1 Introduction
Tiny, resource-constrained wireless sensor network nodes are the basic functional units of sensorless networks, undertaking important tasks such as information collection, data processing, and information transmission.
With advancements in MEMS, microelectronics, network, and computer technologies, wireless sensor networks are gradually becoming a reality. Researchers have developed miniaturized board-level wireless sensor network nodes using embedded technology, something that was merely a concept 30 years ago. Monolithic wireless sensor network nodes have also emerged, but they are still quite far from practical application. To study the networking and energy management technologies of wireless sensor networks, we developed a sensor network node (as shown in Figure 1) using the LPC2138 SoC microcontroller based on the ARM7 core.
2. Node Design Overview
Compared to the significant increases in processor speed and power consumption, improvements in battery performance have been much slower, making energy management the biggest challenge for wireless sensor networks. To save energy, wireless sensor networks require nodes to have Dynamic Power Management (DPM) capabilities, enabling them to enter a low-power state when idle to conserve energy. Implementing DPM requires microcontroller support. Due to the unparalleled advantages of ARM technology in wireless communication, over 85% of wireless communication devices utilize ARM technology. We chose the Philips LPC2138, a high-performance, low-power ARM-based microcontroller, to build the processing unit.
The LPC2138 offers comprehensive DPM support: it features both sleep and power-down low-power states, which can be woken up by external interrupts; in oscillation mode, it supports external crystals from 1 to 30 MHz, and through a phase-locked loop, the CPU can achieve an operating frequency of up to 60 MHz, while an 8 MHz crystal is used for energy saving; in addition to being enabled and disabled via the peripheral power control register, the operating frequency of on-chip peripherals can also be adjusted to 1/2 or 1/4 of the processor clock frequency via a frequency divider. Furthermore, the memory acceleration function can greatly speed up program execution and improve energy efficiency. These features make the LPC2138 suitable for application in low-power systems with considerable processing power.
To enable the node to be powered by two AA batteries, a boost-type DC-DC MAX756 is used to construct the power supply unit. In addition to boosting the voltage, the MAX756 also features power monitoring. When Vin (adjustable via R1 and R3) drops below 1.25V, the LBO pin outputs a low level and sinks current (as shown in Figure 2). While this doesn't accurately indicate the battery's state of charge (SOC), it allows the sensor node to understand that its battery's SOC has dropped to a certain level, rendering the node unsuitable for heavy workloads. This allows for adjustments to the node's operating state, reducing power consumption and extending its lifespan.
The data transceiver unit is built using a ZigBee-compliant RF transceiver chip from Chipcon; the sensing unit consists of a DS1722 temperature sensor and a TSL2561 brightness sensor. Dynamic power management for these sensors is achieved by amplifying the MCU's GPIO drive capability using transistors.
3 Energy Management
The topology of wireless sensor networks (WSNs) creates an imbalance in energy usage among nodes. Therefore, WSNs need to know the battery usage of each node, obtain the state of charge (SBC) of the battery, and adjust the node's role accordingly, dynamically changing the network topology to offset this imbalance. Thus, for WSNs, simply using DPM (Dynamic Power Management) to put nodes into a low-power state without considering battery status cannot optimize energy usage within the network and maximize network lifespan.
3.1 Battery Model
The state of charge of a battery is usually expressed as the ratio of its current available capacity to its rated capacity. It is not a linear function of discharge time and discharge current. It is affected by the inherent properties of the battery, namely the "rated capacity effect" and the "recovery effect". In order to design batteries, evaluate systems and optimize battery usage strategies, researchers have proposed a variety of battery models from different levels. This paper adopts the battery model based on Markov process in reference [7]. This model introduces the minimum available charge unit to represent the state of charge of the battery as a discrete transient random process (as shown in Figure 3). In the figure, N, N-1, ..., 1, 0 in the circle represent the nominal capacity of the battery at a certain moment; qi represents the probability of consuming i charge units in a certain time period. If the battery has N charge units at the beginning and consumes 3 charge units in a certain time period, then the probability of this event is expressed as q3, and the remaining charge units of the battery are N-3.
To describe the battery's "recovery effect," this model divides the battery's recovery capability into f (f=0, 1, ..., fmax) stages based on the strength of the battery's recovery ability during discharge intervals. Within a time step, the probability of the battery recovering one charge unit when it is in state j (j=1, 2, ..., N-1) and stage f is:
In the formula, gN and gC are related to the battery's recovery capability, and q0 is the probability that the battery is in an idle state. Given the recovery probability, the probability that the battery will remain in stage f during a certain idle time while maintaining its charge state can be expressed as:
Compared to battery models described by partial differential equations, this model significantly reduces computational cost while maintaining high accuracy, enabling rapid assessment of the impact of embedded system architecture design on battery state. However, its application in real-time assessment of the state of charge (SOC) of wireless sensor network nodes remains too costly, necessitating further exploration of battery modeling issues.
3.2 Node Power
In practical applications, the node operating current serves as an external basis for evaluating the battery's state of charge. Since a sensorless node is composed of several discrete devices, its power can be obtained from the combination of the effective power consumption states of these discrete devices, as shown in Table 1.
4. Conclusion
Wireless sensor networks (WSNs) have a very broad application prospect and represent a revolutionary information acquisition technology. However, many technical challenges remain unresolved, particularly energy management and large-scale networking. The importance of energy in WSNs necessitates researchers discovering more cost-effective methods to accurately predict battery state of charge to balance network node lifetimes and optimize energy consumption within the network—a crucial issue that urgently needs to be addressed.
