Multi-dimensional radio frequency fingerprint enhancement method based on discrete wavelet transform

Abstract

The invention discloses a multi-dimensional radio frequency fingerprint enhancement method based on discrete wavelet transform. The method comprises the following steps of: 1) weighting and calculating the positions of all APs (Access Points) by utilizing the positions of known fingerprint acquisition reference points and RSSI (Received Signal Strength Indicator) received by the corresponding reference points; 2) arranging the collected RSSI sequences according to the AP position information obtained in the step 1) to form a multi-dimensional fingerprint sequence, and preliminarily enhancing an original fingerprint; 3) converting the multi-dimensional fingerprint sequence in the step 2) into a grey-scale map, and decomposing the grey-scale map by using discrete wavelet transform to obtain a detail coefficient and an approximation coefficient; 4) exchanging detail coefficients of the fingerprint grey-scale maps collected at different positions or in different time periods within the same reference point range; and 5) carrying out wavelet inverse transformation on the exchanged detail coefficient and the non-exchanged approximation coefficient in the step 4) to obtain a final enhanced fingerprint. The method is simple in design, reduces the dependence of the positioning model on the number of label fingerprints, improves the positioning classification accuracy, enhances the robustness of the positioning model, and has flexible practical value and practical popularization significance.

Description

Technical Field

[0001] The invention belongs to the field of indoor positioning and electronic information science, in particular to wavelet transform enhanced radio frequency fingerprint. Background Art

[0002] Wi-Fi fingerprint positioning is often used as the preferred solution for indoor positioning because of its simple and feasible positioning framework, which does not require the additional deployment of positioning beacons and signal receiving equipment. However, in practical applications, Wi-Fi fingerprint positioning requires the collection of fingerprints of known locations as reference fingerprints in advance and stores them in the fingerprint library. Then, the real-time fingerprint is compared with the fingerprint in the fingerprint library to determine the current location. The process of building a fingerprint library is a tedious task, which requires not only collecting fingerprints but also measuring the location information of the collection points. Currently, the burden of fingerprint collection can be reduced through fingerprint enhancement, which is mainly divided into feature extraction type and fingerprint generation type.

[0003] The feature extraction type achieves the purpose of fingerprint enhancement by constructing a feature extraction model with better performance to extract deeper features. Lan T. et al. (Lan T., Wang X., Chen Z., et al, Fingerprint AugmentBased on Super-Resolution for WiFi Fingerprint Based Indoor Localizationz[J]. IEEE Sensors Journal, 22(12): 12152–12162, 2022.) proposed a fingerprint enhancement framework that transforms the fingerprint enhancement problem into a super-resolution problem of fingerprint images. However, the performance of the enhancement framework is closely related to many parameters, such as different reference point distributions and channel characteristics in different environments. Zhang B. et al. (Zhang B., Sifaou H., and Li Y., CSI-Fingerprinting Indoor Localization via Attention-Augmented Residual Convolutional Neural Network[J]. IEEE Transactions on Wireless Communications, 22(8): 5583–5597, 2023.) proposed a novel convolutional neural network with attention-augmented residual to fully utilize the spatiotemporal features in the channel state information to improve the performance of the positioning network.

[0004] The fingerprint generation type constructs new fingerprints through existing fingerprint distribution and other information to expand the fingerprint library. Junoh S. et al. (Junoh S. and Pyun J., Enhancing Indoor Localization with Semi-Crowdsourced Fingerprinting and GAN-Based Data Augmentation [J]. IEEE Internet Things Journal, 11 (7): 11945–11959, 2024.) introduced a method based on generative adversarial neural networks, increasing the amount of training data collected at each reference point to enrich the original data set and reduce manpower investment. Ssekidde P. et al. (Ssekidde, P. Eyobu O., Han D., et al., Augmented CWT Features for Deep Learning-Based Indoor Localization Using WiFi RSSI Data [J]. Applied Sciences, 11 (4): 1806, 2021.) proposed a new feature set for extracting RSSI data using continuous wavelet transform, and used deep learning methods for indoor positioning. Masato Sugasak proposed a new data enhancement method for indoor positioning based on inter-class learning, focusing on the positional relationship between each sampled data at the unsampled location, and constructing a generative model of the Wi-Fi signal of each access point through sparse data, and then forming a synthetic augmented fingerprint dataset. Numan P. et al. (Numan P., Park H., Laoudias C., et al, Dropout Autoencoder Fingerprint Augmentation for EnhancedWi-Fi FTM-RSS Indoor Localization[J]. IEEE Communications Letters, 27(7): 1759–1763, 2023.) proposed a fingerprint enhancement method based on random dropout autoencoder, which significantly improved the accuracy of Wi-Fi indoor positioning by reconstructing unrecorded signal features and applying deep neural network regression.

[0005] It can be seen that although some researchers have used continuous wavelet transform to extract fingerprint features, the continuous wavelet transform process is more complicated than discrete wavelet transform. With the advent of the deep learning wave, most current studies on Wi-Fi fingerprint enhancement are mainly based on deep learning and other similar schemes to learn fingerprint distribution in order to expand the fingerprint library. However, due to the quality of the training sample order of magnitude, the fingerprint distribution extracted by the deep learning model is not necessarily accurate, and the generated new fingerprint may not be conducive to the accuracy of positioning. At the same time, the characteristic distribution of fingerprints changes dynamically, and new samples generated only based on the current fingerprint may not necessarily adapt to the changed environment. In addition, the deep learning method has other disadvantages, such as long training time and large differences between the generated fingerprint and the actual fingerprint. Therefore, designing a lightweight enhanced fingerprint scheme that makes full use of the similarities and differences in the spatiotemporal characteristics of fingerprint samples to adapt to dynamic environmental changes and improve the accuracy of positioning classification is a problem that needs to be solved urgently. Summary of the invention

[0006] In order to solve the problems existing in the prior art, the present invention provides a multi-dimensional radio frequency fingerprint enhancement method based on discrete wavelet transform, which can reduce the dependence of the positioning model on the tag fingerprint and further improve the accuracy of Wi-Fi fingerprint positioning classification.

[0007] The present invention is achieved through the following technical solutions.

[0008] The present invention provides a multi-dimensional radio frequency fingerprint enhancement method based on discrete wavelet transform, comprising the following steps:

[0009] Step 1: Use the known fingerprint collection reference point location and the RSSI (Received Signal Strength Indication) received by the corresponding reference point to weight the location of all APs (Access Points). If the location of nr reference points is {P ri |0 <i≤nr,i∈N +}, the location of the AP can be expressed as:

[0010]

[0011] in, is the estimated position of the jth AP, P rk is the kth known reference point position, w jk Indicates that the jth AP is at the kth reference point P rk The weighted value of It represents the average received signal strength of the j-th AP at the m-th reference point.

[0012] Step 2: Arrange the collected RSSI sequence according to the AP location information obtained in step 1 to form a multi-dimensional fingerprint sequence, preliminarily enhance the original fingerprint, and thus enrich the spatial characteristics of the fingerprint data. The specific construction process is as follows: (1) Receive the signal strength of na APs for n periods at a reference point in the positioning area, and calculate the average signal strength of different APs in a single period; sort them alternately in the order of maximum, minimum, second largest, second smallest, etc. to form a 1-dimensional feature of the positioning fingerprint. (2) According to the estimated three-dimensional position of the AP, sort the APs in the horizontal direction (left-to-right direction), vertical direction (up-and-down direction), and depth direction (front-to-back direction) respectively, so that the other 3-dimensional features of the fingerprint can be obtained. Finally, merge the multi-dimensional feature dimensional sequence to obtain a fingerprint array of size 4×na.

[0013] Step 3: Normalize the multi-dimensional fingerprint sequence values ​​in step 2 to between 0 and 255 and save it as a grayscale image, such as Figure 1 Then the grayscale image is decomposed by discrete wavelet, as shown in Figure 2 shown.

[0014] If the dimension of the grayscale image f(x,y) is M×N, the corresponding discrete wavelet decomposition expression is:

[0015]

[0016] Among them, CA(j0,m,n) is the approximate coefficient, CD l (j,m,n) is the detail coefficient, l∈{H,V,D}, j0 is the scale at which the decomposition starts, and the present invention selects j0=0; the size of the fingerprint is 4×na, so M=4, N=n a ; j = 0, 1, ... min (1, log2n a -1), m=n=0,1,…,2 j -1.

[0017] is a two-dimensional scaling function:

[0018]

[0019] is a two-dimensional wavelet function:

[0020]

[0021] ψ D (x,y)=ψ(x)ψ(y)

[0022] in, is the scaling function, and ψ(x) is the wavelet function.

[0023] Step 4: Exchange the detail coefficients of the fingerprint grayscale images collected at different locations or different time periods within the same reference point range. If the set of detail coefficients after fingerprint decomposition is:

[0024] CD=[CD H ,CD V ,CD D ]

[0025] Then the entire coefficient set can be expressed as:

[0026] SC=[CD,CA].

[0027] (1) Exchange of detail coefficients of fingerprint grayscale images at different time periods:

[0028] Assume that the fingerprint coefficient sets collected at different times t1 and t2 within the same reference point are:

[0029]

[0030] Then the detail coefficients are exchanged:

[0031]

[0032] in, express Medium detail factor Replaced by

[0033] (2) Exchange of detail coefficients of fingerprint grayscale images at different positions within the same reference point range:

[0034]

[0035] in, It represents the coefficient set after the fingerprint detail coefficient at position p1 is swapped with the fingerprint detail coefficient at position p2.

[0036] Step 5: The detail coefficient set SC exchanged in step 4 is transformed by inverse discrete wavelet transform to obtain the final enhanced fingerprint grayscale image, as shown in Figure 3 shown.

[0037] Inverse discrete wavelet transform:

[0038]

[0039] Among them, f R (x,y) represents the final enhanced fingerprint grayscale image. is the replaced two-dimensional detail coefficient, l∈{H,V,D}, and the rest of the symbols are the same as those of the wavelet forward transform.

[0040] The present invention provides a fingerprint enhancement method based on wavelet transform, which first uses the virtual position of Wi-Fi AP to embed spatial information into the fingerprint sequence, and expands the one-dimensional fingerprint sequence into a multi-dimensional fingerprint grayscale image; then based on discrete wavelet transform, different spatiotemporal detail coefficients in the same category are exchanged, thereby expanding the fingerprint library. The Wi-Fi fingerprint enhancement scheme based on wavelet transform proposed in the present invention can effectively reduce the dependence on label fingerprints without significantly increasing the time complexity, is suitable for dynamic environments, and improves positioning accuracy. The present invention is simple in design, reduces the dependence of the positioning model on the number of label fingerprints, improves the accuracy of positioning classification, and enhances the robustness of the positioning model. It has flexible practical value and practical promotion significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is the multi-dimensional fingerprint grayscale image in the present invention.

[0042] Figure 2 It is the discrete wavelet decomposition of the multi-dimensional fingerprint grayscale image in the present invention.

[0043] Figure 3 The enhanced fingerprint grayscale image based on discrete wavelet transform in the present invention, wherein (a) is the fingerprint grayscale image before enhancement, and (b) is the fingerprint grayscale image after enhancement.

[0044] Figure 4 This is a diagram of the CNN structure used for classification prediction in the present invention.

[0045] Figure 5 Schematic diagram of the original 1D fingerprint replication and expansion in the present invention.

[0046] Figure 6 These are the test results of 1D fingerprint and 4D fingerprint in the present invention.

[0047] Figure 7 This is a histogram of the multi-dimensional fingerprint enhancement test results based on discrete wavelet transform in the present invention.

[0048] Figure 8 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0049] In order to verify the feasibility and effectiveness of the present invention, the present invention is further described in detail below. The specific embodiments described herein are only used to explain the present invention and are not used to limit the invention.

[0050] The present invention uses the open source Wi-Fi indoor UJIIndoorLoc dataset (denoted as dataset U) and the self-built dataset (denoted as dataset S) for verification testing. Dataset U has a total of 3 buildings with a total of 520 AP point data. In order to maintain generality, the present invention selects the data of BUILDINGID 0 and FLOOR 0 for data enhancement testing, and the original data totals 1059. However, in some locations, all APs have no signal or some APs have no signal at all locations. By filtering out the above invalid positions and invalid APs, and then using the AP coverage range (the number of reference points that can receive the AP signal) as the standard, the top 16 AP signals are selected as the test fingerprint library, and a total of 246 fingerprint data from 14 locations remain. Dataset S collects fingerprint data from 10 reference locations in total, which are collected in 2 batches, 200 fingerprint data in the morning (denoted as T0 domain) and 200 fingerprint data in the afternoon (denoted as T1 domain).

[0051] The present invention performs all data processing and testing on a desktop computer equipped with an Intel Core i7-13700KF CPU and an NVIDIARTX4080 GPU. Among them, discrete wavelet transform enhancement is completed by the wavelet analysis toolbox (Wavelet Toolbox) in MATLAB2020b software, and a convolutional neural network (CNN) is constructed using the PyTorch framework (python 3.9) as a classification prediction model and classification prediction is performed using enhanced fingerprints. Among them, CNN includes two convolutional layers and two fully connected layers. First, the input single-channel image passes through the first convolutional layer (Conv1), which increases the number of input channels from 1 to 16, and performs a convolution operation through a 3x3 convolution kernel, keeping the output size unchanged. Then, downsampling is performed through a maximum pooling layer (P1) to reduce the size of the feature map. Then, through a second convolutional layer (Conv2), the number of channels is increased to 32, and again through a 3x3 convolution kernel operation, the output size is maintained, and then the maximum pooling (P2) downsampling is performed again. After the feature map is flattened into a one-dimensional vector, it is input into the first fully connected layer (Fc1), and then passes through the second fully connected layer (Fc2), and finally outputs the classification result. The CNN structure is as follows Figure 4 shown.

[0052] During the training process, we set the maximum number of training rounds to 500, the initial learning rate to 0.001, the batch size to 4, and used the Adam optimizer for training. Accuracy was used as an evaluation indicator for location recognition classification. In the test, the Haar wavelet was initially selected for fingerprint enhancement. To simulate the situation of insufficient fingerprint data, the samples were divided into training set and test set at a ratio of 2:8. In the following embodiments, unless otherwise specified, the implementation test conditions are consistent with the above-described conditions.

[0053] Embodiment 1: Verification test of multi-dimensional fingerprint.

[0054] Table 1 Comparison of classification accuracy between 1D fingerprint and 4D fingerprint

[0055]

[0056] In order to verify the effectiveness of the multi-dimensional fingerprint designed by the present invention, tests were conducted on datasets U and S, and compared with the commonly used 1D fingerprint (1D fingerprint). However, the CNN used in the present invention is a 2D input, and 1D fingerprints cannot be directly input. In order to reasonably compare the 1D fingerprint with the multi-dimensional fingerprint (4D fingerprint) proposed by the present invention, the corresponding original 1D fingerprint is directly copied and expanded into 4D (such as Figure 5 As shown in Figure 2), it is used as CNN input. The test results of 1D fingerprint and 4D fingerprint are shown in Figure 2. Figure 6 As shown in Table 1. It can be seen that the classification accuracy of the 4D fingerprint proposed by the present invention is significantly improved compared with that of the 1D fingerprint. Since the 4D fingerprint incorporates the three-dimensional spatial information of the AP point, it is conducive to the recognition of the positioning model. Compared with the original 1D fingerprint, the classification accuracy of the 4D fingerprint designed by the present invention is improved by 19.19% in the data set U and 13.85% in the data set S. The test results verify the effectiveness of the multi-dimensional fingerprint designed by the present invention.

[0057] Embodiment 2: Verification test of multi-dimensional fingerprint enhancement based on discrete wavelet transform.

[0058] In the test, all samples are data set S, of which 20% of T0 domain fingerprints are used as training sets, and the training samples of the training set are enhanced based on discrete wavelet transform using a 4D fingerprint for each reference point in the T1 domain, and the test samples are 80% of the T1 domain fingerprints. For comparison, the T1 domain samples selected for discrete wavelet enhancement are directly added to the original 1D fingerprint library of the T0 domain (which has been copied and expanded to 4D) and the T0 domain 4D fingerprint library that has not been enhanced by discrete wavelet, and are recorded as 1DF and 4DF respectively. At the same time, due to the variety of wavelet basis functions of discrete wavelet transform, the present invention selects commonly used wavelet basis functions such as haar, dbN (N=1,2,3,4), symN (N=1,2,3,4), coifN (N=1,2,3,4) as enhancement comparisons, and the classification accuracy is as follows: Figure 7 And as shown in Table 2.

[0059] Table 2 Comparison of classification accuracy of different wavelet bases

[0060]

[0061] As can be seen from Table 2, the classification accuracy of discrete wavelet transform enhancement based on coif3, haar, coif4 and sym3 has reached more than 90%, which is a maximum improvement of 36.14% compared to 1D fingerprint (1DF) and a maximum improvement of 24.09% compared to 4D fingerprint (4DF). The classification accuracy of 4DF is the second lowest, only 76.25%. The reason is that 4D fingerprint only enhances the fingerprint features of the T0 domain. Although a small amount of the latest T1 domain fingerprints (1 new fingerprint for 1 reference point) are introduced, its features are still far from the T1 domain. Despite this, the classification accuracy of 4DF is still higher than that of 1DF, because 4D fingerprint introduces the spatial position relationship of AP, and the AP position is usually fixed, which can alleviate the dynamic changes of fingerprints to a certain extent. The above tests can prove that the multi-dimensional radio frequency fingerprint enhancement method based on discrete wavelet transform designed by the present invention is effective.

Claims

1. A multi-dimensional radio frequency fingerprint enhancement method based on discrete wavelet transform, characterized in that The following steps are involved: Step 1: Use the known fingerprint collection reference point position and the RSSI received at the corresponding reference point to weight the position of all APs; if the position of nr reference points is {P ri |0 <i≤nr,i∈N + }, the location of the AP can be expressed as: in, is the estimated position of the jth AP, P rk is the kth known reference point position, w jk Indicates that the jth AP is at the kth reference point P rk The weighted value of represents the average received signal strength of the j-th AP at the m-th reference point; Step 2: Arrange the collected RSSI sequence according to the AP location information obtained in step 1 to form a multi-dimensional fingerprint sequence, preliminarily enhance the original fingerprint, and thus enrich the spatial characteristics of the fingerprint data; The construction process is as follows: (1) Receive the signal strengths of na APs for n periods at a certain reference point in the positioning area, and calculate the average signal strength of different APs in a single period; sort them alternately in the order of largest, smallest, second largest, second smallest, etc. to form a 1D feature of the positioning fingerprint; (2) According to the estimated 3D position of the AP, sort the APs in the horizontal, vertical, and depth directions respectively, so that the other 3D features of the fingerprint can be obtained; finally, merge the multi-dimensional feature dimension sequence to obtain a fingerprint array of size 4×na; Step 3, normalize the multi-dimensional fingerprint sequence value in step 2 to between 0 and 255, save it as a grayscale image, and then perform discrete wavelet decomposition on the grayscale image; If the dimension of the grayscale image f(x,y) is M×N, the corresponding discrete wavelet decomposition expression is: Among them, CA(j0,m,n) is the approximate coefficient, CD l (j,m,n) is the detail coefficient, l∈{H,V,D}, j0 is the scale at which the decomposition starts, and the present invention selects j0=0; the size of the fingerprint is 4×na, so M=4, N=n a ; j = 0, 1, ... min (1, log2n a -1), m=n=0,1,…,2 j -1; is a two-dimensional scaling function: is a two-dimensional wavelet function: ψ D (x,y)=ψ(x)ψ(y) in, is the scaling function, ψ(x) is the wavelet function; Step 4: Exchange the detail coefficients of the fingerprint grayscale images collected at different positions or different time periods within the same reference point range; if the set of detail coefficients after fingerprint decomposition is: CD=[CD H ,CD V ,CD D ] Then the entire coefficient set can be expressed as: SC = [CD, CA]; (1) Exchange of detail coefficients of fingerprint grayscale images at different time periods: Assume that the fingerprint coefficient sets collected at different times t1 and t2 within the same reference point are: Then the detail coefficients are exchanged: in, express Medium detail factor Replaced by (2) Exchange of detail coefficients of fingerprint grayscale images at different positions within the same reference point range: in, The coefficient set representing the fingerprint detail coefficient at position p1 is swapped with the fingerprint detail coefficient at position p2; Step 5, the detail coefficient set SC exchanged in step 4 is subjected to inverse discrete wavelet transform to obtain the final enhanced fingerprint grayscale image; Inverse discrete wavelet transform: Among them, f R (x,y) represents the final enhanced fingerprint grayscale image. is the replaced two-dimensional detail coefficient, l∈{H,V,D}, and the rest of the symbols are the same as those of the wavelet forward transform.

Images