Machined Cavity Refined

My momentum slowed a bit this week as I worked to get a proper drawing off to Ryerson Metals for a quote.  The good news is the price for the aluminum cavity including metal and machining might be under a thousand dollars.  I got a quote back for just the aluminum rod 12″ in diameter and 16″ long for only $730 which was a surprise.  My hope is that if I buy the aluminum and get it machined at the same place, I can pay for just the %13 of aluminum remaining, which could save me $635!  I suspect the machining will cost me a lot though because the rod will probably have to spend literally hours on the mill.  I will then have to get the cavity electroplated with copper which is the big unknown – will the copper be thick enough?  Will it be even enough? Can it be made shiny and ultra-smooth?  Can I solder the probes to it?  I expect the quote for machining back early next week and if it is under $500, I might get two!

Most of this week was spend redesigning the cavity again because I wanted to reduce the amount of aluminum required and this is what I came up with, simple and beautiful:

This cavity was built to exactly the dimensions as determined by the cavity exported from the simulations. It was also designed to make mounting the large power probe much easier because it fits exactly:

It is recessed on purpose because to get it flush, the rounded surface will have to be sanded down.

What took the most time this week is that Ryerson wanted a proper 2D drawing including a “tolerance block” and I learned how to do that:

That’s a screenshot of a PDF that allows you to zoom in which is really neat.  I got the template as a DWG file and then combined it with a 2D drawing of the model.  I built the whole thing in Rhinoceros because making the dimensions is trivial and then printed it to a PDF.

I also received a few things I’ve been waiting for this week:

  • Over the many years I have been doing this research, I have been collecting waveguide components and because I was looking at 5.8Ghz, I got a WR-159 waveguide-to-N converter (second from the right):

    The rest of the waveguide components are circulators (typically used to dump unwanted reflected power into a dummy load) and I have a nice cross section of waveguides going from 3.30Ghz up to 8Ghz now.  The waveguides are cheap and plentiful these days, especially WR-90 components (8.20 to 12Ghz) but the sources are still expensive and rare.

  • I also got the copper mesh and the quality looks excellent:

Hopefully in early October I will have a machined cavity to test and then we shall see if the Q measured is anywhere near the Q simulated.  I will be coming up this week to try the copper mesh and get a new lab key.

Machined Cavity and Higher Frequency?

This week has been one of learning because I have been trying to solve two problems: The cavity is not within tolerances necessary to get the high Q required.  As we get a higher and higher Q cavity, it creates another problem – the frequency of that high-Q resonance is very exact and even a 6Mhz variation of the source frequency will take the cavity out of resonance, something that could happen just because the magnetron warms up.  The commercial magnetrons we are working with typically have a 20 to 30Mhz bandwidth, which means it could be at 2.42Ghz or 2.48Ghz!

The good news is that I might have solutions for both problems, a higher Q cavity and a frequency controlled magnetron.

I will deal with the higher Q cavity first and there are two options, both of which require a redesigned cavity.

I started with going to a higher frequency because everything shrinks in size substantially which means costs drop.  A good example is the current 2.45Ghz isolator which cost $250 second hand (which is thousands less then new) and is about the size of a toaster.  A circulator for 5.8Ghz cost $40 on eBay and is the size of a tape measure.  The biggest benefit of going to 5.8Ghz is that I can purchase a 6″ diameter, 8″ long solid rod of aluminum for only $150 and sourced locally!!  That leaves lots of money to have the aluminum machined and then copper plated.  Perfect!  With a computer controlled milling machine, the cavity can be produced almost directly from a file exported from the EM simulation program!

This past Sunday afternoon, Lin let me into the lab (I have to pick up a new key!) and I took the previous cavity designed for 2.45Ghz and rebuilt it for 5.8Ghz.  This was easy – I took the 2.45Ghz cavity and reduced it by 53%, which I figured out by finding the percentage difference between the diagonal dimension of WR-340 wave guide, built for 2.45Ghz and WR-159, built for 5.8Ghz.  I then redid the simulations and as expected, making the cavity smaller raises the frequency of the TE0,1 mode.

Because the large end of the 5.8Ghz cavity was under 6″ in diameter, I spent a few days converting the exported drawings from the simulation program into something that could be made on a CNC mill (ignore the probe bump inside and out):

Getting the cavity made is easy, however, the problem is finding a source for a high power 5.8Ghz signal.  After not getting a response back on a quote for a 5.8Ghz source,  today I phoned John Gerling, of Gerling Applied Engineering, a small California company and learned a lot of things – for example, back in 2001, Panasonic use to make 5.8Ghz magnetrons for a cost of $1.5K each (wooah) but did not find a market and stopped making them.  The 5.8Ghz magnetron is identical to the 2.45Ghz magnetron and takes the same inputs.  However, my hope for finding a high power (there’s the kicker) 5.8Ghz source for a couple thousand dollars was off by an order of magnitude.

John did tell me about TWT or traveling wave tubes and I found a TWT complete with power supply on eBay (link removed) that can vary the power up to 3.3KW, runs at 6Ghz and is only $2500!!  The problem is the sheer size – it is 1200lbs which means shipping alone would be another $2K and then it might not even work.

I came to a realization – by moving away from 2.45Ghz I can build a more accurate, smaller and cheaper cavity, but the source then becomes prohibitively expensive.

The attempts at a 5.8Ghz cavity may have lead to a solution anyway – To machine the 2.45Ghz cavity from a single round metal rod requires one 12″ in diameter and 16″ long.  Nobody I can find makes aluminum rods that large, but I can buy a type of round 4140 alloy steel that is 12″ in diameter and 16″ long for $830 from a local supplier!! (weighing in at 514lbs!!)  I will confirm availability and price on Monday, but first I will see if I can find anybody that is even willing to machine that monster.  It will take a fork lift just to get it into the mill.  If both of those pan out, then I will check to see if it can be electroplated with copper.

On to the second problem which is locking the magnetron to a certain frequency:  As I was talking to John, I mentioned the “Magnetron – A Low Noise, Long Life Amplifier” paper (linked to previously) and it turns out that you can buy magnetrons that use electromagnets instead of permanent magnets and he gave me the part number.  The magnetrons retail for $3K but one recently sold one eBay for $350 (link removed)!  What I am looking for is this:

Note the coils?

It can output 6KW, is water cooled and has electromagnets which can be used to control the frequency!  I am now tracking down what type of power source and controller is needed and then hopefully can find some on eBay.  Given our circulator is already water cooled, a water cooled magnetron is even better because it removes the fan.  John also mentioned that reflected power can change a magnetron’s frequency in a “magnetron pulling” method, but the good news about our setup is that we are using an isolator which means that the reflected power the magnetron see is nearly zero and won’t affect the tuning.  Oh, did I mention the isolator we have can handle 6KW?!

Until next week!

Results – Meh!

We hooked the cavity up to a better network analyzer this week:

and zoomed in on the frequency we are interested in and after tuning a bit, got:

Mehh.  The results show an insertion loss of 1.6dB and after playing with the tuning, it is apparent that the cavity is sensitive to flexing. By putting pressure on the cavity, for some reason the insertion loss would drop to 1dB and the resonance got much better, as deep as 40dB.  The calculated Q for this test was a paltry 360 which is tens of thousands lower then expected.  The likely cause is fabrication tolerances which means a second more accurate method of fabrication resulting in a stiffer cavity may be necessary.  I am currently considering a few alternatives, for example: trying a higher frequency, say 6Ghz, which will drop the size the cavity considerably making it significantly easier to build.  Another option is to purchase a CNC table and do the milling myself.

Another improvement would be better probe mounts, but not a single one of the vendors I sent RFQs to responded, hurumph.

I was also looking at the typical specifications of magnetrons and they produce a frequency of 2.45Ghz ± 10Mhz, which means the frequency could be as high as 2.460Ghz or as low as 2.44Ghz.  One of the tests that will be performed is to measure the center frequency of our commercial grade magnetron and then test to see if it wanders during use or between cycles.  If it is stable, then we should be able to tune the cavity to resonate at just the frequency of our magnetron.  We are waiting on the spectrum analyzer at the university to be fixed.

In order to control the frequency of the magnetron, It might be possible to use a feedback mechanism called a phase-lock-loop to center the magnetron frequency at 2.45Ghz and there is a paper that talks about a simple method here.

I also worked on the vacuum chamber this week, adding a hook to the top in order to keep the acrylic closed.  I got the extra seals this week and will replace the temporary green seal (shown below on the right).  It takes about 15 minutes for the chamber to pump down and the lowest it has gone is 19 microns which is only 2.5% of atmosphere and only 1 micron above the lowest expected (18 microns)

I was happy to note that after a certain pressure, even with the pump running, the gauge starts to show a dropping pressure.

With a vacuum, the 1 inch thick acrylic deflects inward as shown by the shots below:


The center of the acrylic is depressed 9mm!

Below are the calculations I carried out back in September of 2009 that says the windows has a safety factor of 7 times.

I found a handy power point slide here, which shows how to calculate the safety factor for a given strength versus diameter for a vacuum chamber window.  It turns out that for an acrylic window with 25″ diameter of unsupported space and 1″ thick, it has a safety margin of 7x.  In other words, the window can handle over seven times more air pressure before breaking.

Here are the calculations I carried out:

First, calculate the stress on the window for a known thickness versus radius (from the power point slide):

Sm=k(wR^2/t^2) where
Sm – the stress on the window in PSI
k – is the coefficient for circular plate, I used 1.1 which is a conservative estimate.  This constant is described in the power point presentation.
w – the uniform pressure across the window, which because Edmonton is ~2300ft above sea level, means our air pressure is 13.66psi.
R – is the radius of the unsupported part of the window, in our case 12.5″ or half of the 25″ diameter (this was reported by the vendor)
t – is the thickness of the window which in our case is 1″, 1.5″ or 2″

The stress on a 1″ thick acrylic window 25″ in diameter is then Sm=2300psi and the safety factor of cast acrylic is then calculated by:

S.F. = Modulus of Rupture / Max Stress

where the Modulus of Rupture for Acrylic is typically 16,000psi which means a safety factor = 16000/2300=7x.

Analysis of Results & Chamber Complete!

It took years, but what a fantastic week!  Two events, long in the building stage, got completed this week – the vacuum chamber and cavity.

First the vacuum chamber and here is Kashish, one of the masters students, helping me pump it down for the first time:

The gauge with red numerals on the bottom right shows 49.0 microns (vs 760microns at atmosphere), which is the lowest we have pumped the cavity down to.  With a longer pump down, we should be able to get it close to our minimum, 18 microns.  We were using the timer on the iPhone, bottom left, to time how long the vacuum chamber could keep the vacuum.  Even with a temporary seal (the green thing on top) it was only losing about 40 microns of pressure in the first five minutes.  With a proper seal (shipped this past week), we should be able to extend the vacuum time.   Our experiments will be less then a minute long and even at the current state of leakiness, the chamber is useful.
Besides adding a better seal, another modification will be to put the vacuum gauge on it’s own port.  Currently it measures atmosphere (760microns) when the pump is on, which means we have to shut the pump off to see what the pressure is.

Getting a solid vacuum in the chamber is a big step because it verifies all the parts that have been bought off eBay, from the pump, to the cavity, to the strength (so far) of the cell-cast acrylic window.  Awesome!  $3500 and a year in the making! (It was September 8th,2009 when I first purchased the chamber)

The second big event is that it looks like the cavity is working as per our simulations!!

At first, we were getting this:

Really crappy!  High insertion losses (7db) and flat resonance spikes, which means low Q.
After mucking around, it turns out it was a problem with the probes.  In order to solder the probes to the probe mount, I take the bottom plate off to get inside.  After soldering the probe, I then test connectivity between the external probe mount and the cavity, which is a closed loop as it should be.  I then bolt on the bottom plate but unbeknownst to me, doing so warped the plastic just enough to disconnect the probe mounts.  The fix was easy enough, I used a lower power soldering iron to solder the copper mounting plates to the copper wrap!  I will get a picture at some point…

With the probes fixed, we immediately got much better results:

We haven’t calculated Q yet, but the results look very promising, an insertion loss of just 0.6dB and a beautiful narrow deep spike (the narrower the spike the higher the Q)  Below are the simulations results to compare – the “blue” line in the actual results corresponds to “red” in the simulated and yellow corresponds to green:

A quick analysis:

  • The range on the measured results is from 2.4Ghz to 2.5Ghz, the range on the simulation is 2.4274Ghz to 2.4632Ghz
  • Simulated insertion loss is 0.77dB (S2,1), actual is 0.63dB!
  • The measured results show a double resonance (the two blue spikes) as does the simulation (two red spikes)!
  • The maximum S1,1 simulated was -14dB, but actual is -28dB and you can tell the spike goes much much lower where, although I don’t have a picture, it measured -48dB!
  • If the actual results don’t change too much after we zoom in next week, the Q will certainly be lower then the 71K simulated which we can tell by how wide the spike is at the top.  The width of the spike at the top in the simulation is maybe 2 or 3 Mhz wide (take 2.455Ghz minus 2.445Ghz which is 10Mhz and the width looks to be about a fifth that wide), but the actual looks to be twice that (take 2.46Ghz  minus 2.44Ghz equals 20Mhz and the width is roughly fifth of that or, 5 or 6Mhz wide), which means a Q of probably 25K.  We will know more next week when we zoom in and calculate Q.
  • If the depth of the S1,1 curve is a measure of how pure the TE0,1,n mode is, then the actual results looks to be very good.  The difference between our actual (-28dB) to simulated (-14dB) might be explained by not having used enough tetrahedrons in the simulations and/or luck.  It is ironic, that even with our modern technology, we have no way to directly see the wave pattern inside the cavity, and can only strongly guess from indirect results that the cavity is resonating in the right mode.

Having played with the system a bit, a few things are apparent:

  • Tuning works very simply, you move the tuning plate up and the frequency of the resonances moves down.  Move the tuning plate down and the resonances move up.
  • Moving the plate doesn’t move the resonance frequency by very much, in other words, moving the resonance frequency up or down the band can be very fine with the mechanism we have.

As a side story – pictures of the results are taken with a VIXIA HF100 video camera and the lights off in the lab.  Picture a large room with five or six other people in it and every time we take a picture, we kill the lights.

Before going onto the next step, the probe solder points have to be cleaned up to handle high power and the copper mesh needs to be added around the bottom and top seams to ensure solid connectivity and no arcing.

Meanwhile we can move on with the following:

  • Check the attenuation of the reflection port on the circulator – Because resonance occurs at a single frequency with a very narrow bandwidth (the definition of a high Q), an ideal source would only transmit at that one frequency, no matter the condition.  Because we are using a real world and inexpensive magnetron, it will likely not transmit at the correct frequency all the time.  In order to tune the cavity to our magnetron, we have to measure what the magnetron’s center frequency is, and how badly it  wanders.  However, because the magnetron transmits roughly 800W of power, we cannot hook it to the network analyzer but need to attenuate the signal first.  The circulator, which dumps excess reflected energy into a water load, has what looks like a reflected measurement port, which we will see if the power is low enough to measure.  Worst case, the magnetron frequency wanders too much and a magnetron with a phase-locked-loop arrangement will have to be sourced/purchased.
  • Once the issues with the source are cleared up, we can then run the actual experiment and deal with heat problems.

Have a great long weekend! I am off to Calgary and will be back in the lab on Tuesday.