⚡ electrical engineering

Low-Subpacketization MIMO Coded Caching with Flexible Stream Allocation

This paper proposes a low-complexity MIMO coded caching scheme that significantly reduces subpacketization requirements while enabling flexible stream allocation to achieve near-optimal degrees of freedom and improved throughput under linear decodability constraints.

Original authors: Mohammad NaseriTehrani, MohammadJavad Salehi, Antti Tölli

Published 2026-06-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Mohammad NaseriTehrani, MohammadJavad Salehi, Antti Tölli

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Problem: The "Too Many Pieces" Puzzle

Imagine a library (the server) trying to send movies to a group of friends (the users) who all have a small shelf in their house (their cache/memory).

In the past, a clever trick called Coded Caching was invented. Instead of sending the whole movie to everyone, the library sends a giant "puzzle." Each friend already has a few pieces of the puzzle on their shelf. When they get the new puzzle piece from the library, they can combine it with what they have to build their specific movie. This saves a huge amount of time and bandwidth because one single transmission helps everyone at once.

However, there is a catch: To make this work perfectly, the library has to cut every movie into thousands, or even millions, of tiny micro-pieces (called subpackets) before sending them.

The Analogy: Imagine trying to send a pizza to 20 friends. To use this old trick, you'd have to slice the pizza into 10,000 tiny crumbs, label each one with a complex code, and hope everyone gets the right crumbs. If you have more friends, the number of crumbs explodes exponentially. This makes the system too complicated to actually build in the real world.

The New Solution: "Virtual Groups" and "Flexible Streams"

The authors of this paper propose a new way to organize this pizza delivery that keeps the speed benefits but stops the "crumb explosion."

1. The "Virtual Group" Trick (Reducing Complexity)

Instead of treating every single friend as a unique individual with a unique set of puzzle pieces, the authors suggest grouping friends together.

The Analogy: Imagine the 20 friends are sitting at 4 different tables (4 groups). Everyone at Table 1 gets the exact same set of pre-sliced pizza pieces on their shelf. Everyone at Table 2 gets a different identical set, and so on.
Why it helps: The library no longer needs to create unique puzzle pieces for 20 different people. It only needs to create pieces for 4 "virtual groups." This drastically reduces the number of tiny slices (subpackets) needed, making the system manageable even with many users.

2. The "Multi-Antenna" Upgrade (Sending More at Once)

The paper deals with MIMO systems, which means the server has multiple antennas (like a multi-lane highway) and the users have multiple antennas (like multi-lane driveways).

The Analogy: In the old days, the server could only send one "stream" of data to a group at a time. With this new method, because the users have multiple "driveways" (antennas), the server can send multiple streams of data simultaneously to the same group.
The Flexibility: The authors created a system where you can choose how many people to serve at once and how many data streams to send to each person. It's like having a flexible delivery truck that can carry 10 boxes to 5 houses, or 20 boxes to 2 houses, depending on what fits best.

How It Works in Practice

The paper describes a two-step process:

Virtual Planning: They pretend the complex multi-antenna network is a simpler, single-antenna network. They solve the puzzle delivery problem in this "virtual world" where the math is easier.
Real-World Elevation: Once they have the plan, they "lift" it back up to the real multi-antenna world. Because they grouped the users, they can now send multiple data streams (like sending 2 or 3 movies at once to the same group) without the math getting out of control.

The Results: Speed vs. Complexity

The authors tested their idea and found two major wins:

Massive Reduction in Complexity: For the same amount of data delivery, their method requires orders of magnitude fewer tiny puzzle pieces than previous "best" methods.
- Analogy: If the old method required cutting a pizza into 100 million crumbs, their method might only need 100 crumbs. This makes it possible to actually build the system.
Better Real-World Performance: They found that sometimes, sending fewer streams to fewer people at once actually works better in real life (at normal signal strengths) than trying to push the maximum theoretical speed.
- Analogy: Trying to drive 10 cars down a narrow road at top speed causes traffic jams (interference). Their system allows you to slow down and send 4 cars smoothly, which gets everyone to their destination faster than a chaotic 10-car pile-up.

Summary

This paper presents a new way to deliver data to many users with multiple antennas. It solves the problem of the system becoming too complicated by grouping users and flexibly adjusting how much data is sent at once. The result is a system that is much easier to build (low "subpacketization") but still delivers data very fast, especially in real-world conditions.

Technical Summary: Low-Subpacketization MIMO Coded Caching with Flexible Stream Allocation

Problem Statement
Coded caching (CC) is a proven technique for alleviating traffic demands in wireless networks by exploiting user device storage to create global caching gains. While effective in single-input single-output (SISO) and multi-input single-output (MISO) systems, the practical deployment of CC in multi-input multi-output (MIMO) networks faces a critical scalability barrier: subpacketization. In classical CC schemes, the number of file segments required for cache placement and delivery grows exponentially with the number of users. This rapid growth severely limits the feasibility of implementation in large-scale systems. Furthermore, existing low-subpacketization designs for MISO systems often focus exclusively on maximizing the asymptotic Degrees of Freedom (DoF) without providing mechanisms to tune transmission schemes for optimized performance at practical, finite signal-to-noise ratios (SNR). Additionally, general MIMO settings offer receive-side spatial multiplexing capabilities that many existing schemes fail to fully exploit under linear decodability constraints.

Methodology
The authors propose a novel MIMO CC framework based on a virtual broadcast-channel decomposition approach. The core methodology involves three primary stages:

Virtual Decomposition: The original cache-aided MIMO broadcast channel is mapped to an equivalent virtual MISO (V-MISO) network. In this virtual domain, the system is treated as having a virtual base station with $\Delta$ antennas serving single-antenna virtual users (V-UEs).
Group-Aligned Cache Placement: To reduce subpacketization, the authors employ a strategy inspired by the shared-cache paradigm. Users are partitioned into disjoint groups, and all users within the same group are assigned identical cache contents. This reduces the effective placement dimension from $K$ individual users to $P$ cache groups, where $P = K/\Delta$ .
Elevation to MIMO: The placement and delivery algorithms designed in the virtual domain are elevated back to the original MIMO setting. In this step, each virtual stream is replaced by $\beta$ parallel spatial streams for the actual multi-antenna users. The system utilizes Zero-Forcing (ZF) beamforming to manage interference, ensuring that intra-group interference is nulled and inter-group interference is mitigated via cached side information.

The scheme introduces two key design variables: $\Omega$ (the number of simultaneously served users) and $\beta$ (the number of spatial streams per user). These variables are selected from a feasible set defined by linear decodability constraints: $t + 1 \le \Omega \le t + L$ , $\beta \le G$ , and $(\Omega - t)\beta \le L$ , where $t$ is the global caching gain, $L$ is the number of transmit antennas, and $G$ is the number of receive antennas.

Key Contributions

Flexible Stream Allocation: Unlike previous works that target a single optimal DoF point, this framework enables the adaptive selection of feasible $(\Omega, \beta)$ configurations. This provides explicit control over the spatial multiplexing gain, allowing the system to be tuned for specific finite-SNR conditions.
Dramatic Subpacketization Reduction: By utilizing the group-aligned cache placement, the proposed scheme achieves a subpacketization level $\Theta(\Omega, \beta)$ that is orders of magnitude lower than existing DoF-optimized schemes. Theoretical analysis indicates that for large $K$ , the reduction factor is exponential compared to schemes in [9].
Asymptotic DoF Performance: The framework is proven to asymptotically approach the best-known achievable DoF under linear decodability constraints. The gap between the proposed scheme's maximum DoF and the theoretical optimum (from [9]) is bounded by $L-1$ and is often significantly smaller in practical scenarios.
Finite-SNR Optimization: The design acknowledges that maximizing DoF does not always maximize throughput at practical SNRs. By allowing the selection of operating points with lower nominal DoF (fewer streams/users), the scheme can relax interference suppression constraints, leading to higher effective SINR and improved symmetric rates in finite-SNR regimes.

Results
Numerical evaluations validate the theoretical claims:

Subpacketization Efficiency: For a system with $K=24$ users, achieving a DoF of 36 requires a subpacketization level of only 12 in the proposed scheme, whereas the DoF-optimized baseline requires approximately $2.5 \times 10^9$ .
Throughput Gains: Simulations demonstrate that at practical SNR levels, configurations with lower DoF (e.g., $\Omega=18, \beta=1$ ) can outperform higher-DoF configurations (e.g., $\Omega=24, \beta=1$ ) due to more effective beamforming and reduced interference.
Scalability: Under explicit subpacketization constraints (e.g., $\Theta \le 10^4$ or $10^6$ ), the proposed scheme maintains favorable DoF scaling as the number of users increases, outperforming both the constrained Lin-MIMO scheme and the DoF-optimized scheme (which becomes infeasible under such constraints). The proposed design consistently outperforms conventional multi-user MIMO (MU-MIMO) baselines that lack coded caching gains.

Significance
The paper claims that its primary significance lies in bridging the gap between theoretical DoF optimality and practical implementation constraints in MIMO coded caching. By decoupling the subpacketization requirement from the DoF maximization goal through virtual decomposition and flexible stream allocation, the proposed framework offers a practical, low-complexity solution for large-scale MIMO networks. It enables system designers to trade off between asymptotic DoF and implementation complexity (subpacketization) while simultaneously optimizing for finite-SNR throughput, a capability absent in prior art. The work establishes a new paradigm for MIMO CC that is scalable, adaptable to varying channel conditions, and viable for real-world deployment.