Converting Collins 820D-2 AM broadcast transmitter to 4-400As.

[flintstone mop]

Here's the latest update. 

I was able to improve the modulation symmetry by adding an 8H choke on the output of the screen voltage power supply at the input to the screen voltage dropping resistor.

Last night, I took a quick look at the audio output spectrum coming out of the high side of the modulation transformer using a Tektronix P6015 HV probe and a DC block to AC couple the signal into my 7L5 low frequency spectrum analyzer.  With the 10 kHz audio input level adjusted to roughly that which corresponds to 100% modulation, there is a considerable amount of even and odd order harmonic distortion.  The second harmonic is only down 30 dB from the fundamental with the rest of them gradually falling off.  All of this harmonic energy is translated to the RF spectrum when the PA is modulated.  This also occurs at lower audio frequencies, but it isn't quite as severe.

When the audio input level is reduced, the harmonic energy starts falling off rather quickly.  A 10 dB reduction in input level nearly eliminates the harmonic distortion.  The high level waveform isn't clipped.  It seems as though something is being driven into overload, though.  I was suprised to see so much harmonic distortion at the output of the modulation transformer.  Having never looked at the audio waveform at this point before, I would have expected the harmonics to be down 50 dB or more when the modulator section is running at full power (in this case on low power).  Has anyone ever made a measurement at this point before?  What should I expect?

Is there easy way to determine what the proper load impedance for testing the modulation transformer is?  I am not certain that the test I made was valid with the modulator still connected to the PA.  I am beginning to think that I need to isolate the modulator from the PA to verify/troubleshoot the problem.

Of course, there are any number of things that can cause this problem but perhaps there are some thoughts as to where to begin?

Thanks for everyone's help!

Brian, WB6QED

Brian, you may have fixed an old engineering problem with that TX, that previous owners, users, B'casters never noticed or bothered to correct.
It seems that there's a not-very-nice track record of poor performance with 4-400's.

Fred

[WB6QED]

I took a hard look at the audio driver section.  I actually found one small problem, but that wasn't the cure.  The collector voltage to one of the output transistors had an intermittant connection.  While I had the circuit board out, just for fun, I matched all of the transistors using my curve tracer.

The audio driver stage is by no means the all time greatest audio amplifier, but if I can get things to work, its performance should be acceptable.

I measured it without any load by disconnecting it from the input to the modulator tubes.  I measured the distortion and 100 Hz, 1 kHz, and 10 kHz frequency response at 100 V p/p, 200 V p/p, 300 V p/p, 400 V p/p and 500 V p/p output with a sinewave input connected to the transmitter audio input.  The output begins to overload at just under 500 V p/p (open circuit).  THD+N up that point looms in the 0.75% range at 100 Hz, 1 kHz, and 10 kHz.  At 100 Hz, the output level increases +0.5 dB and at 10 kHz it decreases -1.5 dB.  Distortion and frequency response remain consistent at all input and output levels.  The distortion was predominantly 2nd and 3rd harmonic at roughly -44 dB and -50 dB down from the fundamental.

I coupled the signal from the audio driver board using two X10 probes connected to each differential output and connected them to the input of a Tektronix 7A22 differential amplifer.  I used the vertical output from the 7934 oscilloscope to drive the 7L5 and the distortion analyzer.

Things change when the modulator tubes are connected.  The audio driver seems to run out of steam at less than 300 V p/p.  Although, it may not a good measurment point once the tubes are connected to the audio driver?

Any idea what the input impedance to the 4-400As might be?

To look at the distortion at the input to the modulation transformer, I will need to get my hands on another X1000 probe.

I wonder if I have a flat modulator tube???

More to come later.

Brian, WB6QED

[WD5JKO]

Any idea what the input impedance to the 4-400As might be?

Well, if class Ab1 G1 driven, an open circuit with capacitance to the other elements. Once you draw grid current (class AB2), things change rapidly. Look at the 4-400 curves to plot out the grid 1 current and voltage versus plate voltage. The driver load will be non-linear once grid current begins. This is where the driver falls apart if not low impedance enough, or having enough NFB.

Jim
WD5JKO

[WB6QED]

Here are trapezoid photos at 1 kHz and 10 kHz with the transmitter on low power  (250W). 

As you can see with the 1 kHz photo, it is still hitting 100% negative before making 100% positive modulation. 

The 10 kHz photo looks plain ugly!

I sampled the high side of modulator transformer output for the horizontal axis and the modulation monitor sample for the vertical axis.  The carrier reference is 4 divisions.  8 divisions = 100% positive, and the center line = 100% negative.

What do you think???

Brian, WB6QED

[KA2DZT]

You're assuming the amplification of the scope is 100% linear.   Can you be sure it is linear??  One would think that a high quality scope would be linear.

Anytime I ever did tests like this I fed the scope plates directly without any amplifiers.

The 10KC pattern does look sick.  At 10KC, could the screen bypass caps be too large?  Considering the xmtr was used at BC freqs, the screen bypass caps may be much more than the usual .001mfd.

Just a few thoughts,  probably no help at all

Fred

[WB6QED]

I did some additional measurements tonight and all they really did was add to my bewilderment.

I left my high voltage probe connected to the high side of the modulation transformer.  At the same time, I had the differential output of the audio driver connected to my Tektronix 7A22 differential input amplifier with the vertical output of the Tektronix 7844 oscilloscope connected to either the input of my Potomac Instruments AA-51 audio analyzer or the input to my Tektronix 7L5 spectrum analyzer.

The modulating frequency was 1 kHz.  The transmitter was operating at 250W output.

For the most part, the same distortion components that are on the grids of the 4-400A modulators also appear at the output of the modulation transformer.  Interestingly, as the grid bias is made less negative, the audio drive decreases.  When the bias is more negative, there is more audio drive to the modulator grids everything else being equal.  Grid drive at 100% modulation is roughly 300 volts p/p.  At this output level, the waveform at the grid of the 4-400As is beginning to go into clipping.

I then decided to make some further adjustments while looking at both the input to the 4-400As and the output of the modulation transformer since both of these measurement locations have a direct correlation with the demodulated distortion. 

At 50% modulation, best distortion was obtained with about 15 mA of PA grid current, but at 90% modulation the PA wanted all the grid current that I could give it (about 45 mA).  At 90% modulation, I could carefully adjust the modulator bias and balance to minimize the distortion.  When I did this, I ended up with about 2.25% demodulated harmonic distortion (primarily third harmonic).  The interesting thing is that as I reduced the audio input level, the demodulated harmonic distortion did not significantly decrease.  However, if I readjusted the bias to be less negative, I could reduce the harmonic distortion dramatically at lower modulation levels.  The choice seems to be high distortion at all modulation levels or very high distortion at high modulation levels and low distortion at lower modulation levels. 

I don't understand why changing the modulator grid bias is having such a noticeable effect on the grid drive everything else being equal.  As the grid bias is made less negative, the audio drive decreases.  What is happening here?

I took a quick look at 10 kHz, and the audio level at the output of the modulation transformer was greater than the level at 1 kHz.

When I measured the audio driver output at the grid input to the modulators with the 4-400As removed yesterday, the audio driver was capable of 500 V p/p at less than 1% THD+N and the 10 kHz output was -1.5 dB from the 1 kHz reference.

It would be interesting to see what the modulator does when it is disconnected from the PA.  A change in the PA grid bias can be seen all the way back at the grids of the modulators.

Any thoughts or recommendations for additional measurements would be appreciated.

Brian, WB6QED

[WD5JKO]

For the most part, the same distortion components that are on the grids of the 4-400A modulators also appear at the output of the modulation transformer.  Interestingly, as the grid bias is made less negative, the audio drive decreases.  When the bias is more negative, there is more audio drive to the modulator grids everything else being equal.  Grid drive at 100% modulation is roughly 300 volts p/p.  At this output level, the waveform at the grid of the 4-400As is beginning to go into clipping.

   Ok, this is cool, and simplifies the problem. It suggests that cleaning up the grid drive at high drive levels will clean up the modulated B+ waveform. It would be helpful if you could post some schematics. So less bias means higher static modulator plate current, or being closer to class A operation. With class A, the distortion products will be less, but power capability reduced. Increasing the bias (more -), decreases the static plate current, and your into AB1, or AB2 operation where the tubes conduct less than 360 degrees, and higher power is available so long as the grid (G1) can swing into grid current without overloading the audio driver.

  I'd scrutinize what happens as you start to clip the driver output. What is the modulator plate (try both) voltage minimum? Is it below the screen voltage? Is the screen voltage dropping also? Maybe if the modulation transformer turns ratio is too low (close to 1:1), Maybe if the modulation transformer turns ratio is too high, then the modulator runs out of steam as you approach 100% modulation. If so, boosting the audio driver peak drive capability won't help much.

  You mentioned the RF PA grid drive needs to be higher at 100% modulation. This suggests to me that there might be too much fixed bias, and not enough (or any) grid leak bias.

  Any chance the RF PA and modulator grid bias supplies are cross talking sort of like an old Johnson transmitter? If so, fixing this should help. With a Viking I for example, increasing the 4D32 grid drive also adjusts the modulator static plate current. Not good.

  You mentioned the Modulated B+ peak level rises at 10 Khz over that at 1 Khz. You also see audio phase shift on the 10 Khz trapezoid. All audio transformers have a self resonant frequency, and phase shift as we approach this resonant frequency. We like this frequency to be way above the highest audio frequency modulated though. A couple of things will help here. First a series R-C from modulated B+ to ground where the 'R' might be 2X the modulator load impedance, and the 'C' a value where the Xc matches the 'R' at 10 Khz. Just a guess, but an easy test to see what happens. The other thing would be to have a low level LPF to limit your upper audio frequency response to 5 Khz or lower.

  The screen bypass capacitor for the RF PA might be looked at as well as a contributor to the phase shift at 10 Khz.

Jim
WD5JKO

[WB6QED]

The modulation transformer has a 1:1.25 ratio.

Attached are some excerpts from the schematic diagram.

[WB6QED]

I detached the negative feedback and took a look at things.

The output of the audio driver stage cleans up dramatically.  With the negative feedback disconnected, the output from the audio driver looking into the grid of the 4-400A modulators is clean.  At its output, the frequency response and distortion performance roughly correlate with the measurments that I got with it disconnected from the modulator tubes.  It is good for about 450 volts p/p looking into the tubes without feedback.  The audio driver goes into clipping after the output of the modulation transformer has already start to clip.

The demodulated frequency response roughly correlates with the frequency response at the output of the audio driver stage.  +0.5 dB @ 100 Hz and -1.5 dB @ 10 kHz.

At 100% modulation, the distortion at 100 Hz at the output of the modulation transformer goes through the roof.  At 1 kHz it is about 12% and at 10 kHz, it is 0.92% (a huge difference from when the feedback is connected!).

Ok, so the negative feedback is doing something....

See the attached photos.  They show the RF output spectrum, modulation transformer output, and envelope modulation.  Measurments were taken at 100 Hz, 1 kHz, and 10 kHz.

More to think about....

Brian, WB6QED

[WB6QED]

Here are trapezoid shots with the audio feedback disconnected.  100 Hz, 1 kHz, and 10 kHz.

[k4kyv]

Here are trapezoid shots with the audio feedback disconnected.  100 Hz, 1 kHz, and 10 kHz.

If the sample going to the vertical axis from the the modulation monitor is an accurate, undistorted representation of the rf output from the transmitter, it looks like you have excellent modulation linearity in the rf final, so any distortion in the envelope pattern is likely to be from the modulator, not the final.  Another alternative would be to feed the vertical axis directly with an rf sample taken from the output terminal at the transmitter. Assuming you use coax, a tee-connector could be used; feed the rf from the transmitter to the load through the tee, and use the 3rd port on the tee connector to feed  the scope through a coupling capacitor.  A few pf should be enough. Better still, use a small air variable.  The idea is to couple just enough rf to the scope from the feedline to get a good deflection pattern.  If the scope has a direct input connection to the vertical plates that bypasses the vertical amplifier, use it.  You should easily be able to get enough rf voltage from the transmitter at amateur power levels @ 50 ohms to the scope tube to get a good pattern without any vertical amplifier. This way, you know for sure the pattern is not being altered by non-linearity in  the scope's vertical amplifier.  Non linearity shouldn't be a problem with a good lab quality scope, but it may be an issue with cheap bench or monitor scopes.

The "ugly" pattern with the 10 kHz tone is due to phase shift in the coupling network that samples the audio from the mod transformer.  The audio frequency and waveform of  the modulator should have nothing to do with modulating linearity of the final, so you can ignore that flaky pattern.  You could probably get a better picture with a 100X probe, if you can find one that would take the voltage directly off the modulated +HV.

[WB6QED]

If the sample going to the vertical axis from the the modulation monitor is an accurate, undistorted representation of the rf output from the transmitter, it looks like you have excellent modulation linearity in the rf final, so any distortion in the envelope pattern is likely to be from the modulator, not the final.  

The "ugly" pattern with the 10 kHz tone is due to phase shift in the coupling network that samples the audio from the mod transformer.  The audio frequency and waveform of  the modulation should have nothing to do with modulating linearity of the final, so you can ignore that flaky pattern.  You could probably get a better picture with a 100X probe, if you can find one that would take the voltage directly off the modulated +HV.

I am deriving the RF sample directly from the modulation monitor sample BNC connector that has a direct connection to the RF output network.

I am sampling the modulated B+ at the high side of the modulation transformer output using a Tektronix P6015 X1000 probe that I believe is properly compensated to the scope's input.  It has a -6 dB point of 50 MHz, an input resistance of 100 M and an input capacitance of 3 pF so I am surprised that the weird 10 kHz trapezoid waveform is the result of improper coupling.

I am using Tektronix 7000 series oscilloscopes and plug-ins that I believe are in good working order.  The 7844 and 7934 oscilloscopes that I am using have a -6 dB point that is at least 400 MHz.  For single ended measurements, I am using a 7A26 200 MHz vertical amplifier.  For differential measurments, I am using a 7A22 1 MHz differential amplifier.  I am using a 7B92B timebases.  For spectrum measurments, I am using a 7L5.  I am using a Potomac Instruments AG-51 audio generator with its complementary AA-51 audio analyzer.  I am demodulating the modulated RF carrier with the AA-51 using a Potomac Insturments DX-51 precision AM demodulator that connects directly to the AA-51.  For impedance measurments, I am using an HP 4815 vector impedance meter.

I took a little closer look at the output of the modulation transformer on low power (1500V @ 0.25A) with the audio feedback disconnected at various input levels using a 100 Hz sine wave.  I get excellent performance at 2000 volts p/p, but as the output voltage increases, the sine wave starts to tilt.  At 3000 volts p/p, it looks pretty ragged.  It seems that I am running out of modulator linearity much sooner than I should be.  What could be the cause of this?  With the negative feedback disconnected, the output of the audio driver is looks and measures distortion free well after the point at which the modulation transformer output has become non-linear.  If I could get the modulators to follow the audio driver, I would make +125% positive modulation easily!

It would appear that the next things to try are:

1) new modulator tubes
2) test the audio system with and without audio feedback with the modulation transformer terminated into a 6K dummy load.

Brian, WB6QED

[w3jn]

Did you ever try increasing the screen voltage, or comparing your modified xmitter with a GPT-750 as I suggested above?

[DMOD]

I would also check the electrolytic coupling caps in the output of the solid-state driver and replace them with bi-polar caps as a minumum and replace the driver transistors, which could be leaky as well.

My thinking is the 4-400 grids may have a slightly lower impedance and may be loading the collectors of the solid-state driver somewhat higher than did the 5-500's.

Phil - AC0OB

[k4kyv]

As a last resort, you might consider replacing the solid state driver with a R-C coupled tube type driver, consisting of a pair of 12BY7s. That's what the RCA BTA-1R uses to drive a pair of 4-400s to modulate another pair. As I recall, the Collins 20V uses the same line-up. In both those rigs I am pretty sure the modulators operate class AB1 and still deliver enough audio to fully modulate the transmitter at 1KW carrier output, which should be about 1400 to 1500 watts DC in.

I am envious of that perfect straight-sided trapezoid pattern.  I can't quite get that with my push-pull triode final.

[WD5JKO]

Brian,

  I believe that some of your difficulty has to do with the differences between a Pentode and a Tetrode, i.e. the 'Kink' in the plate curve as the plate voltage drops below the screen voltage. This is absent with the pentode, present with the tetrode, and largely fixed with the beam power tube (6L6, 807, 813..)

http://www.r-type.org/static/btet.htm
"The resultant device was a great success, revolutionising HF amplification, but it had a serious problem: the tetrode kink, which rendered it unsuitable for large signal handling (power amplification)."

Now look at the 4-400 spec sheet and curves:
http://www.g8wrb.org/data/Eimac/4-400A.pdf

 Notice the constant current chart with G2 at 500V. Below 500V the plate current curve slope changes, as does the screen current.

  So reading between the lines, to get a larger undistorted plate swing out of the 4-400, we need to reduce the screen voltage. Doing so limits the maximum plate current at zero bias (limit of class AB1 operation). This means we must drive G1 into grid current (class AB2) to get back to where we were with a higher G2 voltage.

  The audio driver schematic posted cannot supply any grid current, so you are limited to class AB1 operation without a major driver rework effort.

  I find the 700 ohm adjustable 300 watt resistor in series with the 50 Henry modulation reactor interesting. What is that used for?

Jim
WD5JKO

[WB6QED]

I find the 700 ohm adjustable 300 watt resistor in series with the 50 Henry modulation reactor interesting. What is that used for?

It is used to adjust the plate voltage of the PA.

I have been doing the latest round of measurements at low power.  The modulator plate voltage gets reduced to 1500 volts from 3000 volts, but the screen voltage remains at 750 volts.

I agree, the "tetrode kink" may be the culprit.  Because of this characteristic of tetrodes, the low power configuration would have impaired modulator performance without a corresponding reduction in screen voltage.  Right???

The 5-500A data sheet dipicts much smoother performance at low voltage.

Interestingly, in the Bauer 707, the Collins 20V series, and the RCA BTA-1R transmitters, low power cutback does not result in a reduction of the modulator high voltage.  The modulator high voltage and screen voltage remains the same and therefore the modulator performance remains identical at high and low power.  Low power cutback is done using dropping resistors that are in series with the PA high voltage supply.

On the other hand, with the 820D-2, the high voltage to the PA and the modulator tubes are cutback, but the screen voltage does not change!!!  For comparable operation, shouldn't all of the parameters be scaled down.  Half high voltage and half screen voltage???

Perhaps, it was the use of the 5-500As that facilitated this design change?

For low power operation, the McMartin BA-1K which uses 4-400As, cuts back both the PA and the modulator high voltage.  Like the 820D-2, it does not alter the modulator screen voltage for low power either.  It too must suffer from this tetrode characteristic.

I guess the next thing is to do some similiar measurements at high power to confirm all this.

I suspect that the high THD at 10 kHz is being caused by the two 1000 pF PA screen bypass capacitors.  The 20V series transmitters used 67 pF on each PA screen.  The Bauer 707 used one 1000 pF for both tubes.

Thanks for the help!

Brian, WB6QED

[WD5JKO]

I find the 700 ohm adjustable 300 watt resistor in series with the 50 Henry modulation reactor interesting. What is that used for?

It is used to adjust the plate voltage of the PA.

   Interesting how they reduce plate voltage in such a lossy way. Since it is there, why not move it to drop voltage on the modulated B+ line instead of the unmodulated B+ line? Then bypass it with a suitable sized oil capacitor. The idea is similar to what we do with a Heising modulation choke system that can only modulate upward to 80% or so. Here we use a DC dropping resistor to drop the modulated B+ line 20% so that we can then modulate the RF final to 100%.

Just an idea...

Jim
WD5JKO

[WB6QED]

Since it is there, why not move it to drop voltage on the modulated B+ line instead of the unmodulated B+ line?

That is the way it is configured.  The center tap on the input side of the modulation transformer is fed from the unmodulated B+ line.  The rheostat is just ahead of the 50 H choke on the modulated B+ line.

I really appreciate your help sorting out this puzzle!

Brian, WB6QED

[KA2DZT]

A resistor in series with the PA B+ was used to allow the rig to reach 100% modulation when using Heising modulation.  THe resistor is not needed when using the reactor with a mod xfmr.  I use a 50hy mid reactor in my HB xmtr and I can easily do 150% positive modulation.  I'm not using any series resistor.

I looked at the schematics.  The one thing I notice is that I don't see any grid leak resistor.  I do see a small resistance in series with the grid choke but there is no bypass cap between the RFC and the small resistor.  Not sure why that resistor is there.

It seems that all the grid voltage is from the the bias supply.  I think it's better to use all grid leak bias when using AM.  At least use some grid leak bias.

Fred

[K5UJ]

The bauer 701 also uses a pair 12BY7s to drive the modulator (4-500As).  It cuts the B+ to the PA from 3 KV to 1.5 when it switches from 1 KW to 250 watts.   Modulator B+ always stays at 3 KV.  Yes that trapezoid looks like a textbook or handbook example.

[WB6QED]

I looked at the schematics.  The one thing I notice is that I don't see any grid leak resistor.  I do see a small resistance in series with the grid choke but there is no bypass cap between the RFC and the small resistor.  Not sure why that resistor is there.

I have tried different values for R15 (220 ohm), and its value has a big effect on the grid bias when the RF drive is on.  The fixed bias supply is -155 volts.  When R220 is changed to 4.7K, the grid bias decreases to -240 volts.  At some point, I will probably adjust the value of R15 for optimum signal to noise ratio and modulation performance.

As I had anticipated, the transmitter is now operating very nicely at high power.  4:1 IMD at 100% modulation is approximately 2%.  THD+N is fairly uniform at approximately 1% across the board.

To improve the performance on low power, I believe that the modulator screen voltage needs to be reduced along with the high voltage.  Evidently, the 5-500As had acceptable performance without changing the screen voltage.

Just for fun, I did an AM-NRSC plot of the transmitter at high power with program audio.  It now looks as good or better than most solid state transmitters currently in AM broadcast service.

Brian, WB6QED

[WD5JKO]

Just for fun, I did an AM-NRSC plot of the transmitter at high power with program audio.  It now looks as good or better than most solid state transmitters currently in AM broadcast service.

  Looks like in the 1980's when the FCC was meddling with the Ham Radio DC input rules, and the AM Grandfather Clause, they were also meddling with making the NRSC rules that applied to AM broadcasters. See attachment.

  So Brian, that audio 10 Khz brick wall filter is a little wide for 160 or 75 meters! 

Looking at your solid state RF driver, I think going to 75M will be a challenge. I'd either replace the driver completely, or just move it on up to 160m..

Jim
WD5JKO

 

[WB6QED]

This morning I spoke briefly with Charlie Goodrich at Goodrich Enterprises who was involved in the design of the McMartin BA-1K which originally used 4-500 tetrodes for the PA and modulators.  The BA-1K uses a screen voltage circuit topology that is similiar to that of the Collins 820D-2.  To confirm my suspicion, I asked him if the modulation performance of the BA-1K at 250 watts carrier was marginal and he said that it was.

Collins understood the benefits of using the 5-500A pentodes and took advantage of them with their implementation in the 820D-1 and the 820D-2.  To my knowledge, these are the only AM broadcast transmitters to have used this unique beam power tube.

Mike Dorrough, KO6NM, has successfully converted the 820D-2's RF driver stage to 75 meters.  He is going to provide me with some of the details when he has an opportunity.

In the meantime, I am going to purchase a couple 100 pF and 500 pF capacitors to use as substitutes for the two 1000 pF PA screen bypass capacitors to try to improve the harmonic distortion performance at 10 kHz.  Once I have the new 15 H screen choke installed, I will try adjusting the PA bias to see if I can improve the signal to noise ratio slightly.

The thing that strikes me the most about the results that I have achieved so far is that when everything is dialed in, these older plate modulated rigs have a very clean RF output spectrum and very good audio performance!  Certainly, they have their shortcomings, but I have done a lot of AM-NRSC measurements on various AM broadcast transmitters over the years and only a few look better than the 820D-2 does now.

Stay tuned!

Brian, WB6QED

[k4kyv]

I find the 700 ohm adjustable 300 watt resistor in series with the 50 Henry modulation reactor interesting. What is that used for?

It is used to adjust the plate voltage of the PA.

   Interesting how they reduce plate voltage in such a lossy way. Since it is there, why not move it to drop voltage on the modulated B+ line instead of the unmodulated B+ line? Then bypass it with a suitable sized oil capacitor. The idea is similar to what we do with a Heising modulation choke system that can only modulate upward to 80% or so. Here we use a DC dropping resistor to drop the modulated B+ line 20% so that we can then modulate the RF final to 100%.

Just an idea...

Jim
WD5JKO

What would be the advantage of doing it that way?  Dropping the voltage on the unmodulated B+ line eliminates the need for the bypass oil capacitor, but otherwise the loss is the same.  My Gates BC1-T does it the same way.  I replaced the rheostat with a bank of fixed WW resistors to drop the 2600 volts on the final down to about 2300 volts, with the full voltage still on the modulator. This gives better positive peak capability.  Performance wise, I don't think it makes any difference whether the voltage dropping is done at the modulated B+ line or unmodulated, but I think it is best to always  choose the  simpler circuit with fewer components whenever possible. If it is working properly, the only loss is the DC voltage drop in the resistor(s).