Murat,

I assume the motor is being driven by a linear amplifier, else you'd not want (need) to convert the PWM signal back to dc.

Since the signal to be filtered is inside the control loop, you want the ripple resulting from filtering to not disturb the controlled position, speed, torque or whatever is the controlled variable. If the mechanical stuff (inertia, friction, springs, ...) removes any ripple in the controlled variable resulting from ripple in motor torque, the loop performance passes muster. But if the amplifier and motor have overheating due to ripple caused by insufficient ripple, the PWM filter bandwidth should be decreased. Decrease it too much and the pole will increase phase lage in the loop and the loop will tend toward instability.

At 5kHz PWM and assumming an open loop bandwidth of, say, 5 Hz; why not try a single pole PWM filter at 50 Hz? Phase lag contributed by the single pole PWM filter to loop stability will be negligible (5 or 6 degrees phase lag) and the ripple will be about 40 dB down from the prefiltered PWM peak to peak variation. If the ripple is still too high, reduce the filter bandwidth in increments of about 5 Hz, watching for closed loop performance beginning to degrade. If you use anything other than a simple pole, be prepared to compute phase lag introduced by the filter and make some educated judgements about whether to apply it.

Good Luck,

Bruce