Zero Delay Coding for Real Time Streams

This direction is split into other directions: real-time communication constraints. Its last version stays readable and citable here.

Version 1 of 199 papers109 conceptsfrozen 2026-10-04

In classical information theory, reliability is a gift from the Law of Large Numbers. To communicate without errors, Claude Shannon taught us to group information into massive blocks. By averaging noise over a long period, we can achieve near-perfect transmission.

But this "block-coding" paradigm has a hidden cost: latency. If a transmitter must buffer thousands of bits before it even begins to encode, the system inherits a fundamental delay. For a file download, this is irrelevant. For a remote-controlled drone, a surgical robot, or a live interactive game, this delay is catastrophic.

A growing body of research into Anytime Coding and Zero-Delay Source-Channel Coding seeks to dismantle the block-based requirement, enabling communication that is both reliable and immediate.

The Conflict Between Blocks and Control

Traditional error correction requires a receiver to collect a complete batch of coded packets before it can reconstruct any part of the original data. This "all-or-nothing" decoding, as noted by Jalaluddin Qureshi and Chuan Heng Foh in Primer and Recent Developments on Fountain Codes, creates a waiting time that is incompatible with real-time feedback loops.

In networked control systems, the goal isn't just to move bits; it's to stabilize an unstable physical process. Anatoly Khina and Wael Halbawi demonstrate in (Almost) practical tree codes that if the communication delay is too high, the controller cannot react fast enough to prevent the system from spiraling out of control. To solve this, we need causal coding: a framework where each symbol is transmitted as it arrives, using only the history of the source.

Anytime Reliability: Errors that Decay with Delay

The breakthrough concept in this field is "Anytime Reliability," pioneered by Anant Sahai. In Anytime coding on the infinite bandwidth AWGN channel, Sahai proves that we don't need to wait for a block to end to gain reliability.

Instead of the error probability being a function of the block length $N$, it becomes a function of the delay $T$. A bit sent at time $t=1$ is decoded with some uncertainty at $t=2$. However, as the transmitter continues to send subsequent bits, it implicitly provides more evidence for that first bit. By the time $t=100$, the receiver's estimate of the first bit has become exponentially more reliable, even though the bit was never "retransmitted."

This is often achieved through Tree Codes. Unlike block codes, which map a message to a fixed-length codeword, tree codes map a sequence of bits to a path in a tree. As the path grows, the distance between the correct path and incorrect branches increases, allowing the decoder to "change its mind" and correct past mistakes as more data arrives.

Streaming Codes for Burst Erasures

While some researchers focus on noisy channels (AWGN), others like Ashish Khisti and Ahmed Badr focus on packet-erasure networks (like the internet). In Streaming Codes for Channels with Burst and Isolated Erasures, they introduce Streaming Codes (SCo).

The intuition here is to interleave data and parity in a way that respects a strict decoding deadline. If a packet is lost, the receiver must recover it within $T$ time steps, or it is useless. Khisti shows that by using non-systematic codes—where the original data isn't sent "as-is" but is instead mixed into every packet—we can achieve the optimal trade-off between throughput and the ability to recover from sudden bursts of losses.

Tracking the State of the World

When the "source" is a physical system (like a robot's position), the coding problem becomes a state estimation problem. Serdar Yuksel and Christoph Kawan have shown in Metric and topological entropy bounds for optimal coding of stochastic dynamical systems that the "complexity" of a system—its topological entropy—dictates the minimum data rate required to track it in real-time.

In this zero-delay setting, Takashi Tanaka and Sanjoy K. Mitter argue in Semidefinite Programming Approach to Gaussian Sequential Rate-Distortion Trade-Offs that the encoder must be "greedy" yet strategic. It must compress the current state as much as possible while considering how its current transmission will affect the receiver's future ability to estimate the system's trajectory.

The Future: Prediction and Semantics

The latest evolution in this field moves beyond pure information theory into the realm of AI. Osvaldo Simeone and colleagues, in Prediction-Powered Communication with Distortion Guarantees, suggest that we can use machine learning models to predict future data, allowing us to send only the "surprising" parts of a signal. By using online conformal prediction, they provide mathematical guarantees that the reconstruction error will stay within bounds, even when the AI's predictions are wrong.

This shift toward semantic and task-oriented communication represents the next frontier: coding not just for bits, but for the immediate, real-time actions those bits must trigger in the physical world.

Papers in this version

Showing 60 of 99 papers.

History & citationVersion 1 · 2026-10-04

Split into other directions · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_zero_delay_coding_for_real_time_streams_ece6cf3e,
  title = {Zero Delay Coding for Real Time Streams},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/zero-delay-coding-for-real-time-streams/rdr_ece6cf3ebbef42979648f2ea54fe7967},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_zero_delay_coding_for_real_time_streams_ece6cf3e_v1,
  title = {Zero Delay Coding for Real Time Streams},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/zero-delay-coding-for-real-time-streams/rdr_ece6cf3ebbef42979648f2ea54fe7967/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}