Place Cells, Time Cells, and Phase Coding as Special Cases of a Generic Spatiotemporal Oscillatory Representation

Hippocampal modeling has largely proceeded phenomenon by phenomenon: separate, specially-tuned mechanisms for place fields, for grid-like oscillatory interference, for time cells, and for theta phase precession. We take a different starting point. We train a Deep Oscillatory Neural Network (DONN) -- an architecture of alternating Hopf-oscillator and dense layers, with a theta-band oscillatory core -- on a single generic task: path-integrating a quadruped agent's position from limb oscillations and a visual cue, with no representation of place fields, time fields, or phase relationships imposed a priori. From the trained network we recover, as emergent properties, spatially and temporally selective units resembling place and time cells, and, more importantly, a population-level code in which position is linearly decodable from the phase of the hidden oscillators while velocity is linearly decodable from their frequency. Motivated by this observation, we outline a theory in which place cells, time cells, and phase-coupled phenomena such as phase precession are not separately engineered outcomes but limiting and mixed sectors of a single spatiotemporal oscillatory basis, indexed by a joint spatial-frequency ($K$) and temporal-frequency ($\Omega$) spectrum. We present this as an early-stage, testable framework rather than a completed proof, and report which parts are supported by our current results and which remain predictions.