Achievable DoF Bounds for Cache-Aided Asymmetric MIMO Communications

1. Problem Statement

The paper addresses the gap in Cache-Aided Multiple-Input Multiple-Output (MIMO) communications regarding asymmetric user configurations.

• Context: While Coded Caching (CC) has been extensively studied for Symmetric MIMO (where all users have the same number of receive antennas, $G$) and MISO (Single-antenna receivers) systems, real-world networks (e.g., 5G/6G) involve heterogeneous devices ranging from high-end smartphones to low-power IoT nodes. These devices possess different numbers of receive antennas ($G_k$).
• The Challenge: In asymmetric settings, existing symmetric strategies fail to optimize the trade-off between global caching gain (multicasting opportunities derived from shared cache content) and spatial multiplexing gain (parallel streams enabled by multiple antennas).
  • Treating all users as having the minimum antenna count ($min-G$) wastes the spatial capabilities of users with more antennas.
  • Treating users separately (Grouping) sacrifices global caching gain by failing to exploit the cumulative cache size across the entire network.
• Objective: To characterize the achievable single-shot Degrees of Freedom (DoF) for MIMO-CC systems with heterogeneous receive antennas and propose strategies that balance these competing gains.

2. Methodology

The authors propose a framework built upon three reference symmetric CC policies, adapted to asymmetric environments through four distinct asymmetric strategies.

A. Reference Symmetric Policies

The strategies are evaluated against three baseline policies that guarantee linear decodability:

1. Optimized Policy ($\pi_{opt}$): Achieves the theoretical DoF-optimal bound by searching for the optimal number of served users ($\Omega$) and streams per user ($\beta$).
2. Combinatorial Policy ($\pi_{cmb}$): Views the system as parallel MISO subsystems. It offers a closed-form DoF expression without parameter search but incurs high subpacketization.
3. Linear/Cyclic Policy ($\pi_{lin}$): Uses a cyclic construction to achieve linear subpacketization growth, offering a favorable complexity-performance trade-off (applicable when $\lfloor L/G \rfloor \ge K\gamma$).

B. Proposed Asymmetric Strategies

The paper introduces four strategies to handle heterogeneous $G_k$:

1. min-G Strategy (Primary):

   • Mechanism: Forces symmetry by treating all users as having $G_{min} = \min_k(G_k)$ antennas.
   • Trade-off: Maximizes global caching gain (largest multicast groups) but severely underutilizes the spatial multiplexing capability of users with $G_k > G_{min}$.

2. Grouping Strategy (Primary):

   • Mechanism: Partitions users into disjoint subsets based on their antenna counts ($K_j$ with $G_j$). Each group is served independently using orthogonal time slots.
   • Trade-off: Maximizes spatial multiplexing gain for each group but sacrifices global caching gain because the cumulative cache size is calculated per group rather than globally.

3. Super-grouping Strategy (Hybrid):

   • Mechanism: A structured combination of the two primary strategies. It merges original user groups into "equivalent" super-sets. Within each super-set, the min-G approach is used (setting the effective gain to the minimum in that set). Between super-sets, the Grouping approach is used.
   • Optimization: The algorithm searches for the optimal partitioning of groups to maximize the total DoF.

4. Phantom Strategy (Hybrid):

   • Mechanism: A more advanced hybrid that introduces "phantom" antennas.
     • Multicasting (MC) Rounds: The system assumes a symmetric setup with the maximum antenna count ($\hat{G}$) among the active users. It generates transmission vectors based on this high capacity.
     • Discarding: For users with fewer actual antennas ($G_k < \hat{G}$), specific subpackets are randomly discarded to ensure linear decodability.
     • Unicasting (UC) Round: The discarded subpackets are delivered in a final unicasting phase.
   • Benefit: This allows the system to exploit the high spatial multiplexing gain of the "phantom" assumption during the MC phase while recovering the lost data later, effectively bridging the gap between min-G and Grouping.

3. Key Contributions

• First Characterization of Asymmetric MIMO-CC DoF: The paper provides the first analytical characterization of achievable single-shot DoF for MIMO-CC systems with heterogeneous receive antennas.
• Four Novel Strategies: Development of min-G, Grouping, Super-grouping, and Phantom strategies, offering a spectrum of solutions from pure caching optimization to pure spatial optimization and hybrid approaches.
• Closed-Form Expressions: Derivation of closed-form DoF expressions and subpacketization scaling laws for all strategy-policy combinations.
• Linear Decodability Proof: Rigorous proof that all proposed strategies maintain linear decodability for all target users, a critical requirement for practical implementation.
• Complexity-Performance Trade-offs: Analysis of the trade-off between achievable DoF and subpacketization complexity (e.g., the Phantom strategy offers high DoF, while the Linear policy reduces subpacketization significantly).

4. Results

Extensive numerical simulations validate the theoretical findings:

• Hybrid Superiority: The Super-grouping and Phantom strategies consistently outperform the primary strategies (min-G and Grouping) across various system configurations.
  • Phantom generally achieves the highest DoF by effectively utilizing the spatial resources of high-capacity users without completely sacrificing the global caching gain.
  • Super-grouping closely approaches Phantom performance with lower implementation complexity.
• Policy Comparison:
  • The Optimized (opt) policy yields the highest DoF but requires parameter search.
  • The Linear (lin) policy achieves DoF very close to the optimized policy in many regimes but with significantly lower subpacketization (linear vs. exponential/combinatorial growth), making it more scalable.
• Impact of Heterogeneity:
  • As the number of users with high antenna counts increases, the Phantom and Super-grouping strategies translate this into significant DoF gains, whereas min-G fails to capture these gains.
  • The strategies effectively adapt to the distribution of antenna counts; for instance, grouping users with similar capabilities (Super-grouping) can sometimes outperform a single global Phantom round depending on the specific user distribution.
• Scaling: The DoF scales linearly with the number of transmit antennas ($L$) and the cache ratio ($\gamma$), with the hybrid strategies showing steeper growth rates than primary strategies.

5. Significance

• Practical Relevance: This work moves beyond idealized symmetric assumptions, addressing the reality of heterogeneous 5G/6G networks where devices have vastly different capabilities.
• Design Guidelines: It provides network engineers with a toolkit to select the appropriate transmission strategy based on system constraints:
  • Use min-G for simple, cache-heavy scenarios with low spatial diversity.
  • Use Grouping for scenarios where spatial diversity is high but global caching is less critical.
  • Use Phantom/Super-grouping for optimal performance in mixed environments, balancing spatial and caching gains.
• Subpacketization Mitigation: By demonstrating that the Linear (cyclic) policy can achieve near-optimal DoF with manageable subpacketization, the paper addresses a major bottleneck in coded caching implementation.
• Theoretical Foundation: It establishes a new theoretical baseline for future research into heterogeneous MIMO networks, proving that heterogeneity can be a resource (via hybrid strategies) rather than just a limitation.