Official Position on Adaptive Predistortion

  • 12
  • Idea
  • Updated 3 years ago

Periodically the topic of Adaptive Predistortion comes up for discussion on this community.  Our team has been consistent in our position on this subject but sometimes it is useful to restate what has been said many times before in the interest of clarity.  Here is a restatement of the facts:

  1. The FLEX-6000 Series hardware has the necessary connections and internal RF port to port isolation to support adaptive predistortion software.  We have verified that transmitter output to receiver input isolation (e.g. ANT1 to RXA) is >80 dB on all bands including 6m, which is more than ample to accommodate adaptive predistortion using an external amplifier in the loop.

  2. FlexRadio remains interested in adaptive predistortion technology but we have not in the past nor will we now make a time frame commitment.  Our internal engineering resources are 100% allocated to complete our published road map for v1.x releases through v2.0.  We believe that the features committed in those releases are currently of higher priority to the vast majority of our customers than is adaptive predistortion.  

  3. We periodically review and rank our software backlog and feature requests.  We look at the ranked priorities and compare the development magnitude for each feature to the available engineering resources within a release cycle.  Based on those factors, we make decisions on what features make the cut in the release planning process.  So far adaptive predistortion has not made the cut.  That doesn't mean it won’t make the cut in a future release.
73 and thanks for your support,
Photo of Gerald - K5SDR

Gerald - K5SDR, Official Rep

  • 794 Posts
  • 1433 Reply Likes

Posted 3 years ago

  • 12
Photo of Ken - NM9P

Ken - NM9P, Elmer

  • 3875 Posts
  • 1181 Reply Likes
Thanks, Gerald.  I will be glad to install APD when you get to it, but for now....First things first.  
Photo of KT0AM  - Mark

KT0AM - Mark

  • 16 Posts
  • 1 Reply Like
Dang... my background is geophysics and data processing and I don't have a clue about adaptive predistortion. I've often thought that some of the techniques used in geophysical data processing, like FK filtering (as opposed to time domain filtering) my be interesting to play with but this is a new concept for me. Anyone have a quick explanation?
Photo of Rob N4GA

Rob N4GA

  • 145 Posts
  • 25 Reply Likes
Basic equivalent in the analog world of negative feedback in a differential amplifier?

But in this case, done digitally?
Photo of Ken - NM9P

Ken - NM9P, Elmer

  • 3859 Posts
  • 1171 Reply Likes
I would say it is more like forward error correction. Though the analogy is not totally accurate. Or like what Bose does on some of its speakers .... It modifies the input into the amp in order to account for the imperfections in the speakers.
Photo of k3Tim


  • 810 Posts
  • 155 Reply Likes
It is more more complicated than that...

Also, 100% agree with the OP also (if that means anything...) 

Photo of Rob N4GA

Rob N4GA

  • 145 Posts
  • 25 Reply Likes

I doubt it's a ton more complicated. What's normally so "complicated" about things like these, are the terms used (adaptive pre-distortion in this case) and the convoluted explanations.

Sounds to me simply that you characterize the PA at a given frequency, mode, load, etc., then modify the input to the PA to optimize for the PA characteristics.

All done digitally.

If that's accurate, then it should be called something like digital error correction, or digital negative feedback, etc.

Not something that sounds like you are introducing distortion.


Last, the bang for the buck is probably minimal, thus the low priority being given by FLEX.

Photo of k3Tim


  • 810 Posts
  • 155 Reply Likes
Sometime ago I worked on a design of an HDTV transmitter that would use the pre-distortion.  This was for 16-QAM and it saved a lot of grief compared to using conventional tuned filters to get sharp skirts.  A PhD from local University did the algorithm design (the University was/is in top 5 Engineering schools).  I recall the h/w design had about 20 FIR taps running at high speed.  I believe the problem is the conditions are changing dynamically (load / PA current / Temps / etc) therefore one has to correct dynamically.

I did see an IEEE paper about using some phase inversion of the 3'rd and higher order products, feeding those back and that would also seem to work.  A  quick review seemed to show it was an easier route but not sure if it would apply to this setup.

I don't particularly care the term Pre-Distortion either.  In the case of the 16-QAM one would know where in the constellation the symbol should be and where it ended up in the PA and then 'offset' it in the input stream to the PA so it landed exactly as expected.  That would seem an easier problem than complex voice signals.  Since the PhD handled the algorithm not sure.

If I come across an interesting / readable paper I'll report back.  Problem is the day job keeps me busy and at night / early morning the 6500 has my full attention.  
What a beast!

Best Regards,

Photo of Bob G   W1GLV


  • 639 Posts
  • 107 Reply Likes
Gerald, we stand behind you on that decision. Your leadership abilities have made what FRS is today. Continue with the fine job your team provides for the Flex community.
Photo of Jay -- N0FB

Jay -- N0FB, Elmer

  • 534 Posts
  • 210 Reply Likes
I agree with Bob.   It sound like you have the correct priorities.  Full steam ahead.  As always, please thank your staff for their hard work.  We do appreciate them.
Photo of WA6FXT Mike


  • 207 Posts
  • 37 Reply Likes
Like Bob says... "We stand behind you ..." In fact, I'm sure others, like myself, stand in front and help try and respond to those who want to interfere with the current roadmap. We ALL want everything (hihi); but, most of us are perfectly happy with the path FRS has chosen. That is not to say some would not have preferred a different sequence (hihi); however, I'm enjoying each and every update. (Bugs included... Heck, I love a challenge.

73's and keep up the great work!

Photo of np2g


  • 271 Posts
  • 29 Reply Likes
It is just a positive event  for a absolutely state of the art  transmitter .One that exceeds the present possibilities .  And smart companies  Like Flex will implement it .

The statement has always been  sooner would be better.

So the next time someone asks if you would like that million dollar signal. Try saying NO.  

When Flex does have the time to implement  Pure signal or whatever they call it  It will be the best .

Just remember the fastest and shortest distance between 2 points  isn't a straight line ,
Photo of Bill N5TU

Bill N5TU

  • 48 Posts
  • 14 Reply Likes
Hi, Gerald:

Thanks for the communication.  I look forward to the implementation of adaptive predistortion, but you have plenty to do right now.

73, Bill, N5TU
Photo of KY6LA - Howard

KY6LA - Howard, Elmer

  • 3471 Posts
  • 1335 Reply Likes
Here are some screen shots from a talk at Friedrichshafen 2014 that may explain ADP better than words

HOWEVER. Flex Need's to finish 1.4, 1.5 and V2.0 BEFORE they start with ADP which is still clearly a "Science Project"

Photo of Rob N4GA

Rob N4GA

  • 145 Posts
  • 25 Reply Likes
Very nice, thank you for posting that.

What would be nice is if APD could eliminate ALL the high energy harmonics so you could eliminate post PA filters all together.

I wonder if this could be accomplished with a different PA topology (vs push-pull) along with APD. Imagine maybe parallel amps with one side used primarily to negate harmonics, all controlled with APD. Execution gets very complex very quickly...

Going form -16 to -53 on IMD3 is pretty darn impressive though!
Photo of k3Tim


  • 818 Posts
  • 158 Reply Likes
The complete PPT is at:

slides 20&21 shows the correction algorithm - doesn't look complicated but the transcendentals take time to compute.  Converting to FPGA code I won't comment on.

In this blog:

that author states only IMD is corrected so it seems harmonics are not.  Also he states:
"badly tuned linear amp stages" can't be corrected.  To me that's the crux of the problem in the worse offenders.



Photo of Rob N4GA

Rob N4GA

  • 145 Posts
  • 25 Reply Likes
Good stuff Tim!

To convert an algorithm to FPGA code, you use EDA tools from Synopsis, Mentor or Cadence.

Mentor's is called Catapult C and costs about $200k fully configured! 

You write Algorithms in C and it converts to VHDL or Verilog for your FPGA designing pleasures.
Photo of k3Tim


  • 817 Posts
  • 156 Reply Likes
Ouch !  $200k.  You would think one could write the C code and have a consultant / contract company convert a project for you to VHDL.  The design / engr / production gets hugely expensive.  I am plugged into this owing to the day job.

I wasn't aware they were converting C to VHDL - thanks for the info - good stuff.

VHDL is a big step up from the old Mead-Conway method:

I also see we have JTAG on the PWB, a big plus for factory testing.

Best Regards,

Photo of np2g


  • 271 Posts
  • 29 Reply Likes
Wait a minute guys 
If we are looking at the present implementation of "Pre   or clear signal " the code is already out there  Yes In open source  .
The hardware is already in the Flex to accomplish this event .
According to Flex the flex was built with the ability to do this.

So     there is a relatively easy to follow  Open source  programming " call it a guide " that if anyone wanted to do this  Anyone!!!!    it could be utilized  . If just to start things off. 

And Now putting the entire package into the FPGA  isn't a real smart idea.

I think it would benefit anyone thinking about  this , a simple You Tube on a present system using the reduction would explain far more than any   techno papers that were written that started this whole thing . 

What you would find that there is USER necessities  presently to set this up. You would find that there is ,When adapted this way, secondary programming  which can reside in the processor that can accomplish this A/B comparison. This gives a relative easy way to continually improve or modify it . 

Simple statement . And I can give you the techno analysis . 

As it has been named It compares the wave form of a transmitted signal in the band pass (PRE) and compares it with the output of the transmitted signal (Both of equal strength) the difference in this wave form  is sees and measures it corrects /  shifts to match the template . 

This goes for any and all  difference it sees (In that band pass) compared to the algorithm it uses as a template. 

The result in the real world is (Seen and heard every day) a CLEAN IMD reduced  SIGNAL Which can be displayed( using the most harsh of tests the 2 tone  test)  . The test  just about every manufacturer stays away from .

And 200K Well  will take time (Programming time)   Someone else did it with volunteers .(A true ham spirit) 

I am glad we are actually understanding what benefits we would get. And how significant utilization of absolutely any system that reduces IMD  is fantastic .

And I am glad some day Flex will Implement it .