Each node, also called a data collector or collector, has a microprocessor, serial interface flash memory, RF transceiver, linear amplifier, low-noise amplifier, RS-485 transceiver, and IrDA module. The basic hardware structure schematic and a photo of the data collector are shown in Figure 2.

Microprocessor: Usually, one of the three microcontrollers, i.e., ATmega128L, MSP430, and Silicon Labs C8051, is used in nodes of the sensor network. After intensive investigations, we chose the ATmega128L microcontroller for our EMMNet platform, because it is a low-power microcontroller based on the AVR-enhanced Reduced Instruction-Set Computer (RISC) architecture. The ATmega128L provides the following features: 128 K bytes of in-system programmable flash with read-while-write capabilities, 4 K bytes Electrically Erasable Programmable Read-Only Memory (EEPROM), 4 K bytes static RAM (SRAM), and 53 general purpose IOs. Making rational use of the microprocessor resources, we leave ample room for expansion. At present, we have used 62.00% of the port, 34.87% of the code memory, 72.75% of the data memory and 88.23% of the EEPROM. Ports assignment is listed in Table 2.

RF Transceiver: We use the Nordic nRF905 transceiver because its carrier frequencies are suitable for automatic electricity meter monitoring and there is no associated proprietary protocol. Its current consumption is very low: in transmit mode, only 9 mA at an output power of −10 dBm, and in receive mode, 12.5 mA. However, EMMNet faces various challenges. Therefore, we add a broadband, fixed-gain, linear amplifier and a low-noise amplifier to the RF circuit.

Memory: A memory bank with relatively large capacity is believed to be beneficial for data storage and buffering in EMMNet since there can be up to 100 electricity meters connected to a single data collector. Therefore, we chose the AT45DB041D, a 4-Mbit serial-interface flash memory. In addition to the main memory, the AT45DB041D also contains two SRAM buffers of 264 bytes each. In our design, we take full advantage of this external SRAM, greatly easing the problem of small SRAM in the ATmega128L.

The key components of the data concentrator are similar to those of the data collector. These devices can share most of the drivers and application codes, which greatly reduces the difficulty of our design and of debugging. However, the data concentrator requires more external memory because it is responsible for transmitting electricity meter data from hundreds of data collectors to the remote centralized server. A GPRS module is also added to the design of the data concentrator. Modularity is a key tenet of our design, reflected in the separation of the GPRS module and the 470 MHz wireless module on the data concentrator. The basic hardware structure is illustrated in Figure 3.

Memory: We use the AT45DB642D instead of the AT45DB041D. Its 69,206,016 bits of memory are organized as 8,192 pages. In addition to the main memory, the AT45DB642D also contains two SRAM buffers of 1,056 bytes each.

GPRS module: We chose the SIM300C for the data concentrator. The SIM300C is a tri-band GSM/GPRS engine that works in the EGSM 900 MHz, DCS 1,800 MHz, and PCS 1,900 MHz frequencies. With a tiny configuration of 50 × 33 × 6.2 mm, SIM300C meets the space requirements in our industrial application. The SIM300C is designed to have power-saving features: the current consumption is as low as 2.5 mA in sleep mode.

The ATmega128L contains 128 K bytes of in-system programmable flash with read-while-write capabilities, 4 K bytes EEPROM, and 4 K bytes SRAM. The in-system programmable flash, which is used for storing program code, does not cause a bottleneck in our design. The problem is that the SRAM is relatively insufficient. It is used for temporary storage of data, produced as a result of processing, until instructions call for the data to be used again in subsequent processing. In addition, during interrupts and subroutine calls, the return address is stored on the stack. The stack is effectively allocated in the general data SRAM, and consequently the stack size is only limited by the total SRAM size and the usage of the SRAM. Therefore, we have to reserve enough SRAM space. As mentioned before, the AT45DB041D contains two SRAM buffers of 264 bytes each, and the AT45DB642D also contains two SRAM buffers of 1,056 bytes each. Therefore, we can take full advantage of these parts of the space and treat them as external SRAM buffers. Thus, the available SRAM of the collector is increased by 12.89%, and that of the concentrator is increased by 50.10%. EEPROM in the ATmega128L is used for storing the device parameters and the routing table. Since each entry of the routing table occupies 5 bytes, the EEPROM can record about 800 entries, which is sufficient for this type of application. The external flash is a type of non-volatile storage that can be electrically erased and reprogrammed. No power is needed to maintain the information stored in the chip. Therefore, all the data obtained from the electricity meters are kept in external flash. The data concentrator is required to record hourly electricity meter data in last fifteen days and daily electricity meter data in last sixty days. And the flash memory size used in a concentrator is 69,206,016 bits, so a single data concentrator can monitor at most 2,300 electricity meters, which is sufficient for the application of the automatic electricity meter monitoring.

Minimizing the energy consumption of our platform is essential, so a dynamic power-management technique is a feature of our system. This technique dynamically reconfigures the system to provide the requested services and performance levels with a minimum number of active components or a minimum activity level on such components. We chose the SP6201 to dynamically control the RF module. The SP6201 is CMOS low dropout regulator (LDO), which offers an extremely low quiescent current. The current of the nRF905 and the SP485E is zero in the sleep mode using the SP6201. Table 3 shows the average current of a data collector under different modes.

The design of an automatic electricity monitoring system needs to consider lightning and fire protection as well as waterproofness. Therefore, transient voltage suppressors are added to protect devices from damage due to lightning strikes by intercepting such strikes and safely passing their extremely high voltage. Thermistors are also included in our design. A thermistor is a type of resistor whose resistance varies with temperature. Thermistors are widely used as inrush current limiters, temperature sensors, overcurrent protectors, and self-regulating heating elements. In addition, waterproof sealing is also needed for collectors and concentrators.

Initially, the DtA value of each data collector equals DtA_Max + 1, where DtA indicates the hop count to the data concentrator, and DtA_Max is the maximum depth of the EMMNet. The DtA value of the data concentrator is set to 0. When the collector is opened, it will regularly send a request message, RM. This request message includes two fields: packet_sender_ID and packet_DtA. Here, packet_sender_ID indicates who sends the message, and packet_DtA describes the hop count between the sender and the concentrator. When any collector receives an RM packet, it can reply if its DtA is smaller than the packet_DtA in RM and does not equals DtA_Max. Therefore, only the data concentrator can reply to collector requests at the beginning. The reply to the request message, RRM, has the following fields: packet_source_ID, packet_DtA, and packet_sender_ID. Here, packet_source_ID denotes the ID of the concentrator.

If any collector receives an RRM and its DtA equals DtA_Max + 1, its DtA is set to packet_DtA + 1 no matter what the packet_DtA value is. Meanwhile, its father_ID list and root_ID are updated according to packet_sender_ID and packet_source_ID, where father_ID list and root_ID are used for recording the potential relay nodes and the concentrator ID.

If a collector receives another RRM and its DtA is smaller than DtA_Max + 1, it should make judgments based on the value of packet_DtA and its own DtA. If packet_DtA + 1<DtA, the value of DtA and root _ID should be updated. Meanwhile, it should clear the father_ID list before inserting packet_sender_ID as a new relay node. If (1) packet_DtA + 1=DtA, (2) packet_source_ID equals root _ID, and (3) packet_sender_ID is not in the father_ID list, packet_sender_ID is added to the end of the father_ID list.

In the electricity meter monitoring application, we not only collect the quantity of electric, but also control the parameters of the electricity meters. Transmission is divided into uplink and downlink channels. The uplink channel is used for transmission of information from the collectors to the concentrator. The downlink channel is used for transmission of information from the concentrator to the collectors, including resetting the sample ratio, altering the collection parameter, and sending query commands. Therefore, we should establish the downlink channel. Each collector needs to transmit a routing message to the concentrator when its DtA is smaller than DtA_Max + 1. If a collector receives routing messages from its child collectors, the routing information is recorded in the routing table and the messages are relayed to its parent collectors.

The structure of the EMMNet after network initiation is shown in Figure 4. Each device has a communication range, which is illustrated by the dotted line. The communication range is not a perfect circle due to the complicated communication environment. The dashed lines indicate relationships among the collectors and the concentrator. The numbers in the circles indicate the value of DtA. The node whose DtA value equals zero denotes the data concentrator. A collector whose DtA value is larger than zero at least has a parent collector, and the DtA value of this collector is larger than that of its parent collector by one.

Every fifteen minutes, collectors report electricity meter data to the concentrator. However, most of the time collectors are in sleep mode. The concentrator transmits the control message to collectors according to the routing table, while collectors report the data based on the father_ID list. If a child collector detects that one of its parent collectors does not respond, it can choose another spare parent collector to relay the message. If all of its parent collectors fail to respond—the situation may be caused by network structure change—it should rejoin the network. In other words, this collector needs to reset DtA to DtA_Max + 1 and send request messages, RM. If a collector receives an RM from its parent collector and its packet_DtA does not equal DtA-1, it needs to remove the corresponding item from the father_ID. A special case is when the father_ID list contains only one single item: the collector should also rejoin the network. Making use of this strategy, EMMNet can maintain connectivity without frequent network reconstruction, even if some collectors die because of non-recoverable errors or malfunction.

We analyze the DT protocol under the ideal case and discuss the problems occurring in practical applications in this section.

First, we make several assumptions to simplify the analysis:

The collectors in the EMMNet are distributed as a homogeneous spatial Poisson process□ in two-dimensional space.

This can then be extended to the situation of 2n hops by concatenating n 2-hop topological graphs. In the same way, 2n + 1 hops can be extended by concatenating n − 1 2-hop topological graphs and one 3-hop topological graph. We can conclude that 2n-hop or 2n + 1-hop requires at least an interval of nr. Thus, with the fact that 2 n hops require at most an interval of 2nr, it could be deduced that on average there are [d/(0.75 r)] hops between any collector and its concentrator.

(4) Taking into account the short distance in one hop, we use the free space channel model for estimating the radio hardware energy dissipation: (1) E T X = E elec k + e fs d 2 k

To receive this message, the radio expends: (2) E RX = E elec k

The electronics energy (E_elec) depends on such factors as digital coding, modulation, and filtering. The variable d indicates the distance to the receiver. The amplifier energy, e_fs, depends on the distance d and the acceptable bit-error rate [21]. k is the length of the message.

According to the results in Reference [22], the expectation value of the total length of all segments connecting the collectors to the concentrator, L_v, is given by: (3) E [ L v ] ≈ 1 2 n 3 / 2where n is the number of collectors. With this formula and the third assumption mentioned above, we can calculate the number of hops, J, in an EMMNet as: (4) J = E [ L v ] r × 3 / 4 = 2 3 n 3 / 2 r

The energy consumption of the communication between the collectors and the concentrator, C, can be evaluated as: (5) E [ C ] = J [ E TX + E RX ] = Jk [ ( E elec + e fs r 2 ) + E elec ]

E[C] is minimized by the value of r that is a solution of: (6) − 2 E elec r 2 + e fs = 0

This equation shows how to maximize the performance of the sensor network. However, a significant gap remains between the assumptions and the application environment. Certain problems may arise in a practical application:

In the conventional literature on sensor networks, protocols are designed on the assumption of reciprocity between uplink and downlink channels. However, affected by the imbalance of the RF circuitries in practice, the reciprocity between the uplink and downlink channels cannot be maintained, which will cause a performance decrease.

We set the packet length of RRM and RM to equal that of the longest data packet to solve the above problems. The collector can only broadcast the RM a limited number of times, marked as LT, when the value of DtA is smaller than DtA_Max + 1. Evaluation results of the DT protocol from several applications show that our strategy greatly improves the performance of the EMMNets.

The experiment environment for the proposed smart sensors is shown in this section. Collaborating with Hangao Electronics Co., Ltd., we deployed our systems in two cities and tested our systems over a three-month period. The data of 2024 electricity meters were +collected. All the devices of the test networks are listed in Table 4 and the elements used in EMMNet are listed in Table 5.

The interval between sending two RMs is 1 minute, the packet length of an RRM or RM equals 330 bytes, and the maximal depth of the EMMNet is 10 in these EMMNets.

Figure 6 shows the architectural plans of the Zhiyang residence community, Yangzhou City, Jiangsu Province, China. We deployed two of our systems in this community. The numbers in circles indicate the last two digits of the collector addresses.

In our original design, there was no limit to the number of times a RM could be sent, even when the collectors had joined the EMMNets. A collector had a great chance of selecting a relay node with a smaller DtA value, when it constantly sent RMs. However, the performance of such relay nodes might be poor. We found that the transmission success rate decreased if a collector chose a parent collector with a very small DtA value. Therefore, we limited the number of broadcast packets when the value of DtA is smaller than DtA_Max + 1.

The distribution of the DtA value is shown in Figure 7, and the transmission success rate is illustrated in Figure 8. From November 3–18, 2009, collectors could broadcast RMs without any limit. As shown in Figure 7, the network structure does not seem very good. All the DtA values of collectors are less than 3, and the average transmission success rate is relatively low. That is because child and parent nodes are too far from each other and the bit error rate is relatively high. Therefore, we reprogrammed the collectors with a hand-held device in the 470 MHz wireless channel. When the value of DtA was smaller than DtA_Max + 1, the collector could only broadcast the RM three times. On Deceber 24, 2009, we reprogrammed the devices and made the EMMNets dynamically choose the limited times that RM packets were broadcasted according to the transmission success rate. At first, the limited-times value LT was set to two; if the performance was poor, the LT was increased by two. Figure 8 shows that average transmission success rate is improved due to the advantage of the network topology. And Figure 9 plots the corresponding network structures of the EMMNets deployed in Zhiyang residence community. With the EMMNets, customers can obtain the energy flow amount at any time through internet and the maintainers can control the parameters of the electricity meters if necessary. The meter monitoring task can be done at the management department by using this system. What should be emphasized is that the operational cost of EMMNet is relatively low, compared with other existing remote meter monitoring systems based on GPRS. Meanwhile, EMMNets can realize automatic charge if they are connected to bank systems.