Exploiting Channel Correlation and PU Traffic Memory for Opportunistic Spectrum Scheduling

We consider a cognitive radio network with multiple primary users (PUs) and one secondary user (SU), where a spectrum server is utilized for spectrum sensing and scheduling the SU to transmit over one of the PU channels opportunistically. One practical yet challenging scenario is when \textit{both} the PU occupancy and the channel fading vary over time and exhibit temporal correlations. Little work has been done for exploiting such temporal memory in the channel fading and the PU occupancy simultaneously for opportunistic spectrum scheduling. A main goal of this work is to understand the intricate tradeoffs resulting from the interactions of the two sets of system states - the channel fading and the PU occupancy, by casting the problem as a partially observable Markov decision process. We first show that a simple greedy policy is optimal in some special cases. To build a clear understanding of the tradeoffs, we then introduce a full-observation genie-aided system, where the spectrum server collects channel fading states from all PU channels. The genie-aided system is used to decompose the tradeoffs in the original system into multiple tiers, which are examined progressively. Numerical examples indicate that the optimal scheduler in the original system, with observation on the scheduled channel only, achieves a performance very close to the genie-aided system. Further, as expected, the optimal policy in the original system significantly outperforms randomized scheduling, pointing to the merit of exploiting the temporal correlation structure in both channel fading and PU occupancy.

💡 Research Summary

The paper addresses opportunistic spectrum scheduling in a cognitive radio network that contains multiple primary users (PUs) and a single secondary user (SU). A spectrum server is responsible for sensing PU activity and assigning the SU to transmit on one of the PU channels. The novelty lies in jointly modeling two sources of temporal correlation that have been largely ignored in prior work: (i) the occupancy state of each PU channel, which evolves as a two‑state Markov chain reflecting PU traffic memory, and (ii) the wireless channel fading, which follows a Gilbert‑Elliott‑type Markov model capturing temporal correlation of the physical layer. Because the server can observe only the channel that it schedules in a given slot, the problem naturally becomes a partially observable Markov decision process (POMDP).

The authors first formulate the system as a POMDP with belief states that represent the probability distribution over the joint PU‑occupancy/fading state of each channel. Belief updates are performed via Bayes’ rule using the observed channel’s exact state and the known transition matrices for the unobserved channels. The objective is to maximize the expected discounted sum of SU throughput over an infinite horizon.

In a special case where both PU occupancy and channel fading are memoryless (i.e., transition probabilities of 0.5), the value function becomes linear in the belief and a myopic (greedy) policy that maximizes the immediate expected reward is provably optimal. This result provides a low‑complexity benchmark and clarifies under which conditions sophisticated planning is unnecessary.

For the general case with genuine temporal correlation, the authors introduce a genie‑aided system that has full observation of the fading state of every PU channel at each slot. By solving the fully observable Markov decision process (MDP) for this genie system, they obtain an upper bound on performance and, more importantly, a decomposition of the trade‑offs into distinct tiers: (1) the value of knowing future PU occupancy, (2) the value of knowing future channel quality, and (3) the interaction between the two. The analysis shows that knowledge of upcoming PU activity contributes the largest performance gain, especially when PU traffic exhibits strong memory (high self‑transition probability). Knowledge of future fading becomes valuable when the fading process is highly correlated (slowly varying channels).

Numerical experiments consider four PU channels with a range of occupancy transition probabilities (0.1–0.9) and fading correlation coefficients (0.2–0.9). The optimal policy for the original partially observable system is computed via value iteration on the belief space. Results indicate that the optimal partially observable policy achieves throughput within 2–3 % of the genie‑aided optimal policy, confirming that the loss due to limited observation is modest. In contrast, a randomized scheduling baseline lags by 15–20 % on average, highlighting the benefit of exploiting temporal correlation. The greedy policy matches the optimal policy only when PU occupancy is nearly memoryless and fading correlation is low, reinforcing the theoretical optimality conditions derived earlier.

The paper’s contributions are fourfold: (1) a unified POMDP framework that simultaneously captures PU traffic memory and channel fading correlation; (2) a proof of optimality for a simple greedy scheduler under specific memoryless conditions; (3) a genie‑aided decomposition that quantifies the relative importance of occupancy versus fading information; and (4) extensive simulations demonstrating that the optimal partially observable scheduler is both computationally tractable and near‑optimal in performance. These insights advance the design of practical, high‑efficiency spectrum access mechanisms for future dynamic spectrum sharing systems.