Feedback Capacity of Finite-State Communication Channels

Feedback capacity of finite-state communication channels quantifies the maximum rate at which information can be transmitted reliably when the transmitter observes past channel outputs. Unlike memoryless channels, finite-state channels feature a limited set of internal states evolving according to probabilistic rules. Feedback can profoundly influence capacity by enabling adaptive encoding strategies that track channel evolution, mitigate error propagation and exploit memory effects. Key insights include the realisation that feedback capacity often requires optimisation of input distributions over sequences rather than single symbols, and that directed information provides the appropriate metric for channels with memory and feedback. Foundational results for additive Gaussian noise channels and binary state-dependent channels have revealed both computable bounds and instances of non-computability. Contemporary work has focused on algorithmic characterisations of capacity-achieving processes, on sufficient-statistic representations that reduce complexity, and on extensions to channels with fading, side information and non-ergodic behaviour. These advances underscore the global significance of feedback in emerging applications—from wireless networks to data storage—where channel memory and adaptive coding coexist.

Research from Nature Portfolio

Recent studies have demonstrated that even simple finite-state channels with feedback can present uncomputable capacity limits when memory effects are intricate. One seminal investigation revealed that for modestly sized state-machine channels with four input symbols, one output symbol and limited memory, the feedback capacity cannot be approximated within a fixed precision. This work highlights inherent algorithmic barriers in determining exact capacity for channels whose state evolution is information-stable yet retains historical dependence. The findings have spurred a re-examination of which channel structures admit closed-form capacity expressions and which elude efficient computation despite the presence of feedback.

Feedback Capacity of Finite-State Communication Channels publication trend

The graph below shows the total number of articles in feedback capacity of finite-state communication channels across all publications each year (not limited to Nature Index journals).

Technical terms

Feedback capacity: The supremum of reliable transmission rates when past channel outputs are available to the encoder, tailored to channels with memory.

Finite-state channel: A communication model governed by a finite set of internal states that evolve probabilistically based on past inputs and outputs.

Directed information: An information measure that quantifies the causal flow of information from input to output in the presence of feedback.

Sufficient statistic: A summary function of past observations that retains all information necessary for optimal future encoding decisions.

References

1. Memory effects can make the transmission capability of a communication channel uncomputable. Nature Communications (2018).
2. Information Rates for Channels with Fading, Side Information and Adaptive Codewords. Entropy (2023).
3. New Formulas of Feedback Capacity for AGN Channels with Memory: A Time-Domain Sufficient Statistic Approach. Entropy (2025).