) plots, but I think you found a very nice way by inserting a gain line based on a fixed effective area. Effective area is (I think) a concept that will be hard to understand by the average reader of your blog and they are more used to gain.
A lot of times people focus too much on gain. Especially for these kinds of wideband antennas, it is purely physics. The antenna will only collect the power density that is 'falling' on the antenna. (P_out(av) = A_eff * S). There is little you can do to the collecting area (effective area). Therefore the gain behavior over frequency is almost fixed by nature. Your graph is a perfect example that the antenna is a very close approximation to fixed effective area.
The way you presented this (maybe move the effective area to a different graph), will be a good explanation for your blog as to why your gain rolls of at lower frequencies. A view of the antenna with a circle/square indicating the effective area would also make it possible to compare the effective area to the scale of the antenna.

[hagster]

That's a really good idea. I think I can add something like that fairly easily.

[hagster]

I have placed a few of the extra Antennas I built for sale on Tindie if anyone is interested. https://www.tindie.com/products/11151/

[hagster]

If anyone is interested this antenna was proffesionally tested 8n an anechoic chamber. Results are here http://antennatestlab.com/antenna-examples/example-7-antipodal-vivaldi-antenna-exponential-slot-edge-pcb?utm_source=HexAndFlex

[Neuromodulator]

That project looks interesting, did you use any antenna design software? It would be interesting to learn how to design a Vivaldi antenna based on antenna requirements (swr, gain, etc).

[hagster]

That project looks interesting, did you use any antenna design software? It would be interesting to learn how to design a Vivaldi antenna based on antenna requirements (swr, gain, etc).

I can neither confirm nor deny the use of any software. I am slowly trying to learn OpenEMS, which should be capable of this kind of simulation.

That said, the process of optimising for Gain/SWR is the same for both EM software and handcrafted prototyping. Generally the EM software takes a long time to calculate solutions over such large frequency sweeps, so doing it experimentally by hand is not a great deal slower so long as you have a good VNA.

First, you can divide the problem up into the feed and the antenna exponential and the corrugations.

Feed
To do this you can first start with a SMA to microstrip to SMA PCB (or copper tape on suitable thickness FR4). Then with a VNA connected you can cut out the ground plane such that it transitions to a broadside couples twinline. Experiment with different tapers and see how this affects the S11 and S22 (you can make do with an S11 measurement and a 50ohm terminator). There will be an upper limit where the S12 starts to drop off and the S11 increases. I have attached a picture of roughly what this test board will look like.

Once this is done, you have nailed down half of your variables and can be confident that the feed works.

Antenna
The exponential opening of the antenna is defined by a formula with an expansion coefficient.

I have uploaded an excel sheet that can generate these curves herehttps://hexandflex.files.wordpress.com/2018/03/vivaldi-curves-hexandflex.xlsx. The value to adjust is alpha(I have spelt is aplha in my sheet).

You will need to guess at the rear cutout of the antenna. I just used an arbitrary oval shape. It is less critical than other factors and you can play with this later if needed.

You can then build and measure various opening coefficients (alpha). Either using a new antenna for each design or starting with the highest alpha first and trimming the antenna down each time (document all measurements, so you can go back).

Corrugations
Once you have picked your optimal opening coefficient you can they experiment with corrugations on the arm edges to try and improve gain/reduce sidelobes. The pattern for the original palm tree antenna IEEE paper and my version seem to work well. Perhaps, you can find a better pattern for these cutouts and create a new category of Vivaldi antenna. The only examples I have seen in literature are rectangular slots and these palm tree cutouts.

If you do give this a go, I would love to see the results.

[Neuromodulator]

Thanks for the detailed response. I found matlab comes with this antenna toolkit which is very nice as it comes with many default antennas, including Vivaldi. Problem is that as you say, its super slow.
Sadly I don't own a VNC, so I can't handcraft antennas to adjust/test them manually. I also found lot of info on vivaldis on IEEE, maybe at some point I'll see what I can build, but its starting to look too annnoyingly laborious.

I wish there were "open antennas" around which anyone could build, as of now buying these PCB antennas is very expensive. For instance rfspace has some nice antennas, but their pricing is crazy. I'll probably  buy yours at some point, as it looks nice for the higher frequency range.

[hagster]

Thanks for the detailed response. I found matlab comes with this antenna toolkit which is very nice as it comes with many default antennas, including Vivaldi. Problem is that as you say, its super slow.
Sadly I don't own a VNC, so I can't handcraft antennas to adjust/test them manually. I also found lot of info on vivaldis on IEEE, maybe at some point I'll see what I can build, but its starting to look too annnoyingly laborious.

I wish there were "open antennas" around which anyone could build, as of now buying these PCB antennas is very expensive. For instance rfspace has some nice antennas, but their pricing is crazy. I'll probably  buy yours at some point, as it looks nice for the higher frequency range.

From experience, many of the 'open antenna' designs on the web don't really perform as well as promised. Some of this is down to manufacturing tolerances, or difference in materials. Without somebody correctly measuring them the loop never closes and errors are never corrected. That said, many 'expensive' commercial antenna's are trash too.

The RF Space antennas you mention are very good. Also, I have good reason to believe the www.wa5vjb.com antennas are very good (and a bit cheaper than me).

For DIY. If you calculate the correct micro-strip width(for your substrate and thickness) and just roughly copy and scale up/down the rest of my design to meet your desired frequency of operation, I dare say you will get an antenna that works reasonably well. It won't be perfect an you won't be able to design it to meet any particular SWR or gain values or be able to accurately tell what these are, but it will be a wideband antenna. Most Alpha coefficients will work to a reasonable level.

[cdev]

The #1 selling point for UWB designs for me is that with many geometry based "frequency independent" construction you often just have to get the shape right, to get a fairly decent antenna, which is easy..

[hagster]

The #1 selling point for UWB designs for me is that with many geometry based "frequency independent" construction you often just have to get the shape right, to get a fairly decent antenna, which is easy..

I agree. For narrow band antennas it is very easy to get the frequency slightly wrong. In fact just putting it near some plastic can easily lower the resonant frequency and turn a good efficient antenna into a very poor one.

The advantage of narrow band antennas is that they also behave as your first stage of frrquency filtering.

[dazz1]

Hi
Just reading your blob.  I like it.

Code: [Select]

A further improvement to the Antipodal antenna aims to improve the polarisation purity. The BAVA antenna uses a 3 layer PCB with one element sandwiched between two layers of board. The effect is to make the antenna symmetrical.
What about using a standard 4 layer board with the middle 2 being identical?  The two inner layers might then look electrically the same as a single middle layer of 3.

Dazz

[dazz1]

Hi

It is possible that the curved lens edge of your board is also reflecting back to the feed at resonance.   In visual optics, lens are coated to improve light transmission.  You could achieve a similar effect by cutting saw tooth edge into the curve.   A bit like an anechoic chamber.   

This should be fairly easy to test (a few minutes with a hacksaw would do the job).  I suspect the effect will be small but no one will know unless someone measures it.

Also, there is obviously a lot going on RF-wise on this antenna.  You could try selectively adding patches of RAM to different parts of the antenna (especially the outer edge) to see what effect individual features have on radiated performance.  Just a thought.

Dazz

[hagster]

Hi
Just reading your blob.  I like it.

Code: [Select]

A further improvement to the Antipodal antenna aims to improve the polarisation purity. The BAVA antenna uses a 3 layer PCB with one element sandwiched between two layers of board. The effect is to make the antenna symmetrical.
What about using a standard 4 layer board with the middle 2 being identical?  The two inner layers might then look electrically the same as a single middle layer of 3.

Dazz

I see no reason why that would not work. It may have an impact of efficiency as more of the fields are within the lossy FR4. Also i'm not clear how I would add the connector.

It is possible that the curved lens edge of your board is also reflecting back to the feed at resonance.   In visual optics, lens are coated to improve light transmission.  You could achieve a similar effect by cutting saw tooth edge into the curve.   A bit like an anechoic chamber.   

This should be fairly easy to test (a few minutes with a hacksaw would do the job).  I suspect the effect will be small but no one will know unless someone measures it.

Also, there is obviously a lot going on RF-wise on this antenna.  You could try selectively adding patches of RAM to different parts of the antenna (especially the outer edge) to see what effect individual features have on radiated performance.  Just a thought.

I don't think there are many reflections heading back into the antenna. This would show up in the SWR measurements.

Your suggestion sounds like creating a soft transition between the Er of the substrate and air. It would be an interesting thing to test if I had more time.

I did look at creating a more aggressive lens using a saw-tooth like pattern(fresnel lens). It does on paper seem like it might work, but becomes a bit ugly.  In fact as the FR4 represents only a thin 2D slice of the propagating field, it seems likely that any changes will only have a very minor effect.

The other thing to try is an ellipsoidal lens. There are a few examples of this sort of thing around and it could be easily tested with some 3D printed shapes. This would allow the beam to be focused in azimuth as well as elevation, but the antenna does become bigger.

[dazz1]

What about using a standard 4 layer board with the middle 2 being identical?  The two inner layers might then look electrically the same as a single middle layer of 3.

I see no reason why that would not work. It may have an impact of efficiency as more of the fields are within the lossy FR4. Also i'm not clear how I would add the connector.

I don't think the FR4 between two middle layers(2,3) would have any effect because if they are identical there should be no voltage between the layers.   It wouldn't be difficult  or expensive to test a prototype.

There could be benefit of using the 2 inner layers. They will add thickness which will add a 3D element to an otherwise planar antenna. This might affect the antenna beam pattern.

Connection to layers 2&3 could be achieved by using a pattern of vias to connect 1&2 ,3&4.    This could be done with blind vias, or more cheaply with full through vias.  The connector could then be soldered to layers 1&4.

I don't think there are many reflections heading back into the antenna. This would show up in the SWR measurements.

I agree.

Your suggestion sounds like creating a soft transition between the Er of the substrate and air. It would be an interesting thing to test if I had more time.

I did look at creating a more aggressive lens using a saw-tooth like pattern(fresnel lens). It does on paper seem like it might work, but becomes a bit ugly.  In fact as the FR4 represents only a thin 2D slice of the propagating field, it seems likely that any changes will only have a very minor effect.

I don't think that the edge shape of the board will have a significant effect because the % of radiated power that reaches the edge of the board (as opposed to the power in free space) will be small.   I don't think a fresnel lens would work at this scale because the lens elements would be too small compared to the wavelengths.  I think it would be more likely to act as a diffraction grating.

Your measured data shows that the antenna works.  Any other features/modifications are likely to provide small incremental improvements.
My old Professor used to say if it isn't measured it isn't true.
The problem with my suggestions is that they are untested, not quantified and therefore largely worthless speculation.

[hagster]

Your measured data shows that the antenna works.  Any other features/modifications are likely to provide small incremental improvements.
My old Professor used to say if it isn't measured it isn't true.
The problem with my suggestions is that they are untested, not quantified and therefore largely worthless speculation.

Your professor was very wise. Measurement is the time consuming bit.

The suggestions are not worthless at all though. They are all good ideas that are worth exploring.

[dazz1]

Hi
Well let me make another suggestion.

I am highly skeptical  about the effects of the palm leaf edge treatment for the simple reason that the palm leaves have no energy dissipation.  Assuming no ohmic losses, any energy that enters those leaves is going to be reflected out again.    If they are radiating energy, you would probably want to know if that has any effect on your antenna performance.   

I suggest the following test sequence:

1.  Make a prototype with straight edges (no palm leaves) and no lens edge. Test and measure. This is the simplest form of your antenna and can be used as a baseline design.

2.   Get some ferrite loaded epoxy or similar and apply along the  outer edges.  Test and measure.   This should be very effective in dissipating any energy that tries to escape down the outer edges where it is not wanted. It should effectively eliminate most parasitic currents.   Adding this material is unlikely to be something you will want to do for production but it should quantify the effect of edge current (radiation) when compared to the untreated PCB edge. 

3.  Take one of the palm leaf antennas and figure out how to add a matched resistive termination at the feedpoint of each leaf. Test and measure. The aim is to absorb any power going into each palm leaf to prevent it being reflected out.    If the termination is a SMD resistor and if this proves effective, that is something that would be easy to added in production. 

These are relatively simple and cheap tests.  If you follow this path, you will end up with a whole lot of data you don't currently have.  When analysed, this data should tell you a lot about the effect (if any) of antenna parasitics and their effect on your antenna performance.

I suspect the greatest potential improvements could be made by refining the main antenna feed point geometry (matching).   Optimising that efficiently probably needs access to a super expensive simulator where multiple geometries can be tested with relative ease. 

Dazz

[hagster]

The following images are from Laboratorio Maxwell where the original creator of the Palm Tree Antenna Dr. Alexandre works.  http://labmax.org/index.php/2016/11/04/labmax-tem-trabalho-de-pesquisa-premiado-no-encom-2016/

You can click on the picture at the bottom of the LABMAX page to see a follow up IEEE article. I can't post the original i'm afraid.





As you say, the cutouts are not absorbing any power as they have no resistive components. Instead they create a wide band high-Z path to reflect the signals back towards the main lobe. I believe the original author arrived at his exact geometry through experimentation. I did the same, but found that in my case(with different substrate and frequencies of interest) I got better results with less leaves.

When I get a chance I will try and do a real world test of this by creating a version without the leaves and do a side by side measurement.

[dazz1]

Hi
Fascinating stuff.  That simulation shows with a minute of study what might take days to find by meticulous measurement.

I can see from the simulation and paper that the palms are being fed rather than just getting parasitic energy escaping from the main antenna feature.

Taking the leaf concept further, I think there would be merit in investigating the antenna conceptually illustrated in the attachment.   This consists of a central radiating feature flanked top and bottom by two secondary directors.  The three are driven by three separate feeds (blue) from the same source.  For the directors to work, they should radiate slightly ahead of the central radiator to focus the main beam.  The central feed needs a small delay line (green) to achieve the correct phase shift compared to the two directors. 

This arrangement would effectively create a passive planar 3 element phased array antenna.  The delay line would need to  be wide band constant angle phase shifter.    There would be enough R&D work here for a solid thesis.

Dazz

[hagster]

Dazz,

It took me a while to get my head round your idea. Your concept is trying to make the emerging field squarer by delaying the phase of the central section. As you say, creating a constant phase delay over a wide-band is difficult.

Hagster

[dazz1]

Dazz,

It took me a while to get my head round your idea. Your concept is trying to make the emerging field squarer by delaying the phase of the central section. As you say, creating a constant phase delay over a wide-band is difficult.

Hagster

Hi
It is not a new problem.  Just search for "Heaviside Condition". 

Dazz