Wireless sensor network link quality regression estimation method

Abstract

The application provides a wireless sensor network link quality regression estimation method, which has the following characteristics: taking the sliding average values of signal-to-interference-and-noise ratio, square and cube of the signal-to-interference-and-noise ratio as input features, and using a linear regression model to estimate the sliding average value of the packet reception rate representing the link quality, comprising the following steps: step S1, establishing a tracking data set and solving model parameters, the signal-to-interference-and-noise ratio of the filling record in the tracking data set taking the minimum value of all records, and the sliding window size of the input features and the average value of the packet reception rate being equal; step S2, according to the channel signal strength collected by any node I in real time and the average value of the received signal strength corresponding to any neighbor node J, the current link quality between the node I and the node J is calculated by combining the model parameters. The method collects data in real time by the receiving node and estimates the packet reception rate, does not depend on the reception of the packet, and can timely and accurately estimate the current link quality under the condition of low calculation complexity.

Description

TECHNICAL FIELD

[0001] The application relates to a link quality estimation method of a wireless network, in particular to a wireless sensor network link quality regression estimation method. BACKGROUND

[0002] In a wireless network, link quality estimation is the basis for route selection, power control and multi-channel communication protocol channel selection, and the accuracy of link quality estimation directly affects the efficiency and reliability of the wireless network.

[0003] Existing methods include methods based on theoretical models, methods based on statistical packet reception numbers, methods directly using physical layer parameters, methods of establishing models online and machine learning methods.

[0004] Methods based on theoretical models, such as the self-adaptive wireless link quality estimation method based on a simplified theoretical model disclosed in patent application No. CN202010135687, use real-time collected parameters to obtain link quality estimation results by substituting the parameters into a calculation formula. Since the physical layer of a wireless sensor network usually uses direct sequence spread spectrum technology and phase shift keying modulation methods, it is difficult to derive an accurate error rate calculation formula, and an upper limit of the estimated error rate is usually used, so the method based on the theoretical model is limited in terms of link quality estimation accuracy. In addition, even if a simplified calculation formula is used, it contains exponential and square operations, and the calculation overhead is high. Moreover, since the error rate is difficult to accurately calculate, the method of counting the number of received packets is currently widely used. According to statistical theory, in order to accurately estimate the packet reception rate PRR, a large number of samples are required, which is difficult to achieve for wireless sensor network nodes with limited computing power and storage resources. Moreover, as the number of samples increases, the dynamic response performance decreases, so it is not suitable for rapidly changing wireless networks, and using a small number of samples will inevitably have a large error.

[0005] Methods that directly utilize physical layer parameters, such as those described in the literature by Sennel M, Chintalapudi K, Lal D, et al. AKalman Filter Based Link Quality Estimation Scheme for Sensor Network, doi:10.1109 / GLOCOM.2007.169, derive quantitative relationships between physical layer parameters such as RSSI, LQI, and / or SNR and PRR through analysis of experimental data. During network operation, PRR is estimated based on the pre-obtained relationships and real-time acquired physical layer parameters. However, because the relationship between physical layer parameters and PRR is significantly affected by environmental and interference factors, it is difficult to obtain mathematical expressions applicable to all links. Therefore, such methods are only suitable for networks with relatively stable link quality or situations where high estimation accuracy is not required.

[0006] Online modeling methods, such as the link quality estimator merging signal-to-noise ratio, link quality index, and packet reception rate disclosed in patent application CN201410597267, and the method for estimating wireless sensor network link quality using a small number of data packets disclosed in patent application CN201410584342, utilize a small amount of online-collected data to solve for the parameters of the link estimation model through regression algorithms. However, because regression algorithms require statistical PRR (Pulse Rate Reduction), and the statistical results often exhibit significant bias when the sample size is small, the accuracy of the link estimation model is affected. Furthermore, these methods rely on RSSI (Receiving Signal-to-Signal Index) and LQI (Low Quality Index) information from received data packets. On the one hand, because missing data is not processed, the model only includes information from successfully received packets, making it difficult to comprehensively estimate link quality; on the other hand, when the data acquisition period is long, these methods cannot respond promptly to dynamic changes in link quality. In addition, the computational load of regression algorithms is also substantial.

[0007] Machine learning methods, such as the method for evaluating link quality using a generalized regression neural network disclosed in patent application number CN201910346281, utilize datasets collected in the offline phase to train a certain model to obtain model parameters. In the online phase, the link quality is estimated using the model established in the offline phase and the input features acquired in real time. Currently disclosed methods use models including logistic regression, support vector machines, and neural networks, among others. Some of these models are highly complex and unsuitable for wireless sensor network nodes with limited computing and storage capabilities. In feature selection, multiple input parameters are generally used, including RSSI, SNR, LQI, and PRR. RSSI represents the signal strength when receiving data packets, but it does not reflect the relative strength of the signal with interference and noise. In dynamically changing environments, RSSI has a low correlation with link quality. Although the standard requires compatible hardware to provide LQI, different manufacturers' products differ in their implementation methods. Chips such as CC2420, CC2530, and SAM-R21 use the average correlation value of multiple consecutive symbols as LQI, while MC13234, K32W061 / 41, and JN5189 / 5188 use the average signal strength value when receiving packets as LQI. Some products do not provide LQI at all. Therefore, models using LQI as a feature have insufficient universality. In addition, LQI depends on packet reception and may not respond promptly to dynamic link characteristics. SNR and PRR have a high correlation, but existing hardware products do not directly provide SNR and require indirect acquisition using RSSI and energy detection (ED) results. Currently available methods generally employ a scheme that uses both SNR and LQI simultaneously. When the LQI provided by the physical layer is a correlation value, using two parameters simultaneously does not significantly improve the estimation accuracy compared to using only one parameter. When the LQI is the received signal strength, the correlation between LQI and PRR is low, and the accuracy of link quality estimation is mainly determined by the SNR component of the model. Methods that use the currently statistically determined PRR as input features leverage the correlation of continuously varying PRR values ​​to improve the accuracy of link quality estimation. However, using both physical layer parameters and PRR simultaneously increases the computational complexity of link quality estimation.

[0008] In summary, existing technologies have the problem of not being able to estimate the current link quality in a timely and accurate manner when computational complexity is low. Summary of the Invention

[0009] This invention is made to solve the above-mentioned problems, and aims to provide a regression estimation method for link quality in wireless sensor networks.

[0010] The present invention provides a method for wireless sensor network link quality regression estimation, which has the following characteristics and includes the following steps: Step S1, collecting data of each area of the wireless sensor network by using a test system, and calculating the coefficient vector w and intercept b of each area according to the collected data; Step S2, any node I in the wireless sensor network calculates the input features according to the real-time collected channel signal strength and the average received signal strength corresponding to any neighbor node J, and combines the coefficient vector w and intercept b of the area where node I is located to calculate the current link quality between node I and neighbor node J. Among them, Step S1 includes the following sub-steps: Step S1-1, dividing the deployment space of the wireless sensor network into each area, and using the test system to collect data in each area respectively. The test system includes multiple sending nodes, one receiving node and one computer; Step S1-2, in each area, the sending nodes of the test system send test packets to the receiving node at a period of T1, and the receiving node collects the channel signal strength at a period of T2 (T2 < T1) in the non-transmission state. Add the channel signal strength and the received signal strength when the receiving node receives the test packet to the received test packet, and then send the test packet to the computer; Step S1-3, according to all the test packets received in the area, construct a tracking data set of this area with records. The tracking data set includes original records and padding records; Step S1-4, respectively calculate the sliding average values corresponding to the signal-to-interference-plus-noise ratio, the square of the signal-to-interference-plus-noise ratio, the cube of the signal-to-interference-plus-noise ratio and the packet reception rate according to the records in the tracking data set, and add each sliding average value to the corresponding record in the tracking data set; Step S1-5, randomly select records from the tracking data set to form a training data subset, and the remaining records are used as a test data subset; Step S1-6, construct an input feature matrix X and an output vector y according to the sliding average values in the training data subset, and calculate the coefficient vector w and intercept b of this area according to the input feature matrix X and the output vector y; Step S1-7, according to the test data subset, test the regression model with the coefficient vector w and intercept b as parameters, and calculate the mean square error MSE and the coefficient of determination As a model performance indicator, step S2 includes the following sub-steps: Step S2-1, any node I in the wireless sensor network periodically collects channel signal strength in a non-transmission state and stores the L most recently collected channel signal strengths as the latest channel signal strength set; Step S2-2, when node I receives a data packet from any neighboring node J, it calculates and updates the average received signal strength of the signal received by node I from neighboring node J as the average received signal strength corresponding to node J; Step S2-3, based on the latest channel signal strength set and the average received signal strength corresponding to node J, it calculates the estimated values ​​of the most recent L signal-to-interference-plus-noise ratios of the signal received by node I from neighboring node J, and calculates the L most recent values ​​of the signal-to-interference-plus-noise ratio based on the L estimated values. Step S2-4: Based on L estimated values, L squared values, and L cubic values, with L as the sliding window length, calculate the sliding average value corresponding to the estimated value, the squared value, and the cubic value, and use it as the input feature; Step S2-5: Construct an input feature vector based on the input feature, and combine it with the coefficient vector w and intercept b of the region where node I is located to calculate the estimated value of the sliding average value of the packet reception rate of node I when neighbor node J sends; Step S2-6: Node I sends the estimated value of the sliding average value of the packet reception rate as the current link quality between node I and neighbor node J, so that neighbor node J can obtain the current link quality of the link with node I.

[0011] The wireless sensor network link quality regression estimation method provided by this invention may also have the following features: In steps S1-3, the original records correspond one-to-one with all received test packets and are arranged in the chronological order of the received test packets. Each original record includes the sending node identifier, packet sequence number, received signal strength, channel signal strength, and transmission power of the corresponding test packet. The filler records correspond one-to-one with the unreceived test packets, i.e., lost packets. The position and number of filler records are determined based on the sending node identifier and packet sequence number in the adjacent original records. The sending node identifier, packet sequence number, and transmission power in the filler records are determined based on the adjacent original records. The channel signal strength of the filler record is the maximum value among the channel signal strengths collected by the receiving node in the original records. The received signal strength of the filler record is the average value of the received signal strengths of all original records with the same sending node identifier and transmission power as the filler record.

[0012] The wireless sensor network link quality regression estimation method provided by this invention may also have the following feature: wherein, in steps S1-4, the formula for calculating the signal-to-interference-plus-noise ratio is: The formula for calculating the moving average of each term is: , , = , = In the formula Let i be the signal-to-interference-plus-noise ratio. Let i be the strength of the received signal. Let i be the signal strength of the i-th channel. , To track the total number of original records and filler records in the dataset, The moving average of the reception rate of the nth packet. The length of the sliding window. The number of original records in the sliding window, when hour, To track the number of original records from record 1 to record n ​​in the dataset, when hour, To track the number of original records in K consecutive records starting from n-K+1 in the dataset, Let n be the moving average of the signal-to-interference-plus-noise ratio (SIR). Let n be the moving average of the squared signal-to-interference-plus-noise ratio (SIR). The moving average of the cube of the nth signal-to-interference-plus-noise ratio (SIR) is calculated. The SIR of all records in the tracking data is calculated, and the minimum value is selected to replace the SIR of all filling records. Based on the replaced SIR, the SIR of the filling records, the square of the SIR, and the moving average of the cube of the SIR are calculated.

[0013] The wireless sensor network link quality regression estimation method provided by this invention may also have the following feature: wherein, in steps S1-5, , The value range is 0.6 to 0.75.

[0014] The wireless sensor network link quality regression estimation method provided by this invention may also have the following feature: wherein, in steps S1-6, the calculation formulas for the feature matrix X, output vector y, coefficient vector w, and intercept b are as follows: , , , , In the formula For the first The moving average of the signal-to-interference-plus-noise ratio of the records. For the first The moving average of the squared signal-to-interference-plus-noise ratio of the records. For the first The moving average of the cube of the signal-to-interference-plus-noise ratio of the records. For the first The moving average of the packet reception rate for each record.

[0015] The wireless sensor network link quality regression estimation method provided by this invention may also have the following feature: wherein, in step S2-3, the formula for calculating the estimated value of the signal-to-interference-plus-noise ratio is: In the formula Let be the estimated value of the signal-to-interference-plus-noise ratio (SIR) of node I when it receives the signal from its neighboring node J. Let J be the average received signal strength of the signal received by node I from its neighboring node J. The signal strength of the i-th channel collected for node I.

[0016] The wireless sensor network link quality regression estimation method provided by this invention may also have the following feature: wherein, in steps S2-4, the calculation formula for the input features is: , = , = In the formula The length of the sliding window. , Let be the estimated value of the signal-to-interference-plus-noise ratio (SIR) for the ni-th node when node I receives the signal from its neighbor node J. Let n be the moving average of the estimated signal-to-interference-plus-noise ratio (SIR) when node I receives the signal from neighboring node J. Let n be the moving average of the squared estimates of the signal-to-interference-plus-noise ratio (SIR) when node I receives signals from its neighboring node J. The nth signal-to-interference-plus-noise ratio (SIR) estimate is the cube moving average of the estimated values ​​when node I receives signals from neighboring node J.

[0017] The wireless sensor network link quality regression estimation method provided by this invention may also have the following feature: wherein, in steps S2-5, the formula for calculating the estimated value of the moving average of packet reception rate is: , In the formula For the input feature vector, Let n be the moving average of the estimated signal-to-interference-plus-noise ratio (SIR) when node I receives the signal from neighboring node J. Let n be the moving average of the squared estimates of the signal-to-interference-plus-noise ratio (SIR) when node I receives signals from its neighboring node J. Let n be the cube moving average of the estimated signal-to-interference-plus-noise ratio (SIR) when node I receives signals from neighboring node J. This is the transpose of the coefficient vector w. This is an estimate of the moving average of the packet reception rate of node I when neighbor node J sends a packet.

[0018] The wireless sensor network link quality regression estimation method provided by this invention may also have the following features: wherein, in steps S2-6, the method for sending the estimated value of the moving average of the packet reception rate includes: Node I periodically broadcasting a control message containing the estimated value of the moving average of the packet reception rate; or the network uses an acknowledgment message ACK, and after each data packet is received, Node I includes the estimated value of the moving average of the packet reception rate in the acknowledgment message ACK and sends it to the neighboring node J that sent the data packet; or the network has a request-response mechanism, and Node I includes the estimated value of the moving average of the packet reception rate in a CTS message and sends it to the neighboring node J that issued the RTS message; or the period during which the neighboring node J sends data is known, and Node I sends the estimated value of the moving average of the packet reception rate to the neighboring node J before the neighboring node J sends the next data packet.

[0019] The role and effect of invention

[0020] According to the wireless sensor network link quality regression estimation method of the present invention, on the one hand, by using the signal-to-interference-plus-noise ratio and its squared and cubic moving averages as input features and employing a linear regression model to estimate the moving average of packet reception rate, high estimation accuracy can be obtained with low computational complexity. On the other hand, by having the receiving node collect data and estimate link quality in real time, it is independent of the packet reception process and can estimate the current packet reception rate in a timely and accurate manner even when the data transmission interval is large. Therefore, the wireless sensor network link quality regression estimation method of the present invention can estimate the current link quality in a timely and accurate manner with low computational complexity. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the wireless sensor network link quality regression estimation method in an embodiment of the present invention. Detailed Implementation

[0022] To make the technical means, creative features, objectives and effects of this invention easy to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate the wireless sensor network link quality regression estimation method of this invention.

[0023] The wireless sensor network link quality regression estimation method of this invention uses the signal-to-interference-plus-noise ratio (SINR) and its squared and cubic values ​​as the moving average as input features. In the offline training phase, a tracking dataset is established using a test system, and missing values ​​are processed. The model parameters corresponding to each segmented region of the wireless sensor network are obtained using a subset of the training data, including the coefficient vector w and the intercept b. In the online phase, wireless sensor network nodes deployed in the corresponding regions collect channel parameters in real time, calculate the input features, and use the model parameters obtained in the offline training phase to estimate the link quality.

[0024] Figure 1 It is a schematic flow diagram of the method for estimating the regression of the link quality of the wireless sensor network in the embodiments of the present invention.

[0025] As Figure 1 shown, the method for estimating the regression of the link quality of the wireless sensor network includes the following steps:

[0026] Step S1, collect data of each area of the wireless sensor network by using a test system, and calculate the coefficient vector w and intercept b of each area according to the collected data. That is, through the off-line training method, use the test system to collect data including channel signal strength and received signal strength in each area, establish a tracking data set, fill in the missing data by using the originally recorded information, calculate the sliding average values of the signal-to-interference-plus-noise ratio, the square of the signal-to-interference-plus-noise ratio, and the cube of the signal-to-interference-plus-noise ratio as the input features of the regression model, and calculate the sliding average value of the packet reception rate as the output quantity of the regression model, so as to obtain the coefficient vector w and intercept b of each area. Specifically, it includes the following sub-steps:

[0027] Step S1-1, divide the deployment space of the wireless sensor network into each area, and use the test system to collect data in each area respectively. The test system includes multiple sending nodes, one receiving node and one computer. The sending nodes and receiving nodes of the test system have the same type of radio communication circuit as the nodes constituting the wireless sensor network. Some nodes of the wireless sensor network can be used as the sending nodes and receiving nodes.

[0028] Step S1-2, in each area, the sending nodes of the test system send test packets to the receiving node at a period of T1, and the receiving node collects the channel signal strength at a period of T2 (T2 < T1) in the non-transmission state, add the channel signal strength and the received signal strength when the receiving node receives the test packet to the received test packet, and then send the test packet to the computer.

[0029] Step S1-3, according to all the test packets received in the area, construct a tracking data set of this area with records. The tracking data set includes original records and filled records.

[0030] The original records correspond one-to-one with all received test packets and are arranged in chronological order of receipt. Each original record includes the sending node identifier, packet sequence number, received signal strength, channel signal strength, and transmission power of the corresponding test packet. The filler records correspond one-to-one with the unreceived test packets, i.e., lost packets. The position and number of filler records are determined based on the sending node identifier and packet sequence number in the adjacent original records. The sending node identifier, packet sequence number, and transmission power in the filler records are determined based on the adjacent original records. The channel signal strength of the filler record is the maximum value among the channel signal strengths collected by the receiving node in the original records. The received signal strength of the filler record is the average of the received signal strengths of all original records with the same sending node identifier and transmission power as the filler record.

[0031] Steps S1-4: Based on the records in the tracking dataset, calculate the signal-to-interference-plus-noise ratio (SIR), the square of the SIR, the cube of the SIR, and the moving average of the packet reception rate, and add the moving averages of each item to the corresponding records in the tracking dataset.

[0032] The formula for calculating the signal-to-interference-plus-noise ratio (SINR) is as follows:

[0033] ,

[0034] Calculate the signal-to-interference-plus-noise ratio (SIR / NNR) of all records in the tracking data, i.e., all filled records and the original records, and then select the minimum SIR / NNR among all records to replace the original SIR / NNR of all filled records.

[0035] The formula for calculating the moving average of each term is:

[0036] ,

[0037] ,

[0038] = ,

[0039] = ,

[0040] In the formula Let i be the signal-to-interference-plus-noise ratio. Let i be the strength of the received signal. Let i be the signal strength of the i-th channel. , To track the total number of original records and filler records in the dataset, The moving average of the reception rate of the nth packet. The length of the sliding window. The number of original records in the sliding window, when hour, To track the number of original records from record 1 to record n ​​in the dataset, when hour, To track the number of original records in K consecutive records starting from n-K+1 in the dataset, Let n be the moving average of the signal-to-interference-plus-noise ratio (SIR). Let n be the moving average of the squared signal-to-interference-plus-noise ratio (SIR). It is the moving average of the cube of the nth signal-to-interference-plus-noise ratio.

[0041] Steps S1-5: Randomly select from the tracking dataset 1 record, constituting a subset of the training data, the remaining These records are used as a subset of the test data.

[0042] in, The value range is 0.6 to 0.75.

[0043] Steps S1-6: Construct the input feature matrix X and output vector y based on the moving average values ​​of each item in the training data subset. Calculate the coefficient vector w and intercept b of the region based on the input feature matrix X and output vector y.

[0044] The formulas for calculating the feature matrix X, output vector y, coefficient vector w, and intercept b are as follows:

[0045] ,

[0046] ,

[0047] ,

[0048] ,

[0049] ,

[0050] In the formula For the first The moving average of the signal-to-interference-plus-noise ratio of the records. For the first The moving average of the squared signal-to-interference-plus-noise ratio of the records. For the first The moving average of the cube of the signal-to-interference-plus-noise ratio of the records. For the first The moving average of the packet reception rate for each record.

[0051] Steps S1-7: Based on the test data subset, test the regression model with parameters as coefficient vector w and intercept b, and calculate the mean squared error (MSE) and coefficient of determination. As a performance metric for the model, namely the mean squared error (MSE) and This is used to measure the accuracy of estimating the quality of inter-node links within a region based on the calculated coefficient vector w and intercept b.

[0052] Steps S1-1 to S1-7 above constitute the offline training phase.

[0053] Step S2: Any node I in the wireless sensor network calculates input features based on the real-time collected channel signal strength and the average received signal strength corresponding to any neighboring node J. Combining this with the coefficient vector w and intercept b of the region where node I is located, the current link quality between node I and its neighboring node J is calculated. This coefficient vector w and intercept b represent the relationship between the moving average of packet reception rate and the moving average of the signal-to-dryness ratio (SDR) and its square and cube, and therefore can be used to estimate the link quality between wireless sensor network nodes in the corresponding region. Specifically, this includes the following sub-steps:

[0054] Step S2-1: Any node I in the wireless sensor network periodically collects channel signal strength in a non-transmission state and stores the L most recently collected channel signal strengths as the latest channel signal strength set.

[0055] Step S2-2: When node I receives a data packet from any neighbor node J, it calculates and updates the average value of the received signal strength of the signal received by node I from neighbor node J, which is used as the average value of the received signal strength corresponding to node J.

[0056] Step S2-3: Based on the latest channel signal strength set and the average received signal strength corresponding to node J, calculate the estimated values ​​of the most recent L signal-to-interference-plus-noise ratios (SIRs) of the signals received by node I from neighboring node J. Based on the L estimated SIRs, calculate the squares and cubes of the corresponding L estimated SIRs.

[0057] The formula for calculating the estimated signal-to-interference-plus-noise ratio is as follows:

[0058] ,

[0059] In the formula Let be the estimated value of the signal-to-interference-plus-noise ratio (SIR) of node I when it receives the signal from its neighboring node J. Let J be the average received signal strength of the signal received by node I from its neighboring node J. The signal strength of the i-th channel collected for node I.

[0060] Step S2-4: Based on the L estimated values ​​of SNR, the squared values ​​of the L estimated values ​​of SNR, and the cubic values ​​of the L estimated values ​​of SNR, with L as the sliding window length, calculate the moving average corresponding to the estimated values ​​of SNR, the squared values ​​of the estimated values ​​of SNR, and the cubic values ​​of the estimated values ​​of SNR, and use them as input features.

[0061] The formula for calculating the input features is as follows:

[0062] ,

[0063] = ,

[0064] = ,

[0065] In the formula The length of the sliding window. , Let be the estimated value of the signal-to-interference-plus-noise ratio (SIR) for the ni-th node when node I receives the signal from its neighbor node J. Let n be the moving average of the estimated signal-to-interference-plus-noise ratio (SIR) when node I receives the signal from neighboring node J. Let n be the moving average of the squared estimates of the signal-to-interference-plus-noise ratio (SIR) when node I receives signals from its neighboring node J. The nth signal-to-interference-plus-noise ratio (SIR) estimate is the cube moving average of the estimated values ​​when node I receives signals from neighboring node J.

[0066] Step S2-5: Construct an input feature vector based on the input features, and combine it with the coefficient vector w and intercept b of the region where node I is located to calculate the estimated value of the moving average of the packet reception rate of node I when the neighbor node J sends the packet.

[0067] The formula for calculating the estimated moving average of packet reception rate is as follows:

[0068] ,

[0069] ,

[0070] In the formula For the input feature vector, Let n be the moving average of the estimated signal-to-interference-plus-noise ratio (SIR) when node I receives the signal from neighboring node J. Let n be the moving average of the squared estimates of the signal-to-interference-plus-noise ratio (SIR) when node I receives signals from its neighboring node J. Let n be the cube moving average of the estimated signal-to-interference-plus-noise ratio (SIR) when node I receives signals from neighboring node J. This is the transpose of the coefficient vector w. This is an estimate of the moving average of the packet reception rate of node I when neighbor node J sends a packet.

[0071] In step S2-6, node I sends the estimated value of the moving average of the packet reception rate as the current link quality between node I and neighbor node J, so that neighbor node J can obtain the current link quality between node I and node I.

[0072] The methods for estimating the moving average of the packet reception rate include: Node I periodically broadcasting a control message containing the estimated moving average of the packet reception rate; or the network using an acknowledgment message (ACK), where Node I, after receiving each data packet, includes the estimated moving average of the packet reception rate in the ACK message and sends it to the neighboring node J that sent the data packet; or the network having a request-response mechanism, where Node I includes the estimated moving average of the packet reception rate in a CTS message and sends it to the neighboring node J that issued the RTS message; or, knowing the data transmission period of the neighboring node J, Node I sends the estimated moving average of the packet reception rate to the neighboring node J before the next data packet transmission by the neighboring node J.

[0073] Steps S2-1 to S2-6 above constitute the online phase.

[0074] The role and effect of the embodiments

[0075] According to the wireless sensor network link quality regression estimation method involved in this embodiment, on the one hand, the signal-to-interference-plus-noise ratio (SINR) and its squared and cubic moving averages are used as input features. A linear regression model is used to estimate the moving average of the packet reception rate, which can achieve high estimation accuracy and low computational complexity. On the other hand, by collecting the SINR in real time through the receiving node, without depending on the packet reception process, the current packet reception rate can be estimated in a timely and accurate manner even when the data transmission interval is large. In summary, this method can estimate the current link quality in a timely and accurate manner with low computational complexity.

[0076] The above embodiments are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention.

Claims

1. A regression estimation method for link quality in wireless sensor networks, characterized in that, Includes the following steps: Step S1: Use a test system to collect data from each region of the wireless sensor network, and calculate the coefficient vector w and intercept b of each region based on the collected data. Step S2: Any node I in the wireless sensor network calculates input features based on the real-time collected channel signal strength and the average received signal strength corresponding to any neighboring node J. Combining the coefficient vector w and the intercept b of the region where node I is located, the current link quality between node I and its neighboring node J is calculated. Step S1 includes the following sub-steps: Step S1-1: Divide the deployment space of the wireless sensor network into the various regions, and use the test system to collect data in each region. The test system includes multiple transmitting nodes, one receiving node, and one computer. Step S1-2: In each region, the transmitting node of the test system sends test packets to the receiving node with a period of T1. The receiving node collects the channel signal strength with a period of T2 in the non-transmission state, where T2 is less than T1. The channel signal strength and the received signal strength when the receiving node receives the test packet are added to the received test packet, and then the test packet is sent to the computer. Steps S1-3: Based on all the test packets received within the region, construct a system with... The tracking dataset for this region contains records, including original records and filled records; Steps S1-4: Based on the records in the tracking dataset, calculate the signal-to-interference-plus-noise ratio (SIR), the square of the SIR, the cube of the SIR, and the moving average corresponding to the packet reception rate, and add each of the moving averages to the corresponding record in the tracking dataset. Steps S1-5: Randomly select from the tracking dataset 1 record, constituting a subset of the training data, the remaining These records are used as a subset of the test data; Steps S1-6: Construct an input feature matrix X and an output vector y based on the moving average of each item in the training data subset; calculate the coefficient vector w and the intercept b of the region based on the input feature matrix X and the output vector y. Steps S1-7: Based on the subset of test data, test the regression model with parameters w as the coefficient vector and b as the intercept, and calculate the mean squared error (MSE) and coefficient of determination. As a performance metric of the model Step S2 includes the following sub-steps: Step S2-1: In the non-transmission state, any node I of the wireless sensor network periodically collects the channel signal strength and stores the L most recently collected channel signal strengths as the latest channel signal strength set. Step S2-2: When node I receives a data packet from any neighbor node J, it calculates and updates the average value of the received signal strength of the signal received by node I from the neighbor node J, and uses it as the average value of the received signal strength corresponding to node J. Step S2-3: Based on the latest channel signal strength set and the average received signal strength corresponding to node J, calculate the estimated values ​​of the most recent L signal-to-interference-plus-noise ratios of the signals received by node I from neighboring node J. Based on the L estimated values, calculate the square value and the cube value of the corresponding estimated values ​​respectively. Step S2-4: Based on L estimated values, L squared values, and L cubic values, with L as the sliding window length, calculate the moving average value corresponding to the estimated value, the squared value, and the cubic value, and use it as the input feature; Step S2-5: Construct an input feature vector based on the input features, and combine the coefficient vector w and the intercept b of the region where node I is located to calculate the estimated value of the moving average of the packet reception rate of node I when the neighbor node J sends the packet. In step S2-6, node I sends the estimated value of the moving average of the packet reception rate as the current link quality between node I and neighbor node J, so that neighbor node J can obtain the current link quality of the link between node I and node I.

2. The wireless sensor network link quality regression estimation method according to claim 1, characterized in that: in, In steps S1-3, each original record corresponds one-to-one with all the received test packets, arranged in chronological order of receipt. Each original record includes the sending node identifier, packet sequence number, received signal strength, channel signal strength, and transmission power of the corresponding test packet. The filling record corresponds one-to-one with the unreceived test packets, i.e., lost packets. The position and number of the filling records are determined based on the sending node identifier and the packet sequence number in the adjacent original records. The sending node identifier, packet sequence number, and transmission power in the filling record are determined based on the adjacent original records. The channel signal strength of the filling record is the maximum value among the channel signal strengths collected by the receiving node in the original records. The received signal strength of the filling record is the average of the received signal strengths of all the original records with the same sending node identifier and transmission power as the filling record.

3. The wireless sensor network link quality regression estimation method according to claim 1, characterized in that: in, In steps S1-4, the formula for calculating the signal-to-interference-plus-noise ratio is: ， The formula for calculating the moving average of each term is: ， ， = ， = ， In the formula Let i be the signal-to-interference-plus-noise ratio. Let i be the strength of the received signal. Let i be the signal strength of the i-th channel. , To track the total number of original records and filler records in the dataset, Let n be the moving average of the reception rate of the nth packet. The length of the sliding window. The number of original records in the sliding window, when hour, To track the number of original records from record 1 to record n ​​in the dataset, when hour, To track the number of original records in K consecutive records starting from n-K+1 in the dataset, Let n be the moving average of the signal-to-interference-plus-noise ratio (SIR). Let n be the moving average of the squared signal-to-interference-plus-noise ratio (SIR). Let be the moving average of the cube of the nth signal-to-interference-plus-noise ratio. Calculate the signal-to-interference-plus-noise ratio (SIR) of all records in the tracking data, select the minimum value among them to replace the SIR of all the filling records, and calculate the SIR, the square of the SIR, and the moving average of the cube of the SIR of the filling records based on the replaced SIR.

4. The wireless sensor network link quality regression estimation method according to claim 1, characterized in that: in, In steps S1-5, , The value range is 0.6 to 0.

75.

5. The wireless sensor network link quality regression estimation method according to claim 1, characterized in that: in, In steps S1-6, the formulas for calculating the feature matrix X, the output vector y, the coefficient vector w, and the intercept b are as follows: ， ， ， ， ， In the formula For the first The moving average of the signal-to-interference-plus-noise ratio of the records. For the first The moving average of the squared signal-to-interference-plus-noise ratio of the records. For the first The moving average of the cube of the signal-to-interference-plus-noise ratio of the records. For the first The moving average of the packet reception rate for each record.

6. The wireless sensor network link quality regression estimation method according to claim 1, characterized in that: in, In steps S2-3, the formula for calculating the estimated signal-to-interference-plus-noise ratio is as follows: ， In the formula Let be the estimated value of the signal-to-interference-plus-noise ratio (SIR) of node I when it receives the signal from its neighboring node J. Let J be the average received signal strength of the signal received by node I from its neighboring node J. The signal strength of the i-th channel collected for node I.

7. The wireless sensor network link quality regression estimation method according to claim 1, characterized in that: in, In steps S2-4, the calculation formula for the input features is as follows: ， = ， = ， In the formula The length of the sliding window. , Let be the estimated value of the signal-to-interference-plus-noise ratio (SIR) for the ni-th node when node I receives the signal from its neighbor node J. Let n be the moving average of the estimated signal-to-interference-plus-noise ratio (SIR) when node I receives the signal from neighboring node J. Let n be the moving average of the squared estimates of the signal-to-interference-plus-noise ratio (SIR) when node I receives signals from its neighboring node J. The moving average of the cube of the estimated signal-to-interference-plus-noise ratio (SIR) when node I receives a signal from neighboring node J is denoted as ...

8. The wireless sensor network link quality regression estimation method according to claim 1, characterized in that: in, In steps S2-5, the formula for calculating the estimated value of the moving average of the packet reception rate is as follows: , , In the formula For the input feature vector, Let n be the moving average of the estimated signal-to-interference-plus-noise ratio (SIR) when node I receives the signal from neighboring node J. Let n be the moving average of the squared estimates of the signal-to-interference-plus-noise ratio (SIR) when node I receives signals from its neighboring node J. Let n be the cube moving average of the estimated signal-to-interference-plus-noise ratio (SIR) when node I receives signals from neighboring node J. This is the transpose of the coefficient vector w. This is an estimate of the moving average of the packet reception rate of node I when neighbor node J sends a packet.

9. The wireless sensor network link quality regression estimation method according to claim 1, Its features are: The method for sending an estimate of the moving average of the packet reception rate in steps S2-6 includes: Node I periodically broadcasts a control message containing an estimate of the sliding average of the packet reception rate; or The network uses an acknowledgment message (ACK). After receiving a data packet, node I includes an estimate of the sliding average of the packet reception rate in the ACK message and sends it to the neighboring node J that sent the data packet; or... The network has a request-response mechanism, whereby node I includes an estimate of the moving average of the packet reception rate in a CTS message and sends it to the neighboring node J that issued the RTS message; or Knowing the data transmission period of the neighbor node J, before the neighbor node J sends the next data packet, the node I sends an estimate of the sliding average of the packet reception rate to the neighbor node J.

Images