If the idea is to set a definite power level and modulate in a definite direction, PWM DC coupled ought to do it.
An ancient technique of DC restoration as used in television transmitters and receivers of the same age as the tube manuals is another way to work the issue.
Here are some resources on DC restoration.https://www.americanradiohistory.com/Archive-Television-Engineering/Television-Engineering-1950-01.pdf Transmitter - page 22. The 4E27s are the DC coupled grid modulator for 4X150s. The DC restorer is before the 4E27s. The variable core inductors are for video equalization as TV is a DC to 7MHz wide signal -just see how the DC restorer is used in the transmitter.
newfangled IC stuff:
https://www.maximintegrated.com/en/app-notes/index.mvp/id/3303https://www.diodes.com/assets/Datasheets/ZXFV4089.pdfDC restoration in early TV signal processing:
https://www.americanradiohistory.com/Archive-Television-Engineering/Television-Engineering-1951-07.pdfpage 12 - article on video switching. DC restoration used with cathode follower.
https://www.americanradiohistory.com/Archive-Television-Engineering/Television-Engineering-1950-10.pdfVideo Amplifiers
TLDR part:Think about this in terms of an analog monochrome television signal and transmitter. Black level is near 100% RF, sync is a bit more. White level is low power, just enough to still have some RF there (see class B television amplifier service for tubes like the 4X250 and others). This convention was chosen because when a person looks at an analog TV image, noise is more noticeable in dark areas compared to bright ones. I'm not saying that these tube manual specs are for TV, but they are obviously for some sort of special use.
With essentially negative modulation, the possibility of grid modulation is palatable since the efficiency at zero mod is going to be high, and only reduced when speaking. This can also be done with a class C stage though the data looks like class B linear amp when TV service is mentioned. The same rule against overmodulation applies.
The only transmitter trouble I can think of right now in making the modulation essentially negative (with or without a 15% positive) is that a random duty cycle waveform created through the usual audio processing, coupled through a non saturated transformer or through a capacitor seems to always find its center line at about 50% on the other side, destroying any DC offset that would have made the modulation essentially negative. Perfect for where modulation is to be essentally well balanced and one is willing to switch polarity to get the little ast bit, but not good where essentially negative modulation is wanted.
What is needed in case the modulating circuit is not fully DC coupled throughout, is DC restoration, a technique used before PWM was practical, and it was used in video output amplifiers of CRT television, as well as anywhere a 'standardized' video signal is needed. The ideas work equally well in transmitters and receivers.
In the earliest days of television reception. it was satisfactory to just put the amplified video signal on the CRT cathode (TV CRTs were often cathode driven) and let loose.
The problem arose with quality improvements in the signals and in the 'effects' used for TV. In the earliest TV sets, if you put a statisitically small amount of white text on a black background, such as on a credits card in front of a TV camera, by the time this got to the receiver CRT and modulated the electron beam, you would see the black background was now gray, and the white was clipped in the CRT since the video output amp could not (and should not be allowed to) drive it into grid current.
The reason this happened is that the bright part of the image was small compared to the black part, so the AC coupled circuit tended to raise or lower the waveform DC level until it was about equal area under the curve if that makes any sense. In a balanced scene, the effect was that people's faces and other well it objects would look pasty as the highlights were compressed at the CRT.
Soon, the DC restorer circuit was being added to consumer TV sets. It's little more than a diode clipper, but at the low power levels used for driving a CRT grid-cathode circuit, about 1-3 Watts, there was little waste and costs were kept down. "My CRT is anode modulated", said no one ever.
Looking at diode type negative peak clippers used by some AM operators and imagining them inverted, we can see the similarity of result to DC restoration, but at a large cost in modulating power and distortion. The DC restorer circuit is designed for low distortion.
If you mod only negative then the other guy with a fast AGC is going to comment on his meter swinging backwards, but a normal or slow AGC is what is best. Does it really matter about meter swing? The RF output meter on the transmitter will swing backwards too, It's totally normal.
A possible benefit, just a supposition, is that when one is not speaking, interferers are more 'blocked out', and when one does speak, the negative-going modulation envelope would seem to also modulate QRM received. This way an interferer's signal is used against it, to fill up only your modulated envelope's negative peaks. May be a good countermeasure for all kinds of noises.