CN111371482A

CN111371482A - Millimeter wave MIMO-NOMA method of downlink broadcast link based on beam aggregation

Info

Publication number: CN111371482A
Application number: CN202010157093.4A
Authority: CN
Inventors: 杨松立; 王爱华; 叶能; 王培森; 董俊伟; 杨佳程
Original assignee: Beijing Institute of Technology BIT
Current assignee: Beijing Institute of Technology BIT
Priority date: 2020-03-09
Filing date: 2020-03-09
Publication date: 2020-07-03
Anticipated expiration: 2040-03-09
Also published as: CN111371482B

Abstract

The invention relates to a millimeter wave MIMO-NOMA method of a downlink broadcast link based on beam aggregation, belonging to the field of communication. In order to solve the problems of user pairing and accurate beam alignment in a downlink broadcast link in the prior art, the invention provides a millimeter wave large-scale MIMO-NOMA method based on beam aggregation. In the invention, a base station generates a large-scale beam set covering the whole cell through beam forming, wherein a plurality of adjacent beams are gathered together to provide service for a group of adjacent users in a non-orthogonal mode, and an approximately optimal digital precoding matrix is designed according to channel state information by constructing a deep learning framework and adopting an offline training online application mode so as to reduce the interference among the users and realize the fairness among the users. The invention does not require strict beam alignment among users and can increase the application possibility of the NOMA scheme in the downlink millimeter wave MIMO.

Description

Millimeter wave MIMO-NOMA method of downlink broadcast link based on beam aggregation

Technical Field

The invention relates to a millimeter wave MIMO-NOMA method of a downlink broadcast link based on beam aggregation, in particular to a millimeter wave large-scale MIMO-NOMA method based on beam aggregation in the downlink broadcast link, which belongs to the field of communication and comprises a large-scale MIMO technology, a non-orthogonal multiple access technology, a mixed beam forming technology and a deep learning technology.

Background

With the explosive growth of connected devices in communication systems, the available radio resources become extremely scarce. In order to alleviate the shortage of radio resources and increase the transmission rate of communication, a direct method is to use a higher frequency band resource. As millimeter wave communication has been developed vigorously in recent years, millimeter waves have been regarded as one of the technologies having great potential because of their ability to provide a wider wireless band.

In millimeter wave communication, massive MIMO (Multi-input Multi-output) is introduced to solve problems inherent to millimeter wave transmission. The attenuation of a millimeter wave system can be compensated by using array gain and aerial multiplexing gain brought by a large-scale antenna, so that the spectral efficiency is improved, and the error rate is reduced. In addition, the arrangement of a large-scale antenna can be realized in a small area by utilizing the characteristic that the wavelength of the millimeter wave is short.

Conventional MIMO systems typically use all-digital precoding techniques, where each antenna is connected to a dedicated Radio Frequency (RF) chain. However, applying this structure in the millimeter wave band results in excessive hardware cost and power consumption. To overcome this problem, researchers have adopted hybrid precoding structures that effectively reduce the number of RF chains without significant performance loss. The hbf (hybrid beamforming) technology converts multi-user communication into multiple point-to-point communications, thereby successfully enabling broadband unicast to be applied in the millimeter wave MIMO system. However, future 5G requires multicast capability on the downlink broadcast channel, and the HBF architecture does not address the need for broadband broadcast/multicast communications.

Non-orthogonal multiple access techniques are a promising scheme that may enable 5G broadcast/multicast at the physical level. At the transmitting end, downlink NOMA (Non-orthogonal Multiple Access) adopts superposition coding, and realizes Multiple unicast or broadcast transmissions on the same frequency and time resources. At the receiving end, the received information is recovered by a multi-user detection algorithm.

The combination of millimeter wave MIMO and NOMA techniques may take advantage of the combined advantages of both, thereby improving the connection and capacity of the downlink broadcast link. Relevant studies show that compared with an orthogonal system, the millimeter wave MIMO-NOMA system has remarkable improvement in the throughput of the whole system. A typical beam-domain MIMO-NOMA system uses linear array antennas, where different signal data streams are distinguished in the power domain, and the number of users supported is greater than the number of rf links.

However, such systems have the disadvantage that power domain NOMA requires strict alignment of users in the corresponding beam direction, and user pairing and beam alignment are two major factors limiting the performance of the system. How to adapt the millimeter wave MIMO-NOMA technology to the 5G broadcast channel still lacks a relevant solution.

Disclosure of Invention

The invention aims to solve the problems of user pairing and accurate beam alignment in a downlink broadcast link in the prior art, and provides a millimeter wave MIMO-NOMA method of the downlink broadcast link based on beam aggregation; the method adopts a plurality of adjacent beams to provide service for a group of adjacent users in a non-orthogonal mode, and designs an approximately optimal digital precoding matrix by constructing a deep learning architecture, so that the interference among the users is reduced, and the difficulty of implementing the NOMA scheme in the MIMO is reduced.

The millimeter wave MIMO-NOMA method of the downlink broadcast link based on the beam aggregation comprises the following steps:

step one, covering a wave beam;

a cell base station is configured with a large-scale antenna array, a large-scale wave beam set capable of covering the whole cell is generated through wave beam forming, and the cell comprises N groups of paired user groups; in each group, N_RFThe adjacent beams are aggregated and cover K users in a non-orthogonal manner, wherein K>N_RF。

Step two, determining a user receiving signal;

digital precoding matrix of F_BB(ii) a Channel with a plurality of channels

Is N_RFA channel coefficient vector between each beam and the kth user; p_BSThe power of the base station is averagely distributed to each wave beam; received signal y at the k-th user_kComprises the following steps:

wherein f is^c _BB,kFor precoding F_BBThe k-th column of the matrix, v_kIs gaussian noise. s_kThe symbol corresponding to the kth user in the source symbol matrix;

the source symbols of the K users are:

wherein the transmission signal of the RF chain is:

wherein x_mIs the transmission signal of the mth RF chain.

Thirdly, recovering symbols of the user;

and in the step I, the user adopts a multi-user detection technology SIC algorithm for detection. The transmission rate of the kth user is R_k，R_kObtained by SIC algorithm.

The signal-to-noise ratio at the kth user is:

wherein the content of the first and second substances,

represents satisfaction

Of the condition

A set of subscripts q of (a); v. of_kIs Gaussian noise with variance of σ_k ². The transmission rate at the kth user is expressed as:

R_k(F_BB)＝log₂(1+SINR_k(F_BB)) (3)

in order to ensure fairness among multiple users, a digital precoding matrix is optimized based on a principle of maximizing a minimum transmission rate among users, and an optimization problem is represented as follows:

wherein

Is a precoding matrix F_BBThe (c) th row of (a),

is composed of

Norm 2 of (d).

The digital precoding matrix in the step two is given by the problem of formula (4a) P1, the problem is converted into an equal cost problem, that is, the maximum-minimum problem is converted into an equivalent equal cost problem, and the optimal precoding matrix for the P1 problem is defined as F, so that the following conclusion is obtained:

wherein R is_l(F^*) And R_m(F^*) Representing the transmission rates of the l and m users.

Therefore, the max-min optimization problem translates into an equal cost problem:

wherein, | R_l(F)-R_m(F) And l is the unfair cost among users, and lambda is the weight of the cost. When the value of lambda is larger, the absolute value of R in the formula is larger_l(F)-R_m(F) The larger the influence of the | cost item is, the more the fairness among the users can be ensured.

And step four, solving the problem of P2 in formula (6a) by adopting a deep learning method to obtain an approximately optimal digital precoding matrix.

The precoding matrix in the second step is given by the problem of formula (4a) P1, the problem is converted into the problem of formula (6a) P2 in the third step, the approximately optimal digital precoding matrix can be effectively obtained by constructing a neural network architecture and adopting a deep learning method, the interference among users is reduced, and the MIMO-NOMA method of the downlink broadcast link based on beam aggregation is realized.

Advantageous effects

1. The millimeter wave MIMO-NOMA method of the downlink broadcast link based on the beam aggregation is characterized in that a group of adjacent beams are aggregated together to cover a group of adjacent users in a non-orthogonal mode, and the method does not require strict beam alignment among the users, so that the range of covering the users is expanded, and the possibility of applying the NOMA scheme in the millimeter wave MIMO is increased finally.

2. The millimeter wave MIMO-NOMA method of the downlink broadcast link based on the beam aggregation designs an exquisite multi-user digital precoder, reduces the interference among users and ensures the fairness among the users.

3. The millimeter wave MIMO-NOMA method of the downlink broadcast link based on the beam aggregation converts the maximum-minimum problem into the equivalent cost problem, and can effectively obtain the approximately optimal digital precoder by adopting a deep learning scheme.

Drawings

FIG. 1 is a flow chart of the steps performed in the present invention;

FIG. 2 is a schematic diagram of an embodiment;

FIG. 3 is a functional block diagram of a deep learning process;

FIG. 4 is an exemplary DNN neural network architecture;

FIG. 5 is a comparison of matching probability with conventional power domain NOMA users based on beam aggregation;

FIG. 6 is a comparison of training performance and testing performance of the DNN neural network of the embodiment;

FIG. 7 is a comparison of the performance of DNN networks at different inputs;

FIG. 8 shows transmission rate for different beam angles;

FIG. 9 shows the transmission rate of the embodiment at different cell radius;

fig. 10 is a cumulative distribution function of transmission rates for the embodiment at different base station powers.

Detailed Description

In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without any creative efforts shall fall within the protection scope of the present invention.

The following describes specific steps of the embodiment of the present invention with reference to specific scenarios:

as shown in fig. 1, the millimeter wave MIMO-NOMA method for a downlink broadcast link based on beam aggregation specifically includes the following steps:

step one, generating beam coverage; arranging an antenna array at a cell base station, wherein 128 antennas are arranged, and users are randomly distributed in a cell range; in the link level simulation, considering a typical NOMA scenario in a downlink broadcast link, as shown in fig. 2, 2 adjacent beams serve 3 users, and the coverage angle of each beam is Δ. For these three users, the first and second users are located in the center region of the cell and at a distance of (0, d) from the base station, and the third user is located in the edge region of the cell and at a distance of [ d, d ']fromthe base station, where d is the radius of the center region of the cell and d' is the radius of the range of the cell.

Wherein

Is the channel gain for the user and,

step two, determining a user receiving signal;

digital precoding matrix of F_BB(ii) a Channel with a plurality of channels

where, i, k ∈ {1,2,3}, f^c _BB,kFor precoding F_BBThe k-th column of the matrix, v_kIs Gaussian noise, P_RFFor the power of each beam, s_kThe symbol corresponding to the kth user in the source symbol matrix;

the source symbols of 3 users are:

the transmission signals of the 2 RF chains are:

wherein x_mIs the transmission signal of the mth RF chain.

The relationship between the source symbol and the transmission signal of the RF chain is:

x＝F_BBS (9)

and thirdly, using an SIC multi-user detector at the user side, and at the user position of the cell center area, firstly recovering and deleting the signals of the users at the cell edge, and then recovering the signals of the users in the cell center area. The signal-to-noise ratio for each user is as follows:

the transmission rate of each user is calculated according to the following formula:

R_k(F_BB)＝log₂(1+SINR_k(F_BB)) (12)

wherein

Is a precoding matrix F_BBThe (c) th row of (a),

is composed of

Norm 2 of (d).

The digital precoding matrix in the step two is given by the problem of a formula (13a) P1, the optimization problem is non-trivial and cannot be directly solved, and the near-optimal multi-user digital precoder is effectively obtained by adopting a deep learning mode. But since the form of the P1 problem is not suitable for deep learning solution, the max-min problem is translated into an equal cost problem.

The proof method that the max-min optimization problem can be converted into an equivalent cost problem is as follows:

setting F^*Being the optimal solution of the P1 problem, i.e. the optimal precoding matrix, then:

wherein

As any other possible solution.

Assuming that the transmission rate of the ith user is minimum, it is

The following assumptions are made in accordance with equation (8), assuming

So that

R_i(F^*)<R_p(F^*)(16)

Wherein

The complex gain on the l-th beam for the i-th user can be expressed as

Wherein

Power allocated to ith user on ith beam, phi_l,iFor the phase of the ith user offset on the ith beam, assume that a positive infinitesimal quantity Δ f is added to

In this case, a new precoding matrix F' is obtained, so:

R_i(F′)＝R_i(F^*)+g_i(|Δf|) (17)

R_p(F′)＝R_p(F^*)-g_p(|Δf|) (18)

wherein g is_i(|Δf|)>0，g_p(|Δf|)>0 represents the amount of change due to Δ f, which causes an increase in the transmission rate for the ith user and a decrease in the transmission rate for the other users, respectively.

There is a sufficiently small variable | Δ f' |, which satisfies:

that is, under the action of | Δ F '|, the transmission rate of the ith user is still the minimum of all users, and the precoding matrix at this time is F', which satisfies:

is obtained in combination (14)

I.e. F^*It is not an optimal solution, and therefore, assuming that equation (16) does not hold, equation (5) holds.

Thus, the max-min optimization problem can be translated into an equivalency problem, which is:

and step four, obtaining an approximately optimal digital pre-coding matrix by adopting a deep learning method. A dnn (deep neural network) deep learning architecture is constructed, an optimal digital precoding matrix is obtained according to channel state information in an offline training and online application manner, in this embodiment, an unsupervised deep learning is adopted to solve the problem of equation (23a) P2 in step three, and a schematic diagram of the method is shown in fig. 3.

DNN is set as follows:

the inputs to DNN are:

or

Wherein

For channel state information, P_BSIs the power of the base station.

The output of DNN is:

wherein

Representing the complex precoding coefficients of the kth user at the mth beam.

For a detailed structure of DNN, a multi-layer backward neural network is used to obtain a non-linear mapping between input and output, and the architecture of DNN is shown in fig. 4, and the DNN architecture includes 1 input layer, 1 output layer and 3 hidden layers.

The mapping of the r-th layer to the r + 1-th layer is:

fl_r＝W_rX+b_r(25)

wherein X is the input of r layers, W_rAnd b_rAnd the parameters of r layers represent a weight matrix and a deviation vector.

By using

Represents the mapping of the entire DNN,

represents:

where σ () represents the activation function,

representing the generated precoding matrix.

Obtaining a final precoding matrix

Subjecting the obtained

Normalization:

wherein

Represents

The (c) th row of (a),

represents

Row j of (2).

The DNN training process is as follows:

during the training process, the parameters of the gradient adjustment DNN are set to minimize the loss function.

First, the loss function is defined as the inverse of the problem of equation (16a) P2 in step three:

wherein F is obtained by the formula (26) N (I).

Defining the parameter of N (I) as w_FIn the process of backward propagation, w_FUpdating by random gradient descent

Wherein

N_batchRepresents the size of the training set during t training sessions and phi (t) represents the rate of training.

Wherein

In one embodiment, the DNN architecture includes one input layer, one output layer, and three hidden layers. The three hidden layers respectively comprise 60, 128 and 36 neuron activation functions f (x) as follows:

base station power P in DNN input_BSThe random selection was within a range, with data set sizes of 100000 and 50000, batch size of 256, and initial learning rate of 0.0003.

The DNN neural network architecture is utilized to effectively obtain an approximately optimal digital precoding matrix, and the MIMO-NOMA method of the downlink broadcast link based on beam aggregation is realized.

In the steps of the above embodiment, simulations are performed in different scenarios, thereby illustrating the beneficial effects of the present invention.

As a simulation result, fig. 5 compares the user matching probability based on the beam aggregation and the normal power domain NOMA, and shows that the amount of users covered by the beam of the base station is obviously increased after the method is implemented.

For example, as shown in fig. 6, the training performance and the testing performance of the DNN neural network are compared, which indicates that the loss function is converged after the network training, and the testing result substantially matches the training result, so that the network can obtain an approximately optimal solution to the P1 problem.

As a result of the simulation FIG. 7, the performance of a DNN network employing two different inputs of DNN in the fourth step is compared, indicating that P is employed_BSThe DNN network as input can achieve lower loss function and has wider application range, which means that the neural network can utilize P_BSInformation and accommodate a wider range of SNRs. At the same time, it can be seen that P_BSAt 40dBm, the performance is worse than 30dBm, indicating a suitable P_BSThe selection may result in better performance.

As shown in simulation results 8 and 9, which tested the average max-min user transmission rate under different simulation conditions, this indicates that the method can obtain better transmission rate than the conventional power domain NOMA. While fig. 8 shows that the average max-min transmission rate decreases as the beam angle becomes larger, fig. 9 shows that the average max-min transmission rate decreases as the radius from the base station becomes larger. Simulation results show that the method can achieve higher fairness among users and weaken the influence of beam deviation.

As shown in simulation result 10, the cumulative distribution function of the transmission rates of the embodiments at different base station powers was tested, which indicates that the method has higher transmission rate in the whole frequency spectrum compared with the normal power domain NOMA.

The above detailed description is intended to illustrate the objects, aspects and advantages of the present invention, and it should be understood that the above detailed description is only exemplary of the present invention and is not intended to limit the scope of the present invention, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims

1. The millimeter wave MIMO-NOMA method of the downlink broadcast link based on the beam aggregation is characterized in that: comprises the following steps:

step one, covering a wave beam;

a cell base station is configured with a large-scale antenna array, a large-scale wave beam set capable of covering the whole cell is generated through wave beam forming, and the cell comprises N groups of paired user groups; in each group, N_RFThe adjacent beams are aggregated and cover K users in a non-orthogonal manner, where K > N_RF；

Step two, determining a user receiving signal;

digital precoding matrix of F_BB(ii) a Channel with a plurality of channels

wherein f is^c _BB，kFor precoding F_BBThe k-th column of the matrix, v_kIs Gaussian noise; s_kThe symbol corresponding to the kth user in the source symbol matrix;

the source symbols of the K users are:

wherein the transmission signal of the RF chain is:

wherein x_mA transmission signal for an mth RF chain;

thirdly, recovering symbols of the user;

in the first step, a user adopts a multi-user detection technology SIC algorithm for detection; the transmission rate of the kth user is R_k，R_kObtained by SIC algorithm;

the signal-to-noise ratio at the kth user is:

wherein the content of the first and second substances,

represents satisfaction

Of the condition

A set of subscripts q of (a); v. of_kIs Gaussian noise with variance of σ_k ²；

The transmission rate at the kth user is expressed as:

R_k(F_BB)＝log₂(1+SINR_k(F_BB)) (3)

wherein

Is a precoding matrix F_BBThe (c) th row of (a),

is composed of

Norm of order 2;

the digital precoding matrix in the step two is given by the problem of the formula (4a) P1, the problem is converted into an equal cost problem, namely, the maximum-minimum problem is converted into an equivalent equal cost problem, and the optimal precoding matrix aiming at the P1 problem is defined as F^*Then, the following conclusions are reached:

wherein R is_l(F^*) And R_m(F^*) Representing the transmission rates of the l and m users;

s.t.||f_j ^r||²≤1，1≤j≤N_RF(6b)

wherein, | R_l(F)-R_m(F) L is the unfair cost among users, and lambda is the weight of the cost;

step four, solving the problem of P2 in formula (6a) by adopting a deep learning method to obtain an approximately optimal digital pre-coding matrix;

2. The method of claim 1, wherein: step three, the lambda is the weight of the cost, and when the value of the lambda is larger, the absolute value of R in the formula (6a) is larger_l(F)-R_m(F) The larger the influence of the | cost item is, the more the fairness among the users can be ensured.

3. The method of claim 1, wherein: the proof method that the max-min optimization problem can be converted into an equivalent cost problem is as follows:

suppose F^*Being the optimal solution of the P1 problem, i.e. the optimal precoding matrix, then:

wherein

As any other possible solution;

assuming that the transmission rate of the ith user is minimum, it is

The following assumptions are made in accordance with equation (8), assuming

P is more than or equal to 1 and less than or equal to K, p is not equal to i, so that

R_i(F^*)＜R_p(F^*) (9)

Wherein

The complex gain on the l-th beam for the i-th user can be expressed as

Wherein

Power allocated to ith user on ith beam, phi_l，iFor the phase of the ith user offset on the ith beam, assume that a positive infinitesimal quantity Δ f is added to

In this case, a new precoding matrix F' is obtained, so:

R_i(F′)＝R_i(F^*)+g_i(|Δf|) (10)

R_p(F′)＝R_p(F^*)-g_p(|Δf|) (11)

wherein g is_i(|Δf|)＞0，g_p(| Δ f |) > 0 represents the amount of change due to Δ f, respectively, Δ f causing an increase in the transmission rate for the ith user and a decrease in the transmission rate for the other users;

there is a sufficiently small variable | Δ f' |, which satisfies:

in combination (7) to obtain

I.e. F^*Is not an optimal solution, so it is assumed that equation (9) does not hold and it is concluded that equation (5) holds;

s.t.||f_j ^r||²≤1，1≤j≤N_RF(16b) 。