Recursive least-squares (RLS). It's in all the books and countless times I have simulated white-noise or a square-wave going through a discrete-time system and used RLS to get the parameters of the model. Works fine. Has anybody used it for real however ie not a simulation. I tried it with a large dc motor. I drove a white-noise generator into PWM and into the dc motor and did tests. I used a quad-detector as the sensor and PWM for driving the motor. All I could get was an integrator representing the position to velocity part. The dominant time-constant of the system it couldn't see. When I put a differentiator to get a speed measurement I got the transfer-function of a differentiator but not the motor dynamics! Could it be that I need to measure the PWM signal and us that as the input rather than the white-noise? (or square-wave). I am beginning to wonder if it is one of these things that appears in books but nobody ever tries. I have used LMS countless times and it works, but not tried on a motor.