Signals can be of different types like
Voice signals,
Vibration signals,
2d signals like picture signals,
Now my question is, how can I differentiate between the signals? How can I say one signal is different from another?
There are many techniques like that that are used in face recognition, or speaker recognition. So what are the best techniques to difference between signals.
I want to know the best methods irrespective of particular type of signal.
Try to use correlation between signals. it will highlight the major difference between any signals,
It depends on your application. For instance you could use spectrum analysis and build a fingerprint for each type of signal you are studying/using.
Another possibility is to use a simhash algorithm (http://matpalm.com/resemblance/simhash/) and check how similar the two signals are OR which are the most similar pairs among a set of signals from different sources.
Please note that, even if this algorithm has been developed for comparing text, it can also be used to compare sequences of ascii datasets, provided that the data files satisfy some conditions, e.g. same field structure and equal length.
This is exactly what I am doing now to compare the similarity of subsets randomly extracted from a main data file.
Dear Suman
the best method to identify your signal is to apply fourier transform, in which you know the frequency band.
Regards
First signal is a quantity which varies with an independent variable(eg. time) .
Any signal can be differentiated by fully understanding the nature of all of them.For example if it is a voice signal, it lies between 0.3 t0 3.4KHz. if it is vibration signal , it may have another spectrum.
Use directly the Fourier Transform to find out the spectrum of the signal under investation
Sorry, there is no best way to compare signals "irrespective of particular type of signal".
It always depends on your application domain and the type of underlying variations of your signals.
Examples: A very simple method is the correlation. But, if your signal varies with intensity and you are only interested in the curve you can use a normalized cross correlation. If your1d- signal is partly shifted you can use dynamic time warping. If you want to compare image patches you may want to consider rotational invariance. For 2d shapes you can compare Fourier descriptors. They can be invariant according to translation, scale and rotation of the shape.
... and so forth
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing