Holographix: a percept-first codec and network substrate for Large Vision Models and Large Audio Models over UDP

HolographiX is a percept-first codec and UDP transport substrate for RGB images and PCM audio operating on impaired links affected by loss, burst loss, jitter, reordering, duplication, and intermittent connectivity. Each signal is represented as a coarse approximation plus an int16 residual; the residual is permuted via a golden-step modular interleaving into a population of holographic chunks so that any subset of chunks contributes globally rather than to a specific local region. Decoding zero-fills missing residual samples and produces an anytime estimate from whatever chunks are available, avoiding all-or-nothing frame failures. Beyond graceful degradation under loss, HolographiX supports controlled refinement: independent partial encodings of the same scene can be fused (stacked) across time and across mesh paths to increase SNR and effective resolution, and the refined estimate can be re-encoded and redistributed as a new chunk population. On the network side, chunks are addressed by content identity (holo://object → content_id), framed over UDP, and replicated by a lightweight mesh. Evaluation is split between codec-level resilience, characterized by PSNR/SNR and subset variance as a function of the number of received chunks k, and full-stack end-to-end behavior on real protocol stacks using a reproducible container-based routing testbed (containerlab + FRR with tc/netem impairments), complemented by ns-3 scenarios for wireless and mobility. In addition to final reconstruction quality, we report perceptual continuity metrics such as time-to-first-useful-percept and area-under-quality curves under arrival-time impairment.