3-Tube AM Stereo Transmitter


I think that a sizeable percentage of the people reading my web pages probably assume that I'm one of those crazy people who believes vacuum tubes are superior to solid state devices. That's not true (except maybe the crazy part). I like vacuum tubes because it's fun and challenging to build circuits using this old technology. It forces the designer to analyse the circuit requirements carefully, and to come up with a minimal optimum design. With solid state, you can throw a few hundred transistors at almost any problem in order to solve it. You can't do that with vacuum tubes, so you have to come up with a good basic design. Therein lies the challenge.

For some time, I've contemplated building an AM-Stereo transmitter using only vacuum tubes, and of course, the fewer, the better. Remarkably, as I will demonstrate, a good quality low power AM-Stereo transmitter can be built using only three vacuum tubes.

AM Stereo in a Nutshell

AM stereo broadcasts began in the USA, Canada, Australia and Japan in the 1980's. Initially, there were several different incompatible systems: Belar, Harris, Kahn, Magnavox and Motorola. All were based on variations of angle modulation of the carrier with the L-R stereo difference signal, and then the resulting signal is amplitude modulated with the monaural L+R signal, making the signal compatible with existing monaural receivers. This also made it reasonably simple to upgrade existing broadcast stations by replacing the existing crystal controlled exciter with an angle modulator unit, which was in the low power section of the transmitter. The high power amplitude modulator could be left essentially unchanged. In fact, the biggest cost usually was in upgrading the audio signal chain from mono to stereo.

Eventually, the Motorola C-Quam system won out and became the standard for AM-Stereo. Over the years, there has been much politically charged debate about the merits of the different systems. I don't intend to get involved with that debate here. However, for the purpose of explaining the following design I will briefly discuss some basic similarities and differences of the Magnavox and Motorola systems.

The Basic Transmitter Design

The Magnavox system is the simplest system, in that the L-R angle modulation, is just a linear phase modulation scheme. The Motorola system, although its full name is 'Compatible Quadrature Amplitude Modulation', it's simply a non-linear phase modulation scheme. The Motorola system uses a lot of circuitry to quadrature modulate the L+R and L-R signals onto the carrier, and then passes the resulting signal through a limiter, which after going to all that trouble, leaves just a phase modulated signal. So why bother? To answer that question would require that we get involved in the politically charged debate. Suffice to say, the difference between the linear and non-linear phase modulation is not significant except in certain special cases such as when an audio signal appears only in one stereo channel at high amplitude. Statistically, this happens rarely in most normal music. For more normal conditions, the output of the C-Quam phase modulator is a reasonably linear function of the L-R signal. For that reason, it appeared to me that the design of a C-Quam compatible transmitter could be greatly simplified by using a simple linear phase modulator circuit. In subsequent tests, I confirmed this. Even when using the worst case audio signals, although there would theoretically be some distortion in the received audio, this distortion was not audible.

To proceed then, I needed to come up with a simple phase modulator, a simple amplitude modulator, and simple audio matixing circuitry. A further requirement is a pilot tone generator. For the C-Quam system, this is a low level 25 Hz tone which tells the receiver that a stereo signal is present.

The audio matrixing combines the left and right audio inputs to create the L+R and L-R audio signals. This is easily accomplished using audio transformers with dual secondary windings, and then by connecting the secondaries as series aiding to get the L+R signal, and series opposing to get the L-R signal.

For the amplitude modulator portion of the circuit, I decided to reuse most of the circuit from The Improved 1-Tube AM Transmitter. It is very simple and works well.

That leaves only the phase modulator part of the circuit. After doing a bit of research, I decided to go with a reactance tube phase shift circuit. Actually, when restricted to vacuum tube circuits, there isn't a lot of choice in the matter.

It's worthwhile discussing the basic reactance tube circuit in a bit of detail,
because its principle of operation is not immediately obvious, and it is a very critical part of the transmitter circuit. The diagram shown here is the most basic reactance tube circuit. Assume that an alternating voltage V is applied to the terminals on the left hand side, as shown. The component X is a pure reactance—either capacitive or inductive. Also assume that the reactance X is a high value so that relatively little current flows through it. Therefore, most of the current I will be due to the plate current of the tube. The plate current of a vacuum tube is given by the following relationship:


Where IP is plate current, gm is the tube's transconductance, and VG is the grid voltage. The grid voltage is proportional to the current through the resistance R which in turn is equal to the current through the reactance X. Since the current through a reactance is 90° out of phase with respect to the voltage across it, the grid voltage VG will be 90° out of phase with respect to the terminal voltage V, and consequently, the plate current will be 90° out of phase with respect to the terminal voltage V. Since this reactive current is also proportional to the tube's transconductance, if the tranconductance changes, then the reactive current also changes. As it happens, a tube's tranconductance is a function of the grid bias voltage, so that if we change the grid bias, then the reactive current can be changed.
Thus, we have at the terminals, effectively, a variable reactor which can act as either a variable capacitance or variable inductance, depending on the component choice for X. This variable reactance can be combined with an additional resistance (R2 in the diagram as shown at the left) to make a simple phase shifting network. Here, the modulation voltage VMOD, controls the grid bias and hence the transconductance and phase shift. While this very basic circuit would work after a fashion, it has a number of drawbacks. One of the most serious is that the output amplitude varies with the amount of phase shift. Almost as simple, but without these drawbacks, is a reactance tube phase modulator circuit from F.E. Terman, Electronic and Radio Engineering, 4th Edition, McGraw-Hill, 1955, page 604. It is shown below.

The details of powering the circuit and biasing have been omitted in this simplified drawing. Terman gives only a very brief discussion of the circuit, but attributes it to S.M. Beleskas, Phase Modulation Circuit, Proceedings of the National Electronics Conference, Chicago, Vol. 3, 1947, pages 654-661, where an in-depth analysis is given.

The components marked X, are reactances of equal value. They may either be both capacitive or both inductive.

VIN is the RF carrier input signal which is to be phase modulated. VOUT is the phase modulated carrier output. I won't give a detailed analysis here, but its main advantages are:

  1. The output amplitude is equal to the input amplitude regardless of the amount of phase shift.

  2. A total phase shift of nearly ±90° is possible.

The phase shift of this circuit is given by the following formula:

ø=2 Arctan|X gm|

where ø is the phase shift, X is the reactance indicated in the schematic, and gm is the vacuum tube transconductance.

Given the Arctan function, it's immediately obvious that the phase shift will be a rather non-linear function of the transconductance. The following graph of phase shift vs. transconductance confirms this.

All is not lost, however. As it turns out, a tube's transconductance is a non-linear function of grid bias. Following is a set of transconductance curves from the General Electric Vacuum Tube Guide for the tetrode section of a 6CQ8 triode/tetrode.

It can be seen that these curves bend in the opposite direction to the phase shift curve. So, by choosing the appropriate screen voltage and grid bias, it may be possible to linearize the phase shift function over some useable range. The data from the transconductance curves were entered into a spreadsheet table, and then applied to the phase shift function. Using 100 pF capacitors for the reactors, and an operating frequency of 1500 kHz (giving a reactance X of 1061 ohms), and a screen voltage of 75 volts, the results are plotted on the chart below.

The blue curve is the transcribed transconductance curve for a screen voltage of 75 V. The red curve is the resulting phase shift as a function of grid bias. It can be seen that with a grid bias of -2.5 volts, the phase shift is quite linear over the range of ±45° which is sufficient for the present application. (For a C-Quam compatible transmitter, we don't need any more than ±30° phase shift.) Just to clarify, with a bias of -2.5 volts, the quiescent point is a nominal phase shift of 75°, and with suitable grid modulation, can vary from 30° to 120° without becoming significantly non-linear. The fact that the quiescent phase shift is 75° rather than 0° is immaterial, because all that matters is the phase shift relative to the quiescent point. The S-shape of the curve is also an advantage, because it partially compensates for the non-linearity of the C-Quam phase modulator when large stereo difference signals occur.

Having found a suitable candidate for the phase modulator, it was time to put together a prototype circuit.  The prototype was built breadboard style on a piece of wood as shown below:

Here is the original prototype schematic. Click on it for a larger version.

In this circuit, V1a is a crystal oscillator which generates the carrier signal. It is a common Colpitts type crystal oscillator. V1b is the phase modulator. Although a triode was shown in the simple circuits discussed earlier, the tetrode is preferable in this application. In a triode, the transconductance will vary with plate voltage, which is undesirable. In a screen grid tube, transconductance is reasonably immune to plate voltage variations. The 220 pF capacitors connected between the plates of V1a and V1b perform the function of the X reactors previously discussed. During testing of the prototype, the value of these capacitors was changed a number of times. During early testing, the 220 pF capacitors gave best performance, but with further circuit adjustments, a value of 100 pF was eventually chosen as the best value for the chosen operating frequency of 1500 kHz. However, there is a lot of leeway in the chosen values. I was able to switch operating frequency from 1500 kHz to 1000 kHz without changing the capacitor values, and only a minor V1b bias adjustment was necessary. Though, if designing the transmitter for a specific frequency, I would recommend choosing capacitors to give a reactance of close to 1000 ohms at the operating frequency.

V2a and V2b perform the same controlled carrier amplitude modulation function as previously described in The Improved 1-Tube AM Transmitter. The only difference is that the modulator section is not self oscillating. It receives the carrier signal (already phase modulated) from the output of V1b. The L+R modulation signal must be attenuated slightly to stay balanced with the phase modulator input. This is accomplished with the 3.3k/1.5k voltage divider between the audio input transformers and the V2a grid.

For initial testing, the pilot generator function was provided by an external signal generator. The pilot tone is combined with the stereo difference signal which is then used as the modulation signal for the phase modulator. The pilot tone does not appear in the amplitude modulated portion of the signal. During testing, it was found that the left and right channels were reversed in the received signal. This is actually good news, as it indicates that the receiver was detecting a stereo signal and except for the reversed channels, was decoding it correctly. It turns out that the phase modulator is reverse acting, and requires an R-L signal instead of an L-R signal. This was easily fixed by changing the secondary connections on the audio transformers.

Here are two audio files transmitted from the prototype transmitter, and as received on my Sony ST-JX450A AM-Stereo receiver:

Soundclip: Johnny Favourite Swing Orchestra

Towards the end of this sample, I unplugged the Right audio input to the transmitter for a few seconds, then plugged it back in and unplugged the Left audio input. You can hear some faint but noticeably distorted audio on the unplugged channel. This doesn't surprise me; in accordance with theory, the worst distortion should occur when there is only one channel present.

Soundclip: Leftfield, Original

I encountered some interference on my receiver while making these clips. However, after I recorded the first one, I managed to adjust the receiver's loop antenna to minimize the noise. So, the second clip is cleaner.

One other problem encountered in testing the prototype was that the centre (L+R) channel audio was shifted slightly over to the right of centre. While not overly offensive, it is audible in the above sound clips, and I felt that it needed to be addressed. After a bit of analysis I determined that if the carrier had any second harmonic content, this would have the identical effect of an off-centre signal, and consequently, the receiver was decoding it as such. It appeared that there were two separate sources of 2nd harmonic distortion. The first was in the crystal oscillator, and the second was due to the carrier input to the phase modulator being at a high enough level that it was causing self-modulation. The solution to the oscillator distortion was to add an LC low pass filter between the oscillator output and the input to the phase modulator. However, fixing the self modulation problem in the phase modulator was not so easy. It would require attenuation of the RF input in order to run at a lower level, which would then require another stage of amplification afterwards. Fortunately, there was an easier fix that didn't require any additional tubes. It is possible to eliminate the shifted audio simply by adjusting the tuning of the antenna matching network.

There is another possible method to eliminate distortion that I have considered, but haven't yet tried. It occurred to me that if the 2nd harmonic distortion of the oscillator could be made equal and opposite to that of the phase modulator, then it would cancel out. This could be done by inverting the output of the oscillator before going into the phase modulator. This inversion could be done using a 1:1 inter-stage transformer, so that additional tubes would not be required. At some point in the future, I may revisit this method, and see how it works. For the present though, the antenna network adjustment method is completely satisfactory.

With the prototype circuit working well, I decided to build a 25 Hz pilot tone generator. I used a 6HA5 triode for no particular reason other than the fact that I had some on hand.
An LC resonant circuit was chosen due to its high Q and good stability compared to other types of low frequency oscillators. The inductance being supplied by a small 1.1 VA power transformer (Tamura 3FD-256). Feedback is taken from the cathode.
Fine tuning of the pilot frequency is done by placing a variable load (using a 5k potentiometer) on one of the transformer windings which, due to the non linearity of the iron core, causes a change in inductance. With the component values that were chosen, the adjustment range is roughly ±3 Hz, which makes it very easy to get the required precise adjustment. Most specifications for decoder chips require the pilot to be 25 Hz ±0.6 Hz.
This oscillator can easily maintain this accuracy even with an unregulated power supply. However, because of the small adjustment range, the tank capacitor must be hand picked to get near the centre of the adjustment range. Fortunately, given the inductance of the chosen transformer, the required capacitance conveniently happens to be the standard value of 0.22 µF. As shown in the scope trace, this oscillator produces a reasonably clean looking sine wave output that doesn't require any additional filtering.

The complete final schematic is shown below. (Click on it for a larger version, which includes an additional multi-function metering circuit.)

Some differences between the preliminary schematic and the final:

  1. Trimmer capacitor added to the crystal oscillator tank. This was done in an attempt to reduce the 2nd harmonic distortion. It didn't help the distortion problem, but it did turn out to be necessary if a ceramic resonator is used instead of a crystal, as it provides fine frequency adjustment.

  2. V1b cathode bias resistor replaced with a potentiometer to allow bias adjustment.

  3. Addition of 680µH/22pF LC low pass filter between V1a output and V1b input. This was described above.

  4. Replacement of the fixed L+R audio attenuator with a 10k potentiometer. This allows optimum stereo separation adjustment.

  5. Inclusion of integral 25Hz pilot tone generator, as previously described.

  6. Injection of 25Hz pilot tone is now in series with R-L difference signal. This eliminates the previously used resistor network, and maximizes the amount of R-L modulation signal.

  7. Replacement of the antenna matching L-network and V2b plate choke with a simpler adjustable inductive plate load (a small slug tuned coil from my junk box) which forms a parallel LC circuit in conjunction with the antenna capacitance. This is simpler, and gives lower losses compared to the earlier circuit. The value of this inductor depends on the antenna length and operating frequency.

  8. Addition of a multi-function meter (shown only on the larger schematic) which is used to display audio input level (L+R), antenna network peaking (diode detector RF probe placed in close proximity to RF output lead), V1b cathode bias voltage level.

It's also worth mentioning that while this circuit was designed around 6CQ8 tubes, I've also tested it with 6GH8A, 6U8A/ECF82 and 6KD8 tubes, which all function identically, with no other circuit changes required. Interestingly, the 6CQ8 is described by the manufacturer, as a triode/tetrode while the others are described as triode/pentodes. In fact, they are all beam tubes, and their internal structures are identical.

The power supply circuit is shown below.

The power transformer is a 20 VA, 120 V isolation transformer removed from an old Razor Outlet. I added a 6.3 V heater winding consisting of 46 turns of #24 AWG wire (0.5 mm) wound over top of the existing secondary winding. The three 0.01 µF X-Y safety capacitors connected across the AC line are important in eliminating 60 Hz hum from the signal at the receiver, because they prevent RF from travelling back through the power mains and causing hum modulation via the receiver's power supply. The fuse shown in the power supply is actually a self-resetting thermal circuit breaker built into the transformer.

Final Construction

I constructed the chassis from a piece of 0.5 mm galvanized steel. This is a bit on the thin side, and subject to warping. As a result, it wasn't appropriate to have a high gloss finish. So, I used a textured matte "Stone" finish paint. Ventilation holes have been drilled for air circulation around the power transformer.

Here is the underside of the chassis. The two terminal strips have been soldered in place to minimize the number of screws visible from the top of the chassis.

Because the chassis is quite compact, I set out to pre-assemble as many things as possible before installing in the chassis. Shown here are the power transformer, power line filter, B+ rectifier and filter, and tube heater wiring.

The audio and pilot oscillator
transformers were designed for printed circuit mounting. So, it was most convenient to build several pre-assemblies using veroboard. These are shown here, along with the pre-assembly for the crystal socket (made from perf-board and a 9-pin miniature tube socket), and the mounting bracket for the antenna tuning coil.

Here, the power transformer, B+ supply, tube sockets, and antenna tuner are installed.

And finally, the photo below shows the underside of the completed chassis.

The cabinet was completed with the addition of front and side panels of MDF (medium density fibreboard). The side panels were stained with medium walnut stain and finished with lacquer. The front panel was painted black. Note that the completed unit is quite small 150 mm (6") square, by 50 mm  (2") high, excluding the side panels.

Checkout and Set-up Procedure

Set-up of the completed transmitter was fairly easy, having already worked out most of the bugs with the prototype. Although there are quite a few things to adjust, they can all be significantly out of adjustment, and the transmitter will still put out a signal. The procedure that I used is as follows:

  1. 1.Connect the antenna, turn on the power, and allow a few minutes for the tubes to warm up and temperatures to stabilize.

  2. 2.Ensure the main crystal oscillator is operating and is on frequency. Adjust the tank trimmer capacitor as necessary, to trim the frequency.

  3. 3.Set the meter to measure output signal strength, and adjust the V2b plate inductor for maximum output signal strength.

  4. 4.Adjust the V1b cathode bias pot to get approximately +2.5 volts on the cathode.

  5. 5.Ensure the pilot oscillator is operating, and is on frequency. Adjust the frequency potentiometer as necessary.

  6. 6.Tune an AM-Stereo receiver to the transmitter frequency, and adjust the pilot level potentiometer until the receiver stereo indicator lights.

  7. 7.Connect an audio source to the Left audio input of the transmitter, leaving the Right input disconnected. (Music seems to work better than a steady fixed frequency tone, for this adjustment.) Adjust the stereo separation control to get the best null on the Right channel of the receiver. Some faint distorted audio will likely be present on the right channel. Touch up the V1b cathode bias pot to get minimum distortion on the Right channel. As the cathode bias adjustment and stereo separation control may interact slightly, both adjustments should be repeated until there is no further improvement.

  8. 8.Repeat step 7, but swapping Left and Right audio channels. If the circuit is working properly, then the adjustments should already be optimum.

  9. 9.Connect a stereo audio source to the transmitter audio inputs and adjust the source audio level to give a good modulation level without distortion.

  10. 10.Readjust the V2b plate inductor as necessary to make sure that centre channel audio from the receiver is properly centred.


This is one of the most satisfying projects that I've built. Considering how few parts were used, and the simplicity of the circuit, the performance is exceptional. In the near future, I will post another audio sample that demonstrates the audio quality of completed transmitter.

Antenna Matching Update - 2015-09-07

Please refer to this page for updated information about:

Matching a Part 15 transmitter to a short antenna

Back to:



This page last updated: July 8, 2016

Copyright 2012, 2015, Robert Weaver