A Multidimensional Indexed Modulation Chaotic Communication System and Detection Method under Complex Fading Channels

Abstract

This invention discloses a multidimensional indexed modulation chaotic communication method and apparatus based on neural network-assisted detection, belonging to the field of wireless communication and chaotic spread spectrum communication technology. The method includes: firstly, segmenting the input bitstream to obtain modulation bits, domain index bits, codeword index bits, and reference index bits; then, selecting the corresponding transform domain codebook based on the domain index and the corresponding codeword basis vector based on the codeword index, while simultaneously generating an orthogonal chaotic sequence pool and selecting the corresponding reference chaotic sequence; next, performing MPSK modulation on the modulation bits, and combining the codeword basis vector and the reference chaotic sequence to generate a multidimensional indexed modulation chaotic signal, which is then transmitted through a wireless channel after multi-carrier modulation; the receiving end demodulates, filters, and samples the received signal, and inputs it into a multi-task convolutional neural network detector to achieve joint detection of modulation symbols, domain index, codeword index, and reference index, thereby recovering the transmitted bitstream.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication and chaotic spread spectrum communication technology, specifically relating to a multidimensional index modulation chaotic communication method and device based on neural network-assisted detection, and particularly to a method for generating, transmitting and intelligently detecting multidimensional index modulation chaotic signals suitable for complex fading channel environments. Background Technology

[0002] Differential chaotic shift keying (DCSK) communication systems have attracted widespread attention in the field of incoherent wireless communication due to their advantages such as not requiring strict carrier synchronization, strong resistance to multipath fading, and good security performance. Traditional DCSK systems typically employ a joint transmission of reference and information signals, which can reduce the synchronization complexity at the receiver, but also suffers from problems such as low spectral efficiency, insufficient energy utilization, and limited bit error rate performance.

[0003] To improve system transmission efficiency, researchers have proposed indexed modulation techniques. These techniques utilize carrier indexes, time slot indexes, or codeword indexes to carry additional information, thereby increasing spectral efficiency without increasing transmit power. However, most existing indexed modulation chaotic communication systems employ a single index dimension, limiting the amount of information they can carry. In complex multipath fading environments, they are susceptible to noise and channel distortion, leading to decreased system reliability. Furthermore, traditional chaotic communication systems typically employ maximum likelihood detection at the receiver. As the index and modulation dimensions increase, the detection complexity significantly rises, and error propagation problems easily arise, affecting the overall system detection performance.

[0004] In recent years, deep learning technology has demonstrated strong feature learning capabilities in the field of wireless communication signal detection. End-to-end feature extraction and classification of received signals using neural networks can effectively improve detection accuracy in complex channel environments. However, most existing deep learning-based chaotic communication detection methods are optimized only for single tasks, lacking the ability to jointly detect multi-dimensional index information, making it difficult to balance spectral efficiency, detection reliability, and system complexity. Therefore, how to construct a chaotic communication method capable of achieving high reliability, multi-dimensional index joint transmission, and low bit error rate detection in complex fading channel environments has become an urgent problem to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide a multidimensional indexed modulation chaotic communication method and apparatus based on neural network-assisted detection, so as to solve the problems of low spectral efficiency, poor bit error rate performance under complex fading environment, and error propagation that are easy to occur in the maximum likelihood detection method in existing chaotic communication systems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A multidimensional index-modulated chaotic communication method based on neural network-assisted detection includes the following steps:

[0008] Step 1: Segment the input bitstream and map it into modulation bits, domain index bits, codeword index bits, and reference index bits respectively;

[0009] Step 2: Select the corresponding transform domain codebook based on the domain index bits, select the corresponding codeword basis vector based on the codeword index bits, and simultaneously generate an orthogonal chaotic sequence and select the corresponding reference chaotic sequence based on the reference index bits;

[0010] Step 3: MPSK modulation is performed on the modulation bits, and a multidimensional indexed modulation chaotic signal is generated by combining the codeword basis vector with the reference chaotic sequence. After pulse shaping and multicarrier modulation, it is sent.

[0011] Step 4: The receiving end performs demodulation, matched filtering, and sampling on the received signal to obtain a discrete received sequence;

[0012] Step 5: Input the discrete received sequence into the multi-task convolutional neural network detector, perform joint detection on the modulation symbols, domain index, codeword index and reference index, and recover the transmitted bit stream.

[0013] Furthermore, in step 1, the input bit stream is divided into four non-overlapping groups according to the information carrying type, which are used to carry modulation information, transform domain selection information, codeword selection information and reference sequence selection information, respectively, forming corresponding modulation bits, domain index bits, codeword index bits and reference index bits.

[0014] Furthermore, the transform domain codebook in step 2 includes the Discrete Cosine Transform (DCT) codebook, the Discrete Sine Transform (DST) codebook, the Hadamard Transform codebook, and the Hartley Transform codebook; each transform domain codebook is obtained by normalizing the corresponding transform matrix so that each codeword vector in the codebook satisfies the unit norm constraint condition.

[0015] Furthermore, in step 2, the orthogonal chaotic sequences are used to generate an initial chaotic sequence through the Chebyshev chaotic mapping, and the initial chaotic sequence is orthogonally processed using the Gram-Schmidt orthogonalization method to obtain an orthogonal chaotic sequence pool; the Chebyshev chaotic mapping satisfies:

[0016] ;

[0017] in, Indicates the current chaotic sequence value. This represents the chaotic sequence value at the next moment.

[0018] Furthermore, in step 2, one of the four transform domain codebooks is selected as the codebook used for the current transmission based on the domain index bits; a sparse codeword basis vector is selected from the selected transform domain codebook based on the codeword index bits; and a path is selected from the orthogonal chaotic sequence pool as the reference chaotic sequence used for the current transmission based on the reference index bits.

[0019] Furthermore, the multidimensional index-modulated chaotic signal in step 3 satisfies:

[0020] ;

[0021] in, and These represent the in-phase and quadrature components of the modulation symbol, respectively. and Represents the codeword basis vector. Indicates a reference chaotic sequence. Indicates code sparsity. This represents the Kronecker product operation. The multidimensional indexed modulated chaotic signal is sequentially subjected to pulse shaping filtering and multicarrier modulation, up-converting the baseband signal into a multicarrier radio frequency signal before being transmitted through the channel.

[0022] Furthermore, in step 4, the receiving end sequentially performs multi-carrier demodulation, matched filtering, and timing sampling on the received radio frequency signal, converting the continuous time-domain signal into a discrete digital receiving sequence as the input for subsequent detection.

[0023] Furthermore, the multi-task convolutional neural network detector in step 5 includes a shared feature extraction module, a modulation classification module, a domain index classification module, a codeword index detection module, and a reference index detection module. The shared feature extraction module uses a multi-layer one-dimensional convolutional network to perform unified feature extraction on the discrete received sequence. The modulation classification module and the domain index classification module use the Softmax function for multi-class detection, while the codeword index detection module and the reference index detection module use the Sigmoid function for binary bit detection. The shared feature extraction module includes a first convolutional layer, a second convolutional layer, an adaptive average pooling layer, and a fully connected layer. The first convolutional layer extracts the energy distribution features and symbol boundary features of the received signal; the second convolutional layer extracts the local phase change features and instantaneous frequency features of the received signal; the adaptive average pooling layer unifies the feature dimension; and the fully connected layer achieves high-dimensional feature fusion and outputs a shared feature vector.

[0024] Furthermore, in step 5, the multi-task convolutional neural network detector takes the discrete received sequence as direct input, without the need for channel estimation and pilot assistance, and realizes joint parallel detection of modulation symbols, domain index, codeword index and reference index, and recovers the transmitted bit stream based on the detection results.

[0025] Furthermore, a multidimensional indexed modulation chaotic communication system under complex fading channels is characterized by comprising:

[0026] The transmitting module is used to segment, index map, and perform chaotic spread spectrum modulation on the input bit stream to generate a multidimensional index-modulated chaotic signal;

[0027] The codebook construction module is used to generate multiple transform domain codebooks and complete the selection of codeword basis vectors;

[0028] The chaotic sequence generation module is used to generate an orthogonal chaotic sequence pool and select a reference chaotic sequence;

[0029] The channel transmission module is used to realize multi-carrier wireless signal transmission;

[0030] The receiving processing module is used to perform carrier demodulation, matched filtering, and sampling processing on the received signal;

[0031] The intelligent detection module is used to achieve joint detection of modulation symbols, domain indexes, codeword indexes, and reference indexes based on a multi-task convolutional neural network.

[0032] The bit recovery module is used to recover the transmitted bit stream based on the joint detection results.

[0033] The advantages of this invention are:

[0034] 1. This invention achieves multidimensional index modulation through joint field index, codeword index, and reference index, thereby improving the system's spectral efficiency and energy utilization efficiency;

[0035] 2. This invention employs an orthogonal chaotic sequence spread spectrum mechanism, which improves the system's anti-interference capability and transmission reliability under complex fading environments;

[0036] 3. This invention uses a multi-task convolutional neural network to achieve joint detection, effectively avoiding the error propagation problem in the traditional maximum likelihood detection method;

[0037] 4. This invention can maintain a low bit error rate under AWGN, complex multipath Rayleigh fading and Rician fading environments, and has good robustness and engineering application value;

[0038] 5. This invention balances spectral efficiency, detection performance, and system complexity, making it suitable for complex wireless communication scenarios. Attached Figure Description

[0039] Figure 1 This is a preferred embodiment of the transmitter structure diagram provided by the present invention;

[0040] Figure 2This is a block diagram of the receiver of the present invention;

[0041] Figure 3 This is a block diagram of the multi-task convolutional neural network structure of the present invention;

[0042] Figure 4 This is the present invention. At that time, the simulated BER value of the system;

[0043] Figure 5 This is the present invention. At that time, the simulated BER value of the system;

[0044] Figure 6 This is the present invention. At that time, the simulated BER value of the system;

[0045] Figure 7 This is the present invention. Comparison of BER curves between the JCTCS-IM-DCSK system and the JCTCS-IM-DCSK system;

[0046] Figure 8 This is the present invention. Comparison of BER curves between the DL-IM-MCDCSK system and the DL-IM-MCDCSK system;

[0047] Figure 9 This is a schematic diagram of a dual-mode differential chaotic shift keying method with joint carrier and time slot index modulation provided by the present invention in a preferred embodiment. Specific implementation methods

[0048] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0049] The implementation of the present invention will now be described in detail with reference to the accompanying drawings. A system launch block diagram is shown below. Figure 1 As shown. First, the input bitstream is processed by a bit splitter and divided into multiple substreams. Each substream is mapped to a symbol and modulated by different functional modules. Specifically, the code index selects a codeword from the codebook of the selected transform domain, while the reference index selects a sequence from the generated chaotic sequence pool. After MPSK modulation, the modulated bits are combined with the outputs of various index selectors in the DCSK modulator to generate the... On the carrier The sampling signal corresponds to each time slot. Finally, after pulse shaping and carrier modulation, the final transmitted signal is obtained in the form of a multi-carrier, multi-time slot waveform. The system is set with a total of Each carrier is divided into 1 carrier. The resulting signal is transmitted through a complex fading channel in each time slot.

[0050] The chaotic sequence generator in this scheme uses a second-order Chebyshev map to generate chaotic sequences:

[0051]

[0052] in It is a second-order Chebyshev mapping. The current sequence value, This is the value for the next sequence. By iterating through this mapping, a chaotic sequence can be generated. , The length of the chaotic sequence. Indicates co-generation The generated chaotic sequences are processed using the Schmidt orthogonalization method to further optimize them. This process aims to transform the generated chaotic sequences into a set of orthogonal sequences, ensuring that they do not interfere with each other, which is crucial for subsequent processing. This step yields an orthogonal chaotic sequence pool containing multiple orthogonal chaotic sequences. This can be used by the reference index selector. Meanwhile, after bit splitting, the number of reference index bits is... After symbol mapping, these bits are used by a reference index selector to select the chaotic sequence used for final modulation.

[0053] In this system, the number of field index bits is A bit signifies the construction of a domain codebook through four orthogonal transformations: Discrete Cosine Transform, Discrete Sine Transform, Hadamard Transform, and Hartley Transform. Each domain provides a set of orthogonal basis vectors for the signal representation. Performing the corresponding transformation on the identity matrix yields the codebook for a specific domain. The codebook order is set to The field matrix is ​​then normalized so that the norm of each row is 1:

[0054]

[0055] Once the codebook is constructed, the signal is represented by a combination of basis vectors selected from the codebook. The selection of the basis vectors is determined by the code index bits. Decision, among which For codebook Okay, from Select elements of size . The basis vectors are randomly selected from all subsets. (Fixed) We obtain the basis vectors I and Q, which represent the in-phase and quadrature components of the signal, respectively.

[0056] The number of modulation bits per time slot is , This represents the modulation order. The MPSK modulation order is selected as follows: a symbol For example, the modulation phase is , The phase disturbance introduced by phase noise. The in-phase and quadrature components of the modulated signal are respectively... , , Let be the signal amplitude. After phase noise perturbation, the signal symbol can be represented as: Under the action of the DCSK modulator, the first The subcarrier The transmitted signals for each time slot are:

[0057]

[0058] in , , This is the Kronecker product. After passing through a pulse shaping filter, the discrete-time signal is converted into a continuous-time signal:

[0059]

[0060] in For the normalized impulse response, The pulse duration of the discrete chaotic signal.

[0061] Because the signal is transmitted via multiple carriers and further modulated, the final transmitted signal is:

[0062]

[0063] in For the first Subcarrier frequencies, The duration of a single time slot, For the first The total time shift of each time slot signal relative to the time origin.

[0064] In summary, the total number of bits transmitted in this scheme is: Correspondingly, the average bit energy The calculation is as follows:

[0065]

[0066] in for The mean square value.

[0067] Figure 2 shows a block diagram of the proposed receiver. After the signal propagates through the channel, the receiver acquires the waveform. The received signal is multiplied and demodulated with carrier signals of different frequencies. The demodulated signal is then sampled after matched filtering to extract each sequence component. The recovered information is... The 3D real-valued vector representation is input into a multi-task convolutional neural network for further demodulation.

[0068] The proposed multi-task learning one-dimensional convolutional neural network adopts an end-to-end architecture, directly using the original sampled sequence of the received signal as input. Training and testing datasets are generated offline according to the transmitter structure. Specifically, the random binary bitstream is mapped according to the designed transmission process into modulation symbols, domain indices, code index combinations, and reference index bits, constructing the corresponding transmitted signal and its label. Before being input into the neural network, the received signal undergoes matched filtering and sampling operations to obtain a discrete baseband received sequence. Subsequently, the received data is shaped and reconstructed to form a one-dimensional input feature vector suitable for the multi-task learning convolutional neural network detector. The input dimension is (batch size, 1, signal length), with each channel corresponding to a real-valued signal representation.

[0069] Figure 3 shows the structure of a multi-task convolutional neural network. The network performs feature extraction through a two-stage convolutional encoder: the first convolutional layer uses a 5-tap kernel, expanding the number of channels from 1 to 32, and outputs a batch-size signal length to capture macroscopic signal features such as symbol boundaries and energy envelopes; the second convolutional layer uses a 3-tap kernel, further increasing the number of channels to 64, keeping the temporal dimension unchanged, and focuses on local signal details such as phase transitions and instantaneous frequency changes to improve representation capabilities. The two-layer convolutional design follows a hierarchical feature extraction strategy: the first layer captures coarse-grained features such as energy distribution and symbol-level structure, while the second layer refines local details such as phase transitions and instantaneous frequencies. This design achieves a balance between feature representation capabilities and computational complexity; increasing the number of convolutional layers significantly expands the parameter space and computational burden, leading to increased inference latency. Subsequently, an adaptive average pooling layer normalizes the variable-length input to a fixed length of 64, outputting a batch-size signal. This operation not only standardizes the input length but also extracts global statistical features of the signal through spatial downsampling. The flattening operation transforms the 3D feature tensor into small-batch 2D feature vectors, which are then input into a shared fully connected layer for advanced feature fusion. The first fully connected layer compresses the 4096-dimensional feature vector to 256 dimensions, introducing nonlinearity through the ReLU activation function. The second fully connected layer further refines the representation into a 128-dimensional shared feature vector, integrating common information required by all downstream tasks. The four task heads share the underlying feature extractor while retaining independent decision layers. During training, joint optimization is achieved through weighted loss functions, including two cross-entropy losses for the classification task and two binary cross-entropy losses for the bit decoding task. This collaborative optimization enables the network to simultaneously learn modulation features, domain distribution features, resource allocation indices, and control information, achieving joint optimized demodulation in complex multipath fading channels. Compared to traditional cascaded processing schemes, this framework significantly improves the overall system performance and robustness.

[0070] During the offline training phase, hyperparameters are configured by combining system experiments and domain knowledge to ensure balanced convergence and good generalization performance of the multi-task learning framework. The initial learning rate is set to... The Adam optimizer is used to adaptively adjust the parameter update step size. A step size decay strategy is adopted, halving the learning rate every 10 epochs to achieve fast convergence in the early stage of training and fine-tuning in the later stage. The batch size is fixed at 64 to balance gradient estimation stability, memory efficiency, and training speed. The number of training epochs is set to 60, determined by early stopping, to balance loss convergence and avoid overfitting. The loss components for multiple tasks are equally weighted (1:1:1:1), and experiments show that the loss magnitudes of the four tasks are similar, and the balanced weights are conducive to simultaneous optimization of tasks. A dropout rate of 0.2 is applied to the fully connected layer to enhance regularization. The hyperparameter configuration is initially determined by grid search on the validation set at a signal-to-noise ratio of 15 dB, and then the robustness is verified under different signal-to-noise ratio conditions to finally obtain a hyperparameter set that balances convergence speed, stability, and generalization performance.

[0071] Figure 4 shows the system bit error rate as a function of bit signal-to-noise ratio under the AWGN channel, and compares the results with different chaotic sequence lengths. The simulation results show that when The system achieves optimal bit error rate performance when the system is in operation, indicating that longer chaotic sequences do not necessarily lead to better performance; rather, there exists an optimal value that minimizes the bit error rate. This phenomenon mainly stems from the trade-off between spreading gain and noise immunity. Longer chaotic sequences generated through Schmitt orthogonalization help construct a sequence pool with better orthogonality, thereby suppressing inter-symbol interference and providing more discriminative feature representations for subsequent deep learning-based detectors. However, when… Beyond a certain range, excessively long sequences introduce additional noise energy, leading to a degraded system performance. Therefore, The selection of the feature should strike a balance between enhancing feature separability and controlling noise accumulation. Figure 5 Displays the number of lines in the codebook The bit error rate performance of the system when taking different values. It can be observed that as... As the codebook count increases from 3 to 6 and then to 9, the overall bit error rate shows an upward trend, with this performance difference being more pronounced in the high signal-to-noise ratio range. The main reason for this phenomenon is the number of codebook lines. The increase in the number of available basis vectors significantly expands the number of information bits carried by code-indexed modulation. However, at the same time, the correlation and mutual interference between different basis vectors become more complex. Under incoherent detection conditions, especially in noise-dominated or severely multipath environments, the expansion of the codebook size will increase the detection burden, forcing the convolutional neural network to make decisions in a higher-dimensional and sparser index space, ultimately leading to a certain degree of decrease in bit error rate performance. Figure 6 Comparison of modulation order The system's bit error rate (BER) performance was analyzed at modulation orders of 4, 8, and 16. It is clear that the BER gradually increases with increasing modulation order. This can be explained by the fact that in MPSK modulation, higher modulation orders reduce the minimum Euclidean distance between constellation points, making the system more sensitive to noise and channel impairments. Although multidimensional indexed modulation and chaotic spread spectrum can mitigate the performance degradation caused by higher-order modulation to some extent, in the incoherent detection framework, increasing the modulation order inevitably increases the uncertainty of symbol decision. Furthermore, higher-order modulation places stricter demands on the classification accuracy of neural network detectors, further increasing the probability of misclassification.

[0072] Figure 7 shows a comparison of the bit error rate (BER) performance of the proposed multidimensional indexed modulation chaotic communication system JCTCS-IM-DCSK under various channel conditions in complex fading channels. It should be noted that the JCTCS-IM-DCSK system is based on traditional incoherent demodulation; therefore, the above comparison essentially reflects the performance difference between learning-based detection and traditional incoherent demodulation. It can be seen that throughout the entire... Within the specified interval, the proposed system consistently maintains the lowest bit error rate (BER), and its BER curve exhibits a steeper decreasing trend with increasing signal-to-noise ratio (SNR). This indicates that traditional incoherent detection schemes struggle to effectively suppress energy dissipation and inter-symbol interference caused by multipath propagation under long time delay conditions. In contrast, the proposed system significantly improves its adaptability to long-delay Rayleigh fading channels by jointly utilizing chaotic spreading, multidimensional indexed modulation, and deep learning-based detection. Figure 8 Furthermore, under long-delay Rayleigh fading channels, the proposed multidimensional indexed modulation chaotic communication system for complex fading channels is compared with other schemes based on neural networks or intelligent detection. Simulation results show that, although some benchmark systems perform better in low-to-medium latency conditions... While the proposed system achieves some performance improvement across the entire signal-to-noise ratio range, it maintains optimal bit error rate performance. This advantage primarily stems from its multi-task learning framework, which enables joint decision-making on modulation symbols, domain indices, and code indices. Consequently, the neural network can more comprehensively capture the signal statistical characteristics under long-delay channel conditions, effectively avoiding local optimum decision-making problems. Overall, compared to the DL-IM-MCDCSK scheme, the proposed system achieves a lower bit error rate mainly due to three aspects: the proposed scheme jointly utilizes multi-domain index modulation and orthogonal chaotic spread spectrum to provide richer discriminative features for neural network-based detection; the proposed multi-task convolutional neural network detector achieves joint optimization of modulation symbols and multiple types of indices, better suppressing inter-index interference under deep fading conditions; and unlike existing deep learning schemes that mainly focus on simplifying index structures, the proposed framework is specifically designed and trained for complex fading channels with long delays and non-uniform gain multipath components, thus possessing stronger robustness and generalization ability.

[0073] This invention proposes a neural network-assisted multidimensional indexed modulation differential chaotic shift keying communication method. Through joint mapping of a four-transform domain sparse codebook and a Chebyshev orthogonal chaotic sequence pool, it achieves joint transmission of four-dimensional information—domain index, code index, reference index, and MPSK modulation—achieving high-speed and reliable transmission without the need for pilot and coherent detection. The system embeds index information into a multi-domain coding structure, significantly reducing the demodulation complexity of index bits. Furthermore, it utilizes chaotic spreading and multi-domain indexing joint design to enhance signal anti-interference capability and physical layer confidentiality. The receiver employs a multi-task convolutional neural network to achieve integrated joint detection of modulation symbols, domain index, code index, and reference index, effectively suppressing inter-index interference and multipath fading effects. Combined with noise suppression mechanisms, it further improves error rate performance. This invention has been verified through Monte Carlo simulations under ten complex channels, including AWGN, Rayleigh, and Rican. Compared with benchmark systems such as JTDIM-DCSK, HR-HIM-QCSK, JCTCS-IM-DCSK, JDOCTF-IM-DCSK, and DL-IM-MCDCSK, this invention exhibits lower bit error rate, higher spectral efficiency and energy efficiency, and superior robustness in channels with large delays, unequal gains, and strong frequency selectivity. The proposed solution achieves a good trade-off between implementation complexity and detection performance, demonstrating strong engineering feasibility. It can effectively improve the reliability and practicality of chaotic communication in complex wireless scenarios such as vehicle-to-everything (V2X), industrial IoT, and UAV communication, promoting the implementation and large-scale application of differential chaotic keying technology in the field of incoherent high-speed communication.

[0074] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0075] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0076] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0077] The above embodiments should be understood as illustrative only and not as limiting the scope of protection of the present invention. After reading the description of the present invention, those skilled in the art can make various alterations or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A multidimensional indexed modulation chaotic communication system and detection method under complex fading channels, characterized in that, include: Step 1: Segment the input bit stream and map it into modulation bits, domain index bits, codeword index bits, and reference index bits respectively; Step 2: Select the corresponding transform domain codebook based on the domain index bits, select the corresponding codeword basis vector based on the codeword index bits, and simultaneously generate an orthogonal chaotic sequence and select the corresponding reference chaotic sequence based on the reference index bits; Step 3: MPSK modulation is performed on the modulation bits, and a multidimensional indexed modulation chaotic signal is generated by combining the codeword basis vector with the reference chaotic sequence. After pulse shaping and multicarrier modulation, it is sent. Step 4: The receiving end performs demodulation, matched filtering, and sampling on the received signal to obtain a discrete received sequence; Step 5: Input the discrete received sequence into the multi-task convolutional neural network detector, perform joint detection on the modulation symbols, domain index, codeword index and reference index, and recover the transmitted bit stream.

2. The method according to claim 1, characterized in that, In step 1, the input bit stream is divided into four non-overlapping groups according to the information carrying type. These groups are used to carry modulation information, transform domain selection information, codeword selection information, and reference sequence selection information, respectively, forming corresponding modulation bits, domain index bits, codeword index bits, and reference index bits.

3. The method according to claim 1, characterized in that, The transform domain codebook in step 2 includes the Discrete Cosine Transform (DCT) codebook, the Discrete Sine Transform (DST) codebook, the Hadamard Transform codebook, and the Hartley Transform codebook. Each transform domain codebook is obtained by normalizing the corresponding transform matrix so that each codeword vector in the codebook satisfies the unit norm constraint.

4. The method according to claim 1, characterized in that, In step 2, the orthogonal chaotic sequences are used to generate an initial chaotic sequence through the Chebyshev chaotic mapping, and the initial chaotic sequence is orthogonalized using the Gram-Schmidt orthogonalization method to obtain an orthogonal chaotic sequence pool; the Chebyshev chaotic mapping satisfies: ； in, Indicates the current chaotic sequence value. This represents the chaotic sequence value at the next moment.

5. The method according to claim 1, characterized in that, In step 2, one of the four transform domain codebooks is selected as the codebook used for the current transmission based on the domain index bits; a sparse codeword basis vector is selected from the selected transform domain codebook based on the codeword index bits; and a path is selected from the orthogonal chaotic sequence pool based on the reference index bits as the reference chaotic sequence used for the current transmission.

6. The method according to claim 1, characterized in that, The multidimensional index-modulated chaotic signal in step 3 satisfies: ； in, and These represent the in-phase and quadrature components of the modulation symbol, respectively. and Represents the codeword basis vector. Indicates a reference chaotic sequence. Indicates code sparsity. This represents the Kronecker product operation; then, the multidimensional index-modulated chaotic signal is sequentially subjected to pulse shaping filtering and multicarrier modulation, and the baseband signal is upconverted into a multicarrier radio frequency signal before being sent into the channel for transmission.

7. The method according to claim 1, characterized in that, In step 4, the receiver sequentially performs multi-carrier demodulation, matched filtering, and timing sampling on the received radio frequency signal, converting the continuous time-domain signal into a discrete digital receive sequence, which serves as the input for subsequent detection.

8. The method according to claim 1, characterized in that, The multi-task convolutional neural network detector in step 5 includes a shared feature extraction module, a modulation classification module, a domain index classification module, a codeword index detection module, and a reference index detection module. The shared feature extraction module uses a multi-layer one-dimensional convolutional network to extract unified features from the discrete received sequence. The modulation classification module and the domain index classification module use the Softmax function for multi-class detection, while the codeword index detection module and the reference index detection module use the Sigmoid function for binary bit detection. The shared feature extraction module includes a first convolutional layer, a second convolutional layer, an adaptive average pooling layer, and a fully connected layer. The first convolutional layer is used to extract the energy distribution features and symbol boundary features of the received signal. The second convolutional layer is used to extract the local phase change features and instantaneous frequency features of the received signal; Adaptive average pooling layers are used to unify feature dimensions; fully connected layers are used to achieve high-dimensional feature fusion and output a shared feature vector.

9. The method according to claim 1, characterized in that, In step 5, the multi-task convolutional neural network detector takes the discrete received sequence as direct input, without the need for channel estimation and pilot assistance, and realizes joint parallel detection of modulation symbols, domain index, codeword index and reference index, and recovers the transmitted bit stream based on the detection results.

10. A multidimensional indexed modulation chaotic communication system under complex fading channels, used to implement the method according to any one of claims 1-9, characterized in that, include: The transmitting module is used to segment, index map, and perform chaotic spread spectrum modulation on the input bit stream to generate a multidimensional index-modulated chaotic signal; The codebook construction module is used to generate multiple transform domain codebooks and complete the selection of codeword basis vectors; The chaotic sequence generation module is used to generate an orthogonal chaotic sequence pool and select a reference chaotic sequence; The channel transmission module is used to realize multi-carrier wireless signal transmission; The receiving processing module is used to perform carrier demodulation, matched filtering, and sampling processing on the received signal; The intelligent detection module is used to achieve joint detection of modulation symbols, domain indexes, codeword indexes, and reference indexes based on a multi-task convolutional neural network. The bit recovery module is used to recover the transmitted bit stream based on the joint detection results.

Images