CN112564881A - 5G communication self-adaptive transmission method based on long-time multi-threshold channel state prediction - Google Patents

Abstract

The invention discloses a 5G communication self-adaptive transmission method based on long-time multi-threshold channel state prediction, which comprises the following steps: 1. acquiring historical channel statistical characteristic samples and corresponding channel state samples by the 5G radio equipment; 2. constructing a neural network based on LSTM; 3. updating the neural network parameters of the LSTM through a self-adaptive gradient descent algorithm; 4. a multi-threshold algorithm is provided to optimize the network; 5. dynamically selecting redundant blocks based on the predicted channel state improves the reliability of data transmission. The invention ensures that the channel prediction considers the time correlation while fitting the transmitted signal to the received signal model, thereby improving the prediction precision, further dynamically adjusting the occupation ratio of redundant blocks according to the channel state and improving the reliability of data transmission in 5G communication.

Description

5G communication self-adaptive transmission method based on long-time multi-threshold channel state prediction

Technical Field

The invention relates to a 5G communication self-adaptive transmission method based on long-time multi-threshold channel state prediction, belonging to the field of wireless communication.

Background

The 5G communication technology has the characteristics of large bandwidth, low time delay, high reliability and the like, reliable data transmission with higher speed is realized in a limited frequency band based on 5G wireless communication, and challenges are brought to the information transmission technology due to the severe characteristics of a wireless channel and complex and changeable external interference. In order to adaptively track dynamic channel variations and thereby introduce FEC (forward error correction) to reduce the impact of channel fading on data transmission, the channel conditions must be predicted.

Traditional prediction methods, such as the AR model, are not suitable for long-term prediction; the calculation amount of the model based on the SOS is large; the self-adaptive algorithm needs more training sets, and the iteration is slow, so that the real-time prediction cannot be realized; adaptive kalman filter prediction requires channel statistics as additional overhead when improving prediction accuracy. The traditional algorithm focuses on predicting the channel state in a short term or predicting the state in a stable channel, and cannot simultaneously guarantee the efficient utilization of the channel and the reliable transmission of data. Therefore, how to simultaneously improve the efficient utilization of the channel and the reliable transmission of data becomes a difficult point in the adaptive channel transmission.

Disclosure of Invention

Aiming at the defects in the prior art, the invention provides a 5G communication self-adaptive transmission mechanism based on long-time multi-threshold channel state prediction, which predicts the channel state by using a historical data set by means of time correlation and obtains the channel state which can be used for self-adaptive coding according to MGcJ judgment, thereby dynamically adjusting the number of RS error correction codes, meeting the condition of successful data transmission and ensuring the data transmission reliability and the high-efficiency utilization rate of a channel of 5G communication.

In order to achieve the purpose, the invention adopts the following technical scheme:

the invention relates to a 5G communication self-adaptive transmission method based on long-time multi-threshold channel state prediction, which is characterized by comprising the following steps:

step 1, acquiring a historical channel statistical characteristic information set X transmitted by N data packets and a channel state set Y corresponding to the historical channel statistical characteristic information set X in an off-line manner through 5G radio equipment to form a historical data set, and dividing the historical data set into M data with the length of k_sizeThe data block of (1); then dividing M data blocks into training set T₁And test set T₂；

Step 2, constructing the input nodes with the number of 1 xk_sizeThe LSTM neural network of (a);

step 3, defining the current iteration number of the LSTM neural network as mu, initializing the current iteration number as mu 0, and setting the maximum iteration number as mu_max(ii) a Carrying out the mu-th random initialization on the parameters of each layer in the LSTM neural network so as to obtain the mu-th iterationAn LSTM neural network;

step 4, defining and initializing a parameter theta of the mu training_μDefine initialization μ -th order moment variable k as 0_μ＝0；

Step 5, training set T from the channel statistical characteristics₁In the mu th random selection of n 1 xk_sizeUsing the channel statistical characteristic information of the transmission time slot of the dimension data block as the sample x selected for the mu time_μInputting the data into the LSTM network of the mu iteration to obtain an n multiplied by 1 dimensional output data set f (x) of the mu iteration_μ,θ_μ)，n＜M；

Step 6, calculating the n multiplied by 1 dimension output data set f (x) of the mu iteration_μ,θ_μ) And the μ -th selected sample x_μCorresponding set of channel states Y_μMean square error e of_μ；

Step 7, calculating the gradient g of the mu iteration by using the formula (1)_μ：

Step 8, calculating the first moment variable k of the mu +1 th iteration by using the formula (2)_μ+1：

k_μ+1＝pk_μ+(1-p)g_μ (2)

In the formula (2), p is an exponential decay rate from the estimate;

step 9, calculating the deviation delta k of the first moment variable of the mu +1 th iteration by using the formula (3)_μ+1：

Δk_μ+1＝k_μ+1/(1-p^μ) (3)

Step 10, calculating the parameter theta of the mu +1 training by using the formula (4)_μ+1：

θ_μ+1＝-hΔk_μ+1/L (4)

In the formula (4), L is a constant in distance estimation, and h is a step constant in distance estimation;

step 11, after assigning mu +1 to muJudging μ > μ_maxIf yes, using the LSTM network model A of the mu iteration_μThe most optimal model; otherwise, returning to the step 5 for execution;

step 12, from the test set T₂N' 1 xk of the total_sizeInputting the statistical characteristic information x' of transmission time slot channel of dimensional data block into the optimal model A_μObtaining a predicted channel state sequence Y '═ Y (1), …, Y (i), …, Y (n')](ii) a Wherein Y (i) represents the transmission time slot channel statistical characteristic information x of the ith data block_i' predicted channel state; i ∈ [1, n']；

Step 13, constructing a multi-threshold channel state judgment model, namely an MGcJ model; and initializing the channel state Y of the i-1 th adaptive coding when i is equal to 1_Gate(i-1)＝0；

Step 14, inputting the predicted channel state Y (i) of the ith data block into the MGcJ model, and judging Y (i) > v ≧ v₁If yes, let the ith adaptively coded channel state Y_Gate(i) And executing step 21, otherwise, continuing to execute step 15; wherein v is₁A threshold value is judged for the channel state;

step 15, judging Y (i) < v₀If true, let Y_Gate(i) And 0, and executing step 21; otherwise, continuing to execute the step 16; wherein v is₀A threshold value is judged for the other channel state; and v is₁＞v₀；

Step 16, judging the channel state Y of the i-1 st self-adaptive coding_GateIf (i-1) is true, executing step 17; otherwise, go to step 19;

step 17, judging Y (i) > v₁₀If true, let Y_Gate(i) And step 21 is executed, otherwise, step 18 is executed; wherein v is₁₀Deciding a secondary threshold value for the channel state;

step 18, determining whether Y (i) > Y (i-1) is true, if true, making Y_Gate(i) 1 and step 21 is executed; otherwise, let Y_Gate(i) And 0, and executing step 21;

step 19, judging Y (i) < v_o1If true, let Y_Gate(i) And 0, and executing step 21; otherwise, step 20 is performed. Wherein v is₀₁A secondary threshold value is determined for another channel state, and v₁₀＞v₀₁；

Step 20, determining whether Y (i) < Y (i-1) is true, if yes, making Y_Gate(i) And 0, and executing step 21; otherwise, let Y_Gate(i) 1 and step 21 is executed;

step 21, assigning i +1 to i, judging whether i > n' is true, and if so, outputting the updated channel state sequence Y_Gate＝[Y_Gate(1)，…Y_Gate(i)，…Y_Gate(n′)]Otherwise, returning to step 14 for execution;

step 22, according to the ith adaptive coding channel state Y_Gate(i) Selecting c in the transmission time slot of the ith data block in 5G wireless communication_γ＝2E_γThe redundant code block of + k-1 bytes transmits the information code block of k bytes of the ith data block after error correction, wherein E_γIs the bit error byte error allowed for data transmission when the channel condition is gamma.

Compared with the prior art, the invention has the beneficial effects that:

1. the invention uses the prediction algorithm based on the LSTM, integrates the time concept into the network, fully utilizes the historical data resources, continuously updates the data set while ensuring the dimensionality of the data set, solves the problem of gradient disappearance in the traditional neural network, and can continuously predict the state change of the channel.

2. The invention uses the adaptive gradient method to update the parameters of the LTSM model, and performs global iteration on the parameters of the model, thereby avoiding the local optimal problem in the traditional BP algorithm, obtaining the global optimal model and improving the accuracy of channel prediction.

3. The invention uses MGcJ decision model to reshape the predicted channel state, avoiding the complex situation of coding scheme caused by scattered channel state in the traditional error correction coding, and adjusting the quantity of redundant blocks more dynamically and efficiently according to the channel state in the transmission process, thereby improving the reliability of data transmission in 5G communication.

Drawings

Fig. 1 is a flowchart of a 5G communication data transmission method according to the present invention;

FIG. 2 is a diagram of the channel state based on LSTM prediction according to the present invention;

fig. 3 is a diagram of multi-threshold channel state decision in accordance with the present invention.

Detailed Description

In this embodiment, referring to fig. 1, a 5G communication adaptive transmission method based on long-time multi-threshold channel state prediction is performed according to the following steps:

step 1, acquiring a historical channel statistical characteristic information set X transmitted by N data packets and a channel state set Y corresponding to the historical channel statistical characteristic information set X in an off-line manner through 5G radio equipment to form a historical data set, and dividing the historical data set into M data with the length of k_sizeThe data block of (1); then dividing M data blocks into training set T₁And test set T₂；

Step 2, constructing the input nodes with the number of 1 xk_sizeThe LSTM neural network of (a);

step 3, defining the current iteration number of the LSTM neural network as mu, initializing the current iteration number as mu 0, and setting the maximum iteration number as mu_max(ii) a Carrying out the mu-th random initialization on the parameters of each layer in the LSTM neural network so as to obtain a mu-th iterative LSTM neural network;

step 4, defining and initializing a parameter theta of the mu training_μDefine initialization μ -th order moment variable k as 0_μ＝0；

Step 5, training set T from channel statistical characteristics₁In the mu th random selection of n 1 xk_sizeUsing the channel statistical characteristic information of the transmission time slot of the dimension data block as the sample x selected for the mu time_μInputting the data into the LSTM network of the mu iteration to obtain an n multiplied by 1 dimensional output data set f (x) of the mu iteration_μ,θ_μ)，n＜M；：

Step 6, calculating the n multiplied by 1 dimension output data set f (x) of the mu iteration_μ,θ_μ) And the μ -th selected sample x_μCorresponding set of channel states Y_μMean square error e of_μ；

Step 7, calculating the gradient g of the mu iteration by using the formula (1)_μ：

Step 8, calculating the first moment variable k of the mu +1 th iteration by using the formula (2)_μ+1：

k_μ+1＝pk_μ+(1-p)g_μ (2)

In the formula (2), p is an exponential decay rate from the estimate;

step 9, calculating the deviation delta k of the first moment variable of the mu +1 th iteration by using the formula (3)_μ+1：

Δk_μ+1＝k_μ+1/(1-p^μ) (3)

Step 10, calculating the parameter theta of the mu +1 training by using the formula (4)_μ+1：

θ_μ+1＝-hΔk_μ+1/L (4)

In the formula (4), L is a constant in distance estimation, and h is a step constant in distance estimation;

step 11, after the value of mu +1 is assigned to mu, judging that mu is more than mu_maxIf yes, using the LSTM network model A of the mu iteration_μThe most optimal model; otherwise, returning to step 5 for execution, in the specific embodiment, manually setting the maximum iteration number mu of the network_max＝300；

Step 12, from the test set T₂N' 1 xk of the total_sizeInputting the statistical characteristic information x' of transmission time slot channel of dimensional data block into the optimal model A_μObtaining a predicted channel state sequence Y '═ Y (1), … Y (i), … Y (n')](ii) a Wherein Y (i) represents the transmission time slot channel statistical characteristic information x of the ith data block_i' predicted channel state; i ∈ [1, n']Detailed description of the preferred embodimentsIn the middle, the obtained predicted channel state is as shown in fig. 2, and under a sample of an massive data set, the channel state obtained by LSTM prediction can better track the change of the actual channel state;

step 13, predicting the obtained channel state non-integer, so that error correction according to channel state self-adaptive coding is particularly complex for ensuring reliable data transmission, and constructing a multi-threshold channel state judgment model, namely an MGcJ model; initializing i to 1, and enabling the i-1 st channel state Y of adaptive coding_Gate(i-1)＝0；

Step 14, inputting the predicted channel state Y (i) of the ith data block into MGcJ model, and judging Y (i) > v ≧ v₁If yes, let the ith adaptively coded channel state Y_Gate(i) And executing step 21, otherwise, continuing to execute step 15; wherein v is₁A threshold value is judged for the channel state;

step 15, judging Y (i) < v₀If true, let Y_Gate(i) And 0, and executing step 21; otherwise, continuing to execute the step 16; wherein v is₀A threshold value is judged for the other channel state; and v is₁＞v₀；

Step 16, judging the channel state Y of the i-1 st self-adaptive coding_GateIf (i-1) is true, executing step 17; otherwise, go to step 19;

step 17, judging Y (i) > v₁₀If true, let Y_Gate(i) And step 21 is executed, otherwise, step 18 is executed; wherein v is₁₀Deciding a secondary threshold value for the channel state;

step 18, determining whether Y (i) > Y (i-1) is true, if true, making Y_Gate(i) 1 and step 21 is executed; otherwise, let Y_Gate(i) And 0, and executing step 21;

step 19, judging Y (i) < v_o1If true, let Y_Gate(i) And 0, and executing step 21; otherwise, step 20 is performed. Wherein v is₀₁A secondary threshold value is determined for another channel state, and v₁₀＞v₀₁；

Step 20, determining whether Y (i) < Y (i-1) is true, if yes, making Y_Gate(i) And 0, and executing step 21; otherwise, let Y_Gate(i) 1 and step 21 is executed;

step 21, assigning i +1 to i, judging whether i > n' is true, and if so, outputting the updated channel state sequence Y_Gate＝[Y_Gate(1)，…Y_Gate(i)，…Y_Gate(n′)]Otherwise, returning to step 14 for execution, in a specific embodiment, the multi-threshold channel decision constructed in steps 14-21 is shown in fig. 3;

step 22, in the transmission time slot of the ith data block, according to the ith self-adaptive coded channel state Y_Gate(i) Selecting c in data transmission for 5G wireless communication_γ＝2E_γThe redundant code block of + k-1 bytes transmits the k-byte normal information code block of the data to be transmitted after error correction, thereby improving the transmission reliability of the data, wherein E_γIs the bit error byte error allowed for data transmission when the channel condition is gamma.

Claims (1)

1. A5G communication self-adaptive transmission method based on long-time multi-threshold channel state prediction is characterized by comprising the following steps:

step 1, acquiring a historical channel statistical characteristic information set X transmitted by N data packets and a channel state set Y corresponding to the historical channel statistical characteristic information set X in an off-line manner through 5G radio equipment to form a historical data set, and dividing the historical data set into M data with the length of k_sizeThe data block of (1); then dividing M data blocks into training set T₁And test set T₂；

Step 2, constructing the input nodes with the number of 1 xk_sizeThe LSTM neural network of (a);

step 3, defining the current iteration number of the LSTM neural network as mu, initializing the current iteration number as mu 0, and setting the maximum iteration number as mu_max(ii) a Carrying out the mu-th random initialization on the parameters of each layer in the LSTM neural network so as to obtain a mu-th iterative LSTM neural network;

step 4, defining and initializing the training of the second timeParameter theta_μDefine initialization μ -th order moment variable k as 0_μ＝0；

Step 5, training set T from the channel statistical characteristics₁In the mu th random selection of n 1 xk_sizeUsing the channel statistical characteristic information of the transmission time slot of the dimension data block as the sample x selected for the mu time_μInputting the data into the LSTM network of the mu iteration to obtain an n multiplied by 1 dimensional output data set f (x) of the mu iteration_μ,θ_μ)，n＜M；

Step 6, calculating the n multiplied by 1 dimension output data set f (x) of the mu iteration_μ,θ_μ) And the μ -th selected sample x_μCorresponding set of channel states Y_μMean square error e of_μ；

Step 7, calculating the gradient g of the mu iteration by using the formula (1)_μ：

Step 8, calculating the first moment variable k of the mu +1 th iteration by using the formula (2)_μ+1：

k_μ+1＝pk_μ+(1-p)g_μ (2)

In the formula (2), p is an exponential decay rate from the estimate;

step 9, calculating the deviation delta k of the first moment variable of the mu +1 th iteration by using the formula (3)_μ+1：

Δk_μ+1＝k_μ+1/(1-p^μ) (3)

Step 10, calculating the parameter theta of the mu +1 training by using the formula (4)_μ+1：

θ_μ+1＝-hΔk_μ+1/L (4)

In the formula (4), L is a constant in distance estimation, and h is a step constant in distance estimation;

step 11, after the value of mu +1 is assigned to mu, judging that mu is more than mu_maxIf yes, using the LSTM network model A of the mu iteration_μThe most optimal model; if not, then,returning to the step 5 for execution;

step 12, from the test set T₂N' 1 xk of the total_sizeInputting the statistical characteristic information x' of transmission time slot channel of dimensional data block into the optimal model A_μObtaining a predicted channel state sequence Y '═ Y (1), …, Y (i), …, Y (n')](ii) a Wherein Y (i) represents the transmission time slot channel statistical characteristic information x of the ith data block_i' predicted channel state; i ∈ [1, n']；

Step 13, constructing a multi-threshold channel state judgment model, namely an MGcJ model; and initializing the channel state Y of the i-1 th adaptive coding when i is equal to 1_Gate(i-1)＝0；

Step 14, inputting the predicted channel state Y (i) of the ith data block into the MGcJ model, and judging Y (i) > v ≧ v₁If yes, let the ith adaptively coded channel state Y_Gate(i) And executing step 21, otherwise, continuing to execute step 15; wherein v is₁A threshold value is judged for the channel state;

step 15, judging Y (i) < v₀If true, let Y_Gate(i) And 0, and executing step 21; otherwise, continuing to execute the step 16; wherein v is₀A threshold value is judged for the other channel state; and v is₁＞v₀；

Step 16, judging the channel state Y of the i-1 st self-adaptive coding_GateIf (i-1) is true, executing step 17; otherwise, go to step 19;

step 17, judging Y (i) > v₁₀If true, let Y_Gate(i) And step 21 is executed, otherwise, step 18 is executed; wherein v is₁₀Deciding a secondary threshold value for the channel state;

step 18, determining whether Y (i) > Y (i-1) is true, if true, making Y_Gate(i) 1 and step 21 is executed; otherwise, let Y_Gate(i) And 0, and executing step 21;

step 19, judging Y (i) < v_o1If true, let Y_Gate(i) And 0, and executing step 21; otherwise, step 20 is performed. Wherein v is₀₁A secondary threshold value is determined for another channel state, and v₁₀＞v₀₁；

Step 20, determining whether Y (i) < Y (i-1) is true, if yes, making Y_Gate(i) And 0, and executing step 21; otherwise, let Y_Gate(i) 1 and step 21 is executed;

step 21, assigning i +1 to i, judging whether i > n' is true, and if so, outputting the updated channel state sequence Y_Gate＝[Y_Gate(1)，…Y_Gate(i)，…Y_Gate(n′)]Otherwise, returning to step 14 for execution;

step 22, according to the ith adaptive coding channel state Y_Gate(i) Selecting c in the transmission time slot of the ith data block in 5G wireless communication_γ＝2E_γThe redundant code block of + k-1 bytes transmits the information code block of k bytes of the ith data block after error correction, wherein E_γIs the bit error byte error allowed for data transmission when the channel condition is gamma.

Images