Semantic Information Encoding in One Dimensional Time Domain Signals

Abstract

A one dimensional time domain analog signal s(t) can be visualized as a trajectory of a moving particle in a force field with one degree of freedom. Then the power of the particle at point t is P(s(t)) = s"(t)s'(t), which is the rate at which kinetic energy is dissipated (assuming the mass of the particle is unit) by the particle in order to create the trajectory or give shape to the signal. Assuming meaning of the signal or the semantic information is in its shape, we can say that P(s(t)) is the rate at which kinetic energy of the particle is dissipated to encode semantic information in s(t) at t. After s(t) is digitized (to make it s[n]) the discrete form P(s[n]) is valid. Considering the sign changes of P(s[n]) it has been shown that in the smallest neighborhood of n, in which n is the middle point, semantic information in s[n] can be encoded in 13 distinct ways. This list is exhaustive. A deterministic finite automaton (DFA) has been designed which can accept any finite length digital signal and therefore collection of all finite length digital signals forms a regular language. The DFA has been generalized to a weighted finite state transducer (WFST), which has been used to identify action potentials in a spike train and also to distinguish two speakers when uttering the same phoneme. It has been shown that in any analog signal semantic information can be encoded at a point in the form of the shape of its infinitesimal neighborhood in 17 distinct ways. The list is exhaustive. A new entropy measure called semantic entropy has been introduced. It has been shown that a signal s(t) is traceable on a piece of paper or in an oscilloscope, only if s"(t) exists on all but at most a finite number of points within any finite interval. This is an essential condition for a signal to be the trajectory of a moving particle.