CN110909863B - Rowland sky-earth wave time delay estimation method based on artificial neural network - Google Patents

Abstract

The invention discloses a method for estimating the Roland space-ground wave time delay based on artificial neural network, which specifically includes designing the artificial neural network structure, collecting samples and performing preprocessing, and using the preprocessed samples to train the artificial neural network, and the training is completed Substituting the Roland space-ground wave signal whose time delay is to be estimated into the trained artificial neural network can output the Roland space-ground wave time delay. The invention not only has high precision in estimating the Roland space-ground wave time delay, but also can be used normally in the environment with low signal-to-noise ratio.

Description

An Estimation Method of Roland Space-Ground Wave Time Delay Based on Artificial Neural Network

technical field

The invention belongs to the technical field of digital signal processing, and in particular relates to an artificial neural network-based method for estimating the time delay of Roland space-ground waves.

Background technique

The Loran C system is a ground-based long-wave wireless navigation system that has long been used as a backup for satellite navigation systems. With the rapid development of radio technology in recent years, the positioning accuracy and conditions of use of the Roland system also need to be further improved. The Roland signal is a low-frequency electromagnetic wave signal, which is greatly affected by terrain and the ionosphere. Some cities inland far from the coast receive signals with a low signal-to-noise ratio, and the existing space-ground wave delay estimation algorithm cannot be used normally. The artificial neural network algorithm is a machine learning algorithm that simulates the structure of the human brain to establish a mathematical model, and adjusts the parameters to approach the actual model through a large amount of data learning. The design is reasonable and can simulate any nonlinear system. Use this algorithm to estimate Roland The space-to-ground wave delay is not only highly accurate, but can also be used normally in low signal-to-noise ratio environments. However, the current Roland space-to-ground wave time delay does not effectively use artificial neural networks for estimation and calculation.

Contents of the invention

The object of the present invention is to provide a method for estimating the Roland space-to-ground wave time delay based on artificial neural network, which can estimate the Roland space-to-ground wave time delay under the condition of low signal-to-noise ratio based on the artificial neural network.

The technical solution adopted in the present invention is a method for estimating the time delay of Roland space-ground waves based on artificial neural network, which is specifically implemented according to the following steps:

Step 1. Design the structure of the artificial neural network, collect the Roland space-to-ground wave signal samples, and preprocess the Roland space-to-ground wave signal samples used for artificial neural network training according to the length and period of the Roland signal;

Step 2. Set the initial parameters of the artificial neural network. The initial parameters include the weight and bias of each layer of artificial neural network, the learning rate of artificial neural network and the activation function between neurons in each layer. At the same time, set the Roland sky-ground wave signal sample The number of training times and the target threshold of the training error function;

Step 3. Use the preprocessed Roland space-to-earth wave signal samples to train the artificial neural network. When the error function value is less than the threshold value or the number of learning times reaches the maximum set value, the training stops, and the trained artificial neural network is obtained;

Step 4. Substitute the Roland space-to-ground wave signal whose time delay is to be estimated into the trained artificial neural network to output the Roland space-to-ground wave time delay.

The present invention is also characterized in that,

Step 1 is specifically implemented according to the following steps:

Step 1.1. Design the artificial neural network structure to be 5 layers, including the input layer, the first hidden layer, the second hidden layer, the third hidden layer and the output space-earth wave delay layer. The input vector of the input layer is 5000x1, and the input vector of the first hidden layer is The number of neurons is 3000 neurons, the number of neurons in the second hidden layer is 10000 neurons, the number of neurons in the third hidden layer is 3000 neurons, and the output vector of the output sky-earth wave delay layer is 2x1;

Step 1.2, collect the Roland sky-ground wave signal sample, the sampling frequency is 10MHz, the single pulse repetition period of the Roland signal is 1 millisecond, and the length of intercepting 5000 points is 500 microseconds;

Step 1.3, the Roland space-to-earth wave signal sample value is planned to [-1,1] interval for normalization processing, and the Roland space-to-earth wave signal preprocessing sample is obtained.

In step 2, the weight and bias value of each layer of artificial neural network are set to 0.5, the learning rate of artificial neural network is 0.01, the activation function between neurons in each layer is the sigmoid function, and the training of Roland sky-earth wave signal samples The number of times is 300, and the training error function target threshold of the Roland space-ground wave signal samples, ie, the error function threshold, is 1e-4.

Step 3 is to input the Roland space-to-earth wave signal preprocessing sample from the input layer, train through the first hidden layer, the second hidden layer and the third hidden layer, and output it from the output layer. Specifically, follow the steps below:

Step 3.1, forward output

Calculate the output vector O _j of each hidden layer of the Roland sky-ground wave signal sample according to the following formula,

O _j ＝f(∑x _n w _nj +θ _j ) (1)

In formula (1), j represents the hidden layer, j=1, 2, 3 respectively represent the first hidden layer and the second hidden layer, x _n represents the Roland sky-earth wave signal sample, n is a natural number not equal to 0, w is the weight, θ represents the bias, f(∑x _n w _nj +θ _j ) is the activation function f(a), and

Step 3.2, reverse error propagation

Calculate the error value Err _k of the output layer and the error value Err _j of each hidden layer according to the following formula,

Err _k ＝O _k (1-O _k )(T _k -O _k ) (3)

Err _j ＝O _j (1-O _j )∑Err _k w _jk (4)

In formula (3) and formula (4), k represents the output layer, T _k represents the target vector of the output layer, O _k represents the output vector of the output layer;

Step 3.3, parameter update

Update the weights and biases according to the error value Err _j of each hidden layer calculated in step 3.2,

The updated weight w _ij =w _nj +ηErr _j O _j (5)

Updated bias θ _j = θ _j +ηErr _j (6)

In formula (5) and formula (6), i represents the input layer, and n is the learning rate;

Step 3.4: Retrain the artificial neural network according to the updated weights and offsets, that is, substitute the updated weights and offsets into the operations of Steps 3.1 to 3.3, and continue the training cycle until the output layer error value Err _k is less than 1e -4 or reach the training times set in step 2, the training stops.

The ground-wave delay and sky-wave delay of the Roland space-to-ground wave signal samples collected in step 1 are between 30 and 200 microseconds, the signal-to-noise ratio is -15dB to 15dB, and the number of Roland space-to-ground wave signal samples is not less than 10,000 .

The beneficial effect of the present invention is that a method for estimating the time delay of Roland space-ground waves based on the artificial neural network of the present invention applies the WRELAX algorithm and the principle of invariance, and converts the multi-dimensional optimization problem into multiple one-dimensional optimization problems, which is simple and clear , easy to operate; the time delay error estimation of the Roland system is accurate, and the positioning accuracy of the Roland system is improved.

Description of drawings

Fig. 1 is the structural diagram of artificial neural network in a kind of Roland space-ground wave delay estimation method based on artificial neural network of the present invention;

Fig. 2 is the basic algorithm block diagram of artificial neural network in a kind of Roland space-ground wave delay estimation method based on artificial neural network of the present invention;

Figure 3 is the error histogram of the performance of the artificial neural network trained under different SNR conditions. Figure 3(a), (b), and (c) are the training conditions with SNR of 10dB, 0dB, and -10dB, respectively. Histogram of Roland space-ground wave delay estimation errors for the completed artificial neural network.

Detailed ways

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

A method for estimating time delay of Roland space-ground waves based on artificial neural network of the present invention is specifically implemented according to the following steps:

Step 1. Design the structure of the artificial neural network, collect the Roland space-to-ground wave signal samples, and preprocess the Roland space-to-ground wave signal samples used for artificial neural network training according to the length and period of the Roland signal:

Step 1.1, as shown in Figure 1, design the artificial neural network structure to be 5 layers, including the input layer, the first hidden layer, the second hidden layer, the third hidden layer and the output space-earth wave delay layer, and the input vector of the input layer is 5000x1 , the number of neurons in the first hidden layer is 3000 neurons, the number of neurons in the second hidden layer is 10000 neurons, the number of neurons in the third hidden layer is 3000 neurons, and the output vector of the output sky-earth wave delay layer is 2x1 ;

Step 1.2. Collect Roland space-to-earth wave signal samples, wherein the ground-wave time delay and sky-wave time delay are between 30 and 200 microseconds, the signal-to-noise ratio is -15dB to 15dB, and the number of Roland space-to-ground wave signal samples is not less than 10,000. The sampling frequency is 10MHz, the single pulse repetition period of the Roland signal is 1 millisecond, and the length of intercepting 5000 points is 500 microseconds;

Step 1.3, the Roland space-to-earth wave signal sample value is planned to [-1,1] interval for normalization processing, and the Roland space-to-earth wave signal preprocessing sample is obtained.

Step 2, the initial parameter of artificial neural network is set, promptly the weight of each layer of artificial neural network and partiality are set to be 0.5, the learning rate of artificial neural network is 0.01 and the activation function between each layer of neurons is sigmoid function, simultaneously , set the number of training times of Roland space-ground wave signal samples to 300 and set the target threshold of the training error function, that is, the error function threshold to 1e-4.

Step 3. Use the Roland space-to-earth wave signal samples to train the artificial neural network. As shown in Figure 2, the Roland space-to-earth wave signal preprocessing samples are input from the input layer, and passed through the first hidden layer, the second hidden layer and the third hidden layer Carry out training and output from the output layer. When the error function is less than the threshold value or the number of learning times reaches the maximum set value, the training stops, and the trained artificial neural network is obtained:

Step 3.1, forward output

Calculate the output vector O _j of each hidden layer of the Roland sky-ground wave signal sample according to the following formula,

O _j ＝f(∑x _n w _nj +θ _j ) (1)

In formula (1), j represents the hidden layer, j=1, 2, 3 respectively represent the first hidden layer and the second hidden layer, x _n represents the Roland sky-earth wave signal sample, n is a natural number not equal to 0, w is the weight, θ represents the bias, f(∑x _n w _nj +θ _j ) is the activation function f(a), and

Step 3.2, reverse error propagation

Calculate the error value Err _k of the output layer and the error value Err _j of each hidden layer according to the following formula,

Err _k ＝O _k (1-O _k )(T _k -O _k ) (3)

Err _j ＝O _j (1-O _j )∑Err _k w _jk (4)

In formula (3) and formula (4), k represents the output layer, O _k represents the output vector of the output layer, and T _k represents the target vector of the output layer;

Step 3.3, parameter update

Update the weights and biases according to the error value Err _j of each hidden layer calculated in step 3.2,

The updated weight w _ij =w _nj +ηErr _j O _j (5)

Updated bias θ _j = θ _j +ηErr _j (6)

In formula (5) and formula (6), i represents the input layer, and n is the learning rate;

Step 3.4: Retrain the artificial neural network according to the updated weights and offsets, that is, substitute the updated weights and offsets into the operations of Steps 3.1 to 3.3, and continue the training cycle until the output layer error value Err _k is less than 1e -4 or reach the training times set in step 2, the training stops.

Step 4. Substitute the Roland space-to-ground wave signal whose time delay is to be estimated into the trained artificial neural network to output the Roland space-to-ground wave time delay.

In order to ensure the performance of the trained artificial neural network is good, in step 1.3, mark 90% of the Roland space-to-earth wave signal preprocessing samples as training samples, and the remaining 10% as test samples, and execute steps 2 to 3 through the training samples , to obtain the trained artificial neural network, input the test sample into the trained artificial neural network, and output the Roland space-ground wave time delay of the test sample. Evaluate the performance of the trained artificial neural network according to the output results, as shown in Figure 3(a), under the environment of SNR=10dB, 90% of the Roland space-to-ground wave time delay estimation errors are concentrated in the interval less than 0.1 microseconds As shown in Figure 3(b), under the environment of SNR=0dB, 75% of the Roland space-ground wave delay estimation errors are concentrated in the interval less than 1 microsecond, and 90% are concentrated in the interval less than 2 microseconds; as shown in Figure 3(c ) shows that under the environment of SNR=-10dB, the Roland space-ground wave time delay estimation error has more than 50% within 5 microseconds, thus it can be found that a kind of Roland space-ground wave time delay estimation method based on artificial neural network of the present invention The artificial neural network trained in the above can still be used in the environment of low signal-to-noise ratio, and the estimated value of Roland space-to-ground wave time delay has high precision. Substituting the Roland space-ground wave signal whose time delay is to be estimated is then substituted into the trained and evaluated artificial neural network to obtain its Roland space-ground wave time delay.

Claims (5)

1. a method for estimating Roland space-ground wave time delay based on artificial neural network, is characterized in that, specifically implements according to the following steps: Step 1. Design the structure of the artificial neural network, collect the Roland space-to-ground wave signal samples, and preprocess the Roland space-to-ground wave signal samples used for artificial neural network training according to the length and period of the Roland signal; Step 2. Set the initial parameters of the artificial neural network. The initial parameters include the weight and bias of each layer of artificial neural network, the learning rate of the artificial neural network and the activation function between neurons in each layer. At the same time, set the Roland sky-earth wave The number of training times of the signal samples and the target threshold of the training error function; Step 3. Use the Roland sky-ground wave signal samples to train the artificial neural network. When the error function is less than the threshold value or the number of learning times reaches the maximum set value, the training stops, and the trained artificial neural network is obtained; Step 4. Substitute the Roland space-to-ground wave signal whose time delay is to be estimated into the trained artificial neural network to output the Roland space-to-ground wave time delay. 2. a kind of Roland space-ground wave time delay estimation method based on artificial neural network according to claim 1, is characterized in that, described step 1 is specifically implemented according to the following steps: Step 1.1, design the artificial neural network structure to be 5 layers, including the input layer, the first hidden layer, the second hidden layer, the third hidden layer and the output space-earth wave delay layer, the input vector of the input layer is 5000x1, the first hidden layer The number of neurons in the layer is 3000 neurons, the number of neurons in the second hidden layer is 10000 neurons, the number of neurons in the third hidden layer is 3000 neurons, and the output vector of the output sky-earth wave delay layer is 2x1; Step 1.2, collect the Roland sky-ground wave signal sample, the sampling frequency is 10MHz, the single pulse repetition period of the Roland signal is 1 millisecond, and the length of intercepting 5000 points is 500 microseconds; Step 1.3, the Roland space-to-earth wave signal sample value is planned to [-1,1] interval for normalization processing, and the Roland space-to-earth wave signal preprocessing sample is obtained. 3. a kind of Roland sky-ground wave time delay estimation method based on artificial neural network according to claim 1, is characterized in that, in described step 2, the weight and the offset value of setting every layer of artificial neural network are 0.5 , the learning rate of the artificial neural network is 0.01, the activation function between neurons in each layer is a sigmoid function, the training times of the Roland space-to-earth wave signal samples is 300 times, and the training error function target threshold of the Roland space-to-earth wave signal samples is the error function threshold is 1e-4. 4. a kind of Roland space-to-earth wave time delay estimation method based on artificial neural network according to claim 2, is characterized in that, described step 3 is that the Roland space-to-ground wave signal preprocessing sample is input by the input layer, through the first hiding Layer, the second hidden layer and the third hidden layer are trained and output by the output layer, specifically implemented in the following steps: Step 3.1, forward output Calculate the output vector O _j of each hidden layer of the Roland sky-ground wave signal sample according to the following formula, O _j ＝f(∑x _n w _nj +θ _j ) (1) In formula (1), j represents the hidden layer, j=1, 2, 3 respectively represent the first hidden layer and the second hidden layer, x _n represents the Roland sky-earth wave signal sample, n is a natural number not equal to 0, w is the weight, θ represents the bias, f(∑x _n w _nj +θ _j ) is the activation function f(a), and

Step 3.2, reverse error propagation Calculate the error value Err _k of the output layer and the error value Err _j of each hidden layer according to the following formula, Err _k ＝O _k (1-O _k )(T _k -O _k ) (3) Err _j ＝O _j (1-O _j )∑Err _k w _jk (4) In formula (3) and formula (4), k represents the output layer, T _k represents the target vector of the output layer, O _k represents the output vector of the output layer; Step 3.3, parameter update Update the weights and biases according to the error value Err _j of each hidden layer calculated in step 3.2, The updated weight w _ij =w _nj +ηErr _j O _j (5) Updated bias θ _j = θ _j +ηErr _j (6) In formula (5) and formula (6), i represents the input layer, and n is the learning rate; Step 3.4: Retrain the artificial neural network according to the updated weights and offsets, that is, substitute the updated weights and offsets into the operations of Steps 3.1 to 3.3, and continue the training cycle until the output layer error value Err _k is less than 1e -4 or reach the training times set in step 2, the training stops. 5. a kind of Roland space-to-ground wave time delay estimation method based on artificial neural network according to claim 1, is characterized in that, in the Roland space-to-ground wave signal sample collected in the step 1, ground wave time delay and sky wave time delay between 30 and 200 microseconds, the signal-to-noise ratio is -15dB to 15dB, and the number of samples of the Roland space-to-earth wave signal is not less than 10,000.

Images