The error to update the weights for LMS is given by:
e(n) = d(n) - y(n)
Assuming e as the error, d as the desired signal and y as the estimated output at given time step n.
How do we know the value of d(n) if we are using it real-time?
In a supervised setting, we would have samples for the desired output but how do we achieve it for real-time adaptive filtering?