timeencoding

Timeencoding refers to representing information primarily by the timing of discrete events or spikes, rather than by sampled amplitudes. In electronics and biology, time-encoding machines convert an analog input signal into a sequence of time stamps that mark when a threshold is crossed by an internal dynamical process. A common implementation uses a leaky integrator that accumulates input; when the integrator's state reaches a threshold, a spike is emitted and the internal state is reset. The resulting spike times t1, t2, ... encode the input signal, with shorter inter-spike intervals corresponding to larger signal values.

Various TEMs exist: asynchronous time encoding uses event-based spikes without a global clock, while clocked variants rely on a periodic reference to regulate spike generation. In neuroscience, time encoding is analogous to how some neural circuits encode information in spike trains. In electronics, time encoding is used in sensors and neuromorphic systems, and relates to time-to-digital conversion and event-based vision.

Decoding reconstructs the original signal from the spike train using the known encoding dynamics, often through inverse methods or filter banks. Advantages include potential high effective bandwidth for sparse signals and robustness to amplitude quantization, and natural integration with asynchronous processing. Limitations include decoding complexity, sensitivity to model mismatch, and inefficiency for signals that are already well represented by amplitude samples.