The Vivaldi Antenna
My chosen antenna design is a version of the Vivaldi antenna. The Vivaldi was named by fellow Brit Peter Gibson. It is believed that the name came out of Gibson’s love of classical music, and in may ways the Vivaldi antenna resembles the horn of a brass instrument.
Why choose this antenna? Well primarily it can be manufactures cheaply with conventional printed circuit technology. This allows me to have it built to a high level of quality and repeatability. Secondly the antenna is capable of incredible bandwidths and also has a reasonable level of gain. In fact, I don’t think there is another antenna in this form factor that can provide more gain over a similar bandwidth.
There is a really great description of this antenna over on Antenna-Theory.com where Pete goes through the process of building one using the time honoured technique of cutting copper tape with an exacto knife. What Pete shows is just how easy these antennas are to construct. The difficult part is optimising the performance so it doesn’t just work, but works well. Here, i will describe some of the design options available and the choices I made.
As described in my previous post, i like to understand how an antenna is structured compared with other antennas. In this vein there are various ways to look at this antenna.
It could be considered to be derived from a Tapered Slot antenna (again see my previous post), in this situation the linear taper has been replaced by an exponential taper. While, im sure there is some maths that explains why an exponential taper is a good choice for this shape, in practice it produces an antenna with a wider bandwidth. You can change the rate of the exponential, and in my case this is what I did to find a value that optimised gain and SWR.
Another way to look at it is as a PIFA antenna which has had the open end flared to increase the bandwidth.
The feed is often the limiting factor of the Vivaldi antenna bandwidth as well as other factors such as polorisation purity and efficiency. The aim of the feed design is to efficiently transfer the power from a (normally) coax transmision line into the antenna with as few reflections and little loss as possible. Below, I describe a number of potential feed mechanisms and explain the decisions I made as part of my design.
Direct Coupled Vivaldi Antenna
In the example above, the antenna is fed with a simple coax feed accross the slot. One of the problems with this is that the optimum distance of the feed to the short end of the slot changes with frequency. Hence there is no single correct distance amd a compromise must be chosen. This in turn limits the bandwidth.
Often the next trick employed to improve the bandwidth is to place a hole at the end of the short. Quite why this works is not totally clear to me, but if it works it works and I have spent enough time with an exacto knife and copper tape to be satisfied that it does. That said, it still seems like a process of trial and error to get the right shape and size.
Microstrip Coupled Vivaldi Antenna
As a further improvement frequently used to improve the feed is to use a microstrip coupled fed. In this technique a microstrip transmission line on the other side of the substrate cuts across the slot. Using the microstrip feed allows the designer additional degrees of freedom to tweak the impedance match. This is a bit of a double edged sword as the additional degrees of freedom dramatically increases the search space for just the right combination to meet the design requirements.
Despite this additional design complexity this antenna is probably the most common that I have seen implemented.
One advantage of both this and the direct feed is that it allows both upper and lower elements to be on the same side of the substrate, or even cut directly from sheet metal without a substrate at all(except perhaps for a mirostrip feed) This creates a well balanced structure with a high degree of symmetry in its radiation pattern.
Antipodal Vivaldi Antenna
The antipodal Vivaldi Antenna is the antenna I chose for my design. The AVA antenna uses a clever tapered feed to gradually transform an unbalanced coax microstrip transmission line into a balanced microstrip line. As show this is achieved by gradually reducing the width of the ground plane(a) until it matches the width of the microstrip(b). This then gradually merges into a overlapped slot-line(c) as the throat of the Vivaldi opens up. So long as all these tapers are done gradually with respect to the wavelength they do not generate significant reflections. This feed results in the highest bandwidth of all the mechanism described in this post.
It should be noted I believe that there is probably room for improvement here. In my design the balanced microstrip line uses two conductors of the same width as the original 50Ω feed from the input connector. The impedance of this new transmission line will be somewhat greater than 50Ω. In theory it should be possible to gradually make the tracks wider at this point to maintain 50Ω for longer and hence increase the lower frequency performance in particular. As the later transition to overlapped slot line occurs over a longer distance as dictated by the opening of the antenna throat any reflections caused by this are likely to be less. I haven’t yet found any papers that describe this additional optimisation, but if anyone does then please leave the details in the comments.
Balanced Antipodal Vivaldi Antenna
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.
The downside is that this design requires a very custom PCB stack-up. While 2 layer and 4 layer PCBs are fairly common. 3 layers is very rare. Due to the additional manufacturing complexity and cost I avoided this design.
The choice of substrate is the next important decision to be made. Many RF engineers will immediately tell you that standard FR4 should not be used for frequencies above 2GHz, but I think it is not so clear cut.
- Dielectric constant (εr)
- Loss Tangent(Tanδ)
- Mechanical strength
- Cost (££££££)
Dielectric constant dictates the effective speed of light(RF) in the medium. Higher values make the antenna look smaller electrically. High performance RF boards will have a very tightly controlled Dielectric Constant and can provide detailed data for this over a range of frequencies. Cheap FR4 is likely to be much less controlled and vary from batch to batch and across frequency. In fact you may be lucky if a manufacturer will provide any figure at all for this, and you could be forced to guess at generic value.
The thickness(along with the Dielectric Constant) will dictate the trace width of a 50Ω transmission line. There are plenty of tools that will let you work out this width. What we really want here is a material and thickness that lets us choose a sensible width. Too thick and our tracks will be very wide and hard to connect to an RF connector. Too thin and the tracks will become very thin and be overly sensitive to manufacturing errors. Additionally if our tracks are too thin then the electrical resistance be too great and limit the efficiency and power handling capability. Remember that skin effect is very significant in the GHz frequencies so we really want a large surface area for our tracks if possible. Again, high performance RF substrates will have a carefully controlled thickness, and generally manufacturers that use this understand the importance of controlling this carefully.
The Loss Tangent is a measure of how much energy a material will absorb. A high loss material will reduce the antenna efficiency and power handling capability. The loss is directly proportional to frequency and hence many RF engineers will normally tell you not use normal FR4 material at GHz frequencies.
Mechanical strength is also important for an antenna, especially when iyou plan to use it without a protective enclosure. Here good old FR4 has your back. It is tough as old boots.
Cost can not be ignored. Choosing a purpose made high performance RF substrate will both increase the cost and reduce you supplier options. I quickly ran some checks with the Price Calculator on Eurocircuits website(one of the few companies to offer a cheap RF pool service). For a 100mm x 80mm 2 layer board the costs come out as ~£43 for Isola ITera, £53 for Rogers 4350, while FR4 comes in at about £18. These numbers are for each PCB when buying in a batch of 5. Eurocircuits are really great btw, and I do recommend them, but by going with OSHpark or a cheap chinese supplier you can really start to get that cost down if you can tolerate generic FR4 material.
As you can see, while other factors are important, there is a really strong motivation to use FR4 if possible. So, for my design, I took the informed choice and am using FR4. To help me come to this conclusion I created my own spreadsheet based upon the calculations shown on Microwave101’s RF encyclopedia and discovered that because the actual length of the microstrip feed in my design was relatively short (~20mm), the actual loss I would expect to get from using FR4 was really not that great. As the signal travels further down the throat of the antenna, my expectation was that the field will be less contained within the substrate and more within the surrounding air. Hence I expect over the length of the antenna the effective dielectric constant and loss tangent will significantly reduce. I do however want to reduce the amount of lossy FR4 in the design and additionally have both antenna arms fairly close together to help improve the polarisation. Hence I decide that a 1mm thick board would provide a reasonable balance for my objective.
Further Improvements and The Palm Tree Vivaldi Antenna
The next couple of improvements come from examples I have seen in published papers(links below).
The first is the use of corrugations in the sides of the antenna arms. These work to suppress currents from flowing in these outer parts of the antenna arms. Any current flowing here will lead to radiation in directions other than the bore-sight, so suppressing this current is a good idea.
One technique uses rectangular cut outs choke off these currents in much the same way as a choke ring antenna. At RF these slots would look like a transmission line with a short circuit at the end. When the length of the cut out is approximately 1/4 lamda the open end impedance becomes very high. Due to this the rectangular cutouts are likely to only be useful over relatively small bandwidth. To combat this, normally multiple rectangular slots are created of differing length.
The technique i used comes from a paper I found here was developed by Dr. Alexandre and his Team at Laboratory Maxwell(named after the great John Clerk Maxwell). It uses exponentially tapered cutouts (of the same geometry as the antenna throat) to ‘in-theory’ suppress the currents over a wider band, and even provide some additional forward gain. The effect makes the antenna look a bit like a palm tree and the creators coined the term ‘Palm Tree Vivaldi Antenna’ , which is more memorable than their other name ‘Exponential Slot Edge Antipodal Vivaldi Antenna’ (ESA-AVA). Overall the boost to gain is not huge, but it does seem to work and it looks good so why not use it. I will admit to being a bit lazy and just using a circular curve rather than the recommended exponential, but my gut feeling was that there were marginal gains to be had and generating circles is much easier than exponential in my CAD, so screw it.
The next improvement is the use of a lens effect. This was achieved by cutting the end of the PCB into an arc rather than leaving it square. The PCB has a dielectric constant of around 4 so the electric field will travel slower in this medium than air. With the field travelling through more PCB in the centre than the edges the effect is to help focus the beam. Now that’s all great in theory, but the steepness of the arc on my antenna is fairly shallow and obviously the PCB is thin and hence does not encompass much of the EM field. With more effort I could have probably worked out the optimal curve for this lens, but again, my gut feeling is that I would have needed to double the length of the antenna to make the effect significant and at that point I would be better just making a bigger antenna.
I have left the small arc in the design as a token effort without dramatically increasing the size and cost of the PCB. I haven’t been able to prove that it helps, but again, it looks good and doesn’t hurt so screw it.
The Final Antenna
So, here is the final antenna. As can be seen, I went with a white solder mask to give it a unique look, and protect the copper tracks from tarnishing. I added a black silkscreen outline of the opposing antenna element. I had thought that the copper would be more visible beneath the white trace than it is. If I had known just how well the white soldermask would obscure the track, I may have done something different with the silkscreen.
The antenna is about 150x90mm. You can fit 4 of them on a sheet of A4 paper(or US letter size). I also added a few mounting holes with a 5mm spacing. These can be used to attach brackets etc, or even to attach multiple antennas together. I intend to design and 3D print some brackets and spacers in due course. These will be shared on my thingyverse page once tested.
In the next post I will show off some of the performance of this antenna. I am really pretty chuffed with how well it works.