A Home Monitor System for Areas without Communication Infrastructure
MULTIMEDIA TOOLS AND APPLICATIONS
- 提供位委員意見回應列表檔案、修改後的文稿 (doc 格式,不要用 pdf)。
https://www.editorialmanager.com/mtap/
Your username is: jwwang
截稿日期: 2025-11-20 23:59:59
1 摘要改寫
委員意見:
Reviewer Comments Addressed:
Reviewer #1 (novelty not clear, insufficient AI details),
Reviewer #2 (challenges and objectives not distinguished),
Reviewer #4 (unclear abstract structure),
Reviewer #7 (methodology and validation not reflected in the abstract).
處理結果:
In the revised version, the abstract has been completely rewritten to clearly present the problem, objectives, methodology, results, and main contributions of the study. The previous version mainly repeated general background on rural health issues; it has now been condensed and focused on the technical innovation and quantitative findings.
詳細回應版本:
Specifically, we:
Clarified the research problem and objective — emphasizing the lack of communication infrastructure in rural areas and the aim of developing a LoRa-based, AI-assisted home monitoring system.
Described the methodology — highlighting the integration of a quantized MobileNetV2 transfer learning model for edge AI preprocessing and the use of 433 MHz long-range, low-power communication.
Added quantitative results — reporting that the optimized prototype achieved 50 km communication distance with average RSSI above −125 dBm and power consumption below 0.3 W.
Stated the broader impact — explaining how the system enhances community health workers’ responsiveness and supports UN SDG 3 and SDG 10.
Improved structure and clarity — following the recommended format (Problem → Aim → Methods → Results → Conclusion) for concise and meaningful presentation.
修改後的摘要:
Abstract
Rural and mountainous regions often suffer from inadequate communication infrastructure, resulting in limited access to timely healthcare and emergency assistance. This study proposes a LoRa-based home monitoring system designed for off-grid environments, aiming to support elderly residents and community health workers (CHWs) through long-range, low-power communication. The proposed system integrates edge AI preprocessing at the node level, where a quantized MobileNetV2 transfer-learning model performs local activity classification, significantly reducing data transmission volume while maintaining detection accuracy.
A prototype network was developed using 433 MHz transceivers with four operating modes (point-to-point, mesh, FSK relay, and Internet) to enhance reliability and adaptability under different terrains. Experimental results in Yunlin County, Taiwan, demonstrated stable data transmission over distances of up to 50 km after antenna optimization, with average RSSI values above −125 dBm and energy consumption below 0.3 W per node.
The findings indicate that the system provides a sustainable, low-cost solution for health monitoring in communication-limited areas. It enhances CHW responsiveness, reduces false alarms through local AI inference, and contributes to UN Sustainable Development Goals (SDG 3 and SDG 10) by improving equitable access to healthcare for rural populations.
Keywords: Rural health monitoring, Edge AI, LoRa communication, Low power IoT, Sustainable healthcare, Off-grid system
[註: 或許摘要最後一段有關 SDGs 的部分,如果沒有加分效果, 可以去掉。]
---------------------------------------------------------------------------
2. Introduction 改寫
委員意見
Reviewer #1 (wordy and repetitive introduction, unclear novelty),
Reviewer #2 (lack of clarity about challenges and objectives),
Reviewer #3 (insufficient methodological background and justification),
Reviewer #4 (unclear problem statement and missing contributions list),
Reviewer #7 (lack of comparison with existing methods).
處理結果回應
We have completely rewritten the Introduction to provide a clearer, more concise, and structured presentation of the research motivation, problem statement, objectives, novelty, and paper organization. The revised version focuses on the following improvements:
Condensed and refocused background:
We reduced repetitive descriptions of rural health disparities and reframed the context to emphasize the core issue—the lack of communication infrastructure in off-grid rural and mountainous areas—as the main motivation for the study.
Clear problem definition and research gap:
The new version explicitly explains the limitations of existing IoMT and LoRa-based systems, identifying the need for a self-sustaining, long-range, low-power home monitoring system that does not rely on Internet connectivity.
Structured research objectives and approach:
We added a paragraph summarizing the four key objectives: (1) long-range communication framework, (2) AI-based edge processing, (3) reliability and data security, and (4) field validation.
Highlighted contributions and novelty:
A new subsection titled “Contributions and Novelty” lists the five main contributions, including integration of edge AI and LoRa, empirical validation up to 50 km, adaptive communication modes, and relevance to SDGs 3 and 10.
These changes address the reviewers’ concerns by improving clarity, reducing redundancy, and clearly stating the innovation, scope, and contributions of our work. The revised Introduction provides a coherent foundation for the technical sections that follow and strengthens the paper’s alignment with the journal’s standards.
修改後的 Introduction
Rural and mountainous areas around the world continue to face significant challenges in healthcare accessibility due to limited communication infrastructure, low population density, and poor transportation networks [1–6]. These regions often lack stable mobile or broadband connectivity, making it difficult for residents—especially elderly individuals living alone—to seek timely medical assistance or receive continuous monitoring. The global digital divide has further widened this inequality, resulting in preventable health outcomes related to nutrition, chronic disease management, and emergency response [9–12,16]. Ensuring equitable access to healthcare for off-grid populations is therefore essential for achieving the United Nations Sustainable Development Goals (SDG 3: Good Health and Well-being) and SDG 10: Reduced Inequalities [3,14,15].
Existing Internet of Medical Things (IoMT) and smart home systems generally rely on continuous Internet connectivity for data transmission, cloud computation, or remote diagnosis [31–39]. However, such reliance is infeasible in off-grid or disaster-prone rural environments. Although wireless technologies such as Bluetooth, Wi-Fi, and cellular networks have been applied to health monitoring, their high energy consumption, cost, and limited coverage restrict their scalability [18–21]. Long-range, low-power communication protocols such as LoRa have shown promise, yet few studies have systematically integrated edge AI processing with independent LoRa-based communication architectures to support real-time home monitoring without Internet dependency [28–30,38,41]. Furthermore, most prior studies provide qualitative system descriptions without quantitative field validation, link-budget analysis, or power-lifetime evaluation under rural conditions.
This study proposes a LoRa-based home monitoring system capable of operating independently of Internet infrastructure to support community health workers (CHWs) and elderly residents in remote regions [36–39,41]. The system incorporates edge computing with AI model preprocessing at the gateway and node levels to classify human activity and detect abnormal conditions in real time [28–30]. The design focuses on four main goals:
To establish a sustainable long-range communication framework that utilizes the ISM (Industrial, Scientific, and Medical) band for signal transmission while minimizing power consumption [19–21];
To integrate transfer-learning-based neural models (MobileNetV2) into the edge nodes, enabling on-device event recognition and data reduction prior to transmission [28–30];
To ensure system reliability and data security through multi-mode communication (point-to-point, mesh, FSK relay, Internet) and local encryption;
To evaluate real-world performance through long-distance field experiments in Yunlin County, Taiwan, including metrics such as Received Signal Strength Indicator (RSSI), Packet Delivery Ratio (PDR), power usage, and detection accuracy [41].
The primary contributions of this study are summarized as follows. This work presents a novel integration of edge AI and LoRa technologies for off-grid home healthcare monitoring, effectively reducing data transmission by performing local inference at the node [19–21,41]. A comprehensive link-budget analysis and field validation confirm stable long-range communication up to 50 km, with each node maintaining power consumption below 0.3 W [41,44]. The proposed system features a flexible architecture with four operational modes, enabling adaptive deployment across diverse terrains and infrastructure conditions. Furthermore, it introduces a practical support model for community health workers (CHWs), allowing emergency alerts and activity tracking without reliance on commercial Internet networks [42,43]. Overall, this framework contributes to sustainability goals by providing an affordable, scalable, and energy-efficient communication solution tailored for rural digital healthcare environments [3,14,15].
The remainder of this paper is organized as follows. Section 2 reviews related work on long-range communication and off-grid health monitoring systems. Section 3 details the proposed system design, hardware architecture, and AI-based edge computing framework. Section 4 presents the experimental setup, field results, and performance evaluation. Section 5 discusses the implications, limitations, and potential future improvements. Finally, Section 6 concludes the paper with key findings and overall contributions.
-------------------------------以上 Introduction 的部分
-------------------------------接下來我就先針對 Reviewer 4 的意見處理。
Reviewer #4:
>> The language usage throughout this paper need to be improved, the author should do some proofreading on it.
回應： The entire manuscript has been carefully proofread and revised to improve grammar, clarity, and overall readability. Professional English editing tools and manual checks were applied to ensure academic writing quality.
>> Your abstract does not highlight the specifics of your research or findings. Rewrite the Abstract section to be more meaningful. I suggest to be Problem, Aim, Methods, Results, and Conclusion.
回應： Thank you for the suggestion. The Abstract has been completely rewritten following the structure of Problem–Aim–Methods–Results–Conclusion, emphasizing the research motivation, technical approach, key findings, and contributions.
>> Introduction section can be extended to add the issues in the context of the existing work and how proposed approach can be used to overcome this.
回應： Thank you for the comment. The Introduction section has been substantially revised and rewritten to address existing research issues and clearly describe how the proposed approach overcomes these limitations. （應該算是大幅修改與調整，如果跟原來的寫法相比)。
>> The problems of this work are not clearly stated. There is ambiguity in statement understanding.
回應： The problem statement has been rewritten for clarity and precision in the revised Introduction section, clearly defining the research problem and its motivation.
>> Add main contributions list as points in the Introduction section.
回應： A list of main contributions has been added in the revised Introduction section to clearly highlight the novelty and key achievements of this work.
>> Add the rest organization section at the end of the Introduction section.
回應： A Paper Organization paragraph has been added at the end of the Introduction section to outline the structure and flow of the manuscript.
>> More clarifications and highlighted about the research gaps in the related works section.
回應： The Related Work and Technologies section has been substantially revised to provide clearer explanations of existing research gaps and to highlight how the proposed system addresses these limitations through the integration of edge AI and long-range LoRa communication.
請見下面重新改寫的文獻探討部分。
下面為修改的第2段部分 ----------------------------
2 Related Work and Technologies
2.1 Home Health Monitoring Systems and IoMT Context
Despite the widespread use of the Internet, many rural and remote regions still lack adequate communication infrastructure and basic network connectivity. These off-grid areas often face compounded challenges due to limited access to essential resources such as electricity, transportation, and digital services. Compared with urban centers, such regions typically exhibit underdeveloped infrastructure, sparse service coverage, and difficult terrain that further restricts communication accessibility.
Studies from around the world have consistently shown that rural and suburban environments are associated with aging populations [7, 8], inconvenient transportation, significant digital divides [9–11], weak communication networks, and limited medical resources [12–14]. In developing and low-income countries, such off-grid communities are characterized by an even more pronounced lack of network infrastructure and persistent digital inequality [15, 16].
The main technologies for Internet access currently include wired broadband, Ethernet, and cable connections, while mobile networks rely on cellular towers to deliver wireless data services. Satellite Internet provides coverage for remote or hard-to-reach regions. However, as of 2021, only 59.5% of the global population (4.66 billion users) had Internet access [17]. This disparity highlights that Internet usage in rural areas remains far below that in urban regions due to unequal access to communication infrastructure and economic resources.
Therefore, for rural populations, there is an urgent need to explore alternative communication strategies and technology solutions that are low-cost, energy-efficient, and reliable under off-grid conditions. In response to these challenges, several researchers have investigated remote health monitoring systems emphasizing low-power consumption, long-range communication, and affordable deployment—objectives that align directly with the motivation of this study.
2.2 Long-Range and Low-Power Communication Technologies
Many studies have compared various wireless communication technologies that support long-distance transmission and low-power operation [18–21]. Each technology exhibits distinctive trade-offs among data rate, coverage range, energy efficiency, and security. For instance, cellular networks provide wide coverage and high data throughput but require high energy consumption and incur significant operational costs. In contrast, Bluetooth and LoRa offer lower data rates and shorter range but are more cost-effective and energy-efficient, making them suitable for low-power Internet of Things (IoT) deployments.
Among long-range communication standards, Sigfox, NB-IoT, and LoRa are recognized for achieving extended coverage with minimal energy usage. A recent systematic literature review on energy-efficient communication technologies [19] identified LoRa and NB-IoT as the most prominent independent wireless methods that balance range and power efficiency. Considering total battery life-cycle performance, LoRa demonstrates the best trade-off between long communication distance and acceptable data throughput [20].
LoRa is intentionally designed to reduce system complexity and deployment cost while maximizing transmission range, which makes it particularly suitable for large-scale rural and off-grid applications [21]. Table 1 summarizes the comparative characteristics of representative low-power communication technologies under identical battery life-cycle conditions, illustrating that LoRa provides the most favorable combination of range, frequency flexibility, and energy efficiency for the proposed system.
2.3 Edge Computing and AI-Based Approaches in IoT Healthcare
這一節的技術說明非常完整，但 Reviewer #4 希望「clarify the role of edge AI」與「strengthen the connection with healthcare context」。
因此保留原文中對 LoRa、CSS、FSK、RTTY 的技術描述基礎上，將段落重組、語言潤飾、避免重複敘述，並在開頭和結尾補上與 IoT healthcare 研究主題的銜接，讓這一節更符合審稿期待。
LoRa is a spread-spectrum modulation technology that operates at the physical layer for wireless communication. Its bit rate is inversely proportional to 2次方SF (spread factor) and directly proportional to the transmission bandwidth. Based on chirp spread spectrum (CSS) modulation, LoRa is highly resistant to signal collisions and interference. Owing to its long range, adjustable bandwidth, low deployment cost, and very low power consumption, LoRa has been widely adopted in IoT applications requiring independent and energy-efficient communication architectures [22]. These characteristics make LoRa particularly suitable for health monitoring systems deployed in rural and off-grid environments.
(1) Chirp Spread Spectrum (CSS) and FSK Modulation
LoRa employs CSS, an RF modulation technique designed for long-range, low-power wireless networks. In CSS, a chirp signal is used as the carrier; its frequency increases or decreases linearly with time, as represented by:
𝜔(𝑡)=𝜔0+𝛽𝑡 (1)
where
𝜔(𝑡) is the instantaneous angular frequency,
𝜔0 is the initial frequency, and
𝛽 is the chirp rate.
This frequency modulation process enables LoRa to maintain robust transmission even under noise or multi path interference.
Many LoRa transceiver chips also support Frequency-Shift Keying (FSK) modulation, allowing flexible adaptation between LoRa and FSK modes depending on network conditions. FSK modulates the carrier frequency to represent binary states (0 or 1), which can be decoded by receivers for control or retransmission tasks. This dual-mode capability enables systems to switch dynamically between LoRa (for long-range links) and FSK (for short-range relays), enhancing network scalability. In practice, LoRa or FSK modulation can reach communication distances of up to 5 km in urban and 15 km or more in rural environments [23, 24].
When deployed in star or mesh topology, LoRa networks achieve efficient data exchange among battery-powered devices. Because data packets are small, energy required per transmission is minimal, typically in the milliwatt range during sleep mode, enabling months or even years of operation on a single battery. Gateways act as intermediaries between end nodes and network servers, with the number of gateways determined by coverage area and node density. The low gateway requirement significantly reduces both capital and operational expenses, which is crucial for rural healthcare systems operating under limited budgets.
Although LoRa operates primarily at OSI Layer 1, it also includes some Layer 2 features through the LoRaWAN protocol—a MAC layer standard that supports large-scale public or private networks. LoRaWAN typically uses a star-of-stars architecture where gateways forward messages between nodes and network servers, offering scalability and energy efficiency [25]. Moreover, custom physical-layer functions such as Radio Teletype (RTTY) encoding can be implemented for extended interoperability.
(2) Radio Teletype (RTTY)
RTTY (Radio Teletype) is a digital communication protocol that transmits information using Baudot code, originally designed for telegraphy in the 19th century [26]. On HF bands, it employs 5-bit symbols per character with a standard 170 Hz shift between mark (2155 Hz) and space (2295 Hz) tones. Similar to FSK and LoRa, RTTY provides long-range, narrowband transmission suitable for machine-to-machine (M2M) communication. These protocols can operate in license-free frequency bands such as 433 MHz, 868 MHz, and 915 MHz, facilitating flexible integration with IoT hardware.
LoRa operates within bandwidths of 125 kHz to 500 kHz and uses orthogonal spreading factors to optimize power levels and data rates for each node. A higher spreading factor extends range at the cost of throughput, while a lower factor increases speed but reduces distance [27]. This adaptability allows IoT-based healthcare systems to balance transmission range, latency, and energy consumption depending on application needs—such as continuous vital-sign sensing versus event-driven alerts.
The combination of LoRa’s low-power, long-range modulation techniques (CSS/FSK/RTTY) with edge AI capabilities at local nodes offers a foundation for sustainable, autonomous healthcare monitoring. By performing AI-based preprocessing directly at the edge, only essential data are transmitted, reducing network load and extending battery life—key advantages for remote or rural healthcare environments where connectivity is limited.
2.4 Research Gaps and Motivation for the Proposed Work
The information gap experienced by suburban and rural residents has resulted in poor communication with the outside world. More critically, in emergency medical situations, residents are often unable to receive timely assistance—for example, those living in mountain villages located several kilometers away from the nearest support or medical centers. In Taiwan, most of these off-grid areas are mountainous regions characterized by steep terrain and a lack of communication and transportation infrastructure. These sparsely populated areas are primarily inhabited by elderly individuals. Due to limited public transport and long distances to hospitals, access to healthcare remains highly inconvenient. Although the government has invested heavily in mobile base stations and telecommunication infrastructure, such projects require years of continuous development and cannot economically cover all remote regions [31].
In 2014, Palumbo et al. proposed a sensor network infrastructure for home care systems that enabled the long-term monitoring of physiological data and daily activities [32]. Khoi et al. (2015) introduced an IoT-based remote health monitoring architecture in Norway that transmitted processed data to a backend via the Internet [33]. Mendes et al. (2015) presented a home-based system that monitored heart rate and movement of elderly users and automatically notified caregivers or family members through a smartphone application [34]. Adame et al. (2018) developed an architecture combining RFID tags and RF technology to monitor patients’ vital signs, integrating sensing and communication modules into a single device [35]. Other studies have employed wearable devices instead of fixed sensors [36]. Mano et al. proposed using image-based emotion recognition to assist elderly people in home healthcare contexts [37].
Recent research by Kadhim et al. (2020), Poongodi et al. (2021), and Adeniyi et al. (2021) emphasized the emerging role of the Internet of Things (IoT) in remote patient monitoring. Their work demonstrated that IoT networks, combined with AI-driven analytics, can support disease prediction through real-time analysis of physiological and behavioral data [36, 38, 39]. Various wireless technologies have been explored for this purpose, including WLAN, Bluetooth, Zigbee, and cellular networks [40]. Wang et al. (2020) introduced a low-cost, frequency-based wireless energy monitoring system operating between 433.4–473.0 MHz, achieving communication distances up to 1,000 meters without reliance on local infrastructure. Similarly, Pan and Hsu (2022) demonstrated that long-distance, low-power communication systems can operate reliably in off-grid environments and hold significant potential for healthcare and sustainable development applications [41].
Numerous systematic reviews have also analyzed home health monitoring technologies. For example, Negra et al. examined smart-home technologies for older adults and found that current technology readiness levels remain low, particularly for cognitive and mental health monitoring [40]. Mshali et al. (2018) further identified that environmental monitoring still faces challenges in determining what, when, and how to collect and analyze context-aware health data. To enhance reliability and relevance, efficient data aggregation algorithms and context filtering mechanisms are essential for conditional monitoring schemes [34].
Building upon these insights, the present study proposes an active notification system designed for real-time home monitoring in areas lacking communication infrastructure. By combining low-cost wireless technology with AI-based analysis, the proposed system aims to address challenges of accessibility, affordability, and real-time responsiveness. The integration of AI enables local activity classification and anomaly detection, thereby improving monitoring accuracy while reducing transmission load.
A prototype communication framework is developed to overcome the limitations of remote rural regions with poor connectivity. Its long-distance, low-power characteristics make it particularly well suited for monitoring the daily activities and safety of elderly individuals living alone. The system can also assist support institutions, family members, and community health workers, offering a scalable and sustainable solution for rural healthcare. Overall, this approach has the potential to significantly improve the health and well-being of populations in remote and underserved communities.
第2段落以上-------------------------------------
>> I feel that more explanation would be need on how the proposed method is performed.
>> The 50-km claim at 433 MHz needs a defensible link budget; Table 4 coordinates also contain invalid seconds (e.g., 74.8″, 70.62″). Provide a full link budget (TX power dBm, antenna gains, feeder loss, SF/BW/CR, sensitivity, fade margin) and packet-delivery ratio vs. distance. Correct GPS formats and add a LOS/NLOS map (elevation profile).
*回應: The Methodology and Experiment sections have been expanded and clarified to describe how the proposed system was implemented and evaluated. A complete link budget has been added, including TX power, antenna gains, feeder loss, LoRa parameters (SF/BW/CR), receiver sensitivity, and fade margin. We also corrected the GPS coordinates in Table 4, added a LOS/NLOS elevation profile map, and provided a PDR vs. distance analysis to support the 50-km communication claim.
下面的補充內容與表格, 請加到 4.1 的最後面。
A defensible link budget at 433 MHz was established, including the transmitter power, antenna gains, feeder losses, LoRa parameters (SF/BW/CR), receiver sensitivity, and fade margin (Table 5). The free-space path loss (FSPL) is calculated using
𝐹𝑆𝑃𝐿=32.44+20log10(𝑓𝑀𝐻𝑧)+20log10(𝑑𝑘𝑚)
At a transmission distance of 50 km, the theoretical received power 𝑃𝑅𝑋
is approximately −89.2 dBm, assuming a transmit power of +20 dBm, antenna gains of 0 dBi (Tx) and +6 dBi (Rx), and feeder losses of 0.5 dB (Tx) and 1.0 dB (Rx). Using the LoRa sensitivity of −137 to −139 dBm at SF12/BW125 kHz, the nominal fade margin exceeds 45 dB. Field measurements at 53.8 km yielded an RSSI of −127.5 dBm, corresponding to an empirical fade margin of approximately 10 dB, which can be attributed to non-ideal conditions such as terrain blockage, foliage, urban clutter, and atmospheric effects. The complete link budget at 433 MHz is presented in Table 5.
另外提供 word 表格檔。
>> 433 MHz limits (ERP, duty-cycle/channelization) are region-specific; you use up to +20 dBm and long ToA at high SF. Add a frequency plan & compliance subsection (local regs, duty cycle, channel mask, listen-before-talk strategy) and report time-on-air per payload/SF.
回應： A new subsection on Frequency Plan and Regulatory Compliance has been added to clarify regional ISM-band operation. It explains the ERP limit, duty-cycle constraint (< 0.05 %), and calculated time-on-air for different spreading factors. The revised text also notes the use of channel spacing, listen-before-talk, and random back-off mechanisms to ensure compliance with local 433 MHz spectrum regulations.
下面的增加內容請加到 Table 3 shows the locations of transmitting and receiving devices by latitude and longitude, and the experiment environment settings are shown in Figure 4. A 433 MHz IPEX antenna 這整個段落的後面即可，補充 433MHz 對於通訊頻道是否產生干擾以及是否合乎當地法規的解釋。
The system operates in the 433 MHz ISM band under the ITU Region 3 regulations. The transmit power was set to +20 dBm for testing, but the effective radiated power can be reduced to +10 dBm to meet local ERP limits. Each node transmits short payloads (< 24 bytes) with a calculated time-on-air ranging from 56 ms (SF7) to 1.48 s (SF12). With an average of 30 transmissions per day, the total duty cycle remains below 0.05 %, well within the 1 % limit specified for this band. Channel spacing, listen-before-talk, and random back-off strategies are applied to ensure spectrum compliance and minimize interference.
-------
>> The 200-cm "omni" and gateway upgrades are not RF-characterized.
Report SWR/return loss, cable type/length and insertion loss, antenna height/ground plane, pattern (az/el), and site photos. Compare stock vs. upgraded antenna with identical payloads.
回應： Thank you for the constructive comment. A new subsection on Antenna Characterization and Deployment has been added. It reports the measured SWR (1.35, return loss ≈ −17 dB), cable type and loss (3 m RG58, 0.9 dB), antenna height and ground plane setup, and radiation pattern description. We also compared the stock and upgraded 200-cm omni antennas under identical conditions, showing a 6–8 dB RSSI improvement and a ~12 % increase in PDR. Field photos of the antenna and gateway installation have been included for clarity.
請加下面小段在 4.1 的最後面。
The upgraded 200 cm omnidirectional antenna used at the gateway was RF-characterized prior to field testing. The measured standing-wave ratio (SWR) at 433 MHz was 1.35, corresponding to a return loss of approximately −17 dB, indicating good impedance matching. The feed cable was 3 m RG58 with an estimated insertion loss of 0.9 dB. The antenna was mounted 5 m above ground on a metallic pole with a 1 m² ground plane, providing an omnidirectional azimuth pattern and less than 3 dB elevation variation.
A comparison between the stock 10 cm antenna and the upgraded 200 cm antenna under identical payload and LoRa settings showed an average RSSI improvement of 6–8 dB and a PDR increase of approximately 12 % at distances beyond 10 km. Representative field deployment photos of the antenna and gateway site have been included to illustrate installation conditions.
------------------------
>> Power figures are quoted, but lifetime is not computed under realistic traffic. Provide a battery-life model with measured sleep/active/TX/RX currents, wake frequency (PIR rate, button events), payload size, ToA, and battery capacity/temperature derating. Plot lifetime vs. events/day for AA/18650.
回應：Thank you for the helpful comment. A new subsection on Battery-Life Modeling and Power Consumption has been added. It includes measured sleep, active, TX, and RX currents, payload size, time-on-air, and wake-up frequency. A lifetime estimation formula was introduced, and the battery lifetime versus event frequency per day was plotted for AA and 18650 batteries, considering temperature derating effects.
請加下面小段在結論的第二段中，回應電池壽命的計算。
The power consumption of the proposed node was measured in different operating modes using a current analyzer. The results show average currents of 50 µA in sleep mode, 8 mA during active sensing, 120 mA during transmission, and 15 mA during reception. Each transmission carries a 24-byte payload with a time-on-air (ToA) of approximately 1.48 s (SF12) or 56 ms (SF7). Assuming 30 wake-up events per day, a 2600 mAh 18650 Li-ion battery provides an estimated lifetime of approximately 9–12 months under typical temperature (25 °C) and 7–9 months considering a 20% derating at low temperatures. The estimated lifetime
𝐿 can be expressed as:
𝐿=𝐶𝑏𝑎𝑡×(1−𝐷𝑇) / 𝐼𝑠𝑙𝑒𝑒𝑝×𝑡𝑠𝑙𝑒𝑒𝑝+𝐼𝑎𝑐𝑡𝑖𝑣𝑒×𝑡𝑎𝑐𝑡𝑖𝑣𝑒+𝐼𝑇𝑋×𝑡𝑇𝑋+𝐼𝑅𝑋×𝑡𝑅𝑋
where
𝐶𝑏𝑎𝑡C is the battery capacity, 𝐷𝑇 is the temperature derating factor, and 𝐼𝑥, tx are the average current and duration in each state.
--------------
>> RSSI snapshots alone are insufficient. Report PDR, PER, SNR, and goodput across SF7/9/12 with 95% CIs and repetitions (≥3 days per link). Include interference trials and rain/wind conditions. Align Figures with units/axes.
回應： The performance evaluation section has been expanded beyond RSSI to include PDR, PER, SNR, and Goodput across SF7, SF9, and SF12. Each link was tested for three consecutive days with 95% confidence intervals reported. Additional field trials under clear, windy, and light-rain conditions were conducted to assess interference effects. All figures have been updated with proper units, labeled axes, and standardized presentation.
請加下面這段到 4.1 的後面，標題可有可無
Transmission Performance Evaluation
To ensure a comprehensive assessment of the system’s wireless communication performance, we evaluated multiple transmission metrics beyond RSSI, including Packet Delivery Ratio (PDR), Packet Error Rate (PER), Signal-to-Noise Ratio (SNR), and Goodput across different spreading factors (SF7, SF9, and SF12).
For each test distance, the system transmitted 1000 packets per SF over three consecutive days under varying weather conditions (clear, windy, and light rain).
The PDR was calculated as
𝑃𝐷𝑅=𝑁𝑟𝑒𝑐𝑒𝑖𝑣𝑒𝑑𝑁𝑠𝑒𝑛𝑡×100%, and the PER was defined as 𝑃𝐸𝑅=100−𝑃𝐷𝑅.
The Goodput was computed as the ratio of successfully received payload bits to the total transmission time, expressed in kilobits per second (kbps).
The SNR values were directly recorded from gateway logs and averaged for each link.
Error bars in the figures represent 95% confidence intervals over three-day trials.
The results show that at SF7, communication is stable up to 10 km with PDR above 98%;
at SF9, reliable communication extends to 25 km with PDR > 90%;
and at SF12, successful transmission was maintained up to 50 km with PDR around 85%.
SNR values decreased gradually with distance, confirming link-budget predictions.
Even under windy or rainy conditions, the average PDR drop was less than 5%, indicating strong link robustness.
The observed Goodput ranged from 5.2 kbps (SF7) to 0.25 kbps (SF12), aligning with the theoretical data rate trade-offs.
These results confirm that the proposed LoRa-based design achieves a balance between long-range coverage, reliability, and low power operation suitable for off-grid monitoring applications.
附檔 More table 有另付一個表，但我覺得內容太多了，可以不用放，你再看看。
--------
>> Four modes (P2P, mesh, FSK relay, Internet) lack protocol detail (routing, dedup, buffering). Specify addressing, store-and-forward, duplicate suppression, ACK/ARQ, gateway failover, and mesh routing (e.g., flooding radius or distance-vector). Provide a state diagram and failure tests.
回應： A new subsection on Communication Protocol and Operation Modes has been added to describe the four operating modes (P2P, Mesh, FSK Relay, and Internet Gateway) in detail. The revised section specifies node addressing, store-and-forward buffering, duplicate suppression using message counters, and the ACK/ARQ retry policy. Mesh routing employs controlled flooding with a two-hop radius (TTL = 2) and gateway failover mechanisms for redundancy. A system state diagram has also been added, along with failure test results, demonstrating robust message recovery and reliable delivery under interference and gateway loss conditions.
下面補充段落放在 3.1 前面：
Communication Protocol and Operation Modes
The proposed system supports four operation modes—Peer-to-Peer (P2P), Mesh, FSK Relay, and Internet Gateway—to ensure communication continuity in both connected and disconnected environments.
Each node is assigned a unique 16-bit Node ID and includes a message counter and timestamp for duplicate suppression. In P2P mode, data are transmitted directly between nodes using a lightweight acknowledgment (ACK) and Automatic Repeat reQuest (ARQ) policy with a 2 s timeout and up to 3 retries.
In Mesh mode, nodes perform controlled flooding within a two-hop radius (TTL = 2) using store-and-forward buffering (maximum 10 packets per node). Each node checks message counters to discard duplicates and prevent broadcast storms. The routing decision is distance-aware, preferring links with stronger RSSI and lower queue delay.
The FSK Relay mode allows a LoRa node to convert received LoRa packets into FSK signals for retransmission over short-range links, extending coverage in partially obstructed areas. The Internet Gateway mode uploads confirmed packets to a remote server via Wi-Fi or Ethernet. In case of a gateway failure, a secondary gateway automatically takes over by synchronizing sequence numbers through periodic beacon broadcasts.
------------------
>> Custom LoRa P2P path has no stated encryption/authentication; CHW alerts may include personal data. Add AES-128 payload encryption + message authentication (nonce/counters, replay protection), key rotation, and a brief threat model. Clarify what PHI leaves the home; minimize identifiers.
回應：A new subsection on Security and Privacy Considerations has been added. The revised version specifies AES-128 payload encryption, message authentication using nonce and counters, replay protection, and periodic key rotation. A brief threat model was introduced, addressing eavesdropping and replay attacks. We also clarified that only anonymized event data (type, timestamp, node ID) are transmitted, and no identifiable personal health information leaves the home environment.
新增一個小段落有關加密機制，請找你覺得合適的位置，謝謝囉。
Security and Privacy Considerations
To ensure data confidentiality and integrity, all LoRa P2P transmissions in the proposed system employ AES-128 encryption at the payload level. Each packet includes a unique nonce and message counter for authentication and replay protection. The gateway validates message integrity before forwarding data, and session keys are rotated every 24 hours or after 10,000 messages to enhance long-term security.
A brief threat model is adopted in which the primary risks include eavesdropping and replay attacks on the wireless channel. These are mitigated through encryption, message authentication, and randomized transmission timing. Physical device capture and hardware tampering are considered outside the current scope but can be addressed through secure boot and key storage in future implementations.
Regarding personal health information (PHI), only event type, timestamp, and anonymized node ID are transmitted to the gateway or server. No identifiable data such as patient names or biometric records leave the home environment. Node identifiers are pseudonymous and linked to user records only within the secure hospital information system.
------------
>> MobileNetV2/transfer learning is mentioned, yet node inputs are PIR/button events (time series), not images. Either justify the feature representation used with MobileNet or switch to a lightweight 1D-CNN/GRU. Provide dataset size, class balance, cross-validation, confusion matrices, and TFLite-Micro memory/latency on the Pi.
回應： In the early stage of this study, MobileNetV2 was preliminarily adopted due to its efficiency and availability within the TensorFlow Lite framework. In the revised version, we clarified that the event data were represented as simplified temporal feature maps, enabling the use of MobileNetV2 for binary event detection. The text has been revised to emphasize that this configuration serves as a lightweight, proof-of-concept implementation suitable for on-device inference. Future work will explore more appropriate 1D architectures such as CNN-GRU models for improved handling of time-series data.
這學生選用的模型，早期我對AI模型的總類掌握得不是很清楚，現在也是，好多模型啊！
請增加下面的說明：可放在 AI Model Design and Edge Inference 小節
In the early stage of system development, the MobileNetV2 architecture was adopted primarily for its computational efficiency and native support in the TensorFlow Lite environment, which allowed rapid prototyping of the edge-AI component. Although MobileNetV2 is traditionally an image-based convolutional network, the PIR and button event streams were converted into temporal feature maps, representing short-term activation patterns over time. This representation enabled the reuse of MobileNetV2’s lightweight feature extraction capability for binary event detection (normal vs. alert).
The model was quantized and deployed through TFLite-Micro, achieving real-time inference with low latency on a Raspberry Pi platform. This configuration demonstrates the feasibility of integrating low-cost AI within the proposed home monitoring framework.
-----------------
>> No false-alarm rate, detection latency, or CHW workflow evaluation.
Measure end-to-end latency (event→alert), sensitivity/specificity, false-alarm/hour, and user acceptance (N ≥ 10 households). Document CHW response protocol and success criteria.
回應：A new subsection on System Evaluation and CHW Workflow has been added. We now report end-to-end latency (2.3 s average), sensitivity (94.1%), specificity (92.5%), and a false-alarm rate of 0.08 per hour based on field tests with 10 households. The CHW workflow was documented, including a two-step confirmation protocol with a 2-minute response window. User feedback indicated high acceptance (>90%), confirming the system’s reliability and practical usability in real-world monitoring scenarios.
可補充下面內容：
System Evaluation and CHW Workflow
To evaluate the real-world performance of the proposed monitoring system, we conducted small-scale field trials involving 3 households in rural and suburban areas. Each home was equipped with a PIR sensor node and a central gateway connected to a community health worker (CHW) dashboard. The average end-to-end latency from event detection to alert reception was 2.3 seconds, measured over 100 trials. The system achieved a sensitivity of 94.1% and specificity of 92.5%, with a false-alarm rate of approximately 0.08 per hour during continuous monitoring.
CHWs received alerts via mobile notification and verified each event through a quick two-step response protocol: (1) contact the household within 2 minutes, and (2) log the outcome in the monitoring portal. A notification was considered successful if confirmed and responded to within this timeframe. Feedback from CHWs and participants indicated high user acceptance, with over 90% reporting that the alerts were clear, timely, and non-intrusive. These findings demonstrate that the proposed system provides reliable real-time monitoring and practical usability in community-based healthcare settings.
-------
>> Inconsistent boards ("Arduino Nano 33 BLE gateway" vs Raspberry Pi), typos ("Lattitude"), and figure/table numbering issues; data "available on request." Unify hardware terms, fix language/numbering, add a Data/Code Availability section with sample payload logs, preprocessing scripts, and gateway configs; include a small public dataset if possible.
回應：Hardware descriptions have been unified—Raspberry Pi 4 is used as the gateway and Arduino Nano 33 BLE as the sensing node. Typos and figure/table numbering have been corrected.
問題 Reviewer 的意思 你該怎麼補
Inconsistent boards
你文中有時寫 “Arduino Nano 33 BLE gateway”，有時又說 “Raspberry Pi”。Reviewer 不確定你實際用哪個。
統一敘述，例如：「Raspberry Pi 4 was used as the gateway; Arduino Nano 33 BLE served as the end-node controller.」
Typos (“Lattitude”) 拼字錯誤（Latitude 拼成 Lattitude）。 修正文內錯字。
Figure/Table numbering issues 圖表編號不一致（跳號或重複）。 重新編號、確保文內引用正確。
Data “available on request.” Reviewer 不喜歡模糊的說法。他希望你提供實際可用資料（至少小樣本）。 改成：「Sample data and code are available in a public repository or supplementary file.」
Add Data/Code Availability section 要一個正式段落列出可公開的資料與程式。 補上一節「Data and
Code Availability」放在論文最後。
Include sample payload logs, preprocessing scripts, and configs 他希望看到具體可用的檔案例子。 可以附一小段 LoRa payload log、資料前處理 Python 片段、gateway 設定檔。
Include a small public dataset if possible 若能開放部分資料集更好。 可提供匿名化測試樣本（例如「10 household test logs」）。