Apparently, Matlab's hmm functions are designed for a system with "discrete" emission values rather than continuously changing values. Is it true that in this case I'll need to find transition function rather than matrix?
If there is a good resource that I can use to learn how to find the hidden states of my signal, please let me know.
Can you be more precise about your signal ? Or, what do you mean by the hidden state ?
You can not acquire signals in continuous state. They must be sampled (made discrete) with a certain step (frequency).
I don't know of any way that would allow you storing a continuous signal with an infinite resolution. In that sense if your signal is stored then it's in a discrete format. Meaning, you can use matlab (or any other signal processing package) to analyse it.
G.
If you have a continuous valued state, observed at discrete sampling times, you should use a discrete-time Kalman filter or extended Kalman filter in the first instance (rather than a HMM). This is a classical MMSE state estimation technique for continuous state space dynamical systems observed in white Gaussian noise. In the simplest case, your signal model can be x(k+1) = x(k) + v(k) where v(k) is white noise. This is a 1-D random walk. You can bring in time dependence by making the state multi-dimensional (position and velocity, for example). Usually, you must supply the system model (transition matrix, measurement matrix) and the noise covariances. The Kalman filter then takes the noisy measurement sequence and filters it to give MMSE estimates of the states and their covariances.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis