Chapter Summary

Chapter Summary

Key Points

• 1.

  The Interference Channel. The capacity of the two-user interference channel remains the most famous open problem in information theory. The Han-Kobayashi scheme (rate splitting into common and private parts) is the best known inner bound. Etkin, Tse, and Wang showed that a simple HK variant achieves within 1 bit of capacity for all Gaussian IC parameters — the problem is "almost solved" for engineering purposes, but the exact capacity in the moderate interference regime remains open.

• 2.

  The General Relay Network. The cut-set bound provides an elegant outer bound for relay networks, but it is not always tight because it ignores causality constraints. Compress-forward achieves within 1/2 bit for the Gaussian relay channel. Noisy network coding generalizes this to multi-relay networks but the gap grows with network size. The exact capacity of even the simplest relay channel (one source, one relay, one destination) remains open in general.

• 3.

  Non-Shannon Inequalities. For four or more random variables, there exist valid information inequalities that cannot be derived from Shannon's axioms. The Zhang-Yeung inequality (1998) was the first such discovery. Non-Shannon inequalities have implications for network coding capacity and distributed source coding, making the characterization of multi-terminal problems fundamentally harder than expected.

• 4.

  Information Theory for 6G. ISAC (integrated sensing and communication) introduces the capacity-distortion tradeoff — the price of sensing in terms of communication rate. Near-field information theory with extremely large arrays opens new spatial multiplexing possibilities. RIS-aided communication and AI-native protocols are active research frontiers where information theory provides the fundamental limits.

• 5.

  Reading and Writing IT Papers. A capacity proof has two halves: achievability (random coding) and converse (Fano's inequality). The most common errors are conflating inner bounds with capacity, ignoring CSI assumptions, mistaking DoF for finite-SNR performance, and applying asymptotic results at finite blocklength. Critical reading starts with the assumptions, not the conclusions.