Post by thread:[PIC:] moving average filter

> Is there a way to do an n sample moving average without storing n > samples? No. > I want to do a 20 sample moving average. Do I need to set aside 20 > locations/variables for this task? Yes. > Is there a slicker way? Probably, depending on your application. It depends on why you want to perform this "average". If it is to reduce random noise there are definitely better ways. This has been discussed many times in depth. See the archives. ***************************************************************** Embed Inc, embedded system specialists in Littleton Massachusetts (978) 742-9014, http://www.embedinc.com -- http://www.piclist.com hint: To leave the PICList .....piclist-unsubscribe-requestKILLspam@spam@mitvma.mit.edu

> From: Robert Mash[SMTP:rmashKILLspamGLOBALCOOLING.COM] > Sent: Thursday, September 18, 2003 12:21 PM > To: .....PICLISTKILLspam.....MITVMA.MIT.EDU > Subject: [PIC:] moving average filter > Is there a way to do an n sample moving average without storing n samples? > I want to do a 20 sample moving average. Do I need to set aside 20 locations/variables for this task? > Is there a slicker way? >Robert Mash That depends on how closely you want to approximate the standard "mean" average. If you want a conventional "boxcar" integration, you must retain all 20 previous values so that you can subtract them one at a time in order as new terms are added. (The name "boxcar" comes from the rectangular shape of the window function that results from adding each term with an equal weighting factor.) If you are willing to implement an exponential filter (exactly analogous to an RC hardware filter), you can do much better. From Z transform theory, a one stage recursive filter can be modelled as y(n) = x(n) + alpha * y(n-1) y(n) is the current output, x(n) is the current input, y(n-1) is the previous output, and alpha is a constant that sets the time interval over which the filter "remembers" the input. If I figured this right, alpha = exp( - T / tau ) where T is the time interval between samples and tau is the effective time constant desired. Rescaling will be needed to find a relationship between tau and "20" (or whatever number is needed). This type of filter discounts earlier samples as time goes by, as opposed to the "mean (boxcar)" filter which treats all values equally right up until they drop out. There are several limitations to this filter. 1. It requires floating point. 2. Either alpha must be precomputed, or the exponential function must be available at run time. John Power -- http://www.piclist.com hint: PICList Posts must start with ONE topic: [PIC]:,[SX]:,[AVR]: ->uP ONLY! [EE]:,[OT]: ->Other [BUY]:,[AD]: ->Ads

John N. Power wrote: > If you are willing to implement an exponential filter (exactly > analogous to an RC hardware filter), you can do much better. From Z > transform theory, a one stage recursive filter can be modelled as > y(n) = x(n) + alpha * y(n-1) This definitely does NOT model an RC filter. For one thing, note that this is an infinitely increasing series even with a constant input and an ALPHA greater than 0. It performs no useful "filtering" function that I can think of. As has been discussed many times here before, the right way to model an RC filter is: FILT <-- FILT + ALPHA * (NEW - FILT) See the archives for many detailed discussions. ***************************************************************** Embed Inc, embedded system specialists in Littleton Massachusetts (978) 742-9014, http://www.embedinc.com -- http://www.piclist.com hint: The list server can filter out subtopics (like ads or off topics) for you. See http://www.piclist.com/#topics

>> Is there a way to do an n sample moving average without storing n samples? >> I want to do a 20 sample moving average. Do I need to set aside 20 locations/variables for this task? >> Is there a slicker way? >>Robert Mash > If you are willing to implement an exponential filter (exactly analogous to an RC hardware filter), > you can do much better. From Z transform theory, a one stage recursive filter can be modelled > as > y(n) = x(n) + alpha * y(n-1) I think I made a mistake in the normalization of this formula. It should be y(n) = (1 - alpha) * x(n) + alpha * y(n-1) John Power -- http://www.piclist.com hint: The PICList is archived three different ways. See http://www.piclist.com/#archives for details.

John N. Power wrote: > I think I made a mistake in the normalization of this formula. It > should be y(n) = (1 - alpha) * x(n) + alpha * y(n-1) Yes this correctly models a simple R/C filter, although it's generally not the way you'd want to implement such a filter on a PIC. ***************************************************************** Embed Inc, embedded system specialists in Littleton Massachusetts (978) 742-9014, http://www.embedinc.com -- http://www.piclist.com hint: The PICList is archived three different ways. See http://www.piclist.com/#archives for details.

plee (EraseMEpaul66spam_OUTTakeThisOuTprimus.ca) wrote: > saw your post. > > I need to design reading of ADXL202 g sensor. > As you surely know it has big noisy and a filter, or > an running avery is critical. > > I didn't find the explaination about your equation. > PICLIST is down. > > Could you help to find detail to decide the factor? > > Paul There is no reason this needed to be a private message to me. The PICLIST is never down long. I just fetched messages and the activity look fairly continuous since the last time I fetched. The longest gap between messages was only 40 minutes. As for the filter, what detail, what factor? ***************************************************************** Embed Inc, embedded system specialists in Littleton Massachusetts (978) 742-9014, http://www.embedinc.com -- http://www.piclist.com hint: The PICList is archived three different ways. See http://www.piclist.com/#archives for details.