Predistortion to reduce IMD, etc.

[AB2EZ]

The use of pre-distortion to reduce IMD and other non-linear artifacts... introduced in ham radio transmitters and amplifiers... is a (good) old idea whose time has come. Although I haven't used it yet with my equipment... I plan to purchase an Apache transmitter in the future.

There are a number of subtle assumptions that are important to the viability/effectiveness of the pre-distortion method.

One of these is that the system to which pre-distortion is being applied is "memory-less". That is, for pre-distortion to provide its expected benefits, the non-linear input-output relationship of the system right now must not depend upon inputs that were applied in the recent past. Therefore, I can measure the input-output behavior at time t, and pre-distort the input at future time t+u to linearize the input-output behavior of the system.

A question I have is whether the amplifiers we employ are sufficiently memory-less to enable pre-distortion to work well.

For example, power supply regulation is probably very critical. If one or more or the amplifier's power supply output voltages drops on a positive modulation peak... because the capacitors in the power supply have lost some charge, it will take some time (depending upon the power supply time constants) for the capacitors to charge back up.

Changes in the power supply voltages at time t will cause changes in the amplifier's input-output characteristics at times that are later than t (by a few time constants).

Stu

[K1JJ]

Predistortion is like a miracle cure to linear amplifiers.  (As opposed to the more wasteful brute force AB1 and class A  heat method)

Rob/ W1AEX can verify the worst case tested by one of the PD gurus:  A trashy CB solid state  amplifier with low 20's IMD turned into a mid 40's Mr. Clean by using PD.

Another guy did it with a tube amp, which some thought wud have problems due to phase shift and tuned circuits - but it worked FB.  

That said, I would still modify or build the linear amp to be as clean as possible BEFORE adding the PD. This includes the stiffest  HV supply possible. This may mean larger filter caps, a choke or even a larger HV transformer.  PS sag needs to be reduced to a minimum, as you say.

I like the idea that PD allows a linear to run up to near saturation and still be clean. THAT in itself is very desirable, since linear efficiency generally gets better as the drive power is increased.  There's nothing like hammering a well cooled amplifier and having it remain clean.

It will be interesting to see what you end up with the Apache SDR, PD and using your tube amps, Stu.


Some may wonder why I spend so much time optimizing for IMD and cleanliness....  I do it to reduce problems with band neighbours.   When running a big amplifier that splatters even a little, we are constantly looking to see who is nearby and who we may be affecting.  It's like going to a party of well-dressed people while we are wearing rags. We think about how bad we look.  But put on a nice suit and we can forget about how we look and just have fun.  Same on the band. With a clean amp that runs -45dB 3rd IMD, we can plop down anywhere (within reason) and not worry about taking out our neighbours. We can forget about ourselves and just have fun.  It's a worthwhile endeavour whether on ssb or AM.

T

[K1JJ]

BTW, while we're on the subject of making linear amplifiers cleaner, I have a quick story to relate:


Last year I was setting up a chain of AB1 / class A  linears for best IMD.  (The Dr. Love system)  The 8877 amp was having big trouble giving better than -30dB 3rd IMD. It should do -40 to -45dB when run at reduced power in AB1.  The other amps in the chain were doing much better. I tried everything I could think of for a month, but NG.

The problem finally solved:  The 8877 uses an indirectly heated cathode, thus requires a single choke from cathode to ground for RF isolation.  I was using one of these BIG  broadcast-type pi-wound chokes with a series resistance of about 20 ohms. BINGO!  That was the problem. In AB1 there is plate current swing and this 20 ohms was swinging the "fixed" cathode bias all over the place.  I wound a single layer choke using #20 wire onto a 3/4" PVC tube.  The resistance was less than 1/4 ohm.   Unbelievable. The IMD shot up to a clean and  easy -45dB 3rd order!  The cathode bias was now set entirely by a string of diodes and solid as a rock.  

So measure that RF choke to be sure it is under one ohm.  Some of these RF chokes can be several ohms and create this problem. If the amplifier is pure class A, then there is no problem.. in fact the resistance will add some negative feedback. But if the plate current swings around like in class B or AB, forget it.  We then need a solid drop as produced by a string of diodes or a zener.

Series resistors  (20 ohms, etc) are OK in the plate HV circuit  to minimize surges into the amplifier, but are not OK in the cathode circuit.

Easy way to test for this problem:  Connect a standard RF choke to the cathode of the tube, at the tube pin. Measure the DC variation with a VOM on the other side of the choke in reference to ground. Add a  .001 disc bypass cap from the RF choke to ground at the VOM end.   Under normal amp operation, from no RF drive to full drive, this DC bias voltage shud remain reasonably steady. Any departure from steady is added distortion.  Hopefully there is no RF getting into the VOM.


BTW, the "grid metering" resistor that goes from the negative lead of the HV power supply to the tube cathode should also be a small value. In some cases this circuit is in series and will create the same problem. I used a 1 ohm resistor there.   (This resistor is used to isolate grid current meter readings from plate current.)


The higher the gain of the tube in GG, the more important it is to have a steady cathode bias. Many of these tubes have an amplification factor of ~200, so just a few volts of bias change can ruin their performance.   5 ohms * 1A = 5 volts variation... yikes!

T

[AB2EZ]

Tom

You wrote: "...It's like going to a party of well-dressed people while we are wearing rags. We think about how bad we look.  But put on a nice suit and we can forget about how we look and just have fun.  Same on the band. With a clean amp that runs -45dB 3rd IMD, we can plop down anywhere (within reason) and not worry about taking out our neighbours. We can forget about ourselves and just have fun.  It's a worthwhile endeavour whether on ssb or AM...."


I think we may be watching too many episodes of "Numb3rs". (-:

Best regards
Stu

[KD6VXI]

What about your typical 5 to 20 ohm rfc from grid to ground a la sb220, l4b, tl922, etc.

IMD drops by REMOVING those stupid bias yoyos.

Make your supplies high,  tight and regulated,   then throw resistance between them and ground,  with varying loads.  

Gain goes up,  allowing most s to reach full Pout with reduced drive,  as well.

--Shane
KD6VXI

[Opcom]

Without costly "Dahl-style" plate iron or grossly oversizing ye olde crufty plate iron, how can good regulation be enforced on a plate supply?

Choke input is the only way mentioned in the books and that is 5 to 10% regulation.

Adding a low voltage high current series regulator in series with the negative side to control that 500V of HV swing? Something could be elegant and ugly all at once.

[WD5JKO]

So measure that RF choke to be sure it is under one ohm.  Some of these RF chokes can be several ohms and create this problem. If the amplifier is pure class A, then there is no problem.. in fact the resistance will add some negative feedback. But if the plate current swings around like in class B or AB, forget it.  We then need a solid drop as produced by a string of diodes or a zener.

I use a cathode side choke with my 8877 GG amplifier. It has 4 ohms resistance. Since I run this amplifier for AM only, I figured it isn't an issue because the average plate current stays constant with or without modulation. I have no data though to support that statement.

On the subject of pre-distortion, I heard someone talking on 20M about this concerning the Flex 6000 series, and a meeting he attended that was conducted by Flex radio at Dayton. Pre-distortion is on their roadmap, and so is finishing the remote networking stuff. A vote was taken, and the crowd overwhelming showed hands for remote/networking as the priority over pre-distortion. I myself wasn't there, but I did hear folks on 20M discussing this.

Jim
Wd5JKO

[K1JJ]

I use a cathode side choke with my 8877 GG amplifier. It has 4 ohms resistance. Since I run this amplifier for AM only, I figured it isn't an issue because the average plate current stays constant with or without modulation. I have no data though to support that statement.

Jim
Wd5JKO

This is a good point, Jim.  Though if you have big positive peaks that are negatively clipped and have some upward carrier shift it could show up.  You might check the bias DC at the cathode with a scope and see what it looks like under heavy modulation just to be sure.

My linear system I described is for ssb only. On ssb the plate current variation is big, of course, so the low resistance is needed.  I use class C rigs for AM.

T

[K1JJ]

Without costly "Dahl-style" plate iron or grossly oversizing ye olde crufty plate iron, how can good regulation be enforced on a plate supply?

Choke input is the only way mentioned in the books and that is 5 to 10% regulation.

Adding a low voltage high current series regulator in series with the negative side to control that 500V of HV swing? Something could be elegant and ugly all at once.

I find the speech syllabic rate on ssb is fast enough so that I don't get much HV sag. The T/C seems slow enough on the filter caps.  (140 uF)  Though on AM, maybe sag won't be an issue as mentioned in the last two posts. The constant carrier certainly is a great bleeder resistor for regulation.  

The best way to find out is run some tone and voice program tests to see.  There are many things that can be tweaked and optimized that all add up in the end.

A high voltage regulator seems unnecessary, though negative side regulation would be an interesting project to look into. I'd like to hear more about that circuit.

T

[W3RSW]

I used a GM3SEK board for my HB amp retrofit. It has a pass transistor stabilized bias set point. Stays constant under varying loads and that was made a selling point.

http://www.ifwtech.co.uk/g3sek/

3 to 33 Vdc adjustable, precision regulation at up to 2.5 amps or more. Resistors included in kit for just about every tube or combination.

[W2VW]

Anything other than changes in amplifier parameters would require different pre-distortion. 

Have not worked with it in hobby radio so not sure how things are done there.

The equipment I've used can store multiple correction data which can be used as a set for a given parameter.

One exciter can be used on many transmitters by loading the proper correction files.

The more expensive exciters include a receiver which is used in determining corrections.

Inexpensive exciters rely on correction files obtained with the better exciter.

A lot different than George Burns' day.

[WBear2GCR]

Can anyone suggest some links with papers on this "predistortion" method(s)??

Interested to read up on it.

[W1ITT]

One place to find some info on PreDistortion is the Apache Labs site       https://apache-labs.com/index.php

Scroll down to the bottom of the page and look at the presentations on PureSignal software for the HPSDR and Apache line of platforms.  Nigel, G7CNF, has some good info. 

Also, if you peruse the HPSDR mailing list archives     http://lists.openhpsdr.org/pipermail/hpsdr-openhpsdr.org/  there is some good stuff there too.    Some of the more impressive demonstrations involve taking the crappy wideband solid state amplifiers  that are lucky to make -20 db IMD and cleaning them up to better than -40 db.   I have long been interested in the G2DAF amplifier design, where a tetrode is used, generating screen voltage by rectifying some of the drive power.  They have a reputation for being temperamental, with a few being fairly clean but most being pretty nasty.  It would be interesting to see what PureSignal would do with them. 
Basically, you are taking a small sample of the output of an amplifier by putting a directional coupler at the output, and feeding an attenuated signal back to a second receiver on the HPSDR transceiver board.  Wizardly software figures out, in real time, the corrections that need to be done to make the output squeaky clean.  It was predicted that memory effects in both solid state and tube amplifiers would be a limiting factor, but if -50 db IMD is good enough for you, it hasn't been much of a problem.  Considering  that much current amateur gear has a hard time making -30 db with a tailwind, that's a useful improvement, and a real breakthrough.  I wonder if the Class E amps will be amenable to this sort of treatment...

[w8khk]

I have long been interested in the G2DAF amplifier design, where a tetrode is used, generating screen voltage by rectifying some of the drive power.  They have a reputation for being temperamental, with a few being fairly clean but most being pretty nasty.  It would be interesting to see what PureSignal would do with them.  

That would be an interesting experiment.  I built a G2DAF amplifier using a pair of 4-400As in 1968 while I was in the Air Force at Davis Monthan AFB, Tucson, AZ.  Although I never measured the IMD, I always got good reports.  I used it until 1980, then my dad, W2DU, used it until about 2000.  I still have it; now that I have equipment to measure IMD, I can determine whether I should save it or tear it down.  I often thought of making a smaller, portable amplifier with either two or four 4CX250B tubes.  I have read often that they are not very clean in grid-driven tetrode configuration, but predistortion may be the answer.

[W1AEX]

A question I have is whether the amplifiers we employ are sufficiently memory-less to enable pre-distortion to work well.

Stu

Stu,

I recall a few threads discussing the issues Warren - NR0V faced when he worked on the Pre-Distortion module for OpenHPSDR and much of it was over my head. There were a lot of variables he had to deal with because of the expectation that end-users would be utilizing anything from broadband solid-state HF amps to tuned input and output tube amplifiers. Somehow he has made it all work, but he has mentioned that some conditions present an optimal correction and some limitations present improvement but with a few compromises. Gary - WA1OXT has probably experimented with the ANAN Pure Signal approach to pre-distortion more than anyone else I know and he has achieved impressive results (better than -50dB 3rd order IMD) with his pair of 8877's. Another guy who had done more than a little tinkering with pre-distortion is David - AC2IQ. He has approached it from the analog side and in fact was able to successfully implement pre-distortion with a Flex 5000A. If you scroll down through David's incredible QRZ page you can see some very interesting stuff he has done with pre-distortion with his own home-brew analog Quadrature modulated rig.

At any rate, back to the question in your original post. If you have not already done so, you might want to join and browse the Apache Labs user group at the link below:

https://groups.yahoo.com/neo/groups/apache-labs/info

In the last post of the thread below Warren mentions that memory effects pose an obstacle that may be difficult to eliminate. One issue that may appear in certain amplifiers is that the 3rd order IMD products appear to be less than the 5th order IMD products. He seems to suggest that maybe the symptom can be minimized by employing delay compensation. I have no idea what that means but the OpenHPSDR guys are apparently squinting to find ways around all the variables found in the amplifiers hams tend to put into service on the bands.

https://groups.yahoo.com/neo/groups/apache-labs/conversations/topics/10179

I have not installed a coupler at the output of either of my amplifiers and have only employed pre-distortion within my ANAN-100 exciter with its 100w PA. It works so well that I'm not really motivated to place a coupler behind the amps and am content with the results I get. For those who have not seen what pre-distortion looks like on an AM signal I did make a YouTube video of it at the link below. Note that the recovered audio in the video is not as clean as the real deal due to an inappropriate level setting in the Windows 7 mixer when I made the video/audio screen capture and also the AGC is pumping pretty badly in the little FiFi SDR I used for reception but the spectrum display shows the amazing effects of pre-distortion.

https://www.youtube.com/watch?v=o8h6ParzsAs

73,

Rob W1AEX

[Steve - K4HX]

Lots of links on the Web. Most any recent RF design text book will cover it in detail. It has been SOP in broadband RF telecom for a long time.

[WBear2GCR]

Any decent RF text?
Like I have any? ha ha... actually I don't.

Would kinda like to focus in on the essence quickly, rather than banging around Gargle search reading assorted marginally useful papers and sites. That's why I asked...

                     _-_-

[W1ITT]

Unfortunately for the casual student, but luckily for us in general, the progress in PreDistortion IMD improvement as regards amateur radio is moving so fast, even week by week, that a text would be out of date before the ink dried.  I understand that there is an article on the HPSDR latest news in QEX this month, but I haven't seen it.  The only thing to do is gird one's loins (and don't ask me for details on how that is actually accomplished) and dive into the HPSDR mailing list archives month by month for the past year or two.  Usually the topic headings will clue you in to the pertinent stuff.  Look for any posting by Warren, NR0V, the prime intellect behind the software and implementation, and follow that thread.  If you download the latest copy of PowerSDRMrx3.2.16 or so, down in the lower right hand corner from the Apache Labs site, you can run it in demo mode.  There is a "help" tab in the predistortion part that is good reading. 
I haven't yet heard that anyone has planned on coming out with PreDistortion For Dummies in a black and yellow book...

[WBear2GCR]

Predistortion in the form of DSP seems to me to be effectively feedforward distortion cancellation where there is no second "amplifier" to null against to produce distortion products to sum against the output signal. In the case of DSP in effect you have a simulation that does the same trick?

Regardless, even if this is not what is being done, I'm loathe to wade through pages of posts to find information by working backwards from some point in a broad discussion. I'm not in the least bit concerned about the specifics of implementation being obsolete, I'm just trying to get a handle on the general idea/principle and how it can be implemented.

Screw this, I'll just go and be another appliance operator, all fat, happy and smug... FB OM?


                           

[Opcom]

Without costly "Dahl-style" plate iron or grossly oversizing ye olde crufty plate iron, how can good regulation be enforced on a plate supply?

Choke input is the only way mentioned in the books and that is 5 to 10% regulation.

Adding a low voltage high current series regulator in series with the negative side to control that 500V of HV swing? Something could be elegant and ugly all at once.

I find the speech syllabic rate on ssb is fast enough so that I don't get much HV sag. The T/C seems slow enough on the filter caps.  (140 uF)  Though on AM, maybe sag won't be an issue as mentioned in the last two posts. The constant carrier certainly is a great bleeder resistor for regulation.  

The best way to find out is run some tone and voice program tests to see.  There are many things that can be tweaked and optimized that all add up in the end.

A high voltage regulator seems unnecessary, though negative side regulation would be an interesting project to look into. I'd like to hear more about that circuit.

T

Here is the complete detail. I have not tried coupling a regulated power supply's error amp to the HV, but have done the work leading up to it so here is the stuff leading to this point.

The idea of regulating the high voltage by using a low voltage power supply between the negative HV terminal and DC ground came from an experiment intended to raise the voltage on a 2200V choke input filter plate supply to 3000V.

It worked by putting a 950V 700mA choke input filtered power supply taken from an MD-141 or MD-129 between the HV supply and GND. They use a 2000VCT 700mA plate transformer. The manuals are here: http://bunkerofdoom.com/lit/mil_uhf_01/index.html but the circuits are straightforward. Still, worth looking at because of the high current at such a low-ish voltage. I do not recall which modulator's transformer was used for the experiment, I have plate iron from both on the shelves.

This was possible or 'safe' because the plate transformers are 3KVA pole pig cores with two well insulated 2400V windings. The windings on each were (are) connected in series for a FWCT arrangement, 4800VCT.

Although those transformers were originally in oil cans, it was for cooling, not for insulation. Things were made so much better in the old days. Without oil, these still deliver 1500WVA without even getting warm, even do well with a C-L-C filter but that is beside the point here.

Various mad scientist devil may care experiments have made me believe that most KW+ size plate transformers can work with the CT several hundred volts above GND, that is, if the secondary is not explicitly grounded and does not obviously use a smaller insulator or other clue, a few hundred volts is perfectly OK. It does after all have to withstand isolation from the mains and core, and a minimal insulation would be up to 1KV anyway.

The point is that a 4800VCT winding on a normal radio-style transformer making 4400V might not stand up to a bridge with the CT at 2200V, but it will stand 700VDC easily, maybe 1000VDC. These are experiences in which I have blown no transformers, but are not 'facts' applicable to all transformers.

The summary then is that elevating the B- end of a 2500VDC supply to 500-700V is safe, usually. Then you get 3000-3200VDC, depends what you got to start with and your luck.

The next experiment which was pretty foolish I admit was to run cables over to a regulated power supply making 0 to 400V at 400mA. Since I run about 300mA on the final it was no issue. I used a 400uF cap across the regulated supply because I was afraid that fluctuations in B+ current to the class C stage under modulation would do something bad to my precious tube type power supply. The result was the ability to vary the B+ very easily over a range of a few hundred volts, as expected.

If the present plate supply is already decent, and has no more than 10% regulation, then an added range of 0-300 to 400V would be enough for even a 4KV supply. The added regulator only has to either make up, or absorb, the fluctuations. It looks easier to stack a little variable regulated B+ supply in series than trying to regulate the whole plate supply with a giant tube.


Through the experiments, I believe it can be made to work from the standpoint of either adding a separate regulated supply (if you want a little more B+) or adding the pass tubes only if you have too much B+. A caveat to pass tubes by themselves is that they will still need auxiliary voltages for the error amplifier.

I have experimented a lot with those tube type regulated supplies including making them follow precisely an audio waveform out to 50-100KHz. An older experiment of that type is the class A series modulator for an 807 shown at http://www.montagar.com/~patj/amxmtr_dc_classamod.html . This means that they are good enough, with some mods for speed, to follow sudden errors in the B+ that occur when the speech modulation is present. The simple hack is to remove a few of the decoupling caps in the error amp circuit.

 
The experiment as I would do it would be to duplicate an existing series regulator design to save headaches. There are a few high voltage regulated supply designs (1000V) on the www. Two can be found on pages 374 and 377 of this document: http://69.36.164.11/kb/goth/AGH-AM-027p.pdf . I am not saying these should be used as-is, only that they deal with a wide range of voltage and the error amps offer good ideas for that.

One uses 1625's, and the other 4CX150's/250's. Those may be impractical. I suggest cheaper tubes that are able to hold off high voltages and also pass high currents with a low voltage drop.

The 6CD6 comes to mind as an octal sweep tube it finds little use today and can be had for less than $10. Five would handle 1A at 200V drop. It has only a 20W dissipation but the beauty of the operation is such that the higher currents are drawn when the voltage drop is lowest and of course this occurs when the desired boost to correct the sagging of the plate voltage is greatest.

It may seem confusing to consider a series regulated 500-1000V power supply in series with a 3000V plate supply, as in, how can this get the error voltage to do the regulation?

It is cleared up by a simple schematic, or at least I think so. The 550K resistor takes the original article's supply from monitoring its own 1000V to monitoring the total HV of 3500V. The capacitors circled in the diagram are those to be removed or made small (high reactance at the highest audio frequency). Their original purpose is to reduce noise on the power supply, but this would be done by putting a similar value cap across the whole shebang. The cap from the 1625 cathodes to the error amp grid may or may not be needed for stability (experiment) but its function should be referenced to the 3500V terminal, because that is not where the error is to be obtained from. Not the low voltage regulated output. Obviously a smaller cap and maybe a RC network is wanted there. Its purpose is to speed up the reaction to variations in the voltage.

That is the whole thing as I see it. The tubes like a 6CD6 can not take a lot of volts on the screen, as a consequence they can give a very low drop. A separate supply might help, or some scheme. No protection has been considered. I would think the little supply would need some protection from the big one, like a stack of zeners, a spark gap, etc. And that is why to use tubes capable of holding off a lot of volts, like sweep tubes or transmitting tubes.

Like I said I have not done this part of it yet, only thought about it a little. I would appreciate comments on this idea.

[W1ITT]

A few days ago in this topic there was some disappointment expressed at the lack of an easy and palatable explanation of the PureSignal software for reducing IMD in RF amplifiers.  That situation should be rectified for most with the posting of a video of Dr Warren Pratt's presentation at Friedrichshafen hamfest in Germany.  Dr Pratt, NR0V, is the fellow who worked most of the details out.  The software works with the TAPR Hermes board, and the closely related Anan series units from Apache Labs.  Due to different architectures, the Flex boxes are not supported.
The video is an MP4, called    PS.MP4.zip   , uploaded on July 5 on   http://www.hamsdr.com/dnld.aspx     .  There are some other videos there too on somewhat related projects, uploaded the same day by DC6NY, for the technically curious,  Video quality itself is not cinema class, but the presentations are excellent.  Typically, the technical content of the presentations at Friedrichshafen is way above most anything found at Dayton.The actual hardware examples shown are done with two-tone and SSB modualtion, but there is no reason why the system wouldn't work just as well to run squeaky clean AM.   It isn't heavy iron, but the real-time linearity scope certainly has coolness.  This is the new state of the art for amateur radio and, while out for public use, the real smart guys are still pushing ahead week by week for yet more improvement.
Norm W1ITT

[Opcom]

That is better explained than most of the SDR and high-software stuff. I like to read on it even though there is no SDR stuff here.

[WD5JKO]

There is no reason we cannot apply some correction with an AM signal while staying all analog. With SSB we cannot easily do this since we have to reinsert the carrier and then detect. With AM we just detect, get the phase correct, and then apply an audio NFB loop.

I once used an EL-34 tube as a class Ab1 linear RF amplifier in a Central Electronics 20a. It was good for about 8 watts AM. The higher the carrier level, the more curvature I saw in the walls of the trapezoid pattern. I applied carrier detected NFB as described to the audio feeding the balanced modulator. The non linear curvature seen on the trapezoid was reduced quite a bit with only about 6 db gain reduction.

It would be interesting to play with the linearity of the AM detector to have a variable control which applies a correction to have a transfer response that counters the typical 'S' curve of a linear amplifier. The idea would be to tune the amplifier with a tone, and use the scope set for trapezoid. Then turn the knob(s) for flat walls on the trapezoid.

The single ended all triode audio amplifier guys play with the operating points of both the driver stage and the output stage. With careful selection of tubes, DC operating points, and AC loadlines, an amplifier with very low THD and IMD can be achieved without the use of an inverse feedback loop.

It would be fun to apply some of these principles to my Flex 3000. There are those who believe that there is no hardware inside an SDR radio's shell. They be wrong.

Jim
Wd5JKO

[AB2EZ]

Jim

Negative feedback is a very useful technique for reducing distortion produced by non-linearities in the input-output response of a system.

For the application at hand (reducing IMD originating in a linear amplifier) negative feedback at baseband (audio-to-audio) is not likely to be of much help. The input-to-output phase shift, as well as the high-audio-frequency rolloff through the system (from the audio signal at the input of the driver - to - the locally-demodulated output of the amplifier) is going to be too large at the frequencies of the distortion products of interest. For example, i.f. band-limiting filters in the driver will be a problem, if you are trying to remove RF IMD products whose frequencies are further away from the RF center frequency than the pass bands of those filters. Closing the feedback loop would result in instability (oscillations) unless the bandwidth of the feedback path were set too low to be effective in reducing those IMD products that are causing co-channel interference (a.k.a. "splatter").

Separately... with respect to the implementation of pre-distortion:

As you know, the processes that produce distortion products in a class AB linear amplifier can be very complex. For example: as the amplitude of the RF signal, driving the input of a vacuum tube linear amplifier, changes, this will produce changes to the shape (not just the amplitude) of the corresponding plate current pulses (whose repetition rate is the fundamental RF frequency).

This, in turn, will produce (non linear) changes in both the amplitude and the phase of the modulated, fundamental frequency, RF output signal that emerges from the output pi-network / tank circuit.

This is made even more complicated because the harmonics in the RF plate voltage waveform (even though they are much smaller than the fundamental), produced across the input of the tank circuit, will (to some extent) affect the plate current.

Nevertheless, with the right algorithms for looking at the output, and estimating the required pre-distortion, the SDR folks are doing remarkably well.

I'm wondering whether neural network techniques (allowing the neural network to use previous inputs to "learn" what pre-distortion algorithm to use) might be applicable in situations where memory effects are important.

Stu

[W1AEX]

A few days ago in this topic there was some disappointment expressed at the lack of an easy and palatable explanation of the PureSignal software for reducing IMD in RF amplifiers.  That situation should be rectified for most with the posting of a video of Dr Warren Pratt's presentation at Friedrichshafen hamfest in Germany.

Norm,

Thanks for posting this link. I really enjoyed the presentation by Warren (NR0V) and learned a lot about the challenges he and others are dealing with as they work out ways to improve the implementation of Pure Signal. I also enjoyed the video presentation by Hermann (DL3HVH) on the CuSDR project and the development path he is working on. I'm fairly ignorant about software programming and understanding how the software and hardware work together to get the job done but they did a great job of explaining it. The next couple of years are going to be quite an adventure for those of us who play with the SDR platforms!

73,

Rob W1AEX