Construction method of anti-delay cross-protocol communication decoder based on DANN network architecture

Abstract

The present application relates to a method for constructing a delay-resistant cross-protocol communication decoder based on a DANN network architecture. By utilizing the characteristic that a domain-adaptive neural network can migrate the feature space mapping of source domain data and target domain data, a coding sequence containing delay is used as target domain data, and a coding sequence without delay is used as source domain data, thereby eliminating the difference in the distribution of spatial features between the two. Finally, a decoder is designed that can extract information from a coding sequence containing delay interference. This decoder improves the reliability and stability of packet-level CTC technology deployed in a multi-device environment.

Description

Technical Field

[0001] The present application relates to the field of communications, and in particular, to a method for constructing a delay-resistant cross-protocol communication decoder based on a DANN network architecture. Background Art

[0002] With the rapid development of the Internet of Things (IoT), packet-level cross-protocol communication technology has been researched and developed due to its flexibility and high compatibility, thereby reducing the cost of communication between heterogeneous devices in coexisting environments and becoming widely used in smart home environments. However, when existing packet-level Cross-Technology Communication (CTC) methods are actually deployed on commercial smart home devices, the CSMA / CA protocol followed by the devices' wireless protocols uses a series of mechanisms to coordinate the data transmission timing between different devices to prevent collisions and data loss and improve network transmission reliability. Since packet-level data packets express information through specific packet timing, the behavior of the CSMA / CA protocol seriously disrupts the orderliness of the packet-level CTC encoding sequence, resulting in loss of information expression. This makes existing packet-level CTC methods ineffective for device information transmission and control in real-world applications, reducing their usability.

[0003] Therefore, for packet-level CTC methods to truly be applied in a multi-protocol home environment, resolving the latency caused by the CSMA / CA protocol and restoring CTC communication reliability presents a significant challenge. While most existing packet-level CTC methods account for coexistence interference in the same environment—that is, packets from other devices can be inserted into the CTC code sequence, disrupting its regularity—they ignore the impact of latency caused by the CTC transmitter's own protocol in a multi-device environment. This latency can also interfere with the receiver's interpretation of the information. Summary of the Invention

[0004] In order to overcome at least one deficiency in the prior art, the present application provides a method for constructing a delay-resistant cross-protocol communication decoder based on a DANN network architecture.

[0005] In a first aspect, a method for constructing a delay-resistant cross-protocol communication decoder based on a DANN network architecture is provided, comprising:

[0006] Preprocessing the RSSI code sequence obtained by the receiving end to obtain a preprocessed RSSI code sequence; the preprocessing includes code segment segmentation and sequence interference elimination;

[0007] The pre-processed RSSI coding sequence is segmented to obtain multiple sequence segments;

[0008] Label each sequence segment and construct training data based on the labeled sequence segments; the labels include delay labels and symbol labels;

[0009] Training the anti-delay cross-protocol communication decoder based on the DANN network architecture based on the training data to obtain a trained anti-delay cross-protocol communication decoder based on the DANN network architecture;

[0010] The delay-resistant cross-protocol communication decoder based on the DANN network architecture includes a feature extractor, a spatial pooling layer, a symbol classifier and a delay classifier; the feature extractor is used to extract features from the training data to obtain a feature vector; the spatial pooling layer is used to perform a maximum pooling operation on the feature vector to obtain a pooled feature vector; the symbol classifier is used to output the symbol category of the sequence segment based on the pooled feature vector, and the delay classifier is used to output the delay category of the sequence segment based on the pooled feature vector.

[0011] In one embodiment, code segment segmentation includes:

[0012] Delete the sampling points with a value less than -75dBm in the RSSI code sequence to obtain a valid RSSI code sequence; the valid RSSI code sequence includes multiple RSSI segments;

[0013] Sequence interference removal includes:

[0014] Determine the characteristic standard value Fea1;

[0015] Calculate the characteristic value Fea2 of each RSSI segment in the RSSI encoding sequence, Fea2 = f1 ± ξ1, f1 is the fingerprint feature group, including the energy variance, energy size and duration of the RSSI segment, ξ1 is the maximum allowed error;

[0016] If the characteristic value Fea2 of the RSSI segment is not equal to the characteristic standard value Fea1, the RSSI segment is deleted to obtain a preprocessed RSSI code sequence.

[0017] In one embodiment, the symbol classifier and the delay classifier have the same network structure but different network parameters. The network structure includes a fully connected layer, a one-dimensional regularization layer, a fully connected layer, a one-dimensional regularization layer, and a fully connected layer.

[0018] In one embodiment, the loss function used in the training process is expressed by the following formula:

[0019] E=Loss y +Loss d

[0020] Among them, E is the loss function, Loss y is the symbol loss, Lossd To determine the loss for delay;

[0021] Symbol loss y The symbol category of the sequence segment based on the output of the symbol classifier and the symbol label of the sequence segment are obtained through the cross entropy function; the delayed discrimination loss Loss d The delay category of the sequence segment based on the output of the delay classifier and the delay label of the sequence segment are obtained through the cross entropy function.

[0022] In a second aspect, a device for constructing a delay-resistant cross-protocol communication decoder based on a DANN network architecture is provided, comprising:

[0023] The preprocessing module is used to preprocess the RSSI code sequence obtained by the receiving end to obtain a preprocessed RSSI code sequence; the preprocessing includes code segment segmentation and sequence interference elimination;

[0024] A sequence segment acquisition module is used to segment the pre-processed RSSI coding sequence to obtain multiple sequence segments;

[0025] The training data construction module is used to label each sequence segment and construct training data based on the labeled sequence segments; the labels include delay labels and symbol labels;

[0026] A training module is used to train the anti-delay cross-protocol communication decoder based on the DANN network architecture based on the training data to obtain a trained anti-delay cross-protocol communication decoder based on the DANN network architecture;

[0027] The delay-resistant cross-protocol communication decoder based on the DANN network architecture includes a feature extractor, a spatial pooling layer, a symbol classifier and a delay classifier; the feature extractor is used to extract features from the training data to obtain a feature vector; the spatial pooling layer is used to perform a maximum pooling operation on the feature vector to obtain a pooled feature vector; the symbol classifier is used to output the symbol category of the sequence segment based on the pooled feature vector, and the delay classifier is used to output the delay category of the sequence segment based on the pooled feature vector.

[0028] In one embodiment, code segment segmentation includes:

[0029] According to the relationship between the value of each sampling point in the RSSI code sequence and the set value, the RSSI code sequence is divided into multiple RSSI segments;

[0030] Sequence interference removal includes:

[0031] Determine the characteristic standard value Fea1;

[0032] Calculate the characteristic value Fea2 of each RSSI segment in the RSSI encoding sequence, Fea2 = f1 ± ξ1, f1 is the fingerprint feature group, including the energy variance, energy size and duration of the RSSI segment, ξ1 is the maximum allowed error;

[0033] If the characteristic value Fea2 of the RSSI segment is not equal to the characteristic standard value Fea1, the RSSI segment is deleted to obtain a preprocessed RSSI code sequence.

[0034] In one embodiment, the symbol classifier and the delay classifier have the same network structure but different network parameters. The network structure includes a fully connected layer, a one-dimensional regularization layer, a fully connected layer, a one-dimensional regularization layer, and a fully connected layer.

[0035] In one embodiment, the loss function used in the training process is expressed by the following formula:

[0036] E=Loss y +Loss d

[0037] Among them, E is the loss function, Loss y is the symbol loss, Loss d To determine the loss for delay;

[0038] Symbol loss y The symbol category of the sequence segment based on the output of the symbol classifier and the symbol label of the sequence segment are obtained through the cross entropy function; the delayed discrimination loss Loss d The delay category of the sequence segment based on the output of the delay classifier and the delay label of the sequence segment are obtained through the cross entropy function.

[0039] In a third aspect, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned method for constructing an anti-delay cross-protocol communication decoder based on the DANN network architecture.

[0040] In a fourth aspect, a computer program product is provided, comprising a computer program / instruction, which, when executed by a processor, implements the above-mentioned method for constructing an anti-delay cross-protocol communication decoder based on the DANN network architecture.

[0041] Compared with the existing technology, the present application has the following beneficial effects: the present application utilizes the characteristics of domain adaptive neural networks to migrate the feature space mapping of source domain data and target domain data, and uses the coding sequence containing delay as the target domain data and the coding sequence without delay as the source domain data, so as to eliminate the difference in the distribution of spatial features between the two. Finally, a decoder is designed that can extract information from the coding sequence containing delay interference. This decoder improves the reliability and stability of packet-level CTC technology deployed in a multi-device environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The present application may be better understood by referring to the following description in conjunction with the accompanying drawings, which together with the following detailed description are incorporated into and form a part of this specification. In the drawings:

[0043] Figure 1 A flowchart of a method for constructing a delay-resistant cross-protocol communication decoder based on a DANN network architecture according to an embodiment of the present application is shown;

[0044] Figure 2 A schematic diagram showing a method of performing segment cutting on an RSSI coding sequence to obtain multiple sequence segments is shown;

[0045] Figure 3 The structural block diagram of the delay-resistant cross-protocol communication decoder based on the DANN network architecture is shown;

[0046] Figure 4 shows the structural block diagram of the spatial pooling layer;

[0047] Figure 5 A structural block diagram of a device for constructing an anti-delay cross-protocol communication decoder based on a DANN network architecture according to an embodiment of the present application is shown;

[0048] Figure 6 Two schematic diagrams of RSSI sequence expression symbol information are shown;

[0049] Figure 7 The delay factor distribution diagrams in three environments are shown, where (a) is the delay factor distribution diagram in the conference room (Mild) environment, (b) is the delay factor distribution diagram in the corridor (Moderate) environment, and (c) is the delay factor distribution diagram in the laboratory (Sever) environment;

[0050] Figure 8 shows the spatial distribution of the encoded sequence features after decoder processing;

[0051] Figure 9 shows the spatial distribution of the encoded sequence features without decoder processing;

[0052] Figure 10 The figure shows the decoding accuracy results at different distances between the transmitter and receiver under different environments. DETAILED DESCRIPTION

[0053] Exemplary embodiments of the present application are described below with reference to the accompanying drawings. For the sake of clarity and conciseness, not all features of actual embodiments are described in this specification. However, it should be understood that in the process of developing any such actual embodiment, many implementation-specific decisions may be made to achieve the developer's specific goals, and these decisions may vary from one implementation to another.

[0054] It is also necessary to explain here that, in order to avoid obscuring the present application due to unnecessary details, the accompanying drawings only show the device structure closely related to the solution according to the present application, while other details that are not closely related to the present application are omitted.

[0055] It should be understood that the present application is not limited to the described embodiments due to the following description with reference to the accompanying drawings. In this document, where feasible, the embodiments may be combined with each other, features between different embodiments may be replaced or borrowed, and one or more features may be omitted in one embodiment.

[0056] This application provides a method for constructing a delay-resistant cross-protocol communication decoder based on the DANN network architecture. The main purpose is to provide a delay-resistant decoder for the packet-level CTC communication system. The decoder is constructed based on the DANN (Domain-Adversarial Training of Neural Networks, deep adversarial neural network) network architecture. The encoder constructed by this method can effectively reduce the reliability of the existing CTC communication system in the multi-wireless node environment of smart furniture.

[0057] The present invention provides a method for constructing a time-delay-resistant cross-protocol communication decoder based on the DANN network architecture. Figure 1 A flowchart of a method for constructing a time-delay-resistant cross-protocol communication decoder based on a DANN network architecture according to an embodiment of the present application is shown. Figure 1 , methods include:

[0058] Step S11 preprocesses the RSSI code sequence obtained by the receiving end to obtain a preprocessed RSSI code sequence; the preprocessing includes code segmentation and sequence interference elimination. Here, the transmitting and receiving ends perform cross-protocol communication in different delay and interference environments. In this embodiment, training data is constructed based on the RSSI code sequence obtained by the receiving end to train the decoder.

[0059] Step S12: Segment the pre-processed RSSI code sequence to obtain multiple sequence segments. Figure 2 A schematic diagram showing how to segment an RSSI coding sequence to obtain multiple sequence segments is shown.

[0060] In step S13, each sequence segment is labeled, and training data is constructed based on the labeled sequence segments. The labels include delay labels and symbol labels. The delay labels include those with delay and those without delay. The symbol labels include symbol types Symbo 1 and Symbo 0. Here, each sequence segment can be labeled manually to facilitate semi-supervised learning. After the labeling operation, the sequence segments are divided into four categories of data: those with delay and with symbol labels (Symbo 1 or Symbo 0), those without delay and with symbol labels (Symbo 1 or Symbo 0), those with delay and without symbol labels, and those without delay and without symbol labels. During training, the sequence segments with delay are used as target domain data, and the sequence segments without delay are used as source domain data.

[0061] Step S14: training the anti-delay cross-protocol communication decoder based on the DANN network architecture based on the training data to obtain a trained anti-delay cross-protocol communication decoder based on the DANN network architecture.

[0062] Figure 3 The structural block diagram of the delay-resistant cross-protocol communication decoder based on the DANN network architecture is shown in Figure 3 The delay-resistant cross-protocol communication decoder based on the DANN network architecture includes a feature extractor, a spatial pooling layer, a symbol classifier, and a delay classifier. The feature extractor is used to extract features from the training data to obtain a feature vector. The spatial pooling layer is used to perform a maximum pooling operation on the feature vector to obtain a pooled feature vector. The symbol classifier is used to output the symbol category of the sequence segment based on the pooled feature vector, and the delay classifier is used to output the delay category of the sequence segment based on the pooled feature vector. Here, the pooled feature vector is a feature vector that can match the input dimension of the classification discriminant layer. The classification discriminant includes a delay classifier and a symbol discriminator.

[0063] Among them, the feature extractor can adopt a CNN neural network, through which the hidden feature information of the coding sequence will be obtained and converted into multiple feature vectors. This embodiment can use RestNet101 as the feature extractor, but it is necessary to remove the average pooling layer and the linear layer to fine-tune the network structure.

[0064] Here, the input training data is a one-dimensional time series, and the feature extractor, using a CNN, is highly capable of extracting features from two-dimensional data. Therefore, a recursive transformation can be used to enhance the training data features and convert the one-dimensional time series into two-dimensional data. This method can reveal the internal structure of short time series, provide prior knowledge of sequence similarity and predictability, monitor the stationarity and inherent similarity of the series, and ensure consistent dimensionality within each batch of data.

[0065] Due to the inconsistency of the training data dimensions, which conflicts with the fixed dimension requirement of the neural network classification layer, the decoding network cannot be designed. Therefore, three pooling strategies with different dimensional sizes are used to construct the spatial pooling layer. Three different maximum pooling operations are performed on the feature vectors extracted from each batch of data, and the results are then merged to make each batch of data have a fixed input dimension. Figure 4 Shows the structural block diagram of the spatial pooling layer.

[0066] In one embodiment, the code segment segmentation in step S11 includes:

[0067] Delete the sampling points with a value less than -75dBm in the RSSI code sequence to obtain a valid RSSI code sequence; the valid RSSI code sequence includes multiple RSSI segments;

[0068] Here, the effective RSSI coding sequence is S0 = {s0, s1, ...s i …,s n}, construct the staggered coding sequence S1 = {0, s0, s1, ... s i …,s n}, the staggered subtraction method is used to calculate the staggered subtraction sequence X = S0-S1. The values ​​in the sequence X are the calculated values ​​of the two sampling points for staggered subtraction. According to the relationship between the calculated value and 0, the index mark of each sampling point in the RSSI coding sequence S0 is determined. Finally, the RSSI segment set with "1" bit in the RSSI coding sequence is {[index0,end0],…,[index m ,end m ]}, the RSSI segment set of "0" bit in the RSSI encoding sequence is {[end0,index1],…,[end m-1 ,index m After segmentation, the lengthy coding sequence is represented by a sequence index array. The index information can be used to obtain the feature array of each symbol, which simplifies the sequence feature extraction process, facilitates subsequent feature recognition, and saves computing resources.

[0069] Since the transmitter's high-frequency transmission and reception of packets will not change significantly within a short period of time after the packet parameters are set, the coding sequence has relatively stable sequence characteristics. Therefore, using the pre-built packet fingerprint can eliminate "1" bit errors caused by other APs. Therefore, after the coding sequence is segmented, sequence interference elimination is performed on the coding sequence, specifically including:

[0070] Determine the characteristic standard value Fea1;

[0071] Calculate the characteristic value Fea2 of each RSSI segment in the RSSI encoding sequence, Fea2 = f1 ± ξ1, f1 is the fingerprint feature group, including the energy variance, energy size and duration of the RSSI segment, ξ1 is the maximum allowed error;

[0072] If the characteristic value Fea2 of the RSSI segment is not equal to the characteristic standard value Fea1, the RSSI segment is deleted to obtain a preprocessed RSSI code sequence.

[0073] Here, the characteristic standard value Fea1 is the characteristic value of the RSSI coding sequence without delay. The specific calculation method is to obtain the RSSI coding sequence without delay, first divide the code segment to obtain multiple RSSI segments, and then calculate the characteristic value of each RSSI segment according to the same calculation method as the characteristic value Fea2, and then calculate the average value to obtain the characteristic standard value Fea1.

[0074] In one embodiment, the network structure of the sign classifier and the delay classifier is the same, but the network parameters are different. The network structure includes a fully connected layer, a one-dimensional regularization layer, a fully connected layer, a one-dimensional regularization layer, and a fully connected layer. Table 1 shows the network structure and parameter settings of the sign classifier and the delay classifier, where N is the input batch.

[0075] Table 1 Network structure and parameter settings of the classification discriminant layer

[0076]

[0077] Specifically, during the training process, the loss function used is expressed by the following formula:

[0078] E=Loss y +Loss d

[0079] Among them, E is the loss function, Loss y is the symbol loss, Loss d To determine the loss for delay;

[0080] Symbol loss yThe symbol category of the sequence segment based on the output of the symbol classifier and the symbol label of the sequence segment are obtained through the cross entropy function; the delayed discrimination loss Loss d The delay category of the sequence segment based on the output of the delay classifier and the delay label of the sequence segment are obtained through the cross entropy function.

[0081] The purpose of network training is to minimize the loss function E. During the training process, the source domain data and the target domain data are randomly mixed for training. The size of each batch is 32, the Adam optimizer is used, the learning rate is 0.001, and the warmup strategy is used to dynamically change the feature extractor learning rate parameter. The number of iterations is 500, and the optimal model is selected as the delay decoder.

[0082] Adopting the same inventive concept as the method for constructing an anti-delay cross-protocol communication decoder based on the DANN network architecture, this embodiment also provides a corresponding anti-delay cross-protocol communication decoder construction device based on the DANN network architecture. Figure 5 The structural block diagram of the anti-delay cross-protocol communication decoder construction device based on the DANN network architecture according to the embodiment of the present application is shown. Figure 5 , the device comprises:

[0083] The preprocessing module 51 is used to preprocess the RSSI code sequence obtained by the receiving end to obtain a preprocessed RSSI code sequence; the preprocessing includes code segment segmentation and sequence interference elimination;

[0084] A sequence segment acquisition module 52 is used to segment the pre-processed RSSI code sequence to obtain multiple sequence segments;

[0085] The training data construction module 53 is used to label each sequence segment and construct training data based on the labeled sequence segments; the labels include delay labels and symbol labels;

[0086] A training module 54 is configured to train the DANN network architecture-based anti-delay cross-protocol communication decoder based on the training data to obtain a trained DANN network architecture-based anti-delay cross-protocol communication decoder;

[0087] The delay-resistant cross-protocol communication decoder based on the DANN network architecture includes a feature extractor, a spatial pooling layer, a symbol classifier and a delay classifier; the feature extractor is used to extract features from the training data to obtain a feature vector; the spatial pooling layer is used to perform a maximum pooling operation on the feature vector to obtain a pooled feature vector; the symbol classifier is used to output the symbol category of the sequence segment based on the pooled feature vector, and the delay classifier is used to output the delay category of the sequence segment based on the pooled feature vector.

[0088] The method for constructing an anti-delay cross-protocol communication decoder based on the DANN network architecture of this embodiment has the same inventive concept as the method for constructing an anti-delay cross-protocol communication decoder based on the DANN network architecture mentioned above. Therefore, the specific implementation method of the device can be seen in the embodiment part of the method for constructing an anti-delay cross-protocol communication decoder based on the DANN network architecture mentioned above, and its technical effects correspond to the technical effects of the above-mentioned method, which will not be repeated here.

[0089] An embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned method for constructing an anti-delay cross-protocol communication decoder based on the DANN network architecture.

[0090] An embodiment of the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the above-mentioned method for constructing an anti-delay cross-protocol communication decoder based on the DANN network architecture.

[0091] In order to further verify the effectiveness of the decoder construction method of the present application, the following experiments were conducted.

[0092] Experiment 1

[0093] To obtain reliable model training data, this experiment evaluated the data collection environment and built transceiver equipment to collect data. The core components of the WiFi transmitter and ZigBee receiver are the Intel-5300 AGN wireless network card and the MicaZ cc2420 node, respectively. The WiFi wireless card uses the Atheros CSI Tool to configure the Intel-5300 hardware transmission parameters from the Linux kernel to send WiFi packets with different packet sizes, rates, and other characteristics. Figure 6 Schematic diagrams showing two RSSI sequences expressing symbol information.

[0094] To obtain reliable training data in a valid environment, we evaluated the system in three campus environments with varying levels of interference, using the average number of AP nodes, channel occupancy, and energy intensity as criteria. Because the wireless transmission and reception protocols use the CSMA protocol to avoid collisions, ED is a key method used by CSMA to determine whether a channel is busy. When the energy level is high and the number of APs is dense, the CSMA protocol must operate frequently. The node status in different environments is shown in Table 2.

[0095] Table 2 Wireless signal conditions in different experimental environments

[0096] Environment Type Number of nodes Channel occupancy % Average energy intensity dBm Meeting Room (Mild) <12 5％ -90 Corridor (Moderate) <18 20％ -87 Laboratory (Sever) <23 50％ -81

[0097] At the same time, in order to focus on the impact of time factors related to signal delay in the environment on system performance, the separability of the three interference environments is further evaluated by analyzing the duration (Duration), interval (Interval) of high-energy signals (>75db) in the environmental interference and the energy value variance (Variance) of low-energy signals. Figure 7 The delay factor distribution diagrams under three environments are shown, among which (a) is the delay factor distribution diagram under the conference room (Mild) environment, (b) is the delay factor distribution diagram under the corridor (Moderate) environment, and (c) is the delay factor distribution diagram under the laboratory (Sever) environment.

[0098] In the medium and heavy interference environments described above, the commercial WiFi and Zigbee devices mentioned above were used to send two different CTC coding sequences. The coding sequence is Figure 6 The scheme shown in [1] uses a 360-byte data packet to represent a "1" bit and a 0 bit to represent a "0" bit, using 0.625 transmitter silence duration to construct a code sequence. A 2T = 9.46ms interval is added between each code sequence for identification and labeling. Each data set is transmitted for 3 minutes. A fingerprint is constructed based on the normal code sequence during stable transmitter transmission in the experimental environment. An interference cancellation algorithm is then used to process each code sequence, filtering out invalid data such as missing bits. Finally, 5,000 distinct RSSI segments with and without delay interference are obtained to represent CTC code sequences with and without interference. "0" and "1" are used as labels for different symbols in the code sequence. The dataset used for network training is divided into test data and training data. The training data consists of RSSI segments with and without delay interference, representing two-fifths and one-fifth of the dataset, respectively. The test data and training data contain the same data types, each comprising one-fifth of the collected data.

[0099] Experiment 2

[0100] This experiment tests the decoding accuracy of the decoder under three interference scenarios to evaluate the effectiveness of this method.

[0101] The optimizer for decoder training uses Adaptive Moment Estimation (Adam) with a learning rate of 0.001. A layered learning rate adjustment strategy is adopted, with the feature extraction layer decaying to 0.1 times the original value each time. The other layers use a warm-up learning strategy to adjust the learning rate. In the coefficient of the gradient reversal layer, t is the ratio of the current number of iterations to the total number of iterations.

[0102] In order to prove that the decoder of this application can transfer and fuse the features of delayed coding sequence and non-delayed coding sequence, and improve the decoding accuracy of delayed coding, this experiment visualizes the feature graph of the coding sequence in the network and analyzes the decoder performance from the distribution after PCA dimensionality reduction. Figure 8 shows the spatial distribution of the encoded sequence features after decoder processing, Figure 9 The feature space distribution of the coded sequence without decoder processing is shown in Figure 2. Figure 8 In the image, most of the feature points of the Symbol “0” sequence without delay (red) and with delay (green) are clustered and separated from the Symbol “1” sequence (blue and yellow). Figure 9 There are obvious differences in the coding sequence characteristics when this method is not used, which shows that decoding can distinguish coding sequences containing delays and eliminate the delays at the same time.

[0103] Experiment 3

[0104] Since there are multiple devices distributed in a typical smart home environment, this experiment conducted unobstructed direct transmission and reception measurements from 1 to 5, where the WiFi transmitter transmission power was set in the range of 12+2dBm. Figure 10 The chart shows decoding accuracy results for different distances between the transmitter and receiver in different environments. As the distance between devices increases, the decoder DAComm maintains a low bit error rate (SER) of less than 1% in both moderate and mild environments. Even in heavy interference, the accuracy remains at 82.18% at the longest distance and reaches 88.74% at the closest distance. This is a slight decrease compared to the accuracy on the test dataset. This is due to the unavoidable decoder error in the variable test environment, but it is within a controllable range.

[0105] In summary, this application has the following technical effects:

[0106] This application utilizes the property that domain adaptive neural networks can migrate the feature space mapping of source domain data and target domain data, and uses the coding sequence with delay as the target domain data and the coding sequence without delay as the source domain data, so as to eliminate the differences in the distribution of spatial features between the two. Finally, a decoder is designed that can extract information from the coding sequence with delay interference. This decoder improves the reliability and stability of packet-level CTC technology deployed in a multi-device environment.

[0107] The above descriptions are merely examples of various embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any modifications or substitutions that can be readily conceived by a person skilled in the art within the technical scope disclosed in the present application should be included within the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for constructing a delay-resistant cross-protocol communication decoder based on the DANN network architecture, characterized in that: include: Preprocessing the RSSI code sequence obtained by the receiving end to obtain a preprocessed RSSI code sequence; The pre-processing includes code segment segmentation and sequence interference elimination; Segmenting the preprocessed RSSI code sequence to obtain multiple sequence segments; Labeling each of the sequence segments, and constructing training data based on the labeled sequence segments; the labels include delay labels and symbol labels; Training the anti-delay cross-protocol communication decoder based on the DANN network architecture based on the training data to obtain a trained anti-delay cross-protocol communication decoder based on the DANN network architecture; The anti-delay cross-protocol communication decoder based on the DANN network architecture includes a feature extractor, a spatial pooling layer, a symbol classifier and a delay classifier; The feature extractor is used to extract features from the training data to obtain a feature vector; the spatial pooling layer is used to perform a maximum pooling operation on the feature vector to obtain a pooled feature vector; the symbol classifier is used to output the symbol category of the sequence segment based on the pooled feature vector, and the delay classifier is used to output the delay category of the sequence segment based on the pooled feature vector; The code segment segmentation includes: Deleting sampling points with values ​​less than -75dBm in the RSSI code sequence to obtain a valid RSSI code sequence; the valid RSSI code sequence includes multiple RSSI segments; The sequence interference elimination includes: Determine the characteristic standard value Fea1; Calculate the characteristic value Fea2 of each RSSI segment in the RSSI coding sequence, Fea2=f1±ξ1, f1 is a fingerprint feature group including energy variance, energy size and duration of the RSSI segment, and ξ1 is a maximum allowable error; If the characteristic value Fea2 of the RSSI segment is not equal to the characteristic standard value Fea1, the RSSI segment is deleted to obtain a preprocessed RSSI coding sequence.

2. The method according to claim 1, wherein The network structure of the symbol classifier is the same as that of the delay classifier, but the network parameters are different. The network structure includes a fully connected layer, a one-dimensional regularization layer, a fully connected layer, a one-dimensional regularization layer and a fully connected layer.

3. The method according to claim 1, wherein The loss function used in the training process is expressed by the following formula: E=Loss y +Loss d Among them, E is the loss function, Loss y is the symbol loss, Loss d To determine the loss for delay; The symbol loss Loss y The symbol category of the sequence segment output by the symbol classifier and the symbol label of the sequence segment are obtained through the cross entropy function; the delay discrimination loss Loss d The delay category of the sequence segment output by the delay classifier and the delay label of the sequence segment are obtained through a cross entropy function.

4. A device for constructing a time-delay-resistant cross-protocol communication decoder based on a DANN network architecture, characterized in that: include: The preprocessing module is used to preprocess the RSSI code sequence obtained by the receiving end to obtain a preprocessed RSSI code sequence; The pre-processing includes code segment segmentation and sequence interference elimination; A sequence segment acquisition module is used to segment the pre-processed RSSI code sequence to obtain multiple sequence segments; A training data construction module is used to label each of the sequence segments and construct training data based on the labeled sequence segments; the labels include delay labels and symbol labels; A training module, configured to train the anti-delay cross-protocol communication decoder based on the DANN network architecture based on the training data to obtain a trained anti-delay cross-protocol communication decoder based on the DANN network architecture; The anti-delay cross-protocol communication decoder based on the DANN network architecture includes a feature extractor, a spatial pooling layer, a symbol classifier and a delay classifier; The feature extractor is used to extract features from the training data to obtain a feature vector; the spatial pooling layer is used to perform a maximum pooling operation on the feature vector to obtain a pooled feature vector; the symbol classifier is used to output the symbol category of the sequence segment based on the pooled feature vector, and the delay classifier is used to output the delay category of the sequence segment based on the pooled feature vector; The code segment segmentation includes: Dividing the RSSI code sequence into a plurality of RSSI segments according to a magnitude relationship between a value of each sampling point in the RSSI code sequence and a set value; The sequence interference elimination includes: Determine the characteristic standard value Fea1; Calculate the characteristic value Fea2 of each RSSI segment in the RSSI coding sequence, Fea2=f1±ξ1, f1 is a fingerprint feature group including energy variance, energy size and duration of the RSSI segment, and ξ1 is a maximum allowable error; If the characteristic value Fea2 of the RSSI segment is not equal to the characteristic standard value Fea1, the RSSI segment is deleted to obtain a preprocessed RSSI coding sequence.

5. The device according to claim 4, characterized in that The network structure of the symbol classifier is the same as that of the delay classifier, but the network parameters are different. The network structure includes a fully connected layer, a one-dimensional regularization layer, a fully connected layer, a one-dimensional regularization layer and a fully connected layer.

6. The device according to claim 4, characterized in that The loss function used in the training process is expressed by the following formula: E=Loss y +Loss d Among them, E is the loss function, Loss y is the symbol loss, Loss d To determine the loss for delay; The symbol loss Loss y The symbol category of the sequence segment output by the symbol classifier and the symbol label of the sequence segment are obtained through the cross entropy function; the delay discrimination loss Loss d The delay category of the sequence segment output by the delay classifier and the delay label of the sequence segment are obtained through a cross entropy function.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method for constructing an anti-delay cross-protocol communication decoder based on a DANN network architecture as described in any one of claims 1 to 3.

8. A computer program product, characterized in that It includes a computer program / instruction, which, when executed by a processor, implements the method for constructing an anti-delay cross-protocol communication decoder based on a DANN network architecture as described in any one of claims 1 to 3.

Images