Thermal Design

[James Ahlstrom]

Hello Group,

I personally find thermal design to be incredibly boring, but unfortunately it is often the most important part of amp design.  I usually just slap on a ridiculously large heat sink and ignore it.  We don't have that luxury here.

Manufacturers often engage in specmanship with power figures.  I have seen devices claim (say) 6 watts, only to find out that it applies to a 25 degree case temperature, and you don't get that without water cooling.  We need to look at the degrees per watt and do an actual design.

The SMD devices we are discussing depend on the circuit board for a heat sink, as the package itself dissipates little heat.  Claudio's amps Steve referenced earlier use a special thin circuit board to decrease via length, and have the PC board edges soldered to the box, and they still run hot.  This is a great design, and I love it, but I am just pointing out that we can't do that for HL.  If we use SMD power devices, the heat will follow the copper.  We don't want the heat on the main HL board, but where is it going to go?  Putting the finals near the edge and leaving the solder mask off at the rail might help a little.

I always assumed we would use TO-220 devices bolted to the case, but that makes sliding the boards in and out a problem.  Still, I think it is a good option.  Another thought is to use the 4012H-6N  62 mm high box in a two board design.  The bottom board is the HL main board.  The upper board has the power devices mounted on top with finned heat sinks.  This makes the top space an oven, but the heat is insulated from the lower main board by the poor conductivity of the FR-7 top board, and we have the whole area of the aluminum top to get rid of heat.

It would be nice to have 5 watts continuous output for digital modes.  But it is enough to have 5 watts PEP for SSB and CW, and a lower continuous rating, and that cuts our power budget considerably.

Can someone suggest some more thermal designs?  What are our power devices and where do we put them?

Jim

N2ADR

[Steve Haynal]

Hi Jim and Group,

This paper is very useful for this problem. I've thought a bit about heat dissipation strategies in the HL 2.0 that could accommodate this new transistor and John's new 10W push pull and will share later when I have time.

73,

Steve

KF7O

[in3otd]

Hello Jim,
I agree that there are no easy solutions for heat sinking those small SMD devices; one option could be having the (small) PA board with the bottom layer entirely GND, without even a signal via, so it could be fixed to a metal plate (top, bottom or even front/bottom plate of the case, maybe) with some thermal paste in between. But it won't be a good solution for a box were the top and bottom cannot be removed and the PCBs have to be slided in. Hammond (and probably others) have enclosures where the top can be slided out (like this) so one could fix the PCBs on the top plate, with standoffs if needed; still, inserting one or two PCBs and the top plate at the same time may not be so easy.
A time-proven solution, as you suggested, is to use TO-220 devices and fix them to one of the end plates; as discussed previously this should be adequate for 5 W of dissipation, at least, with the enclosure size we are targeting. One could even use a real heat sink as end plate, for higher powers, as proposed here, but making the openings for all the connectors may be difficult.

73 de Claudio, IN3OTD / DK1CG

[John Williams]

I think this is the way to go. To keep the heat somewhat away using the
board itself, kind of like how they built attics in houses. Heat
radiated up and out. I have my current HL in a similar config.

[Glenn P]

If the TO-220 is used, depending on height available in the chosen enclosure, the slide in could be a piece of 1/16" aluminium sheet, (instead of the PCB itself)  to which the H-L boards are then mounted.  This way also, the whole assembly can be slid out of the box to do work on it and the heatsink is retained and is also fairly large in area. Conversly, H-L can be built up on this al. plate, tested etc, then simply slid into the enclosure.

Picture is of John's Pre V2 PA/LPF and Rx BPF on top.  {spacer on bottom to aluminium plate is 6mm high (1/4") The TO-220 sits under the PCB. Overall height of this assembly is about 40mm}.  >Note:- the spacers between the boards have been reduced to 19mm (3/4") since picture was taken.

glenn
vk3pe
.

[James Ahlstrom]

Hello Claudio and Group,

After some more thought, I think we should pursue TO-220 devices.  Putting the heat on the board and then taking it off is just too complicated.  The RD06HHF1 is the obvious choice, but expensive.  Alternatives are the IRF510, STF2N80K5, STF5N60M2 and FQPF3N25, all in TO-220F.  The TO-220F package has an insulated tab, and can be bolted directly to the case.  The thermal resistance is 4 to 6 degrees per watt, well within our heat budget.  I no longer think the TO-220 will be a problem mechanically.  We can mount it directly at the OPA2677 on the under side of the main board with holes to access the screws.  To mount it, we slide in the board and mark the case at the screw holes.  Then we drill the case, place the TO-220's under the board, slide the board in, screw down the TO-220's, and then solder the three leads.  To remove the board, we unscrew the TO-220's and slide the board out.  If it does not slide easily, we heat all three leads at once with a soldering iron and pull up on the middle lead which we have left a bit long.  Removing the board and then reinstalling it with heat sink compound is a bother, but not that difficult.  We would test the HL main board completely before installing the TO-220's.  All other amplifier parts including relays and filters will be on the top surface at the front, and so can be assembled and changed without removing the board.

I looked at a two board design, but two boards are more expensive than one, and raise the problems of interconnection and excessive distance between the OPA2677 and the sensitive MOSFET gates.  If we use Steve's 4004H-6N  10x15 cm box, and if the HL main board can fit into 10x10 cm, then we have 5X10 cm for the power amp, filters and relays.  That should be enough for a few bands.  Just like SoftRock, the user will build the bands she wants.  So there it is, a complete ready-to-go QRP rig on one board.  We need a BNC antenna jack at the back.  We need to route the RF power and DC amp power carefully.  Since we have four layers, we can use stripline instead of microstrip; or mini coax.  We can leave off a strip of the solder mask between the HL main board and filters to solder down a shield.

For more bands we need a second board and a taller box, such as the 4006H-6N or 4012H-6N.  The TO-220's must stay where they are, close to the OPA2677, so that the distance to the MOSFET gates is minimized.  We add a second 10x15 cm board in a slot above the first board.  To interconnect the two boards, we design the main board with a row of through-hole pads for IO along one edge.  These match up with an identical row on the second board, and are wired together with jumpers.  Connecting RF is more complicated, but we now have single ended 50 ohm RF, so we can use mini coax.  Again, all parts on the second board are on the top and can be assembled and changed without removing or separating the boards.  I think this should provide enough space for John's amp and filters.

The design is not ideal for those who want to use HL without the amp.  They are stuck with a 10x15 board instead of a 10x10 board.  But they have the extra 5 cm of space and a blank front panel that may be useful.  It also requires a more expensive second-tier PC board instead of the cheaper 10x10 board.  But at least it is just one board.  I am also assuming that the OPA2677 directly connected to MOSFET gates is a good design, but that has not been proven.

Jim

N2ADR

[Steve Haynal]

Hi Jim,

Just a few quick thoughts:

1. I think we should capture the many ideas for enclosures and thermal design on a wiki page that all can edit. Much has already been discussed in the past but it is hard to find on this list.
   
2. The tab on the IRF510 is the drain, not source for easy grounding/thermal dissipation. People have objected to the IRF510 in the past for this reason. I didn't check the other alternatives you list.
3. I'm not as concerned about board separation in a two board solution as you are. Consider the FLEX1500 pictures Alan Hopper posted to this list on May 10. The RF passes through a 5x2 header with long pins. I estimate the distance to be around 2cm. Consider how the Ultimate3 is stacked with filters. RF passes through total header length of 2-3 cm. For a balanced feedline, just a twisted pair of wires will have impedance in the low 100 Ohms, which looks just about in the right range going from your OPA2677 design to an RF board, assuming all or at least some termination resistance is on the PA board.
4. Engineers are using the AirFast transistor family in real designs, so how are they dissipating heat? I've read this paper and the basic idea for the least expensive solution is that thermal vias are enough to transfer the heat to the other side of the PCB, even for standard 1.6mm thick PCBs. Once the heat is on the other side, an exposed ground plane must abut to a thermal sink. If a small RF PA board were entirely surface mount (which wouldn't be to hard to do), the entire board could be bolted to the top or bottom of the enclosure to use the enclosure as the thermal sink. If the PA was on one main board, enough area on the bottom could be designated as flat and component free so that area can abut to a small aluminum plate which then abuts to the enclosure. See again the pictures from Alan and the picture from Glen. Another alternative is to leave a slot in the PCB for copper tape. One end of the short copper tape is soldered to the tab, the other goes through the slot and is bolted to the enclosure. I would provide short lengths of copper tape with PCB orders. I still need to go through the calculations (in the paper I linked to 2-3 days ago) to see if this solution is adequate.
5. I am not against a TO-220 device, and want a solution that allows for the use of TO-220 devices as well as others. the RD16HHF1 is a tried and true device, and John and company have spent much time building and testing with these devices. If it comes down to only being able to support once device, I will stick with the RD16HHF1 as that is the most proven. I do not like the price point or the fact that these are not available from Digi-Key or Mouser. 
6. I prefer the split enclosures over single body enclosures as you can take a top or half off to gain access to the board. People will need to do this to connect JTAG for initial and sometimes subsequent FPGA programming.
7. Enclosures with card slots have some attraction at first, but I wonder if we are better off not using card slots (or at most one slot) so we have freedom in how boards are spaced. 
8. In quantity 50 batches from Elecrow, 4 layer PCBS are $1.90 for max 5x5, $3.66 for max 10x10 and $4.83 for max 10x16 mm boards. The price difference between a 10x15 + 5x5 PA, or two 10x10 is not significant overall. Interconnect costs will be the bigger concern. I do not think I should try to fit everything including the PA on one board. I want some room to simplify initial layout and routing and allow for late changes. I think the remaining alternatives are a two board, each 10x10, with PA and Filters on one board, or a two board, 10x15 and 5x5, with PA on 5x5.   
9. A two board 10x10 solution would work well in this enclosure. Although not clear from the pictures, it is split body, not equal halves. Given one saves $8-$10 on the price compares to other enclosures, this more than makes up for additional cost of some interconnect. It is high enough that three board can fit, as there is interest in a separate RX BPF board. There is just a single card slot. The PA/filter board could use that and the other boards would stack on with standoffs. We'd control the spacing. The other boards would have to be slightly less than 10 cm wide. 
10. If people really believe minature coax is better for board to board RF connection, then something like this cable, cut in half with cut ends soldered directly to PA board and SMA connectors on the main HL2.0 board could be a cost-effective solution. These cables are even less in quantity. We'd have to give up with balanced feed from the op-amp to PA and go with 50 Ohm though. I've been debating the value of trying to use balanced feed. Obviously there are fewer baluns, which are hard to find and scary for many to wind on their own. The 50 Ohm is a standard though.

Okay, that was longer than I expected, but those are the ideas I've been thinking about recently.

73,

Steve

KF7O

[Steve Haynal]

Hi John,

I've always imagined the RF transistors thermally dissipating using the bottom of the enclosure. But the same think can be flipped over (bottom becomes top) for an "attic" effect.

73,

Steve

KF7O

[James Ahlstrom]

Hello Steve,

After thinking about your comments, I believe your design is a good one.  The Hermes-Lite 10x10 board without power amp satisfies those who want to use HL either barefoot, or as an instrument.  And given the pricing you quoted for board sizes, there is no compelling advantage to staying with a one-board design.  Many will want to experiment with amps and filters, and your design will accommodate that.

I looked at the 10x10 box you referenced, and it is 50 mm tall and good for a two-board design as you say.  I looked on DigiKey for a 10x10 box and didn't find one.  They mostly offer Hammond, and the 10x12 box is very suitable and only 34 mm tall.  I didn't check Mouser, but there are many boxes designed for a 100 mm circuit board, I guess because of the EuroCard standard.  So we will have many box choices.

As far as interconnect, it can't go on the end and still have the board fit in a 10x10 box, so it goes on the top, and 0.1 inch pin headers are one option.  These can take a ribbon cable, or plug into the mating pins of a board above the main board.  I think I remember that manufacturing boards with them adds cost, because they must be accurately perpendicular to the board.  If so, we can leave them off and let the builder solder them in their pads.

Jim

N2ADR

[ZL2APV]

I too like the 10x10 board. In earlier years we built softrock SR6.3's and mated them with a project called Mobo for filters, amplification and microprocessor control. The connection was with piggy back 0.1 inch headers and the usual method was to solder the plugs onto the SR6.3 RF board and then plug in the header sockets, drop the Mobo onto them and solder down giving a perfect mating fit. Most of us made extension cables so we could test and work on the two boards and gain access to the components. The extension cable in my case had RG174 coax on the RF paths and ribbon cable on dc paths. In this case the Hermes-Lite could be manufactured with headers in place and the mating boards could have hand soldered matching connectors.

In my personal case I will be forever experimenting with the radio and lots of flexibility with what I connect to the Hermes-Lite board is highly desirable. I would like to see headers on the HL board for power and control and 2 pin headers for the RF ports. I think that Steve already has something like this in mind and it also suits Jim.

73, Graeme zl2apv

[Steve Haynal]

Hi Jim and Graeme,

Thanks for all the input. I really am still on the fence regarding a two board solution (2 at 10x8 or 10x10 cm) versus a single board (1 at 10x15) solution. For a single board solution, I think it only make sense if it truly is single board and doesn't require a second board for the PA. I need to progress farther with the schematic revisions (which I am doing) and layout to make an informed decision.

73,

Steve

KF7O

[James Ahlstrom]

Hello Group,

This is a report on some tests of the TO-220 devices I mentioned in my previous post here June 29.  First the good news.  I tested a pair of RD06HHF1 in a simple AB push-pull amp modeled on the Penny Whistle amp from the HPSDR group.  There is no optimization of any kind.  The output transformer is 2 turns CT primary, 4 turns secondary, and there is no DC supply choke.  Bias is 70 ma each transistor.  There is no input transformer.  The HL op amp including the 24 ohm resistors is directly connected to the gates through 100n capacitors.  I set the Quisk Spot control to give 5 watts on 40 meters.  Then without changing anything, I measured the output on different bands and got this:

Freq MHz       1.9     3.8     7.2    14.2    21.3    29.6

Output dBm  37.0   37.1   37.0    36.9    35.7    34.6

The IMD was measured at 7.2 MHz, and was 33 dB below PEP.  The voltage at the gates for 7.2 MHz was 5.85 Vpp, which is below the maximum 7.5 Vpp available.  The 2.4 dB drop on 10 meters is unwelcome, but perhaps could be improved or compensated in software.  This amp is basically a toy, but I think that the results show that directly connecting the op amp to the gates should work, and is an option at 5 watts.

Although the RD06HHF1 works well, it costs $4.75 and is not available from Digikey or Mouser.  So I tried using some switching MOSFETS and a switching BJT in the amp because they are really cheap and readily available.  Because the tab is the drain/collector, I focused on TO-220F "full pack" devices which have a plastic tab, and need no insulation.  All of the switching devices failed except for the IRF510, and that worked poorly.  Now that I understand this better, I see why they failed.  But finding a suitable device is tedious because you must read each data sheet and the relevant information, such as Miller capacitance, is probably not provided.

Back in the old days, we had the MRF475 and life was good.  Then these disappeared, but we could use CB radio devices like the 2SC1969, but now these are no longer made.  We have lots of $1.00 switching devices, but where are the cheap linear RF devices for HF?  They must be around somewhere.

Jim

N2ADR

[ZL2APV]

That is a very useful report and I think a wiki with engineering reports like this should be set up as there is so much activity on the group that these relevant little goldmines get lost in the sheer volume of data. I would be reluctant to see op-amps used in a PA situation as they would be susceptible to high VSWR voltage spikes. Something like an autotuner can induce these as the reactance relays switch giving momentary open circuit. The Mitsubishi fets are as close to bullet proof as they can be reasonably made with today's technology and Jim's tests confirm that a cheap and reliable amplifier can be reasonably be made from them.

The output of Hermes-Lite is as Alan Hopper suggests already clean enough as to not require filters but a following amplifier would introduce spurs and would need cleaning up with a filter but a discussion on that is not for this thread. However the case used in the RD06HHF1 is going to be the simplest to mount on to a heat sink on a removable board particularly if it gravitated to a single board design. RD06HHF1's are under $5.00 each at retail outlets but there seems to be no advantage from aliexpress as their prices are similar. Other RF types are available and a typical supplier is here but of course there are others. For thermal design we need to remember that we need to heat sink the devices when they are in the case and we want to be able to clip on heat sinking for testing when the board is out of the case too.

73, Graeme zl2apv

[Steve Haynal]

Hi Jim,

That is very good to hear. I am working on the schematics this weekend and will appreciate your feedback on the PA given your recent experience. I also have an enclosure and thermal solution that I am warming up to (pun intended) which accommodates both the TO-220 devices and AirFast devices that I will share in a later post. 

The Mitsubishi devices are available on www.aliexpress.com. There are some big price variations, but the RD15HVF1 appears to have the lowest average price in the low $3.00 range. This device is used in the KX3 and the original Penny Whistle. Since they are for VHF, they may perform a bit better on 10M. What do you think?

73,

Steve

KF7O

[in3otd]

Hello Steve,
[snip]

The Mitsubishi devices are available on www.aliexpress.com. There are some big price variations, but the RD15HVF1 appears to have the lowest average price in the low $3.00 range. This device is used in the KX3 and the original Penny Whistle. Since they are for VHF, they may perform a bit better on 10M.

I have noticed that here the RD15HVF1 costs the more or less the same as the RD16HHF1; I have quickly compared their electrical specs and done some simulations in the past, and it seems that the former can give a couple of dB more gain at HF. I'll expect it should work fine well over the 10 meters band.

[James Ahlstrom]

Hi Steve,

I think the RD15HVF1 is well worth trying, but I am not sure it will work as well as the RD06HHF1.  Look at these specs.  I also included the RD16HHF1, but I am actually using the RD06HHF1.

                                            RD06HHF1             RD15HVF1           RD16HHF1

Max Vdss (breakdown)               50 volts                30 volts                   50 volts

Ciss (input capacitance)             25p                       57p                         55p

S21  at 50 MHz                       24.979                 23.274                    22.018

Gain at 30 MHz                         16 dB                                                   16 dB

Gain at 175 MHz                                                   14 dB

The graph of Vds versus Ids for the RD15HVF1 is not as nice as the RD06HHF1, and the breakdown is a lot less.

It is hard to know how these devices will perform without trying them.

Jim

N2ADR

[Steve Haynal]

Hi Jim,

Thanks for the data. I agree that it is best to try. I am shooting for a generic enough PA design in HL2 so that people can try various devices with a few other discrete component changes. Is there any data from the Penny Whistle or KX3 on how well the RD15HVF1 is working?

73,

Steve

KF7O

[James Ahlstrom]

Hello Steve,

On Sunday, July 10, 2016 at 1:11:15 AM UTC-4, Steve Haynal wrote:

Is there any data from the Penny Whistle or KX3 on how well the RD15HVF1 is working?

I am using the Penny Whistle as a driver for my 100 watt amp.  It is excellent quality and it works fine, but I remember being  disappointed by the IMD spec, and it uses 0.9 amps bias for an amp with a draw of a bit less than 3 amps at max output.  So it dissipates 12 watts quiescent.  This is not a problem in my application.  But for HL it shows the need for a switch to turn off the finals when not transmitting.  This is easy with a transistor, or a small voltage regulator with an enable pin.  But it also means that HL will need sequencing for the amp, TR relay and output in order to be a real rig.

I could measure the Penny Whistle, but that would not be meaningful as it uses a 1:9 Z output transformer instead of my 1:4 Z.

When driving the MOSFET gates directly with the op amp, I noticed that the op amp runs hot.  Not necessarily hotter than when driving 50 ohms, but hot.  Since Misubishi gives the S-parameters, it would be nice to calculate the input gate impedance, and thus the load on the op amp.  Unfortunately, I don't know how to do that; can anyone help?.  Also, we really need a serious effort to heat sink the metal pad on the bottom of the op amp.  My footprint with the thermals on the pad is not adequate.

Jim

N2ADR

[John Williams]

The bias regulator on my PA board is only enabled during TX using the Ptt signal.

--
You received this message because you are subscribed to the Google Groups "Hermes-Lite" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hermes-lite...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Steve Haynal]

Hi John and Jim,

For the PA and associated LDMOS/MOSFET device parameters, I eventually plan to simulate with QUCS. I've used this in the past and it is quite nice. There is an article describing how to use QUCS for basic simulation, including S-parameters, on page 54 of this German article.  

The shutdown on the LP2951 John is using is very nice. I plan to use a separate signal for this, so that the biasing of the PA can be turned on/off in a proper sequence, not in lock step with /PTT. However, I plan to use the LP2985. The cost is the same, the footprint is smaller and there are no required resistors to set an adjustable voltage. I am also considering a digital potentiometer for the biasing. This is actually less expensive than the two manual ones John uses. I've seen this used successfully in PA bias circuits. It provides additional freedom and can eventually ease the use of a PA.

Regarding the heat from the OPA2677, I think we can improve the footprint a bit by moving the large hole for manual soldering to the side, and by including some copper area on the top and bottom layers right around the opamp for dissipation. Someone could also solder a small copper fin to this exposed copper area if desired.

73,

Steve

KF7O

On Sunday, July 10, 2016 at 7:50:57 AM UTC-7, John Williams wrote:

The bias regulator on my PA board is only enabled during TX using the Ptt signal.

[John Williams]

Digital bias is an excellent idea.

[Graeme Jury]

On my other rig the PA bias is controlled from a microprocessor driving an LM317 and I can set the PA bias current from the front panel monitored with an LCD display. The control is a rotary interrupter and is multifunction for other settings too with each function being set from a menu. To get this sort of functionality in Hermes-Lite is why I am considering incorporating and Arduino for control. Of course PA bias delay and smart filter switching etc. can be done via the Arduino also. Currently my PA bias delay is an RC time constant to a gate and was set by measuring my antenna changeover relay delay. Now that I am using a PIN diode TR switch the delay is hardly necessary. My external linear uses carrier detection and has its PA bias delay built in.

Back on topic, A small pin grid array with thermally conducting silicone rubber adhesive stuck on top seems to be helping with the present setup although better copper area on the PCB with as Steve suggests a small copper tab soldered on (maybe top and bottom) should improve on this.

73, Graeme zl2apv

You received this message because you are subscribed to a topic in the Google Groups "Hermes-Lite" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/hermes-lite/-7ZDEBz25ss/unsubscribe.
To unsubscribe from this group and all its topics, send an email to hermes-lite...@googlegroups.com.

[in3otd]

Hello,
I usually do all my RF simulations in Qucs, so I already have a few examples for the LDMOS amplifiers. I'll try to clean them up and put the Qucs projects on my website so they can be used a starting point for further simulations.
One common problem in using the devices S-parameters provided by the manufacturer is that very often these do not cover the HF bands, but only the VHF and above. The simulator can extrapolate the S-parameters but of course the results may not be good, if too far away from the original frequency range. For this reason I usually derive a (small-signal) model from the S-parameters that can be used at "any" frequency.

Qucs works well with frequency-domain simulations, less well for transient (time-domain) simulations and not so well for harmonic balance (large signals, frequency domain). QucsStudio, used in the article linked below, is a fork of Qucs which may have an improved simulator, but I do not use it as it is Windows-only.

BTW, there are only a few people currently working on the Qucs development; I have recently joined the development team to try helping a little, more help will be welcome. If someone is interested, besides the C++ coding/bug fixing for the GUI and simulator also checking/updating documentation, tidying up the project website, reporting bugs/issues for the various platforms, etc., all will be useful!

73 de Claudio, IN3OTD / DK1CG

[Steve Haynal]

Hi Claudio,

That would be great if you could post a few sample Qucs RF amplifier designs/simulations!

73,

Steve

KF7O

[in3otd]

Hello,
enclosed is a zipped Qucs project with a few files for simulating some basic HF PAs. All the schematics use small-signal equivalent circuits, so you cannot compute the compression point or harmonics or other non-linear effects, but they are useful to have an idea of the amplifier gain and input (and output) impedances. At higher power the small-signal approximations are of course no longer valid, measurements on the bench will tell you how the amplifier behaves there, hi.
Some of the schematics have also examples of parameter sweeping and/or optimization.
The project was done with Qucs 0.0.19 but it should work fine also on 0.0.18.

73 de Claudio, IN3OTD / DK1CG

[Steve Haynal]

Hi Claudio,

Thanks for the example files. I tried them out this evening, and am also ramping up on QUCS by working through a few simpler examples. Claudio's examples have small signal models for the RD16HFF1 RD06HFF1 and RD15HVF1 as well as several single device and push pull amplifiers using these models. It is worthwhile to take a look. What tool did you use to extract the small-signal model from the manufacturer's s-parameter data? I assume you can also use the s-parameter data directly in a simulation, provided you are happy with the Idq used to generate the s-parameter data?

73,

Steve

KF7O

[in3otd]

Hello Steve,
to extract the small-signal models for the FETs I use a GNU Octave script that first tries to estimate the equivalent circuit values directly from the S-parameters and then optimizes the equivalent circuit to match the S-parameters. But I think that the optimizer is actually doing most of the work, hi.
As said above, the S-parameters can certainly be used directly (for a given Iddq) but often they do not cover the lowest part of the HF or not even the HF bands. The S-parameters files for the RD06HHF1 and RD16HHF1 start from 10 MHz, 50 MHz for the RD15HVF1 and 200 MHz or more for the AirFast devices.

73 de Claudio, IN3OTD /  DK1CG

On Wednesday, July 20, 2016 at 8:38:11 AM UTC+2, Steve Haynal wrote:
[snip

[Steve Haynal]

Hi Claudio,

Do you have a pointer to the Octave script, or is it something you wrote that you are willing to share?

73,

Steve

KF7O

[in3otd]

Hello Steve,
the Octave script I wrote is quite messy and, as said, not always providing good starting values to the optimizer, which then has to do most of the work. So I have just prepared a Qucs project which can do the MOS model optimization (enclosed), starting from manually entered initial values for the model parameters. The optimizer used in Qucs is quite good and not so sensitive to the initial choice of parameters, so I think you could leave the default values for most of the low/medium power device planned to be used for the H-L PA.
Try to take a look at the example provided and let me know in case of issues.

73 de Claudio, IN3OTD / DK1CG

[Steve Haynal]

Hi Claudio,

Thanks! I will try it out. I haven't done Analog or RF design since school and will use this to brush up. I learn best by running and playing with software and simulations.

73,

Steve

KF7O