R help - Jan 2006 - Selecting significant peaks in periodograms

Greetings all,
 
 I am using Fourier analysis to search for periodicities in IP network traffic
by generating periodograms and then visually examining them for large, distinct
peaks.
 
 However, in many cases it is not readily apparent where there are
periodicities.  I have no experience with discrete maths so I've come up
against a block here: How do I define what the "noise floor" is and
what peaks rising above it are significant enough to warrant further
investigation?
 
 I had thought to try to detect peaks as outliers by using confidence intervals
(assuming that the "Power" vector was normally distributed) but
I'm not sure if this is statistically valid.  If anyone can provide help, or
point me to some resources on the subject, then I'd appreciate it.
 
 Incidentally, I have tried to use other methods (the Lomb-Scargle method in
particular) but haven't found any especially well-suited to the problem.
 
 Best regards,
 Pete
 

		
---------------------------------

	[[alternative HTML version deleted]]

I just got 2 hits for 'RSiteSearch("significant peaks in 
periodogram")', the first of which was 
"http://finzi.psych.upenn.edu/R/Rhelp02a/archive/61482.html".  If you 
would like more help from this group, PLEASE do read the posting guide! 
"www.R-project.org/posting-guide.html".  Anecdotal evidence suggests 
that posts more consistent with that guide tend to get more useful 
replies quicker.

	  spencer graves

Pete Cap wrote:
> Greetings all,
>  
>  I am using Fourier analysis to search for periodicities in IP 
network traffic by generating periodograms and then visually
examining them for large, distinct peaks.
>  
>  However, in many cases it is not readily apparent where there 
are periodicities.  I have no experience with discrete maths so
I've come up against a block here: How do I define what the
"noise floor" is and what peaks rising above it are significant
enough to warrant further investigation?
>  
>  I had thought to try to detect peaks as outliers by using 
confidence intervals (assuming that the "Power" vector was
normally distributed) but I'm not sure if this is statistically
valid.  If anyone can provide help, or point me to some
resources on the subject, then I'd appreciate it.
>  
>  Incidentally, I have tried to use other methods (the 
Lomb-Scargle method in particular) but haven't found any
especially well-suited to the problem.
>  
>  Best regards,
>  Pete
>  
> 
> 		
> ---------------------------------
> 
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide!
http://www.R-project.org/posting-guide.html