I want to adjust to time series, sample by sample. Assuming that one series could be a stretched and/or compressed version for the other one, also that there could be noise in both series (consistent and inconsistent).
I'm currently looking into dynamic time warping, but I'm wondering if there is a more sophisticated method do this.
Hi André, from my point of view (as an author of a dynamic time warping manuscript), the time warping method is quite sophisticated. Nevertheless there are other methods to consider. If there are enough features in the time series you could try a windowed cross-correlation with a sliding window. Here, noise could be treated by a smoothing algorithm (e.g. simple moving average). You could also try a deconvolution algorithm - but this is usually applied for time series on the whole and of course not sample by sample.
Hi Andreas,
Thank you for your answer.
In fact I'm trying the Dynamic Time Warping (DTW), but as far as know, it only adjust the sample interval in time series (stretching and shrinking), but how to scale the amplitudes after the DTW?
Hi André,
yes, the time warping is, at least this is the way we applied it, just a method to adjust the transient time lag between the two time series. In our case this lag was the product we wanted to have. To make the process easier, we scaled the time series to zero mean and variance of 1. To be honest: I have no idea how to scale the amlitudes within this process.
Hi Andre! A more complicated approach that could give good results depending on the domain/problem is the combined use of a temporal abstraction method and a edition distance to allign the time series
Why do not make the regression of one sample as a function of the other?
You said that the one time series is the stretched or compressed time series of the other. So, this means you make time scaling only either enlarge or reduce the the time duration of the series. You do not made any amplitude scaling.Logically if an amplitude scaling is made it must be given such that one can reconstruct the original signal. If the sampling is uniform then the time scaling is uniform and one can get the time scale factor by measuring the time duration of the two series.
Best wishes
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Electrodes