home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!spool.mu.edu!yale.edu!jvnc.net!rutgers!rochester!cantaloupe.srv.cs.cmu.edu!crabapple.srv.cs.cmu.edu!sundar
- From: sundar+@cs.cmu.edu (Sundar Vallinayagam)
- Newsgroups: sci.math
- Subject: Help! how to compute this derivative ?
- Summary: differentiating running sum frequency
- Keywords: derivative w.r.t. frequency
- Message-ID: <C053oF.FJ5.1@cs.cmu.edu>
- Date: 31 Dec 92 20:02:37 GMT
- Sender: news@cs.cmu.edu (Usenet News System)
- Organization: School of Computer Science, Carnegie Mellon
- Lines: 34
- Nntp-Posting-Host: speech1.cs.cmu.edu
-
- Howdy!
-
- I want to reach the minimum of an error surface by a gradient
- descent procedure. The error criterion is least-squared error,
- i.e., of the form ||y - X*inv(X'*X)*X'*y||.
-
- The k-th column of X would be: cos(2*pi*k*f*n) n=0,1,...,N-1.
- That is, the frequency is k*f. For a gradient descent procedure
- I need to calculate the derivative of the pseudoinverse, which needs
- the calculation of the derivative of the k-th column of X w.r.t. f.
- This is simply -2*pi*k*n*sin(2*pi*k*f*n), n=0,1,...,N-1.
- No problems with this simple case.
-
-
- Next consider the more general case where for the k-th column
- the frequency f is not fixed but is changing for each value of n,
- i.e., f now becomes f(n). I do not have a closed form expression
- for the frequency variation but have a set of N measurements.
- The k-th column of X now becomes: cos(2*pi*k*F(n))
- where F(n)--the phase--is the running sum of f(n)--the frequency.
- Now if I want to differentiate the k-th column w.r.t. frequency,
- how is the derivative defined and what would be a numerical procedure
- to implement it ?
-
- Thanks a lot in advance.
-
-
-
-
-
- --
- **************************************************
- ramli@orca.ele.uri.edu
- **************************************************
-