Frequency spectrum prediction and frequency hopping method and system in sanitation equipment SoC ad hoc network

Abstract

The invention relates to a frequency spectrum prediction and frequency hopping method and system in an SoC ad hoc network of environmental sanitation equipment, and the method comprises the following steps: frequency spectrum monitoring and data collection, HMM modeling and training, frequency spectrum prediction and frequency hopping decision making, and a dynamic adjustment mechanism. It is ensured that an efficient data transmission link can be maintained even in a complex and changeable electromagnetic environment.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent environmental sanitation, and particularly relates to a spectrum prediction and frequency hopping method and system in the SoC self-organizing network of environmental sanitation equipment. Background Art

[0002] In the modern urban environment, as a node of the environmental sanitation system, the garbage collection station (GCS) exchanges working status information in real time through a wireless network, such as the degree of garbage fullness and the operation progress. This information is crucial for optimizing the efficiency of environmental sanitation services. With the development of smart cities, more and more sensors are deployed on GCSs, resulting in a significant increase in data volume, which poses higher requirements for wireless communication.

[0003] The SoC self-organizing network is a technology that connects multiple SoC modules with computing and communication capabilities to form a temporary network, which allows data transmission between GCSs without relying on fixed infrastructure. In a typical urban environment, GCSs may be distributed in residential areas, commercial areas, near schools, etc. Each GCS is equipped with an intelligent terminal with a built-in SoC module, which is responsible for monitoring the local electromagnetic environment and synchronizing with other GCSs. However, in practical applications, the communication quality between GCSs is affected by various factors, as follows:

[0004] Increased data transmission requirements: With the development of smart cities, more and more sensors are deployed on GCSs, which not only increases the data volume but also poses higher requirements for wireless communication. A large amount of data needs to be transmitted quickly and reliably to ensure timely update and processing, which poses a challenge to the existing wireless communication infrastructure.

[0005] Limitations of fixed positions and complex and variable electromagnetic environment: Since GCSs are usually located in fixed places, although the electromagnetic environment around them has a certain stability and repeatability, it is more likely to be affected by specific interference sources. For example, although the GCSs are in fixed positions, the urban environment where they are located is full of various radio interference sources, including but not limited to Wi-Fi routers, mobile communication base stations, household appliances such as microwave ovens, and occasional temporary interferences brought by large-scale activities. Especially during peak hours or special events, the number and intensity of electromagnetic interference sources may suddenly increase, making the originally stable communication link unstable. Taking the GCS near a school as an example, there will be two obvious Wi-Fi interference peaks in the morning and afternoon on weekdays, while in the commercial area, the working hours of surrounding enterprises will also affect the change of the electromagnetic environment. This periodic interference pattern not only increases the risk of communication interruption but also may lead to waste or congestion of frequency band resources.

[0006] Lack of intelligent prediction mechanism: Current technical solutions often lack intelligent prediction functions and are unable to predict the changing trends of the electromagnetic environment in advance, thus increasing the risk of communication interruption. This means that the system cannot effectively anticipate potential problems and take preventive measures to avoid communication issues.

[0007] Insufficient resource utilization: The spectrum resources are not fully utilized, especially when facing a rapidly changing electromagnetic environment, which is prone to causing frequency band congestion or waste.

[0008] Poor adaptability: It responds slowly to the changes in the electromagnetic environment under different scenarios and is difficult to achieve real-time optimization and adjustment. A more flexible and responsive system can better adapt to different environmental conditions and ensure the continuity and stability of communication.

[0009] Inadequate fixed-frequency strategy: Traditional fixed-frequency or simple frequency hopping strategies are difficult to cope with complex and changing electromagnetic environments. Especially during peak hours or special events, the number and intensity of electromagnetic interference sources may suddenly increase, further exacerbating the difficulty of stable communication.

[0010] In summary, the current GCS wireless communication technology faces many challenges in achieving efficient and stable data transmission. Therefore, it is necessary to develop a spectrum prediction and frequency hopping method and system in the self-organizing network of sanitation equipment SoC. Summary of the Invention

[0011] The purpose of the present invention is to provide a spectrum prediction and frequency hopping method and system in the self-organizing network of sanitation equipment SoC to ensure efficient data transmission links even in complex and changing electromagnetic environments.

[0012] In the first aspect, the spectrum prediction and frequency hopping method in the self-organizing network of sanitation equipment SoC according to the present invention includes the following steps:

[0013] The GCS node continuously collects the surrounding electromagnetic environment data through the spectrum monitoring unit of its built-in SoC module, including frequency usage, interference intensity, signal-to-noise ratio, bit error rate, and geographical location information; performs local preprocessing on the collected electromagnetic environment data, extracts key features and converts them into a standardized format, and then periodically uploads the processed data to the cloud server for data sharing;

[0014] Uses the Hidden Markov Model to model and train the uploaded electromagnetic environment data, including: defining the state space, generating an observation sequence based on the key features, constructing a transition probability matrix based on the observation sequence, constructing an emission probability matrix, determining the initial state distribution, and optimizing the parameters of the Hidden Markov Model through the Baum-Welch algorithm;

[0015] Based on the trained Hidden Markov Model, predict the electromagnetic environment, including: calculating the likelihood of each state at each time point using the forward algorithm and finding the most likely state sequence using the Viterbi algorithm; according to the predicted state sequence, combined with the predefined spectrum resource allocation rules, select the best hopping frequency band for communication and measure the communication quality after frequency hopping through a reward value;

[0016] After each frequency hopping, record the communication success rate and interference situation at the new frequency and feedback them to the Hidden Markov Model, use the online learning algorithm to update the parameters of the Hidden Markov Model in real time, and periodically retrain the Hidden Markov Model using the batch learning algorithm to achieve self-optimization of the Hidden Markov Model.

[0017] Optionally, the local preprocessing step includes:

[0018] Extract key features including center frequency, bandwidth, and power density and convert them into a standardized format; then mark and store the processed data according to the time stamp. This step further refines the local preprocessing process, ensures the effective extraction of key features and the standardized storage of data, and provides a reliable basis for subsequent data processing and analysis.

[0019] Optionally, use the Hidden Markov Model to model and train the uploaded electromagnetic environment data, specifically including:

[0020] Define the state space: Based on the key features after local preprocessing, according to the actually monitored electromagnetic environment change data and the preprocessed key features, define the electromagnetic environment configuration, and set the state space S = {s1, s2,..., s n}, where each state is composed of a set of feature vectors, describing the electromagnetic environment characteristics within a specific time period;

[0021] Generate the observation sequence: Extract features from the actually monitored electromagnetic environment change data to form an ordered observation sequence, and set the observation sequence O = {o1, o2,..., o t}, where each observation value o t is composed of a set of feature vectors, specifically including center frequency, bandwidth, power density, signal-to-noise ratio, and bit error rate, describing the electromagnetic environment state at the current moment;

[0022] Construct the transition probability matrix: Based on the observation sequence, judge the probability of a state transitioning to another state, and set the transition probability matrix A = [a ij , where a ij represents the probability of transitioning from state s i to state s j ;

[0023] Construct the emission probability matrix: Calculate the probability of emitting at state sj The probability of observing specific electromagnetic environment characteristics o t Let the emission probability matrix B = [b j (o t )], where b j (o t ) represents the probability of observing specific electromagnetic environment characteristics o j under state s t .

[0024] Determination of the initial state distribution: Based on the prior probability obtained from historical data statistics, let the initial state distribution π = {π1, π2,..., π n} represent the prior probabilities of the occurrence of each state;

[0025] Model training: Use the Baum-Welch algorithm to perform the expectation-maximization iteration until the hidden Markov model converges.

[0026] Optionally, use the Baum-Welch algorithm to perform the expectation-maximization iteration until the hidden Markov model converges, specifically including:

[0027] E step: Calculate the likelihood of each state at each time point.

[0028] M step: Update the transition probability matrix A, the emission probability matrix B, and the initial state distribution π according to the results of the E step;

[0029] During the model training process, the E step and the M step are alternated. The transition probability matrix A, the emission probability matrix B, and the initial state distribution π are updated in each iteration until the hidden Markov model converges. The alternation of the E step and the M step is the core step of the Baum-Welch algorithm. By continuously iterating and optimizing the model parameters, the model can gradually approximate the real electromagnetic environment distribution and improve the prediction performance of the model.

[0030] Optionally, use the forward algorithm to calculate the likelihood of each state at each time point, specifically:

[0031] Based on the trained hidden Markov model, for a given observation sequence O = {o1, o2,..., o t}, use the forward algorithm to calculate the likelihood of being in the states of "strong Wi-Fi interference", "weak Wi-Fi interference", or "no obvious interference" at each time point.

[0032] Optionally, use the Viterbi algorithm to find the most likely state sequence, specifically: Based on the results calculated by the forward algorithm, use the Viterbi algorithm to find the most likely state sequence.

[0033] Optionally, according to the predicted state sequence and in combination with the predefined spectrum resource allocation rules, the best hopping frequency band is selected for communication. Specifically:

[0034] Based on the predicted state sequence and in combination with the predefined spectrum resource allocation rules, a frequency that is expected to provide the best communication quality and is not overused is selected as the next hopping frequency band.

[0035] Optionally, the calculation formula for the reward value is:

[0036]

[0037] where R is the reward value, c is the communication success rate, d is the delay, e is the bit error rate, u is the spectrum occupancy, and w1, w2, w3, and w4 are weight coefficients, and the sum of the four is equal to 1.

[0038] Optionally, the parameters of the hidden Markov model are updated in real time using the incremental Baum-Welch algorithm; the hidden Markov model is retrained regularly using the standard Baum-Welch algorithm.

[0039] In a second aspect, a spectrum prediction and frequency hopping system in a sanitation equipment SoC ad hoc network according to the present invention includes a plurality of GCS nodes, and each node is equipped with an intelligent terminal with a built-in SoC module for performing the steps of the dynamic spectrum prediction and adaptive frequency hopping method as described in the present invention.

[0040] Advantages of the present invention:

[0041] (1) Effectively cope with the increasing data transmission demand: Through the intelligent spectrum prediction and frequency hopping method, a large amount of data can be transmitted quickly and reliably, meeting the data transmission demand brought about by the increasing number of sensor deployments under the development of smart cities, ensuring timely update and processing of information, and effectively alleviating the pressure on existing wireless communication infrastructure.

[0042] (2) Overcome the limitations of fixed positions and the influence of complex electromagnetic environments: By using the introduced HMM model to predict the trend of electromagnetic environment changes in advance, the system can sense and adapt to complex electromagnetic environments in advance, effectively reducing the risk of communication interruption caused by specific interference sources (such as Wi-Fi routers, mobile communication base stations, household appliances such as microwave ovens, and temporary interference caused by large-scale activities), avoiding waste or congestion of frequency band resources, and improving the stability of communication links. For example, in different scenarios such as near schools and commercial areas, the communication strategy can be adjusted in advance according to the predicted electromagnetic interference pattern, reducing the possibility of communication interruption.

[0043] (3) Build an intelligent prediction mechanism: Introduce the HMM model to achieve the intelligent prediction function, anticipate the changing trend of the electromagnetic environment in advance, enable the system to take preventive measures proactively, avoid the occurrence of communication problems, significantly reduce the risk of communication interruption, and enhance the reliability and stability of the system.

[0044] (4) Achieve full utilization of spectrum resources: Through a reasonable spectrum resource allocation algorithm, combined with spectrum prediction and frequency hopping methods, in the face of a rapidly changing electromagnetic environment, it can dynamically adjust spectrum usage, avoid frequency band congestion or waste, improve the overall utilization efficiency of spectrum resources, and enhance communication performance.

[0045] (5) Enhance system adaptability: Design a more flexible and responsive spectrum prediction and frequency hopping system that can quickly sense the changes in the electromagnetic environment under different scenarios and perform real-time optimization and adjustment to ensure the continuity and stability of communication under various complex environmental conditions, and meet the communication requirements of GCS in different regions (near residential areas, commercial areas, schools, etc.).

[0046] (6) Improve the fixed frequency strategy: Abandon the traditional fixed frequency or simple frequency hopping strategy, and adopt an intelligent frequency hopping method based on spectrum prediction, which can dynamically adjust frequency hopping parameters according to the real-time electromagnetic environment, effectively cope with the complex and changing electromagnetic environment, especially during peak hours or special events, reduce the impact of the sudden increase in the number and intensity of electromagnetic interference sources on communication stability, and ensure communication quality.

[0047] In summary, the spectrum prediction and frequency hopping method and system in the self-organizing network of the sanitation equipment SoC provide effective solutions for the numerous challenges faced by the current GCS wireless communication technology, significantly improving the stability, efficiency, and adaptability of communication and reducing the maintenance cost. Brief Description of the Drawings

[0048] Figure 1 is the flowchart of the spectrum prediction and frequency hopping method in the self-organizing network of the sanitation equipment SoC described in the embodiment of the present application;

[0049] Figure 2 is the flowchart of modeling and training the uploaded electromagnetic environment data using the hidden Markov model in the embodiment of the present application;

[0050] Figure 3 is the flowchart of predicting the electromagnetic environment based on the trained hidden Markov model in the embodiment of the present application. Detailed Embodiments

[0051] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention, rather than for limiting the protection scope of the present invention.

[0052] As Figure 1 shown, in the embodiment of the present application, a spectrum prediction and frequency hopping method in a sanitation equipment SoC self-organizing network includes the following steps:

[0053] Spectrum monitoring and data collection: The GCS node continuously collects the surrounding electromagnetic environment data through the spectrum monitoring unit of its built-in SoC module, including frequency usage, interference intensity, signal-to-noise ratio, bit error rate, and geographical location information; locally preprocesses the collected electromagnetic environment data, extracts key features and converts them into a standardized format, and then periodically uploads the processed data to the cloud server for data sharing;

[0054] HMM modeling and training: Use the hidden Markov model to model and train the uploaded electromagnetic environment data, including: defining the state space, generating an observation sequence based on the key features, constructing a transition probability matrix based on the observation sequence, constructing an emission probability matrix, determining the initial state distribution, and optimizing the parameters of the hidden Markov model through the Baum-Welch algorithm;

[0055] Spectrum prediction and frequency hopping decision: Based on the trained hidden Markov model, predict the electromagnetic environment, including: calculating the possibility of each state at each time point using the forward algorithm, and using the Viterbi algorithm to find the most likely state sequence; according to the predicted state sequence, combined with the pre-defined spectrum resource allocation rules, select the best frequency hopping band for communication, and measure the communication quality after frequency hopping through the reward value;

[0056] Dynamic adjustment mechanism: After each frequency hopping, record the communication success rate and interference situation at the new frequency and feedback them to the hidden Markov model, use the online learning algorithm to update the parameters of the hidden Markov model in real time, and use the batch learning algorithm to retrain the hidden Markov model regularly to achieve self-optimization of the hidden Markov model.

[0057] In the embodiments of the present application, a spectrum prediction and frequency hopping method in the SoC ad-hoc network of sanitation equipment is applied to the SoC ad-hoc network between garbage collection stations (GCS) in the urban sanitation system. Each GCS will independently perform spectrum monitoring, prediction, and adjustment, but data sharing and model synchronization are achieved through the cloud server to ensure the coordination and consistency of the overall network.

[0058] The following is a detailed description of each step:

[0059] 1. Spectrum monitoring and data collection

[0060] This step provides the necessary input data for subsequent HMM modeling to ensure the accuracy and integrity of the data, as follows:

[0061] 1.1 Continuously collect data on changes in the surrounding electromagnetic environment

[0062] Each GCS is equipped with an intelligent terminal with a built-in SoC module, which is responsible for monitoring the local electromagnetic environment and recording the success rate of each communication and the interference encountered. The SoC module is built with a high-precision spectrum monitoring unit to continuously collect data on changes in the surrounding electromagnetic environment, including the following types of information:

[0063] Frequency usage: Which frequency bands are being used.

[0064] Interference intensity: How strong the signals are on different frequency bands and whether there is strong interference.

[0065] Signal-to-noise ratio (SNR): The ratio of the useful signal to the background noise.

[0066] Bit error rate (BER): The proportion of errors that occur when transmitting data.

[0067] Geographical location information: At which specific location these changes occur.

[0068] Suppose GCS1 in a residential area has a Wi-Fi signal strength of 75 dBm, an SNR of 25 dB, and a BER of 0.05% during the period from 8 pm to 10 pm; while GCS2 in a commercial area has a mobile base station signal strength of 80 dBm, an SNR of 20 dB, and a BER of 0.1% during the period from 9 am to 5 pm on weekdays.

[0069] 1.2 Local preprocessing

[0070] All the collected data will be preliminarily processed and stored in the local database of the SoC module, and marked according to the time stamp.

[0071] Extract key features such as center frequency, bandwidth, power density, etc., and convert them into a standardized format. These key features are extracted from the original acquisition data and are the most important parameters reflecting the changes in the electromagnetic environment. Suppose the original data is Wi-Fi signal strength of 75 dBm, signal-to-noise ratio of 25 dB, and bit error rate of 0.05%. Then the possible standardized features after preprocessing are: center frequency of 2.4 GHz, bandwidth of 20 MHz, and power density of 75 dBm / Hz.

[0072] 1.3 Periodic synchronization

[0073] At regular time intervals (e.g., every hour), upload the processed data to the cloud server.

[0074] Data compression and encryption: Reduce the transmission bandwidth and protect privacy.

[0075] In summary, the spectrum monitoring and data acquisition in this step are the basis for the subsequent steps. By continuously collecting and processing data, it provides the necessary input for HMM modeling.

[0076] 2. HMM Modeling and Training

[0077] As Figure 2 shown, in this step, construct a Hidden Markov Model (HMM) that can capture the time series characteristics of the electromagnetic environment, and continuously optimize its parameters through training to improve the prediction accuracy, as follows:

[0078] 2.1 Define the state space

[0079] Based on the key features after local preprocessing: According to the actually monitored electromagnetic environment change data and the key features after preprocessing, define the possible electromagnetic environment configurations. Let the state space S = {s1, s2,..., s n}, where each state is composed of a set of feature vectors, describing the electromagnetic environment characteristics within a specific time period. For example:

[0080] s1: Strong Wi-Fi interference, with features of center frequency 2.4 GHz, bandwidth 20 MHz, power density > 70 dBm / Hz, signal-to-noise ratio < 25 dB, and bit error rate > 0.1%.

[0081] s2: Weak Wi-Fi interference, with features of center frequency 2.4 GHz, bandwidth 20 MHz, power density 50 - 70 dBm / Hz, signal-to-noise ratio 25 - 35 dB, and bit error rate 0.05% - 0.1%.

[0082] s3: No obvious interference, with features of center frequency 2.4 GHz, bandwidth 20 MHz, power density < 50 dBm / Hz, signal-to-noise ratio > 35 dB, and bit error rate < 0.05%.

[0083] s4: Strong base station interference, characterized by a central frequency of 900 MHz, a bandwidth of 200 kHz, a power density > 80 dBm / Hz, a signal-to-noise ratio < 20 dB, and a bit error rate > 0.1%.

[0084] s5: Weak base station interference, characterized by a central frequency of 900 MHz, a bandwidth of 200 kHz, a power density of 50 - 80 dBm / Hz, a signal-to-noise ratio of 20 - 30 dB, and a bit error rate of 0.05% - 0.1%.

[0085] 2.2 Observation sequence generation

[0086] Based on the specific values defined in the state space: Extract features from the actually monitored electromagnetic environment change data to form an ordered observation sequence. Let the observation sequence O = {o1, o2,..., o t}, where each observation value o t is composed of a set of feature vectors (specifically including the central frequency, bandwidth, power density, signal-to-noise ratio, and bit error rate), describing the electromagnetic environment state at the current moment.

[0087] Suppose that at a certain moment, the measured Wi-Fi signal strength is 80 dBm, the signal-to-noise ratio is 20 dB, and the bit error rate is 0.1%. The key features obtained after preprocessing are: a central frequency of 2.4 GHz, a bandwidth of 20 MHz, and a power density of 80 dBm / Hz. These key features constitute an observation value o1. At another moment, the measured mobile base station signal strength is 85 dBm, the signal-to-noise ratio is 18 dB, and the bit error rate is 0.12%. The key features obtained after preprocessing are: a central frequency of 900 MHz, a bandwidth of 200 kHz, and a power density of 85 dBm / Hz. These key features constitute another observation value o2.

[0088] 2.3 Transition probability matrix

[0089] The transition probability matrix represents the probability of transitioning from one electromagnetic environment state to another, which helps the system predict the change trend of the electromagnetic environment in the future for a period of time, so as to better guide the frequency hopping decision. Specifically as follows:

[0090] Based on the observation sequence to judge the probability of a state transitioning to another state: Represents the probability of transitioning from one electromagnetic environment state to another. Let the transition probability matrix A = [a ij , where a ij represents the probability of transitioning from state s i to state s j .

[0091] If the observation sequence shows that the state of "no obvious interference" is often followed by the state of "strong Wi-Fi interference", then the corresponding transition probability can be set relatively high. For example, a31 = 0.6. If the observation sequence often shows the "weak base station interference" state after showing the "strong base station interference" state, the corresponding transition probability can be set relatively high, for example, a 45 = 0.7.

[0092] 2.4 Emission Probability Matrix

[0093] The emission probability matrix represents the probability of observing specific electromagnetic environment characteristics in a certain state, which helps the system determine which specific observed values are more likely to occur in a given state, thereby improving the accuracy of the model. Specifically as follows:

[0094] Used to calculate the probability of observing specific electromagnetic environment characteristics in a certain state: Represents the probability of observing specific electromagnetic environment characteristics in a certain state. Let the emission probability matrix B = [b j (o t )], where b j (o t ) represents the probability of observing specific electromagnetic environment characteristic o j under state s t .

[0095] For example, in the "strong Wi-Fi interference" state, the probability of observing a Wi-Fi signal strength of 80 dBm is 80%, that is, b1(80 dBm) = 0.8. In the "strong base station interference" state, the probability of observing a base station signal strength of 85 dBm is 75%, that is, b4(85 dBm) = 0.75.

[0096] 2.5 Initial State Distribution

[0097] The initial state distribution represents the possibility of each state occurring in the initial stage, which is the prior information of the model and affects the initial setting of the model and subsequent parameter updates. Specifically as follows:

[0098] Prior probability statistically obtained based on historical data: Let the initial state distribution π = {π1, π2,..., π k ,..., π n} represents the prior probability of each state occurring. The prior probability is statistically obtained based on the historical data obtained in the spectrum monitoring and data acquisition steps, reflecting the possibility of each state occurring in the initial stage.

[0099] Suppose the initial probability of the "no obvious interference" state is 30%, then π3 = 0.3 can be set. This prior probability is statistically obtained based on historical data, reflecting the possibility of each state occurring in the initial stage.

[0100] 2.6 Model Training

[0101] Continuously optimize the model parameters through the Baum-Welch algorithm: Use the Baum-Welch algorithm to perform the Expectation-Maximization (EM) iteration until the model converges. Each iteration consists of two main steps:

[0102] E step (Expectation Step): Calculate the likelihood of each state at each time point. Specifically, for a given observation sequence O, calculate the likelihood of being in state s at each time point t i . The observation sequence here can be either historical data or real-time data, depending on the requirements of the application scenario. When the model is first trained or readjusted, historical data is usually used to initialize the parameters and establish the initial transition probability matrix, emission probability matrix, and initial state distribution. This is because historical data provides rich prior information, which helps the model quickly converge to a reasonable starting point. Once the model has been initially trained, online learning can start using real-time data, which enables the model to adjust according to the latest electromagnetic environment changes and ensures that it is always in the optimal state.

[0103] M step (Maximization Step): Update the transition probability matrix A, emission probability matrix B, and initial state distribution π according to the results of the E step.

[0104] Among them, update the transition probability matrix: According to the likelihood of states at each time point calculated in the E step, re-estimate the probability of transitioning from state s i to state s j .

[0105] Among them, update the emission probability matrix: According to the likelihood of states at each time point calculated in the E step, re-estimate the probability of observing a specific electromagnetic environment feature o j under state s t .

[0106] Among them, update the initial state distribution: According to the likelihood of the initial state calculated in the E step, re-estimate the prior probability of each state. Even if the historical data does not change, as new data arrives, the initial state distribution π will be adjusted according to the new observation sequence to reflect the latest electromagnetic environment changes.

[0107] In summary, this step realizes HMM modeling and training. The definition of the state space depends on the key features after local preprocessing, defining the possible electromagnetic environment configurations. The generation of the observation sequence is based on the state space definition, extracting features from the actual monitoring data to form an ordered time series. The transition probability matrix and the emission probability matrix are based on the observation sequence, representing the transition probabilities between states and the probabilities of observing specific electromagnetic environment features. The initial state distribution is based on the prior probabilities obtained from historical data statistics, affecting the initial setting of the model and subsequent parameter updates. The model training iteratively optimizes the model parameters through the EM algorithm to ensure that the model can capture the time series characteristics of the electromagnetic environment, providing a basis for subsequent spectrum prediction.

[0108] 3. Spectrum Prediction and Frequency Hopping Decision

[0109] As Figure 3 shown, this step selects the best frequency hopping band based on the prediction results of the HMM model to ensure the optimization of communication quality, specifically as follows:

[0110] 3.1 Forward Algorithm Calculation

[0111] Based on the trained HMM model: For a given observation sequence O = {o1, o2,..., o t}, use the forward algorithm to calculate the likelihood of each state at each time point.

[0112] Assume the current observation sequence is O = {o1, o2,..., o t}, calculate the likelihood of being in the states of "strong Wi-Fi interference", "weak Wi-Fi interference", or "no obvious interference" at each time point.

[0113] 3.2 Viterbi Algorithm Path Decoding

[0114] Based on the results calculated by the forward algorithm: Use the Viterbi algorithm to find the most likely state sequence Q * such that Q * =

[0115] argmax Q P(Q | O).

[0116] Assume that the most likely state sequence decoded by the Viterbi algorithm for a future period of time is "no obvious interference" -> "strong Wi-Fi interference" -> "weak Wi-Fi interference".

[0117] This is not limited to Wi-Fi interference, but also includes other common electromagnetic environment interference sources such as mobile base stations, Bluetooth devices, and microwave ovens.

[0118] 3.3 Frequency Hopping Frequency Selection

[0119] Predicted state sequence: Combining predefined spectrum resource allocation rules, select those frequencies that are expected to provide the best communication quality and are not overused as the next-hop frequency band.

[0120] If the prediction result shows that there is strong Wi-Fi interference in the electromagnetic environment corresponding to the next state, preferentially select frequencies far from the Wi-Fi frequency band; if a certain frequency has a high historical communication success rate, preferentially consider this frequency as a candidate.

[0121] Of course, a reward function can also be introduced to select the best hopping frequency band to ensure the optimization of communication quality, as follows:

[0122] Define the following reward function formula to measure the communication quality after each frequency hop:

[0123]

[0124] Where R is the reward value, c is the communication success rate, d is the delay, e is the bit error rate, and u is the spectrum occupancy. w1, w2, w3, and w4 are weight coefficients, and the sum of the four is equal to 1.

[0125] For example: GCS1 is located in a residential area and is mainly affected by home Wi-Fi devices. Especially between 8 pm and 10 pm, the Wi-Fi signal strength is high, the signal-to-noise ratio is low, and the bit error rate is high.

[0126] Initial state: The current time is 8 pm, and the HMM model predicts that there is strong Wi-Fi interference in the electromagnetic environment corresponding to the next state.

[0127] Candidate frequencies: There are two candidate frequencies to choose from:

[0128] Frequency 1: 2.4 GHz frequency band, with an expected communication success rate c = 0.7, delay d = 10, bit error rate e = 0.2, and spectrum occupancy u = 0.8.

[0129] Frequency 2: 5 GHz frequency band, with an expected communication success rate c = 0.9, delay d = 5, bit error rate e = 0.92, and spectrum occupancy u = 2.

[0130] Then, according to the calculation formula of the reward value, it is calculated that the reward value of Frequency 2 is higher. Therefore, the 5 GHz frequency band is selected for communication to improve the communication success rate and bandwidth utilization rate while reducing the bit error rate.

[0131] In summary, the forward algorithm in this step calculates the likelihood of each state at each time point, providing input for the Viterbi algorithm path decoding. The Viterbi algorithm path decoding is used to find the most likely state sequence, providing a basis for the hopping frequency selection. The hopping frequency selection is based on the predicted state sequence to select the best hopping frequency band, ensuring the optimization of communication quality.

[0132] 4. Dynamic Adjustment Mechanism

[0133] This step ensures that the HMM model is always in an optimal state through immediate feedback and parameter updates, and continuously improves the prediction accuracy through self-optimization, as follows:

[0134] 4.1 Immediate Feedback

[0135] Based on the results of the hopping frequency selection: Whenever the SoC module completes a frequency hop, it records the communication success rate and interference situation at the new frequency and feeds this information back to the HMM model.

[0136] Suppose the 5GHz frequency band is selected for communication. Calculate the ratio of the number of successfully transmitted data packets to the total number of transmitted data packets (communication success rate), measure the signal power density at the new frequency, and identify whether there are strong interference sources.

[0137] 4.2 Parameter Update

[0138] Based on the data of immediate feedback: Use an online learning algorithm (such as the incremental Baum-Welch algorithm) to update the parameters of the HMM model in real time to ensure that the model is always in an optimal state.

[0139] If it is found that the communication success rate of the 5GHz frequency band is very high, update the transition probability matrix and emission probability matrix to make the HMM model more inclined to select this frequency band.

[0140] 4.3 Self-optimization

[0141] Based on the results of parameter updates: Over time, continuously accumulate more data and periodically retrain the HMM model through a batch learning algorithm (such as the standard Baum-Welch algorithm) to further improve the prediction accuracy.

[0142] All SoC modules of the GCS regularly upload local data to the cloud server to form a global spectrum environment database. Utilize the distributed computing resources in the cloud to accelerate the training process of the HMM model. Customize and train a dedicated HMM model for each GCS according to the electromagnetic environment characteristics of different regions to achieve more accurate spectrum prediction and frequency hopping decision-making.

[0143] In summary, the real-time feedback in this step provides the latest data input for parameter update, ensuring that the model can quickly adapt to the changes in the new electromagnetic environment. The parameter update continuously optimizes the model parameters through an online learning algorithm, ensuring that the HMM model is always in an optimal state. Self-optimization retrains the HMM model regularly through a batch learning algorithm to further improve the prediction accuracy and ensure long-term stable performance improvement.

[0144] The following is a typical example:

[0145] Suppose there are three GCSs in a certain city, located in a residential area (GCS1), a business district (GCS2), and near a school (GCS3) respectively. Each GCS is equipped with an intelligent terminal with a built-in SoC module, which is responsible for monitoring the local electromagnetic environment and recording the success rate of each communication and the interference situation encountered.

[0146] 1. Spectrum monitoring and data collection:

[0147] GCS1 is in a residential area and is mainly affected by household Wi-Fi devices, especially between 8 pm and 10 pm.

[0148] GCS2 is in a business district and is mainly affected by the Wi-Fi of surrounding enterprises and mobile base stations, especially between 9 am and 5 pm on weekdays.

[0149] GCS3 is near a school and is mainly affected by campus Wi-Fi devices, especially between 7 am and 9 am and between 3 pm and 5 pm on weekdays.

[0150] Each GCS continuously collects data on the surrounding electromagnetic environment and uploads the processed data to the cloud server.

[0151] 2. HMM modeling and training:

[0152] Based on the above collected data, define the state space, such as "strong Wi-Fi interference", "weak Wi-Fi interference", "no obvious interference", "strong base station interference", "weak base station interference", etc.

[0153] Generate an observation sequence according to the specific values actually monitored.

[0154] Calculate the transition probability matrix and the emission probability matrix, and set the initial state distribution.

[0155] Use the Baum-Welch algorithm to train the HMM model to ensure that the model can capture the time series characteristics of the electromagnetic environment.

[0156] 3. Spectrum prediction and frequency hopping decision:

[0157] At 7:00 in the morning on a weekday, GCS3 detects that it is about to enter the Wi-Fi interference peak. The HMM model predicts that there is strong Wi-Fi interference in the electromagnetic environment corresponding to the next state, and it preferentially selects the 5GHz band or other frequencies with less interference for communication.

[0158] Similarly, at 9:00 in the morning on a weekday, GCS2 detects that it is about to enter the peak period of commercial activities. The HMM model predicts that there is strong Wi-Fi and mobile base station interference in the electromagnetic environment corresponding to the next state, and it preferentially selects frequencies far from these bands for communication.

[0159] 4. Dynamic adjustment mechanism:

[0160] After each frequency hopping, the SoC module records the communication success rate and interference situation at the new frequency and feeds this information back to the HMM model for real-time updating of the model parameters.

[0161] Through instant feedback and parameter update, it is ensured that the model is always in the optimal state and the prediction accuracy is continuously improved through self-optimization.

[0162] 5. Multi-GCS cooperation mechanism:

[0163] Personalized model: Each GCS has its own HMM model, which is personalized trained according to its location and the characteristics of the electromagnetic environment.

[0164] Data sharing: The SoC modules of all GCSs regularly upload local data to the cloud server to form a global spectrum environment database for other GCSs to refer to.

[0165] Model synchronization: The cloud server uses distributed computing resources to accelerate the training process of the HMM model and distributes the updated model parameters to each GCS to ensure the coordination and consistency of the overall network.

[0166] In the embodiment of the present application, a spectrum prediction and frequency hopping system in a self-organizing network of an environmental sanitation equipment SoC includes multiple GCS nodes, and each node is equipped with an intelligent terminal with a built-in SoC module for executing the steps of the dynamic spectrum prediction and adaptive frequency hopping method as described in the present invention.

[0167] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement methods and shall be included in the protection scope of the present invention.

Claims

1. A spectrum prediction and frequency hopping method in the SoC self-organizing network of sanitation equipment, characterized in that, It includes the following steps: The GCS node continuously collects the surrounding electromagnetic environment data through the spectrum monitoring unit of its built-in SoC module, including frequency usage, interference intensity, signal-to-noise ratio, bit error rate, and geographical location information; performs local preprocessing on the collected electromagnetic environment data, extracts key features and converts them into a standardized format, and then periodically uploads the processed data to the cloud server for data sharing; Use the Hidden Markov Model to model and train the uploaded electromagnetic environment data, including: defining the state space, generating an observation sequence based on the key features, constructing a transition probability matrix based on the observation sequence, constructing an emission probability matrix, determining the initial state distribution, and optimizing the parameters of the Hidden Markov Model through the Baum-Welch algorithm; Based on the trained Hidden Markov Model, predict the electromagnetic environment, including: using the forward algorithm to calculate the possibility of each state at each time point, and using the Viterbi algorithm to find the most likely state sequence; according to the predicted state sequence, combined with the predefined spectrum resource allocation rules, select the best frequency hopping band for communication, and measure the communication quality after frequency hopping through the reward value; After each frequency hopping, record the communication success rate and interference situation at the new frequency and feedback them to the Hidden Markov Model, use the online learning algorithm to update the parameters of the Hidden Markov Model in real time, and use the batch learning algorithm to retrain the Hidden Markov Model regularly to achieve the self-optimization of the Hidden Markov Model.

2. The spectrum prediction and frequency hopping method in the sanitation equipment SoC ad-hoc network according to claim 1, characterized in that The local preprocessing step includes: Extracting key features including center frequency, bandwidth, and power density, and converting them into a standardized format; then marking and storing the processed data according to the timestamp.

3. The spectrum prediction and frequency hopping method in the sanitation equipment SoC ad hoc network according to claim 1, characterized in that Use the Hidden Markov Model to model and train the uploaded electromagnetic environment data, specifically including: Define the state space: Based on the key features after local preprocessing, according to the actually monitored electromagnetic environment change data and the preprocessed key features, define the electromagnetic environment configuration. Let the state space S = {s1, s2,..., s n}, where each state is composed of a set of feature vectors, describing the electromagnetic environment characteristics within a specific time period; Observation sequence generation: Extract features from the actually monitored electromagnetic environment change data to form an ordered observation sequence. Let the observation sequence O = {o1, o2,..., o t}, where each observation value o t is composed of a set of feature vectors, specifically including the center frequency, bandwidth, power density, signal-to-noise ratio, and bit error rate, which describe the electromagnetic environment state at the current moment; Construction of the transition probability matrix: Based on the observed sequence, judge the probability of a state transitioning to another state. Let the transition probability matrix A = [a ij , where a ij represents the probability of transitioning from state s i to state s j . Emission probability matrix construction: Calculate the probability of observing a specific electromagnetic environment feature o j in state s t . Let the emission probability matrix B = [b j (o t )], where b j (o t ) represents the probability of observing a specific electromagnetic environment feature o j in state s t . Initial state distribution determination: Based on the prior probability obtained from historical data statistics, let the initial state distribution π = {π1, π2,..., π n}, representing the prior probability of each state occurring; Model training: Use the Baum-Welch algorithm to perform the expectation-maximization iteration until the Hidden Markov Model converges.

4. The spectrum prediction and frequency hopping method in the sanitation equipment SoC ad-hoc network according to claim 3, characterized in that Use the Baum-Welch algorithm to perform the expectation-maximization iteration until the Hidden Markov Model converges, specifically including: E step: Calculate the possibility of each state at each time point. M step: Update the transition probability matrix A, the emission probability matrix B, and the initial state distribution π according to the results of the E step; During the model training process, the E step and the M step are performed alternately, and the transition probability matrix A, the emission probability matrix B, and the initial state distribution π are updated in each iteration until the Hidden Markov Model converges.

5. The spectrum prediction and frequency hopping method in the sanitation equipment SoC ad-hoc network according to claim 1, characterized in that Use the forward algorithm to calculate the possibility of each state at each time point, specifically: Based on the trained Hidden Markov Model, for a given observation sequence O = {o1, o2,..., o t}, the forward algorithm is used to calculate the likelihood of being in the states of "strong Wi-Fi interference", "weak Wi-Fi interference", or "no obvious interference" at each time point.

6. The spectrum prediction and frequency hopping method in the sanitation equipment SoC ad hoc network according to claim 1, characterized in that Use the Viterbi algorithm to find the most likely state sequence, specifically: Based on the results calculated by the forward algorithm, use the Viterbi algorithm to find the most likely state sequence.

7. The spectrum prediction and frequency hopping method in the sanitation equipment SoC ad-hoc network according to claim 1, characterized in that According to the predicted state sequence, combined with the predefined spectrum resource allocation rules, select the best frequency hopping band for communication, specifically: Based on the predicted state sequence, combined with the predefined spectrum resource allocation rules, select the frequency that is expected to provide the best communication quality and has not been overused as the next frequency hopping band.

8. The spectrum prediction and frequency hopping method in the sanitation equipment SoC ad hoc network according to claim 1, characterized in that The calculation formula of the reward value is: Among them, R is the reward value, c is the communication success rate, d is the delay, e is the bit error rate, u is the spectrum occupancy, and w1, w2, w3, and w4 are weight coefficients, and the sum of the four is equal to 1.

9. The spectrum prediction and frequency hopping method in the sanitation equipment SoC ad-hoc network according to claim 1, characterized in that Use the incremental Baum-Welch algorithm to update the parameters of the hidden Markov model in real time; use the standard Baum-Welch algorithm to retrain the hidden Markov model regularly.

10. A spectrum prediction and frequency hopping system in the SoC self-organizing network of sanitation equipment, characterized in that, It includes multiple GCS nodes, and each node is equipped with an intelligent terminal with a built-in SoC module for performing the steps of the dynamic spectrum prediction and adaptive frequency hopping method described in any one of claims 1 to 9.

Images