How to recognize the signal?

On Thursday, September 12, 2013 11:06:37 AM UTC-4, LVitya wrote:
> >Run your signal through a slope filter
> 
> >
> 
> >See:
> 
> >
> 
> >http://www.claysturner.com/dsp/FIR_Regression.pdf
> 
> >
> 
> >Eqn 8 gives you the coefs for your filter.
> 
> >
> 
> >IHTH,
> 
> >
> 
> >Clay
> 
> >
> 
> >
> 
> 
> 
> I will try to do it in Matlab. But first I need to understand how to do it.
> 
> I'm not familiar with slope filtering.	 
> 
> 
> 
> _____________________________		
> 
> Posted through www.DSPRelated.com


Pick the length of the filter to be longer than the narrow spikes you are trying to ignore.

Clay

>> _____________________________		
>> 
>> Posted through www.DSPRelated.com
>
>
>Pick the length of the filter to be longer than the narrow spikes you are
trying to ignore.
>
>Clay
>

What narrow spikes you are saying about? Could you mark them on my images?	


_____________________________		
Posted through www.DSPRelated.com

On Friday, September 13, 2013 2:56:35 AM UTC-4, LVitya wrote:
> >> _____________________________		
> 
> >> 
> 
> >> Posted through www.DSPRelated.com
> 
> >
> 
> >
> 
> >Pick the length of the filter to be longer than the narrow spikes you are
> 
> trying to ignore.
> 
> >
> 
> >Clay
> 
> >
> 
> 
> 
> What narrow spikes you are saying about? Could you mark them on my images?	
> 
> 
> 
> 
> 
> _____________________________		
> 
> Posted through www.DSPRelated.com

Narrow is a relative term, but I'm referring to the spikes that are around 20 to 30 units wide on your graph. Let the slope filter span for example 50 units worth of data. By unit I'm referring to the scale of your horizontal axis.

I'm assuming what you want is the average slope of the portions in between the spikes.


Clay

Ok, Clay. I run it thru slope filter, which have coefficients from formula
(8), N = 50.
http://postimg.org/image/feycm1u8p/
But I can not understand how it can help.
	 

_____________________________		
Posted through www.DSPRelated.com

On Wednesday, September 11, 2013 2:16:54 PM UTC+2, LVitya wrote:
> I have a signal from sensor. It look like upper part of this pictures
> 
> http://postimg.org/image/l51neenup/
> 
> http://postimg.org/image/ddkxfujpd/
> 
> 
> 
> I need to recognize two oposite forms of signal: 1) when it goes up then
> 
> down; 2) when it goes down then up. For example
> 
> http://postimg.org/image/vxuhwk611/
> 
> On this picture 1st forms are marked, 2nd is unmarked.
> 
> 
> 
> I tried to use cross-correlation (xcorr in Matlab). The 2nd (reference)
> 
> signal for correlation looks like
> 
> http://postimg.org/image/8pq93yzyr/
> 
> It has 2nd form.
> 
> The results are presented on the lower part of pictures. It is useful in
> 
> most cases. But sometimes it is wrong. For example
> 
> http://postimg.org/image/f1kg2px7r/
> 
> On this waveform marked place have 2nd form, same as reference. So positive
> 
> value of correlation must be higner than negative. But it is not.
> 
> Does anyone have any suggestion how to make algoritm that always work
> 
> correct? It can be something different than correlation or maybe improvment
> 
> of this.

What is the variability in your signal? The general approach to such problems (indeed any problem) is Bayesian:

P(signal present | data, K) \propto P(data | signal present) P(signal present | K) ,

where K is the knowledge you have about the situation. You build probabilistic models summarizing this knowledge and then you calculate/compute. 

illywhacker;