Coded caching has emerged as a powerful technique to alleviate the burden on backhaul networks by pre-storing popular content in user devices. By leveraging these local caches, a base station can satisfy multiple distinct requests with a single multicast transmission, achieving a "multicast gain" that scales with the aggregate cache size. However, translating these theoretical gains to wireless environments is challenging due to channel fading.
The Scalability Bottleneck in Wireless Coded Caching
In a standard multicast scenario, the transmission rate is limited by the user with the worst channel condition. As the number of users ($K$) increases, the probability of encountering a "bottleneck" user with a very poor signal increases significantly. In large-scale networks, this causes the per-user throughput to vanish at a rate of $O(1/K)$, effectively negating the benefits of coded caching. This paper addresses how to maintain high speeds without leaving weak users behind through opportunistic scheduling.
System Model
The paper considers a base station serving $K$ users over a fading channel. Each user has a local cache of size $M = mF$, where $F$ is the file size and $m$ is the fraction of files stored.
Opportunistic Scheduling: From Full CSI to 1-Bit Feedback
To overcome the weakest-user bottleneck, the authors propose opportunistic scheduling. Instead of serving all users in every time slot, the scheduler selects a subset of users with favorable channel conditions. The authors compare two primary approaches:
- Optimal Gradient Scheduling: This approach maximizes a utility function based on full Channel State Information (CSI). While optimal, it requires the base station to know the exact signal-to-noise ratio (SNR) of every user at every moment, leading to high feedback overhead and $O(K^2)$ computational complexity.
- Threshold-Based Scheduling: The authors introduce a simplified policy where a user is scheduled only if their channel gain exceeds a pre-defined threshold $c$. This requires only 1-bit of feedback per user (indicating whether they are above the threshold) and reduces complexity to $O(K)$.
Asymptotic Optimality and Fairness
The core contribution of this work is the proof of asymptotic optimality. The authors demonstrate that as the number of users $K$ grows, the performance of the simple 1-bit threshold policy converges to that of the complex full-CSI optimal scheduler. By utilizing the $\alpha$-fairness framework, the system can be tuned to balance total throughput ($\alpha=0$) against strict user fairness ($\alpha \to \infty$).
Key Results and Implications
Numerical simulations validate the theoretical findings, showing that for networks with as few as 50 to 100 users, the 1-bit threshold scheme achieves nearly the same utility as the full-CSI scheme. Unlike baseline multicast methods, the proposed scheduling maintains a stable, non-vanishing per-user rate regardless of network size. This suggests that coded caching can indeed be scaled to large wireless deployments using remarkably simple feedback and scheduling rules, provided that local memory is used as a buffer to wait for optimal channel conditions.
Summary updated