topic Re: Using Cross Correlation to determine time delay between two signals in SAS Forecasting and Econometrics

Using Cross Correlation to determine time delay between two signals

EBB — Fri, 02 Mar 2018 15:46:08 GMT

What I have is two signals in time (x=time in seconds, y=Force) and they are lagged (see graph bellow). The lag is unknown to me and I have more than 5000 of this pair of curves. I need to align all this pairs or curves to enable other posterior analysis.

To do that, I would normally use the cross correlation method to find the the point in time where the two signals are best aligned in seconds . After I have the this optimal set point, I can use this number to correct one of the time series and make them match.

The problem is that I never did this in sas and I was wondering if there is a procedure that I can use or I need to code this in sas base from scratch myself. I google quite a bit and I'm not sure if what I call a cross correlation could be done with proc timeseries, or proc ARIMA in EG or even with some TS node in EM.

I would take any recommendation regardless of it is the name of a procedure, a sas base program, or an em node.

Thanks!!

Re: Using Cross Correlation to determine time delay between two signals

Puwang — Fri, 02 Mar 2018 14:50:39 GMT

You can call PROC TIMESERIES to compute cross correlation of two time series. See the following for a simple code example:

proc timeseries data=sashelp.pricedata out=_NULL_ outcrosscorr=crosscorr;
 crosscorr _ALL_ / nlags=12;
 id date interval=month;
 var sale price;
 by region line product;
run;

You can also check http://support.sas.com/documentation/cdl/en/etsug/66840/HTML/default/viewer.htm#etsug_timeseries_syntax05.htm for more details on the PROC and related functionalities.

Re: Using Cross Correlation to determine time delay between two signals

mitrov — Fri, 02 Mar 2018 14:52:17 GMT

You can use PROC TIMESERIES to compute crosscorrelations. You need to use the CROSSCORR and CROSSVAR statements. Here is simple example.

data temp;
  call streaminit(123); 
  do i=1 to 103;
     x = rand("Uniform");
     y = lag3(x);
     if i >  3 then output;
  end;
  drop i;
run;

proc timeseries data = temp crossplot=all outcrosscorr=outcrosscorr;
   crosscorr lag n ccov ccf ccfstd;
   var x;
   crossvar y;
run;

I am not sure exactly what the purpose of your analysis is, but you may want to look also at the SIMILARITY procedure. It allows you to compute a distance matrix between series using time warping rather than cross-correlation.

Re: Using Cross Correlation to determine time delay between two signals

EBB — Fri, 02 Mar 2018 14:55:48 GMT

Thanks for that! I also thought about DTW but it does not fit to what I need to do afterwards with the data.

Re: Using Cross Correlation to determine time delay between two signals

mitrov — Fri, 02 Mar 2018 15:15:01 GMT

BTW, I forget to mention that if you use SAS Studio, the forecasting tasks provide an easy point-and-click interface that generates code for this type of analysis and much more.

I should put this sentence as my signature

Re: Using Cross Correlation to determine time delay between two signals

mitrov — Fri, 02 Mar 2018 15:33:58 GMT

Links:

Time Series and Forecasting tasks

SAS Studio

Time Series Data Preparation

Time Series Exploration

Time Series Modeling and Forecasting

Re: Using Cross Correlation to determine time delay between two signals

mitrov — Fri, 02 Mar 2018 16:21:10 GMT

Let's see how my signature looks now.