pith. sign in

arxiv: 2605.03214 · v1 · submitted 2026-05-04 · 📡 eess.SP

Canonical Optimization for MIMO MAC Design

Muhammad Umer , Muhammad Ahmed Mohsin , Ahsan Bilal , John M. Cioffi This is my paper

Pith reviewed 2026-05-08 17:12 UTC · model grok-4.3

classification 📡 eess.SP
keywords MIMO MACconvex optimizationcapacity regionresource allocationpolymatroidL-BFGSOFDMBC-MAC duality
0
0 comments X

The pith

The MIMO MAC admits canonical convex formulations with four solvers that characterize its capacity region.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that resource allocation for the MIMO multiple access channel, long treated as non-convex and intractable, actually reduces to convex problems. It presents four solvers—maxRMAC for weighted sum-rate maximization under per-user energy limits, minPMAC for minimum energy to meet target rates, maxRESMAC for sum-rate under total energy, and admMAC for feasibility testing—that together map the full capacity region. These solvers rely on the rate region's polymatroid structure and the separability of the dual Lagrangian across frequency tones, turning the task into independent per-tone covariance optimizations solved by limited-memory BFGS on Cholesky factors. Experiments on correlated MIMO OFDM channels confirm the solvers match commercial convex software in quality while running up to 100 times faster and handling larger instances. The same machinery transfers to the MIMO broadcast channel through duality.

Core claim

The MIMO MAC admits canonical convex formulations, and the four solvers maxRMAC, minPMAC, maxRESMAC, and admMAC together characterize its capacity region. They exploit the polymatroid structure of the MAC rate region and the separability of the dual Lagrangian across frequency tones to reduce each problem to parallel per-tone covariance optimizations solved via L-BFGS over Cholesky-like factors. The resulting algorithms match commercial convex solvers in solution quality, run up to two orders of magnitude faster, scale to regimes where commercial solvers time out, and extend via BC-MAC duality to optimal precoder design for the MIMO broadcast channel.

What carries the argument

Polymatroid structure of the MAC rate region together with separability of the dual Lagrangian across frequency tones, which reduces the optimization to independent per-tone covariance problems solved by L-BFGS over Cholesky-like covariance factors.

If this is right

  • The four solvers achieve solution quality identical to commercial convex solvers while running up to two orders of magnitude faster.
  • They remain tractable on problem sizes where commercial solvers exceed time limits.
  • The same four algorithms, via broadcast-to-MAC duality, yield optimal precoders for the MIMO broadcast channel.
  • Open-source code makes the methods directly usable on realistic spatially correlated OFDM channels.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The per-tone reduction may allow parallel hardware implementations for real-time adaptation in frequency-selective channels.
  • Similar polymatroid-plus-separability arguments could apply to other multiuser MIMO settings that share the same structural features.
  • The explicit convex forms reduce the need for machine-learning approximations once the canonical reduction is recognized.

Load-bearing premise

The MIMO MAC rate region possesses a polymatroid structure and its dual Lagrangian separates across frequency tones.

What would settle it

A spatially correlated MIMO OFDM channel instance where one of the four proposed solvers returns a noticeably different rate-energy point from a commercial interior-point solver, or where the proposed methods fail to finish while the commercial solver succeeds.

Figures

Figures reproduced from arXiv: 2605.03214 by Ahsan Bilal, John M. Cioffi, Muhammad Ahmed Mohsin, Muhammad Umer.

Figure 1
Figure 1. Figure 1: MIMO MAC system model. U transmitters with per-user antenna counts Lx,u communicate to a single receiver with Ly antennas across N¯ frequency tones. The receiver applies SIC in an order determined by the rate weights θ. optimal BC precoder design. II. PRELIMINARIES AND SYSTEM MODEL A. Channel model Consider a U-user MIMO MAC with Ly receive antennas and N¯ orthogonal frequency tones, as shown in view at source ↗
Figure 2
Figure 2. Figure 2: Relationships among the four canonical MAC solvers. MAXRMAC and MINPMAC are Lagrangian duals, where rate weights θ in MAXR￾MAC become dual variables in MINPMAC and energy multipliers w swap roles correspondingly. MAXRESMAC is a scalar-constraint specialization of MAXRMAC. ADMMAC calls MAXRMAC as a subroutine in an outer feasibility loop. TABLE I SUMMARY OF CANONICAL MAC SOLVERS. Solver Problem Outer method… view at source ↗
Figure 5
Figure 5. Figure 5: Time-sharing structure (U = 3, Ly = 2, Lx,u = 1, frequency￾selective channel). (left) Time-sharing probability vs. N¯ and loading factor ρ. (right) Dominant fraction αmax (left axis) and time-sharing probability (right axis) at ρ= 0.85. Time-sharing vanishes as N¯ grows. that WMMSE and SCA saturate below the optimum at high SNR, with a gap that widens under asymmetric weights as the non-convexity of their … view at source ↗
Figure 4
Figure 4. Figure 4: Minimum weighted energy vs. target sum-rate. (left) Equal per-user rate allocation. (right) Asymmetric allocation bu ∝ [4, 2, 1, 0.5]. MINPMAC matches the commercial solver; all baselines require strictly more energy. IV. EXPERIMENTAL RESULTS A. Experimental setup All experiments consider a MIMO orthogonal frequency￾division multiplexing (OFDM) uplink with U transmitters sharing a common Ly-antenna receive… view at source ↗
Figure 6
Figure 6. Figure 6: Runtime vs. problem size at SNR = 15 dB with i.i.d. Rayleigh channels. Sweeps over U, N¯, Ly, and Lx,u; open markers denote timeouts. The proposed solvers are 10 to 100× faster than the commercial solver across all dimensions. b1 (bpcu) 0 50 100 150 200 b2 (b p c u) 0 50 100 150 200 sum-rate decode user 2 first decode user 1 first view at source ↗
Figure 7
Figure 7. Figure 7: MAC capacity region traced by ADMMAC for a two-user channel with Ly = 4, Lx,u = 2, N¯ = 16, and SNR = 15 dB. The two corner points correspond to the two SIC orderings, and the intervening face is recovered by repeated feasibility tests. around 1 s. The per-iteration cost of the L-BFGS inner loop is O(N¯(UL2 xL 2 y + U 2L 4 x )), multiplied by O(U 2 log(1/ϵ)) ellipsoid iterations for the outer loop. SCA and… view at source ↗
read the original abstract

Resource allocation in the multiple-input multiple-output (MIMO) multiple access channel (MAC) is a fundamental problem in multiuser communications, yet it is increasingly treated as non-convex and computationally intractable. This has motivated a large body of heuristic machine learning and successive-approximation methods. Results here show that the MIMO MAC admits canonical convex formulations and present four solvers that together characterize its capacity region. maxRMAC performs weighted sum-rate maximization under per-user energy constraints, minPMAC finds the minimum weighted energy required to support target rates, maxRESMAC performs weighted sum-rate maximization under a total energy constraint, and admMAC tests rate-region feasibility. The solvers exploit the polymatroid structure of the MAC rate region and the separability of the dual Lagrangian across frequency tones, which reduces the problem to parallel per-tone covariance optimizations solved via limited-memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) over Cholesky-like covariance factors. Experiments on spatially correlated MIMO orthogonal frequency-division multiplexing (OFDM) channels show that the proposed solvers match a commercial convex solver in solution quality while running up to two orders of magnitude faster and scaling to regimes where the commercial solver times out. Through broadcast channel (BC) to MAC duality, the same solvers also enable optimal precoder design for the MIMO BC. All solvers are open-sourced and available at https://github.com/muhd-umer/canonical-mac.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 3 minor

Summary. The manuscript claims that the MIMO MAC admits canonical convex formulations via its polymatroid structure and the separability of the dual Lagrangian across OFDM frequency tones. It introduces four solvers (maxRMAC for weighted sum-rate maximization under per-user energy constraints, minPMAC for minimum weighted energy to meet target rates, maxRESMAC for weighted sum-rate under total energy, and admMAC for feasibility testing) that reduce to parallel per-tone covariance optimizations solved by L-BFGS on Cholesky factors. Experiments on spatially correlated MIMO-OFDM channels demonstrate that these solvers match a commercial convex solver in solution quality while running up to two orders of magnitude faster and scaling to larger regimes. The approach extends to MIMO BC precoder design via duality, with all code open-sourced.

Significance. If the results hold, the work is significant because it establishes that MIMO MAC resource allocation admits exact convex formulations and efficient solvers, countering the trend toward heuristics and successive approximations. The explicit use of polymatroid properties and dual separability, combined with reproducible open-source implementations and demonstrated scalability on realistic correlated channels, provides practical tools for capacity region characterization and BC design. This could reduce reliance on approximate methods in multiuser communications.

minor comments (3)
  1. Abstract: The statement that solvers 'match a commercial convex solver in solution quality' would be clearer if it referenced the specific metric (e.g., rate error or duality gap) and the commercial solver name used in the experiments section.
  2. The description of the L-BFGS implementation over Cholesky-like factors in the solvers section would benefit from an explicit statement of the variable substitution and any projection steps to enforce positive-semidefiniteness.
  3. Experiments section: Timing comparisons would be more informative with details on the hardware platform, number of Monte Carlo trials, and the precise commercial solver configuration (e.g., CVXPY with MOSEK or SDPT3).

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive and accurate summary of our work, the recognition of its significance in providing exact convex formulations and scalable solvers for the MIMO MAC, and the recommendation for minor revision. The referee's description correctly captures the polymatroid structure, dual separability, L-BFGS-based per-tone optimizations, experimental results on correlated channels, BC duality extension, and open-source release.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The derivation rests on the standard polymatroid structure of the MIMO MAC rate region (subset mutual-information inequalities) and the additive separability of the dual Lagrangian over OFDM tones once per-user energy multipliers are fixed. These are classical results from multiuser information theory, not self-derived or fitted within the paper. The four solvers reduce to independent per-tone convex covariance optimizations solved by L-BFGS on Cholesky factors; no step renames a fitted input as a prediction, imports uniqueness via self-citation, or smuggles an ansatz. BC-MAC duality is likewise external. The chain is therefore self-contained against external benchmarks with no load-bearing reduction to its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard domain assumptions from information theory and convex optimization with no free parameters or new invented entities.

axioms (2)
  • domain assumption The MIMO MAC rate region is a polymatroid.
    Invoked to enable the convex formulations and solver structure.
  • domain assumption The dual Lagrangian separates across frequency tones in OFDM.
    Allows reduction to parallel per-tone covariance optimizations.

pith-pipeline@v0.9.0 · 5560 in / 1259 out tokens · 20053 ms · 2026-05-08T17:12:05.303704+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

13 extracted references · 13 canonical work pages

  1. [1]

    Multiaccess fading channels. I. Polymatroid structure, optimal resource allocation and throughput capacities,

    D. Tse and S. Hanly, “Multiaccess fading channels. I. Polymatroid structure, optimal resource allocation and throughput capacities,”IEEE Trans. Inf. Theory, vol. 44, no. 7, pp. 2796–2815, 1998

    work page 1998

  2. [2]

    Iterative water-filling for gaussian vector multiple-access channels,

    W. Yu, W. Rhee, S. Boyd, and J. Cioffi, “Iterative water-filling for gaussian vector multiple-access channels,”IEEE Trans. Inf. Theory, vol. 50, no. 1, pp. 145–152, 2004

    work page 2004

  3. [3]

    Optimized transmission for fading multiple-access and broadcast channels with multiple antennas,

    M. Mohseni, R. Zhang, and J. Cioffi, “Optimized transmission for fading multiple-access and broadcast channels with multiple antennas,”IEEE J. Sel. Areas Commun., vol. 24, no. 8, pp. 1627–1639, 2006

    work page 2006

  4. [4]

    On the duality of Gaus- sian multiple-access and broadcast channels,

    N. Jindal, S. Vishwanath, and A. Goldsmith, “On the duality of Gaus- sian multiple-access and broadcast channels,”IEEE Trans. Inf. Theory, vol. 50, no. 5, pp. 768–783, 2004

    work page 2004

  5. [5]

    Sum capacity of Gaussian vector broadcast channels,

    W. Yu and J. Cioffi, “Sum capacity of Gaussian vector broadcast channels,”IEEE Trans. Inf. Theory, vol. 50, no. 9, pp. 1875–1892, 2004

    work page 2004

  6. [6]

    Spectrum- efficient user grouping and resource allocation based on deep reinforce- ment learning for mmWave massive MIMO-NOMA systems,

    M. Wang, X. Liu, F. Wang, Y . Liu, T. Qiu, and M. Jin, “Spectrum- efficient user grouping and resource allocation based on deep reinforce- ment learning for mmWave massive MIMO-NOMA systems,”Sci. Rep., vol. 14, no. 1, p. 8884, 2024

    work page 2024

  7. [7]

    A survey of deep learning based NOMA: State of the art, key aspects, open challenges and future trends,

    S. A. H. Mohsan, Y . Li, A. V . Shvetsov, J. Varela-Ald´as, S. M. Mostafa, and A. Elfikky, “A survey of deep learning based NOMA: State of the art, key aspects, open challenges and future trends,”Sensors, vol. 23, no. 6, p. 2946, 2023

    work page 2023

  8. [8]

    Weighted sum-rate maximization using weighted MMSE for MIMO- BC beamforming design,

    S. S. Christensen, R. Agarwal, E. de Carvalho, and J. M. Cioffi, “Weighted sum-rate maximization using weighted MMSE for MIMO- BC beamforming design,” inIEEE Int. Conf. Commun. (ICC), pp. 1–6, 2009

    work page 2009

  9. [9]

    An iterative water-filling algorithm for maximum weighted sum-rate of gaussian MIMO-BC,

    M. Kobayashi and G. Caire, “An iterative water-filling algorithm for maximum weighted sum-rate of gaussian MIMO-BC,”IEEE J. Sel. Areas Commun., vol. 24, no. 8, pp. 1640–1646, 2006

    work page 2006

  10. [10]

    On the limited memory BFGS method for large scale optimization,

    D. C. Liu and J. Nocedal, “On the limited memory BFGS method for large scale optimization,”Math. Program., vol. 45, no. 1, pp. 503–528, 1989

    work page 1989

  11. [11]

    Boyd and L

    S. Boyd and L. Vandenberghe,Convex Optimization. Cambridge Uni- versity Press, 2004

    work page 2004

  12. [12]

    3GPP TR 38.901 channel model,

    Q. Zhu, C.-X. Wang, B. Hua, K. Mao, S. Jiang, and M. Yao, “3GPP TR 38.901 channel model,” inThe Wiley 5G Ref.: The Essent. 5G Ref. Online, pp. 1–35, Wiley Press, 2021

    work page 2021

  13. [13]

    J. M. Cioffi,Data Transmission Theory. Stanford University, 2026. Book chapters available at:cioffi-group.stanford.edu

    work page 2026