Today, to build a 10GHz SSB transverter is much easier than, say, 10 years ago. This is for two main reasons: there are good commercial kits on the market for medium-skilled hobbyists, and the  significant price drop of 10-12 GHz active components (GaAsFETs, Schottky Diodes) now widely used in Sat-TV equipments. If a Kit is still too expensive for your budget, there are also a lot of projects to try, but all requiring a quite high level of expertise and instrumentation, plus a good experience in precision PCB design and etching.

But there is a third choice, that I discovered after visiting the interesting website of G4HJW (www.g4hjw.metahusky.net) that shows the many uses of sections taken from commercial Ku band LNBs. There are many other projects from different sources of microwave projects built with modified LNBs, so you may think that I'm inventing the "Hotwater", but when I opened a Cambridge dual output LNB, I said: "Hey, this can become my first 10GHz transverter!"

This approach requires a deep tailoring of the project to the characteristics of the LNBs you can find, and is not easy because it needs a lot of time for adjustments and optimization. Really good is that with little microwave design experience you can get decent results if you have the right instrumentation to verify what you are doing. Perhaps you will never copy this project, but at least I hope to give some good ideas. 

-IMPORTANT- I did not believe, at the beginning, but it is absolutely true; GaAsFETS are veeeeery delicate. Anything metallic that gets in touch with their leads (no matter if they are on the PCB or not) can damage them. Use only non-metallic tools, hot-air soldering station and ESD precautions (the simplest one is not to wear shoes). Luckily, you can get plenty of low-power GaAsFETs if you have many LNBs to dismantle.    

REUSE OF LNBs 

My choice has been to put SMA female connectors to arrange the necessary transitions from the stripline tracks,  and at 10GHz it is not simple. The best solution I have seen is to drill a 1mm hole trough the PCB, exactly at the end of the RF stripline we want  to port out. Then, flip the PCB with the ground plane facing upwards, and with a 3mm drilling bit carefully rotated by hand,  mill about 0.5mm of GND copper around the hole. Prepare a female SMA for PCB vertical mount, removing all 4 GND corner pins and filing the little flange to be totally flat. Place the SMA connector with its center pin through the hole. and the ground flange well  flat on the GND  plane. Cut the center pin leaving no more than 0.5mm extra length over the PCB and solder it to the stripline with minimum solder quantity. Flip the PCB and solder the GND flange letting the solder fill any gap between the GND plane and the flange.

With all SMA ports in place, soldered on the GND side of the PCB, you cannot remount the PCB into the original shell, unless you make some holes for the SMAs. But this can be very unpractical because of wall thickness and because the feedhorn waveguide (in one with the shell) is  always where it should not be... Better to build a customized case for the PCB, and here I explain my way to do it.  All LNBs I've seen are built as a "sandwich", where the PCB is  between the outer diecast aluminum shell and the internal aluminum cover shaped to keep each section shielded from the others. Some self-screwing screws keep those three things together. The principle is to keep this sandwich structure, but replacing the outer shell with an aluminum plate that has been formerly cut and drilled to accommodate SMA, alignment pins (if any) and the screws. Plate thickness can be 2 or 2.5mm, and if you cut it  few mm longer than the PCB length or width you also get the room to drill the extra holes for screw mounting to a chassis (see picture).

-Hints- To easy locate where to drill the SMA holes in the aluminum plate, I mark the hole centers using the PCB as a template just after having drilled on it the 1mm holes for the SMA center pins. You can choose whether to drill round holes or cut square slots on the plate for the SMAs. I prefer the second choice using a jigsaw and a set of files. 
PCB material of LNBs is very soft and doesn't like heat too. It is very easy to bend or twist it when the SMA are connected to the testing instruments, and this will likely damage SMD components. 
And two further remarks: the screws that now tighten the aluminum plate to the shielding cover (with the PCB in the middle) can be non-self-screwing, to help mounting and dismounting. Verify if on the GND side of the PCB there are any tracks bringing supply, biases or other signals that will be shorted when the sandwich is assembled. Those tracks must be covered with good quality insulating tape (I've used Polyimide adhesive tape, that is also needed for other purposes in this project).

Another key technique in the art of LNB modification is the "gain block reversal". In all LNBs the signal path goes from the antennas to the IF output(s), but in the TX section of a transverter this path is the opposite. The pin arrangement of GaAsFETs helps a lot to do a clean job: there are two Source pins,  with Gate and Drain at the opposite sides of the package. This suggests that a GaAsFET can be desoldered and then resoldered rotated by 180 degrees. Yes, the impedance matching is not the same, but by patient cutting and adding stubs on the striplines you can still rely on 7-8dB gain from a reversed GaAsFET. 
Of course the Gate and Drain biases must be exchanged, but this is quite easy by cutting and crossing the tracks that go from the bias transistor to the RF decoupling capacitors. 

In the process of building this transverter I had to lower the resonating frequency of the 11GHz image filters. These filters are often made by coupled tiny halfwave striplines, that cannot be easily made longer. The resonating frequency of a stripline filter depends on the dielectric constant of the substrate, but  a part of the E-field lines also run in air, so the trick can be to artificially increase the dielectric constant of the air! I easily got good results using little pieces of Polyimide adhesive tape (Epsilon-r ~ 4), sticked here and there over the filter. A single piece covering the whole filter works well most of the times, but sometimes the frequency shifts too much and a smaller piece is better. That's real "cut and trial"! 

If you like to build a similar transverter, you must understand which LNB type suits well. First of all we have to consider LNBs with dual polarization (they have two LNAs and two mixers), with mixers being of the diode balanced type. It is not important if they are dual-band or single band because the DRO is unsuitable for narrow-band and it must be removed, but it is important that it is a dual-output type because there is a PIN diode matrix that can be used as RX/TX IF switch, making life much simpler.

Here is the block diagram of the AE2 LNB by Cambridge Industries that has inspired this project:

  Back to top

The Mix Unit we can build with a modified LNB needs few other blocks to make a working transverter. The most important is a precise and low phase-noise Local Oscillator (LO), then we need a PA and an antenna relay. I also added the automatic RX/TX switch over the IF cable (requires a modification in the IF transceiver), an output power indicator that also serves as a DC voltage check, a remote control connector (for pole mounting in fixed station) and a PLL to lock the LO to an external 10MHz frequency standard. Here is the block diagram of the complete project:

Let's now see how each section of the transverter has been made. 

MIX Unit
As told before, the MIX Unit has been made from a modified AE2 Cambridge LNB. Each block is indicated on the photo of the PCB (shield is removed), and for some of them I give some comments.

20dB attenuator: it is in place of one of the two INA10386 IF buffers, and works in both RX and TX. It is needed to reduce the power from the IF transceiver to a safe level for the RX/TX diode switch. The attenuation it introduces in RX is recovered by the gain of the IF RX buffer (~20dB). To sustain the 1W max power from the IF (0.5WPEP operating), the 55ohm resistor is made with 4x220ohm 1/2W SMD resistors in parallel. 

RX/TX Switch: the original switch was a matrix arrangement, that has been simplified to a SPDT arrangement. Original design significantly cuts signal below 700MHz, so, to work well at 432MHz, two stripline chokes have been replaced with SMD RF inductors.

6dB attenuator: it is in place of one of the original IF buffers (mark C1H) to further reduce the power from the IF transceiver to a suitable level for the TX mixer (~2dBm). Low power SMD resistors are OK.

TX Mixer: it works reversed, but no modification is needed because diode mixers are bi-directional. The little piece of fiberglass laminate is glued over the stripline balun for a better balance, in a position where the maximum LO rejection is obtained. This trick is not needed for the RX Mixer

Image BPFs: one for RX and one for TX. Both have been re-centered to 10.5GHz using Polyimide adhesive tape as described previously 

TX Buffer: it is a gain block that has been reversed as described here. The GaAsFET is not the original one and Drain resistor has been reduced to 27ohm to get a little more power. Two tiny copper foil stubs have been used to optimize gate and drain impedance matching. 

TX Driver: another reversed gain block. The GaAsFET is not the original one and Drain resistor has been reduced to 10ohm. Luckily no retuning has been needed to reach the target output power (11dBm).

RX LNA: these two stages have not been touched until an upgrade was done. I have just added a 1/4lambda short to GND at the RX IN connector for ESD protection.

IF RX buffer: also this stage has not been touched. Its gain (~20dB) recovers the losses of the 20dB attenuator.

LO splitter: the only modification has been to put the SMA connector to inject the LO frequency. The components of the original DR Oscillator have been removed. 

GaAsFET bias: apart of the Gate/Drain bias wires swap for the two reversed gain blocks in the TX path, this block has been modified to properly activate the RX and TX RF stages together with the PIN diode RX/TX switch and the Antenna Relay. This modification is quite simple, just requiring to selectively apply the Vbb  to the PNP bias transistors using the inverters of the 74HC14 (see the two 8.2K resistors and the two brown wires).

  Back to top

LO Generation
Unfortunately I did not have any LO source ready, like a cavity multiplier, so I had to build a good 10GHz LO source from the scratch, and this took the larger part of the time I've allocated to the whole project. My approach has been a "counter-stream" design: from the final frequency to the oscillator  because I had no previous experience on 10GHz multipliers and I tend to solve first the biggest problems... 

Once determined that the optimal LO power for the Mix Unit was something between 2 and 4dBm, my idea was to reach a 2.5GHz frequency by affordable multiplications, further multiplied by 4 with a portion of LNB implementing a saturated gain stage  followed by a stripline filter. This is very similar to what proposed by G4HJW, so I made some experiments, discovering that this kind of  multiplier is quite deaf and critical, but can work, provided that two gain stages are used and that the stripline filter accepts to be retuned down to 10GHz. With a +4dBm output level at 10GHz, spurious x3 and x5 multiplications were well below -40dB, so next efforts were  concentrated on earlier multiplications, oscillator stability and phase noise. 

LO Multiplier
For this transverter a good fit is a 5th overtone 92MHz crystal, that  multiplied by 108 goes to 9.936GHz: the exact value for a 432MHz IF. So the multiplication scheme I chose is:

92x3=276MHz
276x3=828MHz
828x3=2484MHz
2484x4=9936MHz

This is basic schematic of the LO Multiplier:

First two multipliers are a classical design, and have been realized "Manhattan style" on a dual-sided FR4 laminate. The 276MHz coils are on the backside, and all other components plus the the 828 MHz filter are on the front side. Last two multiplications have been implemented reusing some sections of a modified LNB. An interesting point is that the 2484MHz filtering is greatly simplified by using miniature monolithic filters for WiFi, that I took from an old dismantled WiFi access point.  Two 2484MHz  stages were needed to reach enough drive level for the GaAsFET x4 multiplier (~5dBm). This last multiplier is a 2-stage gain, where I added a trimmer to adjust the bias of the two FETs to let them work as much as possible in a non-linear region. Besides, this trimmer can adjust the LO level to an optimal value: if too low there will be less TX power and less RX gain, but if too much the LO leakage and the image level will rise dramatically. Of course, the image filter frequency has been lowered by putting some dielectric material on it.
In later 10GHz LO multipliers I made (for Andrea's transverter and for the CW beacon), I got a much higher efficiency for this "LNB multiplier" creating an intermediate "weak" x2 multiplication. 

The difference stays in the coupling between the two GaAsFETs, that instead of the SMD capacitor, is done with two coupled lines. This coupler is equivalent to a capacitor when the two lines resonate as 1/4 lambda, so the trick is to make them resonating at about 5GHz. Without any calculation, I simply took the 1/2 lambda resonators length of the 11GHz image filter, that means 1/4 lambda at half the frequency. Then I've easily engraved the coupler pattern on the existing transmission line using a cutter, as shown in the picture.  This improved multiplier can be driven by just -2dBm signal at  2484MHz, so only one MMIC amplifier stage is enough.

92MHz Oscillator
Working at relatively low frequency, this section in not particularly critical, and could be as simple as a one-transistor crystal oscillator. But for best performance a more complex design is mandatory. The key parameters, as widely described in literature, are the low phase-noise, the frequency stability and the frequency accuracy. To get the first one I simply took the project of a 5th overtone oscillator+buffer  by I2SG, who has optimized it for phase-noise performance.  Then I have added the varicap frequency adjust and a prescaler buffer to phase-lock it to an external 10MHz reference, like a good quality OCXO or my GPS-based Clock Generator. This last  would ensure a very high absolute frequency accuracy (around ±100Hz at 10GHz), but to keep a reasonable degree of accuracy when no external source is available, I put a small TCXO that keeps my transverter within ±10KHz with a very compact design. The TCXO insertion happens automatically when no external source signal is detected. 
I've seen in practice the advantages of having a fairly good accuracy without relying on auxiliary equipments during quick lab tests or for line-of sight contacts,  but with the possibility to stay within 10ppb when  challenged by a long-range contact or for beacon monitoring/hunting. 

The Oscillator and the PLL have been realized on experimental pre-drilled small boards, mounted inside a tinned shielding box. An internal shield keeps the oscillator  separated from the PLL. The big 92MHz crystal is on the backside of the oscillator board. Similarly, the TCXO sub-module, that has been added later, has found his place at the backside of the PLL board. Room is quite tight, but I like this compact arrangement that still allows a good access to all internal components and adjustments. The TCXO is now replaced by an OCXO following an upgrade work.

PA Unit and Antenna Relay
The PA unit is built from a part of another LNB. Since only one gain stage is needed, the simplest LNB is the best. The modification  could have been easy,  but the 20dBm GaAsFET I had was a flange-mount type and I also wanted to add an RF detector.  So I carefully cut a rectangular slot in the PCB exactly the size of the GaAsFET flange (a little, sharp cutter does the job), and now the GaAsFET is tightened to the aluminum plate with two small screws. The RF coupler is done with a small strip cut from a copper foil, glued to the PCB close to the output stripline. At one end of the added strip there is the 50ohm termination, while at the other end there is a microwave schottky diode taken from the unused mixer of the LNB. During RX, the GaAsFET is forced to cut-off, again by playing with the Vbb of the PNP bias transistor. 
The GaAsFET matching has been optimized seeking the maximum (stable) power by cut and trial with small stubs cut from a copper foil. This technique is normally used when tuning microwave amplifiers, but needs patience and extreme ESD precautions (here GaAsFETs start to be highly priced...) . At the end I got the 20dBm with 9dB gain: very close to the transistor specifications. Here is the picture of the  PA unit PCB (without shielding cover):

A new 2-stages PA has replaced this little boy, and currently output power of the transverter is a respectable 1WPEP. A picture of  the new PA here.

AUX Unit 
This last block includes a few circuits that are not part of  RF processing:
- Antenna relay driver
- Generation of -5V supply
- Remote control and monitoring
- Supply Voltage check
The antenna relay driver gives a short current pulse to each relay coil by discharging the energy stored in a 220uF capacitor, that was previously charged at 17V (nominal) through a 1K current limiting resistor . The -5V generator is not strictly necessary, but I've put it to have the negative supply ready  when I (hopefully) will add a more powerful PA. In the meanwhile, having the -5V rail put in series with the Transverter supply, I can reach the specified 15V relay coil drive also when the supply voltage drops from the nominal 12V down to 10V. This gives a good margin for portable operation with a Pb-gel 12V battery. And for a pushbutton check of the battery voltage, a simple indicator circuit drives the same microammeter used for RF TX monitoring. 
Not strictly necessary is also the DC amplifier block, that I have included to remotely monitor the RF TX power when the Transverter is pole-mounted in a fixed station. For a complete operation  monitoring, also the LEDs drive voltages are made available at the Supply/Remote connector.  

 The AUX unit is also a kind of hub for supply rails and for the connections between the various other units. I built it on a pre-drilled experimental board, and is mounted inside the Transverter without any shielding box.

ASSEMBLY

This form-factor is different from the more common "desktop style", and in fact you cannot stack anything on it, but I found it more suitable to be mounted close to the antenna. A foldable collar on the back allows to hang the transverter to the vertical mast that holds the antenna. Alternatively, with a U-shaped bracket, it can be steadily attached to the back of a dish, and the two front handles that protect the controls and the connectors can serve to steer the microwave beam... 

The enclosure I found perfect for the purpose is a 1590D diecast aluminum box by Hammond, that I have used "reversed". All modules, switches, connectors LEDs and the microammeter are mounted on the box cover that actually is the front panel, while the box has become in turn the cover. So, when the enclosure is opened, everything comes out in one piece with the panel: you have 5 sides open, therefore all modules and internal connectors are easily accessible while the whole Transverter is still operational. 

These are 5 views on the inside of the Transverter. You can identify each block of the diagram.

And the next two photos have been taken with the MIX Unit removed, to show the inner view of controls and connectors. 

UPGRADES

No project is perfect, and homebrewing gives a lot of room for upgrades. Three main parts of this transverter have been addressed for improvements:
- TX power
- RX conversion gain
- Replace the TCXO with an high stability OCXO

The original 100mW PA is a single stage with 9dB gain, so to reach a significant increase of power it has been replaced by a dual stage PA. This time, the LNB reuse was abandoned, and a specific teflon PCB from an italian hamfriend was adopted. Luckily, a very small dual stage PCB was available, so it has been possible to replace the old 100mW PA keeping the same mechanical arrangement. The new PA delivers 1W (30dBm) with the same driving power from the Mix Unit (11dBm). Its full project is described in another page of this website, so here we just show how it fits inside the transverter.

Because of the architecture chosen in the TX/RX PIN switch section of the Mix Unit, the 20dB gain of the IF buffer just compensates the attenuation of the 20dB patch from/to the IF transceiver. As a result, overall conversion gain in RX is just 12dB, that could be low if the transverter is mounted on the rooftop with a long IF cable. To get a little improvement, also on the NF side, the GaAsFETs of the 2-stages RX front-end have been substituted by an higher gain pair. Original transistors of the AE2 Cambridge LNB were FHX06LG (10.5dB gain, 1.1dB NF), now replaced by two NE32584 (13dB gain, 0.4dB NF). Measured conversion gain shows a 4dB improvement given by the new transistors: not a big jump, but not bad without making radical changes in the architecture, and with a theoretical improvement in the NF. 

Last upgrade was the replacement of the TCXO with an high stability Ovenized Oscillator. In the starting concept of this transverter, mostly driven by a compact design, the only reference frequency  source should be external, like a GPS Disciplined Oscillator. Then, as a back-up, a small TCXO was placed inside the 92MHz PLL oscillator to give a decent accuracy without relying on an external source. Having recently bought on e-Bay some  nice small-sized OCXO by Piezo Technik,  I found a little room inside the transverter to put one of them and finally get rid of the GPS DO unit when operating outdoor. Furthermore, after warm-up, current drain of the OCXO is much less than the GPS DO,  leaving longer operation time with batteries.  Of course, the OCXO is routinely calibrated in my lab with the GPS...

All test results of next section reflect the upgrades just described.

TESTS AND CONCLUSIONS

A thorough test has not been possible, because not all necessary instruments are in our hands. For example, RX Noise Figure should be measured now that new Low-noise transistors have replaced the original ones. The few tables below give the main characteristics of the transverter, plus the breakdown of the contributions to overall signal gains or levels. 

Another transverter, based on the same concepts and similar architecture, has been built by Andrea (IW9HJV). It's shown working with the IF transceiver and the GPS frequency reference. With these two transverters we have started our experimentation on the 3cm bands, establishing interesting contacts. This activity is described in our Ham Radio pages.  

  Back to top