I'm sure VK3YE has more experience at DSB projects than I have!
My problem was actually with an SSB rig, when I replaced the
RF board on my Ten-Tec Argosy with one from the newer digital
version. The transmit mixer had changed from an LM1496 to a
diode ring, and the existing VFO didn't like the lower impedance
it had to drive. I got several complaints of "bad audio", but it took
me several years to get around to tracking down the problem since
I work mostly CW. Turns out the VFO output was saturating due to
the heavy loading. I had a great idea for an AGC-controlled RF
amp to maintain constant injection level on the mixer, but settled
for a simple 4 : 1 transformer to provide a better impedance match
between the stages.
One of the reasons why you see several stages of amplification
following the DSB generator is because they have to be linear, and
will typically have less gain than the class-C stages usually used
in CW transmitters. The low-level stages will probably be class-A,
which means significant current draw even when there is no signal.
Overdriving, clipping, or any other non-linearities in the amplifier
stages after the modulator will generate a wide, "garbage" signal.
If you are "borrowing" the RF stages from another design, make sure
they are designed for SSB instead of CW. (But SSB rigs typically
have some IF amplification in addition to the RF stages.)