The RixPix

My Version of the Pixie Transceiver

3/18/2005 by Rick Andersen, KE3IJ

The Pixie, Pixie-II, etc., are a very popular group of QRP tranceivers among Hams who are interested in building "minimalist" rigs. The original Pixie design, which can be found at many sites on the Web, consisted of a crystal oscillator followed by a one stage, milliwatt power amp, with a Pi output filter feeding a 50 ohm antenna. When the key was not being pressed, the emitter current to that "power amp" (a 2N3904 or 2N2222 transistor) was limited to a small value by a 10K ohm emitter resistor; in this "receive mode" the power amp base-emitter junction was being used as a diode detector while the crystal oscillator was beating with signals coming into the antenna and Pi filter-- a rudimentary Product Detector.

The RF was filtered out of the audio beat note by a small-value capacitor around that same emitter resistor, and the emitter served as a pick-off point for the audio which was now amplified by an LM386 audio amp IC, configured for its high-gain mode. The output was either an earphone or a tiny speaker. There was no audio filtering (other than the emitter resistor's parallel cap) so the CW (Morse Code) signals heard were "hi-fi" but it was impossible to listen to any one signal individually-- you had to train your ear to pick one pitch out of the jumble (part of the price one pays for minimalist circuitry).

When the key was pressed down, the emitter was grounded directly, which served the dual purpose of 1) greatly increasing the emitter current, so that the "power amp" was now operating as such; this was the "transmit mode". And, 2) the audio amp's input was shorted to ground, preventing the transmit signal from overloading the audio stage.

Compared to the usual transceiver circuits out there, this little gem had practically no fancy circuit tricks normally needed to switch over from transmit to receive, except for a diode-resistor pair that interrupted power to the LM386 audio amp in the transmit position. A very clever design that provided basic transmit/receive functionality in an extremely simple circuit that, some would say, "shouldn't even work".

Many QRPers and clubs have made the Pixie available as a kit. Many Hams have built it. I built one (from scratch, using "Ugly Construction" over copper-clad board as a ground plane) and actually made a couple of contacts using a flea-power 2N3904 transistor in the output stage. That was neat... but not always easy! I didn't measure my power out, but I'd bet it was well below 100 milliWatts.

So these Pixies do work. But they have their drawbacks, as you would expect from any rig this simple.

30 Meters

My first Pixie clone was built for 30 Meters (10.108 MHz crystal).

Immediately I ran into these problems:

1) AM (SW) broadcast interference/intermod/overload: There are a lot of strong shortwave broadcast stations around 9 MHz, and more around 11 MHz; when your input filter is a Pi-type (a low pass filter with reactances around 50 ohms), there is next to no selectivity at the front-end (the receiver is not able to pass just the narrow band you're interested in, while rejecting the surrounding interference several KHz away. Instead, it passes everything below the Pi network's cutoff frequency. Of course, only the frequencies adjacent to the VXO get converted down to baseband; yet, a broad front end invites AM overload problems).

I tried changing the Pi filter to one of several variations on a parallel LC tank circuit, tapped down for impedance matching to the 50 ohm antenna, but invariably I found that when the filter's "Q" was high enough (bandwidth was narrow enough) to reject the surrounding broadcast stations, the circuit would become unstable (crystal would "chirp" when keyed, or tank would go into self-oscillation).

2) The end-fed long-wire antenna (with antenna tuner) that I originally used, seemed very prone to picking up noise and common mode hum, which is that annoying, raspy hum that Direct Conversion receivers are particularly bad at rejecting (actually, they generate it, indirectly, when harmonics of the receiver's oscillator ride into the DC power supply rectifiers and re-enter the receiver through the power buss). Even when I used a 9v battery and discarded the well-regulated DC supply, I still experienced a similar hum when the antenna tuner was tuned to its resonant point.

3) Later I changed my station's antenna to a full-wave horizontal loop, with an antenna tuner, and noticed some improvement with the common mode hum problem; this seemed to be in line with advice from other QRPers that one ought to use a tuned, resonant antenna (like a dipole), or at least a balanced antenna, rather than an end-fed one which is unbalanced electrically. Changing back to a Pi filter, now that I was using a loop antenna and tuner, seemed to eliminate most of this hum problem.

But overall, I still didn't like the tiny rig's performance.

Changing to 80 Meters

More recently, a fellow Ham got bitten by the Pixie bug, and bought a kit from the Internet. Our experiments focused on 80 meters this time, using a 3.579 MHz colorburst crystal. This worked much better, I found, because there aren't many megawatt European shortwave stations operating below 5 MHz. On 80 meters, you can get away with a front-end whose selectivity is as "broad as a barn door", as they say. I also tried a 7.040 MHz crystal (40 meters) and, while it certainly worked, I found that the old AM broadcast blanketing was happening again. Ditto with 14.060 MHz (20 meters), where I thought I would not have that problem. So I switched back to the 80 meter colorburst crystal, and the only real hassle I experienced was when W1AW would come on the air with its nightly code practice... although a couple of KHz away from my crystal frequency, the signal blasts into my South Central Pennsylvania QTH with a strength of around +30 dB over S9 some nights, and the high pitch is piercing enough to make me want to sail the ol' Pixie right out the basement window!

So I decided that I couldn't live with the bare-bones Pixie... time to modify and enhance her!

KD1JV's "AP-80" Alternative Pixie

I know some of my readers will wince at the suggestion-- why would I want to "complexify" the Pixie when it was meant to be a "minimalist" rig? And, If you're gonna double or triple the parts count, why not just build a "real" transceiver, or a "real receiver [a Superhet]", etc., etc.

Well, just when I started feeling guilty and un-American about all this, I ran across KD1JV's web page, and read about his majorly-modified Pixie, the AP80. While I certainly admire what he has done, he also makes a few criticisms of the Pixie that I have to take exception to:

"The transmitter was unstable" -- Not my experience, unless I tried a parallel tank in place of the output Pi network.

"Couldn't hear anything but exceptionally strong signals" because the detector "has large conversion losses since it's not configured in a very optimum way." -- Nope, mine is quite sensitive, maybe because of all the audio gain and filtering I put in.... see below.

"PA got very hot putting out what little power it did." -- No, I was getting S6-S7 reports from my buddy over 30 miles away (lots of Pennsylvania's rolling hills and small mountains between us) using just a 2N3904 running off a 9 volt battery! And it stayed perfectly cool.

CAUTION--I did find that, when I used a 2N3053 for the output transistor, it got hot if there was any reflected power showing on the SWR/Power Meter, so be careful to tune your antenna system to resonance and proper Z-match with a tuner. Once tuned to maximize forward power and eliminate all noticeable reflected power, the transistor stayed relatively cool to slightly warm, with no heat sinking. I recommend you use a small heat sink-- one with fins that slips down over the metal body of the 2N3053 transistor.


"All the receiver gain is supplied by the LM386 audio amp, which is strapped at full throttle. What we end up with is a high level of background noise." -- I use 2N3904 NPN transistors throughout, followed by an external Radio Shack amplified speaker. Very pleased with the results.

KD1JV presents a fairly complex makeover of the Pixie, and, don't get me wrong, I'm not at all criticizing his work-- this is what Ham radio is all about, and we need more innovators in our ranks!-- but when I saw how big his circuit got after his modifications were implemented, I thought "Gee, I think I can build me a Pixie with a lot less complexity, but that still represents an improvement over the original." So I got over my guilt trip and wasted some 2N3904's and circuit board space in coming up with my own mutant Pixie-style transceiver.

My version of the Pixie

Here's the schematic for my version, which I named the "RixPix", cuz it's, well, you know-- "Rick's Pixie"! At least it distinguishes my modified circuit from the others, should anyone out there actually want to try building it.

First off, notice that there's no LM386 IC. I have built enough toy projects over the years (as well as taught Electronics Labs at local tech schools) to have built up a real animosity toward that chip... if it can find a way to motorboat or squeal, it will. All "band-aids" notwithstanding, like the .1uF cap and 10 ohm resistor the applications notes recommend you hang off the output pin to ground. IF THE DAMN THING IS SO PRONE TO OSCILLATE WITHOUT THAT BANDAID, WHY DIDN'T THEY PUT IT INSIDE THE CHIP ALREADY!? Deep breath.

So call me an old fart, but I like 2N3904 bread and butter transistors. And I use them everywhere I can.

I'm also a fan of Radio Shack's little beige Amplified Speaker ($11.95), which I tack on the output of most of my radio projects. One of the few good things they still sell. And before some wise-acre points out that there's an LM386 inside that Radio Shack amp, all I can say is, If they were able to tame that bronco, more power to 'em. I'd rather not be bothered with pulling my hair out trying to figure out the optimum parts layout in every project-- just so the 386 doesn't squeal every time I turn the thing on. Besides, the little beige plastic case gives the 2" speaker just enough resonance to sound like it has some "body" and not like a piece of tinfoil.

Now let's back up and start with the crystal oscillator.

I had a spare 100 pF variable cap; that got mounted in the front panel of my L-shaped copper-clad board-chassis. With 100 pF, and a single crystal, I find that I can get a frequency swing of about 1 KHz, maybe a little less, from a colorburst crystal. The frequency goes up to about 3.5805 KHz when the cap plates are totally unmeshed; 3.5795 KHz when fully meshed.

Based on a trick I saw somewhere, I paralleled another crystal of the same frequency across the original; I was delighted to find that I can now "rubber" the VXO (Variable Crystal Oscillator) almost double the frequency range as before-- about 1.75-1.8 KHz in the 80 meter band. I imagine one could get 2 or more KHz at 40 meters and above, but I haven't tried it yet. (Yes, I did try putting a third crystal in parallel with the other two; no real improvement, so I stuck with just two.)

You will notice that I don't have any provision for RIT (the ability to automatically offset my receive frequency 600 to 800 Hz from my transmit frequency so that I'll be a little bit away from zero-beat with the other Ham with whom I'm having a QSO-- so I'll actually hear him!). I tried adding a transistor switch and capacitor (as many have done) but don't like how the frequency offset changes so drastically from one end of the VXO tuning to the other. Still working on this... will probably just throw in a toggle switch that either puts the 100 pF variable in circuit or shorts the bottom of the crystal directly to ground. I'll label the switch "Tx/Rx" or "XMT/RCV". Or if I'm particularly lazy that day, "T/R".

The feedback caps in the VXO are 100 and 150 pF; no special reason except they're the values I happened to have (the original used 100pF for both).

In the output stage (Q2), you may have noticed that I don't have any base bias (pullup) resistor at all, even though the original Pixie had a 33K or similar. I found, through repeated experimentation, that it makes absolutely no difference whether the resistor is there or not. So I opted to leave it out, assuring me that the drive from the oscillator must be high enough to be operating the second transistor in Class-C mode, which should keep the transistor the coolest anyhow.

I found that the output power could be increased, to a point, by increasing the value of the coupling cap between Q1 and Q2 -- mine started showing some instability at .001uF (1000pF) so I went back down a notch and left the value at 470 pF.

Again, the original Pixie had a 22uH choke in the collector of the power amp Q2. I surmised that the original designer must have chosen that value because it ought to be at least 10 times the value of the inductor in the Pi network. The Pi network L=2.2uH (for 80 meters). I determined that any available value of inductance should work as Q2's collector choke, if it was 22uH or greater, so I originally had a 1 millihenry inductor in its place. Cuz I had one.

Later, I discovered through computer modelling that, if that collector choke were reduced to match the inductor in the Pi filter, a nice resonant peak occured which significantly boosted the output power... up to 1 Watt with a 2N3053 output transistor and a 12 volt DC power supply (rather than the usual 9 volt battery). By nature a Pi filter is a lowpass cutoff filter; here we give it an additional resonant peak just before cutoff, which increases the output power.

My Pixie has had a 2N3904, 2N2222A, and a larger 2N3053 as its output power amplifier. They all work. I finally left the 2N3053 in as my preferred choice.

So the modifications mentioned above enabled me to get more power out than the design of the original Pixie allows for.

Though I don't have an accurate QRP wattmeter, I did the next best thing: I hung a .01uf cap off the output antenna jack, a 10k resistor to ground on the other side of the cap, a 1N34A germanium diode (anode toward antenna; cathode [stripe] toward voltmeter) also off the cap and top of the 10k, a .001uF cap to ground after the diode's cathode, and a DIGITAL voltmeter across that .001uF cap. With the voltmeter set to the 20 VDC range, I measured about 10 VDC across a 50 ohm "dummy load" when the Pixie's inductors were peaked. [See schematic below:]

The calculations are as follows:

First, the DVM, with its 10 Megohm input resistance, draws almost no current through the germanium diode; therefore, the voltage drop across it is negligible. So I take the 10v reading at face value.

This voltage is PEAK. To calculate power from voltage and resistance, I need to convert the peak voltage to RMS form, which is done by multiplying the 10vpk by .707, giving 7.07vrms.

Now I use the formula, P=V^2/R.... 7.07 squared, divided by the 50 ohm load.

The answer is .999 Watts (999 milliWatts)-- close enough to be called "1 Watt" to me!

In QSOs between my Ham buddy and me, he has been giving me consistent S5-S7 reports, early in the evening on 80 meters, with about 30 miles or so between us, with mountains in-between. He uses a dipole and I use a horizontal loop, so it's debatable whether we're communicating with ground waves or warming the clouds with straight up-and-down, very short "skip". [These experiments were performed in late March 2005.]

[3/26/05 UPDATE: My buddy took his Pixie to Pittsburgh (western end of Pennsylvania) while visiting relatives for Easter. Breaking all the rules about using "balanced antennas", etc., he used a slingshot to get a piece of half-wave wire (130 feet long) into a very tall tree, and, using only a 9 volt battery and a homemade antenna tuner, he worked me (near Harrisburg, south-central Pennsylvania) using that end-fed half-wave antenna, and no ground to speak of. The QSO took place (with simultaneous telephone call, for backup) around 8 PM on 3.579 MHz. His power must have been only hundreds of milliwatts; my Pixie put out about a Watt into my horizontal loop antenna through a Heathkit tuner. We could not carry on a full QSO-- our weak signals were fading in and out with the band that night-- but we were able to copy about half of the characters sent. Like kids with new toys, he and I continue to be amazed that we can communicate with these ultra-simple transceivers putting out a Watt or less!]

[4/2/05 UPDATE: Tonight I worked a station in Connecticut with my 1 Watt RixPix, on 3710 KHz. The CT. station gave me an RST of 599. Not to be outdone, my local friend (mentioned above) worked Oklahoma on 30 meters; his kit-built Pixie puts out something like 250 mW or so. We're both convinced that these little buggers work just fine!]

The theoretical 50-ohm calculations for the Pi output filter are 2.22uH for L and 890 pF for the two caps. I used 1000pF (.001uf) mylar caps and wound about 22 turns of #26 or 28 magnet wire on an Amidon T50-2 red-gray toroid, adjusting the spacing on the turns a little for maximum output of the transmitter into a 50 ohm dummy load. (Do the same with the other toroid that I have used in place of the Q2 collector choke used in the original Pixie.) The point: Use the nearest parts values you can, and don't sweat it with trying to be exact. Just adjust the spacing of the turns on the coils for highest power into a 50 ohm load.

Now for the audio section of the receiver circuit.

I like 2N3904's, I like grounding the emitter and using collector feedback bias in my preamp stages, but I hate putting blocking (coupling) caps in every stage. So, after the initial blocking cap, I use DC coupling, which then necessitates using a rather large 270K resistor in the collector of Q3, since that resistor doubles as Q4's base bias resistor. The .022uF cap across the 270K is a lowpass filter that mellows the audio and gets rid of any vestige of RF carrier missed by the cap across Q2's emitter resistor. (By the way, the .01uF for the initial blocking cap may seem too small for a transistor amp, but I verified by experiment that anything larger (such as the usual 4.7-10uF electrolytic) is totally unnecessary here... perhaps because the big resistors limit the current and cause the input impedance of this stage to look larger than usual.)

Q5 and Q6 are my pride and joy. Each one is a transistor version of the textbook op amp circuit known as the Multiple Feedback Bandpass Filter, with a center frequency of about 750 Hz (I aimed for 800 but with practical parts values, I got 750. What's 50 Hz among friends?). I used two of them in series or cascade because I wanted relatively steep rejection of CW tones outside the passband, with as little "ringing" as possible... See, I stubbornly tried to "cheat", for a long time, by trying to squeeze too much performance out of a single filter stage. Like my LC tank problem mentioned several paragraphs back, I would find that I needed high Q (and maybe some Gain to boot) to make the single stage's 6 dB per octave rolloff seem like it was actually earning its living, but would run into instability problems (that's when sharp filters become sine-wave oscillators and "ring" continuously at their center frequency). Finally I bit the bullet and built two stages (giving a total rolloff of 12 dB/octave) and decided I could stagger their tuning a little, if I wanted. Then when I heard the result, even with identical parts values (and identical center freq's), I was satisfied enough with the gentle "sea-shell at the ocean" sound (no ringing, but a little distortion on W1AW's monster signal) that I left things alone and settled for the circuit values you see in the schematic above.

[3/25/05 UPDATE: I changed the 2nd filter's ground resistor (nominally 560 ohms) to a 1K ohm trim potentiometer, so I could vary the 2nd filter's center frequency a little... after a while the constant "seashell" sound of a narrowband filter gets tiring to listen to.]

Now W1AW's signal is suppressed enough, when I'm tuned a couple of KHz away, that it doesn't pierce through annoyingly; yet any weak signals in the passband are accentuated just enough to make their presence known. RTTY sounds downright musical!

The audio is then routed to the output jack, mounted in the front panel of the copper chassis, into which I plug a 1-ft audio cable leading to the Radio Shack amplified speaker.

So yes, I do have quite a bit of audio gain going on here. But I have had very little problem with squealing or howling or motorboating... sometimes if the Radio Shack speaker is too close to the receiver box, a little bit of acoustical resonance (vibration) builds up when listening to a strong CW tone, right in the middle of the filter's passband. I simply move the amplified speaker a little farther away and that usually solves the problem.

Strange Experience

The only really strange "funny" that I've experienced was some intermodulation-- one weekend when there was a lot of strong RTTY on 80 meters (probably a contest), I was mystified by a most unusual "new" form of modulation on the air... I heard Morse Code (CW), but instead of the dots and dashes being made up of pure tone, they were made up of the "beadle-beadle" sound of a RTTY transmission! In other words, I was hearing some RTTY being "gated" on and off by a station transmitting straight CW! It took me a couple of minutes to figure out (with the help of my "real" Kenwood TS-520S receiver) that there were indeed two strong stations, one sending CW and the other sending RTTY, some KHz above my colorburst crystal frequency of 3.58 MHz, and my RixPix front-end was mixing them in its overloaded detector stage. So, Pixie builders, beware of strong signals lurking just beyond your crystal's neighborhood. A better receiver would reject them as out of the immediate tuning window.

Future Enhancements

What next for the RixPix? Oh, I don't know. If you're like me, you go through alternating spurts of creativity and longer intervals of laziness where you just don't feel like "getting into it"... so I don't know if I'll ever really add all of the following, but I'd like to:

A decent RIT

Sidetone so I can hear my CW as I send

Some easy way of switching crystals and Pi filters for multiband operation

More power out while preserving most of the "spirit" and design of the original Pixie

An enclosure that, for once, looks halfway professional (my mechanical skills suck, in general)

Maybe - just maybe - taming an LM386 chip and building the entire audio amp and speaker into that dream enclosure.

.... And then I woke up.