Quote:
Originally Posted by e*clipse
Ok, so if the switching frequency is 8kHz, that 800Hz sine wave would be broken into 10 pieces - right? It seems to me - and this is really a guess - that breaking the sine wave into only 10 parts would be rather rough, kind of like 6 step switching. Are there any efficiency advantages to a closer approximation? Also, this is a 12,000 RPM - much more normal speeds would be half that. Running at 16kHz would double that resolution - how would the motor respond?
Also, this is merely the switching frequency - how many AtoD samples of the rotor position and current should be done per swich, just to get an average or clean data?
|
The frequency change would be seamless to the motor. When I was doing the hissing spread spectrum switching with the DC motor, it was completely indistinguishable as far as how the acceleration and constant speed felt, even though the switching frequency was randomly spread across like 7kHz to 12KHz, changing every cycle. You could change to 16KHz once the rpm was above a particular value. Then, the added heat from the switching losses would only be under the unusual circumstance of 12,000 RPM. You could even have the switching frequency be set by the RPM. Maybe some minimum value of 8KHz for 0rpm - ??? rpm, and then linear ramp of frequency from 8KHz to whateverKHz from RPM_START_RAMP to RPM_MAX.
The advantage of doing a single A/D read sync'ed to the PWM period is that the measurements take place when the IGBTs are not switching (center aligned mode). So, you are always taking a current snapshot at the same point in the pwm duty waveform. So there is no concern about noise from the IGBT switching on reading #1, no igbt switching noise on reading #2, and averaging apples to oranges.... And these LEM swiss sensors are pretty dang quiet. I did a buck converter charger that was using a single current sample and got 0.2% accuracy for commanding the output current over most of the current range. Their datasheet says 1%.