Hello, I have a general question: Does anyone know of any algorithms for segmenting a discrete signal into regions which are "predictable" and "non-predictable" or stationary and non-stationary. By predictability, I mean, for example, predictability using a predictive FIR filter. Any other suggestion would be highly appreciated. Thanks much. Oz.
suggestions on signal segmentation
Started by ●January 12, 2012
Reply by ●January 12, 20122012-01-12
doublehelics wrote:> Hello, I have a general question: Does anyone know of any algorithms for > segmenting a discrete signal into regions which are "predictable" and > "non-predictable" or stationary and non-stationary. By predictability, I > mean, for example, predictability using a predictive FIR filter.If prediction error is lower then threshold, then signal is predictable.> Any other > suggestion would be highly appreciated. Thanks much.How much is your high appreciation ?
Reply by ●January 12, 20122012-01-12
On 1/12/2012 1:07 AM, doublehelics wrote:> Hello, I have a general question: Does anyone know of any algorithms for > segmenting a discrete signal into regions which are "predictable" and > "non-predictable" or stationary and non-stationary. By predictability, I > mean, for example, predictability using a predictive FIR filter. Any other > suggestion would be highly appreciated. Thanks much.Well, I have a specific question: what is a predictive FIR filter? Could one tell me the outcome of the next race at Hyaleah? Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●January 13, 20122012-01-13
On 1/12/2012 8:31 AM, Jerry Avins wrote:> On 1/12/2012 1:07 AM, doublehelics wrote: >> Hello, I have a general question: Does anyone know of any algorithms for >> segmenting a discrete signal into regions which are "predictable" and >> "non-predictable" or stationary and non-stationary. By predictability, I >> mean, for example, predictability using a predictive FIR filter. Any >> other >> suggestion would be highly appreciated. Thanks much. > > Well, I have a specific question: what is a predictive FIR filter? Could > one tell me the outcome of the next race at Hyaleah? > > JerryThe drag races on Hyaleah Rd. aren't scheduled AFAIK. Otherwise, there are NO races at Hyaleah. But there are sanctioned drag races at 19999 Okeechobee Rd. :-)
Reply by ●January 13, 20122012-01-13
*Hialeah. Ozgun.>On 1/12/2012 8:31 AM, Jerry Avins wrote: >> On 1/12/2012 1:07 AM, doublehelics wrote: >>> Hello, I have a general question: Does anyone know of any algorithmsfor>>> segmenting a discrete signal into regions which are "predictable" and >>> "non-predictable" or stationary and non-stationary. By predictability,I>>> mean, for example, predictability using a predictive FIR filter. Any >>> other >>> suggestion would be highly appreciated. Thanks much. >> >> Well, I have a specific question: what is a predictive FIR filter?Could>> one tell me the outcome of the next race at Hyaleah? >> >> Jerry > >The drag races on Hyaleah Rd. aren't scheduled AFAIK. >Otherwise, there are NO races at Hyaleah. >But there are sanctioned drag races at 19999 Okeechobee Rd. > >:-) > > >
Reply by ●January 13, 20122012-01-13
On Jan 12, 7:07�pm, "doublehelics" <ozgun.harmanci@n_o_s_p_a_m.gmail.com> wrote:> Hello, I have a general question: Does anyone know of any algorithms for > segmenting a discrete signal into regions which are "predictable" and > "non-predictable" or stationary and non-stationary. By predictability, I > mean, for example, predictability using a predictive FIR filter. Any other > suggestion would be highly appreciated. Thanks much. > > Oz.A test for whiteness. White noise cannot be predicted whereas coloured noise can. however, you will need a model of teh signal generation process in order to make your predictor. Perhaps a self-tuning predictor (Google that - goes way back - Wittenmark in the late 1970s). Hardy
Reply by ●January 13, 20122012-01-13
On Jan 13, 5:31�am, Jerry Avins <j...@ieee.org> wrote:> On 1/12/2012 1:07 AM, doublehelics wrote: > > > Hello, I have a general question: Does anyone know of any algorithms for > > segmenting a discrete signal into regions which are "predictable" and > > "non-predictable" or stationary and non-stationary. By predictability, I > > mean, for example, predictability using a predictive FIR filter. Any other > > suggestion would be highly appreciated. Thanks much. > > Well, I have a specific question: what is a predictive FIR filter? Could > one tell me the outcome of the next race at Hyaleah? > > Jerry > -- > Engineering is the art of making what you want from things you can get. > �����������������������������������������������������������������������Don't jest, the stock exchange uses similar methods every day of the week. Forecasting of time-series is an old "art". Hardy
Reply by ●January 13, 20122012-01-13
Great idea. Thanks a lot. I will look into the reference. Ozgun.>On Jan 12, 7:07=A0pm, "doublehelics" ><ozgun.harmanci@n_o_s_p_a_m.gmail.com> wrote: >> Hello, I have a general question: Does anyone know of any algorithmsfor>> segmenting a discrete signal into regions which are "predictable" and >> "non-predictable" or stationary and non-stationary. By predictability,I>> mean, for example, predictability using a predictive FIR filter. Anyothe=>r >> suggestion would be highly appreciated. Thanks much. >> >> Oz. > >A test for whiteness. White noise cannot be predicted whereas coloured >noise can. however, you will need a model of teh signal generation >process in order to make your predictor. Perhaps a self-tuning >predictor (Google that - goes way back - Wittenmark in the late >1970s). > > >Hardy >
Reply by ●January 15, 20122012-01-15
On Thu, 12 Jan 2012 00:07:48 -0600, "doublehelics" <ozgun.harmanci@n_o_s_p_a_m.gmail.com> wrote:>Hello, I have a general question: Does anyone know of any algorithms for >segmenting a discrete signal into regions which are "predictable" and >"non-predictable" or stationary and non-stationary. By predictability, I >mean, for example, predictability using a predictive FIR filter. Any other >suggestion would be highly appreciated. Thanks much. > >Oz.I remember a book I believe might help you. Its application section was mainly about segmentation into stationary segments for time series processing. Had a few pseudo code examples. http://www.amazon.com/Time-Frequency-Representations-Numerical-Harmonic-Analysis/dp/0817639187/ref=sr_1_28?s=books&ie=UTF8&qid=1326665560&sr=1-28 Mark DeArman
Reply by ●January 15, 20122012-01-15
Oz, it's kinda interesting how many people danced around an answer. i thought what you are asking about is Linear Predictive Theory. in basic LPC theory, given some statistical information like the power spectrum or auto-correlation of the data you're trying to predict with, it can come up with an FIR filter that you can predict the following sample with some sorta error metric (i think mean-square) that gets minimized, according to the auto-correlation of the input. if the input is white noise, the auto-correlation is a discrete impulse function and LPC can't really make a guess other than the latest sample less a little gain if the input is known to be DC free (but white everywhere else below Nyquist). i dunno. could be completely wrong about it. maybe you wanna predict farther into the future. i s'pose a sorta complete Kalman filter theory (which i don't know anything about since my skool daze) has some way to exchange prediction accuracy for the amount of time it's predicting. i remember it went the other way. the more delay you allowed a Kalman filter to estimate a signal (or, more specifically, the states that define the signal), the lower the prediction error. but somebody else has to show you how a Kalman filter works. i don't really remember anymore. r b-j On 1/13/12 1:03 PM, doublehelics wrote:> Great idea. Thanks a lot. I will look into the reference. > > Ozgun. > > >> On Jan 12, 7:07=A0pm, "doublehelics" >> <ozgun.harmanci@n_o_s_p_a_m.gmail.com> wrote: >>> Hello, I have a general question: Does anyone know of any algorithms for >>> segmenting a discrete signal into regions which are "predictable" and >>> "non-predictable" or stationary and non-stationary. By predictability, >>> I mean, for example, predictability using a predictive FIR filter. Any >>> other suggestion would be highly appreciated. Thanks much. >>> >>> Oz. >> >> A test for whiteness. White noise cannot be predicted whereas coloured >> noise can. however, you will need a model of teh signal generation >> process in order to make your predictor. Perhaps a self-tuning >> predictor (Google that - goes way back - Wittenmark in the late >> 1970s). >> >> >> Hardy >>
