Had the same problem when "designed" a 3-phase bldc motor controller + amp with 629 back in late 80s, there was that constant pwm-noise due to the way how 629 dig-filter pll-pwm's and eventually was told by more experienced to add series L, this really reduced both the noise, and improved the motor performance radically, smoothing everything nicely. The small motor was too fast, andresponded too fast to the 629 "pwm-noise"while large motor would have self-lowpassedthe noise.
Found the serial L values moderately quick
experimentally, could have tried calclating
but bein lazy ...
Whole design, motor board, amp-board, and
writing software for 6803, a board stack of
2.5x5 inches,took less than 2 months, and
out of the 2 months, 90 % was the sw part,
few thousand lines of straight 6803 assembler + debugging it with piggy-back
rom 6803 plus L.A. , and tuning the pid
using F-V + scope was included in the total
time.
Still have the design and code somewhere,
but better designs can be done now 14
years later for sure.
The series L issue was one of the most
important, and another was using a look-up
table for allowing dynamic chancing of PID
values on the run, as the motor-pump
combo had non-linearities at low and high
rpm end to piece-wise-linearize the deal
which extended the dynamic range of the
motor compared to single fixed PID set for
all RPM range. Hope this is of any help.