Final Report for University of Alberta Collaboration

Dear Rambabu and Kevin,

Here is an a final email to summarize where things have ended.  As usual, you can find the entire thing on the research blog.

I have moved all of my equipment, including the vacuum chamber out of the University and have stored it.  Neil, my engineering friend, who was going to design the magnetron feedback loop has sent me back the network analyzer and the thermal camera.  He still has the magnetron, waveguide assembly, power supply and an oscilloscope.

As things stand, there are three phases left in order to get a high Q before any testing can commence:

  1. A working power supply with feedback loop (more below)
  2. The cavity – The cavity needs a couple thousand dollars of alterations (both suggested by Shawyer) to make it resonate with a Q of 50K.
  3. The two then need to be combined and the setup tested for deficiencies (i.e. the coax cable connecting the supply to the cavity may be prone to overheating).

As it turns out, the complicated feedback loop I had dreamed up may not be necessary because, as I have mentioned previously, it sounds like the magnetron gets “pulled” into the right frequency because of the strong resonance of the cavity.  How this works exactly, however, is not clear to me and I am not sure if some type of external feedback loop is necessary.  From the pictures of Shawyer’s demonstration engine, it would seem to have a computer controlled tuning wall at the back of the cavity.  If I had to guess – the feedback loop works like this:

  • The magnetron is turned on and allowed to warm up, the cavity is then tuned to whatever the magnetron frequency settles on (with the screw actuated tuning plate at the back).  The hard thing is that the resonant frequency bandwidth of a cavity with a Q of 50K is a couple Khz wide, but a commercial magnetron can wander up to 30Mhz on either side of a target frequency.  Once the frequency of the Magnetron gets close to the resonant frequency of the cavity, it is “pulled” into the right frequency because of a reduction in reflections (??).  I am pretty fuzzy on how it works.

Let me tie up a few loose ends mentioned in previous blog entries:

  • I decided not to go through with becoming a CNC programmer/operator for personal reasons – after talking to my working classmates, I realized I don’t want shift work.  Larger shops who can pay more usually run their machines in two shifts, one from 6am to 3 pm and one from 4pm to 12pm.   I have too many activities in the evenings to take on that type of lifestyle.
  • Although I simulated a bunch of simple “test cavities”, I never did get the OK from NAIT to use their CNC machines.

Here is a summary of where the cavity was left:

  • The unmodified cavity has a TE0,1,4 mode with a  Q of 2500, an insertion loss of 2.2dB and a resonant frequency at 2.449Ghz instead of the simulated 2.467Ghz and a Q in the tens of thousands.
  • I have created a model (v4) with Shawyer’s proprietary modifications and have simulated it with the Eignemode solver which resulted in a Q of 62K at 2.43Ghz.  The model doesn’t have probes or optimizations (i.e. moving the tuning plate) and needs to be simulated with the Frequency domain solver.

On the plus side, Shawyer asked me today if I wanted to talk to a “CEO of a Canadian space company” who wants to do a “demonstration test”.  As excited as I am by potentially getting paid to do this work, I don’t hold out much hope.

While the project has been winding down over the past year, I have been thinking about what I would have done differently and, I would go the other way with a larger cavity at a lower frequency.  It has a number of improvements:

  • The Q will be higher – as you get into smaller and smaller cavities, the same current gets concentrated onto smaller and smaller cavity walls, making the conductivity of the walls more important.  With a larger cavity, the same current gets spread out over a much larger area and even copper or aluminum can make for a fantastic Q.
  • I would make the tapered part of the cavity out of aluminum sheeting, but the end caps out of copper.  As I mention in my blog, simulation of even the 2.45Ghz cavity in Aluminum had little difference because the TE mode works such that circular fields are zero along the walls.
  • High power HAM radio equipment is readily available and relatively inexpensive compared to high-power 2.45Ghz systems and can pump up to 5000Watts of power.
  • HAM radio equipment is designed to output finely tuned frequencies with narrow bandwidths, unlike commercial magnetrons which have a large and wandering bandwidth.
  • Because the Mhz frequency has a longer wave length, it means the tolerances for the cavity are not nearly as tight as the Ghz range, which makes for less expensive cavities.

One downside will be a cavity measured in feet making it difficult to handle.

Machinist Training Almost Finished and Modified Cavity Simulations


  • The Faraday cage at the University is being replaced by a modern RF anechoic chamber.
  • I signed an NDA with Roger Shawyer and I am now simulating a cavity with the modifications suggested.

Since the last update, I am now two classes away from completing the “CNC Certificate”program at NAIT which means I will start applying for jobs as a “CNC Operator”.  With my strong programming background and 3D CAD experience, I plan to quickly transition into CNC programming.  In the last class, we cut an inside bore and then an external tapered NPT thread.

Tapered external NPT 3/4 thread in steel

I have been looking at local machine shops and have found a bunch I have short listed, both for working at and for building the parts I am currently simulating.  Here are few that stand out:

The great news is that all of them are within a twenty minute drive.

As for the cavity, as per the agreement with Shawyer, I can’t say much about the alterations necessary to get a high Q, but now that I have seen them, they all make a lot of sense.

On the simulation front, I tried using a modification of my previous model, but the results haven’t been satisfactory and I am rebuilding it from scratch.  The problem is that my previous model used a vacuum as the background material to which was added a thin skinned copper cavity.  However, the AKS Eignemode solver then treats the external air as a valid location for modes which are useless.  The new model will use a metal background into which an appropriately shaped vacuum will be inserted which means a a simpler model with fewer tetrahedral mesh cells.  The mesh cells only need to make up the vacuum and the probes.  For example here is one of my test structures from a month ago:

As an aside – I have decided not to build the test structures because, as confirmed by Shawyer, the Q will be too low (as is expected with small cavities at 6Ghz, the tolerances required a very tight and wall heating is high).  With the necessary alterations for the 2.4Ghz cavity, the test structures are redundant.  The test structures were really plan B if Shawyer decided not to respond.  NAIT has also not responded to my request to use their CNC machines, which is where I had planned to build the test structures.

Talked to Roger Shawyer


  • Machining coming along
  • Introduced Roger Shawyer of our work and he is going to run our machined cavity dimensions through his simulation software.

Although my last entry was a number of months ago, a few things have been getting done.  I have been enjoying my machinist training immensely and we have been learning how to make “O.D.” cuts or “Outside Dimension” cuts.  For example, here was our project from last weekend:

NAIT MAC302 Lab Object

This Saturday we are learning how to cut threads and then in MAC303, we are finally learning how to do ID cuts or inside dimensions.

I have also settled on my first simple test structure, a tube that should resonate in the TE0,1 mode at 6Ghz and have a Q of 25,000.:

Simple Round Aluminum Test Structure with two ports and a Q of 25K

It should be easy to create from a 3″ diameter 5″ long aluminum bar I can get from the local Metal Supermarket for $33.  The great thing about aluminum is that it is so soft that the inserts (the blade used to cut) for machining have almost negligible wear.

I also decided to cross that PONR or Point Of No Return and contacted Roger Shawyer directly.  We had an interesting conversation last week and I have point him to this blog.  We talked about the following:

  • Other universities have tried building cavities but ultimately failed where we have – their simulations don’t match actual results.
  • Chinese Results:
    • Have been one (of few?) to have reproduced results.
    • Yang Jung group have built actual cavities and have gotten results of 750mN using 3Kw power.
    • The work has gone black/military and moved to Beijing, no more papers are being published.
    • Yang Jung was “disappeared” for six months with the implication of instilling dogma.
    • No one outside of China has seen their cavity, although Shawyer did traveled there and given talks.
    • They rebuilt commercial simulation software which took 19 months in order to deal with the discontinuities of asymmetrical cavities.
  • A Big Obvious Enterpris-ING company is the US contractor Shawyer has been working with
  • Sounds like the trick to powering the cavity at the resonant frequency, is to tune the cavity to the magnetron after it has warmed up, and the because of reflection, the cavity will pull the magnetron frequency and lock it in.
  • A simply geometry will get the cavity into a Q of thousands, but some type of internal shaping is required to get Qs of tens of thousands and the reason is because of phase continuity or coherence has to be maintained.

I have posted a summary of everything mentioned in this blog on the Research Summary pages.

Magnetron Testing

As usual, I underestimated how much it would cost to get the cavity machined and I have three quotes back now, for $1600US, $1900US and $2150CND, all of which includes the aluminum.  None of these include copper plating which will run me another $300 or so and ironically, I haven’t gotten a quote back from Ryerson for whom I went to all the trouble of getting a 2D drawing made up.  Just for fun, I also got a quote through for a gold plated cavity and it came out to  be $6K, not bad!  I have been impressed with both their software which includes a mini-quoting system with many types of materials and finishes.  The design was easy to import because, first a 2D wire frame profile was loaded from a DWG file and then a revolve tool was used to make the cavity.  All told, I had my drawing imported and the quote generated in less than an hour, very cool.  I will probably end up going with them because their price was the most reasonable and I can include 0.005″ of copper plating too.  I have to confirm the quote with them yet though.

I was also up at the university for a day this week testing the magnetron output and in it’s current form, it is unusable.

First, here is the initial test setup:

We tested to see what the attenuation was on the coupling port on the circulator and it turns out to be -62.6dBm.  If the input power is 1000W or 60dBm, then the remaining power at the coupling port is roughly 0.6mW (-2.6dBm where -3 dBm = 0.5 mW)   We then hooked up the magnetron like this:

And attached the coupling port to another 70dB attenuator for safety and then hooked the output into a power meter which read between -58dBm and -52dBm depending on how long we left the microwave on.

We then hook the output up to the spectrum analyzer and got this (The picture should be an animated GIF):

The GIF contains about fifteen pictures taken over a thirty second test and here is a quick analysis:

  • The signal is intermittent – this is likely because the power supply is a voltage doubler which only has a %50 duty cycle, which are cheap to make but not useful for our tests.
  • The primary signal starts at roughly 2.466Ghz and moves downward – The cause is pretty simple, the magnetron is heating up, going from 33c to about 90c, most likely because of the short and reflected power. (Err forgot my thermal imaging camera at home, oops).  The center frequency is moving with temperature.

What we would have like to have seen was this:

A clean sharp spike at 2.45Ghz (The above was created by a frequency generator, and would be perfect for our needs, but look at the power!  mW!  Scotty, we must have more power!)
The magnetron still may be useful, but we will need a water cooled one to keep the frequency stable and a continuous wave power supply.  I will start searching eBay and other sources next week.  Meanwhile I will confirm the quote with and see if I can get the process started.

I also have upcoming programming work with TOF in October which will probably consume considerable time, but generate funds to finance this endeavor.