Block Diagram of the transverter.
Click on picture for a larger version.

Homebrew transverter:

This is page has been quickly thrown together as a place to stick some pictures and descriptions of this 10 GHz transverter and other 10 GHz-related activity.

The signal flow is like this (See the block diagram to the right):

• Original master frequency reference:
• Originally, a 5 MHz ovenized master reference oscillator is responsible for maintaining the frequency accuracy.  Initial tests indicate that this oscillator is capable of holding the frequency to 1 part in 10^9, or about 100 Hz error at 10 GHz, but sitting on a bench I've seen it hold steady at at 2-3 parts in 10^10 (20-30 Hz error) for hours at a time.  (My GPS reference typically holds in the several parts in 10^11 range.)  In actual fact, bouncing around in a car may degrade this error to about 500 Hz error, but I don't know yet...
• This 5 MHz ovenized oscillator is doubled to 10 MHz:  I would have preferred to use a 10 MHz oscillator originally, but the 5 MHz one was already on hand and construction of a doubler was pretty easy.  The doubler circuit consists of a 2N3904 running in class-C followed by a two-stage bandpass filter consisting of two 10.7 MHz IF cans.  (Yes, 10.7 MHz IF cans easily tune to 10 MHz with no modification.)  As it turns out, crystal oscillators in the 5 MHz range are the quietest and most stable, anyway.

• New frequency reference:
• More recently, I have switched to using an external 10 MHz frequency reference rather than the original on-board unit.  This made things a bit simpler and saved a bit of power as only this reference needs to be kept powered-up, ready for use.  This change was done because of the completion of a 24 GHz transverter that also used a 10 MHz reference and it made more sense to use just one reference for everything.  For more information on the frequency references that I use, see the:
• 10 MHz Oven-based frequency reference.  This is a quartz-based 10 MHz reference with accuracy and stability sufficient to maintain accuracy to within a few hundred Hz at 10 GHz.
• 10 MHz Rubidium frequency reference.  Using a surplus 10 MHz Rubidium frequency reference module, this will provide accuracy to within a few hertz at 10 GHz - far closer than the IF rig would be to frequency at 70cm!

• Main VCXO:
• The main VCXO runs at 18.4 MHz using a 2-stage butler oscillator.  This is tripled - and then doubled to obtain a 110.4 signal MHz for application to the "lock input" of the "brick" oscillator.
• A portion of the 110.4 MHz is mixed with the 10 MHz reference with a harmonic mixer and the result is a 400 kHz signal.  This 400 kHz signal is compared with 400 kHz divided down from the 10 MHz reference using a hybrid flip-flop/XOR phase/frequency detector design based on the AD9901 chip (but using two 74HC74's, a 74HC86 and a 74HC00.)  This particular design is unique in that it has properties of the 4046's flip-flop frequency detector in that it will pull up/down if the frequency is low/high - BUT it does not have a "deadband" in the middle of the lock range like the 4046:  This deadband phenomenon can cause "bouncing" that could be catastrophic in a noise-sensitive application like this.  Practically speaking, a simple XOR-gate based phase comparator (preceded with divide-by-two flip-flops to guarantee square waves on both inputs) would have worked equally well, but I wanted to try this design to see how well it worked.)
• The 110.4 MHz signal, firmly locked to the the 10 MHz reference, is output to the "Brick" oscillator.  A typical "Brick" oscillator consists of a high-powered (1/2 watt or so) cavity oscillator in the 1.5-1.7 GHz range followed by a snap-diode multiplier and filter.  In this case, the oscillator runs at one-sixth of the output frequency of (for 9936 MHz, this would put the oscillator at 1656 MHz) and is locked to the 15th harmonic of the 110.4 MHz input.  The wide loop-filter bandwidth allows this otherwise free-running oscillator to take on the stability of the original 110.4 MHz reference.  In this way the 110.4 MHz input is "multiplied" 90 times to 9936 MHz.

• "Brick" oscillator:
• The "brick" oscillator operates at 9936 MHz and is used as the local oscillator (about 5-10 milliwatts) which is fed into a mixer. In this case, the sum of the LO frequency plus the IF (e.g. 9936 MHz plus 432 MHz) puts the final mixing product at 10368 MHz - the SSB/CW calling frequency (which, itself, is 72 times 144 MHz or 24 times 432 MHz.)

• IF Chain:
  
• The 70cm output of this mixer at UHF is applied to a 70cm GaAsFET preamplifier (using an MRF966 - it's a preamp that I happened to have laying around...) and then to the IF radio (an FT-817 in this case.)  The purpose of this amplifier is simply to prevent the IF receiver from being the limiting factor in post-mixer sensitivity.  Ultimately, this preamplifier will be switched in and out on receive/transmit and may be completely unneeded when a 3cm preamplifier is used.  The details of this preamp may be found in older editions of the ARRL Radio Amateur's Handbook:  It appears in my 1987 and 1991 versions and no doubt the years in between (and a few years before and after.)

• RF Chain:
  
• The RF input port of the mixer (or output, on transmit) is connected to an interdigital bandpass filter.  This filter has minimal insertion loss (1 db or so) over most of the 3 cm band but it has over 50 db of loss at 9936 MHz and much higher at even lower frequencies.  This filter serves several purposes:
• Keeping local oscillator bleedthrough out of the transmit chain.  If this filter were not present, the 9936 LO output would appear in the transmitted signal, only 15-25 db down or so from the peak output power.
• Suppressing the transmit image.  On transmit, were this filter not present, difference output (at 9504 MHz) would have just as much energy as the original.  Not only would this be out-of-band, but it would be wasting 3db of transmitter power.
• Suppressing the receive image.  Although it is unlikely that there are any signals at the difference frequency (9504 MHz) this filter prevents those from appearing.  More importantly, if a preamp precedes the mixer, this filter provides a significant improvement in noise figure in that it removes the "image noise" - that is, the thermal noise that appears around 9504 MHz - from the preamp.  Note that this benefit only occurs if there is gain preceding the bandpass filter.

• There were way too many clip leads and flying wires.  Some of these clip leads/wires carried low-level/high frequency signals, such as the PLL output, reference signals, and power supply lines with circulating currents.

• The loop filter/amplifier was almost nonexistent.  The output voltage swing of the PLL's frequency/phase detector was only +- 1 volt about 2.5 volts:  This wasn't quite enough to provide proper frequency control on the VCXO.  I simply tacked a capacitor and a couple of resistors to make a really cheesy loop filter and was surprised that it worked at all.

• On the VCXO module, nothing had been "tidied up."  The various multiplier stages had combinations of poorly-bypassed power supply leads and un-dressed signal feeds running around, all unresolved remnants of the initial testing/construction.  The result was that the 110.4 MHz output signal wasn't as clean as it could have been.

• There was inadequate output of the 110.4 MHz signal from the VCXO module to the brick.  This signal was at about 0 dbm - but it should have been higher than this for the brick to be happy.  As it was it would sometimes takes several "power cycles" of the brick before it locks up.

• Nothing was in a box.  Ultimately, these things will be in their own enclosures.  (The main 10 MHz reference will be its own, self-contained unit, however.)

• There is no really convenient way to see if everything has locked up, yet.

Since the "benchtop and clipleads" picture was taken, I screwed the various modules to a piece of plywood:  While not elegant, it allows the unit to be transported and easily worked on.

• With just the mixer, the maximum output is very low - about 30 microwatts at the desired frequency.

• Using my FT-817 as an "IF Radio" (on 432 MHz) I can easily detect thermal noise from the earth when using a military surplus 3 cm preamplifier. Pointing a 17 dBi gain horn at the sky, there is an S-0 noise reading while pointing the antenna at the ground causes a reading of S-2.

Note:  The circuits depicted are drawn as-built.  It is likely that further improvement could be made in terms of optimizing performance and/or simplification.  These diagrams are provided simply to show one possible approach to solving the technical problems presented and are, in no way, intended to represent the ideal or state-of-the-art.

These schematics are hand-drawn and then scanned.  While reasonable efforts have been made to assure accuracy and legibility, it's possible that errors may be present.  If you have any questions, please feel free to send an email.

Note that not all bypass capacitors (on ICs, etc.) are shown on the schematic:  These capacitors (0.01 or 0.1 uf) are liberally sprinkled about the various circuits to provide bypassing to prevent unwanted cross-coupling of energy on power supply lines.

Since drawing the schematics, a few changes have been made - not all of which have been mentioned on these web pages.

NOTE about the "brick" power supply:

Due to a mishap, I managed to blow up my original "brick" oscillator.  Fortunately, I was able to procure another "brick" of the same manufacture and simply fit the multiplier module to it, retuning the main oscillator.  Instead of requiring +24 volts, it operated from -20 volts so I had to constructed a -20 volt power converter for it.  At the moment I don't have a diagram of that converter on this web page, but that can be supplied upon request.

Schematic-  Power converter:  This is a very simple boost-type voltage converter using the ubiquitous TL494 PWM switching controller IC - the same IC seemingly found in almost every PC power supply ever made.  This converter operates with an input voltage in the range of 8 to 16 volts (the upper end mainly limited by the voltage rating of the power supply bypass capacitors) and can supply at least 1 amp at 28 volts.

Operation:  The input voltage is filtered and bypassed, using low-ESR electrolytic capacitors and a series choke.  Other than providing a low impedance voltage source for the switcher, this filtering prevents switching frequency components (22 kHz and harmonics thereof) from being superimposed on the power supply line, possibly working their way into other circuits.

The TL494 consists of an oscillator, a voltage reference, some comparators, and some output switching logic. The square wave output of the oscillator is fed to the chip's driver transistors (through the switching logic) which are, in this case, configured as a totem-pole driver to a power MOSFET transistor, a MT5N05EL which is a generic 60 watt, 15 amp, 50 volt device (with a 0.14 ohm ON resistance with 5 volts on the gate) that I pulled out of my parts bin:  This transistor is nothing special except that I bought several rails for nearly nothing at a hamfest swapmeet several years ago and it is likely that a beefier transistor (with lower ON resistance) might improve efficiency somewhat.  When the transistor is on, current through the 20 uH inductor causes energy to be stored magnetically:  When the transistor is switched off, this energy is released as the field collapses and the resultant current flows through the diode (a high-speed diode pulled from a defunct computer power supply - a "normal" diode like a 1N4001 will not work well!) and into the output capacitors.

• The 20 uH inductor was another item pulled from the miscellaneous parts drawer and it probably came from a computer power supply.  Its core is about 2.5 cm long with the windings covering 1.25 cm of it and the wire used appears to be 18 gauge. Several inductors in the calculated range (15 to 30 uH) were tried and this is the one that ran coolest under full load.  There's no magic here - just do pick an inductor in the 15-30uH range, try it in the circuit and see which one runs the coolest while providing the necessary output voltage at the expected load.  Note that reducing the input supply to 10 volts or so will put more stress on the voltage converter and is a good way to verify that the coil being tested will be suitable.

• While this inductor will likely get "warm", it should never get too hot to touch.  If it does, it wire may be too thin and/or its core may be saturating, indicating that it may be too low-power a device simply the wrong type!

• It is possible to build a simple fixture and use a standard computer sound card and program to measure capacitance and inductance with reasonable precision.  Rather than place a link that will likely become outdated here, simply put "use sound card to measure inductance" into your favorite search engine and a number of articles will show up.
  

The TL494's internal 5 volt reference is compared with a resistively-divided voltage from the output capacitors and this divisor ratio (in this case, 10k and 47k) is used to set the output voltage.  When the output voltage exceeds the threshold, the oscillator's output is gated off, shutting off drive to the switching transistor.  In this application, a portion of the "feedback" signal (output line from the voltage comparators) is fed into the inverting input of the comparator:  This feedback allows duty-cycle modulation of the drive signal from the oscillator, allowing for smoother, more-responsive regulation.  Without this feedback, the converter would operated in a "discontinuous" mode where the drive was abruptly started and stopped.  In applications (such as this) that could be potentially sensitive to power supply fluctuations, the discontinuous mode of operation may pose stability/purity problems.

The output from the storage capacitors goes through a 80 uH choke followed by more filtering.  This filtering, too, is important to prevent bleedthrough of the switching signal through the output voltage and because the switching currents present in the 20 uH inductor are mostly removed by the filter capacitors, the magnetic characteristics of this choke aren't as critical and almost all of the heating will be due to IR losses.  Note that the schematic indicates the use of a "common-point ground":  When using a switching supply with high currents, good-quality capacitors, a robust ground plane, and good wiring techniques are required to prevent the energy from the switching frequency getting into everything and  one of the most important of these techniques is the use of a common-point ground - a single location where all high-current conductors (and components) tie together.  Were this not done, I/R losses would impose the switching frequency on otherwise-unrelated components, reducing the effectiveness of output filtering.  Note that the other chokes, because they aren't exposed to high switching currents, need not be as "heavy duty" as the 20 uH choke used in the voltage converter.

It's worth mentioning that low-ESR capacitors are absolutely essential in this application where noted!

While every low-ESR capacitor that I have seen has stamped on it a temperature rating of 105 degrees C, not every 105 C capacitor is of the low-ESR version!  Often, low-ESR capacitors may be found in switching supplies - but be absolutely certain that the capacitor itself isn't bad and wasn't the reason why the supply itself was junked.  How important is the use of a low-ESR capacitor?  Just for grins, I put a "normal" 470 uF electrolytic capacitor in place the two paralleled 220 uF units:  The 470 uF capacitor immediately got too hot to touch and the efficiency of the voltage converter plummeted.  Had I left the unit running, the capacitor would certainly have vented or simply blown up!  In contrast, the two low ESR 220 uF units barely run warmer than ambient temperature.  In the other positions calling for low-ESR capacitors, the AC current isn't nearly as high, but a "normal" capacitor doesn't work there, either because they have high enough impedance at the switching frequency that they do not effectively filter out the residual ripple.

The two devices that require this upconverted voltage (the 5 MHz ovenized oscillator - which was replaced with an external reference and the "brick" oscillator) actually operate at 24 volts.  To provide this voltage - as well as to further improve power supply cleanliness and stability, each of these other devices has its own 7824 3-terminal regulator.  While the addition of a linear regulator decreases overall efficiency somewhat, the relatively small input/output differential voltage makes these losses fairly small.

To conserve power, I added some power switches to allow me to turn off the "Brick" oscillator and the VCXO module and lock unit and all that remained powered up was the 5 MHz ovenized oscillator.  Being able to power down these other circuits slashes the current consumption but allows the most critical component of the system - the 5 MHz reference - to remain warm and stable.  The "Brick" and the lock unit stabilize within a second or two of being powered up.

Schematic- VCXO Module:

Page 1:

The heart of this module is the 18.4 MHz oscillator.  A fundamental-mode oscillator was used (rather than starting out with an overtone oscillator at 110.4 MHz) for two important reasons:

• Fundamental mode oscillators are generally more stable in terms of frequency and phase noise.
• It is difficult to frequency-pull overtone oscillators.  Because this was to be a VCXO, it is important that electronic frequency control be reliable and have sufficient tuning range to accommodate all anticipated temperature conditions.

This oscillator is the well-known "Butler" type.  While slightly more complex than other types this oscillator was chosen because it is cleaner and more stable than many others.  The crystal is placed in the feedback path of two amplifier stages, and with the relatively low crystal drive, internal heating is minimized, aiding stability.  Also, with the crystal in series with the signal path (and with both amplifiers operating well within their linear regions) the output signal is effectively filtered through the crystal itself.  Electronic tuning is accomplished by using a varactor diode (an NTE614) in the crystal's feedback path allowing about 1 kHz of "warpage" at 18.4 MHz - more than enough range to accommodate thermal/age-related drift.  Note that in low phase-noise oscillators, it is common to use back-to-back varactor diodes to minimize incidental phase modulation:  I may make this change at some point.

Also attached to the tuning line is a very simple loop filter.  The output of the phase/frequency detector (see below) is essentially a square wave with a varying duty cycle (at least when at/near lock) and this filter integrates the voltage to a DC level proportional to the duty cycle.  A simple transistor "inverter" is used to increase the voltage range from the phase/frequency detector:  The output swing voltage of the phase/frequency detector, after filtering, is from approximately 1 volt to 4 volts, but the transistor, being operated from the 9 volt supply, extends this to about 8 volts on the high end, increasing the range of electronic tuning.  Of course, the output of the phase/frequency detector must be "pre-inverted" to accommodate the inversion performed by this transistor.  The loop filter shown is quite minimal:  In the future, this will be improved to reduce the amplitude of the reference sidebands.

Following the 18.4 MHz oscillator is a FET-based buffer:  This provides minimal loading of the oscillator and drives a class-C tripler that multiplies the 18.4 MHz crystal frequency to 55.2 MHz.  A two-stage L/C filter isolates the 55.2 MHz component, removing most of the original 18.4 MHz energy.

Page 2:

Following the tripler is a frequency doubler that produces the 110.4 MHz signal.  It, too, has a two-stage L/C filter that removes 18.4 MHz and 55.2 MHz energy and the output from this filter is passed to an additional stage (also with filtering) that further amplifies/cleans up this signal.

Page 3:

The output from the 110.4 MHz amplifier is sent to two places:  One of these is a 110.4 MHz buffer comprised of an emitter-follower.  It is the output of this stage that is fed to the "brick" oscillator that produces the 9936 MHz local oscillator signal.

The 110.4 MHz output signal is also sent to another, identical emitter-follower amplifier that buffers the 110.4 MHz signal that is then applied to the base of the harmonic mixer transistor.  10 MHz energy (amplified/filtered from the master reference) is fed into the emitter of this stage and the collector, connected through a 455 kHz IF transformer, is used to extract and filter the minute amount of 400 kHz energy resulting from the mix between the 11th harmonic of the 10 MHz reference (110 MHz) and the 110.4 MHz signal.

Two stages of filtering clean up the resulting 400 kHz signal and it is amplified by a FET and buffered by a bipolar transistor stage.  The output of this bipolar stage is then lowpass filtered to remove traces of higher-frequency components and made available to the lock unit.

Schematic- Lock Unit:

Page 1:

Both the 400 kHz signal from the harmonic mixer and the 10 MHz reference pass through similar amplifiers that convert their respective inputs to logic-level signal suitable for driving digital dividers.

The outputs of these amplifiers are applied to XOR gates for buffering:  U1A for the 10 MHz reference signal, U1C for the 400 kHz output from the harmonic mixer.

The 10 MHz output from U1A is passed to U2 and U3, each of which is wired as a divide-by-5 counter, and the final result is a 400 kHz signal based on the stable 10 MHz reference oscillator.

Page 2:

The two 400 kHz signals (one from the reference and the other from the harmonic mixer) go to the two inputs of the phase/frequency detector.  Flip-flop counter U4A and U4B divide each of the signals by two, ensuring a perfect square wave from each source and these outputs are sent into U1D, an XOR gate.  If the frequencies are very close to each other, the output of U1D is a square wave with a duty cycle that varies in a way proportional to the phase difference.

The more different they get from each other, however, U5A and U5B begin to put clock glitches on the output waveform.  The ultimate result is that if the input frequency is higher than the reference frequency, the output from U6B (the combined output of the XOR gate and U5) will spend the majority of its time low - and vice-versa if the input frequency is lower than the reference frequency, allowing this circuit to provide PLL action over an extremely wide tuning range (unlike a simple XOR-based circuit.)  The action of the loop filter (on the VXCO module) integrates these clock glitches to a pure DC voltage for frequency control.  U6C is simply a logic inverter to accommodate the presence of the inverter/amplifier transistor at the loop filter on the VCXO board.

This sort of phase/frequency detector is similar to that found in the Analog Devices AD9901 and it has the advantage of not having the "deadband" problem that the phase/frequency detector used in the 4046:  In this condition, when the PLL is locked, the 4046's phase/frequency detector goes into a "tri-state" condition and when a frequency correction is needed, pulses (either high or low, depending on what is needed to provide the frequency correction) are output to correct the tuning.  The problem is that these correction pulses may occur infrequently enough that their frequency components are below that removed by the loop filter and the result is that "bouncing" can occur and low frequency noise may be introduced into the system.  In the phase/frequency detector shown, the XOR gate is doing most of the work when in lock and it is continually putting out a square wave of varying duty cycle and can never have any spectral components below this frequency (aside from the phase/frequency corrections, of course.)

Note that in this application, a simple XOR-gate phase/frequency detector would probably have sufficed, but I wanted to try this circuit to see how well it worked.
 

Improvements:

There is a lot of room for improvement.  A few things that come to mind immediately are:

• Improve the cleanliness of the 110.4 MHz reference.  At the moment, the harmonic filter is located right next to the 110.4 MHz output.  The result of this is that a small amount of 400 kHz energy gets coupled into the 100.4 MHz output, resulting in weak reference sidebands that are about 55 db below the 110.4 MHz carrier.  Unfortunately, the magnitude of these weak sidebands is amplified by the multiplication to 9936 MHz and, at that frequency are only 30-35 db down at worst and fall off rapidly within a few MHz on either side.  While undesirable, these sidebands are of insignificant total power and all fall within the amateur band, so there is no real risk of QRM.  Nevertheless, I plan to relocate the harmonic mixer to a separate module which should afford much better isolation.

• Improve the loop filter.  The loop filter is very minimal and contributes (slightly) to the presence of 400 kHz reference sidebands on the LO.  At some point I'll re-do the loop filter (after relocating the harmonic mixer and cleaning up those reference sidebands.)
  

• Re-box the VXCO.  The 18.4 MHz multiplier unit will go into its own box, and the lock unit and harmonic mixer will go into another, while all of these modules will be put into a single enclosure to replace the plywood substrate.

• Better T/R switching.  The 70cm post-mixer "preamp" is simply "transmitted through" (via a 10db pad) to get TX energy into the mixer.  While this works (and the relatively low amount of power does not harm the preamplifier at all) it is quite a kludge.  Improved T/R switching will also ease the interfacing of the transverter to other modules - such as an output power amplifier/RX preamplifer.

Related page:

• Avantek 3cm preamplifier details - A few details concerning the use of a surplus 3cm military surplus preamplifer that is used with this transverter.

Go to the KA7OEI main page.
 

20140812