Re: Entropy -What is it?
I agree, that entropy reveals the extent of randomness and uncertainty of the system. Nevertheless, if you pay attention to the definition of this notion in Matlab, concerning signals, you'll see some strange things:
Shannon Entropy: E = sum (si^2*log2(si^2)) - Matlab's definition. si - samples of the discrete-time signals
1) according to the expression above, entriopy can be negative (less than zero) whereas this facts contradicts to the intuitive sense of entropy and the definition of it in information theory which maintain that entropy>=0!!!!
2) Imagine a constant signal, with the values very large and equal for any time sample. It will have enormous entropy, but it's easily predictive (because it's a constant) therefore should have small entropy. However, fast oscillation noise with zero mean and unit standard deviation must have the entropy larger than the constant signal. But according to the formula it's vice versa!!!!
How can you explain this curious facts?
With respect.
Dmitrij