Enclosure for microwave circuits

Hi All,

Any advice on the design of metal enclosures for microwave circuits? Dimensions, positioning of connectors and dc input etc....


Thanks!

If possible keep the width of the enclosure less than 1/2 wavelength at highest operating freq to prevent waveguide coupling.

To avoid box resonances, be sure at least two of the three enclosure dimensions are < 0.5 wavelength. Problem in EM filelds is known as a "dielectric loaded rectangular resonantor". Solving for resonant frequencies is easy, but requires iteration. You can also use the free EM software SonnetLite (I work for Sonnet) from www.sonnetsoftware.com. The xgeom option "Analysis->Estimate Box Resonances" lists off all the resonances in whatever box/dielectric geometry you have specified. It is very upsetting to have a beautiful filter response and when you put the lid on there is a suck-out right in the middle of the passband!

If you want to analyze the effect of the enclosure on your circuit you can use either Sonnet or EMsight, both very efficient for shielded circuits. One example of the effect of a box on a filter is in G. L. Matthaei, J. C. Rautio, and B. A. Willemsen, "Concerning the influence of housing dimensions on the response and design of microstrip filters with parallel-line couplings," IEEE MTT Transactions, Vol. 48, August 2000, pp. 1361 –1368. You can get that from IEEE Explore, or I will email you a pdf. The Matthaei by the way, is from the Matthaei, Young, and Jones of filter fame.

Thanks for the reply guys!

My circuit is rather big compared to the wavelength (6 GHz) since I'm working on a active circuit. I'll try to make some quick calculations and CAD before I finalise the box.

About the resonant problem, I heard you can easily overcome this by, say, screwing on an extra piece of metal on the inside?

You can glue a piece of ECCOSORB Cavity Resonance Absorber to the inside of the enclosure cover lid.

h**p://www.eccosorb.com/catalog/eccosorb/index.asp
h**p://www.eccosorb.com/cavity_resonance.asp

It is possible to move a resonance by adding metal on the inside, or to reduce or eliminate it by adding absorber. Be careful with absorbing matrial if you are working with power or low loss is important. Also, out gassing can be a problem long term in some applications. You want to be sure actual circuit operation is boring, not exciting!

Both absorber and extra metal can be less effective than you might think, depending on the mode you are trying to modify. Some modes have most of their energy concentrated in the dielectric. Do whatever you like in the air and it doesn't care. Those kinds of modes can be real hair-pullers.

Not all circuits are affected by all resonant modes. Best thing to do is to analyze some kind of representation of your circuit in the actual box size. If you see a suck-out at a known resonant frequency, you know you have problems.

If your top cover gets real close to the circuit, it will modify your circuit performance. If you take that route, be sure to do shielded EM analysis and compensate the design.

If you just say to heck with it, and leave the cover off, the resonance will (probably) dissappear, but you will get radiation. If there are other components around that can couple out of, or into your circuit, there will be problems. In fact, one way to do a quick check on radiation is to do an unshielded EM analysis (Momentum, IE3D, Ensemble), followed by a shielded analysis (Sonnet, EMSight) and see if there is any significant difference.

If it is a pure rectangular box, you are best using a shielded planar EM analaysis like our own Sonnet. You can include absorber easily by specifying a lossy dielectric layer. Chunks of metal can be included as well, but that starts to use up memory and analysis time. At some level of complexity, you would want to switch over to a volume mesher to see what happens when the lid goes on. We work a lot with CST here (support and sales in North America) and I would recommend CST for this application. The broad band high resolution frequency domain data you get from a time domain analysis like CST is really good when searching for narrow high-Q resonances.

As I mentioned above, if you want a quick list of resonant frequencies, download SonnetLite htt://www.sonnetsoftware.com, enter your box and dielectric geometry and click Analysis->Estimate Box Resonances.

Even if there are no resonances, many microwave engineers (including me) have had the joy of seeing a beautiful amplifier response on the ANA, then when we put the cover on, it becomes a beautiful oscilator. Use your circuit analysis tools, combined with planar EM analysis of the passive portion of your circuit (use internal ports for connecting your transistor) and look at stability factor...with the cover on.

Hi,

since we have attention of Rautio I would like to ask something about enclosures myself. Is there any other mechanism of coupling in encloused amplifier apart from box resonances? I had a terrible time trying to eliminate oscillations from an amplifier. Microstrip portion was EM simulated and even bias circuit was simulated as well and transistors placed on internal ports as you suggested. Bias was introduced trough quarter wavelength lilnes with radial stubs. Amplifier simulations showed uncoditional stability on all frequencies. I played a bit with my fingers and absorbers and actually discovered that a coaxial connector was a small discontinuity exciting the whole box in a spectrum of modes that I didn't see in simulations. Some of those modes made I/O bias lines coupled although they were separeted 3-4 wavelengths! They were made paralell, after puting them at 90 deg one with respect to the other the problem was stil present but was possible to remove with absorbers. Also, bias was introduced via feed-trough capacitors. Small pieces of leads of this capacitors were also coupled although on different walls and taking part in oscillations as well! Amplifier was supposed to work above 20GHz and the problems were occuring on much lower frequencies.

My second question will be, how the things are changing by introducing non-ideal box, i.e. by having holes for connectors, feed-troughs, screws, technology required holes in corners etc?

flyhigh

Hi Flyhigh -- OK to call me "Jim".

You just described the classic amplifier/oscilator problem. If it didn't happen so often it would actually be funny to say when you design one you get the other!

When you put a lot of gain in one package, just about anything you do will change an amplifier into an oscilator. Absorber is a nice solution provided the problems described above are not an issue. Years ago I designed microwave stuff for spacecraft (Landsat IV, DSCS III) and absorber was not an option due to outgassing.

If you have a non-resonant package and the package introduces output-to-input coupling so you get oscilation, one solution is to put less gain in the package. If you have two stages, split the stages up into two packages. But I bet you only have one stage. In that case, another solution is to put a wall of some kind between the two input and output. Easy if you are on a metal fixture and you have separate substrates for in and out matching. Take it to the top cover if you have to.

Another solution is to bring the top cover down as close as possible. I have never done that, I am thinking it would reduce out-to-in coupling, but maybe not. Easy to try in EM analysis. Doing such will change the response of your circuit, but it is no problem to design the circuit with a requirement that the top cover be in place for proper operation. Heck, take the top cover down so it sets almost right on the tallest component in your circuit! The response will look terrible with the cover off, but who cares!

Another solution would be to introduce negative feedback, to cancel the positive feedback introduced by the package. With your transistor at ports 3 and 4, look at Y34 (Y-parameters). Y34 = minus the admittance between ports 3 and 4. So pick some circuit that adds that admittance in between ports 3 and 4. Now your passive matching network will have zero Y34, and no chance to oscillate, at least at that frequency. I have never done that myself either, but seems like it would be straightforward, the main problem would be finding a circuit that minimizes Y34 at all frequencies (very low and very high) that your transistor has gain.

One source of coupling might be the vertical feed through pins. For a lot of modes (especially with a low cover), the E field will be mostly Ez (veritcal). Such modes are easily excited by veritcal conductors (think vertical antenna). Easy to check in EM simulation.

For all the access holes, rounded inside corners in milled packages, etc. you can check with a volume mesher. If resonances are not a problem and because you don't need the last word on accuracy for this problem, any volume mesher should be OK. Analysis time and ease of use will be determining factors.

Good luck! It is a tough problem, but it is a rite of passage for microwave designers. You just gotta believe, yes, the problem really does have a solution.