Abstract
This paper establishes a rigorous information-theoretic foundation for "green" communication by characterizing the Shannon capacity of energy-harvesting (EH) sensor nodes. It explores how nodes that harvest energy from stochastic environmental sources can maximize data transmission rates under practical constraints, including processor energy costs, storage inefficiencies, and fading channels.
Introduction
Traditional wireless sensor network (WSN) research typically assumes a fixed, non-rechargeable battery, where the primary goal is maximizing network lifetime. Energy harvesting shifts this paradigm: while a node can potentially operate indefinitely, its power supply becomes a stochastic, time-varying process. This paper addresses the fundamental question of the maximum reliable data rate (Shannon capacity) for such nodes, moving beyond simple models to include processor energy, storage leakage, and channel fading.
System Model
The authors model the energy harvesting process as a system where energy arrives stochastically and is stored in a buffer before being used for transmission.
As shown in Figure 1, the energy buffer is analogous to a G/G/1 queue. Here, "customers" represent units of harvested energy ($Y_k$), and "service" corresponds to the energy consumed during transmission ($T_k$). The encoder maps the source data $S_k$ into a transmitted signal $X_k$, constrained by the available energy $E_k$ in the buffer.
Capacity Analysis and Methodology
To handle the non-stationary nature of the energy buffer, the paper employs the framework of Asymptotically Mean Stationary (AMS) sequences. The communication process is modeled as a channel with side information, where the harvesting process influences the channel state.
The analysis reveals that for an ideal system with an infinite, perfectly efficient battery, the stochastic nature of energy harvesting does not reduce capacity. In this case, the capacity is equivalent to that of a node with a constant power constraint equal to the average harvesting rate: $C = 0.5 \log(1 + E[Y]/\sigma^2)$.
Practical Constraints: Processor Energy and Storage
The research goes beyond ideal models to include "Processor Energy" (PE)—the energy required for sensing and processing regardless of transmission.
- The Shift to Discrete Signaling: While Gaussian signaling is optimal for standard AWGN channels, the introduction of PE makes the optimal input distribution discrete. The node adopts a "sleep-wake" strategy, using a randomized policy to decide when to remain active.
- Storage Inefficiencies: Real-world batteries exhibit leakage and charging losses. The authors compare Harvest-Store-Use (HSU) and Harvest-Use-Store (HUS) architectures. They find that if storage efficiency is low, it is optimal to use energy immediately (Harvest-Use) to bypass battery leakage.
Key Results
- Theorem 1: Confirms the capacity for the ideal EH system is $0.5 \log(1 + E[Y]/\sigma^2)$.
- Theorem 2 & 4: Characterize the capacity when processor energy is included, showing that a randomized sleep policy is necessary to achieve optimal rates.
- Efficiency Impacts: The paper quantifies how leakage ($\beta_2$) and storage efficiency ($\beta_1$) degrade the achievable rate, providing a formula for the HSU architecture: $R_{HSU} = 0.5 \log(1 + \frac{\beta_1 E[Y] - \beta_2}{\sigma^2})$.
Conclusion
This work establishes that while infinite buffers allow for average-power equivalence, practical constraints like processor energy and finite, leaky storage necessitate sophisticated energy management policies and non-Gaussian signaling. While the paper provides achievable rates for finite buffers, finding the exact Shannon capacity for the finite-state case remains an open problem for future research.
Summary updated