Gruending Design and Test Setup


  • Neil is now setup to go, need better magnetron power supply

My trip to West Kelowna was a success, I survived the 24 hours of driving (there and back) and we have a better idea of the challenge ahead.

The first thing we did was set everything up including the water cooling and magnetron and ended up with the following

On the left is the water pump and radiator, then the magnetron connected to the circulator which was connected to a section with tuning pegs finally ending in a short (metal bolted to the end). In the foreground you can see a multimeter and a high voltage probe we used to sample the magnetron power supply. We also used a microwave detector to test for stray radiation and the thermal imager to make sure the magnetron didn’t overheat.

We then hooked up the variable step attenuator, programmed it for 70dB and connected the Signalhound to the isolation port on the circulator. After getting a few sparks when the high voltage for the magnetron arced through the wire shielding, we recorded a few results.

It was soon apparent that the magnetron was jumping frequency too much to even show up on the Signalhound which takes measurements every 150mseconds. (As a reference, the HP 8753D analyzer at the university samples at 50msec, and even there it was jumping around way too much).

We then hooked up a high voltage probe to record what the magnetron input voltage looked like and I was surprised to find this:

From Neil’s blog he writes:

  • The peak to peak AC voltage is about 4.4kV and is shown AC coupled. The current waveform is DC coupled and is changing directions which makes sense in an AC powered system. The peak current draw is about 12A with a RMS current of about 8A. These numbers don’t make sense though because the RMS power consumed using these numbers would be 16kW (2kV * 8A) which is quite a bit more than a 110V plug can supply. I will need to try and repeat these measurements once I go over the test setup.

I was under the incorrect assumption that domestic magnetron power supplies used some type of rectification in order to create a 4.2Kv DC cathode voltage, but apparently you can run the magnetron off AC as well. We are guessing that the reason the magnetron output jumps frequencies, is because of AC input. I plan to send Neil another newer microwave oven that uses a different kind of power supply to find out if it too drives the magnetron with AC, or a DC voltage and if it makes any difference in the magnetron output.

The conclusion of our trip is that we need to test more magnetrons and magnetron power supplies in order to find a setup that has a much steadier output. Once we have a steady output that can be measured by the Signalhound, we can then use a feedback loop to center it on the cavities resonant frequency.

Along that line, I found a paper that mentions the magnetron theory of operation which had some interesting points:

  • “For most magnetrons the temperature coefficient is negative (frequency decreases as temperature increases) and is essentially constant over the operating range of the magnetron.”
  • “The “automatic” synchronism between the electron spoke patter and the r.f. field polarity in a crossed field device allows a magnetron to maintain relatively stable operation over a wide range of applied input parameters. For example, a magnetron designed for an output power of 200kw peak will operate quite well at 100kw peak output by simply reducing the modulator drive level.”
  • “The pushing figure of a magnetron is defined as the change in magnetron frequency due to a change in the peak cathode current.”

Here are my takeaways from the paper:

  • Frequency control may be easier by controlling the magnetron cathode current with a constant temperature then controlling the magnets

Or, it may be a combination between controlling the cathode current and magnets that gives the best result.

Here is my todo list:

  • Send Neil the other microwave to see what waveform that microwave uses to power the magnetron (TODO)
  • Send an email to John Gerling to see if we can get the specs for the ASTeX magnetron head because his company repairs them.  and John responded with “Sorry, can’t help you any further”.  It turns out that John was one of the original designers for the ASTeX magnetron and Gerling spun out the ASTex company which means he considers it IP. Darn!  Have to find another source…
  • Get schematics and/or repair manual for the MDX-10 power supply in case we can use any of the parts. (TODO)

High Voltage High Power Fun

A friend stopped by this weekend and dropped off some fun toys.

He needs a frequency locked 1000W 2.45GHz signal based on a microwave magnetron. I like the magnetron solution because it’s a cheap way to generate such a high power RF signal but they wander between  2.42GHz and 2.48GHz which is a problem for his application. I have volunteered to help him figure out a method to lock the output frequency. I think it will make a great writeup for the blog since I expect it will take a significant amount of reverse engineering and experimentation to make it work.

I would like to point out at this time that the output from a microwave magnetron is extremely dangerous. Microwaves like to boil water and people are 75% water. All of my experiments will be done in carefully controlled conditions and have been checked with a microwave leakage detector. The microwave energy in the experiments will be dissipated using a water load and also the test setup itself, not me.

The first thing we did is setup all of the equipment for the tests. The magnetron and control circuit are from a microwave oven and all of the waveguides are recycled from old equipment. The water cooling setup is a pump and radiator from an old CPU water cooling setup. Unfortunately it takes a lot more than an old microwave for a test setup. Here’s a block diagram of the basic test setup (you might have to click on the image to see it clearly):

With this test setup, the magnetron can only be run for short periods of time. For example, after 20 seconds of operation the magnetron case temperature can reach 50C even with the cooling fan from the microwave blowing across it. Since the magnetron can run for many minutes inside of a microwave the test setup needs to be reviewed to see if the microwaves are being reflected back into the magnetron. Since the magnetron output is shorted, I would expect to see most of the power reflected back towards the magnetron. The isolator should be attenuating this energy by 20db or so but that means that 900W (1000W-1000W/10) would be absorbed by magnetron. Here is a good application note on how isolators work. Section 7.9 suggests a method of frequency locking a magnetron which might come in handy.

We first tried to observe the magnetron output using a Signal Hound spectrum analyzer connected to the isolator reflected power output with an external attenuator. When we tried a 50MHz span centered at 2.45GHz, the sweep rate was about 300 to 400ms. Occasional peaks were observed but it was pretty obvious that either the signal wasn’t there or it was hopping around faster than the Signal Hound could capture. Different attenuation settings didn’t help. My friend had previously tried the same test using a newer Agilent spectrum analyzer and even with a sweep rate of 50mS he observed that it was difficult to see the magnetron output. Even when the Signal Hound was set to a 5MHz span the sweep times were still 150ms to 250ms. I will have to investigate further to see if it’s possible to speed up the sample rates using either a faster PC or with different sample settings.

We also measured the input voltage to the magnetron using a Fluke 87III multimeter and a Fluke 80K-40 probe and got a DC voltage of about -2.2kV and a ripple of about 2.2kV. This surprised my friend since he thought that the magnetron was powered with a rectified DC voltage. I added a CT238 current probe and captured the following waveforms with my Tek 754D oscilloscope:

The oscilloscope attenuation factors were set so that the displayed voltage from the 80K-40 probe was approximately correct on channel 1 (the black trace). The oscilloscope was also set so that the voltage displayed from the CT238 current probe is actually in amps, ie 1V = 1A. Normally you can’t use the 80K-40 as an oscilloscope probe because it has a 3dB bandwidth of about 400Hz, but in this case the waveform is about 60Hz so it’s an acceptable approximation. The CT238 probe has a frequency response of 250kHz which is more than adequate for the signal measured here.

The peak to peak AC voltage is about 4.4kV and is shown AC coupled. The current waveform is DC coupled and is changing directions which makes sense in an AC powered system. The peak current draw is about 12A with a RMS current of about 8A. These numbers don’t make sense though because the RMS power consumed using these numbers would be 16kW (2kV * 8A) which is quite a bit more than a 110V plug can supply. I will need to try and repeat these measurements once I go over the test setup.

So far there are more questions than answers, but that’s what makes this project so interesting. After a few hours of playing here are the next steps:

  • Contact Signal Hound to see if sampling can be sped up. Right now it’s too slow too see the magnetron output if it’s constantly changing frequency.
  • Verify the power test configuration. The numbers measured aren’t making sense.
  • Try to measure how much power is being reflected back into the magnetron. The large reflected power could cause the magnetron to change frequency rapidly.
  • Can magnetrons operate on DC voltage instead of AC? Maybe a DC voltage will help stabilize the magnetron.

(Original Post here.)

Signalhound Comparison and Attenuator Test


  • Signalhound and 70dB step attenuator seem to work fine

I was up at the university yesterday in order to test the Signalhound and the 70dB step attenuator and both of them seem to work fine.  Here is the test setup for the attenuator:

To change the levels of attenuation, a 15v power supply was required and the attenuator was hooked up between the ports of the network analyzer.  The attenuation was as expected and it was neat to hear the attenuator make a solid click when changing between attenuation levels.

The network analyzer was then used as a super accurate signal generator and hooked to both the Agilent spectrum analyzer and Signalhound at the same time.  The spectrum analyzer was attached to channel two and the signal hound attached to channel one.

The results were then compared:

Looks pretty good, although the Agilent is cleaner and faster, the Signalhound is certainly usable.

Many thanks to Kevin who took a few minutes to configure the network analyzer and spectrum analyzer to make the measurements.

I then packed up all the waveguide components, power supplies, pump, radiator, and magnetrons and carted them out to the car in preparation for my drive to Kelowna on Friday, October 22nd.

Bent probes and cavity a no-go and CNC Machinist


  • Bent Probes and cavity a no-go and CNC Machinist

Last week I was up at the university and tested the bent probes.  The results were bad enough that I didn’t record any results.  The problem is that there doesn’t seem to be much of correlation between what is expected and what is measured.  Worse, I don’t know why.

To move forward, I have to step back to simpler cavities and make sure I can create cavities that match measured results to simulations.  Once there, I can then make and test incremental changes, until I arrive back at the original shape with all the improvements necessary.

Making the cavities by hand, i.e. folding copper into a tube, isn’t an option because I can’t get enough precision.  Using 2.45Ghz requires using cavities that are on the order of the wavelength at that frequency, which is roughly 10 to 12cm or larger (depending on propagation in free or enclosed spaces).  A crappy metal lathe starts at $600 ( – I like the word “Metal Worker” on the side 🙂 ) and a good one goes for a few multiples of that.

Simple small tube cavities are easy on a manual lathe, but eventually, making large asymmetrical ones, which are at the heart of the EMdrive, requires a CNC controlled lathe.  Second hand CNC lathes can be had for about $12K and up but require 220V or 480V and require knowledge about the controllers.  I have neither the knowledge, money, electrical power or space in my garage for such a beast.   Because all the cavities are one-offs and can be large, local machine shops won’t touch them, as I found out when trying to get the $2K cavity built (I think I contacted ten different machine shops).   The cavity is large enough, 280mm dia. at the large end, that it is outside the build envelope of a lot of lathes, and the larger lathes that are capable are all tied up doing more valuable work.   Ordering from is too slow, as the lead time is months and I want to iterate through cavity designs weekly.

I have decided to kill a few birds with one stone by becoming a CNC machinist, my third career.  My first career was a design/marketing engineer at a small BC company, my third was Levitee research, which I haven’t given up, and now a machinist.  I have been working on research now for ten years, four of which was Levitee specific, slowly burning through my savings ($30K left!!).  I also need to again start saving for retirement.  Again?  Yes, I was fortunate enough to have enough money from my first job, and willing parents, to “retire” at the age of 28 when I started down this crazy path. I need to turn on the money faucet once again.

The three birds I will kill with one stone are – learn how to be a CNC machinist, make cavities and money.

The good news is that there is a solid demand for CNC machinists in Edmonton, in large part, thanks to the oil industry.  To that end, I have, through a friend, applied at a local company which has some CNC positions open and without much surprise, they are reluctant to hire me.  To overcome those fears and because I don’t know anything about machining, I have now enrolled in an evening/weekend CNC machinist training course at NAIT and should be done by May 12th, 2012 (costing roughly $2K and 200 hours in time).

On the Levitee front, a few things did get done, for example, I received the Signalhound and I bought cables for Neil.  I plan to be up at the university next week to test the Signalhound against a more expensive spectrum analyzer to give Neil a better idea of the differences.  Things are still moving forward, even if I have to take a side step at this point.  Besides working with Neil to get the magnetron feedback loop working and simulating test cavities, things will slow down.