Channel and question
- Input
- At time \(t\), each encoder uses its message and common past outputs \(Y^{t-1}\).
- Output
- \(Y=X_1+X_2+Z\), returned noiselessly to both encoders.
- Law
- \(Z\sim\mathcal N(0,N)\) iid, with powers \(P_1,P_2\).
- Quantity
- Feedback capacity region \(\mathcal C_{\mathrm{GMAC,fb}}\), measured in rate pairs in bits per channel use.
Criterion. Vanishing average joint decoding error. Resource averaging is fixed by the explicit model assumption below.
- Feedback is causal and noiseless.
- Messages are independent.
- For each user j, E_{M_1,M_2,Z^n}[sum_t x_{j,t}(M_j,Y^{t-1})^2] <= n P_j. The messages are independent and uniform. Every message-pair-conditional energy is integrable. The power bounds are separate and not pathwise.
Current status
| Result | Relation | Method | Year |
|---|---|---|---|
| Inner Region | \(\mathcal R_{\mathrm{Ozarow}}\subseteq\mathcal C_{\mathrm{fb}}\) | Linear feedback creates controlled correlation between users. | 1984 |
| Outer Region | \(\mathcal C_{\mathrm{fb}}\subseteq\mathcal R_{\mathrm{Ozarow}}\) | A dependence constraint and Gaussian extremality. | 1984 |
Formal verification
New statements await mathematical review and proofs. Power convention: SeparateMessageNoiseAveragePowerAdmissible. No equivalence to another power convention is assumed.
Claims
- Ozarow's capacity region under separate expected-power constraints and noiseless output feedback.
operational-capacity· exact capacity · solved · Formally stated · v1
Lean declarations (1)
CapacityAtlas.Claims.gaussianMACFeedbackclaim · operational-capacity
lean/CapacityAtlas/Claims/GaussianMACFeedback.lean — Ozarow's capacity region under separate expected-power constraints and noiseless output feedback.
References
- Lawrence H. Ozarow (1984). The Capacity of the White Gaussian Multiple Access Channel with Feedback. IEEE Transactions on Information Theory. DOI 10.1109/TIT.1984.1056935.
- Abbas El Gamal and Young-Han Kim (2011). Network Information Theory. Cambridge University Press. DOI 10.1017/CBO9781139030687.