Output Filters

Project Gamma

Output Filters

Current Status: Boards are back and one is assembled. Qualification is proceeding, but so far the results are excellent. Pics are in the photo gallery.
The board design (done in ExpressPCB's free design software, available from http://www.expresspcb.com/ and last updated on June 8th, 2005) can be downloaded from here .

I have uploaded the results of the initial qualification of the filter board. See the following files for details on each band. This data was gathered using a small 50 ohm non-inductive resistor as the high pass dump resistor, and an HP Vector Network Analyzer. The VNA output was dumped to a text file then processed in Excel.

10 Meters
15 Meters
20 Meters
40 Meters
80 Meters
160 Meters

We have also run high power tests, with excellent results. Each filter was subjected to 1500 watts key down solid dits (resulting in a 50% duty cycle) for three minutes. Any component failure was noted (there were none), and temperature of the cores, caps and wire was checked. Then for the three low band filters, the high pass filter was subjected to 100 watts key down solid dits for three minutes at two bands above the fundamental (so the 3.5 mHz filter got 100 watts on 14 mHz). Again, we checked for any issues, and again, we found none. While it is quite possible that we will find issues in actual contest use, our initial high power tests revealed no problems.

-----------------------

Notes on output filters:

We need output filters, or some kind of output filter, because solid state amplifiers generate a lot of high energy harmonics.

In a tube amplifier, the output impedance of the tube is matched to the antenna (50 ohms nominal) through a tunable output tank circuit of some sort (usually at pi-L circuit, I think). This does two things - it does the impendance matching, but it also serves as a fairly narrow pass-band tuned filter. In that capacity, it knocks down the harmonics generated by the tube to acceptable levels.

But in a solid state amplifier, we typically use a broad-banded impedance transformer rather than a tuned circuit. That transformer doesn't have the same band pass filter effect as the tube amps output circuit, and so we have bad harmonics in our output. We need to get rid of those harmonics if we want to build a clean amplifier that meets the standards of good engineering practice, never mind FCC regulations!

So in essence we have to options. One, we can build a set of low pass filters (because the harmonics are always higher in frequency than the fundamental that we want, we don't need band pass filters) and switch them in using some kind of (hopefully automated) relay arrangement. Or alternatively, we could build a tuned output circuit similar to that used in tube amps, and motor drive it to auto-tune.

Now the idea of an auto-tuning output circuit is very attractive. In essence, we would be building a 50 ohm input, variable output antenna tuner. We could do it with a fairly simple pi-L circuit design - WC6H has designed a 50 to 1000 to 20-200 ohm circuit using motor driven vacuum variable capacitors and switched tapped coils. There are, however, a couple of problems. First, the parts are expensive - expecially the vacuum variable caps. Second, building such a beast requires both mechanical and electical design skills that are beyond my capability. So at least for now, I am focusing my energy on the switched filter bank idea.

The design of the filters is one major question, what to do about the third harmonic is the other.

By using a push-pull design for our RF section, we get rid of the second harmonic. Well, we don't get rid of it, but we knock it down to the point where it doesn't require special consideration. But the third harmonic, now that is a problem. The third harmonic in a broad- banded amplifier is going to be around 14 db down worst case (see the application note for details). 14 db is roughly equal to a factor of 25, which means that if our fundamental is 1500 watts then our third harmonic is going to be 60 watts or so. That is quite a bit of power to disappate in a filter.

In Mechanicals of the amp from the front of the QST that included Helge's article, we can clearly see the mechanical construction of the filters that he used. Each of the filters uses the same schematic, and the same mechanical construction. In fact, I think he used the same PCB for each filter, and just adjusted the component values.

We built a set of filters using Helge's design, and they worked perfectly. Unfortunately, they were not designed to handle 1500 watts of output in a contest situation - and they proved unable to handle that power level. The iron powder cores and copper wire in the inductors heated up so much during testing that we feared disaster in a real world situation. Back to the drawing board.

Back in 1999, William Sabin, W0IYH wrote an excellent article in QEX describing what he called a "Diplexer" filter for use in solid state amplifiers. Sabin noted that a standard low pass filter design has a very high SWR in the stopband. This causes the harmonic output of the amplifier to be reflected back to the RF section, possibly causing unexpected intermodulation distortion and other issues. More recently, Steve Friis (WM5Z) designed a set of these diplexer filters, based on the QEX article, for use in his 600 watt EB-104 based solid state amp. He was very satisfied with the result, and published a very nice web site describing his design. The Project Gamma team decided to adapt this design to 1500 watts for our purposes. We designed a printed circuit board to hold all six filters, and selected components that would handle the expected power level. We applied our learning from the first set of filters to this one to ensure adequate power handling capability.

Some resources:

The ExpressPCB design file for this board.

Pictures detailing the assembly of the filter board.

The parts list and shopping list for the filter board.

Some random thoughts and notes follow.

JW1FSN-DC12V relays are $1.89 each quantity 25 from Digikey. Panasonic 3 kv capacitors have 7.5mm lead spacing, roughly 7 mm diameter, and 3.5mm thickness. 14 gauge wire is .064" diameter. T130 torroids are 1.3" x .437". That makes the inductors total footprint 1.428" x .565". There is a great article/chart related to toroids and inductors (impedance per # of turns, wire required per turn, etc.) here . Panasonic JGE series high voltage ceramic disc capacitors look like a good choice. They are rated at 6 KVDC, perform well at higher frequencies according to the datasheet , and a reasonably priced.

The article is missing one capacitor value, for C30 - the last cap in the 160 filter. I downloaded a very nice simulator from Ansoft that can simulate filters, and found that the value of that cap is non-critical. Any value between 1200 pF and 1500 pF seems to change the depth of the first notch a bit, but not move the frequencies at all. I will continue to experiment, but for now will assume that I can find a value that will work.

Turns out, little did I know, that you want the return loss to be as high as possible, while the insertion loss is as low as possible. The optimal value for this filter appears to be 1090 pF. Higher is ok, lower isn't.

According to a very cool little tool that I found on the web, if you want to make a microline (that is what they call a stripline that is made with just one ground layer, not two) for 50 ohms out of FR4 PC board material (which has a dialectric thickness of .059" or 59 mils)using one ounce copper (industry standard) traces, then the trace width for the signal trace is 2.7 mm, or 106 mils assuming a relative permittivity of 4.6.

I gave George Cutsogeorge, W2VJN (co-counder of Top Ten Devices ) a call. He told me that when they designed their original products, they started out with a 50 ohm stripline design on standard thickness FR4. He found that the dialectric heating of the FR4 was so great that he was worried that the board would fail - so he relieved the ground plane under the RF conductor (essentially, he removed the ground plane directly under the trace and for a small distance on either side). He reports that this results in acceptable SWR and much lower temperatures. I plan to do the same thing for the filter board design.

Ok, that was the old plan :-) New Plan: The trace width for a microstrip using one layer of standard thickness FR4 (59 mils) is about .1". That width trace can't carry the current that we need safely without getting hot. So, I am going to a two layer design. The main PCB that carries the filter components will have ground on the back. A second layer of FR4 (that I will make from a blank from Mouser) will be mounted on top of the main board, and will carry the hot trace. That will get me two layers of FR4, total thickness 118 mils, between ground and hot. That makes the hot trace width 217.5 mils, which is wide enough to carry the current we want. As a side effect, both the additional thickness and the wider trace should reduce the dialectric heating on the FR4 board material that George was worried about. It will be a bit more tedious to manufacture, but I think that is well worth it.

According to my handy dandy micro-strip calculator at http://www.emclab.umr.edu/pcbtlc/microstrip.html , the capacitance of our micro strip is 2.9 pF/Inch . We need to either remove that much capacitance from C1 of each effected filter (probably the higher frequency ones) or cancel it out with inductors as Helge did. Of course it also has 7.1 nH of inductance per inch, what impact does that have??

According to TLW from the ARRL, at 2:1 SWR the max RMS voltage is 380 volts and the max RMS current is 7.5 amps. At 3:1 SWR the max RMS voltage is 483 volts at 9.6 amps.

The Elecraft KAT-100 uses 0 - 20 uH in 256 steps and 0 - 2400 pF in 256 steps.

The LDG AT-11 uses .11, .22, .39, .59, 1.25, 2.5, 5, 10 uH inductors (0-20 uH), and 0 - 3900 pF.

Per Title 47 Chapter 1 Part 97.307, spurious emmissions must be 43 db down and less than 50 mw. 50 mw is slightly less than 45 db down from 1500 watts. 43 db down is 75 milliwatts, too much. I think we should spec our filters to deliver 40 db of attentuation. Given that the worst case harmonic (3rd) will be at least 10 db down to start with, 40 db of filter attenuation gets us 50 db of total attenuation - more than enough.

Some learnings about designing filters for high power applications:

Capacitors: There are two issues with the capacitors. First is voltage rating. Remember that caps are typically rated for DC voltage, but not for RF or AC voltage. We need to derate the caps significantly to take the frequency into account. W3NQN uses 2 KV caps in his high power filters.

The second issue is heating due to circulating currents. Here the key question in the dialectric material, which we would like to be very low loss. W3NQN uses very low loss Tusonix parts with a dissapation factor of .1%.

Inductors: Figuring out how to design inductors for high power use is tricky, once you know how it isn't too hard. First, we need to select wire or tubing that can handle the power at frequency. Remember that the skin effect at RF means that the center of the wire isn't carrying any current - so bigger wire is needed than for 60 hz.

Enter content here

Enter supporting content here