A portable blood pressure health monitoring gateway system and method based on a large language model

Abstract

The application discloses a portable blood pressure health monitoring gateway system and method based on a large language model. The system comprises: a sensing layer for synchronous signal acquisition; a gateway layer adopting a double-core architecture, responsible for receiving and forwarding sensing layer data and local interaction, solving the interference problem caused by single-chip concurrent communication; a cloud platform layer deploying a blood pressure algorithm module and a large language model interaction module, used for receiving data, estimating blood pressure and returning results, supporting user voice query of historical data, and realizing voice-based natural health Q&A and personalized suggestions by constructing a structured prompt word containing historical blood pressure characteristics. The application has the advantages of low power consumption of wearable devices, stable communication, high intelligence, and no need to rely on mobile phone terminals, and is especially suitable for middle-aged and elderly family health monitoring scenes.

Description

Technical Field

[0001] This invention belongs to the fields of intelligent health management and Internet of Things technology, specifically relating to a portable gateway device and its cloud service system that can collect physiological signals, perform intelligent voice interaction and blood pressure health management, and coordinate edge gateway and cloud platform. Background Technology

[0002] Existing cuffless blood pressure monitoring systems typically employ a "data acquisition terminal-gateway-server" architecture. However, current technologies generally suffer from the following problems: gateway devices have limited functionality, serving only as a data transmission bridge and lacking local processing and user interaction capabilities; for middle-aged and elderly individuals, operating a mobile app to obtain their health information involves complex interfaces, cumbersome Bluetooth pairing steps, and a high cognitive barrier related to physiological values; cloud servers serve only as storage, lacking the ability to receive voice queries, provide large language model-assisted interpretation, and push personalized health advice to users.

[0003] Given that most existing monitoring systems are passive, providing only cold, impersonal data after user measurements, and lacking targeted, personalized interpretation and proactive care, there is an urgent need for a portable blood pressure health monitoring solution that can operate independently without a mobile phone, use large language models for interpretation, continuously track data, provide more natural user interaction, and efficiently collaborate with edge gateways and cloud platforms. Summary of the Invention

[0004] The technical problem this invention aims to solve is to address the shortcomings of existing cuffless blood pressure monitoring technologies, such as the lack of intelligent voice interaction, large-scale model data-assisted interpretation, and a collaborative architecture between edge gateways and cloud platforms. This invention proposes a portable blood pressure health monitoring gateway system and method based on a large language model. This system can detect physiological parameters through wearable devices, achieve data transmission and intelligent interaction through a dual-core gateway, and complete blood pressure calculation and large-scale model voice question-and-answer through a cloud platform, forming a closed loop from signal acquisition to result feedback and interpretation.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A portable blood pressure health monitoring gateway system and method based on a large language model, comprising a sensing layer, a gateway layer and a cloud platform layer;

[0007] (a) Sensing layer: includes a wearable device that integrates a PPG sensor, ECG electrode and Bluetooth communication module, used to synchronously collect the user's PPG signal and ECG signal and send them to the gateway layer via Bluetooth Low Energy;

[0008] (ii) Gateway Layer: This layer includes a main control chip and a communication chip, as well as input / output peripherals related to human-computer interaction. The communication chip communicates and exchanges data with the wearable device; the main control chip communicates with the communication chip and integrates a WiFi communication module to connect to the cloud platform layer, as well as a driver module to drive the input / output peripherals. The gateway layer acts as a local hub, responsible for receiving data from the sensor layer, uploading it to the cloud platform, displaying blood pressure results, and facilitating voice interaction.

[0009] (III) Cloud Platform Layer: This layer includes an MQTT broker, a blood pressure algorithm module, and a time-series database. It receives data uploaded from the gateway layer via the MQTT protocol, performs blood pressure calculations, and returns the calculation results to the gateway layer for display or broadcast. The cloud platform layer also integrates a module for interacting with a large language model, which responds to user voice queries and generates personalized health recommendations based on historical blood pressure data.

[0010] Preferably, the gateway layer adopts a dual-chip solution, with the main control chip being ESP32-S3 and the communication chip being nRF52840, to achieve functional separation of Bluetooth communication and network communication.

[0011] As a further preferred embodiment, the blood pressure algorithm module is configured to: extract time-stamp-aligned ECG and PPG signal segments from a time-series database, detect the R-wave peak point of the ECG signal and the feature point of the PPG signal, calculate the time difference between the two as the PTT value, and substitute the PTT value into the blood pressure calculation model to obtain systolic and diastolic blood pressure.

[0012] Compared with existing technologies, the beneficial effects of the portable blood pressure health monitoring gateway system and method based on a large language model provided by this invention are as follows:

[0013] To address the issues of existing cuffless blood pressure monitoring systems relying on mobile apps and being cumbersome to operate, this invention integrates interactive modules such as microphones and speakers into the gateway device, enabling the system to operate independently. Users no longer need to go through tedious Bluetooth pairing or operate complex mobile phone interfaces, making it particularly suitable for middle-aged and elderly people unfamiliar with smartphone operation, thus improving the user experience.

[0014] Edge gateways and cloud platforms can collaborate efficiently. Specifically, the sensing layer is only responsible for signal acquisition and data transmission, without undertaking physiological parameter calculation or inference tasks, resulting in extremely low power consumption and allowing wearable devices to be made smaller and lighter. The gateway layer is responsible for local communication, human-computer interaction, and real-time uploading, with fast response speed. The cloud platform is responsible for all heavy computing (including PTT extraction, model inference, and large model dialogue). Each of the three terminals performs its own function, avoiding the problems of insufficient computing power and battery life bottlenecks of a single device.

[0015] The dual-core edge gateway design makes communication more stable during data forwarding, ensuring real-time communication and data integrity. The communication chip is dedicated to the BLE protocol stack, while the main control chip is dedicated to handling WiFi networking, cloud platform interaction, and user interface. This eliminates data packet loss and delay caused by multi-protocol concurrency and time-division multiplexing antennas in single-chip solutions, ensuring real-time signal data transmission and command transmission.

[0016] The cloud platform integrates a time-series database and a large-scale model interface, enabling full lifecycle management of data. The time-series database retains complete blood pressure waveforms and historical measurement records, providing a data foundation for subsequent data analysis and personalized model training, as well as rich context for large-scale models in the cloud. Through voice interaction, users can query historical data and obtain intelligent interpretations. The system has evolved from a traditional passive detection system into a proactive health management assistant, improving data readability.

[0017] The system architecture is flexible and highly scalable. Blood pressure algorithms and large models can be iterated and upgraded in the cloud without requiring users to replace or upgrade edge gateway hardware, thus reducing long-term maintenance costs. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall system architecture according to an embodiment of the present invention;

[0019] Figure 2 This is a schematic diagram of the gateway layer hardware structure according to an embodiment of the present invention;

[0020] Figure 3 The overall system workflow diagram provided for embodiments of the present invention. Detailed Implementation

[0021] See Figure 1 , Figure 1 This is a schematic diagram of the overall system architecture according to an embodiment of the present invention. The system in this embodiment includes a sensing layer 100, a gateway layer 200, and a cloud platform layer 300. The sensing layer 100 is used to collect PPG and ECG signals and transmit them via Bluetooth; the gateway layer 200 acts as a local hub, responsible for receiving signals, uploading to the cloud, displaying results, and providing voice interaction; the cloud platform layer 300 is responsible for data storage, blood pressure calculation, and large-scale intelligent question answering.

[0022] In this embodiment, the sensing layer employs a wristband-style wearable device, integrating a PPG sensor, an ECG analog front-end, and a Bluetooth SoC. The device is worn on the user's wrist, with the ECG electrodes positioned at the bottom to contact the wrist skin and at the front for contact with the fingers of the other hand, forming a single lead. After the user initiates a measurement command through the gateway layer, the sensing layer simultaneously acquires 30 seconds of ECG and PPG signals at a sampling rate of 250Hz, and packages and sends them to the gateway layer via the BLE protocol. Specific signal processing by the sensing layer, such as filtering and noise reduction, uses conventional methods and will not be elaborated here.

[0023] like Figure 2 The gateway layer hardware structure diagram shows a dual-chip solution: the main control chip is ESP32-S3 (running FreeRTOS), and the communication chip is nRF52840. The nRF52840 only runs the BLE protocol stack, configured as a GATT Client, establishing a connection with the wearable device in the sensing layer (acting as a GATT Server) and subscribing to notifications to receive real-time signal data. The ESP32-S3 runs a WiFi network protocol stack, an MQTT client, voice wake-up and interaction model, and is connected to a 1.3-inch OLED screen, three physical buttons, a microphone and speaker, LED indicators, a Flash cache module, and a power supply and battery module. Specifically: the LED indicators show system, network, and power status; the Flash cache module temporarily stores physiological signal data when the network is disconnected and retransmits it when the network connection is restored; the power supply and battery module is responsible for voltage conversion, voltage regulation, and battery charging management; the buttons respond to user measurement commands and voice start / stop commands; and the screen displays blood pressure and historical trends, as well as dialogue content between the user and the large model.

[0024] To more clearly illustrate the technical solution of this embodiment, the workflow of this system is described below:

[0025] After the gateway layer powers on, it automatically scans for specific wearable devices in the broadcast and initiates a connection. Once the connection is established, the gateway layer writes a capture / stop command via BLE. After the wearable device responds to the capture command, it begins capturing signals for a preset time and uploads data packets via Notification, packaging them into sets of 10 sampling points. Upon receiving this, the nRF52840 immediately notifies the ESP32-S3 to retrieve the data via UART.

[0026] The ESP32-S3 temporarily stores the received signal data in a local circular buffer. After acquisition is complete, it assembles the complete signal segment along with the user ID and timestamp into a JSON format: {

[0027] “user_id”: “U123”,

[0028] "timestamp": "2026-06-08T10:30:00Z",

[0029] “ecg”: [...],

[0030] “ppg”: [...]

[0031] }

[0032] Then, it publishes the data to the cloud platform topic / measure / raw via the MQTT protocol, with QoS=1, ensuring at least one delivery. If WiFi is temporarily unavailable, the ESP32-S3 will store the data in local Flash memory and retransmit it after the network is restored.

[0033] The gateway layer runs a voice wake-up model in real time. When a user wakes up the model using a specific prompt and says a voice query such as "How is my blood pressure now?", the microphone captures the audio, and the ESP32-S3 streams it to the large language model interaction module on the cloud platform via WebSocket. After the cloud platform returns the TTS audio, the ESP32-S3 outputs it to the speaker for playback via I2S. Simultaneously, the ESP32-S3 displays the text suggestions returned by the large model on the screen in a scrolling manner.

[0034] The cloud platform deploys EMQX as an MQTT broker, listening on the default port 1883. After the data published by the gateway layer arrives, the broker triggers the rules engine: writing the raw JSON data to the time-series database InfluxDB, storing it in partitions by user ID, and preserving the original waveform for subsequent algorithm recalculation or model retraining.

[0035] The blood pressure calculation module is implemented in Python. It subscribes to the / measure / raw topic, and when new data arrives, the module performs the following steps:

[0036] (1) Read the latest ECG and PPG records of the current user from the time series database, or process the data that was just written directly;

[0037] (2) Bandpass filtering of the ECG signal with a frequency band of 0.5Hz to 40Hz was performed, and the peak point of the R wave was detected using the Pan-Tompkins algorithm;

[0038] (3) The PPG signal is low-pass filtered with a cutoff frequency of 10Hz, and the maximum rising edge or peak point is located using first-order differential positioning as the PPG feature point.

[0039] (4) Calculate the time difference between the peak point of the R wave and the corresponding PPG characteristic point in each cardiac cycle to obtain multiple PTT values, and take the median as the final PTT;

[0040] (5) Input the PTT (Personal Tolerance Time) along with the user's personal information (including age, gender, height, weight, and heart rate, where height and weight are converted to BMI) into the pre-trained blood pressure model. This embodiment uses a lightweight XGBoost regression model with the following structure: input feature dimension 5 (PTT, age, gender encoding, BMI, heart rate), output systolic / diastolic blood pressure. The model is trained using publicly available data from the MIMIC-Ⅲ database. The mean error on the independent test set meets the AAMI requirement;

[0041] (6) After the calculation is completed, the results (systolic blood pressure, diastolic blood pressure, heart rate, PTT value, measurement time) are encapsulated in JSON format and published to the topic / measure / result / {user_id} via MQTT. The gateway layer subscribes to the topic and displays or broadcasts the results via voice.

[0042] The cloud platform also deploys an API for an intelligent interaction service. The processing flow of this service is as follows:

[0043] (1) Receive the real-time audio stream uploaded by the gateway layer and call the third-party ASR interface to convert it into a text query;

[0044] (2) Based on the user ID, retrieve the blood pressure measurement records for the past 7 days from the time-series database. Each record includes systolic blood pressure, diastolic blood pressure, and measurement time.

[0045] (3) Construct a prompt: "You are a health assistant. The following is the blood pressure data of user {user_id} for the past 7 days: {data list}. User asks: {user query}. Please provide a brief and friendly suggestion based on the blood pressure guidelines. Note that the suggestion is for reference only. If you feel unwell, please seek medical attention."

[0046] (4) Call the general large model API and set temperature=0.7, max_tokens=200;

[0047] (5) Convert the text returned by the model into audio through the TTS service and return it to the gateway layer for playback.

[0048] Preferably, the specific method for constructing the prompt words described above is as follows:

[0049] The system first performs text-based preprocessing on the retrieved blood pressure data from the past 7 days. For example, if it detects that a user's recent average systolic blood pressure is higher than 140 mmHg and shows an upward trend, it generates a prompt fragment: "Systolic blood pressure remains high, above 140 mmHg, and fluctuates significantly." This prompt fragment is then combined with preset role settings and the user's query content to form the final prompt:

[0050] "You are a professional cardiovascular health assistant. Please respond to user {user_id} based on the following user data: {User Data}. Data Summary: Systolic blood pressure has been consistently high recently and is showing an upward trend. User Question: {User Query}. Requirements: Use a friendly tone, provide brief advice, and emphasize that if you feel unwell, please seek medical attention."

[0051] By constructing structured prompts, large language models can be guided to focus on data features, thereby generating more accurate and personalized suggestions and avoiding perfunctory or unexpected responses.

[0052] Combining the above modules, the complete workflow of this system includes two parts: the cuffless blood pressure measurement process and the intelligent health interaction process.

[0053] I. Cuffless blood pressure measurement procedure:

[0054] Step S1: The user presses the gateway layer button, and the ESP32-S3 notifies the nRF52840 to send a measurement command to the wearable device via UART; the wearable device simultaneously collects ECG and PPG signals for a preset duration and uploads them to the gateway layer via BLE;

[0055] Step S2: The nRF52840 forwards the data to the ESP32-S3. The ESP32-S3 packages the signal data into JSON format and uploads it to the cloud platform layer via WiFi using the MQTT protocol. If WiFi is unavailable, the signal data is temporarily stored in the Flash cache module and retransmitted after the network is restored.

[0056] Step S3: The cloud platform MQTT broker receives data and stores the raw data in the time series database; the blood pressure calculation module directly reads the current data in memory or extracts the data written to the time series database to calculate PTT;

[0057] Step S4: Substitute the PTT into the preset XGBoost blood pressure calculation model to estimate the user's systolic and diastolic blood pressure;

[0058] Step S5: The cloud platform layer returns the estimated blood pressure result to the gateway layer via MQTT. The ESP32-S3 displays the blood pressure value on the screen in real time and can optionally announce it via speaker.

[0059] II. Intelligent Health Interaction Process:

[0060] Step S6: The user directly sends a voice wake-up word to the gateway layer or presses the query trigger button to ask a health-related question (such as "How is my blood pressure now?"). The gateway layer uploads the real-time collected audio stream to the cloud platform layer.

[0061] Step S7: The intelligent interaction service of the cloud platform layer calls the ASR interface to convert speech into text, and retrieves the most recent blood pressure historical data from the time series database based on the user ID as the context of the big language model; then the cloud platform layer constructs prompt words containing the user's blood pressure historical data, calls the big language model API, and generates personalized health advice text for the user.

[0062] Step S8: The cloud platform layer converts the generated health advice text into an audio stream via TTS, returns it to the gateway layer, and then plays it out through the audio playback module of the gateway layer, while scrolling the advice text on the screen.

[0063] Furthermore, those skilled in the art will readily recognize that the wearable device can also be in the form of a clip-on or ring; the main control chip in the dual-core gateway layer solution can also be replaced with other chips integrating WiFi, and the communication chip can also be replaced with other chips integrating BLE. The blood pressure model can also be implemented using linear regression or 1D-CNN, both of which fall within the scope of protection of this invention.

Claims

1. A portable blood pressure health monitoring gateway system based on a large language model, characterized in that, include: The sensing layer includes a wearable device that integrates a PPG sensor, an ECG electrode, and a Bluetooth communication module for synchronously acquiring PPG signals and single-lead ECG signals and transmitting them to the gateway layer via Bluetooth Low Energy. The gateway layer includes a main control chip, a communication chip, a Flash cache module, a power supply and battery module, and input / output peripherals. The input / output peripherals include an audio acquisition module, an audio playback module, a visual interface module, a physical interaction module, and a status indicator module. The communication chip is used to communicate and exchange data with the wearable device. The main control chip communicates and interacts with the communication chip, and integrates a WiFi communication module to connect to the cloud platform layer, drive the input / output peripherals, and manage the Flash cache module. The cloud platform layer includes an MQTT broker, a blood pressure algorithm module, a large language model interaction module, and a time-series database. The time-series database stores raw waveform data with user IDs and timestamps, as well as blood pressure calculation results. The large language model interaction module is configured to: respond to user voice queries, retrieve the user's historical blood pressure data from the time-series database, construct prompt words containing historical data, and call the large model to generate personalized health suggestions.

2. The system according to claim 1, characterized in that: The gateway layer supports Bluetooth Low Energy (BLE) and Wi-Fi communication protocols. As a BLE central device (GATT Client), the gateway layer communicates with the sensing layer and connects to the existing Wi-Fi network through client mode (STA) to achieve network connectivity and communicate with the cloud platform layer.

3. The system according to claim 1, characterized in that: The wearable device is configured to trigger and acquire synchronous ECG and PPG waveforms within a preset duration in response to instructions transmitted by the gateway layer via BLE, for calculating pulse wave conduction time (PTT).

4. The system according to claim 1, characterized in that: The audio acquisition module includes a microphone, the audio playback module includes a power amplifier and a speaker, the visual interface module includes a screen, the physical interaction module includes buttons, the status indicator module includes LEDs, and the power supply and battery module includes a voltage conversion chip, a charging chip, and a battery.

5. The system according to claim 1, characterized in that: The gateway layer adopts a dual-chip solution. The main control chip is ESP32-S3, which runs an MQTT client and a voice wake-up model. It is connected to a visual interface module, a physical interaction module, an audio acquisition module, an audio playback module, a status indicator module, a Flash cache module, and a power supply and battery module. It is responsible for WiFi network communication, interaction with the cloud-based large model, audio input and output, and local display of blood pressure values. The communication chip is nRF52840, which only runs the BLE protocol stack and is responsible for Bluetooth connection and data exchange with the wearable device.

6. The system according to claim 1, characterized in that: The wearable device is in the form of a wristband, a clip, or a ring.

7. The system according to claim 1, characterized in that: The blood pressure algorithm module is configured as follows: Timestamp-aligned ECG and PPG signal segments are extracted from the time-series database. The R-wave peak point of the ECG signal and the feature point of the PPG signal are detected. The time difference between the R-wave peak point and the PPG feature point is calculated as the PTT value. The PTT value, along with the user's physiological characteristic data, is substituted into a preset blood pressure calculation model to calculate the systolic and diastolic blood pressure. The physiological characteristic data includes age, gender, height, weight, and heart rate.

8. The system according to claim 1, characterized in that: The large language model interaction module integrated in the cloud platform layer performs the following steps: (1) Receive user voice queries forwarded by the gateway layer and convert them into text using automatic speech recognition (ASR); (2) Retrieve the user's most recent historical blood pressure data from the time-series database based on the user ID, wherein the record includes at least systolic blood pressure, diastolic blood pressure, and measurement timestamp; (3) Extract features from the historical blood pressure data to generate feature text data, which includes blood pressure fluctuation trends and outlier markers; (4) Construct prompt words from user voice query content, feature text data, and role setting instructions; (5) Input the constructed prompt words into the large language model interface and receive the returned text suggestions.

9. A method for blood pressure health monitoring based on the system according to any one of claims 1 to 8, characterized in that, Includes the following steps: Step S1: The wearable device receives and responds to the measurement command triggered by the user through the button of the gateway layer via BLE, or synchronously collects the user's PPG signal and single-lead ECG signal according to a preset period, and sends them to the gateway layer via BLE; Step S2: In the gateway layer, the communication chip forwards the received data to the main control chip, and the main control chip uploads it to the cloud platform layer via WiFi using the MQTT protocol; Step S3: After receiving the data, the cloud platform layer stores the data in the time series database, and then extracts the ECG and PPG signals from the time series database to calculate the PTT; Step S4: Substitute the PTT value into the blood pressure calculation model to estimate the user's systolic and diastolic blood pressure; Step S5: Return the estimated blood pressure value to the gateway layer and display it on the screen of the gateway layer, and / or broadcast it via a speaker; Step S6: The gateway layer receives the voice commands issued by the user and uploads the collected audio stream to the cloud platform layer; Step S7: The cloud platform layer converts speech into text, retrieves historical blood pressure data from the time-series database based on the user ID, then constructs prompt words that include the historical blood pressure data, and calls a large language model to generate personalized health advice text; Step S8: The cloud platform layer converts the health advice text output by the large language model into a speech stream and sends it back to the gateway layer. The gateway layer plays the speech through the speaker and displays it on the screen.

Images