Negative Feedback audio in PWM

[VE3ELQ]

Have just completed the design and build of a high side (current source) Pulse Width Modulator for a 160 volt 200W carrier full bridge RF deck. When fed with a 500 hz audio triangle wave the filtered output into a resistive dummy load looks pretty good on the scope but shows a very slight non linear skew in the ramp up and down of the waveform.  I thought that by wrapping this back to the Pulse Comparator with negative feedback it might ensure better linearity as is done in audio amps.  But the output is phase shifted about 40 degrees probably by the 4 pole low pass filter.  So I'm stuck on how to do this without making it into a high power oscillator.  Any ideas out there would be appreciated.
73s  Nigel  VE3ELQ

[PD0RTT]

Well anything  what cause distortion is the pwm conversion and the filters beyond it.
A form of negative feedback reduce phase shift errors what occurs in the pwm filter, it also flattens the audio response to a flat curve in the audio bandwidth.

Regards,

Martin

[WD5JKO]

Nigel,

   It would be interesting to see more information such as a block diagram, and what devices are used in the modulator and RF PA stage.

In regards to your question, I presume the 500 hz triangle ramp non linearity is when the PWM approaches 0% duty cycle and 100% duty cycle. If so, this would mean the switching times are becoming significant with respect to either the off time or the on time of the PWM output. This leads to a departure from a linear ramp. If the non linearity is at the middle of the waveform, then the PWM ramp generator likely has a "bow" in it. Not related here, but some of the digitally synthesized audio waveforms have a glitch in the middle as the MSB toggles on or off.

A high side switcher is difficult to achieve good isolation and at the same time have fast rise and fall times from the driver. I think this is the crux of your problem. Beefing up the driver, and/or using PWM power devices (If they are FET's) with lower gate-source capacitance and/or lower reverse transfer capacitance should help. The IXYS folks seem to be leading the way here with big HV power FET's.

One other way around the PWM extremes is to widen the dynamic range of the switcher such that you would normally never exceed the 10-90% duty cycle range. If the minimum modulation percentage were 5%, I'd say this is a good thing. My G-76 with negative peak attenuation (not a clipper) goes from about 10% to 150% at the modulated envelope, exactly what my goal was.

Maybe a little further input from you will help use better understand what you are doing.

Jim
Wd5JKO  

[Opcom]

I'd sure like to see a diagram as well. I tried NFB on a home made class D audio amp with poor results and instability.

[VE3ELQ]

Thanks for the replies.  All my schematics are hand drawn, one day will get some cad app.  Attached is the modulator schematic as requested.

Nigel  VE3ELQ

[WD5JKO]

  Nigel,

   Your circuit is quite interesting. Having a linear ramp for the PWM is very important, and your implementation of a J310 as a CCS to feed the ramp will work, but I'd wager the CCS capability of a J310 is not a straight line as the VDS goes from zero to Vdd. Many IC Chip PWM's exist that are much better. If you prefer to keep the PWM discreet, perhaps try something with a PNP transistor:

http://www.tedpavlic.com/teaching/osu/ece327/lab5_adc/lab5_adc_ramp_generator.pdf

  Look at page 7, and replace the three diodes with an LED.

Jim
Wd5JKO

[steve_qix]

Putting negative feedback around a PWM is highly problematic unless the switching frequency is very high and the corners of the filters are WELL outside of the audio range.  Otherwise, the phase shift in the various filters is a big problem.

But, back to your original issue - the ramp generator is most likely the culprit.  I used to build discrete PWM generators, and getting the ramp just perfect was a challenge.  Yes, it can be done, but it's not necessarily  trivial.  Finally found the UCC25701 PWM controller IC, which has an EXTREMELY linear ramp generator.  I tried a lot of other chips, and this one was the best.

What is the frequency of your RF amplifier (just curious)?

Regards,  Steve

[VE3ELQ]

Jim, Steve,
Thanks for the advice.  Based on your comments I re-visited the ramp generator and had a close look at it with the scope full screen and yes it has some slight exponential skew.  I was sure it was linear when I first made it.  So Jim will try a high HFE PNP at 12v source instead of 6v and see if I can get it better.

Steve thanks for the comments. I was pretty sure negative feedback was going to be next to impossible but there are some clever folks out there that may have figured it out.  

Couple pics attached. The bridge amp will do 8 mhz using a IXDD604 DIP8 driver for each FET but the pulse transformers need careful optimization for each band.


73,s  Nigel

[Opcom]

TI took the UCC1570 off the recommended for new designs list. It's still available.

Replaced By "UCC25701 – The device has the SAME FUNCTIONALITY and PINOUT as the compared device but is NOT an exact equivalent." Probably as linear, can't see them degrading it. So maybe that is an alternative for the future.

[Opcom]

But, back to your original issue - the ramp generator is most likely the culprit. 
Regards,  Steve

Had you considered a digital ramp generator? By this I mean one with maybe 256 steps, made of counting ICs?

I built one previously for a totally different project. It was as linear as anything would be where voltages are added precisely. The ramp frequency was low, about 780Hz IIRC the crystal clock was 200Khz. It might be possible to build a 100KHz one with modern ICs. Could be a bad idea as well but I thought it worth mentioning.

[VE3ELQ]

Jim, Opcom,
Im trying to keep this simple yet functional as one of the design goals, its passing the winter when Im not snow blowing.
I redesigned the CCS in the ramp gen with a PNP BJT and a 2.4v zener and ran it at 9v.  After a bit of fiddling got a very nice ramp.  Think I will leave it alone now and get on with the metal bashing.  I want this on the air soon.
Regards es 73s,  Nigel VE3ELQ