阅读说明：本技术 利用回声状态网络预测混沌基带无线通信解码阈值的方法 (Method for predicting chaotic baseband wireless communication decoding threshold value by using echo state network ) 是由 任海鹏 尹慧平 李洁 于 2020-11-12 设计创作，主要内容包括：本发明公开了一种利用回声状态网络预测混沌基带无线通信解码阈值的方法,步骤包括：步骤1、确定回声状态网络参数；步骤2、对储备池状态进行初始化；步骤3、进行输出权值的训练；步骤4、预测当前符号的解码阈值；步骤5、符号解码；步骤6：如果n＜L-F,L-F为一个数据帧的总长度,转向步骤4；步骤7：令F-r＝F-r+1,n-0＝0,n-(max)＝n-t,r(n-0)＝r(n),如果F-r＜N-F,N-F为传输总帧数,转向步骤3,继续解码后续更多帧的符号信息；否则,结束完成所有解码。本发明的方法,利用回声状态网络良好的短期记忆能力,准确地预测当前符号对应的解码阈值,更加有效地减小了多径传输引起的码间干扰,降低了误码率,降低了计算量。(The invention discloses a method for predicting a chaotic baseband wireless communication decoding threshold by using an echo state network, which comprises the following steps: step 1, determining echo state network parameters; step 2, initializing the state of the reserve pool; step 3, training an output weight; step 4, predicting the decoding threshold value of the current symbol; step 5, decoding the symbol; step 6: if n < L F ，L F Turning to step 4 for the total length of one data frame; and 7: let F r ＝F r +1，n 0 ＝0,n max ＝n t ，r(n 0 ) R (n), if F r ＜N F ，N F Turning to step 3 for transmitting the total frame number, and continuously decoding the symbol information of the subsequent more frames; otherwise, finishing all decoding. The method of the invention uses the good short-term memory ability of the echo state network to accurately predict the decoding threshold value corresponding to the current symbol, and furtherThe method effectively reduces the intersymbol interference caused by multipath transmission, reduces the error rate and reduces the calculated amount.)

1. A method for predicting a chaotic baseband wireless communication decoding threshold value by using an echo state network is characterized by comprising the following steps:

step 1, determining echo state network parameters;

step 2, initializing the state of the reserve pool;

step 3, training an output weight;

step 4, predicting the decoding threshold value of the current symbol;

and step 5, the symbol is decoded,

when the over-sampling rate is N_sThen, the maximum SNR point is the second one corresponding to the symbolA sampling point for decoding the symbol s according to the value of the maximum SNR point and the following decoding rule formula (7)_nThe expression is as follows:

to this end, the direct use of the sign s of the prediction threshold is accomplished_nDecoding of (1);

step 6: if n < L_F，L_FTurning to step 4 for the total length of one data frame;

and 7: let F_r＝F_r+1，n₀＝0,n_max＝n_t，r(n₀) R (n), if F_r＜N_F，N_FTurning to step 3 for transmitting the total frame number, and continuously decoding the symbol information of the subsequent more frames; otherwise, finishing all decoding.

2. The method for predicting the chaotic baseband wireless communication decoding threshold value by using the echo state network according to claim 1, wherein the specific process of step 1 is,

the echo state network consists of input layer, dynamic reserve pool and output layer, the number of input nerve cells is K, and the input vector is u (t) ═ u₁(t),…,u_K(t))^TThe superscript T represents the vector transposition; the number of neurons in the dynamic reserve pool is N, and the state vector is r (t) ═ r₁(t),…,r_N(t))^T(ii) a The number of output neurons is Q, and the output vector is v (t) ═ v₁(t),…,v_Q(t))^T(ii) a The connection weight matrix input to the reserve pool is W_in∈R^N×KExpressing N rows and K columns of real number matrix, and taking the connection weight matrix of the internal neurons in the dynamic reserve pool as a sparse matrix W e R^N×NFeedback weight matrix W from output unit to reserve pool_fb∈R^N×QAll three are randomly generated and take the value of [ -1,1]The distribution is uniform, the change is not needed in the training process, wherein the spectrum radius of the sparse matrix W, rho (W) < 1 is the condition for ensuring the network echo state attribute, and the connection sparsity SD value between neurons in the reserve pool is changed between 1% and 5% for ensuring the diversification of the reserve pool; the connection weight matrix from the reserve pool to the output unit is W_out∈R^Q×(K+N)The connection weight matrix is the only connection weight matrix to be trained by the network.

3. The method for predicting the chaotic baseband wireless communication decoding threshold value by using the echo state network according to claim 2, wherein the specific process of the step 2 is,

in the chaotic baseband wireless communication system, training data is used as frame header data and information data to jointly form a data frame for transmission, and the bit number of the training data is assumed to be n_tThe data frame count is F_rInitial value F_rThe initial value of the state vector r is r (0) ═ 0, so as to avoid different initial values of r to output weight matrix W_outThe effect of (2) requires inserting n in front of the first frame training sample of the transmitted data₀Obtaining all training samples u (n) (n is 1, …, n) corresponding to the first frame by using the randomly generated sign bit₀,…,n_max)；

Desired decoding threshold θ (n) (n is 1, …, n) of training data₀,…,n_max) Calculated according to formula (1), the expression is as follows:

wherein n is 1, …, n₀,…,n_max，Indicating intersymbol interference caused by past 3-bit symbols,representing intersymbol interference caused by future 3-bit symbols, L representing a multipath number, according to a chaotic characteristic I_l,iCalculated by formula (2), the expression is as follows:

wherein alpha is_lAnd τ_lBoth the parameters are channel parameters, both omega and beta are chaotic shaping filter parameters, and the parameters meet the condition that omega is 2 pi f, beta is fln2, and f is a fundamental frequency; the channel parameters are obtained using a conventional channel identification method,

first n to be used for initialization₀The samples are input into the network in sequence, and the state is updated according to the formula (3), wherein the expression is as follows:

r(n+1)＝tanh(W_inu(n+1)+Wr(n)+W_fbθ(n))， (3)

wherein n is 0, …, n₀-1 until a state vector r (n) is obtained₀) Namely, the initialization work of the reserve pool state is completed.

4. The method for predicting the chaotic baseband wireless communication decoding threshold value by using the echo state network as claimed in claim 3, wherein the specific process of step 3 is,

continuing to train the training sample u (n +1) (n ═ n)₀,…,n_max-1) input into the network in sequence, combined with a threshold θ (n) (n ═ n)₀,…,n_max-1) obtaining the state vector r (n) in turn according to the state update formula (1)₀+1),…,r(n_max) Corresponding to a threshold value theta (n)₀+1),…,θ(n_max) Calculating by formula (1) and formula (2) to obtain network target output;

let R and T be the state matrix and the target output matrix respectively, input vector u (n) and state vector R (n) (n ═ n)₀+1,…,n_max) Redefining a combined matrixPutting the corresponding target threshold value theta (n) into T to obtainSubstituting into formula (4), calculating weight W_out∈C^{Q ×(K+N)}The expression is as follows:

W_out＝TR^T(RR^T+λ_rI)^-1， (4)

where I is the corresponding identity matrix, λ_rIs a regular term coefficient smaller than 1 and,

at this point, the output weight W required for decoding one data frame is completed_outAnd save the state vector r (n)_max) When n is equal to n_max。

5. The method for predicting the chaotic baseband wireless communication decoding threshold value by using the echo state network as claimed in claim 4, wherein the specific process of the step 4 is,

let n be n +1, use the ESN obtained by training for prediction of decoding threshold, and combine the current input vector u (n), the combination state vector r (n-1) and the decoding threshold theta (n-1) orAnd updating according to the formula (5) to obtain a new state vector r (n), wherein the expression is as follows:

r(n)＝tanh(W_inu(n)+Wr(n-1)+W_fbθ(n-1))， (5)

output weight W obtained by r (n), u (n) and training_outPredicting the symbol s according to equation (6)_nThe expression of the decoding threshold of (1) is as follows:

predicting the resulting symbol s_nDecoding threshold ofAnd storing r (n),the value of (c).

Technical Field

The invention belongs to the technical field of artificial intelligence and wireless communication, and relates to a method for predicting a chaotic baseband wireless communication decoding threshold value by using an echo state network.

Background

The chaos has wide application prospect in the communication field due to the inherent characteristics of the chaos, such as high initial condition sensitivity, broadband property, orthogonality, easy generation and the like. In the early 90 s, chaos was applied to communication in two main aspects, one is secure communication, and the other is spread spectrum communication. Research into spread spectrum communications has led to some local area network communication standards, such as IEEE 802.15.6. Since chaos has been successfully applied to optical fiber communication channels and higher bit rates are obtained, the research focus of chaotic communication is also shifted from an ideal channel of gaussian noise to an actual communication channel with complex constraint conditions. In recent years, researches find that the chaos is suitable for new characteristics of communication application, for example, a very simple matched filter can be used for chaotic signals, and the signal-to-noise ratio can be improved to the maximum extent. And certain conditions are met, and the transmission information is not lost after the chaos passes through a multipath wireless channel. In addition, the chaotic nature can be used to reduce the effects of inter-symbol interference (ISI). Compared with the traditional non-chaotic communication, experiments show that chaotic baseband wireless communication obtains better performance.

Although Chaotic Baseband Wireless Communication Systems (CBWCS) have achieved better performance, information is decoded using only a sub-optimal decoding threshold that contains past inter-symbol interference or an improved sub-optimal decoding threshold that adds predicted one-bit symbol information in the future. Since it is difficult to predict more future symbol information, it is difficult to obtain the corresponding optimal decoding threshold value by calculation. This is an obstacle to further improving the performance of the chaotic baseband wireless communication system. Therefore, finding a better decoding threshold has become a technical problem to be solved urgently.

Disclosure of Invention

The invention aims to provide a method for predicting a chaotic baseband wireless communication decoding threshold by using an echo state network, which solves the problem that the optimal solution threshold in the prior art cannot be calculated at the current moment.

The technical scheme adopted by the invention is that a method for predicting a chaotic baseband wireless communication decoding threshold value by using an echo state network is implemented according to the following steps:

step 1, determining echo state network parameters;

step 2, initializing the state of the reserve pool;

step 3, training an output weight;

step 4, predicting the decoding threshold value of the current symbol;

and step 5, the symbol is decoded,

when the over-sampling rate is N_sThen, the maximum SNR point is the second one corresponding to the symbolA sampling point for decoding the symbol s according to the value of the maximum SNR point and the following decoding rule formula (7)_nThe expression is as follows:

to this end, the direct use of the sign s of the prediction threshold is accomplished_nDecoding of (1);

step 6: if n < L_F，L_FTurning to step 4 for the total length of one data frame;

and 7: let F_r＝F_r+1，n₀＝0,n_max＝n_t，r(n₀) R (n), if F_r＜N_F，N_FTurning to step 3 for transmitting the total frame number, and continuously decoding the symbol information of the subsequent more frames; otherwise, finishing all decoding.

The beneficial effects of the invention are that the invention comprises the following aspects: 1) according to the characteristics of the chaotic baseband signal, the method accurately predicts the decoding threshold value corresponding to the current symbol by utilizing the good short-term memory capacity of the echo state network. Compared with a method of calculating the threshold value by only using the past symbol and the past symbol plus the future one-bit symbol, the new method obtains a decoding threshold value which is closer to the optimal value, more effectively reduces the intersymbol interference caused by multipath transmission and reduces the error rate; 2) the invention does not need channel identification in the prediction process, simplifies the decoding process and reduces the calculation amount.

Drawings

FIG. 1 is a block diagram of a chaotic baseband wireless communication system used in the method of the present invention;

FIG. 2 is a schematic diagram of an echo state network structure used in the method of the present invention;

FIG. 3 is a functional block diagram of an implementation of the method of the present invention;

FIG. 4 is a bit error rate simulation result obtained by decoding different thresholds under a Gaussian channel when the method of the present invention is verified;

FIG. 5 is a bit error rate simulation result of different decoding thresholds under a time-invariant two-path channel when verifying the method of the present invention;

FIG. 6 is a bit error rate simulation result for different decoding thresholds under a time-invariant three-path channel when verifying the method of the present invention;

FIG. 7 is a bit error rate simulation result for different decoding thresholds under a time-varying two-path channel when the method of the present invention is verified;

FIG. 8 is a simulation result of bit error rates of different decoding thresholds under a time-varying three-path channel when the method of the present invention is verified.

Detailed Description

The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

Referring to fig. 1, binary symbol information s in a chaotic baseband wireless communication system_nObtaining a chaotic signal x (t) as a baseband signal through a chaotic forming filter; after the upper carrier frequency, the modulation signal is transmitted through a wireless channel, the impulse response of the channel is represented by h (t), the receiving end obtains a receiving baseband signal represented by r (t) through the lower carrier frequency, the receiving baseband signal r (t) is sent into a matched filter for filtering, and the sampling value of the output signal y (t) of the filter is compared with the corresponding decoding threshold value to recover the sending bit information. According to the characteristics of the forming filter, the current sign bit information is mainly influenced by historical 4-bit signs and future 4-bit signs, in order to predict the decoding threshold value corresponding to the current sign, the network needs to memorize historical 4-bit, current 1-bit and future 4-bit sign information, but the training time is longer due to larger corresponding training data amount, therefore, only the past 3-bit, current 1-bit and future 3-bit sign information is considered in the step process of the method of the invention and is used for predicting the decoding threshold value of the current sign.

The method of the inventionThe communication method used is to transmit information frame by frame and recover information frame by frame. Each frame comprises two parts of frame header data and information data, wherein the frame header data is used for frame synchronization, channel parameter identification, frequency offset correction and network training at the same time. Completing one network training by using frame header data in each frame, namely using output weight W obtained by frame header training_outAnd keeping the same frame transmission data information when decoding the same frame transmission data information, and carrying out symbol decoding on the frame symbol information according to the bit-by-bit prediction decoding threshold value of the weight obtained by training. According to the characteristics of chaotic signals, a training data set is generated by traversing and combining past 3-bit symbols, current 1-bit symbols and future 3-bit symbol information, the training data is used as frame header data and information data to be combined into a frame to be sent, and N is supposed to be sent in total_FFrame data, total length of data per frame L_F。

Referring to fig. 2 and 3, the working principle of the method of the present invention is that a dynamic reserve pool of an echo state network has a plurality of sparse neurons, and the training contains the operation state of the system, so that the method has a short-term memory function. Firstly, determining parameters of a network, wherein the parameters comprise K input neuron number, N neuron number and Q output neuron number in a dynamic reserve pool; the connection weight matrix input to the reserve pool is W_inThe connection weight matrix of the neuron in the dynamic reserve pool is a sparse matrix W, and the feedback weight matrix from the output unit to the reserve pool is W_fb(ii) a Outputting a weight matrix W on the training sample set_outAnd using the W obtained by the training_outThe corresponding decoding threshold is predicted from the current input signal. As shown in FIG. 3, assume the oversampling ratio N of the signal_s16, the matched filter output signal sample value (y) corresponding to the nth symbol in the communication_n(1),…,y_n(16))^TAs input to the network, the output is the decoding threshold of the corresponding symbolMaximum signal-to-noise ratio point y₈(n) decoding the received signal by comparing the decoded threshold value with the predicted decoding threshold value, and recovering the transmission bit information

Based on the principle, the method is implemented according to the following steps:

step 1, determining echo state network parameters,

the echo state network consists of input layer, dynamic reserve pool and output layer, the number of input nerve cells is K, and the input vector is u (t) ═ u₁(t),…,u_K(t))^TThe superscript T represents the vector transposition; the number of neurons in the dynamic reserve pool is N, and the state vector is r (t) ═ r₁(t),…,r_N(t))^T(ii) a The number of output neurons is Q, and the output vector is v (t) ═ v₁(t),…,v_Q(t))^T(ii) a The connection weight matrix input to the reserve pool is W_in∈R^N×KExpressing N rows and K columns of real number matrix, and taking the connection weight matrix of the internal neurons in the dynamic reserve pool as a sparse matrix W e R^N×NFeedback weight matrix W from output unit to reserve pool_fb∈R^N×QAll three are randomly generated and take the value of [ -1,1]The distribution is uniform, the change is not needed in the training process, wherein the spectrum radius of the sparse matrix W, rho (W) < 1 is the condition for ensuring the network echo state attribute, and the connection sparsity SD value between neurons in the reserve pool is changed between 1% and 5% for ensuring the diversification of the reserve pool; the connection weight matrix from the reserve pool to the output unit is W_out∈R^{Q ×(K+N)}The connection weight matrix is the only connection weight matrix to be trained by the network.

In the embodiment, the number K of input neurons and the number Q of output neurons are 16 and 1, the spectral radius ρ (W) of the sparse matrix W is 0.9, and the degree of sparsity SD is 0.02. Because the method of the invention considers the influence of past 3-bit symbols and future 3-bit symbols on the waveform of the current symbol, in order to predict the decoding threshold value corresponding to the current symbol, the network needs to memorize the historical 3-bit symbols, the sampling points corresponding to the information of the current 1-bit symbols and the future 3-bit symbols, and when the oversampling rate N is higher_sWhen the number of the past 3 bits is 16, the current 1 bit and the future 3 bits of symbol information correspond to 16 × 7 to 112 sampling points, so that the number of neurons N in the reserve pool is 112.

Step 2, initializing the state of the reserve pool,

in the chaotic baseband wireless communication system, training data is used as frame header data and information data to jointly form a data frame for transmission, and the bit number of the training data is assumed to be n_tThe data frame count is F_rInitial value F_rThe initial value of the state vector r is r (0) ═ 0, so as to avoid different initial values of r to output weight matrix W_outThe effect of (2) requires inserting n in front of the first frame training sample of the transmitted data₀Obtaining all training samples u (n) (n is 1, …, n) corresponding to the first frame by using the randomly generated sign bit₀,…,n_max)；

Desired decoding threshold θ (n) (n is 1, …, n) of training data₀,…,n_max) Calculated according to formula (1), the expression is as follows:

wherein n is 1, …, n₀,…,n_max，Indicating intersymbol interference caused by past 3-bit symbols,representing intersymbol interference caused by future 3-bit symbols, L representing a multipath number, according to a chaotic characteristic I_l,iCalculated by formula (2), the expression is as follows:

wherein alpha is_lAnd τ_lBoth the parameters are channel parameters, both omega and beta are chaotic shaping filter parameters, and the parameters meet the condition that omega is 2 pi f, beta is fln2, and f is a fundamental frequency; the channel parameters are obtained by using a conventional channel identification method (such as a least square method).

In a special case, the decoding threshold of the first bit symbol is calculated by using only future 3-bit information, the decoding threshold of the second bit symbol is calculated by using the past 1-bit and future 3-bit symbol information, and the decoding threshold of the third bit symbol is calculated by using the past 2-bit and future 3-bit symbol information; starting from the fourth bit, the decoding threshold is completely calculated according to formula (1) and formula (2), until the decoding threshold of the last bit symbol is calculated by the past 3-bit symbol information and the future 2-bit symbol information, the decoding threshold of the last bit symbol is calculated by the past 3-bit symbol information and the future 1-bit symbol information, and the decoding threshold of the last bit symbol is calculated by using the past 3-bit symbol information only.

First n to be used for initialization₀The samples are input into the network in sequence, and the state is updated according to the formula (3), wherein the expression is as follows:

r(n+1)＝tanh(W_inu(n+1)+Wr(n)+W_fbθ(n))， (3)

wherein n is 0, …, n₀-1 until a state vector r (n) is obtained₀) And finishing the initialization work of the state of the reserve pool, and preparing for the next network training.

In the embodiment, the combination of the historical 3-bit symbol, the current 1-bit symbol and the future 3-bit symbol is 2⁷A case where each case corresponds to 7 bits, so the total traversal case corresponds to 896-bit symbols, with 896-bit as the training data source for the network, i.e., n_t896. Since the current data frame is the first frame, i.e. F_r1, different initial values of r need to be considered for the output weight matrix W_outPlus the first 100 random sign bits, i.e., n, used to initialize the network₀＝100，n_max996 bits total as a frame header for network training. In order to initialize the network, the first 100 training samples are sequentially input into a reserve pool, and the state is updated according to the formula (3) by combining with the corresponding decoding threshold values calculated by the formula (1) and the formula (2), so that a state vector r (100) is obtained, and the method is ready for the next network training.

Step 3, training the output weight value,

continuing to train the training sample u (n +1) (n ═ n)₀,…,n_max-1) input into the network in sequence, combined with a threshold θ (n) (n ═ n)₀,…,n_max-1) obtaining the state vector r (n) in turn according to the state update formula (1)₀+1),…,r(n_max) Corresponding to a threshold value theta (n)₀+1),…,θ(n_max) Calculating by formula (1) and formula (2) to obtain network target output;

let R and T be the state matrix and the target output matrix respectively, input vector u (n) and state vector R (n) (n ═ n)₀+1,…,n_max) Redefining a combined matrixPutting the corresponding target threshold value theta (n) into T to obtainSubstituting into formula (4), calculating weight W_out∈C^{Q ×(K+N)}The expression is as follows:

W_out＝TR^T(RR^T+λ_rI)^-1， (4)

where I is the corresponding identity matrix, λ_rIs a regular term coefficient less than 1.

At this point, the output weight W required for decoding one data frame is completed_outAnd save the state vector r (n)_max) When n is equal to n_max。

In the embodiment, according to the initialized state vector r (100) and the state updating formula (3), training samples u (n) (101, …,996) are sequentially input to the reserve pool, and corresponding state vectors r (101), …, r (996) are obtained. Collecting the state vector and the input value at each moment to obtain a state matrix:

the corresponding target output matrix is T ═ θ (101), …, θ (996)]Substituting the matrix R and the matrix T into the formula (4), and calculating to obtain an output weight matrix W of the network_outWhere n is 996.

Step 4, predicting the decoding threshold value of the current symbol,

let n be n +1, use the trained ESN (echo state network) for the prediction of decoding threshold, combine the current input vector u (n), the combined state vector r (n-1) and the decoding threshold theta (n-1) orAnd updating according to the formula (5) to obtain a new state vector r (n), wherein the expression is as follows:

r(n)＝tanh(W_inu(n)+Wr(n-1)+W_fbθ(n-1))， (5)

output weight W obtained by r (n), u (n) and training_outPredicting the symbol s according to equation (6)_nThe expression of the decoding threshold of (1) is as follows:

predicting the resulting symbol s_nDecoding threshold ofAnd storing r (n),the value of (c).

In an embodiment, the current input vector u (997) ═ y₁(997),…,y₁₆(997))^TCombining the state vector r (996) obtained by updating the formula (3) and the target threshold theta (996), updating the state vector r (997) by the formula (5), and then obtaining W according to training_outThe symbol s is obtained according to equation (6) in combination with the current input vector u (997) and the state vector r (997)₉₉₇Threshold prediction value of

And step 5, the symbol is decoded,

when the over-sampling rate is N_sThen, the maximum SNR point is the second one corresponding to the symbolA sampling point for decoding the symbol s according to the value of the maximum SNR point and the following decoding rule formula (7)_nThe expression is as follows:

to this end, the direct use of the sign s of the prediction threshold is accomplished_nAnd (4) decoding.

In an embodiment, the oversampling ratio is N_sAnd (3) 16, the maximum signal-to-noise ratio point is the 8 th sampling point, and the current symbol information is decoded according to the value of the maximum signal-to-noise ratio point and the following decoding rule, wherein the expression is as follows:

to this end, the direct use of the sign s of the prediction threshold is accomplished₉₉₇And (4) decoding.

Step 6: if n < L_F(n is a cyclic variable whose value is changed in step 4 by making n ═ n +1), L_FFor the total length of one data frame, go to step 4,

in the embodiment, when the value of n is smaller than the length of one data frame, the process goes to step 4 until the decoding of all information symbols in one data frame is completed.

And 7: let F_r＝F_r+1，n₀＝0,n_max＝n_t，r(n₀) R (n), if F_r＜N_F，N_FTurning to step 3 for transmitting the total frame number, and continuously decoding the symbol information of the subsequent more frames; otherwise, finishing all decoding.

In the embodiment, after all the information symbols in the first frame are decoded, the data frame number is made F_r＝F_r+1, the symbol decoding work of the next frame is started in the same way. Wherein, when the data frame is not the first frame, there is no need for an initialization procedure of the reserve pool, so n here₀＝0，n_max＝896。

In order to verify the actual performance of the method, the simulation results of the Gaussian channel and the multipath channel are compared as follows: four different decoding thresholds are used, respectively, the first one being θ ═ 0, in which case the inter-symbol interference caused by past and future symbols is completely ignored; the second is the literature [ Jun Liang Yao, Yu-Zhe Sun, Hai-Peng Ren, Celso Grebogi, Experimental Wireless Communication Using Chaotic base and Waveform, IEEE Transactions on Vehicular Technology,2019,68(1):578-]The method takes theta as I_pastThe decoding threshold takes into account only intersymbol interference caused by past sign bits; the third is the document [ Ren H P, Yin H P, Bai C and Yao J L, "Performance Improvement of visual basic and Wireless Communication Using Echo State Network," IEEE Transactions on Communications,2020, vol.68, No.10, pp.6525-6536]The method takes theta as I_fut1+I_pastNot only past sign bits are considered, but also intersymbol interference caused by predicted future 1-bit signs is considered; the fourth is the decoding threshold value obtained by adopting the prediction of the invention

Simulation verification:

1) error rate under gaussian channel:

the simulation adopts a Gaussian channel model to test the error rates under different thresholds, and the obtained simulation result is shown in FIG. 4, so that compared with the first three threshold decoding methods, the error rate corresponding to the obtained decoding threshold is directly predicted to be the lowest.

2) Bit error rate under multipath channel:

the multipath parameters in different frames are different, assuming that the multipath parameters in the same frame are not changed. L represents the number of multipaths, gamma and alpha_l＝e^-γτFor the multipath fading parameter, τ is the multipath delay. Similarly, the method of the invention uses the threshold value theta as 0 and theta as I_pastAnd θ ═ I_fut1+I_pastAnd (5) comparing the three methods.

a) Simulation results under the condition of time-invariant channel parameters:

in the simulation, when L is 2, the parameter is α₀＝1,τ₁＝1,γ is 0.6, and the simulation result is shown in fig. 5; when L is 3, α₀＝1,τ₁＝1,τ₂＝2,γ is 0.6, and the simulation result is shown in fig. 6. As can be seen from fig. 5 and 6, the best performance is obtained with the method of the present invention, θ ═ I_fut1+I_pastIs less than theta ═ I_pastIn the case of (a), θ ═ I_pastIs lower than the case where θ is 0, and θ is 0, the performance is the worst. In particular, in the case of a two-path channel, at high snr, the prediction threshold used in the method of the invention is compared with θ ═ I_fut1+I_pastThe method has the advantage that the error rate performance is improved by 0.5dB compared with the method that theta is I_pastBy this approach, the bit error rate performance is improved by about 1dB, and the bit error rate increases with the number of multipaths.

b) Simulation results under time-varying channel parameters:

in actual communication, channel conditions are variable, and channel parameters are unknown. Considering the multipath parameter unknowns per frame, assume that the parameter γ obeys 0.3,0.9]The simulation result is shown in fig. 7 when L is 2, and in fig. 8 when L is 3. The simulation results of fig. 7 and 8 are slightly worse than the corresponding simulation results of fig. 5 and 6 due to channel estimation errors. As can be seen from fig. 7 and 8, the prediction threshold used in the method of the present invention is still better than the threshold θ ═ 0, θ ═ I_pastAnd θ ═ I_fut1+I_past。

In conclusion, the method is completely suitable for chaotic baseband communication, and the decoding threshold value obtained by prediction is superior to that of a comparison method.