Satellite communication system

Abstract

The present invention relates to a satellite communication method, and the method may comprise the steps of: acquiring at least one of vehicle information and driver information on the basis of at least one of collected voice data, sensor data, and user input data; on the basis of the acquired information, classifying a driver situation into at least one of a plurality of context types; generating a command message on the basis of the context type and the acquired information; performing data lightening such that the size of the command message is reduced; and transmitting the lightened command message to a satellite.

Description

satellite communication system

[0001] The present invention relates to a satellite communication system. More specifically, it relates to a satellite communication system and method that provides data transmission and intelligent emergency response services optimized for vehicle and driver conditions by linking an on-device AI of a user terminal with a context-aware AI assistant of a server in a non-terrestrial network (NTN) environment.

[0002] Due to recent advancements in telecommunications technology, Non-Terrestrial Network (NTN) technology is gaining attention for providing seamless connectivity in areas not covered by Terrestrial Networks (TN), such as maritime and aviation regions. In particular, communication services utilizing Low Earth Orbit (LEO) or Geostationary Orbit (GEO) satellites are establishing themselves as essential means of communication for emergency rescue requests or disaster situations.

[0003] However, these satellite communication technologies have the following limitations.

[0004] First, there is the issue of accessibility caused by the high costs and bandwidth constraints of satellite communication. Generally, satellite-based voice calls incur significantly higher charges compared to terrestrial networks, and data transmission speeds are slow with limited bandwidth. Consequently, existing systems rely on methods that simply transmit text messages or send low-quality voice data as is. This approach not only increases communication costs by failing to efficiently reduce data packet sizes but also poses a risk of missing the "golden time" during emergencies due to transmission failures or delays caused by network congestion.

[0005] Second, there is the problem of inaccurate situational awareness resulting from simple data transmission. Conventional emergency rescue systems are limited to transmitting only fragmentary information, such as text input by the user or GPS coordinates. However, at actual accident scenes, it is common for drivers to panic and find it difficult to input accurate information, or to lose consciousness and be unable to operate the system. Relying solely on sensor values ​​or text makes it difficult to comprehensively assess the driver's psychological state, the extent of specific injuries, and the noise level at the scene, thereby limiting the rescue center's ability to perform an appropriate initial response tailored to the situation.

[0006] Third, there is a lack of limited two-way communication and user experience (UX). Existing satellite terminals are primarily focused on one-way rescue requests, making it difficult to engage in two-way interaction, such as receiving feedback after a rescue request or specific instructions (guides) before the rescue team arrives. Furthermore, there are issues where communication with local rescue teams is impossible due to language barriers in the event of an accident while traveling abroad, or where access to daily convenience services—such as restaurant reservations or general inquiries—is restricted due to the lack of voice call support or the burden of charges.

[0007] Therefore, there is a growing technical demand for intelligent satellite communication systems that can improve communication efficiency by drastically reducing data packets even in bandwidth-limited satellite communication environments, while utilizing AI technology to accurately recognize the driver's situation and provide active two-way guidance. In particular, there is an increasing need for an integrated solution that can reduce communication costs through text-to-text (STT / TTS) and compression technologies, and flexibly respond to both emergency and general situations through a server-based AI assistant.

[0008] The present disclosure aims to solve the aforementioned problems and other problems, and to provide a satellite communication system and a method of operation thereof for optimizing data by utilizing on-device AI in situations where communication resources are limited, such as in a non-terrestrial network (NTN) environment, and for providing intelligent emergency response services by linking with an AI assistant on a server.

[0009] Another objective of the present disclosure is to overcome the high costs and bandwidth constraints of satellite communication by converting voice data into text within the terminal and selectively extracting, compressing, and transmitting only the essential data according to the situation type, thereby minimizing data packets and maximizing transmission efficiency.

[0010] Another objective of the present disclosure is to provide a satellite communication method that overcomes the limitations of conventional methods that merely transmit text or sensor values, performs accurate context awareness by comprehensively analyzing the driver's voice, psychological state, and location information through on-device AI, and can precisely determine the level of emergency based on this.

[0011] Another objective of this disclosure is to provide seamless two-way communication and an enhanced user experience (UX) by utilizing server-based conversational AI and large language models (LLM) even in text-based satellite communication environments to provide users with situation-specific customized response guides or to perform voice calls with emergency centers on their behalf.

[0012] A satellite communication method according to the present disclosure may include: a step of acquiring at least one of vehicle information and driver information based on at least one of collected voice data, sensor data and user input data; a step of classifying a driver situation into at least one of a plurality of context types based on the acquired information; a step of generating a command message based on the context type and the acquired information; a step of performing data lightweighting so as to reduce the size of the command message; and a step of transmitting the lightweighted command message to a satellite.

[0013] According to an embodiment, the method may further include the step of extracting necessary data associated with the situation type; and the step of adding the necessary data to the command message.

[0014] According to an embodiment, the method further includes a step of determining whether there is an emergency situation or an urgency level based on the acquired information; and the situation type may be classified based on whether there is an emergency situation or an urgency level.

[0015] According to an embodiment, the situation type may include at least one of an emergency call, an emergency action guide, a phone reservation, and an information search.

[0016] According to an embodiment, the step of determining the emergency level may include: a step of calculating a driver psychological level by analyzing the acoustic features of the voice data; and a step of determining the emergency level by reflecting the driver psychological level.

[0017] According to an embodiment, the step of determining the emergency severity level may include: a step of obtaining location information from the sensor data and calculating a risk weight corresponding to the location information; and a step of determining the emergency severity level by reflecting the risk weight.

[0018] According to an embodiment, the method may further include the step of converting the voice data into text data using a speech-to-text (STT) model; and the step of adding the text data to the command message.

[0019] According to an embodiment, the method may further include a step of determining whether the text data and the sensor data are different from each other; and a step of performing content correction on the text data by applying a priority to the sensor data if the text data and the sensor data are different from each other.

[0020] According to an embodiment, the method may further include the step of receiving a response message for the transmitted command message from a server via the satellite; and the step of performing a response action associated with the received response message.

[0021] According to an embodiment, when the situation type is an emergency call or emergency action guide, the required data may include at least one of whether a collision has occurred, vehicle identification information, location information, vehicle speed, and occupant information, and when the situation type is a phone reservation or information search, it may include at least one of location information, user information, vehicle information, and service request information.

[0022] According to an embodiment, the step of performing data lightweighting includes the step of performing summarization or compression of the command message, and the summarization level or compression rate of the command message may be applied differently based on at least one of the situation type and satellite communication link quality.

[0023] According to an embodiment, if the situation type is an emergency call or emergency action guide, or if the satellite communication link quality is below a threshold standard, the command message can be compressed at a first compression rate, and if the situation type is a phone reservation or information search, and the satellite communication link quality is above a threshold standard, the command message can be compressed at a second compression rate lower than the first compression rate.

[0024] According to an embodiment, a server that receives the command message may include the steps of: decompressing the received command message; analyzing the command message to identify a situation type included in the command message; and summarizing or compressing a response message containing the result of performing a situation-specific operation according to the identified situation type and transmitting it to a user terminal via the satellite.

[0025] According to an embodiment, if the identified situation type is an emergency call, the method may further include the steps of: analyzing the command message to obtain rescue request information; activating a server-based interactive AI to connect the emergency response center with the emergency call; performing a voice-based call conversation with the emergency response center based on the rescue request information; and generating the response message to include feedback information including dispatch information of emergency personnel or the result of action.

[0026] According to an embodiment, if the identified situation type is an emergency response guide, the method may further include the steps of: configuring an LLM request requesting situation analysis and transmitting it to the LLM; receiving an LLM response containing situation analysis results from the LLM in response to the transmission; and analyzing and reconstructing the LLM response.

[0027] According to an embodiment, if the identified situation type is a phone reservation, the method may further include the steps of: analyzing the command message to obtain reservation information; activating a server-based conversational AI to connect a call with the reservation target included in the reservation information; performing a voice-based conversation with the reservation target based on the reservation information; and generating the response message to include feedback information including whether the reservation was successful or details of the reservation.

[0028] According to an embodiment, if the identified situation type is information search, the method may further include: a step of obtaining search information by analyzing the command message; a step of generating a search query based on the search information and performing an information search in conjunction with an external search engine; a step of performing a text summary by extracting key information corresponding to the user's intent from the result data of the information search; and a step of generating the response message to include the summarized text data.

[0029] According to an embodiment, the step of configuring an LLM request requesting a situation analysis may include: selecting a situational prompt model based on at least one of a situational type, required data, and text data included in the command message; creating a prompt set or a situational query using the selected prompt model; and configuring an LLM request including the prompt set or the situational query.

[0030] According to an embodiment, the method further comprises the steps of: obtaining a plurality of step-by-step guide texts based on the reconstructed LLM response; and generating a response message including the plurality of step-by-step guide texts, wherein the response message may further include a TTS command that instructs an on-device artificial intelligence model to provide voice guidance for the plurality of step-by-step guide texts.

[0031] According to an embodiment, the method may further include the step of re-determining whether the driver situation is an emergency call situation based on the reconstructed LLM response, and if the re-determination result determines that it is an emergency call situation, the step of activating a server-based conversational AI to connect the emergency call with an emergency response center; the step of performing a voice-based conversation with the emergency response center based on the reconstructed LLM response; and the step of generating the response message to include feedback information including dispatch information of emergency personnel or the result of action.

[0032] According to an embodiment, in the step of performing the response operation, if the response message includes a TTS command, the method may include the step of sequentially performing the operation of converting a first guide text among a plurality of step-by-step guide texts included in the response message into speech and outputting it, detecting a user's response input to the output speech, and, when the response input is detected, converting a second guide text into speech and outputting it.

[0033] According to an embodiment, the step of performing the response operation may include: a step of obtaining at least one of location information and phone number information to identify a country code; a step of determining a target language based on the identified country code; and a step of automatically translating the received response message into the determined target language and outputting it.

[0034] A vehicle performing satellite communication according to the present disclosure includes: a communication module configured to transmit and receive data to and from a server by linking with a non-terrestrial network (NTN) via a satellite; and a processor operably connected to the communication module, wherein the processor acquires at least one of vehicle information and driver information based on at least one of collected voice data, sensor data and user input data, classifies a driver situation into at least one of a plurality of context types based on the acquired information, generates a command message based on the context type and the acquired information, performs data lightweighting so as to reduce the size of the command message, and transmits the lightweighted command message to the satellite.

[0035] In a computer-readable storage medium according to the present disclosure, the computer-readable storage medium stores at least one computer program including instructions that, when executed by at least one processor, cause the at least one processor to perform operations, wherein the operations include acquiring at least one of vehicle information and driver information based on at least one of collected voice data, sensor data and user input data, classifying a driver situation into at least one of a plurality of context types based on the acquired information, generating a command message based on the context type and the acquired information, performing data lightweighting so as to reduce the size of the command message, and transmitting the lightweighted command message to a satellite.

[0036] According to the present disclosure, voice data is converted into text (STT) using on-device AI within a terminal in a non-terrestrial network (NTN) environment, and only essential data is selectively extracted, compressed, and transmitted according to the type of situation, thereby effectively solving the problems of high cost and bandwidth constraints of conventional satellite communication. Accordingly, communication charges can be reduced by minimizing the size of data packets, and the success rate and speed of data transmission can be dramatically improved even in network congestion situations.

[0037] According to the present disclosure, unlike conventional methods that simply transmit text or sensor values, the occurrence of an accident and the level of urgency can be precisely determined by comprehensively analyzing the driver's voice characteristics (tone, intensity, etc.), psychological state, and location information through an on-device AI model. This reduces false alarms and accurately recognizes actual emergency situations, thereby establishing a foundation for rescue centers to perform optimal initial response within the golden hour.

[0038] According to the present disclosure, even in a text-based satellite communication environment, server-based conversational AI and a Large Language Model (LLM) can be integrated to provide users with situation-specific customized first aid guides or to perform voice calls with emergency centers on their behalf. Accordingly, seamless two-way communication is possible even in situations where it is difficult for the driver to make a call directly, and through automatic translation and information search functions, the user experience (UX) can be significantly expanded to areas such as overseas travel and general convenience services.

[0039] The accompanying drawings, which are included as part of the detailed description to aid in understanding the present invention, provide embodiments of the present invention and explain the technical concept of the present invention together with the detailed description.

[0040] FIG. 1 is a conceptual diagram illustrating an embodiment of a non-terrestrial network system according to the present disclosure.

[0041] FIG. 2 illustrates a block diagram of a user terminal according to the present disclosure.

[0042] FIG. 3 illustrates a block diagram of an AI assistant server according to the present disclosure.

[0043] FIG. 4 illustrates an operation flowchart of a situation recognition and corresponding situation-specific service provision method performed in a satellite communication system according to the present disclosure.

[0044] FIG. 5 is a conceptual diagram illustrating the data flow and interoperability between components of different situation types performed in a satellite communication system according to the present disclosure.

[0045] FIG. 6 is an operation flowchart for a method of generating and transmitting a command message performed at a user terminal according to the present disclosure.

[0046] FIG. 7 illustrates the data flow for the command message generation, summarization, and compression processes performed at a user terminal according to the present disclosure.

[0047] FIG. 8 is an operation flowchart of the situation analysis and response mode determination process based on LLM integration performed in an AI assistant server according to the present disclosure.

[0048] FIG. 9 illustrates an example of data processing and mode switching in which an AI assistant server according to the present disclosure reconstructs LLM analysis results and provides them to a user.

[0049] FIG. 10 illustrates an embodiment of data reconstruction, summarization, and compression performed in a user terminal and an AI assistant server according to the present disclosure.

[0050] FIG. 11 is a flowchart of the operation in an emergency call mode performed by an AI assistant server according to the present disclosure.

[0051] FIG. 12 illustrates an example of a server-based interactive TTS conversation performed in an emergency call mode according to the present disclosure.

[0052] FIG. 13 is a flowchart of the operation in the emergency action guide mode performed by the AI ​​assistant server according to the present disclosure.

[0053] FIG. 14 illustrates an example of an on-device-based interactive TTS conversation performed in an emergency response guide mode according to the present disclosure.

[0054] FIG. 15 is a flowchart of the operation in a phone reservation mode performed by an AI assistant server according to the present disclosure.

[0055] FIG. 16 illustrates an example of a server-based interactive TTS conversation performed in a phone reservation mode according to the present disclosure.

[0056] FIG. 17 is a flowchart of the operation in an information search mode performed by an AI assistant server according to the present disclosure.

[0057] FIG. 18 illustrates an embodiment of a server-based search and search result display method performed in an information search mode according to the present disclosure.

[0058] FIG. 19 illustrates an example of a Minimum Set of Data (MSD) and a data set that are selectively configured according to a situation type according to the present disclosure.

[0059] It should be noted that technical terms used in this specification are used merely to describe specific embodiments and are not intended to limit the invention. Additionally, singular expressions used in this specification include plural expressions unless the context clearly indicates otherwise. The suffixes "module" and "part" for components used in the following description are assigned or used interchangeably solely for ease of drafting the specification and do not inherently possess distinct meanings or roles.

[0060] In this specification, terms such as "composed of" or "comprising" should not be interpreted as necessarily including all of the various components or steps described in the specification, and should be interpreted as potentially excluding some of the components or steps, or including additional components or steps.

[0061] In addition, when describing the technology disclosed in this specification, if it is determined that a detailed description of related prior art could obscure the essence of the technology disclosed in this specification, such detailed description is omitted.

[0062] In addition, the attached drawings are intended only to facilitate understanding of the embodiments disclosed in this specification, and the technical concept disclosed in this specification is not limited by the attached drawings; it should be understood that they include all modifications, equivalents, and substitutions that fall within the spirit and technical scope of the present invention. Furthermore, not only each of the embodiments described below, but also combinations of the embodiments may fall within the spirit and technical scope of the present invention as modifications, equivalents, and substitutions that fall within the spirit and technical scope of the present invention.

[0063] Hereinafter, embodiments disclosed in this specification will be described in detail with reference to the attached drawings.

[0064] FIG. 1 is a conceptual diagram illustrating an embodiment of a non-terrestrial network system according to the present disclosure.

[0065] Referring to FIG. 1, a non-terrestrial network system according to the present disclosure may include at least a user terminal (100), a satellite (200), and an AI assistant server (300). Additionally, it may further include external service providers that interact with the AI ​​assistant server (300), such as an emergency response center (410) and a telephone reservation center (420).

[0066] A user terminal (100) according to the present disclosure may be configured to transmit an emergency message from a vehicle (101).

[0067] In this specification, the user terminal (100) is not limited to portable devices such as smartphones. In the embodiments, it should be noted that the vehicle (101) itself may perform the role of the user terminal (100), or that components of the user terminal (100) may be implemented integrally within the vehicle (101). Accordingly, in this specification, the user terminal (100) and the vehicle (101) may be used interchangeably or interpreted as the same entity depending on the method of implementation, and the scope of the invention is not limited by a specific form.

[0068] The user terminal (100) can collect user voice data, vehicle sensor data, and UI input through the user terminal (100).

[0069] Here, vehicle information can be understood as a comprehensive concept that includes not only vehicle state information, such as vehicle speed, steering angle, acceleration, and collision sensor data, but also current vehicle location information, terrain information along the driving path, and communication environment information. Furthermore, driver information may include not only driver state information, such as patient information and psychological state, but also extensive user information, including the number of occupants in the vehicle, user identification information per occupant, user preferences, and past usage history.

[0070] In an embodiment, the user terminal (100) can collect voice data from a user (102) (e.g., driver, passenger, etc.) located inside or outside the vehicle (101) and convert it into data for transmission via a satellite network.

[0071] In this regard, the user terminal (100) may be configured to convert the voice data of the user (102) into text data and to transmit and receive the converted text data to and from the satellite (200). To this end, the user terminal (100) may perform interactive text-to-speech functions on-device. That is, the terminal itself may be configured to perform functions such as converting user voice input into text data or converting a text-format response received from the AI ​​assistant server (300) into synthesized speech and outputting it. This may be particularly useful in cases where available wireless resources are limited due to the characteristics of satellite communication in an NTN environment, or where it is difficult to transmit large volumes of voice data in emergency situations such as emergency rescue.

[0072] Meanwhile, the user terminal (100) may, in some cases, directly transmit and receive voice data with the satellite (200). In this case, the user terminal (100) may be configured to compress the received voice signal using a predetermined voice codec or convert it into a packet-unit voice frame for transmission. Additionally, the voice data received from the satellite (200) may be restored and output to the user.

[0073] In another embodiment, the user terminal (100) may collect sensor data from one or more sensors included in the vehicle (101). In this regard, the sensor data may include data regarding the vehicle state, the vehicle surrounding environment, and the driver state obtained through a driving sensor, a bio-sensor, a posture recognition sensor, a camera, a microphone, or a collision detection sensor included in the vehicle (101).

[0074] In another embodiment, the user terminal (100) may receive not only voice data of the user (102) but also input data generated by the user (102) operating the terminal (100). In this regard, the input data may refer to user interface (UI) input generated by using button input, touch input, UI menu selection, or emergency call function.

[0075] Additionally, the user terminal (100) can use an on-device artificial intelligence model to analyze at least one of voice data, sensor data, and user interface input (UI Input) to obtain at least one of vehicle information and driver information.

[0076] The user terminal (100) can classify driver situations by situation type based on acquired vehicle information and driver information.

[0077] Here, "Context Type" may be defined to represent a driver situation. In an embodiment, the context type may be set to Type 0 to Type 4.

[0078] For example, 'Type 0' may refer to 'emergency call' situations requiring immediate rescue (e.g., a Direct 911 call). Types 1 and 2 may refer to 'first aid guide' situations where a first aid manual is needed for medical advice or simple measures, even though life is not in danger. Additionally, Type 3 or Type 4 may refer to 'phone reservation' or 'information search' situations involving phone reservations and information searches for general situation-based information conversations or guidance services.

[0079] Meanwhile, the classification system of situation types in this specification is not limited to the aforementioned Types 0 through 4. In the embodiments, new situation types may be additionally defined as the types of collected multimodal data expand or as the service scenarios provided through the system according to the present disclosure become more diverse. Furthermore, it should be noted that multiple pre-configured situation types may be integrated into a single category and operated according to the system operation method. Accordingly, Types 0 through 4 exemplified in this specification are merely functional classifications to aid in understanding the invention, and the scope of the invention is not limited by specific names or numbers.

[0080] Additionally, the situation type can be determined based on a trigger type that indicates an input type triggered by any one of user voice input, sensor input, and UI operation input.

[0081] For example, if a vehicle collision sensor is detected or the tone and intensity of the user's voice are analyzed to be above a preset threshold, the situation type may be determined as 'Type 0 (Emergency Call)', which is the highest priority. In another example, if the location information included in the input information is analyzed and the vehicle is located in an area with a high risk of secondary accidents, such as a highway or a tunnel, or corresponds to a communication blind spot, the user terminal (100) may assign a preset weight to increase the emergency urgency level and determine the situation type accordingly.

[0082] Next, the user terminal (100) can generate a corresponding command message based on vehicle information and driver information. For example, if the vehicle information and driver information indicate an emergency situation such as sudden stop, risk of collision, decreased consciousness, or abnormal vital signs, the user terminal (100) can generate a command message that responds to a rescue request.

[0083] Additionally, the user terminal (100) may be configured to extract necessary data corresponding to a classified situation type. In an embodiment, if the situation type is an emergency call or an emergency action guide, it may include at least one of whether a collision has occurred, vehicle identification information, location information, vehicle speed, and occupant information. In another embodiment, if the situation type is a phone reservation or information search, it may include at least one of location information, user information, vehicle information, and service request information.

[0084] Next, the user terminal (100) can generate a command message including text data corresponding to the situation time, extracted necessary data, and voice data, summarize and compress it, and transmit it to the satellite (200).

[0085] At this time, the user terminal (100) may determine the summary level or compression rate differently based on the situation type and the quality of the satellite communication link. For example, in an emergency situation or when the communication quality is low, the text summary level may be increased or a high compression algorithm may be applied to increase the data transmission success rate.

[0086] The AI ​​assistant server (300) can transmit and receive voice data or text data with the satellite (200) and provide a service that responds to the user's situation by linking with the emergency response center (410) and the telephone reservation center (420).

[0087] The AI ​​assistant server (300) according to the present disclosure may operate in conjunction with a gateway infrastructure operated by a satellite communication operator. For example, the AI ​​assistant server (300) may operate at a location adjacent to a gateway device connecting the satellite (200) and a terrestrial network, or may be configured as a server cluster within the gateway device. In another embodiment, the AI ​​assistant server (300) may be deployed in an external cloud environment built on a public or private network to receive and process data transmitted through the satellite gateway.

[0088] Additionally, the AI ​​assistant server (300) may be configured to operate in conjunction with a Large Language Model (LLM) (310). Here, the LLM (301) may refer to an artificial intelligence model that performs natural language understanding and response generation for user input by pre-training a large-scale corpus. Additionally, the LLM (301) may be provided in the form of a cloud-based AI service.

[0089] The AI ​​assistant server (300) according to the present disclosure may be configured to analyze a command message received from a user terminal (100) and perform an operation corresponding to a situation type included in the message.

[0090] In an embodiment, when a command message with a situation type of Type 0 is received, the AI ​​assistant server (300) can be configured to activate a server-based conversational AI to communicate with an emergency response center (410) to transmit an emergency rescue request and to perform a voice call on behalf of the user.

[0091] In another embodiment, when a command message of type 1 or type 2 is received, the AI ​​assistant server (300) can generate a prompt suitable for the situation and provide it to the LLM (301). In response, the server (300) receives an LLM response containing the situation analysis results from the LLM (301) and can re-analyze it. At this time, if the situation is determined to be serious, it can immediately switch to an emergency call mode and perform fail-safe logic to connect to an emergency response center (410). Otherwise, the server (300) can reconstruct the LLM response and transmit it to the user terminal (100).

[0092] In another embodiment, when a command message of type 3 or type 4 is received, the AI ​​assistant server (300) may provide a corresponding service by interacting with a phone reservation center (420) or an external search engine. To this end, the AI ​​assistant server (300) may generate a response message including the result of performing an automatic reservation or searching for information.

[0093] Afterwards, the user terminal (100) can perform two-way communication by receiving a response message through the satellite (200), translating the text data included in the response message into a target language, or converting it into voice through TTS to guide the user.

[0094] Meanwhile, the classification of situation types processed by the AI ​​assistant server (300) in this specification and the corresponding service modes are not limited to the aforementioned Types 0 to 4. The server (300) according to the present disclosure may additionally define new situation types and dynamically expand new service modes corresponding thereto as the types of multimodal data collected from the user terminal (100) expand or as the service scenarios provided through the system according to the present disclosure become more diverse. Furthermore, depending on the load distribution or operation policy of the system, a plurality of pre-configured situation types and service modes may be merged into a single integrated intelligent service category and operated, or conversely, a specific type may be differentiated and operated according to detailed situations. Therefore, it should be noted that each type (Type 0 to 4) and the service functions matched thereto exemplified in this specification are merely exemplary classifications to aid in understanding the invention, and the technical scope of the invention is not limited by specific names, numbers, or fixed mapping relationships.

[0095] FIG. 2 illustrates a block diagram of a user terminal (100) according to the present disclosure.

[0096] Referring to FIG. 2, a user terminal (100) according to the present disclosure may be configured to include at least a sensor unit (110), a data processing unit (120), an audio processing unit (130), a memory (140), a communication module (150), a GPS module (160), a processor (170), a display (180), and an AI module (190). Each of the above components transmits and receives data to and from one another through a bus or other electrical connection means and may be connected so as to be operable under the control of the processor (170).

[0097] The sensor unit (110) is an integrated sensing unit for detecting changes in the physical state of the vehicle and the driver, and can perform motion detection, vehicle state, and biosignal detection functions depending on the purpose of detection.

[0098] In the embodiment, the sensor unit (110) includes an IMU (Inertial Measurement Unit) to detect in real time whether the vehicle has suddenly decelerated, impacted, or overturned. Additionally, the sensor unit (110) can cross-verify whether an accident has occurred by collecting airbag deployment signals and tire pressure information in conjunction with an internal vehicle network. Furthermore, the sensor unit (110) can determine the driver's consciousness status and the level of emergency by measuring the driver's biometric signals. The data collected from each sensor is transmitted to the data processing unit (120), undergoes synchronization and noise filtering processes, and can then be used as input data for situational judgment by the AI ​​module (190).

[0099] The data processing unit (120) may refer to a preprocessing module that processes data (Raw Data) collected from the sensor unit (110) into a form suitable for processing by the AI ​​module (190).

[0100] In an embodiment, the data processing unit (120) can apply a low-pass filter or Kalman filter algorithm to remove noise generated during driving and extract only meaningful impact signals. Additionally, the data processing unit (120) can ensure data consistency by correcting or interpolating the sampling rate based on the timestamps of heterogeneous sensor data collected at different intervals. Furthermore, the data processing unit (120) can optimize computational efficiency by normalizing the collected data and converting it into a vector or tensor form that fits the input specifications of the AI ​​module (190). The preprocessed data can be transmitted to the AI ​​module (190) and used as core input data for accident type classification and urgency determination.

[0101] The audio processing unit (130) can perform the role of converting an analog voice signal input through a microphone into a digital signal. To this end, the audio processing unit (130) may be configured to include a digital signal processor (DSP) that extracts voice data and acoustic features through a preset algorithm.

[0102] In an embodiment, the audio processing unit (130) can perform the function of converting a voice signal into text data using an STT model stored in memory (140). At the same time, the audio processing unit (130) can analyze the input voice waveform in the frequency domain or time domain to calculate an acoustic feature vector including pitch, intensity, frequency variation, amplitude variation, and speech rate. Subsequently, the audio processing unit (130) can be configured to transmit the calculated feature vector to an AI module (190).

[0103] Additionally, the audio processing unit (130) may include a function to convert text data into a synthesized speech signal by utilizing a TTS model stored in memory (140). Specifically, the audio processing unit (130) may convert a response message in text format received from a communication module (150) or an AI module (190) into a synthesized speech signal through the TTS model. Subsequently, the audio processing unit (130) may control the converted signal to be output to a speaker via a digital-to-analog converter (DAC).

[0104] Memory (140) may refer to a storage medium that stores data, programs, and algorithms necessary for the operation of a terminal. In an embodiment, memory (140) may be configured to include at least one of various types of non-volatile or volatile memory, such as flash memory, RAM, and ROM.

[0105] Additionally, the memory (140) can perform the function of temporarily or permanently storing programs or applications for controlling the processor (170), or input / output data. In an embodiment, the memory (140) can store offline map data containing road network and terrain information in preparation for situations where network connectivity is unavailable. Through this, the terminal can identify geographical environment information even in communication blind spots.

[0106] In addition, the memory (140) can store pre-trained artificial intelligence models and weight parameters required for the operation of the AI ​​module (190) described later. Through this, the memory (140) can provide an environment in which the terminal can independently perform on-device operations without communication with an external server.

[0107] The GPS module (160) may refer to a positioning device that receives signals from at least one Global Navigation Satellite System (GNSS) and calculates the terminal's current location coordinates (latitude, longitude, altitude, etc.) and movement speed in real time. In an embodiment, the GPS module (160) may be configured to integrally support various global satellite navigation systems including GPS, GLONASS, Galileo, and BeiDou.

[0108] Additionally, the GPS module (160) may include a function to generate uninterrupted location information by performing dead reckoning in conjunction with IMU data from the sensor unit (110) for precise positioning in shadow areas where satellite signal reception is difficult, such as tunnels or underground parking lots.

[0109] This GPS module (160) operates independently regardless of whether it is connected to a terrestrial network, and in particular, when an emergency situation based on satellite communication occurs, it can acquire precise coordinate information of the accident location and transmit it to a server (300) or an emergency response center (410).

[0110] Meanwhile, the GPS module (160) and the memory (140) can work together to perform location-based situational awareness of the terminal. In an embodiment, the processor (170) can match the current coordinates of the terminal, tracked in real time through the GPS module (160), with offline map data stored in the memory (140). Through this, the processor (170) can determine specific geographical environments, such as whether the vehicle is currently on a highway or in a mountainous area. This location information can be used as an important basis for determining the level of urgency in the situational awareness process described later.

[0111] The AI ​​module (190) is activated by loading an AI model stored in memory (140) and may refer to an on-device artificial intelligence processing unit that comprehensively determines the situation based on collected multimodal data. The AI ​​module (190) may be implemented as a separate neural network processing unit (NPU) or installed in the form of a software algorithm within the processor (170). Additionally, a lightweight model may be used to enable independent inference even when the network connection is disconnected.

[0112] In an embodiment, the AI ​​module (190) can perform multimodal situation analysis. Specifically, the AI ​​module (190) can fuse and analyze text data from the audio processing unit (130), impact data from the sensor unit (110), and geographical environment information. Through this, the AI ​​module (190) can finely classify whether the situation is a 'simple contact accident' or a 'life-threatening rollover accident' and calculate a severity level accordingly.

[0113] Additionally, the AI ​​module (190) can perform a function to analyze the driver's psychological state. To this end, the AI ​​module (190) can infer the psychological state by analyzing acoustic characteristics such as the pitch, tremors, and speech rate of the voice waveform. For example, the AI ​​module (190) can determine that severe voice tremors indicate a 'panic' state, and that unclear speech indicates a 'decreased consciousness' state. This psychological state information can be included as metadata in the distress request message.

[0114] Furthermore, the AI ​​module (190) can perform text optimization suitable for a satellite communication environment. The AI ​​module (190) can remove unnecessary particles or redundant expressions from the text and reconstruct sentences by extracting only core keywords. Through this intelligent compression process, the AI ​​module (190) can be configured to minimize the size of data packets to increase the transmission success rate.

[0115] The communication module (150) may refer to a hybrid communication interface that supports both terrestrial network communication and non-terrestrial network communication. Specifically, the communication module (150) may be configured to include a first communication modem and a second communication modem that are physically or logically separated.

[0116] In the embodiments, the first communication modem may refer to a broadband mobile communication modem that communicates with a ground base station, such as LTE or 5G NR (New Radio). This can be used for the vehicle's infotainment or general data communication in normal times. Additionally, the second communication modem may refer to a satellite-dedicated modem (e.g., NR-NTN, IoT-NTN modem) that communicates directly with a Low Earth Orbit (LEO) or Geostationary Orbit (GEO) satellite.

[0117] The processor (170) can control the communication module (150) to select a modem suitable for the situation. For example, in urban areas where the signal strength (RSRP) of the terrestrial network is above a threshold, the first communication modem can be activated to maintain high-speed communication. On the other hand, the location of the accident may be determined to be a terrestrial network blind spot, such as a remote mountainous area or the sea. Alternatively, the antenna of the first communication modem may be damaged due to the impact of the accident, making connection impossible. In such cases, the processor (170) can immediately activate (Wake-up) the second communication modem to form a satellite link. This dual modem-based automatic switching can be configured to reliably secure an external transmission path for the rescue signal even when the communication environment is poor.

[0118] The display (180) may refer to an input / output device that provides visual information to the user and receives touch input. In terms of hardware, the display (180) may perform the role of generating a driving signal by converting video signals, data signals, OSD (On Screen Display) signals, etc., into R, G, and B signals, respectively.

[0119] In an embodiment, the display (180) can visually output system status information under the control of the processor (170). Specifically, the display (180) can display the transmission progress status of a command message and the satellite connection strength, etc.

[0120] Additionally, the display (180) may be configured to render a graphical user interface (GUI) corresponding to the situation type identified by the processor (170). For example, the display (180) may display a map based on location information from the GPS module (160) and map data from the memory (140). Alternatively, it may output emergency first aid guide images and text data.

[0121] In addition, the display (180) can support the function of converting and outputting the received response message into the user's native language under the control of the processor (170).

[0122] The processor (170) may refer to a central processing unit (Main Controller) that controls the overall operation of the user terminal (100) and performs calculations. The processor (170) may be configured to analyze data collected from each module to determine an emergency situation and to control the generation of command messages. In terms of hardware, the processor (170) may be implemented as an application processor (AP), a central processing unit (CPU), or a microcontroller (MCU).

[0123] The processor (170) can perform integrated logic to control the signal flow between each component. Specifically, the processor (170) can recognize a physical impact signal exceeding a threshold value as an interrupt signal when input from the sensor unit (110). Accordingly, the processor (170) can control the application of a wake-up signal to the GPS module (160) and the AI ​​module (190) that were in standby mode.

[0124] Additionally, the processor (170) can receive the criticality determination result and the terrestrial network connection failure signal output from the AI ​​module (190). Based on this, the processor (170) can control the communication module (150) to instruct it to immediately switch the communication path to the satellite modem.

[0125] Furthermore, the processor (170) can identify the situation type by analyzing the header of the received response message. Then, the processor (170) can generate video and audio data signals corresponding to the identified type and control them to be transmitted to the display (180) and the audio processing unit (130), respectively.

[0126] FIG. 3 illustrates a block diagram of an AI assistant server (300) according to the present disclosure. Referring to FIG. 3, the AI ​​assistant server (300) according to the present disclosure may be configured to include at least a TTS / STT module (310), a data processing unit (320), an audio processing unit (330), a database (340), a communication module (350), an external linkage module (360), a processor (370), an LLM module (380), and a prompt generation module (390). Each of the above components transmits and receives data to and from one another through a bus or other electrical connection means and may be connected so as to be operable under the control of the processor (370).

[0127] The TTS / STT module (310) may refer to an engine that performs server-based high-performance voice conversion and dialogue processing functions based on a signal received from the audio processing unit (330).

[0128] In an embodiment, the TTS / STT module (310) can convert compressed voice data transmitted from the terminal (100) into high-precision text data through a STT (Speech-to-Text) function. In particular, it can be configured to secure a high recognition rate even in emergency situations where noise is mixed in or speech is unclear by utilizing the large-capacity computing resources of a server.

[0129] Additionally, the TTS / STT module (310) can incorporate an Interactive TTS (Text-to-Speech) function. This allows for the synthesis of speech that reflects intonation and emotion appropriate to the situation, going beyond simple text reading. For example, in emergency call mode, a voice response can be generated in a calm and clear tone that can reassure the driver, and in a natural conversational tone in phone reservation mode.

[0130] Furthermore, the TTS / STT module (310) can support multilingual interpretation and translation functions. Through this, when an accident occurs overseas or a reservation is made, it can perform real-time voice conversion between the user's native language and the local language to overcome language barriers.

[0131] The data processing unit (320) may refer to a preprocessing module that processes a compressed command message received through the communication module (350). Specifically, the data processing unit (320) may perform a decompression function that restores the received compressed packet to its original data form.

[0132] Subsequently, the data processing unit (320) can precisely analyze the header field and payload of the command message. Through this, the data processing unit (320) can separate multimodal data such as text, sensor values, and audio data, and ensure the time-series consistency of each data. In addition, it can perform the role of normalizing the separated data into a standard format (e.g., JSON, Vector, etc.) that can be processed by the processor (375) or LLM of the server (300).

[0133] Additionally, the data processing unit (320) can perform a variable compression function for a response message to be transmitted to the terminal (100). The data processing unit (320) can dynamically determine the compression rate based on the current satellite communication link quality or the urgency of the identified situation. For example, if the communication status is unstable or the urgency is high, the data processing unit (320) can apply a text-based high-compression algorithm. This can be configured to minimize the data size and reduce transmission delay.

[0134] The audio processing unit (330) may refer to a signal processing device that sets up and manages the voice conversation environment of the server (300). Specifically, the audio processing unit (330) may perform the role of controlling acoustic signals generated in a call line with an external organization (e.g., emergency center, restaurant, etc.).

[0135] In an embodiment, the audio processing unit (330) may apply high-performance echo cancellation (AEC) and noise reduction algorithms to the received voice signal. Through this, the audio processing unit (330) can refine the signal to maintain the purity of the voice signal even in a noisy environment. Subsequently, the audio processing unit (330) can transmit the refined audio signal to the TTS / STT module (310). This can serve as an essential preprocessing step to maximize the voice recognition rate of the STT engine.

[0136] Additionally, the audio processing unit (330) can perform the role of a gateway that converts the synthetic voice data generated by the TTS / STT module (310) into a codec suitable for an external telephone network (PSTN) or VoIP protocol and transmits it.

[0137] The database (340) may refer to a large-capacity storage medium that stores data and algorithms necessary for the operation of the server (300). In an embodiment, the database (340) may store essential configuration data for linking with an external LLM (301). Specifically, the database (340) may store prompt templates for each situation type used by the prompt generation module (390).

[0138] Additionally, the database (340) can store API authentication keys and connection information for external service integration. Through this, the database (340) can be configured to support an environment in which the server (300) can seamlessly integrate with the LLM (301) to provide intelligent services based on natural language processing.

[0139] The communication module (350) may refer to a hybrid communication interface that supports both terrestrial network communication and non-terrestrial network communication. Specifically, the communication module (350) may include a satellite network gateway interface for communication with a user terminal (100). Additionally, it may be configured to include an external public network interface for interoperability with external services.

[0140] In an embodiment, the communication module (350) can receive a command message transmitted from a user terminal (100) via a satellite. Additionally, it can perform the role of transmitting a response message processed within the server back through the satellite network.

[0141] Furthermore, the communication module (350) can form a communication channel with the emergency response center (410) and general service providers (e.g., telephone reservation center (420)) through an external public network. Through this, the communication module (350) can perform the function of immediately transmitting accident situation information to the emergency center or transmitting and receiving restaurant and accommodation reservation data with the service provider server.

[0142] At this time, message transmission and reception between satellites can be configured to be performed in compliance with standard non-terrestrial network (NTN) specifications such as NR-NTN and IoT-NTN.

[0143] The external linkage module (360) can act as a gateway to transmit the processing results of the server (300) to an external service and integrate the results.

[0144] In an embodiment, the external integration module (360) can determine the connection target according to the identified situation type. For example, in the case of an emergency situation, the external integration module (360) can establish an immediate call connection with the emergency response center (410) or transmit accident data. Additionally, in the case of an information search or reservation situation, the external integration module (360) can transmit API queries to an external search engine (430) or a general service provider (420) server. Subsequently, the external integration module (360) can collect and refine search results or reservation confirmation information received from the outside. Through this, the data can be integrated into a form that the server (300) can process and transmitted to the processor (375).

[0145] The LLM module (380) may refer to a computing device that performs unstructured data analysis by linking with an external LLM (301) that provides cloud AI services. In an embodiment, the LLM module (380) may serve as an interface that transmits a query received from the prompt generation module (390) to the external LLM (301). Subsequently, the LLM module (380) may transmit the analysis results received from the external LLM (301) to the processor (370).

[0146] The prompt generation module (390) may refer to a software engine that analyzes structured terminal information received from the data processing unit (320) and dynamically generates a natural language-based 'context-aware prompt' so that the LLM module (380) can perform optimal inference. Specifically, the prompt generation module (390) can reconstruct heterogeneous multimodal data, such as text transmitted from the terminal, sensor values, and biosignals, into a single integrated context that the LLM can understand.

[0147] In an embodiment, the prompt generation module (390) can refine system commands by utilizing situational templates stored in the database (340). For example, when rollover accident data is detected, the quality of the LLM response can be controlled by assigning a role such as "You act as a skilled emergency rescuer from now on." Additionally, the prompt generation module (390) can adjust the urgency and response style of the LLM request based on geographical location (e.g., a communication dead zone) and the driver's psychological state (e.g., fear, unconsciousness). This allows for configuration to elicit concise and clear guidance in emergency situations, and detailed and friendly responses in general reservation situations.

[0148] The processor (370) may refer to a central processing unit (Main Controller) that controls the overall operation of the server (300) and performs operations for large-scale data processing. Specifically, the processor (370) may perform the function of identifying the type of situation the terminal is currently in by analyzing the header information or payload of a received command message.

[0149] Additionally, the processor (370) can perform integrated control logic that determines the operation of the system according to the identified type. The processor (370) can selectively activate the LLM module (380), TTS / STT module (310), and external linkage module (360) depending on the situation. Through this, the processor (370) can be configured to provide services by branching to processing logic optimized for each operation mode, such as emergency calls, information search, and reservations.

[0150] Meanwhile, the user terminal (100) and AI assistant server (300) illustrated in FIGS. 1 to 3 are merely embodiments of the present invention, so some of the illustrated components may be integrated, added, or omitted depending on the specifications of the actual implemented user terminal (100) and AI assistant server (300).

[0151] That is, as needed, two or more components may be combined into a single component, or a single component may be subdivided into two or more components. In addition, the functions performed in each block are intended to explain embodiments of the present invention, and the specific operations or devices thereof do not limit the scope of the present invention.

[0152] Hereinafter, the configuration and transmission of command messages based on vehicle and driver status according to the present disclosure, and the corresponding situational response operations are described. In this regard, FIG. 4 illustrates a flowchart of the operation of a situation recognition and a situational service provision method corresponding thereto performed in a satellite communication system according to the present disclosure.

[0153] Referring to FIGS. 1 to 4, a service provision method based on driver conditions performed in a satellite communication system can be divided into a process of determining driver conditions and generating and transmitting related command messages by a user terminal (100) (S401 to S406), and a process of receiving and processing the same by an AI assistant server (300) (S407 to S414).

[0154] First, the user terminal (100) can acquire at least one of vehicle information and driver information (S401) based on at least one of collected voice data, sensor data and user input data.

[0155] In this regard, the user terminal (100) can collect data through sensors, microphones, cameras, or a user interface (UI) inside and outside the vehicle. Specifically, the terminal (100) can receive sensor data detecting whether the vehicle has collided, voice data including the driver's speech, and touch input from the user by monitoring in real time.

[0156] Next, the user terminal (100) can classify the driver situation into at least one of a plurality of situation types (S402) based on the acquired information.

[0157] In this regard, the user terminal (100) can determine whether there is an emergency situation or the level of emergency based on the acquired information. Additionally, the situation type can be classified based on whether there is an emergency situation or the level of emergency.

[0158] In an embodiment, the user terminal (100) can determine whether there is an emergency situation or the level of emergency urgency by analyzing vehicle information and driver information using an on-device AI model. At this time, the situation type may be classified based on whether there is an emergency situation or the level of emergency urgency, and may include at least one of an Emergency Call, an Emergency Action Guide, a phone reservation, and an Information Search. For example, if the determination result indicates an emergency situation, the user terminal (100) may classify the driver situation as a situation requiring an Emergency Call or an Emergency Action Guide. On the other hand, if it is determined not to be an emergency situation, the terminal (100) may classify the driver situation as a general assistant service request situation, such as a phone reservation or an Information Search.

[0159] Meanwhile, the user terminal (100) can determine the level of emergency by reflecting the driver's psychological state and location characteristics.

[0160] In an embodiment, the user terminal (100) can calculate a driver psychological level by analyzing acoustic features including at least one of tone, intensity, and speech rate of voice data. Here, the driver psychological level may refer to a quantitative numerical value or grade that quantifies a driver's psychological state, such as panic, fear, or composure. Subsequently, the terminal (100) can determine an emergency level by reflecting the calculated driver psychological level.

[0161] In another embodiment, the user terminal (100) can obtain location information from sensor data and calculate a risk weight corresponding to the location information. Here, the sensor data can be understood as a concept that includes not only location data obtained from the GPS module (160), but also location and terrain information estimated based on the inertial sensor data of the sensor unit (110).

[0162] In this regard, the user terminal (100) can analyze collected sensor data to obtain current vehicle location information and calculate a risk weight corresponding to the topographical or communication environment characteristics of the location. For example, if the vehicle (101) is located on a highway, in a tunnel, or in a remote area that is a communication blind spot, a higher risk weight may be assigned because the risk of accidents or the difficulty of rescue is high. The user terminal (100) can determine the emergency urgency level by reflecting the calculated risk weight.

[0163] Additionally, if the calculated emergency urgency level is above the first threshold, the user terminal (100) can classify the driver situation as 'Type 0 (Emergency Call)' and control an immediate connection with the emergency response center (410). On the other hand, if the emergency urgency level is below the first threshold but above the second threshold, it can classify the situation as 'Type 1 (Emergency Measure Guide)' by determining that an immediate emergency call is not necessary but medical advice is required. In this case, the user terminal (100) can control the AI ​​assistant server (300) to receive an emergency measure guide optimized for the driver situation and provide it to the user (102).

[0164] Next, the user terminal (100) can generate a command message (S403) based on the situation type and acquired information.

[0165] In this regard, the user terminal (100) can use a speech-to-text (STT) model to convert voice data into text data and add the converted text data to a command message.

[0166] In an embodiment, the user terminal (100) may perform signal correction by removing background noise from the voice data before converting the voice data into text. This is to prevent the voice recognition rate from being degraded due to driving wind noise or accident impact noise, and to obtain a clear voice signal.

[0167] Additionally, the user terminal (100) may perform cross-verification with sensor data to ensure the reliability of the converted text data. In an embodiment, the user terminal (100) may determine whether the text data and the collected sensor data are different from each other. If the text data and the sensor data are different from each other, the user terminal (100) may apply priority to the sensor data and perform content correction on the text data.

[0168] For example, if the driver says "I'm fine" due to the impact but the sensor data indicates "airbag deployment" and "vehicle rollover," the user terminal (100) can generate a command message by prioritizing the sensor data and forcibly inserting or modifying the phrase "[Automatic detection: Rollover accident occurred]" into the text data.

[0169] In an embodiment, the generated command message may be generated with a structure including a header field in which an identified situation type is recorded, and a payload field in which the converted or corrected text data is recorded.

[0170] Next, the user terminal (100) can extract necessary data associated with the situation type and add it to the generated command message (S404).

[0171] In this regard, the user terminal (100) can selectively extract necessary data corresponding to a situation type determined from memory or vehicle network (CAN).

[0172] In an embodiment, when the situation type is an emergency call or first aid guide, the required data may include at least one of whether a collision occurred, vehicle identification information, location information, vehicle speed, and occupant information. For example, collision sensor values, vehicle identification number (VIN), GPS coordinates, speed at the time of the accident, and the number of occupants may be included, and this may constitute a minimum data set (MSD) essential for rescue operations.

[0173] On the other hand, if the situation type is a phone reservation or information search, the required data may include at least one of location information, user information, vehicle information, and service request information. For example, it may include the current vehicle location, the name of the reservation holder, the vehicle type, and search keywords entered by the user (e.g., 'nearby restaurants') or reservation time information.

[0174] The user terminal (100) can construct a transmission message by adding the necessary data extracted according to the situation to the payload of the command message.

[0175] Next, the user terminal (100) can perform summarization or compression on the command message (S405).

[0176] In this regard, the user terminal (100) may perform data lightweighting to reduce the size of the command message in order to increase data transmission efficiency in consideration of the limited satellite bandwidth. Here, data lightweighting may mean summarizing or compressing the command message. At this time, the summarization level or compression rate for the command message may be applied differently based on at least one of the situation type and the satellite communication link quality. That is, the terminal (100) may perform a variable compression policy depending on the urgency of the situation or the communication status.

[0177] In an embodiment, if the situation type is an emergency call or an emergency action guide, or if the satellite communication link quality is below a threshold standard, the user terminal (100) can compress the command message with a first compression rate. Here, the first compression rate may be a lossless compression method that ensures the accuracy of information by minimizing data loss, or a high compression method that drastically reduces the packet size by prioritizing the transmission success rate.

[0178] On the other hand, if the situation type is a phone reservation or information search and the satellite communication link quality is above a threshold standard, the user terminal (100) can compress the command message at a second compression rate lower than the first compression rate. This is to preserve more of the quality or details of the data in situations where the communication environment is good and not urgent.

[0179] Additionally, the user terminal (100) may apply different summarization levels depending on the situation. For example, in an emergency situation, it may perform a 'high-intensity summarization' that extracts only core keywords, such as "collision detected, unconscious." In addition, in a non-emergency situation, it may perform a 'low-intensity summarization' that maintains the context of the sentence, such as "hope to make a reservation at an Italian restaurant near Gangnam Station."

[0180] Next, the user terminal (100) can transmit a lightened command message, i.e., a summarized or compressed command message, to the satellite (S406).

[0181] Meanwhile, the AI ​​assistant server (300) that receives a command message via the satellite (200) can analyze the received command message (S407). To do this, the server (300) can first decompress the received command message.

[0182] Next, the server (300) can analyze the command message to identify the type of situation included in the command message and obtain situation information (S408).

[0183] In this regard, the server (300) can parse the header information of the restored command message to identify a situation type indicating the type of service requested by the terminal (100). Subsequently, the server (300) can obtain situation information by analyzing the command message. In this regard, the server (300) can obtain at least one of rescue request information, emergency measure information, reservation information, and search information according to the identified situation type.

[0184] In an embodiment, rescue request information may include the vehicle's latitude, longitude, collision location, and whether the airbag deployed. Emergency response information may include the patient's state of consciousness, location of pain, and breathing status. Reservation information may include the reservation holder's name, number of people, desired time, and type of location. Search information may include search keywords, search radius, and category information.

[0185] Next, the server (300) may operate in at least one of an emergency call mode, an emergency action guide mode, a phone reservation mode, and an information search mode depending on the identified situation type. That is, the server (300) may branch to and operate in at least one of the following steps (S410 to S413) based on the identified situation type.

[0186] In an embodiment, if the identified situation type is an emergency call, the AI ​​assistant server (300) can operate in an emergency call mode (S410).

[0187] In this regard, the server (300) can enable server-based conversational AI to connect emergency response center (410) with emergency calls.

[0188] Next, the server (300) can perform a voice-based conversation with the emergency response center (410) based on the acquired rescue request information. Specifically, the server (300) can process the real-time conversation by converting the received text data into speech (TTS) or converting the rescuer's voice into text (STT).

[0189] Afterward, the server (300) can generate a response message to be transmitted to the user terminal (100) to include feedback information including emergency personnel dispatch information and the results of the action.

[0190] Meanwhile, if the identified situation type is an emergency response guide, the server (300) may perform a situation analysis (S409) in conjunction with the LLM (301) prior to the emergency response guide step (S411).

[0191] In this regard, the above situation analysis (S409) may include the following steps. First, the server (300) may construct an LLM request (LLM Request) requesting a situation analysis using a situation-specific prompt model and send it to the LLM (301).

[0192] Next, the server (300) can receive an LLM response (LLM RESPONSE) containing the results of the situation analysis in response to the transmission of the LLM request. Next, the server (300) can analyze and reconstruct the received LLM response.

[0193] Additionally, based on the result of reconstructing the LLM response, if the server (300) determines that the analysis result of the LLM is greater than or equal to a preset level of urgency, it can immediately switch to an emergency call mode (step S410) and connect an emergency call with an emergency response center.

[0194] Next, the server (300) can continue to operate as an emergency response guide (S411), analyze the reconstructed LLM response to obtain multiple step-by-step guide texts, and generate a response message containing the obtained multiple step-by-step guide texts.

[0195] In another embodiment, if the identified situation type is a phone reservation, the AI ​​assistant server (300) can operate in phone reservation mode (S412).

[0196] In this regard, the server (300) can automatically perform the reservation procedure based on the acquired reservation information. First, the server (300) can activate a server-based conversational AI to connect a call with the reservation destination included in the reservation information (e.g., the telephone reservation center (420) of FIG. 1).

[0197] Next, the server (300) can perform a voice-based conversation with the reservation target based on the reservation information. This may mean a process in which the server (300) communicates with the other party, such as a restaurant employee, using TTS / STT technology to confirm the reservation on behalf of them.

[0198] Finally, the server (300) may generate a response message that includes feedback information including whether the reservation was successful and details of the reservation. This response message may be transmitted to a user terminal (100) and used to inform the user of the reservation confirmation details.

[0199] In another embodiment, if the identified situation type is information retrieval, the AI ​​assistant server (300) can perform an information retrieval mode operation (S413).

[0200] In this regard, the step of operating in information search mode can be performed as follows. First, the server (300) can generate a search query based on the acquired search information. Then, the server (300) can perform information search by linking with an external search engine.

[0201] Next, the server (300) can extract key information corresponding to the user's intent from the result data of the information search and perform text summarization based thereon. For example, the server (300) can lighten the data by summarizing only essential information, such as addresses or contact information for the top 2 or 3 best places, instead of dozens of search result links.

[0202] Finally, the server (300) can generate a response message containing summarized text data. The response message can be transmitted to the terminal (100) via the satellite (200) and provided to the user.

[0203] Meanwhile, the AI ​​assistant server (300) can transmit (S414) a generated response message, including the results of each step (S410 to S413), to the user terminal (100). In an embodiment, the server (300) can summarize and compress the generated response message and transmit it to the user terminal (100) via the satellite (200). At this time, the response message may include a control command that causes the user terminal (100) to provide voice guidance via on-device TTS or to output an automatically translated message according to the country code.

[0204] Accordingly, the user terminal (100) can receive a response message for a transmitted command message from the server (300) via the satellite (200). Additionally, the user terminal (100) can perform a response action associated with the received response message. In an embodiment, depending on the TTS command included in the response message, the user terminal can perform actions such as providing voice guidance on emergency action guide content or displaying reservation confirmation details or summary results of information search on a UI screen.

[0205] Meanwhile, although the above description describes steps S407 to S414 as being performed by an AI assistant server (300), the technical concept of the present disclosure is not limited thereto. According to various embodiments of the present disclosure, at least some of the processes of steps S407 to S414 may, depending on the case, be performed independently by a processor and an on-device AI model installed in the user terminal (100). Additionally, it should be noted that the processes may be organically distributed and processed among the user terminal (100), the satellite (200), and the server (300) depending on the system environment.

[0206] Hereinafter, data flow and interoperability methods between components according to situation types performed in a satellite communication system according to the present disclosure will be described. In this regard, FIG. 5 is a conceptual diagram illustrating data flow and interoperability between components according to situation types performed in a satellite communication system according to the present disclosure.

[0207] Referring to FIGS. 1 to 5, a method for providing driver-based services performed in a satellite communication system can be implemented through mutual interaction between a user terminal (100), a satellite (200), an AI assistant server (300), and an external service provider. In an embodiment, the external service provider may include an LLM (301), an emergency response center (410), a general service provider (420) such as a phone reservation, and an external search engine (430).

[0208] First, the user terminal (100) can configure a command message (S10) based on vehicle information and driver information.

[0209] In this regard, the user terminal (100) can primarily determine the vehicle status and driver status through on-device AI and classify the driver situation by situation type. Additionally, the user terminal (100) can extract necessary data based on the vehicle status and driver status and generate a command message containing the situation type and necessary data. In some cases, text data corresponding to voice data may be further included in the command message. The user terminal (100) can summarize and compress the generated command message and transmit it to the AI ​​assistant server (300) via a wireless link with the satellite (200).

[0210] The AI ​​assistant server (300) that receives the command message can decompress and analyze the message to identify the situation type (S20), which is the type of service requested by the terminal.

[0211] Next, the server (300) can obtain specific situation information corresponding to the identified situation type from the command message (S30). For example, the server (300) can obtain at least one of rescue request information, emergency action information, reservation information, and search information depending on the situation type.

[0212] Additionally, the server (300) can operate in conjunction with different external interfaces (410, 420, 430) according to the acquired information.

[0213] In an embodiment, if the situation type is identified as 'emergency call' in step S20, the server (300) may acquire rescue request information in step S30 and operate in an emergency call mode (S40). In this case, the server (300) may establish a communication channel with the emergency response center (410) and activate a server-based conversational AI. Through this, the server (300) may convert a rescue request received as text into speech (TTS) and deliver it to a rescue worker. Additionally, the server (300) may recognize the rescue worker's voice response as text (STT) and process it.

[0214] In another embodiment, if the situation type is identified as 'phone reservation' in step S20, the server (300) may obtain reservation information including the reservation date and time and location information in step S30 and operate in a phone reservation mode (S60). In this case, the server (300) may establish communication with a service provider (420), such as a restaurant or an airline, and perform an automatic reservation call through a server-based conversational AI.

[0215] In another embodiment, if the situation type is identified as 'information search' in step S20, the server (300) may acquire search information including search keywords, etc. in step S30 and operate in an information search mode (S70). In this case, the server (300) may perform an information search by linking with an external search engine (430) API and extract and summarize only the core information that matches the user's intent from the search results.

[0216] In another embodiment, if the situation type is identified as 'Emergency Measure Guide' in step S20, the server (300) may acquire emergency measure information including patient symptom information in step S30 and operate in an emergency measure guide mode (S50). In this case, the server (300) may perform a situation analysis (S51) in conjunction with an external large-scale language model (LLM, 310) to provide an additional accurate emergency measure guide. Specifically, the server (300) may transmit a prompt (LLM REQUEST) containing situation information to the LLM (301) and receive an analysis result (LLM RESPONSE) therefrom. Subsequently, the server (300) may reconstruct the response from the LLM to generate a step-by-step emergency measure manual (S52) and transmit it to a user terminal (100).

[0217] Finally, the AI ​​assistant server (300) can generate a response message containing the final result performed in each of the above modes (S40, S50, S60, S70) and transmit it to the user terminal (100) via the satellite (200) (S80). Upon receiving the response message, the user terminal (100) can activate a built-in on-device conversational AI function to convert the received text information into synthesized speech (TTS) and provide guidance to the user (102) or display it on the screen.

[0218] Hereinafter, the operation of generating and transmitting a command message performed in a user terminal according to the present disclosure will be described. In this regard, FIG. 6 is an operation flowchart of a method for generating and transmitting a command message performed in a user terminal according to the present disclosure. FIG. 7 also illustrates the data flow for the process of generating, summarizing, and compressing a command message performed in a user terminal according to the present disclosure.

[0219] Referring to FIGS. 1 to 7, the operation of generating and transmitting a command message can be performed by the processor (170) of the user terminal (100).

[0220] First, the processor (170) can receive a trigger input from the user terminal (100) and check the input type (S11). The trigger input may include at least one of user voice, vehicle sensor data, or UI input. Here, the UI input may refer to input data generated by the user (102) operating the terminal (100).

[0221] Specifically, user voice is a natural language-based utterance input through an in-vehicle microphone, which can then be converted into text through correction and STT processes. Vehicle sensor data can be acquired through accelerometers, collision detection sensors, airbag sensors, biometric sensors, or posture recognition sensors, and can reflect the vehicle state and driver state. In the embodiments, vehicle sensor data may further include information regarding the vehicle's surrounding environment based on ambient images and noise acquired through cameras or microphones. Additionally, UI input may include an emergency call button, touchscreen operation, UI menu selection, or input via a vehicle control panel.

[0222] The processor (170) can determine the trigger type based on the source and content of the trigger input. In an embodiment, the processor (170) can determine (S12) whether the trigger input is a sensor input as the top priority for priority processing.

[0223] In the case of sensor input ('Yes' in S12), the processor (170) may further determine (S13) whether there is consciousness or a reaction of the driver based on driver information. At this time, the processor (170) may detect the driver's movements or groans by analyzing in-vehicle camera footage or microphone input using an on-device AI model. If the determination result indicates that consciousness or a reaction is present, it may be interpreted that the user is in a state where they can check the situation. Since this may not be an emergency situation requiring immediate rescue, the processor (170) may return to step S11 or switch to a waiting state to wait for the next trigger input.

[0224] On the other hand, if there is no consciousness or response ('No' in S13), it is determined that the driver is in an emergency situation where normal operation or response is impossible, and the driver situation can be classified as TYPE 0. Here, TYPE 0 may mean an 'emergency call' situation. Subsequently, the processor (170) can generate a command message corresponding to the classified type and set the generated message to include the corresponding type (S15). In this case, a message containing a Minimum Set of Data (MSD) for an emergency rescue request may be generated.

[0225] Meanwhile, if it is determined in step S12 that the input is not a sensor input ('No' in S12), the processor (170) can determine whether the input is due to UI operation (S14). If it is a UI input ('Yes' in S14), the processor (170) can determine that the user intended a general service request or an information search call. Accordingly, the processor (170) can classify the driver situation as TYPE 3 or TYPE 4. Here, TYPE 3 may mean a 'phone reservation' situation and TYPE 4 may mean an 'information search' situation. Subsequently, the processor (170) can generate a command message corresponding to the classified type and set the generated message to include the corresponding type (S15).

[0226] Additionally, if it is determined in step S14 that it is not a UI input ('No' in S14), the processor (170) may determine that the trigger input is a user voice. At this time, the processor (170) may perform corrections on the input voice data, such as removing ambient noise, and then perform STT to convert it into text data. Additionally, based on the converted text content and the input situation, the processor (170) may classify the driver situation type as corresponding to either TYPE 1 or TYPE 2. Here, TYPE 1 and TYPE 2 may refer to 'emergency action guide' situations. Subsequently, the processor (170) may be configured (S16) to set the classified type and generate a text-based command message corresponding to that type.

[0227] Afterward, the processor (170) can extract necessary data corresponding to the set situation type (S17). Depending on the situation type, the necessary data may include various data required for type-specific processing, such as MSD, location information, sensor information, STT converted text, and reservation information.

[0228] Next, the processor (170) can construct a final transmission message (S18) by combining the extracted necessary data, command message, and situation type. Subsequently, to increase transmission efficiency in an NTN environment, the transmission message can be summarized and compressed (S19). At this time, the processor (170) can apply a compression rate determined based on the urgency of the situation type or the communication quality. Finally, the compressed transmission message can be transmitted via satellite communication using a satellite communication modem.

[0229] In this regard, referring to FIG. 7, the user terminal (100) can obtain various input information (701).

[0230] In an embodiment, the input information (701) may be directly obtained in the form of raw data such as user voice, vehicle identification number (VIN), altitude (alt), number of occupants (occu), and blood type (ABO). In another embodiment, the input information may be vehicle and driver information obtained by processing through an on-device AI model. For example, patient information may include gender (male), age (55 years), blood type (B), disorientation, head, and chest pain. Additionally, vehicle information may include airbag deployment, vehicle rollover X, 60 km / h, etc.

[0231] That is, the user terminal (100) can analyze the received voice data, sensor data, and UI input using an on-device AI model and convert them into vehicle information and driver information necessary for situational awareness to obtain them.

[0232] Next, the user terminal (100) can form an initial data set (702) by combining text data that has undergone STT conversion with an initial MSD (needed data). This initial data set (702) contains all collected raw information and may be, for example, about 366 bytes in size.

[0233] Next, the user terminal (100) can perform text data summarization and necessary data selection processes to maximize message transmission efficiency. Through the above process, an optimized data set (703) is generated, which includes only situation-specific essential MSD data and can be reduced in size, for example, to about 269 bytes.

[0234] Next, the user terminal (100) can apply a compression algorithm to the summarized and selected data set to generate a final compressed data set (704). The compressed data set (704) is generated with a size of, for example, about 217 bytes, and then the user terminal (100) can transmit it to the AI ​​assistant server (300) via the satellite (200).

[0235] Meanwhile, the AI ​​assistant server (300) that receives the compressed command message (704) can analyze the received compressed data set.

[0236] First, the server (300) can analyze the received command message to perform situation recognition corresponding to the driver's situation. That is, based on the situation type, text data, and necessary data of the command message, it can identify the specific situation of the driver and obtain information necessary for that situation (rescue request information, reservation information, etc.).

[0237] Next, the server (300) may perform an initial step of selecting an optimal prompt model to query an external LLM (301) based on at least one of the identified situation type and acquired situation information. At this time, the external LLM (301) may be implemented as a server (Cloud AI Server) or platform that provides cloud-based AI services.

[0238] Hereinafter, a method for analyzing LLM integration situations performed by an AI assistant server according to the present disclosure is described. In this regard, FIG. 8 is an operation flowchart of the process for determining a response mode based on LLM integration performed by an AI assistant server according to the present disclosure. FIG. 9 also illustrates an example of data processing and mode switching in which an AI assistant server according to the present disclosure reconstructs LLM analysis results and provides them to a user.

[0239] Referring to FIGS. 1 to 9, the process of determining the situation analysis and response mode based on LLM integration can be performed by the processor (370) of the AI ​​assistant server (300).

[0240] If, as a result of identifying the situation in step S20 of Fig. 5, the situation type is identified as an emergency response guide, the processor (370) of the server (300) can perform additional situation analysis in conjunction with the LLM (301).

[0241] To this end, the processor (370) can configure an LLM request (LLM REQUEST) requesting situation analysis in conjunction with the LLM (301) and send it to the LLM (301).

[0242] In this regard, the processor (370) can first analyze the received command message (S801). Then, the processor (370) can select a situational prompt model (S802) based on at least one of the situation type, required data, and text data corresponding to the user's voice included in the command message.

[0243] In this specification, the prompt model may refer to a predefined query template model configured to enable the LLM (301) to determine the driver situation more specifically and accurately, i.e., a template model for generating LLM inputs. Additionally, the prompt model may include a query structure, a situation description template, or a response induction rule appropriate to the situational context. For example, if the command message is a message indicating the occurrence of an accident, the processor (370) may select an accident response-specific prompt model to analyze the accident situation.

[0244] In this regard, the prompt model can be implemented as an artificial intelligence model based on machine learning or deep learning. The prompt model can be pre-trained based on various emergency situation scenarios, stored in the memory of the server (300), and loaded and executed by the processor (370).

[0245] In an embodiment, the training process of the prompt model can be performed as follows. First, the training dataset may consist of 'input data' consisting of command messages in a shortened form that reflect the constraints of the satellite communication environment (e.g., situation type codes, numerical sensor data, fragmentary voice keywords), and 'label data (Ground Truth)' consisting of pairs of 'optimal natural language queries' written by experts so that the LLM can understand the situation.

[0246] Furthermore, the prompt model can be supervised to generate a sentence with high similarity to the optimal natural language query when the abbreviated command message is input. For example, if vector data such as "Type=0, Impact=High, HR=Unstable" is given as input, the weights of the prompt model can be optimized to expand this into a context-rich prompt such as, "A high-speed collision has been detected, and the driver's heart rate is unstable, making it an emergency situation requiring immediate medical attention. Please suggest priority actions for this."

[0247] Furthermore, the prompt model can be continuously tuned through reinforcement learning. That is, by setting the quality (accuracy, usefulness) of the answer derived when a generated prompt is input into the LLM as the reward function, it can be configured to autonomously learn a prompt structure that elicits better answers from the LLM.

[0248] Next, the processor (370) can create a prompt set and a situational query (S803) using a selected prompt model. In this regard, the prompt set may be defined as a set of questions in the form of multiple sentences or multiple queries generated based on the selected prompt model. For example, the prompt set may include a basic prompt such as "Please determine whether there is a vehicle accident and the circumstances of the accident." Additionally, the situational query may be an additional query designed to enable the LLM (301) to perform an accurate situational judgment by reflecting necessary input data such as command messages, sensor data, and MSD. For example, it may include "A value of airbag=1 was detected. Is the collision at a level requiring emergency measures?"

[0249] Subsequently, the processor (370) may construct an LLM REQUEST containing a prompt set and a contextual query and transmit it to the LLM (301) (S804). Additionally, the LLM REQUEST may include text corresponding to user utterances, and in some cases, the LLM REQUEST may include additional necessary data. For example, the LLM REQUEST may include text data such as "It's hard to move" or "My car hit something" corresponding to user utterances. In some cases, the LLM REQUEST may include additional necessary data such as location information, speed information, and whether an airbag has deployed.

[0250] In response to the transmission of an LLM REQUEST, the processor (370) may receive an LLM response (S805) from the LLM (301) that includes the results of a situation analysis. In this regard, the LLM RESPONSE may be a natural language-based response generated by the LLM (301) by analyzing the user utterance text, prompt set, situational query, and necessary data included in the LLM REQUEST. In an embodiment, the LLM RESPONSE may include an assessment of the accident situation, a determination of the need for emergency measures, and driver stabilization measures.

[0251] For example, as illustrated in FIG. 9, the LLM RESPONSE (901) may include a statement regarding the location and situation of the accident, such as "You're near the Las Vegas Strip area." It may also include a statement regarding emergency and stabilization measures, such as "Please stay calm and check for any injuries," and "Avoid moving if you're injured, and wait for help to arrive."

[0252] Meanwhile, the processor (370) can analyze and reconstruct (S806) the received LLM RESPONSE. As illustrated in FIG. 9, the original LLM RESPONSE (901) may generally contain a long natural language sentence of several hundred bytes (e.g., about 426 bytes). It may not be easy to extract key information for determining whether an emergency response is necessary from such a long response. Additionally, considering transmission efficiency, it may not be suitable for transmission as is in a satellite communication environment.

[0253] Accordingly, the processor (370) can remove long natural language responses, redundant information, or unnecessary descriptions included in the LLM RESPONSE, and extract only the core information necessary for performing an emergency response and convert it into a structured form. For example, as illustrated in FIG. 9, the contents of the LLM RESPONSE (901) can be analyzed to generate a restructured LLM response (902) (e.g., 360 bytes) composed of core instructions.

[0254] Next, the processor (370) can analyze the reconfigured LLM RESPONSE to determine whether emergency measures are required (S807). That is, the processor (370) can determine whether to switch to an emergency call mode based on the reconfigured LLM RESPONSE.

[0255] For example, as illustrated in FIG. 9, the reconfigured LLM RESPONSE (902) may include words or sentences that directly instruct to call external rescue services (e.g., 911), such as "Call emergency services (911) if you can. If not, signal for help." In this case, the processor (370) may determine that immediate emergency response is required. On the other hand, if the reconfigured LLM RESPONSE includes only instructions for emergency steps, such as "Avoid moving if you're injured" or "Wait for help to arrive," and does not include direct rescue service call phrases, the processor (370) may determine that immediate response is not required and that it is appropriate to provide the user with an emergency response manual guide.

[0256] In an embodiment, if it is determined that immediate emergency response is necessary ('Yes' in S807), the processor (370) may operate in an emergency call mode (S808). At this time, the processor (370) may activate a "conversational AI" (905 in FIG. 9) to perform server-based interactive TTS. Additionally, the processor (370) may control the execution of an Emergency Call to the Emergency Response Center (410) based on a reconfigured LLM RESPONSE. That is, by activating the server-based conversational AI, the Emergency Call may be connected with the Emergency Response Center (410), and a voice-based conversation may be performed with the Emergency Response Center based on the reconfigured LLM RESPONSE.

[0257] Here, "server-based interactive TTS" may refer to a method in which the generation of emergency response guidance sentences, speech synthesis, and conversation flow control are performed on the server (300). Additionally, it may be a real-time voice interaction method in which the server (300) directly performs voice output or call connection control. As a result, even in situations where it is difficult for the driver to make a call, it may be possible for the server (300) to immediately start voice guidance or perform a proxy connection with a rescue agency.

[0258] Additionally, the processor (370) can obtain feedback information including dispatch information and action results of emergency personnel. Subsequently, the processor (370) can generate a response message containing the feedback information and transmit it to the user terminal (100) via the satellite (200).

[0259] In another embodiment, if it is determined that immediate emergency response is not necessary ('No' in S807), the processor (370) may continue to operate in emergency response guide mode (S809).

[0260] At this time, the processor (370) can analyze the reconstructed LLM response to obtain multiple step-by-step guide texts to be provided to the user sequentially. Subsequently, the processor (370) can generate a response message containing multiple step-by-step guide texts. Here, the response message may further include an 'on-device interactive TTS command' that instructs the on-device artificial intelligence model of the user terminal (100) to activate the text-based guide and provide voice guidance.

[0261] In an embodiment, the processor (370) can summarize the reconstructed LLM RESPONSE in consideration of the data transmission efficiency of the satellite communication. Through this, the processor (370) can generate a summarized response that abbreviates the original text (see 903 in FIG. 9, e.g., about 181 bytes).

[0262] Additionally, the processor (370) can compress the summary response (903) back into a binary form to generate a compressed data set (see 904 in FIG. 9, e.g., about 163 bytes) and transmit it to a user terminal (100) via a satellite (200).

[0263] In response to this, the processor (170) of the user terminal (100) can restore the received compressed data set (904) and identify the included TTS command. If the TTS command is identified, the user terminal (100) can immediately operate as an on-device interactive TTS. Accordingly, the processor (170) can perform the operation of converting the restored step-by-step guide text into a synthesized speech signal and sequentially guiding the driver through the speaker.

[0264] Hereinafter, a method for data reconstruction, summarization, and compression suitable for an NTN-based satellite communication environment, performed in a user terminal and an AI assistant server according to the present disclosure, is described. In this regard, FIG. 10 illustrates an embodiment of data reconstruction, summarization, and compression performed in a user terminal and an AI assistant server according to the present disclosure.

[0265] Referring to FIGS. 1 to 10, data reconstruction, summarization, and compression methods can be performed by the processor (170) of the user terminal (100) and the processor (370) of the server (300), respectively.

[0266] First, the processor (170) of the user terminal (100) can receive voice data (S1001) including the user's voice. In an embodiment, the voice data may consist of data approximately 11 seconds in length (e.g., approximately 11,000 bytes).

[0267] In this case, the voice data may include not only the user's voice but also ambient sounds such as vehicle driving noise, external crash sounds, tire friction sounds, or siren sounds. Additionally, the included user voice may include unstructured sentences, interjections, and incomplete sentences uttered in a driving environment or immediately after an accident.

[0268] Accordingly, the processor (170) according to the present disclosure may perform correction to remove background noise included in the input voice data or to increase the clarity of the voice signal. This may be intended to prevent the voice recognition rate from being degraded due to wind noise, engine noise, or popping sounds that may occur during driving or collision. Subsequently, the processor (170) may convert the corrected voice data into text data (S1002) using an STT model. The converted text data is, for example, "text: I've got a car accident and I think it's hard to drive right now. I think I was passed out right before the crash. Maybe my car seems to have hit a tree or something like that."

[0269] Next, the processor (170) can combine the converted text with the entire data of the necessary data (e.g., MSD) to generate a message in the form of "text + MSD data" (S1003). That is, the processor (170) can perform the step of adding the converted text data to a command message. At this time, the entire MSD data may correspond to 702 in FIG. 7 and may include not only basic information such as location coordinates (lat, lon) and vehicle speed, but also additional information such as the last stop point, previous driving history, vehicle interior temperature, and tire pressure. In this case, the "text + MSD data" may be approximately 366 bytes in size.

[0270] In an embodiment, the processor (170) may select only some items necessary for accident determination (e.g., whether the airbag deployed, location coordinates, speed information, etc.) from the entire MSD data and generate a message in the form of "text + MSD selected data" together with the converted text (S1004). At this time, "text + MSD selected data" may correspond to 703 in FIG. 7.

[0271] Additionally, although not described, the processor (170) may generate a message in the form of "text + MSD data" by summarizing it to reduce the message size. For example, the summarization may be performed by removing unnecessary conjunctions, repetitive expressions, and contextually redundant descriptions from the text to leave only the key sentences.

[0272] Next, the processor (170) can compress "text + MSD selected data" (S1005) based on the summarized text and selected MSD data.

[0273] In this process, the processor (170) may apply different summarization levels or compression rates to command messages based on at least one of the situation type and satellite communication link quality.

[0274] In an embodiment, the processor (170) may apply a first compression rate that minimizes data loss and a high-intensity summary focused on key keywords when there is an emergency call or emergency action guide situation or when communication quality is low. On the other hand, when there is a phone reservation or information search situation and communication quality is good, the processor (170) may optimize data transmission by applying a second compression rate that increases transmission efficiency and a low-intensity summary that maintains context. Through this process, compressed data can be generated at a level of approximately 217 bytes, and the processor (170) can transmit this to an AI assistant server (300) via satellite communication.

[0275] Meanwhile, the processor (370) of the AI ​​assistant server (300) can receive a compressed message from the terminal (100), decompress it, and perform subsequent processing according to the situation type. For example, in the case of an accident or emergency situation classification such as TYPE 1 or TYPE 2, the processor (370) can prioritize performing an LLM-based situation judgment procedure.

[0276] In this regard, the processor (370) can send an LLM REQUEST to the LLM (301) and receive a corresponding LLM RESPONSE (S1006). In an embodiment, the LLM RESPONSE may correspond to 901 of FIG. 9 and may consist of natural language text (e.g., about 426 bytes) containing an accident situation assessment and an emergency guidance sentence.

[0277] Next, the processor (370) can reconstruct the LLM RESPONSE (S1007). For example, redundant sentences or unnecessary descriptions can be removed from a long LLM RESPONSE (e.g., 426 bytes), and only the key instructions necessary for actual emergency response judgment can be selected to generate a reconstructed response (902 in FIG. 9) of about 360 bytes in size.

[0278] The processor (370) can then summarize the reconstructed LLM RESPONSE (S1008). In an embodiment, the processor (370) can convert the emergency guidance sentence into a number-based step instruction form or organize only the key action instructions to generate a concise summary response (e.g., about 181 bytes).

[0279] Finally, the processor (370) can compress the summarized LLM RESPONSE again (S1009). In an embodiment, the compressed summarized response may correspond to 904 in FIG. 9 and may be generated with a size of approximately 163 bytes. This can ensure transmission efficiency and minimize delay in an NTN-based satellite communication environment. The generated summarized and compressed message may be transmitted to a user terminal (100) via a satellite (200).

[0280] Finally, the processor (170) of the user terminal (100) can decompress the compressed summary response received from the server (300) and play the response content (S1010) using an on-device based interactive TTS.

[0281] Accordingly, the user terminal (100) and the AI ​​assistant server (300) according to the present disclosure may apply an intelligent data processing method that selects only the necessary data during the message generation process or summarizes and compresses text according to the situation, taking into account the transmission efficiency of the NTN-based satellite communication environment.

[0282] Hereinafter, a method of operation in an emergency call mode in a satellite communication system according to the present disclosure will be described. In this regard, FIG. 11 is a flowchart of operation in an emergency call mode performed by an AI assistant server according to the present disclosure. In addition, FIG. 12 illustrates an example of a server-based interactive TTS conversation performed in an emergency call mode according to the present disclosure.

[0283] Referring to FIGS. 1 to 12, operation in emergency call mode in a satellite communication system can be implemented through mutual interoperability between a user terminal (100), a satellite (200), an AI assistant server (300), and an emergency response center (410).

[0284] First, the user terminal (100) can transmit a compressed command message generated based on vehicle information and driver information to the server (300) via the satellite (200).

[0285] In this regard, the user terminal (100) can generate a trigger signal when it detects an impact from a sensor mounted inside or outside the vehicle. Specifically, the terminal (100) may determine that the impact is valid only when an impact exceeding a pre-learned hysteresis range is detected, in order to exclude noise or minor vibrations in the sensor values.

[0286] At this time, the terminal (100) can perform a probe (User consciousness probe) procedure to check for a response from the driver. If the driver's voice or touch response is confirmed, the terminal (100) may ignore the corresponding trigger, but if there is no response within a preset time (No response), it may determine this as an actual emergency situation.

[0287] Accordingly, the user terminal (100) can classify the driver status as an 'emergency call (Type 0)' state and generate a command message containing corresponding text data. In an embodiment, the text data may be an automatically generated phrase summarizing the situation, such as "[Automated Alert] Crash detected. User unresponsive."

[0288] Additionally, the user terminal (100) can add situation type, trigger subject information (e.g., Sensor), and location information to the command message, compress it, and transmit it to the server (300).

[0289] Meanwhile, the AI ​​assistant server (300) can receive a compressed data set transmitted from the terminal (100) and decompress it. Then, the processor (370) of the server (300) can analyze the text data and necessary data included in the command message to perform situation recognition and identify the situation type (S20).

[0290] In an embodiment, if the processor (370) determines that the analysis result is a trigger by the sensor, it can process it with the highest priority and determine that it is an emergency situation requiring an emergency rescue request (911 Call).

[0291] Next, the processor (370) can acquire situation information (S30) according to the identified situation type. In an embodiment, if the identified situation type is TYPE 0 (emergency call), the processor (370) can combine the recognized specific accident situation and MSD data (location, whether airbags are deployed, etc.) to generate a rescue request sentence to be delivered to rescue personnel. In the present disclosure, the generated rescue request sentence can be acquired as rescue request information.

[0292] In this case, the processor (370) may decide to immediately enter emergency call mode (S40) and connect a 911 call.

[0293] On the other hand, in the case of TYPE 1 or 2 (emergency action guide), the processor (370) can perform LLM-linked situation analysis (S51) for precise situational judgment. At this time, the processor (370) can re-evaluate the urgency of the situation by analyzing the LLM REQUEST and LLM RESPONSE transmitted and received with the LLM (301). If it is determined that immediate rescue is required based on the reconstructed RESPONSE, the processor (370) can obtain rescue request information from the corresponding RESPONSE and enter an emergency call mode (S40).

[0294] Afterwards, the processor (370) can operate in an emergency call mode (S40) based on the acquired rescue request information.

[0295] When the emergency call mode (S40) is executed, the processor (370) of the AI ​​assistant server (300) can activate (S41) the "conversational AI" module inside the server. That is, the emergency call mode (S40) may refer to an AI-based emergency response system mode. In this regard, the conversational AI can perform functions such as converting text data into speech (TTS), converting input speech into text (STT), and managing the flow of natural language conversation.

[0296] Next, the processor (370) can make an emergency call (S42) to the emergency response center (410) based on the acquired information (location, vehicle information, patient status, etc.).

[0297] Once a connection is established, the processor (370) can operate in server-based interactive TTS mode to perform a conversation.

[0298] In the embodiment, when a query asking about the patient's condition or the situation at the scene is received (S43) from an agent at the emergency response center (410), the processor (370) can recognize this in real time. Subsequently, the processor (370) can generate an appropriate response based on sensor data or LLM analysis results within a pre-secured command message, convert it into synthesized speech, and transmit the response to the agent (S44). Through this, even in a situation where the driver loses consciousness or is unable to make a call, the AI ​​assistant server (300) can exchange detailed situational explanations with the rescue worker on behalf of the driver.

[0299] For example, referring to FIG. 12, the processor (370) can analyze a command message or a reconfigured LLM response in step S30 or step S51 of FIG. 11 to obtain rescue request information including driver information and vehicle information.

[0300] In this regard, the processor (370) can obtain patient information and vehicle information indicating the driver's condition and the circumstances of the accident as shown in FIG. 12.

[0301] Additionally, the processor (370) can enable server-based conversational AI to perform voice conversations based on acquired information with an emergency response center agent.

[0302] At this time, when a query such as “How is the patient’s condition?” is received from an agent of the emergency response center (410), the processor (370) can generate and transmit a natural language response such as “The patient is a 55-year-old male. He is currently unconscious and complaining of pain in his head and chest.” based on the driver information already acquired.

[0303] Then, when an additional inquiry is received from the agent saying, "Please tell me the circumstances of the accident and vehicle information," the processor (370) can convey detailed information in voice based on the vehicle information, such as, "It was a single-vehicle accident involving a Hyundai Sonata vehicle, the airbag deployed, and there was no rollover. It collided at a speed of 60 km / h."

[0304] Finally, when an agent asks for a location to dispatch an ambulance, the processor (370) can support rapid dispatch by providing the agent with accurate GPS-based latitude / longitude information or converted address information.

[0305] Finally, the processor (370) may generate feedback information (S45) as a result of the emergency call and construct a response message containing the information. In this regard, the feedback information may include dispatch information of emergency personnel and the results of the action. In an embodiment, the processor (370) may summarize and compress the response message considering the satellite communication bandwidth and then transmit it to the user terminal (100) via the satellite (200).

[0306] Hereinafter, a method of operation in an emergency response guide mode in a satellite communication system according to the present disclosure will be described. In this regard, FIG. 13 is a flowchart of the operation in an emergency response guide mode performed by an AI assistant server according to the present disclosure. FIG. 14 also illustrates an example of an on-device-based interactive TTS conversation performed in an emergency response guide mode according to the present disclosure.

[0307] Referring to FIGS. 1 to 14, operation in the emergency guide mode in a satellite communication system can be implemented through mutual interaction between a user terminal (100), a satellite (200), and an AI assistant server (300).

[0308] First, the user terminal (100) can transmit a compressed command message generated based on vehicle information and driver information to the server (300) via the satellite (200).

[0309] In this regard, the user terminal (100) can receive the user's voice through a microphone. For example, if the user speaks vaguely, such as "There was an accident... uh... I don't know what to do," the terminal (100) can detect this and convert the user's voice into text data by performing an STT function.

[0310] At this time, the user terminal (100) can identify the user's core intent from the converted text data using an on-device AI model and correct the text based on this. For example, the AI ​​model can analyze the rambling speech content and refine it into a sentence with a clear meaning, such as 'reporting an accident and requesting an emergency response guide.'

[0311] Next, the user terminal (100) can extract necessary data (e.g., vehicle location, speed, etc.) associated with the corrected text data, include it in a command message, compress it, and transmit it to the server (300).

[0312] Meanwhile, the AI ​​assistant server (300) can receive a compressed data set transmitted from the terminal (100) and decompress it. Then, the processor (370) of the server (300) can analyze the text data and necessary data included in the command message to perform situation recognition and identify the situation type (S20).

[0313] Next, the processor (370) can obtain situation information (S30) according to the identified situation type. In the embodiment, if the identified situation type is an 'emergency action guide (TYPE 1 or 2)', the processor can obtain basic emergency action information (S30) by analyzing the command message.

[0314] Subsequently, the processor (370) can perform a situation analysis (S51) in conjunction with the LLM (301) to provide accurate accident situation analysis and emergency response guides. Specifically, the processor (370) can transmit an LLM REQUEST containing situation information to the LLM (301) and receive a corresponding LLM RESPONSE. The processor (370) can analyze the received LLM RESPONSE and extract and reconstruct only the key step-by-step instructions required for emergency response.

[0315] Meanwhile, even when triggered by user voice and transmitted with the situation type set to 'Emergency Measure Guide', the processor (370) can re-evaluate the urgency of the situation based on the reconfigured LLM RESPONSE. At this time, if it is determined that immediate professional action is required (decision that emergency measures are needed), the processor (370) can automatically switch from the Emergency Measure Guide mode to the Emergency Call mode (S40) and operate.

[0316] Next, the processor (370) can generate a response message (S52) based on the reconstructed LLM RESPONSE and transmit it to a user terminal (100) via a satellite (200). At this time, the processor (370) can summarize and compress the response message to maximize transmission efficiency before transmission. Additionally, the response message may further include an on-device interactive TTS command that instructs an on-device artificial intelligence model to activate a plurality of step-by-step guide texts to be voiced. In an embodiment, the command may be implemented by including a field within the response message that sets the 'Interactive TTS mode (True / False)'. For example, if an emergency response guide is required, the server may set the corresponding field to 'True' and transmit it.

[0317] The user terminal (100) can receive a response message from the server (300), decompress it, analyze it, and configure an emergency response manual (S53). At this time, if the Interactive TTS mode in the response message is set to 'True', the processor (170) of the user terminal (100) can operate in an on-device based interactive TTS mode.

[0318] Here, "on-device based interactive TTS mode" may refer to a method in which the terminal (100) independently converts text into speech and detects the user's response to proceed with guidance for the next step without real-time communication with a server. This is intended to provide uninterrupted guidance without being affected by satellite communication latency.

[0319] Specifically, the processor (170) can convert a first guide text among a plurality of step-by-step guide texts included in the response message into speech and output it (S54). Then, the processor (170) can detect a user's response input to the output speech (S55). At this time, when a response input is detected, the processor (170) can sequentially perform the operation of converting a second guide text into speech and outputting it. Here, the response input may include the user's voice confirmation ("Yes," "Next," etc.) or a terminal button input.

[0320] In this regard, referring to FIG. 14, the server (300) can analyze the situation of a 55-year-old male driver complaining of unconsciousness and chest pain and transmit a compressed text set containing step-by-step instructions to the terminal (100), such as "1. Take deep breaths calmly. 2. Check the injured area. 3. Do not move if there is an injury. 4. Wait for rescue."

[0321] The user terminal (100) receives this and decompresses the response, and then outputs the first guide text, "Stay calm. Take deep breaths." as voice. Afterward, when a response input is detected, such as the user taking deep breaths or pressing a confirmation button, the terminal (100) can immediately provide the second guide text, "Check for injuries." as voice.

[0322] Through this process, even in situations where the communication connection is unstable, the user can be guided step-by-step through safety rules such as "Don't move if you're injured" and "Wait for help to arrive" under the guidance of the terminal (100).

[0323] Hereinafter, a method of operation in a phone reservation mode in a satellite communication system according to the present disclosure will be described. In this regard, FIG. 15 is a flowchart of the operation in a phone reservation mode performed by an AI assistant server according to the present disclosure. FIG. 16 also illustrates an example of a server-based interactive TTS conversation performed in a phone reservation mode according to the present disclosure.

[0324] Referring to FIGS. 1 to 16, operation in phone reservation mode in a satellite communication system can be implemented through mutual interaction between a user terminal (100), a satellite (200), an AI assistant server (300), and a service provider (420).

[0325] First, the user terminal (100) can transmit a compressed command message generated based on vehicle information and driver information to the server (300) via the satellite (200).

[0326] In this regard, the user terminal (100) may receive a trigger signal including user voice or UI input. For example, the user may say, "Please make a reservation for 3 people at a nearby Italian restaurant in one hour," or select a reservation menu through the touchscreen of the terminal.

[0327] At this time, the user terminal (100) can extract reservation request information (Req Info) from the input and add it as necessary data for a command message. Here, the reservation request information may include the name of the person making the reservation (e.g., Justin), the number of people (e.g., 3 people), the desired time and location type, etc. Then, the user terminal (100) can compress the generated command message and transmit it to the server (300).

[0328] Meanwhile, the AI ​​assistant server (300) can receive a compressed data set transmitted from the terminal (100) and decompress it. Then, the processor (370) of the server (300) can analyze the text data and necessary data included in the command message to perform situation recognition and identify the situation type (S20).

[0329] Next, the processor (370) can obtain situation information (S30) according to the identified situation type. In an embodiment, if the identified situation type is 'phone reservation (TYPE 3)', the processor can obtain phone reservation information (S30) by analyzing a command message. For example, the reservation information may include the reservation type (restaurant, hotel, etc.), the name of the person making the reservation, the number of people, the type of preferred dish, distance-based conditions (e.g., possible to arrive within 1 hour), etc.

[0330] Meanwhile, in this embodiment, the service target of TYPE 3 (telephone reservation) is not limited to restaurants. For example, the service target may include not only general service providers such as hotels, airlines, and car rental companies, but also various targets requiring reservation or reception, such as disaster centers, hospital reception desks, or public institutions depending on the situation.

[0331] Afterwards, the processor (370) can operate in phone reservation mode (S60) based on the acquired reservation information.

[0332] First, the processor (370) can activate the "conversational AI" module inside the server. Then, the processor (370) can activate the server-based conversational AI to connect a call (S61) with the reservation destination (service provider, 420) included in the reservation information. In an embodiment, the processor (370) can search for a restaurant that meets the conditions by considering the user's current location and travel path, and connect a call to the restaurant.

[0333] Once a connection is established, the processor (370) can operate in a server-based interactive TTS mode to perform real-time voice conversation with the reservation target (e.g., restaurant staff).

[0334] In this regard, referring to FIG. 16, the processor (370) can analyze a received command message to obtain reservation information necessary for executing a reservation. In an embodiment, the reservation information may include at least one of a reservation type (e.g., restaurant, hotel), the name of the person making the reservation (e.g., Justin), the number of people (e.g., 3 people), a preferred menu (e.g., Italian, steak), and a location-based travel time (e.g., arrival within 1 hour).

[0335] At this time, the processor (370) can recognize the other party's speech and generate a response appropriate to the situation based on the acquired reservation information. Specifically, when an inquiry "How can I help you?" is received from a restaurant employee, the processor (370) can generate a natural language query in real time, such as "Hello. We are a party of 3 people including Justin. We would like to inquire if we can have a meal between 7:00 PM and 8:00 PM," and transmit it as synthesized speech.

[0336] Then, when an additional information query is received from an employee, such as "7:30 is possible. How many people are in your party?" (S62), the processor (370) can generate and transmit an information-based response (RESPONSE) such as "7:30 is fine. My family has 3 members" based on the previously obtained information (S63).

[0337] Finally, when the employee asks for the name of the person making the reservation, the processor (370) can confirm the reservation by answering, "It is Justin."

[0338] Finally, the processor (370) can analyze the call results to generate feedback information including whether the reservation was successful and details of the reservation. Then, the processor (370) can generate a response message (S64) containing the feedback information and transmit it to the user terminal (100) via the satellite (200). The user terminal (100) can display the confirmed reservation location (address), time, and number of people information on the screen or provide voice guidance through the received response message. Additionally, if necessary, the reservation location can be automatically set as the destination of the navigation.

[0339] Hereinafter, a method of operation in an information search mode in a satellite communication system according to the present disclosure will be described. In this regard, FIG. 17 is a flowchart of the operation in an information search mode performed by an AI assistant server according to the present disclosure. FIG. 18 also illustrates an embodiment of a server-based search and search result display method performed in an information search mode according to the present disclosure.

[0340] Referring to FIGS. 1 to 18, operation in the emergency response guide mode of a satellite communication system can be implemented through mutual interaction between a user terminal (100), a satellite (200), an AI assistant server (300), and an external search engine (430).

[0341] First, the user terminal (100) can transmit a compressed command message generated based on vehicle information and driver information to the server (300) via the satellite (200).

[0342] The processor (370) of the server (300) can analyze the received message to identify the situation type (S20). Subsequently, the processor (370) can obtain situation information (S30) according to the identified situation type.

[0343] In an embodiment, when the identified situation type is TYPE 4 (information search), the processor (370) can obtain search information by analyzing a command message. The search information may include search keywords desired by the user (e.g., gas station, repair shop, restaurant, etc.), search conditions (within a radius of 10 km, rating of 4.0 or higher, etc.), and current location information.

[0344] Afterward, the processor (370) can operate in an information search mode (S70) based on the acquired search information. Specifically, the processor (370) can generate a search query based on the acquired search information and perform an information search (S71) in conjunction with an external search engine (430). That is, the processor (370) can generate a query such as "gas stations within 10km of the current location (latitude / longitude)" and send it to Google or Naver Map APIs, etc.

[0345] When the search is completed, the processor (370) can receive the search result (RESPONSE) (S72) from the search engine (430). At this time, the original search result may contain dozens of lists and unnecessary metadata, making it unsuitable for transmission via satellite communication. Therefore, the processor (370) can perform text summarization by extracting key information that matches the user's intent from the result data of the information search. For example, the processor (370) can lighten the data by selecting only the top two best places in order of distance or rating among the search results and keeping only essential information such as business name, location, and phone number.

[0346] Referring to FIG. 18 in this regard, when a user requests a "gas station search," the server (300) can obtain information such as "Current lat / long (current location)," "Radius: 10km (radius)," and "Key: gas station (keyword)" from a JSON-formatted message. Then, the server (300) can perform an internet search to obtain information on multiple gas stations, and then filter this information to extract only the information on the two closest gas stations (Gas station 1, Gas station 2).

[0347] Finally, the processor (370) can generate a response message (feedback information) containing summarized text data (S73) and transmit it to a user terminal (100) via a satellite (200).

[0348] As illustrated in FIG. 18, the server (300) can finally send a structured and compressed JSON response message (Compress textset) to the terminal (100), such as "cnt: 2 (count)", "station: [{name: 76, lat:..., lon:...}, {name: 7-Eleven...}]".

[0349] In response to this, the user terminal (100) can parse the received message and display the locations of the searched gas stations as icons on a map. Additionally, the user terminal (100) can provide a function to display detailed information about a specific location as a popup or to immediately set the location as a navigation destination to provide route guidance based on user input.

[0350] Hereinafter, the configuration and extraction method of necessary data for each situation type according to the present disclosure will be described. In this regard, FIG. 19 illustrates an example of a Minimum Set of Data (MSD) and a data set that are selectively configured according to the situation type according to the present disclosure.

[0351] Referring to FIGS. 1 to 19, the processor (170) of the user terminal (100) can extract data optimized for a classified situation type and configure a command message.

[0352] In the embodiments, the required data may be configured to include different items depending on the situation type.

[0353] For example, if the situation type is 'emergency call (TYPE 0)' or 'emergency action guide (TYPE 1, 2)', the processor (170) can extract information essential for accident handling and rescue as necessary data. In this case, the necessary data may include at least one of whether a collision occurred (crash), vehicle identification information (vin), location information (lat, lon, alt), vehicle speed (speed), and occupant information (occu, ABO, sex, age). On the other hand, if the situation type is 'phone reservation (TYPE 3)' or 'information search (TYPE 4)', the processor (170) can extract information necessary for service performance as necessary data. In this case, the necessary data may include at least one of location information, user information (reservation name, etc.), vehicle information, and service request information (reservation time, search keywords, etc.).

[0354] In this regard, referring to FIG. 19, the processor (170) can filter data based on a situational essential item table. For example, in the case of a 'traffic accident' situation, crash, vehicle identification number (vin), location (lat, lon), and speed are key elements of accident analysis, so they are designated as essential items ('O') and must be included in the message. In contrast, in the case of a 'distress' situation, vehicle speed information is of relatively low importance and may be omitted or treated as an optional item ('M'). Instead, sex, age, and blood type (ABO) information may be included as essentials for identifying the person being rescued and for medical treatment.

[0355] Additionally, the necessary data according to the present disclosure may further include information on the driver's psychological state. As illustrated in FIG. 19, the processor (170) may include a psychological value (e.g., value 3 - fear / panic state) calculated through voice analysis in the data. Through this, the rescue center can determine an appropriate level of response by comprehensively understanding not only simple physical collision information but also the driver's emotional urgency.

[0356] In this way, the processor (170) does not unconditionally transmit all collected data, but selects and transmits only the necessary data through situational filtering logic defined as in FIG. 19, thereby reducing satellite communication costs and efficiently increasing the success rate of data transmission.

[0357] Meanwhile, a satellite communication method performed by a user terminal and an AI assistant server according to another aspect of the present disclosure is not limited thereto and can be applied in various ways by combining with the embodiments of FIGS. 1 to 19.

[0358] The satellite communication system based on vehicle and driver information and the method of operation thereof according to the present specification have been described above. The technical effects of the satellite communication system and the method of operation thereof according to the present specification may be summarized as follows, but are not limited thereto.

[0359] According to the present disclosure, voice data is converted into text (STT) using on-device AI within a terminal in a non-terrestrial network (NTN) environment, and only essential data is selectively extracted, compressed, and transmitted according to the type of situation, thereby effectively solving the problems of high cost and bandwidth constraints of conventional satellite communication. Accordingly, communication charges can be reduced by minimizing the size of data packets, and the success rate and speed of data transmission can be dramatically improved even in network congestion situations.

[0360] According to the present disclosure, unlike conventional methods that simply transmit text or sensor values, the occurrence of an accident and the level of urgency can be precisely determined by comprehensively analyzing the driver's voice characteristics (tone, intensity, etc.), psychological state, and location information through an on-device AI model. This reduces false alarms and accurately recognizes actual emergency situations, thereby establishing a foundation for rescue centers to perform optimal initial response within the golden hour.

[0361] According to the present disclosure, even in a text-based satellite communication environment, server-based conversational AI and a Large Language Model (LLM) can be integrated to provide users with situation-specific customized first aid guides or to perform voice calls with emergency centers on their behalf. Accordingly, seamless two-way communication is possible even in situations where it is difficult for the driver to make a call directly, and through automatic translation and information search functions, the user experience (UX) can be significantly expanded to areas such as overseas travel and general convenience services.

[0362] The foregoing disclosure may be implemented as computer-readable code on a medium on which a program is recorded. A computer-readable medium includes all types of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable media include Hard Disk Drives (HDDs), Solid State Disks (SSDs), Silicon Disk Drives (SDDs), ROMs, RAMs, CD-ROMs, magnetic tapes, floppy disks, optical data storage devices, etc., which may constitute non-transitory computer-readable storage media. It may also include implementations in the form of carrier waves (e.g., transmission over the Internet), but the computer-readable storage media described in the claims of this disclosure should be interpreted to mean the non-transitory storage media. Furthermore, the computer may include a control unit of a terminal. Accordingly, the above detailed description should not be interpreted restrictively in all respects and should be considered exemplary. The scope of the invention should be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of the invention are included within the scope of the invention.

Claims

1. In a satellite communication method, the above method is, A step of acquiring at least one of vehicle information and driver information based on at least one of collected voice data, sensor data and user input data; A step of classifying the driver situation into at least one of a plurality of context types based on the above-mentioned acquired information; A step of generating a command message based on the above situation type and the above acquired information; A step of performing data lightweighting so that the size of the above command message is reduced; and A satellite communication method comprising the step of transmitting the above-mentioned lightweight command message to a satellite.

2. In Paragraph 1, A step of extracting necessary data associated with the above situation type; and A satellite communication method further comprising the step of adding the necessary data to the command message.

3. In Paragraph 1, The method further includes a step of determining whether there is an emergency situation or the urgency level based on the above-mentioned acquired information; A satellite communication method in which the above situation type is classified based on whether the above emergency situation is or the above emergency severity.

4. In Paragraph 3, The above situation type is, A satellite communication method comprising at least one of an emergency call, an emergency action guide, a phone reservation, and an information retrieval.

5. In Paragraph 3, The step of determining the above emergency urgency is, A step of calculating a driver psychological map by analyzing the acoustic characteristics of the above voice data; and A satellite communication method comprising the step of determining the emergency level by reflecting the above driver psychology map.

6. In Paragraph 3, The step of determining the above emergency urgency is, A step of obtaining location information from the sensor data and calculating a risk weight corresponding to the location information; and A satellite communication method comprising the step of determining the emergency urgency level by reflecting the above risk weighting.

7. In Paragraph 1, A step of converting the speech data into text data using a speech-to-text (STT) model; and A satellite communication method further comprising the step of adding the text data to the command message.

8. In Paragraph 7, A step of determining whether the text data and the sensor data are different from each other; and A satellite communication method further comprising the step of performing content correction on the text data by applying priority to the sensor data when the text data and the sensor data are different from each other.

9. In Paragraph 1, A step of receiving a response message for the transmitted command message from a server via the satellite; and A satellite communication method further comprising the step of performing a response action associated with the received response message.

10. In Paragraph 2, The above required data is, If the above situation type is an emergency call or emergency action guide, it includes at least one of whether a collision occurred, vehicle identification information, location information, and occupant information, and A satellite communication method comprising at least one of location information, user information, vehicle information, and service request information, wherein the above situation type is telephone reservation or information search.

11. In Paragraph 1, The step of performing the above data lightweighting includes the step of performing summarization or compression of the above command message, and A satellite communication method in which the summary level or compression rate for the above command message is applied differently based on at least one of the situation type and satellite communication link quality.

12. In Paragraph 11, If the above situation type is an emergency call or emergency action guide, or if the above satellite communication link quality is below a threshold standard, the command message is compressed at a first compression rate, and A satellite communication method that compresses a command message with a second compression rate lower than the first compression rate when the above situation type is telephone reservation or information search and the above satellite communication link quality is above a threshold standard.

13. In Paragraph 9, A server that receives the above command message, decompresses the received command message; A step of analyzing the command message to identify the situation type included in the command message; and A satellite communication method comprising the step of summarizing or compressing a response message containing the result of performing a situation-specific action according to the identified situation type and transmitting it to a user terminal via the satellite.

14. In Paragraph 13, If the above identified situation type is an emergency call, A step of obtaining rescue request information by analyzing the above command message; A step of activating server-based interactive AI to connect the emergency response center with the emergency call; A step of performing a voice-based conversation (Call Conversation) with the emergency response center based on the above-mentioned structural request information; and A satellite communication method further comprising the step of generating the response message to include feedback information including dispatch information or action results of emergency personnel.

15. In Paragraph 13, If the above identified situation type is an emergency first aid guide, A step of configuring an LLM request requesting situation analysis and transmitting it to the LLM; A step of receiving an LLM response (LLM RESPONSE) including a situation analysis result from the LLM in response to the above transmission; and A satellite communication method further comprising the step of analyzing and reconstructing the above LLM response.

16. In Paragraph 13, If the above identified situation type is a phone reservation, A step of obtaining reservation information by analyzing the above command message; A step of activating a server-based conversational AI to connect a call with the reservation destination included in the reservation information; A step of performing a voice-based conversation with the reservation target based on the above reservation information; and A satellite communication method further comprising the step of generating the response message to include feedback information including whether the reservation was successful or details of the reservation.

17. In Paragraph 13, If the above identified situation type is information retrieval, A step of obtaining search information by analyzing the above command message; A step of generating a search query based on the above search information and performing information retrieval by linking with an external search engine; A step of extracting key information corresponding to the user's intent from the result data of the above information search and performing a text summary; and A satellite communication method further comprising the step of generating the response message to include the summarized text data above.

18. In Paragraph 15, The step of configuring an LLM request that requests the above-mentioned situation analysis is, A step of selecting a situational prompt model based on at least one of the situation type, required data, and text data included in the above command message; A step of creating a prompt set or contextual query using the above-mentioned selected prompt model; and A satellite communication method comprising the step of configuring an LLM request including the above prompt set or the above situational query.

19. In Paragraph 15, Based on the above-mentioned reconstructed LLM response, a step of obtaining a plurality of step-by-step guide texts; and The method further includes the step of generating a response message containing the above-mentioned plurality of step-by-step guide texts; The above response message is, A satellite communication method further comprising a TTS Command that activates an on-device artificial intelligence model to instruct the plurality of step-by-step guide texts to be spoken.

20. In Paragraph 15, The method further includes a step of re-determining whether the driver situation is an emergency call situation based on the reconstructed LLM response, and if the re-determination result determines that it is an emergency call situation, A step of activating server-based conversational AI to connect emergency response centers with emergency calls; and A step of performing a voice-based conversation with the emergency response center based on the above-mentioned reconstructed LLM response; and A satellite communication method further comprising the step of generating the response message to include feedback information including dispatch information or action results of emergency personnel.

21. In Paragraph 9, In the step of performing the above response operation, if the response message includes a TTS command, Among the plurality of step-by-step guide texts included in the above response message, the first guide text is converted into speech and output, and Detects user response input to the above-mentioned output voice, and A satellite communication method comprising: a step of sequentially performing an operation of converting a second guide text into speech and outputting it when the above response input is detected.

22. In Paragraph 9, The step of performing the above response operation is, A step of identifying a country code by obtaining at least one of location information and phone number information; A step of determining a target language based on the identified country code; and A satellite communication method comprising the step of automatically translating the received response message into the determined target language and outputting it.

23. In a vehicle performing satellite communication, A communication module configured to transmit and receive data to and from a server by linking with a non-terrestrial network (NTN) via a satellite; and It includes a processor operably connected to the above communication module, and the processor, Based on at least one of collected voice data, sensor data and user input data, at least one of vehicle information and driver information is obtained, and Based on the above-mentioned acquired information, the driver situation is classified into at least one of a plurality of context types, and Based on the above situation type and the above acquired information, a command message is generated, and Data lightweighting is performed to reduce the size of the above command message, and A vehicle that transmits the above-mentioned lightweight command message to a satellite.

24. In a computer-readable storage medium, The computer-readable storage medium stores at least one computer program comprising instructions that, when executed by at least one processor, cause the at least one processor to perform operations, and said operations are Based on at least one of collected voice data, sensor data and user input data, at least one of vehicle information and driver information is obtained, and Based on the above-mentioned acquired information, the driver situation is classified into at least one of a plurality of context types, and Based on the above situation type and the above acquired information, a command message is generated, and Data lightweighting is performed to reduce the size of the above command message, and A computer-readable storage medium for transmitting the above-mentioned lightweight command message to a satellite.

Images