Design and Implementation of Temperature and Humidity Monitoring System Using LPWAN Technology

Several works were conducted using standard wireless network technologies for temperature and humidity (TH) measurement and monitoring. For instance, a wireless sensor network (WSN) is an essential wireless communication technology that delivers more reasonable edges than wired communication in cost, scalability, and robustness. They are broadly used with various microcontrollers (MCU) like Arduino, Raspberry Pi, etc., for prototyping and real-life deployment of IoT systems where energy, memory, reliability, etc., are crucial. WSNs employed a multi-hub path (a connection of numerous sensor nodes connected in several topologies such as star and mesh configuration) to relate sensor data to the nearest sink nodes (gateway or base station) [13]. The gateways send these sensor data to the internet via wired Ethernet, Wi-Fi, or cellular networks. Gao et al. [14] proposed a WSN-based remote THMS. The study used ATmega 128L as the MCU unit that provides the suggested system’s network controlling requirement. Also, DS18B20 and SHT75 sensors were used for TH measurement, where nRF905 wireless transceiver modules transmitted the sensor data to the cloud. The system proved scalable and automated in wirelessly sending the sensor data over a considerable long range. Similarly, Guan et al. [15] designed and implemented a short-range wireless communication-based real-time THMS in the agricultural and industrial sectors. The research work introduced an algorithm that intelligently assigned frequency, reduced energy consumption, and solved resource allocation issues. They used an SHT11 sensor to send the sensor data to the STC89C52 MCU unit. The data was then transmitted using the nRF905 wireless transceiver module and MAX232 chip for serial interfacing between these devices. Radio-frequency integrated circuits (RFICs) are analog circuits operating from 3 kHz to 2.4 GHz [16]. Chang and Hung [17] exploited RFIC’s capabilities incorporated with RH/T sensors to directly initiate a continuous measurement of the internal TH of concrete in real-time. Later on, they obtained a remarkable performance compared to the temperature measurements with a Pt100 resistance thermometer and eventually analyzed and deduced that the quantity of reinforcing steel and the strength of the RF signal in the concrete were negatively correlated. Wang and Chi [18] also used an AVR MCU unit-based approach to design and implement a system that sensed specific environmental TH conditions with a DHT11 sensor using a serial peripheral interface (SPI) bus communication. The nRF24L01 wireless transceiver module was used to transmit sensor data, and a Dot matrix LCD12864 display module displayed the results of the measurements to enable wireless communication and real-time monitoring. In addition, a potentiometer was used to set low and high sensor data levels for proper regulation.

The positive contributions of Machine Learning (ML) to various aspects of human lives cannot be overemphasized. When appropriately trained, ML models are famously known for their high accuracy and performance in prediction in multiple applications such as image and speech recognition, natural language processing, self-driving cars, etc. In light of this, several research works have been conducted to predict environmental TH values with ML models. For instance, Bellido-Jiménez et al. [19] proposed a novel ML-based approach that used several ML classifiers on the intra-daily temperature dataset to predict solar values in different geo-climatic conditions such as elevation, sea distance, and aridity. They compared their outcome to empirical methods and found that root-mean-square error (RMSE) values were 7.56% in arid, and 45.65% in humid sites, Nash-Sutcliffe efficiency coefficient (NSE), and R2 value was up to 60% in summer. Then they evaluated the results of the ML classifiers and concluded that Support Vector Machine (SVM) and Random Forest (RF) classifiers were the most preferred in terms of overall performance. Although very challenging due to its nonlinearity, relative humidity prediction is another essential research dimension. Qadeer et al. [20] proposed an RF classifier algorithm-based ML model to estimate an air-based energy system’s relative humidity (RH). Their study linked the Aspen HYSYS v10 process simulator with MATLAB 2019a to provide an environment necessary for data mining processes. They employed the dry and wet bulb thermometer to compare the results of their model, where an absolute deviation was observed. They further analogized the RF performance to an SVM and discovered a better overall performance of about 74.4%. Also, some researchers opted to exploit the high prediction accuracy of convolutional neural network (CNN) models. For instance, Jung et al. [21] used a CNN to monitor and regulate different inhabitants’ indoor temperature and energy consumption. Their work used a 1D CNN to recognize inhabitants’ activities and reinforce learning (RL) models for indoor temperature control. Their findings showed that the system automatically controlled the indoor temperature in real-time, with a 10.9% decrease in thermal discomfort and preserved energy consumption. Kreuzer et al. [22], however, used long short-term memory (LSTM) and convolutional LSTM (convLSTM) networks to enhance temperature forecasting compared to the traditional naive forecast approach. They used five different weather stations to put up a case study in Germany and observed that their models worked best in longer horizons prediction. After using numerous datasets to check the systems’ performances, they deduced that convLSTM provided the best results.

In attempting to improve performance accuracy and efficiency in IoT devices, scores of researchers embedded LPWAN technologies with either machine learning (ML) or deep learning (DL) techniques. Parvez et al. [23] suggested a development architecture for the LoRa gateway communication technology by employing Artificial intelligence (AI) techniques. The study used the MySignals development shield, Arduino medical devices, and eHealth development platform to remotely monitor and maintain the sensor. Reduced complexity, minimum decision-making delay, and less deployment cost were the crucial findings of such integration. LPWAN technologies are specifically faced with payload size constraints compared to other wireless technologies. To minimize these issues, Bernard et al. [24] embedded a deep LSTM model in the LoRa sensor and network infrastructure to improve data transmission efficiency by using the occupancy dataset obtained from cellular base stations to train the model. They suggested a lossy compression technique, significantly reducing the gateway’s sensor data traffic. They further analyzed the system’s lossy compression error rate, energy consumption, and model performance, and the outcome was remarkable. Another noteworthy investigation was conducted by Ouameur et al. [25] that employed ML and DL models to measure and localize an indoor location by leveraging the LoRaWAN channel state information in a dataset depicting four different spots. The framework further provided entry to the sensor data and physical layer where information like the spreading factor, the receiver’s signal strength, frequency hopping signature, etc., can also be accessed. They used a multilayer neural network to predict these locations with 98% accuracy. Wearable devices such as smartphones and glasses are becoming more and more integrated into our lives. Despite the immense benefits of monitoring peoples’ fitness, health conditions, and so on, they are greatly limited by relying on smartphones and tablets for decision-making and data transmission to the internet. Sanchez-Iborra [26], in his work, reviewed and examined the aspects of embedding LPWANs (LoRaWAN) with tiny ML (TinyML). His research work further conducted an experimental study and investigated coverage and intelligence of the integration. The investigation eventually inferred that these paradigms fit the next generation of wearable devices. Loukatos et al. [27] used a combination of LoRa and Wi-Fi networks and remote cloud service-based voice recognition (VR) engines to enable smartphones to operate specific farming chores for the elderly and crippled remotely. They further used the embedded boards to facilitate offline VR in case of a shortage of internet which warranted remote connection to the VR servers. Also, the end nodes were able to reply by voicing the status of the current farming activity by playing a prerecorded message or via speech synthesis technique.

This section provides information on the materials and methods used to develop the designed system. The system comprises two parts: the end device responsible for sensing changes in temperature and humidity and the gateway that receives the transmitted data from the end device. The gateway then encrypts and forwards the final packet to TTN. Both parts require software and hardware components to accomplish the mentioned task. The hardware provides the avenue for interacting with the physical environment and houses the MCUs and other essential system elements. The software contains the programs and algorithms that control the hardware through designated libraries and modules like the SPI modules for MCU to I/O units communication, radio head module for RF transceiver breakout configuration, and DHT library for controlling the DTH11 sensor, and many more.

4.1 Software requirement

Arduino IDE: The Arduino Integrated Development Environment (IDE) is open-source software that enables coding, debugging, and code upload to the Arduino physical board. The Arduino IDE is available for Windows, Linux, macOS, and even web versions for working online and is frequently updated for bugs and necessary changes [31]. This study utilized Windows version 1.8.19 of the Arduino IDE.

Raspberry Pi Imager: The Raspberry pi OS, formerly known as Raspbian, is a free Debian-based operating system improved for the Raspberry Pi system-on-chip hardware. It has more than 35,000 precompiled software bundles for use by the hardware [32]. However, the Raspberry Pi Imager is software that provides a fast and straightforward approach to installing the different versions of the Raspberry Pi OS on the microSD card, which is later used by the hardware [33].

PuTTY: PuTTY is also an open-source secure shell (SSH) and Telnet client software for reliable network services [34]. The software enables us to connect to the Raspberry Pi desktop environment via the headless connection method. PuTTY is also available for Windows and Linus OS.

Single Channel Packet Forwarder: A packet forwarder is a program that facilitates the transmission of uplink and downlink packet data between a concentrator (LoRa transceiver module) and the network server via IP/UDP communication protocol [35]. There are two packet forwarders types: single-channel packet forwarders and multi-channel packet forwarders. The implemented system used the single packet forwarder to develop a Raspberry Pi LoRa gateway. However, only the multi-channel packet forwarder is fully LoRaWAN compatible.

RadioHead Library: The RadioHead library is a radio packet library enabling the transceiver module to send and receive a message in a packet format through the embedded MCU [36]. It allows the RFM95W transceiver module on the end device side to transmit sensor data with the help of an embedded microprocessor, in this case, Arduino UNO.

Adafruit DHT11 Library: The DHT11 library is a simple Arduino library designed to control the low-cost TH sensor provided by Adafruit Industries (an open-source hardware company that designs and manufactures numerous electronic components and various products).

Arduino LMIC Library: Arduino LMIC library is a modified version of the IBM LMIC (LoraMAC-in-C) designed to operate in the Arduino environment. This library permits SX1272/76 transceivers and several RFM9x compatible transceiver modules for data transmission [37].

4.2 Hardware requirement

Raspberry Pi 3 B+: The raspberry Pi, 3 Model B+, is used in implementing the single-channel LoRa gateway. The Pi board is the final version of the third generation single-board computers manufactured by Raspberry Pi Foundation, a partner of Broadcom based in the United Kingdom. Raspberry Pi 3 B+ board is a powerful chip-based computer with several impressive capabilities and specifications. The board is built with a 1.4GHz 64-bit quad-core processor, faster Ethernet, dual-band wireless LAN, and Power-over-Ethernet (PoE) support. It also has 1GB RAM, a 4.2/BLE Bluetooth, 4 USB 2.0 ports, a 40 GPIO pin, full-size HDMI, a microSD port for OS, and data support, to name a few [38]. It is a fantastic board that is very affordable and useful in developing many stunning applications.

Arduino Uno: The second outstanding MCU unit employed is the Arduino Uno board. Arduino Uno board is a popular board that has been widely used to make many applications due to its simple design and functionality. It is widely considered the starting point for developing IoT applications. Arduino Uno is a 63.6 x 53.4 mm board designed with an Atmega328P microcontroller. It has 14 digital I/O pins (out of which six (6) offer Pulse-width modulation PWM) and six (6) analog input pins. Uno board has other features like 16 MHz clock speed, 2 KBstatic random-access memory (SRAM), and 1 KB electrically erasable programmable read-only memory (EEPROM), etc. Uno is derived from Italian, which means one, and it marks the first Arduino IDE version 1.0. [39].

Adafruit RFM95W LoRa Radio Transceiver Breakout 868 MHz: RFM95W LoRa radio module is the key component in this implementation. For long-distance wireless data transfer, two endpoints, typically a transmitter and receiver (or transceivers), are necessary to send and receive data. Two Lora radio modules are required to develop the system in the same light. These radio modules enable packetized data transmission between the end node and gateway, provided their radio modules are designed with the same frequency and use the same encryption key. They are also easily compatible with Arduino Uno and Raspberry Pi boards. Both use the same 868 MHz European license-free ISM bands following the LoRaWAN regional parameters with a simple wire antenna. The transceivers are based on the SX1276 LoRa module with the SPI interface capable of transmitting sensor data over about 2 Km depending on several factors such as antenna type, frequency, obstacle, and power out [40].

Whip or Wire Antenna: Antennas are essential components in wireless communication. They act as an interface between the transmitted radio waves and the electric current moving in the metallic conductors of the system. The LoRa radio modules facilitate using different antennas connection like the regular wire, u.FL or RP-SMA connection. This system used the wire antenna for cost-effectiveness and range up to 2 km [41]. The length of the antenna was computed using the quarter-wave whip antenna approach and was obtained by dividing the radio transceivers’ wavelength by a value of 4, as given below.

$\lambda=\frac{v}{f}$      (1)

where, λ=Wavelength, v=Velocity, f=Frequency.

Here, the velocity of a radio signal in a vacuum is 299,792,458 m/s. And the frequency of the radio module is 868 MHz. Therefore, we can convert it to Hz by adding six zeros resulting in 868,000,000 Hz. Therefore,

$\lambda=\frac{299,792,458}{868,000,000}, \lambda \approx 0.345 \mathrm{~m}, \lambda \approx 34.5 \mathrm{~cm}$      (2)

After dividing the wavelength by 4, we obtained the desired length of the wire antenna as follows:

$l=\frac{\lambda}{4}, l=\frac{34.5}{4}, l \approx 8.6 \mathrm{~cm}$     (3)

DHT11 TH Sensor: The DHT11 sensor is a low-cost, basic device widely employed in several long-term IoT prototyping for measuring TH in its surrounding environment. The component uses a capacitative humidity sensor and a thermistor to make these measurements. It contains an 8-bit chip for analog and digital signal conversion. The chip also emits the TH digital signal values, which are easily detected by the Arduino board. It can measure temperature between 0 to 50℃ with a 2℃ error margin. Also, with the DHT11, values ranging from 20 to 90% for the humidity measurement are detected with a 5% error margin. The DHT11 used in this design is a three-pin type, eliminating the need for a resistor. Other features of the DHT11 sensor include 3 to 5 Vpower with an I/O pin, 2.5 mA maximum current in requesting data, less than 1 Hz sampling rate, etc. [42].

Other Components: Other essential but straightforward components used in accomplishing the design include two half-size breadboards for testing, a 9 V battery power supply for the Arduino board, a 5 V regulated adapter, and a 16 GB microSD memory card for the Rasberry Pi board, soldering iron and lead, and 22 AWG solid-core hook up wires.

4.3 Simulation and breadboarding

As can be seen in Figure 3. the complete setup of the end device and gateway developed with fritzing design automation software was appropriately shown.

3.png

Figure 3. Breadboard designs of the end device and raspberry Pi 3 B+ LoRa gateway

The software provides a simple GUI for the full system realization. For the end device, all the required components, namely Arduino Uno board, DHT11 sensor, Adafruit RFM95W Lora radio module, 10 kΩ resistors, and antenna, were arranged and connected accordingly. The LoRa radio module and DHT11 libraries had to be imported into the software as they were initially unavailable. According to the design, all connections were ensured, and the system was set and ready for testing with the breadboard. The same approach was extended to the gateway setup, comprising only three components: Raspberry Pi 3 B+, Adafruit RFM95W Lora radio module, and antenna. The fritzing breadboard setups for the end device and gateway are ready for implementation.

4.4 System implementation and TTN registration

The Adafruit RFM95W Lora radio modules come with a set of 0.1-inch male header strips that need to be soldered to facilitate their usage with the breadboard or a universal printed circuit board (PCB) for a more concrete system realization. However, the universal PCB was not used in our cases as the complete soldering of the system is not required.

Table 1. Arduino Uno, RFM95W Radio Module, and DHT11 Wiring Connection

Table 2. Raspberry Pi and RFM95W Radio Module Wiring Connections

Only the male header strips were soldered to the radio module for proper deployment on the breadboard—Tables 1 and 2 outline the wiring connections for the end node and gateway and their respective components. The end device and gateway units were later configured after soldering the male headers to the radio modules and properly making the connections shown below. To configure the gateway for remote connection, the Raspberry Pi OS was written on the microSD card with the help of the imager. Afterward, the memory card is inserted into the Raspberry board to enable the desktop environment. There are many ways of accessing the Raspberry OS ranging from the headless connection via the SSH protocol to using software that allows virtualization. The board can also be directly connected to an external monitor through the HDMI port and can be controlled with a keyboard and monitor just like a regular computer. The approach used in this implementation is the headless connection approach to minimize cost. The PuTTY software facilitated the SSH headless connection. Then the Raspberry OS was updated through the command prompt, and the SPI function was allowed for communication between the board and its pins. Other requirements for the board are installing the wiringPi library and cloning the single-channel packet forwarder repository from Github, which can be found here [43]. The final actions for the gateway realization are setting the frequency of the radio module (868 MHz) and TTN network server IP (63.34.215.128) in the C++ file (“main.cpp”) from the clone folder. These actions were essential for the remote connection to TTN.

After setting up the gateway and making the end device ready, they were registered on TTN for online access. An account had to be created to enable the gateway, application registration, and further configuration and can be found here [44]. Upon completion of the above process, the network session key (NtwkSKey), application session key (AppSKey), and device address (DevAddr) were taken for activation. There are two activation methods on TTN: over-the-air activation (OTAA), which automatically activates the devices, and activation by personalization (ABP). The required details are manually typed in the end device program for successful packet transmission from the gateway to TTN. This design employed the ABP activation method, and the three necessary parameters NTWKSK, APPSK, and DEVADDR were used in the code. Using the Arduino IDE, the Arduino-LMIC code [45] was updated with the above parameters and other necessary details such as the Arduino activation code for the DHT11 sensor, the pin mapping for the RFM95W LoRa radio module, and the end node frequency. The updated program was compiled and uploaded to the Arduino board, which sent the LoRaWAN packet with TH sensor values as payload to TTN with a matching frequency and encryption. After finally completing the system realization, the Raspberry pi gateway successfully sent the encrypted uplink message. However, to see the actual sensor value, TTN provides several payload formatters like JavaScript, Cayenne low power payload (LPP), etc., to decrypt the payload into human-readable formats just like in the Arduino IDE serial monitor. As simple as it seems, this task was challenging due to so many changes and recent updates in TTN. However, despite such challenges, the payload was finally decrypted using the JavaScript payload formatters option.