SS Amplifier IMD Testing (New Start rolled from Command Technologies Thread)

[KE5JPP]

Gentlemen (and Allison KB1GMX if she's around), we've talked enough about IMD so how about if we show the manufacturers a good example.

The Challenge
=========
Build an amplifier chain with 0dBm input and 50dBm output. Any technology can be used but the amplifier chain must cover 1.8 MHz to 30 MHz and be easily constructed using standard parts.

The Test
======
A two tone signal at 14.200 and 14.210 MHz fed into the amp, lowest IMD wins. No prize, bragging rights only.

We have a lot of expertise on these forums and it's about time we put it to good use. Who's interested?

Tanakasan

This is a good idea, but I doubt it will get too many takers.  There are way too many armchair engineers out there that like to pontificate about how it should be done and criticize everyone, but ACTIONS speak louder than WORDS.  Instead of trashing the manufacturers and engineers (which involves WORDS), let's see how many of these experts (that are so fast to criticize designs) come up with a real world reproducible example design (ACTIONS) that meets the criteria they are so quick to criticize others for.  Oh, and let's see you do it using in-production parts - no out-of-production, impossible to source parts.

Gene

[W8JI]

##  what's wrong with 200 hz or similar narrow tone spacing??   For now, we will assume the VL-1000 [rated for 1 kw pep out]  will have no dynamic reg problems/bias  when run at 500/600/700/800 w pep out on ssb.

Everything is wrong with 200Hz spacing in a two-tone test, unless you are looking for a power supply or bias issue that shows at 200Hz loading.

With wider tone spacing, the analyzer can be swept faster and skirt selectivity is much less of an issue.

The better question is, why would anyone want to use 200 Hz test spacing on a transmitter? What would be gained for the added analyzer issues?

Also, a notched noise test is often not good on a transmitter. It will totally hide dynamic regulation problems, because supply loading is constant. A much better test is a three-tone if we want to include supply issues. A low frequency tone can be used to modulate the normal two-tone to test dynamics.

73 Tom

[KB1GMX]

Tanakasnan>

The Challenge
=========
Build an amplifier chain with 0dBm input and 50dBm output. Any technology can be used but the amplifier chain must cover 1.8 MHz to 30 MHz and be easily constructed using standard parts.
<

If you are buying I'm building.   First I want to see how clean that 0dbm source is.

I've used both bipolar and LDMOS FETs for power amps and driver chains and it's all about
heat and power.  You have to burn a bit of  power and get rid of the heat. Most all the
homebrews I have sound good for a reason, that reason.  that also doesn't mean being
stupid about it. the other is design for 3 to 6DB more output than will be used.  Half the
signals (even some that sound good) I hear are wide because they are running right
up to the 1db compression point and that can dirty an amp up regardless of the device
used.  It also means the power source has to be able to support the load!!!

FYI: find an old AIL wideband amp,  10mhz to 400mhz, flat gain of 50DB, 50W
(100w saturated) class A, only 80 pounds and oh about 20+ years old.  Did I
mention solid state and bullet proof?

Seriously it can be done and is being done.  You have to be willing to pay for it and accept that the
unit will be large.  Look at the AR313 or AR305 design as the staring point, we are talking 300W.
Those Grandberg/MOtorola designs are very clean and with care can put serious gain and clean power.
They are also antiques or more than 20  years old.  So the arguement is not how or with what
but cost.

Every stage leading to that has to be clean and often that's where people screw up.

NOTE: while the K3 has a poor IMD we do NOT know if that due to the power stages or earlier
in the system.

Some observations.  All the better and cleaner solid state linear amplifiers run on 28, 36 and 50V.
There a reason for that, it's I^R and also heat.  as you go up the impedences become more
manageable even for LDMOS parts.  Amplifiers used for analog TV and DIgital YV are very linear
and none I know of work at less than 28V even for the 25 or 50W drivers.

Right now I'm completing two amps, both take less than 4W.  One is a pair of BLF177s push pull
and the output match is optimized for 20 through 6M (my span of interest for it ).  The base
Phillips/NXP design was for 1.6 to 28 at 300W, 22db gain (300w watts with less that 2W drive).
Worst case IMD (3rd is -33 below carrier at 30W and better at 300, 5th is -36 at 30W) I think
a bit more bias current should clean the IMD hump up some as it is not uniformly that level
and at some frequencies it's closer to -45!  This is a simple amp.  My 2M is a variation on the
granberg/PK0V/NXP designs and runs a BLF278 for the same power I expect similar numbers.


My change allows operation though the 6M band with the BLF177s as the ferrite used in the apnote.
is very lossy at VHF and lower loss ferrites limit the lower HF usability and efficiency.  It only takes
10 or 15 of those 300W to make the ferrite get very (80C) hot if its wrong.

So the NXP/Phillips amp plus a driver that is very clean for the needed 4W (2w + 6db headroom)
is a trivial task and there are many good sub 12W  LDMOS devices that run at 28V with 20+ db gain
over that range and the apnotes for them.  Add a driver for that and run them with heavy feedback
for a total of say 30db of clean gain.

Theres your amp.  An easy 47db(or more) of gain and with 0Dbm in that should yield more than
100W of very clean power (its capable of 300).

Of course TR switching, power supply, supervisory (Over current, high SWR, and high input power
protections), cute metering and low pass filters that can stand power (and deliver suitable 2 and 3rd harmonic suppression) and their switching.

Are there tricks. well yes and no.  This is not a slam it together on perfboard thing.  It requires a
well laid out board and more than trivial mechanical work to lay it out and build it.  there are odd
material used like UT141-25 (25 ohm copper jacket coax) and parts like ATC 100B caps, ELENCO
metal cad mica caps, oddball and large ferrites of the correct types.  The mechanical magic is a
slab of .25" thick copper 4"x6" to move the heat to the aluminum heatsink and who knows how
many hours of polishing the copper and the matching surface of the heatsink for flatness. Add
a few days of taping holes for screws. You don't do this and run full power and it goes pop,
don't cry to me.

So you issue a challenge to engineers that need skills of a solid sate power designer, a healthy dose
of mechanical skills, and knowledge of mechanical fabrication techniques.  And the heart of a lion
as your spending $100 for the BLF177s and maybe $100 more for a set of spares in case an Oops
happens. Sure and you want this cheap in kit form that's foolproof.

There's little rocket science in doing that just hard work.

But:
The the user turns up the compression and clipping, replaces the SSB filter with a wider
essb filter, adds base boost treble boost, and runs power till the average reading meter says
"yep the amps doing what you paid for".  With all that the LO phase noise and the noise
from 50DB(or more) of amplifiers doing their work products are at the level that people
for 10 miles around can hear them across the entire band before you start to call CQ. 
We blame the amp but the signal was already trash by time it got to the maybe -10dbm
post filter amp never mind the power stages.

Remember GIGO, garbage in, garbage out.


Allison

[M0HCN]

The the user turns up the compression and clipping, replaces the SSB filter with a wider
essb filter, adds base boost treble boost, and runs power till the average reading meter says
"yep the amps doing what you paid for".  With all that the LO phase noise and the noise
from 50DB(or more) of amplifiers doing their work products are at the level that people
for 10 miles around can hear them across the entire band before you start to call CQ. 
We blame the amp but the signal was already trash by time it got to the maybe -10dbm
post filter amp never mind the power stages.

Very true, you got to make a clean signal to start with or all the work on the PA is in vain.

Difficult to avoid needing at least 50dB post mixer gain, which is boring, but even level 23 mixers tend not be be particularly clean above 0dBm or so. I suspect the other biggie is getting the matching right (particularly the match out of the mixer), but I have seen some very strange looking matching networks out there.

There is only so much you can do about the screwdriver experts fiddling with the drive power trimmer or modulation bandwidth, try to design "attractive nuisance" trimmers like "Drive power" and "Deviation" out of your design (Digipots are sometimes handy here, or do it as a calibration directly in the DSP). In fact try to keep trim pots out generally, trim caps are not always avoidable, but pots can usually be designed out.

And yea, if you are prepared to pay for it (in capital cost, size and heat) superior IMD is **EASY**, getting nice numbers without using twice the amplifier sand needed and without masses of excess heat is the trick.

Also I hate the 160 - 6M expectation these days, makes magnetics selection more then a little tricky.

Incidentally a good source for small quantities of copper sheet in the UK would be appreciated if anyone has a suggestion, I have been thinking about those copper 'bullion' sellers on ebay and face milling top and bottom, thick copper sheet seems to be a hard problem unless you want square metres of the stuff.

73, Dan.

[W6RMK]

The real problem with this sort of thing is that it ignores things like production cost & efficiency (Anyone can build a clean 100W by burning 500W DC input in a class A stage).

Even without efficiency limits, it's a challenge.  20% efficiency in Class A with very good IMD over multiple octaves isn't easy.

[W6RMK]

Incidentally a good source for small quantities of copper sheet in the UK would be appreciated if anyone has a suggestion, I have been thinking about those copper 'bullion' sellers on ebay and face milling top and bottom, thick copper sheet seems to be a hard problem unless you want square metres of the stuff.

Oooh.. face milling or grinding or lapping copper.  that's a chore and not as easy as, say, aluminum or steel.  Talk to your machinist before you chose that approach.

What about cutoffs from a big billet or bar? 

How much thermal conductivity do you need?  Yeah copper is almost as good as silver but only about twice as good as aluminum.   Folks use copper. but I'm not sure it's actually needed, it might just be convenient because it's solderable, it's cheaper than gold or silver, and it's soft enough that you can squish something harder down on it and make good thermal contact.
Is the "spreading" the important thing (would a thin copper layer over an aluminum substrate work?  Yes, there's all kinds of CTE mismatch issues to deal with but maybe they'd be easier.  You can use explosive bonding or diffusion bonding for instance. )

or what about using a thermal transfer medium/gasket (grafoil, Chotherm, NuSil). (I'm not wild about those, but they might get you where you need to go.)

I think, though, that this discussion points up that designing a high power  low IMD amp is more a mechanical and thermal engineering problem than an electrical one.

[M0HCN]

Yea, I know all about bloody copper on the mill, do my own machining.

High power is always mainly about thermal and mechanical.

Is copper needed? Valid question actually, I suspect yes if running full transistor ratings into the smallest possible heatsink, but possibly not if running into a somewhat over sized heatsink. This really needs some FEM modelling to see for sure (Or an actual experiment).
Certainly I have an old Redifon Mel PA that does not use copper, but the heatsinks are **Massive** and it makes 250W from a pair of BLW96 (Nominal device rating 200W each). It might be that the copper thing is mainly about allowing the use of physically smaller heatsinks by allowing them to run hotter. 

Some of the 'phase change' heat transfer materials are possibly interesting, or maybe even something like a very thin indium foil (a VERY soft metal).

73 Dan.

[VE7RF]

The real problem with this sort of thing is that it ignores things like production cost & efficiency (Anyone can build a clean 100W by burning 500W DC input in a class A stage).

Even without efficiency limits, it's a challenge.  20% efficiency in Class A with very good IMD over multiple octaves isn't easy.

##  yaesu has already done it... Class A... and they get 25% eff... all bands 160-6m.  But it's a blast furnace on both my yaesu MK-V's.    In fact, it's even worse when the ptt is keyed....and you don't taki into the mic.  Then it's  30vdc @ 10A CCS..and zero watts out ! = 300w ccs diss !  At least when you drive it..it's then 300-75=225 watts diss.

##  IMO, none of this is rocket science.  You take a 300w pep out devices and run em at 100w pep out..then u get a clean -40db pep imd3 signal.  You design the amp around the 100w pep out level too...like they did with the yaesu 767GX years ago. [ which is -40db pep IMD3].   The 767GX used a pair of MRF-422's  with 24 vdc on em.   The combiners, LPF's, power supply, cooling, heat sinks etc, are all designed around the desired 100w pep output. The eff is high, since the combiner, etc, is designed around the  load Z you have, when running the 422's at the 100w level...with 24 vdc.


##  yaesu then comes out with it's infamous FT-1000D..which uses the same pair of MRF-422's...and then runs em at 200++ watts out..with 30 vdc.  Imd is not as good, BUT the eff is "high" [ 50%], since it was designed around the desired 200w level.  The 1000-D is a heavy brute, 56 lbs !  Now if I run the 1000-D at just 100w pep out, eff goes to hell..and down to 36%.  Drop the po down to just 50w...and eff gets even worse.  Same deal on my MK-V.

##  If you wanted a clean 1000w pep out, you would need  16 x MRF-150's to pull it off..and operate em with 50 vdc.   That would increase costs by a whole bunch though.  The heatsinks and power supply would NOT change though...just the number of devices.   We are not generating any more heat folks..we are just using more devices to get the same power we had before  with 8 x MRF-150's.

##  Bottom  line is... good IMD does NOT sell these days  with most hams.  IF an amp with 16 x 150's cost way more than the competition's amp, with just 8 x 150's... the cleaner amp with 16 x devices...and way more $$  would be outa business real soon.  Now if you could use 4 x 150's, all in one die [ which they make]..then imd would not be an issue.   That Finnish co is getting 1 kw pep out  of just one free scale device.  Now maybe if you took 2 or 4 x of the biggest freescale devices they make..and built a linear around it... you would get your low imd.

Jim  VE7RF

[VE7RF]

Yea, I know all about bloody copper on the mill, do my own machining.

High power is always mainly about thermal and mechanical.

Is copper needed? Valid question actually, I suspect yes if running full transistor ratings into the smallest possible heatsink, but possibly not if running into a somewhat over sized heatsink. This really needs some FEM modelling to see for sure (Or an actual experiment).
Certainly I have an old Redifon Mel PA that does not use copper, but the heatsinks are **Massive** and it makes 250W from a pair of BLW96 (Nominal device rating 200W each). It might be that the copper thing is mainly about allowing the use of physically smaller heatsinks by allowing them to run hotter. 

Some of the 'phase change' heat transfer materials are possibly interesting, or maybe even something like a very thin indium foil (a VERY soft metal).

73 Dan.

##  per Eimac's care and feeding book, Silver is 1.05   Copper is  1.00   Aluminum is just .57   Everything else is worse than AL.  After running  CU barstock and also AL barstock  into my 12" disc grinder, I can tell you for a fact that CU conducts heat WAY better than AL.  And both would burn my hands with no gloves on..cu being the hotter of the two.   Now stuff some steel barstock in there, and I don't feel any heat at all, none, zero.   Of course the real trick is... all this heat now into the heatsink now has to be transferred to the air in the end.  LOTS of fins, and really thin, and made from CU is the ultimate heatsink.

## Ever look inside a large Eimac  triode, like a 3CX-10,000A7, etc.  Loads of paper thin fins, and all are silver plated  CU...and not much thicker than a razor blade either.  I'm not talking about the struts at the top and bottom either.  Those things are thicker, and are used to hold the cylinder in place.  If u shine a light down inside..with the tube sitting on a sheet of white paper, you will see the real cooling fins.  They fold em back onto themselves, when they reach the outside edge too..to increase their surface area.

##  I always found it amusing that the typ heat sink for any RF amp consists of AL..and uses thick fins..and not many of them.  OK, now blow lotsa air through em.  Machine a real heatsink out of a solid block of CU..and you will see what I mean right away.   Then insulate the heatsink from the chassis with ceramic stand off's....then no need for joint compound nor thin mica insulator's.  I tried this  with zeners years ago for an experiment, and it worked good.  CU is 45 heavier than steel.  And steel is 2.87 times as heavy as AL.   CU heatsinks end up being triple the weight of an AL heatsink.   I silver plate all my own CU..so no big deal.

Later... Jim  VE7RF 

[W8JI]

Heatsinks are the way they are because thermal engineers looked at designs carefully over the years, and picked a variety of optimum materials and construction based on application.

It isn't because engineers at Wakefield and other thermal management companies are dumb.

If transistors could be mounted  in the center of the cooler, and spread evenly over a large area of the surface, they would look a lot like a tube cooler.  Since semiconductors present small areas of localized heat, and devices must be accessible for service, heatsinks for semiconductors don't look at all like tubes, or use the same materials.

73 Tom

[VE7RF]

Heatsinks are the way they are because thermal engineers looked at designs carefully over the years, and picked a variety of optimum materials and construction based on application.

It isn't because engineers at Wakefield and other thermal management companies are dumb.

If transistors could be mounted  in the center of the cooler, and spread evenly over a large area of the surface, they would look a lot like a tube cooler.  Since semiconductors present small areas of localized heat, and devices must be accessible for service, heatsinks for semiconductors don't look at all like tubes, or use the same materials.

73 Tom

##  all as I'm saying is.... why not make the  entire heat sink from Cu..instead of AL.  Then use more fins, and thinner fins.   You would extract a helluva lot more heat  vs AL...like abt 75% more.

later... Jim  VE7RF

[KB1GMX]

The reason for copper is simple. That and its cheaper than transistors. The high power devices
all their heat into a footprint of around 1 square inch (6.45cm^2).  Some less or more but not
by much.  the goal is to get that heat from the device with the lowest thermal resistance to
where you can get rid of it.  There is only one way to do that with hundreds of watts, conduction.
All that heat come from the die that is much smaller.  Also hotter directly impacts the MTBF
of that transistor/MOSFET.

The typical pallet amp and we are talking about over 100W size amps the copper is there to
make up for a rather small flange on the device and to spread the heat out to a larger and
better radiator usually aluminum.

the typical amp I've seen and built that means .2 to .25 inch (5 to 6.35mm) copper plate
and a aluminum heat sink that has a base thickness of not less than .375 inch (9.525mm)
not including the height of the fins but only the base they extend from.

The goal to keep the thermal resistance so that at full sustained power the die never hits
200C  junction temperature and melts.  That's not easy when you look at say a MRF150 with a
die to flange of .6 C/W and the flange to heat spreader might add .15 C/W and so on.  Rough
translation is for a perfect heattsink the die will heat .75 degrees C for every watt dissipated
in the device.  That alone limits the device to 300W.  When you add that you have to derate
the device 1.71W for every degree C  to stay in the safe area you'd better keep it cool or
stay under 100W power dissipated.  [thats why some designs use more at lower power
than might be achievable.] 



Some numbers: 
   100C is boiling water at sea level. 
   200C is 392F, at 361 solder (Sn63/37Pb) melts.


I'd add that to the user of a solid state amp running that blower used to keep the 3-500Z cool
is out of the question due to it's noise.  Seems rather odd that both might dissipate the same
heat for about the same power yet the poor transistors are on their own hoping for convective
cooling.  Does that seem right?  We need the blower to move air, with a large finned heat sink
and no back pressure we can do this quieter than trying ram air past the pins of a 8877 under
pressure.
 
The approaches to get clean power and meet the thermal challenges many but the most common is
to use push-pull class B or more commonly AB1 where the idle current is somewhere around 3 to
10% maximum.  For a 300W amp that is about 500mA per device at 50V for 8% (50W of heat
just idling!).  It  can go higher but there is a point where more current gets very little
improvement in IMD.

The class A single ended will need a lot more power and produce heat at a 1:1 ratio for the
power in.  The efficiency is never more than 40% and typical is less.  But we may use it to get
a few watts as a driver where the power budget is far lower.  Even then we may run 12W of power
just to get 2-3 clean watts.  Yet we see radios where the driver is not class A but likely class AB1.

Looking back the driver and finals are sitting there pumping 72W of heat and there no intelligence
going out yet. This is just a back of the envelope and we are at the .1-.2W input level.  That means
we need more gain most likely class A to get there.  Or differently put yet more watts of power
and heat for gain.   

Sounds awful, but that 3-500Z is running 50W of heat for the filaments.  Though it does look cool
if you can see it.



Allison

[ZS6BIM]

Question – have you ever seen broadband solid state amplifiers, or transceivers IMD tested into loads other than 50, j0 Ohms?

On the other hand how many amateurs operate their transceivers into 50, j0 loads?
 
An SWR of 1,5:1 represents an apparent reflected power of just 4%, hardly visible on the scale of the normal amateur power meter and yet this represents a 50% load variation on the output stage of the power amplifier.

Even when using the typical auto ATU an SWR considerably better than 1,5:1 is seldom achieved.

As many (most) solid state PA’s are operated (as determined by the ALC loop) very close to clipping, usually within 1dB, what happen to linearity and stability when the load changes by 50% or more at all phase angles?

With the need to demonstrate constant forward power into a mismatched load (a customer requirement) many manufacturers resort to only reducing power when the apparent reflected power exceeds some set level, usually resulting from a 1,5 or even 2:1 SWR load!

I guess they know it’s unlikely their equipment will ever be tested into such loads.

ALC loops should be designed to reduce the load power by 1/SWR to ensure linearity and stability is maintained when operating into a mismatched load.

If an amplifier can supply say 100W into a 2:1 miss-match then it would be able to supply 200W into a 50, j0 load with the same linearity; so why not then specify the amplifier as a 200W amp and reduce power by 1/SWR! 

73
Mike
ZS6BIM

[M0HCN]

If you fed back the load phase angle as well you could get smarter about it, 2:1 can represent an impedance magnitude of either 100 or 25 ohms after all, and on the low side it might be possible to drop the PA supply voltage to remain within the safe operating area while supplying more current to the low Z load.....

But yea, not running into the supply rail is slightly important!

The reason to shy away from rating that 100W PA at 200W and 1/SWR is probably mainly down to heatsink and DC supply limitations.

The modern trend for built in L network auto tuners probably does much to improve IMD simply because it improves the match in most cases.

Regards, Dan.

[G3RZP]

Dan,

Copper in small quantities in the UK.

Model engineering suppliers. I'm in Germany right now, so the names may not be 100%, but google should find them.

Folkestone Model engineering
College Metal supplies.
metal to metal (M2M?)
Milton Keynes metals

Get a copy of 'Engineering in Miniature' (TEE Publishing) and/or Model Engineer for metal suppliers.

73

Peter DL/G3RZP/P this week