WB4SON

Of course the Pixie final transistor, Q2 S8050, gets darned hot, and I’ve blown one out in the past just holding the key down more than a few seconds.  With my Pixie + Red Core Filter, I was seeing about 250 mW out on my QRP meter.

I decided to replace Q2 with an old 2N2219 NPN transistor (one of the few metal can devices I had in my part box).  Sure enough, that old device, circa 1975, had better performance than the S8050.  My output meter was now about 320 mW out (SA was reading +24.3 mW or 270 mW), and the 2nd harmonic was 50 dB down.  A new 4th harmonic showed up, but was also 50 dB down.

Playing around with the output transistor will likely yield some improvements.

When viewed from above, with the board rotated so the BNC connector would be on the right-hand side, Q2 will be flat-side toward the bottom edge of the board, and the pin-out is EBC from left to right, as shown below.

After experimenting with T37-6 yellow core inductors, it was time to change to the T37-2 red core.  These have a lower operating frequency, up to 10 MHz, and it is possible that they might have a higher attenuation on 14 MHz and up because of that.  The toroid calculator indicated that 17 turns would produce a similar 1.2 uH coil.  That worked out to 12 inches of #20 wire, still more than a single layer can deal with, but I’ll put up with 4 turns of overlap.

It was immediately apparent that the filter response was different.  The loss at 7.040 MHz was lower and the attenuation at 14/21/28 was higher by a few dB.

When added to the output of my Pixie, which was producing lots of harmonics (14 MHz was only 10 dB down from the carrier), the red core filter was a substantial improvement compared to the yellow core.

The second harmonic has been reduced 46.4 dB below the carrier.  The Pixie is Passing the FCC limit of -43 dBc.

The Pixie has the Red Core Filter in series with its output.  The output of the filter is fed into an external 40 dB attenuator.  The attenuator output is fed into the input of the Spectrum Analyzer.  Thus the measured 7.040 MHz signal at -16.6 dBm is actually -16.6 dBm + 40 dB, or +23.4 dBm.  That is 219 mW.

As one might expect the output of the Pixie now looks like a real sine wave with the harmonics suppressed.  By measurement, the signal is 9.4 vpp, or 3.3 vrms.   That corresponds to 221 mW into 50 Ohms.

As I mentioned the 40-meter Pixie QRP rig has pretty horrible harmonics, well above the maximum allowable FCC limit (-43 dBc).  One Pixie was about -9 dBc, and the other was about -20 dBc.

The output filter in the Pixie is a simple 3-pole PI network, clearly not up to snuff.  With judicial parts placement, an extra 5-pole filter could be installed, but it probably makes sense to install it in an external box (Like a Pomona 3231 or 2391).

Several club members and I embarked on experiments to build a prototype filter, with some different choices of -3 dB cutoff frequency, output ripple, etc.  We also discovered that while many of the online calculators produced very similar results, at least one of them does not.  Filters built using those values did not match up well with predicted results.

John King, WA1ABI, suggested using “Elsie” from Tonne Software.  Far more flexible than the online calculators, it produced output values that were consistent as long as the initial parameters were correctly selected.  The software can be downloaded here:

http://www.tonnesoftware.com/elsiedownload.html

I used an online calculator that produced identical results to Elsie:

http://www.calculatoredge.com/electronics/ch%20pi%20low%20pass.htm

Input values were a cutoff frequency of 7.5 MHz, 50 Ohm impedance, 1 dB ripple, and a 5-pole Chebyshev filter.  That produced the C1/C2/C3 & L1/L2 values shown below:

C1 & C3 were fabricated from a 680 pF in parallel with a 220 pF, for a total of 900 pF (less than 1% low).

C2 was fabricated from three caps in parallel; 470 pF, 470 pF, and 330 pF, for a total of 1270 pF (less than 1% low).

My inductor was set to 1.2 uH, about 4% high, but that required 20 turns on a #6 core.

I had chosen T37 cores for their small size and had initially planned on winding them with #26 wire.  However I switched to #20 for mechanical reasons.  A T37 core can hold about 13 turns of #20 in a single layer, and 20 turns of #20 in a dual layer — perfect.  14″ of #20 magnet wire was enough to wind 20 turns with 1.5″ long pigtails.  It was tight, but things fit.

The final breadboard circuit looked like this:

The test setup looked like this, with the breadboard hooked up in series between the Tracking Generator output and Spectrum Analyzer input:

The measured response was:

• 7.055 MHz -2.6 dB (Operating Frequency — loss is higher than expected)
• 7.429 MHz -3.0 dB (Cutoff Frequency — a bit lower than expected)
• 14.112 MHz -35.1 dB (2nd Harmonic)
• 21.168 MHz -29.7 dB (3rd Harmonic)