Medium Power Amplifiers for 24Ghz

By John Hazell G8ACE

 






The Background.

This project came about from building the DB6NT Transverter and Amplifiers. These excellent designs form the basis of equipment used by many 24Ghz activists. The progression of construction from transverter through three stage low noise amplifier producing tens of milliwatts using the inexpensive NEC NE32584 GaAsFet device is straight forward. Skill with surface mount construction being the main requirement to produce the expected results. The DB6NT Power Amplifier design uses the Mitsubishi device MGF1303 and whilst it was possible eventually to achieve 100mw of output power many hours were spent by the author adjusting the foil matching tabs to achieve the full output. It seems from corresponding with locally active 24Ghz stations that I was not alone in finding it difficult to achieve the full output power from this amplifier, some others achieving 60mw from a pair of MGF1903 (Birkett short lead version of the MGF1303) devices.

As a constructor I wanted to experiment further for a little more power. Some Rogers Duriod PCB material type 5870 was acquired and using the Laser jet printing method to produce layouts some experiments began. The output circuit of the DB6NT amplifier was copied and a pair of NE325 devices which have a high IDSS were used as an experiment to see what might be achieved. The NE325 is a lower dissipation and lower voltage device than the MGF1303 the literature for the NE325 stating a maximum of 4 volts at the drain. Since the DB6NT circuits already constructed used a 5 volt supply and no loss of devices had been suffered, 5 volts was initially tried for the +ve supply voltage. Negative bias is needed to keep the drain current and dissipation in check. This is an advantage in so far that the recommendation is that devices should be biased at ½ IDSS for linear amplifier application. The results were immediately promising and the drain voltage was pushed beyond 5 and it was found by experiment that output power peaked at 5.2 volts and then fell away again beyond this figure. This seems to be fairly consistent across a number of devices. An output power of 100mw was obtained from this circuit driven directly from the 3 stage low noise preamplifier, the measured gain being 6db, the dissipation limits being exceeded somewhat for the devices but no GaAsFet smoke being visible unless the supply voltage was extended towards 7 volts. I had ideas to print multiple 3db coupler connected output stages for more power but I had been corresponding with Neil, G4BRK during these experiments and he sent me an article from Microwave Journal. This article then formed the basis for a new design phase, three prototype amplifiers producing 125mw, 250mw and 500mw output.
 
 

The 125mw Amplifier Design

The Microwave Journal article discusses the use of Wilkinson combiners for the distribution of power to a phased antenna array, Fig 1. I could see that this arrangement might lead to the possibility of using a number of GaAsFet devices in parallel to achieve more power from easily available devices at a realistic price.
 

Fig.1 

One of the difficulties at such a high frequency is achieving consistent component positioning within the divider network and therefore repeatable results, Fig2. This problem is addressed by using a ¾ wave network with ½ wave arms connecting to the terminating resistor, Fig3.
 

         Fig.2

          Fig.3

A penalty of this arrangement is that the operating bandwidth is reduced. An prototype PCB layout was drawn using a PC Paint Package and the layout printed onto the PCB substrate, again using the Laser jet printing method, Fig4
 

     Fig.4 

The PCB track dimensions for the combiner and dc feed lines are dictated by the wavelength and reference to the parameters for the substrate. The answer to the dimension for the lines connecting to the devices was found in a Siemens application note, No. 022.    A ½ wave line is recommended with move able matching tab. The resultant layout provides good physical separation of the devices and is open and easy to work on.

For simplicity it was decided to arrange for the bias voltage to be applied equally to both devices and for the drain current to be applied without individual current monitoring. The bias circuitry was also not included on the PCB simply because it could easily be included as part of the PSU board.

Resistor positions are included to stop the gates floating when the board is being handled and for the drain voltage adjustment which provides some supply lead rf isolation.

It was decided that a milled box using integral wave guide connections below the PCB would afford the best results to properly access the performance. The input and output from the board is via 0.085" rigid coaxial probes into the wave guide. This arrangement, although not the best due to the discontinuity in the transmission line enables the board to be easily removed from the box.

Construction

The construction is along the same lines as for the DB6NT PA Amplifier. Constructional experience is necessary but since preceding amplifiers are needed to provide enough RF drive for this amplifier it is assumed that this will have given experience in the techniques needed. So far the amplifier has been constructed with wave guide terminations only which eliminates the need for dc isolating capacitors in the input and output rf connections.

The board is first drilled in two positions for the coax probes and then between the GaAsfet positions for the M2 screw fixing to aid heat sinking. These dimensions are then marked through to the box housing. Next, slots in four positions are made for the GaAsFet source lead earthing foils. These foils are 2 mm wide, 0.1 mm. thick copper. These are also used to provide as much thermal conduction through to the back plane of the board as possible. It is vital that the thermal gradient across the foils is as low as possible to ensure in operation the source leads are held close to ambient temperature. The foils are soldered on both sides of the board. Solder wick was used to mop up surplus solder on the top side of the board but the solder was puddled on the ground plane side right around all the foils and around the fixing hole to afford the best flat heat transfer surface area to contact with the box.

The wave guide probes are the most difficult part of the board assembly. The wall thickness from the box to the integral guide was made between 1 and 2 mm thick in the milled box housing and therefore the probe outer coax must be cut to this length. This is achieved by rolling a short length of 0.085 rigid coax on a cutting board under a sharp knife. This will cut through the copper sheath far enough for it to be broken away allowing the inner PTFE to also be trimmed away. This probe construction is an important element in obtaining good output power and eliminates the need to fix the board into the box with silver epoxy, so that the PCB can be assembled and easily removed again if required. The ideal length of the copper sheath is a fraction longer than the hole thickness in the box. The trimming of the ground plane around the probe feed through to the PCB track also calls for care as the board is only 0.010" thick. Carefully use a small drill to remove the copper ground plane leaving a countersink hole which goes almost to the track foil on the other side. The diameter of the removed copper needs to match the inner of the coax as closely as possible. The coax probe inner is first soldered in place to the PCB track, this then holds the probe whilst the outer copper sheath is soldered to the board. The excess inner can be cut away on the track side of the board and the assembly checked for short circuits between inner and outer of the probes. Finally any excess inner protruding above the PCB track can be trimmed down carefully with a small file and excess solder removed with solder wick. At this stage the board can be tested for assembly into the box. If the probe outer finishes flush with the wave guide inner surface all is well. This is unlikely however, so if it has not protruded enough either countersink the probe hole in the bottom of the box to allow the solder around the probe to seat better and or remove a little excess solder from around the probe. If the probe outer protrudes into the wave guide as long as it is not excessive it can be ignored at this stage. Remove the board and fit the surface mount resistors and chip capacitors. The GaAsFet are fitted before the board is placed in the box. The heat sinking will make the source leads very difficult to solder to the foils once the board is in the box if it is done well. If you are confident that you have your static discharges under control then both GaAsFets can be fitted. If not then connect dc supply leads to the board and fit the first of the two devices. Apply about 1.0 volts negative bias to the gate and then around 3 volts to the drain. Measure the drain current and change the gate voltage to prove you have an operational device. Note the current, switch off and fit the second device. Switch on again and note you have an increase in current proving the second device is good. The board can now be inserted into the box. Further work shows the power supply leads are best connected by running up through the box metal work, through the underside of the PCB to the track. This way any variation of wire position above the PCB surface in the rf field which may affect performance is eliminated. Feed the supply wires through their respective holes Apply heat sink compound to the soldered area around the foils on the board underside and push home into the box. Secure the board with an M2 screw between the devices. Turn the assembly over and cut the probe inner to 2.2 mm. Adjust the probe outer to be flush with the inner surface of the guide and using a cocktail stick with a fine point dribble silver loaded paint (car windscreen heater repair paint) between the outer of the probe and the hole through the guide. This forms the rf earthing between the amplifier board assembly and the wave guide. The board ends may well be flexed upward slightly within the box housing due to the probe final adjustment. The paint around the probes will be strong enough whilst the corners of the board are pushed down and secured with a blob of instant glue. Both the paint and the glue will break away if the board needs to be removed again. The small variable space between the underside of the PCB and the box seems to have no rf consequence. It is unlikely the board will be perfectly flat after construction anyway, but this does not seem to affect performance to any extent.
 
 

Alignment and Performance

Check that the PSU board is providing 1.0 volts for the gate supply before connecting the +ve supply to the drain dc connection. A bench PSU is helpful for the positive supply as it allows the +ve drain voltage to be set low during initial tune up with gradual increases until 5.2v at the drain is reached. Use say 4 volts initially. Do not exceed a supply voltage of 6 volts.

Connect a load and output monitor. Apply some rf drive, a three stage DB6NT low noise amplifier is a good source, if it is supplying 20mw or so. Typically at this initial stage only the same amount or maybe even less rf will come out from the amplifier.

Place a small tuning tab on the input line and adjust for more output. The tab should be the same width as the line ideally. A wooden cocktail stick is useful for moving the tab without having any rf effect itself. The starting position is opposite the dc feed line. An increase in output should be observed. Move to the output line and adjust the tuning tab again starting opposite the dc feed line. . Finally tabs can be added to the gates and then to the drains. These may make a small improvement otherwise leave them out. It is a good idea to tin the back of the tuning tabs prior to fitting to aid soldering down using the minimum of solder.

The supply voltage, drive, and matching tabs are all interactive so it is necessary to go around several times and readjust until maximum output is achieved. It is important to make final tuning adjustments at full output power as the device matching characteristics will almost certainly be changing at increased power levels . The lid will enhance the output further when adjusted to the optimum height feeding radiated power back in phase to the output. If the lid is fitted too close to the board eddy current losses will increase in the metal cover. Between 12mm and 15mm lid height gave good results on the 125mw prototype. The input and output line matching tabs appear to compensate well for the guide match. No need for tuning screws in the guide has been found if a matched load is available during alignment. If a matched load is not available then matching screws can be added to the guide for comfort, ensuring full power transfer can be achieved .

The GaAsFets are run beyond the manufactures recommended ratings leading to elevated channel temperature, hence the importance of the heat sinking. The prototype was finally set to the following dc conditions with no input signal. Drain current, 40-45mA per GaAsFet, set with the gate bias voltage at a drain voltage of 5.2 volts. Output levels of 100mw with 1db compression and 135mw saturated power were obtained. As saturation commences gate current will flow and the drain current will increase accordingly.

Gain was measured at 7db for power levels just below compression on the authors prototype. A Beta amplifier constructed by Arie, PA0EZ produced performance shown in the graph below. Thanks to Arie for permission to reproduce his results graph.

Power Supply

It is not intended to deal with the PSU construction in detail since this will be well within the capability of the 24Ghz constructor. Some suggestions however. Since the high dc input power is only required in transmit mode a suitable PSU can be arranged where the receive mode applied voltage is reduced with the Tx/Rx logic. This will reduce the dissipation within the GaAsFets to a more acceptable level. A power supply due to F1GHB can be easily modified for this, The basic circuit is shown in Fig5. The negative bias for the gates, circuit not shown here, should be derived from a low impedance source such that the voltage is not modified substantially by the flow of gate current. This is particularly important for the higher power amplifiers.
 
 

Fig.5

A simple +ve supply using a 7806 regulator is in use by the author. The regulator is housed separately so that its dissipation is not conducted to the GaAsfets.
 
 

   Fig 6.   Interior View of the 125mw amplifier

Since this picture was taken the resistors in the Wilkinson divider network have been changed for 0603 types. The supply leads are also routed away by going down below the board via through holes in the box.

The 250mw Amplifier

Once the basic module described above was proven, the artwork was copied and pasted using Paintshop Pro to produce an amplifier using four GaAsfets. The Wilkinson divider was also copied and pasted to split and combine the pairs of GaAsFets. Curved transmission feed lines were chosen to avoid any right angle bends which might cause additional radiation loss. The artwork is shown in Figure 7.

Fig 7.

Fig 8.

The construction was again in a milled box with the wave guide in the reverse side. Figure 8 shows the completed amplifier after tuning. The tuning tabs were applied to the input and output lines first and then to the gate and drain lines as with the two GaAsFet amplifier. It is necessary to go around several times to optimize for maximum output. Some signs of low level oscillations where initially encountered and rf absorbing foam can be seen in Figure 8 across the centre of the PCB which cured this problem. However once the 0805 resistors in the Wilkinson combiners were changed for 0603 types this problem disappeared and the foam was removed. The lid height was optimized for maximum output power with 250mw saturated output power being achieved. A beta amplifier has been produced by Eric, F1GHB and his finished amplifier is shown in Figure 9.
 
 

Fig. 9

The 500mw Amplifier

With good results being obtained from four GaAsFets the PCB artwork for eight devices was created again by copying and pasting the layout using Paintshop Pro. Additional input and output dividers were added with associated connecting transmission lines to complete the circuit. The PCB layout is shown in Figure 10.
 
 

Fig.10

Fig.11

The completed amplifier is shown in Figure 11. The resistors shown in the dividers in this picture are the correct 0603 types. Matching tabs can be seen on the input and output lines. The tabs on the gates and drains could not be adjusted properly on this amplifier due to the effects of the hand over the PCB. A few tenths of a db were obtained with tabs on the feed lines to each group of four GaAsFets, both on the input and output. Tabs are shown only on the output side in the picture. The optimum lid height above the PCB was also lower on this amplifier than on the lower power versions. This amplifier gave a saturated output of 520mw, measured at the Martlesham Microwave Round Table in November 1998.

I would like particularly like to record my thanks for help and ideas to the following 24Ghz Golfists. Bob, G3GNR . Neil , G4BRK. & Chris, G8BKE.

References:

Microwave Newsletter May 1998 Issue, Making Microwave PCBs.

Microwave Journal November 1995 Issue, Modified Wilkinson power Dividers, Pages 98-104

Siemens Application Note No.022

NEC Data Sheet for NE32584C

Rogers Duroid Web site