[Paper Review] A Cross-layer Perspective on Energy Harvesting Aided Green Communications over Fading Channels

Tian Zhang, Wei Chen+2 more|arXiv (Cornell University)|Nov 23, 2012

Energy Harvesting in Wireless Networks4 citations

TL;DR

This paper proposes a cross-layer optimization framework for energy harvesting green communications that jointly minimizes long-term average buffer delay while satisfying an average grid power constraint. By modeling transmission rate and battery power allocation as a constrained Markov decision process (MDP), the authors derive structural properties of the optimal policy and present low-complexity heuristic policies, proving that greedy battery usage is optimal under certain conditions and validating performance via simulations.

ABSTRACT

We consider the power allocation of the physical layer and the buffer delay of the upper application layer in energy harvesting green networks. The total power required for reliable transmission includes the transmission power and the circuit power. The harvested power (which is stored in a battery) and the grid power constitute the power resource. The uncertainty of data generated from the upper layer, the intermittence of the harvested energy, and the variation of the fading channel are taken into account and described as independent Markov processes. In each transmission, the transmitter decides the transmission rate as well as the allocated power from the battery, and the rest of the required power will be supplied by the power grid. The objective is to find an allocation sequence of transmission rate and battery power to minimize the long-term average buffer delay under the average grid power constraint. A stochastic optimization problem is formulated accordingly to find such transmission rate and battery power sequence. Furthermore, the optimization problem is reformulated as a constrained MDP problem whose policy is a two-dimensional vector with the transmission rate and the power allocation of the battery as its elements. We prove that the optimal policy of the constrained MDP can be obtained by solving the unconstrained MDP. Then we focus on the analysis of the unconstrained average-cost MDP. The structural properties of the average optimal policy are derived. Moreover, we discuss the relations between elements of the two-dimensional policy. Next, based on the theoretical analysis, the algorithm to find the constrained optimal policy is presented for the finite state space scenario. In addition, heuristic policies with low-complexity are given for the general state space. Finally, simulations are performed under these policies to demonstrate the effectiveness.

Motivation & Objective

• To minimize long-term average buffer delay in energy harvesting wireless networks under grid power constraints.
• To jointly optimize physical layer transmission rate and battery power allocation with upper-layer buffer dynamics.
• To account for the combined impact of random data arrivals, intermittent energy harvesting, and fading channels on system performance.
• To develop low-complexity heuristic policies that achieve near-optimal performance with practical feasibility.
• To prove that greedy battery usage is optimal under the derived structural properties of the MDP framework.

Proposed method

• Formulates a stochastic optimization problem integrating physical layer power allocation and application layer buffer delay under random energy and channel conditions.
• Reframes the constrained MDP problem as an unconstrained MDP using Lagrangian relaxation to derive the optimal policy structure.
• Models the system as a two-dimensional MDP with state variables including buffer size, channel gain, queue state, battery level, and energy arrival.
• Proves that the optimal policy is non-decreasing in buffer size and battery energy, and derives structural properties using convexity and dynamic programming techniques.
• Proposes a finite-state algorithm for exact optimal policy computation and heuristic policies (two deterministic and one mixed) for general state spaces.
• Validates the framework through simulations under various policies, demonstrating delay reduction and energy efficiency gains.

Experimental results

Research questions

• RQ1 How can transmission rate and battery power allocation be jointly optimized to minimize average buffer delay in energy harvesting networks?
• RQ2 What structural properties does the optimal policy exhibit under random data arrivals, fading channels, and intermittent energy harvesting?
• RQ3 Under what conditions is greedy battery power allocation optimal for minimizing delay and grid power usage?
• RQ4 How does the inclusion of circuit power affect the optimal power allocation strategy compared to models that consider only transmission power?
• RQ5 Can low-complexity heuristic policies achieve near-optimal performance while maintaining practical implementability?

Key findings

• The optimal policy for the constrained MDP is shown to be equivalent to solving an unconstrained MDP, enabling efficient computation via Lagrangian relaxation.
• The optimal transmission rate policy is non-decreasing with respect to buffer size, ensuring higher rates as backlog increases.
• The optimal battery power allocation policy is non-decreasing with respect to battery energy level, and greedy usage (prioritizing battery over grid) is proven optimal.
• For large discount factors, the optimal policy converges to a greedy battery allocation strategy under any fixed rate policy.
• The proposed heuristic policies achieve significant delay reduction with minimal complexity, outperforming baseline strategies in simulations.
• Simulation results confirm that the framework effectively balances grid power usage and buffer delay under realistic stochastic channel and energy arrival models.