CN112291005A - A receiver signal detection method based on Bi-LSTM neural network - Google Patents

Abstract

The invention discloses a receiving end signal detection method based on Bi-LSTM neural network. First, a bidirectional long-short-term memory neural network Bi-LSTM is constructed; the original data sent by the transmitting end and the processed signal data received by the receiving end are used as the constructed signal. The training data set of Bi-LSTM neural network; use the training data set to train the Bi-LSTM neural network, and obtain the network parameters after training; place the receiving end in different positions to receive different Signal-to-noise ratio signal, input the received signal into the trained Bi‑LSTM neural network, perform real-time signal detection, and obtain the corresponding raw data. The above method adopts a bidirectional long-short-time neural network model to realize signal detection at the receiving end, which can effectively suppress inter-symbol crosstalk and nonlinear distortion, thereby improving the performance of visible light communication.

Description

A receiver signal detection method based on Bi-LSTM neural network

technical field

The invention relates to the technical field of visible light communication, in particular to a method for detecting a signal at a receiving end based on a Bi-LSTM neural network.

Background technique

With the significant increase in the number of mobile devices and users, the demand for transmission bandwidth is increasing, and visible light communication (VLC) has attracted more and more attention due to its potential high transmission bandwidth, high transmission rate, data security and no electromagnetic interference. growing research interest. Currently, various VLC transmission methods have been proposed to improve communication capabilities.

As an important supplement to radio frequency (RF) communication, VLC can be used in electromagnetic sensitivity, submarine communication and other fields, and is considered to be one of the most important green information technologies. However, visible light communication still faces many problems. The biggest challenge is The modulation bandwidth of LEDs, and the modulation bandwidth of general phosphor LEDs is only a few megahertz, which severely limits the rate of visible light communication. In order to improve the transmission rate, in addition to expanding the bandwidth of the LED from the structure of the LED and the design of the pre-equalization circuit, high-order modulation and multi-carrier transmission schemes can also be used. However, this greatly increases the complexity of the receiving end, and in the signal-to-noise ratio When it is low, high-order modulation and multi-carrier will lead to a more obvious impact on the peak-to-average power ratio, which in turn affects the visible light communication rate. In addition, the LMMSE equalizer and Volterra equalizer in the prior art can suppress inter-symbol crosstalk, but at low signal-to-noise conditions. When the ratio and crosstalk are more serious, the performance needs to be improved.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a receiving end signal detection method based on Bi-LSTM neural network. The method adopts the bidirectional long and short-term neural network model to realize the receiving end signal detection, which can effectively suppress the inter-symbol crosstalk and nonlinear distortion, thereby improving the Visible light communication performance.

The purpose of this invention is to realize through the following technical solutions:

A method for detecting a signal at a receiving end based on a Bi-LSTM neural network, the method comprising:

Step 1. Build a bidirectional long-short-term memory neural network Bi-LSTM;

Step 2, take the original data sent by the transmitter and the processed signal data received by the receiver as the training data set of the constructed Bi-LSTM neural network;

Step 3, using the training data set to train the Bi-LSTM neural network to obtain the network parameters after training;

Step 4: Place the receiving end at different positions to receive signals with different signal-to-noise ratios, input the received signals into the Bi-LSTM neural network that has been trained, perform real-time signal detection, and obtain corresponding raw data.

It can be seen from the technical solutions provided by the present invention that the above method adopts a bidirectional long-short-term neural network model to realize signal detection at the receiving end, which can effectively suppress inter-symbol crosstalk and nonlinear distortion, thereby improving the performance of visible light communication.

Description of drawings

In order to illustrate the technical solutions of the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without any creative effort.

1 is a schematic flowchart of a method for detecting a signal at a receiving end based on a Bi-LSTM neural network provided by an embodiment of the present invention;

2 is a schematic diagram of the internal structure of a single Bi-LSTM unit provided by an embodiment of the present invention;

3 is a schematic diagram of a signal processing process of visible light communication according to an embodiment of the present invention;

4 is a schematic diagram of a comparison curve between a detection method provided by an embodiment of the present invention and a traditional method in terms of performance and receiver position;

FIG. 5 is a schematic diagram of a comparison curve between the detection method provided by the embodiment of the present invention and the traditional method in terms of performance and sampling rate.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present invention.

The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings. FIG. 1 is a schematic flowchart of a method for detecting a signal at a receiving end based on a Bi-LSTM neural network provided by an embodiment of the present invention, and the method includes:

Step 1. Build a bidirectional long-short-term memory neural network Bi-LSTM;

In this step, by constructing a bidirectional long-short-term memory neural network Bi-LSTM, it is introduced into the signal detection of the receiving end of the visible light OOK modulation communication system, and the signal vector received by the receiving end is used as the input signal. Forward calculation, output the symbol estimated value corresponding to the sampling value of the receiving end, the Bi-LSTM neural network constructed in this example specifically includes:

A two-layer Bi-LSTM neural network layer, a fully-connected layer with N ₁ neurons, a fully-connected layer with N ₂ neurons, a fully-connected layer with 2 neurons, One softmax layer and one classification layer.

In the specific implementation, a single Bi-LSTM neural network layer includes K Bi-LSTM units, the number of hidden layers of each Bi-LSTM unit is N ₃ , and the input dimension is M×1, as shown in FIG. 2 , which is an embodiment of the present invention A schematic diagram of the internal structure of a single Bi-LSTM unit is provided. A single Bi-LSTM unit contains an input gate, an output gate and a forget gate, where:

The forget gate decides which information should be discarded or retained, and the specific expression is as follows:

f _t =σ(W _f ·[h _t-1 , x _t ]+b _f )

Among them, σ( ) represents the sigmoid function; h _t-1 and x _t represent the output of the previous moment and the input of the current moment, respectively; W _f and b _f represent the parameter matrix and bias matrix of the forget gate, respectively; f _t represents The output of the forget gate;

The input gate is used to update the state, and the specific expression is as follows:

i _t =σ(W _i ·[h _t-1 , x _t ]+ _bi )

Among them, Wi and _bi represent the parameter matrix and bias matrix of the input gate, respectively; _{i t} represents the output of the input gate;

The output gate is used to update the state, and the specific expression is as follows:

o _t =σ(W _o ·[h _t-1 , x _t ]+b _o )

Among them, W _o and b _o represent the parameter matrix and bias matrix of the output gate, respectively; o _t represents the output of the output gate;

The operation process of the single Bi-LSTM unit is expressed as:

h _t =o _t ×tanh(C _t )

C_t-1以及C_t分别表示临时单元细胞状态、上一个时刻的单元细胞状态以及当前时刻的单元细胞状态。Among them, W _c is the parameter matrix; b _c is the bias matrix; tanh( ) represents the tanh function; h _t represents the output at the current moment;

C _t-1 and C _t represent the temporary unit cell state, the unit cell state at the previous moment, and the unit cell state at the current moment, respectively.

Step 2, taking the original data sent by the transmitter and the processed signal data received by the receiver as the training data set of the constructed Bi-LSTM neural network;

In this step, FIG. 3 is a schematic diagram of the signal processing process of visible light communication according to the embodiment of the present invention. The raw data sent by the transmitter is modulated by on-off keying (OOK) to drive the LED, and transmitted to the receiver by light waves. ;

At the receiving end, the APD receiver performs photoelectric conversion, power amplification, and analog-to-digital conversion on the received signal, and then restores the original binary bits through finite impulse response FIR low-pass filtering, averaging filtering, sliding correlation synchronization, and signal detectors stream, where:

表示经过FIR低通滤波的信号采样值，采样频率为N倍过采样；

表示经过平均滤波和归一化处理后的信号采样值；N/M表示平均滤波窗口，其中N表示接收机对单个符号的采样数，M表示输入Bi-LST神经网络的单个符号对应的采样数；

Indicates the signal sampling value after FIR low-pass filtering, and the sampling frequency is N times oversampling;

Represents the signal sample value after average filtering and normalization; N/M represents the average filtering window, where N represents the number of samples of a single symbol by the receiver, and M represents the number of samples corresponding to a single symbol input to the Bi-LST neural network ;

Y_n＝[y_n，1，y_n，2，...，y_n，M]，其中y_n，k表示第n个符号的第k个采样值。The eigenvalue corresponding to the nth symbol is

Y _n =[yn _{, 1} , yn _{, 2} , . . . , yn _{, M} ], where yn _{, k} represents the k-th sampled value of the n-th symbol.

In specific implementation, 15,000 sets of data can be collected for neural network training, and 2,000 sets of data are used for neural network cross-validation.

Step 3, using the training data set to train the Bi-LSTM neural network to obtain the network parameters after training;

Wherein, the network parameters include the input gate, the output gate, the parameter matrix and the bias matrix of the forget gate, etc.;

The process of training the Bi-LSTM neural network is as follows:

The Bi-LSTM neural network is optimized by the Adam optimizer based on the cross-entropy loss function, and the specific loss function is expressed as:

表示第n个符号对应的预测值；N表示接收机对单个符号的采样数。Among them, x _n represents the label corresponding to the nth symbol, that is, 0 or 1;

represents the predicted value corresponding to the nth symbol; N represents the number of samples of a single symbol by the receiver.

In the specific implementation, the total number of samples is not less than 15000, the proportion of training samples is 0.8, the proportion of cross-validation samples is 0.2, the maximum number of iterations is 10, the minimum batch is 30, and the learning rate is 0.001.

Step 4: Place the receiving end at different positions to receive signals with different signal-to-noise ratios, input the received signals into the Bi-LSTM neural network that has been trained, perform real-time signal detection, and obtain corresponding raw data.

The process of the method of the present invention will be described in detail below with a specific example. This example is a receiving end signal detection method in a visible light non-line-of-sight communication system. The 3dB bandwidth of the LED transmitter of the system is 2.01MHz, and the modulation rate is 50Mbps. The specific process of this example is as follows:

Step 1. Build a bidirectional long-short-term memory neural network Bi-LSTM;

In this step, a bi-directional long-short-term memory neural network Bi-LSTM is constructed, and it is introduced into the signal detection of the receiving end of the visible light OOK modulation communication system, and the signal vector received by the receiving end is used as the input signal. Forward calculation, output the symbol estimated value corresponding to the sampling value of the receiving end, the Bi-LSTM neural network constructed in this example specifically includes:

A two-layer Bi-LSTM neural network layer, a fully-connected layer with 30 neurons, a fully-connected layer with 10 neurons, a fully-connected layer with 2 neurons, a fully-connected layer with 2 neurons softmax layer and one classification layer.

In the specific implementation, a single Bi-LSTM neural network layer contains 7 Bi-LSTM units, the number of hidden layers of each Bi-LSTM unit is 30, and the input dimension is 7×1.

Step 2, using the original data sent by the transmitter and the processed signal data received by the receiver as the training data set of the constructed Bi-LSTM neural network;

In this step, as shown in Figure 3, the raw data sent by the transmitter is modulated by on-off keying (OOK) to drive the LED, and transmitted to the receiver by light waves through a non-line-of-sight reflective link;

表示经过FIR低通滤波的信号采样值，采样频率为20倍过采样；

表示经过平均滤波和归一化处理后的信号采样值；N/M表示平均滤波窗口，其中M＝5，N＝20；第n个符号对应的特征值是

Y_n＝[y_n，1，y_n，2，y_n，3，y_n，4，y_n，5]，其中y_n，k表示第n个符号的第k个采样值。At the receiving end, the APD receiver performs photoelectric conversion, power amplification and analog-to-digital conversion on the received signal, and then restores the original binary bits through finite impulse response FIR low-pass filtering, averaging filtering, sliding correlation synchronization and signal detector. flow. in,

Indicates the signal sampling value after FIR low-pass filtering, and the sampling frequency is 20 times oversampling;

Represents the signal sample value after average filtering and normalization; N/M represents the average filtering window, where M=5, N=20; the eigenvalue corresponding to the nth symbol is

Y _n =[yn _,1 , yn _,2 , yn _,3 , yn _,4 , yn _,5 ], where yn _,k represents the kth sample value of the nth symbol.

In the specific implementation, 15,000 sets of data are collected for neural network training, and 2,000 sets of data are used for neural network cross-validation.

Step 3, using the training data set to train the Bi-LSTM neural network to obtain the network parameters after training;

The specific parameters include the parameter matrix and bias matrix of the input gate, output gate, forget gate, etc.;

In this step, the process of training the Bi-LSTM neural network is as follows:

The Bi-LSTM neural network is optimized by the Adam optimizer based on the cross-entropy loss function, and the specific loss function is expressed as:

表示第n个符号对应的预测值，N表示接收机对单个符号的采样数。总样本数为15000，训练样本占比0.8，交叉验证样本占比0.2，最大迭代次数为10，最小batch为30，学习率为0.001。Among them, x _n represents the label corresponding to the nth symbol, that is, 0 or 1;

represents the predicted value corresponding to the nth symbol, and N represents the number of samples of a single symbol by the receiver. The total number of samples is 15000, the proportion of training samples is 0.8, the proportion of cross-validation samples is 0.2, the maximum number of iterations is 10, the minimum batch is 30, and the learning rate is 0.001.

Step 4: Place the receiving end at different positions to receive signals with different signal-to-noise ratios, input the received signals into the Bi-LSTM neural network that has been trained, perform real-time signal detection, and obtain corresponding raw data.

FIG. 4 is a schematic diagram showing the comparison curve between the detection method provided by the embodiment of the present invention and the traditional method (the LMMSE equalizer, Volterra equalizer, and DCO-OFDM are shown in the figure) in performance and receiver position. It can be seen that the detection method of the present invention obtains the best performance and has good generalization in different positions.

FIG. 5 is a schematic diagram showing the comparison curve between the detection method provided by the embodiment of the present invention and the traditional method (the LMMSE equalizer, the volterra equalizer, and the DCO-OFDM are shown in the figure) in terms of performance and sampling rate. It can be seen from FIG. 5 It is concluded that the detection method of the present invention obtains the optimal performance under the three sampling rates, and has good generalization.

It should be noted that the content not described in detail in the embodiments of the present invention belongs to the prior art known to those skilled in the art.

To sum up, the detection method according to the embodiment of the present invention effectively suppresses inter-symbol crosstalk and nonlinear effects, and when the signal-to-noise ratio is low and nonlinear distortion exists, OOK modulation is used to realize multi-carrier and high-order modulation that cannot be achieved. The transmission rate, without channel estimation, achieves better performance than traditional equalization algorithms.

The above is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited to this. Substitutions should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims (6)

1. A receiving end signal detection method based on a Bi-LSTM neural network is characterized by comprising the following steps:

step 1, constructing a Bi-directional long-and-short-term memory neural network Bi-LSTM;

step 2, taking the original data sent by the transmitting end and the processed signal data received by the receiving end as a training data set of the constructed Bi-LSTM neural network;

step 3, training the Bi-LSTM neural network by using the training data set to obtain trained network parameters;

and 4, placing the receiving end at different positions to receive signals with different signal-to-noise ratios, inputting the received signals into the trained Bi-LSTM neural network, and carrying out real-time signal detection to obtain corresponding original data.

2. The method for detecting a receiving-end signal based on the Bi-LSTM neural network as claimed in claim 1, wherein in step 1, the constructed Bi-LSTM neural network specifically comprises:

two Bi-LSTM neural network layers and one layer with N neuron number₁The number of the full-connection layer and one layer of neuron is N₂The full-connection layer, the full-connection layer with the neuron number of 2, the softmax layer and the classification layer.

3. The method for detecting the signal of the receiving end based on the Bi-LSTM neural network as claimed in claim 2, wherein a single Bi-LSTM neural network layer comprises K Bi-LSTM units, and the number of hidden layers of each Bi-LSTM unit is N₃The input dimension is mx 1, and a single Bi-LSTM cell contains an input gate, an output gate, and a forgetting gate, wherein:

the forgetting gate decides which information should be discarded or retained, and the specific expression is as follows:

f_t＝σ(W_f·[h_t-1，x_t]+b_f)

wherein σ (·) represents a sigmoid function; h is_t-1And x_tRespectively representing the output of the previous time and the input of the current time; w_fAnd b_fRespectively representing a parameter matrix and a bias matrix of the forgetting gate; f. of_tAn output representing a forgetting gate;

the input gate is used for updating the state, and the specific expression is as follows:

i_t＝σ(W_i·[h_t-1，x_t]+b_i)

wherein, W_iAnd b_iRespectively representing a parameter matrix and an offset matrix of the input gate; i.e. i_tRepresents the output of the input gate;

the output gate is used for updating the state, and the specific expression is as follows:

o_t＝σ(W_o·[h_t-1，x_t]+b_o)

wherein, W_oAnd b_oRespectively representing a parameter matrix and an offset matrix of the output gate; o_tRepresents the output of the output gate;

the operation process of the single Bi-LSTM unit is represented as:

h_t＝o_t×tanh(C_t)

wherein, W_cIs a parameter matrix; b_cIs a bias matrix; tanh (·) represents a tanh function; h is_tAn output representing a current time;

C_t-1and C_tRespectively, the temporary cell state, the cell state at the previous time, and the cell state at the current time.

4. The receiving-end signal detection method based on the Bi-LSTM neural network of claim 1, wherein in step 2, the original data sent by the transmitting end is on-off keying modulated to drive the LEDs and transmitted to the receiving end by the light waves;

at the receiving end, the APD receiver performs optical-to-electrical conversion, power amplification and analog-to-digital conversion on the received signal, and then recovers the original binary bit stream through finite impulse response FIR low-pass filtering, average filtering, sliding correlation synchronization and a signal detector.

5. The receiving-end signal detection method based on the Bi-LSTM neural network as claimed in claim 4,

representing the signal sampling value after FIR low-pass filtering, wherein the sampling frequency is N times of oversampling;

representing the signal sampling value after average filtering and normalization processing;

N/M represents an average filtering window, wherein N represents the number of samples of a single symbol of a receiver, and M represents the number of samples corresponding to the single symbol of the input Bi-LST neural network;

the characteristic value corresponding to the nth symbol is

Y_n＝[y_n，1，y_n，2，...，y_n，M]Wherein y is_n，kThe kth sample value representing the nth symbol.

6. The method for detecting a receiving-end signal based on the Bi-LSTM neural network as claimed in claim 1, wherein in the step 3, the process of training the Bi-LSTM neural network specifically comprises:

and optimizing the Bi-LSTM neural network by adopting an Adam optimizer based on a cross entropy loss function, wherein the specific loss function is expressed as:

wherein x is_nA label corresponding to the nth symbol is represented;

a predicted value corresponding to the nth symbol is represented; n represents the number of samples of a single symbol by the receiver.

Images