Adaptive noise cancellation

I am working on a project where i am trying to remove noise from signal in
real time.

I have decided to use adaptive LMS algorithm.

If x(n): input signal corrupted with noise
y(n): output from filter;
e(n): the error signal that controls the weights of filter

Now do i need to generate another local noise source and feed it to the
filter?

Does the corrupted signal go directly into the filter or only the e(n) and
the locally generated noise(refrence signal) go into the filter?

Does the filter work just like it does in a linear predictor mode?

Is the output of filter (called anti-noise) add up with the corrupted
signal to produce e(n)? Is this e(n) the output (or the noise free
signal)?

Hello derif,

there is an article [1] of Bernard Widrow "Adaptive Noise Cancelling:
Principles and Applications". In Chapter "F. Cancelling Periodic
Interference without an External Reference Source" he explains this
problem.
Right before the FIR filter a fixed delay is inserted. This allows to have
correlated signals (eg. single tone) separated from uncorrelated (eg.
noise).

If your 'noise' signal is known you may generate it yourself and insert
this signal as usual in an LMS algorithm in the reference input.

Regards
Alois

[1] http://www-isl.stanford.edu/~widrow/publications.html , Papers nr. 31