CN115632726B - Model-driven ISTA-Net uplink scheduling-free large-scale MIMO-NOMA signal detection method - Google Patents

Model-driven ISTA-Net uplink scheduling-free large-scale MIMO-NOMA signal detection method Download PDF

Info

Publication number
CN115632726B
CN115632726B CN202211652733.4A CN202211652733A CN115632726B CN 115632726 B CN115632726 B CN 115632726B CN 202211652733 A CN202211652733 A CN 202211652733A CN 115632726 B CN115632726 B CN 115632726B
Authority
CN
China
Prior art keywords
ista
net
signal detection
noma
model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202211652733.4A
Other languages
Chinese (zh)
Other versions
CN115632726A (en
Inventor
李飞
宋晶科
曹倩
王媛媛
李汀
余杰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangsu Broadcasting Cable Information Network Co ltd
Nanjing University of Posts and Telecommunications
Original Assignee
Jiangsu Broadcasting Cable Information Network Co ltd
Nanjing University of Posts and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Jiangsu Broadcasting Cable Information Network Co ltd, Nanjing University of Posts and Telecommunications filed Critical Jiangsu Broadcasting Cable Information Network Co ltd
Priority to CN202211652733.4A priority Critical patent/CN115632726B/en
Publication of CN115632726A publication Critical patent/CN115632726A/en
Application granted granted Critical
Publication of CN115632726B publication Critical patent/CN115632726B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B17/00—Monitoring; Testing
    • H04B17/30—Monitoring; Testing of propagation channels
    • H04B17/309—Measuring or estimating channel quality parameters
    • H04B17/336—Signal-to-interference ratio [SIR] or carrier-to-interference ratio [CIR]
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B1/00—Details of transmission systems, not covered by a single one of groups H04B3/00 - H04B13/00; Details of transmission systems not characterised by the medium used for transmission
    • H04B1/69—Spread spectrum techniques
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B17/00—Monitoring; Testing
    • H04B17/30—Monitoring; Testing of propagation channels
    • H04B17/391—Modelling the propagation channel
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
    • H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
    • H04B7/0413—MIMO systems
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • Electromagnetism (AREA)
  • Quality & Reliability (AREA)
  • Mobile Radio Communication Systems (AREA)
  • Radio Transmission System (AREA)

Abstract

Aiming at the huge challenge brought to signal detection by a receiving end of a non-scheduling NOMA system that the active state information of a user is completely unknown, the traditional ISTA algorithm is improved in a momentum acceleration mode, and then the improved ISTA algorithm is deeply networked by using a model-driven idea to form an ISTA-Net signal detection scheme. On the basis, according to the signal detection result of the ISTA-Net, an active user judgment scheme of 'obvious jump for the first time' is adopted to further improve the detection performance. Finally, experiments prove that the ISTA-Net detection scheme based on the judgment of the active user with the first obvious jump has feasibility in a large-scale MIMO-NOMA system without scheduling.

Description

Model-driven ISTA-Net uplink scheduling-free large-scale MIMO-NOMA signal detection method
Technical Field
The invention belongs to the field of signal detection of a large-scale MIMO-NOMA system, and relates to an ISTA-Net uplink scheduling-free large-scale MIMO-NOMA signal detection method based on model driving.
Background
In recent years, the access demand of wireless communication is exponentially increased, and overload access on the same time-frequency resource is inevitable. The NOMA technology can enable a plurality of devices to communicate on the same time-frequency resource at the same time, and the requirement of a large-scale wireless network is met. Therefore, the NOMA is introduced into the massive MIMO system, so that the performance of the massive MIMO system is obviously improved, and overload access is realized.
The scheduling-free access mechanism and the scheduling-free NOMA combined with non-orthogonal signal superposition meet the requirements of low-delay and wide access in the current 5G and future 6G mobile communication systems. But also brings some problems, because of the distributed instant transmission in the non-scheduling NOMA system, the receiving end is unknown to the state information of the active user, which brings huge challenges to the signal detection of the receiving end. But fortunately, compressed sensing provides theoretical guarantee for sparse signal recovery, and is widely applied to detection of uplink non-scheduling NOMA active users.
At present, many schemes for uplink non-scheduling NOMA active user detection are analyzed and solved based on CS theory, such as OMP and IHT algorithms; however, most of these methods analyze with the knowledge of the number of active users, which is generally not relevant to the actual communication scenario. Meanwhile, most uplink scheduling-free NOMA active user detection schemes consider a single-antenna scene at a base station end, and multi-antenna signal detection is more appropriate to a real communication scene. Therefore, in the uplink scheduling-free massive MIMO-NOMA system, it is important to find a novel signal detection scheme which does not need to know the number of active users and has good detection performance.
Disclosure of Invention
The invention mainly aims at the signal detection problem of uplink scheduling-free large-scale MIMO-NOMA and provides an ISTA-Net detection scheme based on 'first obvious jump' active user judgment. In order to realize the technical purpose, the technical scheme adopted by the invention comprises the following processes:
an ISTA-Net uplink scheduling-free large-scale MIMO-NOMA signal detection method based on model driving comprises the following steps:
step 1, obtaining an improved ISTA algorithm based on momentum acceleration;
step 2, carrying out deep networking on the improved ISAT algorithm by using a model driving idea to obtain an ISTA-Net signal detection scheme based on model driving;
and 3, for the application scene of multiple antennas at the base station end, combining a first significant jump scheme on the basis of an ISTA-Net signal detection scheme, regarding the front of a jump point as an inactive user, regarding the rear of the jump point as an active user, finding the index value of the active user, and obtaining the support set of the active user
Figure 793791DEST_PATH_IMAGE001
；
Step 4, aiming at the support set in the step 3
Figure 18099DEST_PATH_IMAGE001
One antenna is selected, and a zero forcing algorithm is used to obtain high-precision signal detection.
Further, in step 1, the specific steps of Momentum acceleration are as follows:
Figure 17279DEST_PATH_IMAGE002
Figure 301630DEST_PATH_IMAGE003
wherein the content of the first and second substances,
Figure 714156DEST_PATH_IMAGE004
is an attenuation factor of less than 1 and,
Figure 679838DEST_PATH_IMAGE005
for the step size, will
Figure 861421DEST_PATH_IMAGE006
Specifically, the improved ISTA is:
Figure 51094DEST_PATH_IMAGE007
Figure 387135DEST_PATH_IMAGE008
Figure 422087DEST_PATH_IMAGE009
due to each
Figure 192597DEST_PATH_IMAGE010
、
Figure 756434DEST_PATH_IMAGE011
Attenuation factors are all less than 1, hence are taken for
Figure 612394DEST_PATH_IMAGE012
Of greater influence
Figure 716616DEST_PATH_IMAGE010
Namely:
Figure 341633DEST_PATH_IMAGE013
in this case, the modified ISTA is:
Figure 873108DEST_PATH_IMAGE014
Figure 950786DEST_PATH_IMAGE015
wherein
Figure 97514DEST_PATH_IMAGE016
，
Figure 577037DEST_PATH_IMAGE017
Residual residue for t iterationsThe difference vector is a vector of the difference between the two vectors,
Figure 279414DEST_PATH_IMAGE018
a factor for controlling the residual vector; then, after writing the ISTA to the general expression:
Figure 109967DEST_PATH_IMAGE019
Figure 555991DEST_PATH_IMAGE020
Figure 155600DEST_PATH_IMAGE021
Figure 28878DEST_PATH_IMAGE022
and using the approximate LMMSE matrix in OAMP,
Figure 346727DEST_PATH_IMAGE023
；
the improved ISTA scheduling-free NOMA system is obtained based on the combination of active user detection and data recovery.
Further, in step 2, for the control residual vector factor in the improved ISAT algorithm
Figure 65284DEST_PATH_IMAGE024
、
Figure 784979DEST_PATH_IMAGE025
And threshold factor in soft threshold function
Figure 829158DEST_PATH_IMAGE026
The ISTA and the deep learning DL are combined, the improved ISTA is developed through the DL, uncertain parameters in the ISTA scheme are learned through data, and the ISTA scheme based on model driving is called as ISTA-Net signal detectionAnd (4) scheme.
Further, for control residual vector factors
Figure 634303DEST_PATH_IMAGE024
、
Figure 422130DEST_PATH_IMAGE025
Each iteration of the modified ISTA is expanded to a layer while leaving the residual factors
Figure 261910DEST_PATH_IMAGE027
Which is different in each iteration and which is,
Figure 975526DEST_PATH_IMAGE028
then the DL learns continuously through back propagation, which yields:
Figure 2388DEST_PATH_IMAGE029
Figure 593906DEST_PATH_IMAGE030
。
further, for the threshold factor
Figure 22614DEST_PATH_IMAGE031
Learning based on neural networks, making the threshold factor a learnable parameter in ISTA-Net
Figure 408596DEST_PATH_IMAGE033
Meanwhile, a common nonlinear activation function relu in the neural network is used for partially replacing the soft function of soft; wherein, the relu function formula is as follows:
Figure 188333DEST_PATH_IMAGE034
then the activation function of ISTA-Net
Figure 317963DEST_PATH_IMAGE035
Comprises the following steps:
Figure 601177DEST_PATH_IMAGE036
therefore, a learnable threshold parameter is included
Figure 689218DEST_PATH_IMAGE033
The output of the ith layer of the ISTA-Net is as follows:
Figure 690672DEST_PATH_IMAGE037
。
further, step 3 includes the following steps:
step 3-1, N antennas at the base station end respectively utilize an ISTA-Net detection scheme to obtain estimated values of N sending signals, namely
Figure 623993DEST_PATH_IMAGE038
；
Step 3-2, according to the judgment scheme of the active users with the first significant jump, a support set of the active users is obtained
Figure 761714DEST_PATH_IMAGE039
(ii) a N in the above steps
Figure 755077DEST_PATH_IMAGE040
Accumulating the absolute values to obtain an accumulation result
Figure 243828DEST_PATH_IMAGE041
Sorting and finding out a jump point; wherein
Figure 980839DEST_PATH_IMAGE042
Is shown for N
Figure 238645DEST_PATH_IMAGE043
Accumulating the absolute values and sequencing the absolute values in an ascending order;
Figure 402911DEST_PATH_IMAGE044
there are a total of K elements, i.e. the total number of users,
Figure 113378DEST_PATH_IMAGE045
then represent
Figure 654080DEST_PATH_IMAGE046
The kth largest component;
Figure 297551DEST_PATH_IMAGE047
a jump between an inactive user and an active user is indicated; so that the index value of the active user is satisfied
Figure 632718DEST_PATH_IMAGE048
The subscript index value of (2), the set of subscript index values is the support set of active users
Figure 361639DEST_PATH_IMAGE039
：
Figure 611093DEST_PATH_IMAGE049
Figure 843491DEST_PATH_IMAGE050
Figure 83979DEST_PATH_IMAGE051
Figure 300197DEST_PATH_IMAGE052
Figure 917123DEST_PATH_IMAGE053
。
Further, in step 4, the zero forcing algorithm is used to obtain signal detection specifically, if the jth antenna at the base station end is selected, thenSignal
Figure 4028DEST_PATH_IMAGE054
。
The invention has the beneficial effects that:
1) The traditional ISTA algorithm is improved in a momentum acceleration mode, and the problems that the signal detection performance of the traditional ISTA algorithm in a scheduling-free NOMA system is not ideal and the signal detection is difficult in the scheduling-free large-scale MIMO-NOMA system are solved;
2) The improved ISTA scheme is deeply expanded through a DL (downlink differential signaling) technology to obtain an ISTA-Net signal detection scheme based on model driving so as to determine parameters in an improved ISTA algorithm, thereby further improving the signal detection performance and having remarkable performance in processing the problems of scheduling-free NOMA (non-uniform access multiple access) combined active user detection and signal recovery;
3) The method is applied to the problem of scheduling-free large-scale MIMO-NOMA signal detection, and when a base station is expanded from a single antenna to multiple antennas, the number of active users at the end of the multi-antenna base station can be accurately obtained;
4) An ISTA-Net scheme based on judgment of 'first significant jump' active users is provided to process the problem of scheduling-free large-scale MIMO-NOMA signal detection, and simulation experiments show that the scheduling-free large-scale MIMO-NOMA signal detection method has superiority in an uplink scheduling-free large-scale MIMO-NOMA system, and is low in bit error probability BER and high in detection precision.
Drawings
FIG. 1 is a model of a NOMA system in an embodiment of the present invention.
Fig. 2 is a diagram of a transmission procedure of an uplink non-scheduled NOMA in the embodiment of the present invention.
Fig. 3 is a diagram of a model-driven DL framework structure in an embodiment of the present invention.
Fig. 4 is a flow chart of the flow of the tth of the detection of the ISTA-Net signal in the embodiment of the present invention.
Fig. 5 is a BER performance curve for each detection scheme at a system overload rate of 200% in an embodiment of the present invention.
Fig. 6 is a BER performance curve for each detection scheme at a system overload rate of 150% in an embodiment of the present invention.
Fig. 7 is a graph of the BER performance of ISTA-Net for different numbers of antennas at the base station in the embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further explained in detail by combining the drawings in the specification.
Consider an uplink scheduling-free massive MIMO-NOMA system, in which a base station end has N antennas to receive signals transmitted from K single-antenna devices. When transmitting data, the mobile device randomly selects a frequency band from a set of available frequency subbands. In the embodiment of the present invention, considering the overload situation, the device K in the cell is greater than the number of orthogonal subbands M, i.e. M < K.
In particular, bit streams transmitted by users are mapped to signals via modulation
Figure 680997DEST_PATH_IMAGE055
Then using the specific code word of each device
Figure 853352DEST_PATH_IMAGE056
And spreading the transmitted signal, wherein the length of the spreading sequence is M, and finally, overlapping and transmitting the information of all users to a base station end through M orthogonal subcarriers. Here, the spreading code words of K users are non-orthogonal. Receiving signal on jth receiving antenna of base station end
Figure 742811DEST_PATH_IMAGE057
Can be expressed as:
Figure 215381DEST_PATH_IMAGE058
Figure 797672DEST_PATH_IMAGE059
therein are
Figure 457323DEST_PATH_IMAGE060
The spreading code representing the k-th user,
Figure 416052DEST_PATH_IMAGE061
for the channel gain vector between the kth user and the current jth receiving antenna at the base station,
Figure 976084DEST_PATH_IMAGE062
it indicates the received signals of M sub-carriers on the jth antenna of the base station,
Figure 260435DEST_PATH_IMAGE063
is additive white Gaussian noise in the signal transmission process, and satisfies the mean value of 0 and the variance of
Figure 407382DEST_PATH_IMAGE064
I.e. by
Figure 904223DEST_PATH_IMAGE065
Wherein
Figure 820226DEST_PATH_IMAGE066
Is an M-dimensional unit matrix.
It is known that in the expression of the received signal,
Figure 9899DEST_PATH_IMAGE067
therefore, the received signal on the jth antenna of the base station
Figure 644143DEST_PATH_IMAGE068
The equivalent expression below can be written:
Figure 944674DEST_PATH_IMAGE069
Figure 980763DEST_PATH_IMAGE070
Figure 341337DEST_PATH_IMAGE071
wherein the content of the first and second substances,
Figure 197298DEST_PATH_IMAGE072
representing the channel gain matrix for the jth antenna with K users,
Figure 35941DEST_PATH_IMAGE073
for the spreading code word matrix of all K users,
Figure 192116DEST_PATH_IMAGE074
which represents the hadamard product, is,
Figure 723591DEST_PATH_IMAGE075
then represent
Figure 66848DEST_PATH_IMAGE076
The equivalent channel matrix of (2). Since the number M of available orthogonal resources is smaller than the number K of users served, the above equation is an underdetermined equation.
At this time, the received signals y on all N antennas at the base station end are obtained as follows:
Figure 974761DEST_PATH_IMAGE077
and is
Figure 719863DEST_PATH_IMAGE078
，
Figure 422240DEST_PATH_IMAGE079
Figure 518372DEST_PATH_IMAGE080
，
Figure 214931DEST_PATH_IMAGE081
。
Figure 80119DEST_PATH_IMAGE082
Representing a set of complex matrices.
In this embodiment, a traditional ISTA algorithm is continuously improved to finally obtain an ISTA-Net signal detection scheme based on "first significant jump" for detection and data recovery of active users in an uplink scheduling-free massive MIMO-NOMA system.
Firstly, the signal detection of the traditional ISTA algorithm in the non-scheduling NOMA system is introduced, and because the ISTA does not need to know the number of active users, the signal detection of the non-scheduling NOMA system is carried out by utilizing the ISTA algorithm in the CS.
The CS solves the sparse linear equation by solving the following optimization problem and derives a received signal from the received signal
Figure 953397DEST_PATH_IMAGE083
To reconstruct the transmitted signal x:
Figure 271246DEST_PATH_IMAGE084
Figure 255382DEST_PATH_IMAGE085
wherein
Figure 975076DEST_PATH_IMAGE086
For controlling the degree of fidelity of the image,
Figure 19256DEST_PATH_IMAGE087
for controlling the level of sparsification, and,
Figure 89980DEST_PATH_IMAGE088
to regularize the parameters, the balance between sparsity and minimization of error is controlled. The following describes the ISTA algorithm in detail to provide a theoretical basis for the following algorithm improvement.
First consider solving alone common
Figure 877807DEST_PATH_IMAGE089
The problem is that,for such a problem, a projective gradient descent method may be used, and each iteration is:
Figure 452008DEST_PATH_IMAGE090
wherein
Figure 667089DEST_PATH_IMAGE091
Is the iteration step size.
It is also possible to use a second order Taylor for f (x) solved separately
Figure 693951DEST_PATH_IMAGE092
Point approximation is unfolded, then
Figure 285469DEST_PATH_IMAGE093
Figure 714176DEST_PATH_IMAGE094
The minimum value of the above equation is:
Figure 100158DEST_PATH_IMAGE095
in the formula
Figure 879896DEST_PATH_IMAGE096
In a Taylor expansion
Figure 9526DEST_PATH_IMAGE097
In that
Figure 791274DEST_PATH_IMAGE098
The second derivative of the point. Now, a sparse term g (x) is added to solve the original problem, and the gradient descent method is used, which is equivalent to that each step iteration is as follows: (wherein
Figure 879316DEST_PATH_IMAGE099
And
Figure 880770DEST_PATH_IMAGE100
to be generalized, i.e. shaped like such an optimization problem)
Figure 548512DEST_PATH_IMAGE101
Figure 951811DEST_PATH_IMAGE102
Is easy to know, shaped as
Figure 945175DEST_PATH_IMAGE103
Is solved as soft: (
Figure 168346DEST_PATH_IMAGE104
;
Figure 905358DEST_PATH_IMAGE105
) The soft function is specifically:
Figure 163164DEST_PATH_IMAGE106
note that this should be for
Figure 61850DEST_PATH_IMAGE100
Each element in (a) performs a soft projection operation one by one. Wherein the content of the first and second substances,
Figure 37896DEST_PATH_IMAGE107
represent
Figure 313020DEST_PATH_IMAGE108
The kth element in (1). Handle type
Figure 425332DEST_PATH_IMAGE109
Bringing in type
Figure 760499DEST_PATH_IMAGE110
Zhongling
Figure 958262DEST_PATH_IMAGE111
Then can obtain the first
Figure 302655DEST_PATH_IMAGE112
Step iteration
Figure 535054DEST_PATH_IMAGE113
Specifically, the solution of (c) is:
Figure 775542DEST_PATH_IMAGE114
Figure 224716DEST_PATH_IMAGE115
Figure 841642DEST_PATH_IMAGE116
wherein, the first and the second end of the pipe are connected with each other,
Figure 928547DEST_PATH_IMAGE117
. The soft function of soft in the above equation can be equivalent to:
Figure 339936DEST_PATH_IMAGE022
wherein
Figure 777871DEST_PATH_IMAGE118
Representing updates one by one
Figure 464067DEST_PATH_IMAGE012
K elements in (c). The complete ISTA is expressed as the above two equations.
Through the detailed introduction of the ISTA algorithm, it is found that the performance of the traditional ISTA algorithm directly applied to the non-scheduling NOMA system for signal detection is not very good. Therefore, the specific steps of the embodiment of the invention for improving Momentum acceleration of the traditional ISTA by a Momentum acceleration method are as follows:
Figure 671058DEST_PATH_IMAGE002
Figure 784507DEST_PATH_IMAGE003
wherein the content of the first and second substances,
Figure 709738DEST_PATH_IMAGE119
is an attenuation factor of less than 1 and,
Figure 871729DEST_PATH_IMAGE120
for the speed at the time t, the speed,
Figure 198805DEST_PATH_IMAGE121
for the step size, will
Figure 483156DEST_PATH_IMAGE122
Specifically, the improved ISTA is:
Figure 630103DEST_PATH_IMAGE007
Figure 330206DEST_PATH_IMAGE123
Figure 246210DEST_PATH_IMAGE009
wherein
Figure 435882DEST_PATH_IMAGE125
Is that
Figure 70126DEST_PATH_IMAGE097
In that
Figure 869193DEST_PATH_IMAGE126
The second derivative of the point. Due to each
Figure 639702DEST_PATH_IMAGE127
、
Figure 277DEST_PATH_IMAGE128
Attenuation factors are all less than 1, hence are taken for
Figure 121816DEST_PATH_IMAGE012
Of greater influence
Figure 226039DEST_PATH_IMAGE119
Namely:
Figure 116634DEST_PATH_IMAGE013
in this case, the modified ISTA is:
Figure 648110DEST_PATH_IMAGE014
Figure 991366DEST_PATH_IMAGE015
wherein
Figure 633700DEST_PATH_IMAGE129
，
Figure 378802DEST_PATH_IMAGE130
For the remaining residual vector of iteration t times,
Figure 346758DEST_PATH_IMAGE131
to control the factors of the residual vector. Then, after writing the ISTA to a more general expression:
Figure 177311DEST_PATH_IMAGE019
Figure 623336DEST_PATH_IMAGE020
Figure 488524DEST_PATH_IMAGE021
Figure 361802DEST_PATH_IMAGE022
and using the approximate LMMSE matrix in OAMP,
Figure 414072DEST_PATH_IMAGE132
。
therefore, a combined active user detection and data recovery procedure based on the improved ISTA schedule-free NOMA system can be obtained as shown in algorithm 1.
Algorithm 1 based on improved ISTA Schedule NOMA Signal detection
Inputting: receiving signal at base station end
Figure 663787DEST_PATH_IMAGE133
Channel matrix
Figure 649061DEST_PATH_IMAGE134
A matrix of spread spectrum code words S,
maximum number of iterations max _ iter
And (3) outputting:
Figure 958820DEST_PATH_IMAGE135
initialization:
Figure 763964DEST_PATH_IMAGE136
equivalent channel matrix
Figure 259449DEST_PATH_IMAGE137
，
Figure 99229DEST_PATH_IMAGE138
，
Figure 579889DEST_PATH_IMAGE139
for t=1: max_iter
Figure 137909DEST_PATH_IMAGE140
Figure 729427DEST_PATH_IMAGE141
Figure 626976DEST_PATH_IMAGE142
End
Compared with the traditional ISTA algorithm, the improved ISTA scheme has improved performance in NOMA signal detection without scheduling, but factors for controlling residual vectors in the algorithm
Figure 278537DEST_PATH_IMAGE143
、
Figure 58275DEST_PATH_IMAGE144
And threshold factor in soft threshold function
Figure 453484DEST_PATH_IMAGE145
It is always poorly determined.
The deep learning DL is known to have strong learning capability, and the model-driven deep expansion has the advantages of both a model algorithm and DL learning data. Therefore, in the embodiment, the ISTA and the DL technology in the CS are combined, the improved ISTA is developed through the DL, meanwhile, the uncertain parameters in the ISTA scheme are learned through data, so that the detection performance of the ISTA is further improved, and meanwhile, the ISTA scheme based on model driving is called the ISTA-Net signal detection scheme.
In the general expression of ISTA, the factors controlling the residual vector
Figure 267856DEST_PATH_IMAGE146
、
Figure 28002DEST_PATH_IMAGE144
Not only is it difficult to determine, but it is also fixed in each iteration, so that the detection performance is limited. Inspired by the model-driven thought, each iteration of the improved ISTA is expanded into one layer, and meanwhile, the residual factors are enabled to be
Figure 295035DEST_PATH_IMAGE147
Which is different in each iteration and which is,
Figure 493935DEST_PATH_IMAGE148
then the DL is continuously learned through back propagation, and then:
Figure 162814DEST_PATH_IMAGE029
Figure 156178DEST_PATH_IMAGE030
in addition to this, a threshold factor
Figure 582611DEST_PATH_IMAGE145
(i.e., the regularization parameter above, which may be considered as a regularization parameter in the foregoing, evolves to the meaning that becomes the threshold in the following) is more difficult to determine. It is known that the threshold factor plays an important role in NOMA active user detection, which will decide whether the number of active users eventually recovered matches the actual number of active users. Therefore, the present embodiment will make the threshold factor be the learnable parameter in ISTA-Net by the learning capability of the neural network
Figure 585202DEST_PATH_IMAGE149
Meanwhile, a non-linear activating function relu commonly used in a neural network is used for partially replacing the soft function of soft. Wherein the content of the first and second substances,the relu function is formulated as follows:
Figure 108587DEST_PATH_IMAGE034
then the activation function of ISTA-Net
Figure 272852DEST_PATH_IMAGE150
Comprises the following steps:
Figure 514478DEST_PATH_IMAGE036
therefore, including a learnable threshold parameter
Figure 491399DEST_PATH_IMAGE149
The output of the ith layer of the ISTA-Net is as follows:
Figure 869291DEST_PATH_IMAGE151
in summary, the learnable parameter for ISTA-Net non-scheduling NOMA active user signal detection is
Figure 470036DEST_PATH_IMAGE152
Wherein, L is the network layer number of the ISTA-Net signal detection, and the total learning parameter number is 3L. The flowchart of the t-layer of the ISTA-Net non-scheduled NOMA signal detection is shown in FIG. 4.
In order to further improve the performance of signal detection of ISTA-Net in a non-scheduling NOMA system, aiming at the condition that a large-scale MIMI-NOMA system base station end is configured with multiple antennas, the embodiment of the invention provides a scheme of 'significant jump for the first time' active user judgment to improve the performance of signal detection.
When the base station end is no longer a single antenna, then N antennas of the base station end will obtain an estimated value of a transmission signal by using L-layer ISTA-Net signal detection schemes according to the above ISTA-Net detection schemes, that is, at this time, the absolute values of the estimated values of the N transmission signals are accumulated and sorted. Since the data of the inactive user approaches to 0, a jump point occurs from the inactive user to the active user, i.e. the scheme of "significant jump for the first time" proposed in this embodiment. At this time, all users in front of the jump point are considered to be inactive users, and all users behind the jump point are active users, so far, the index value of the active users, namely the support set, is found. Because the support set is obtained from N estimation signals, the accuracy is higher, and the data of the active users can be recovered only by using a simple zero forcing algorithm. The method comprises the following specific steps:
1. n antennas at the base station end respectively utilize an ISTA-Net detection scheme to obtain estimated values of N transmitted signals, namely
Figure 933379DEST_PATH_IMAGE153
。
2. According to the scheme of judging the active users by 'significant jump for the first time', a support set of the active users is obtained
Figure 277772DEST_PATH_IMAGE039
. Concretely, N in the above steps are used
Figure 447854DEST_PATH_IMAGE154
Accumulating the absolute values to obtain an accumulation result
Figure 219500DEST_PATH_IMAGE041
And sorting to find the jump point. Wherein
Figure 170139DEST_PATH_IMAGE155
Is shown for N
Figure 52644DEST_PATH_IMAGE156
The absolute values are accumulated and the results are sorted in ascending order.
Figure 405128DEST_PATH_IMAGE157
There are a total of K elements, i.e. the total number of users,
Figure 19780DEST_PATH_IMAGE158
then represent
Figure 192136DEST_PATH_IMAGE155
The kth largest component.
Figure 612753DEST_PATH_IMAGE159
It indicates a jump between an inactive user and an active user, and it is assumed in the embodiment of the present invention
Figure 554164DEST_PATH_IMAGE160
. So that the index value of the active user is satisfied
Figure 900569DEST_PATH_IMAGE161
The set of subscript index values is the support set of active users
Figure 560221DEST_PATH_IMAGE039
Figure 784529DEST_PATH_IMAGE049
Figure 846026DEST_PATH_IMAGE050
Figure 599218DEST_PATH_IMAGE051
Figure 746166DEST_PATH_IMAGE052
Figure 508585DEST_PATH_IMAGE053
3. The support set obtained according to the above steps
Figure 424589DEST_PATH_IMAGE039
Selecting one antenna, and using simplest zero forcing algorithm to obtain high-precision signal detection. For example: selecting the jth antenna at the base station end, then
Figure 879841DEST_PATH_IMAGE162
。
In conclusion, the specific flow of scheduling-free massive MIMO-NOMA signal detection and data recovery based on ISTA-Net is shown in algorithm 2.
Algorithm 2 is based on ISTA-Net scheduling-free large-scale MIMO-NOMA signal detection judged by 'first significant jump' active users
Inputting: the base station end receives the signal y, the channel matrix H, the spread spectrum code word matrix S, the network layer number L
And (3) outputting:
Figure 248505DEST_PATH_IMAGE135
initialization:
Figure 549037DEST_PATH_IMAGE136
equivalent channel matrix
Figure 585126DEST_PATH_IMAGE137
，
Figure 945700DEST_PATH_IMAGE139
Estimated signals on N antennas:
for j = 1:N
Figure 67240DEST_PATH_IMAGE163
for t = 1:N
Figure 905883DEST_PATH_IMAGE164
Figure 796478DEST_PATH_IMAGE165
Figure 327954DEST_PATH_IMAGE166
end
End
active user support set
Figure 936790DEST_PATH_IMAGE167
：
Figure 579123DEST_PATH_IMAGE168
Figure 822761DEST_PATH_IMAGE169
Figure 525138DEST_PATH_IMAGE170
Figure 355690DEST_PATH_IMAGE171
Figure 801715DEST_PATH_IMAGE172
And (3) data recovery:
selecting the jth antenna at the base station end, then
Figure 666903DEST_PATH_IMAGE173
Simulation experiments of the embodiment of the invention are all built on python and Tensorflow platforms. Firstly, setting parameters of a simulation experiment, and then verifying the feasibility of the ISTA-Net signal detection scheme based on the judgment of the active users with the first significant jump based on the different numbers of the active users, the different numbers of the sub-carriers (overload rates) and the different numbers of the antennas at the base station end.
Uplink scheduling-free MIMO-NOMA active user detection parameter based on ISTA-Net
The settings are shown in table 1, where the user spreading code sequence uses a pseudo random noise sequence.
TABLE 1 parameter settings for ISTA-Net uplink scheduling-free MIMO-NOMA based detection scheme
Figure 540181DEST_PATH_IMAGE174
The simulation results are shown in fig. 5, 6, and 7. It can be seen from fig. 5 and fig. 6 that, no matter the overload rate of the system is 200% or 150%, compared with the existing ISTA and OMP detection schemes, the detection scheme based on DL ISTA-Net has lower bit error probability BER and higher detection accuracy, and verifies the superiority of the ISTA-Net detection scheme in non-scheduling NOMA signal detection. In fig. 5, as the number of active users increases, the BER of each detection scheme gradually becomes larger, i.e., the detection performance is degraded. However, it can be seen that ISTA-Net can be better than the detection performance of OMP with an active user number of 10 when the active user number is 12, and can be better than the improved ISTA even at high signal-to-noise ratio.
Fig. 7 compares BER performance in scheduling-free massive MIMO-NOMA signal detection based on the "first significant hop" ISTA-Net scheme at different antenna counts at the base station side. With the increase of the number of the antennas at the base station end, the BER of the ISTA-Net detection scheme based on the first obvious jump is gradually reduced, which shows that the detection performance is gradually improved, and the feasibility of the ISTA-Net detection scheme based on the judgment of the active users of the first obvious jump in the uplink scheduling-free large-scale MIMO-NOMA is verified.
The above description is only a preferred embodiment of the present invention, and the scope of the present invention is not limited to the above embodiment, but equivalent modifications or changes made by those skilled in the art according to the present disclosure should be included in the scope of the present invention as set forth in the appended claims.

Claims (7)

1. A model-driven ISTA-Net uplink scheduling-free large-scale MIMO-NOMA signal detection method is characterized by comprising the following steps: the method comprises the following steps:
step 1, obtaining an improved ISTA algorithm based on momentum acceleration;
step 2, carrying out deep networking on the improved ISAT algorithm by using a model driving idea to obtain an ISTA-Net signal detection scheme based on model driving;
and 3, for the application scene of multiple antennas at the base station end, combining a first significant jump scheme on the basis of an ISTA-Net signal detection scheme, regarding the front of a jump point as an inactive user, regarding the rear of the jump point as an active user, finding the index value of the active user, and obtaining the support set of the active user
Figure 236322DEST_PATH_IMAGE001
；
Step 4, aiming at the supporting set in the step 3
Figure 503356DEST_PATH_IMAGE001
One antenna is selected, and a zero forcing algorithm is used to obtain high-precision signal detection.
2. The model-driven ISTA-Net uplink scheduling-free massive MIMO-NOMA signal detection method according to claim 1, characterized in that: in step 1, the specific steps for Momentum acceleration of Momentum are as follows:
defining parameters
Figure 764573DEST_PATH_IMAGE003
And
Figure 167872DEST_PATH_IMAGE004
：
Figure 426815DEST_PATH_IMAGE005
Figure 181145DEST_PATH_IMAGE002
wherein, the first and the second end of the pipe are connected with each other,
Figure 918156DEST_PATH_IMAGE006
for a decay factor less than 1 at time t,
Figure 753126DEST_PATH_IMAGE007
for the speed at the time t, the speed,
Figure 917391DEST_PATH_IMAGE008
for step size, the reconstructed transmission signal
Figure 627858DEST_PATH_IMAGE009
Solving function by projection gradient descent method
Figure 902982DEST_PATH_IMAGE010
Specifically, the improved ISTA is:
Figure 546453DEST_PATH_IMAGE011
Figure 740674DEST_PATH_IMAGE012
Figure 204016DEST_PATH_IMAGE013
wherein
Figure 17251DEST_PATH_IMAGE014
Is that
Figure 249649DEST_PATH_IMAGE016
In that
Figure 365504DEST_PATH_IMAGE018
Second derivative of point, due to each
Figure 316143DEST_PATH_IMAGE006
、
Figure 933069DEST_PATH_IMAGE019
Attenuation factors are all less than 1, hence
Figure 285553DEST_PATH_IMAGE020
Of greater influence
Figure 962522DEST_PATH_IMAGE021
Namely:
Figure 259511DEST_PATH_IMAGE017
in this case, the modified ISTA is:
Figure 414549DEST_PATH_IMAGE022
Figure 621539DEST_PATH_IMAGE023
wherein
Figure 203830DEST_PATH_IMAGE024
，
Figure 237383DEST_PATH_IMAGE025
For the remaining residual vectors of the iteration t times,
Figure 196112DEST_PATH_IMAGE026
in order to control the factors of the residual vector,
Figure 257609DEST_PATH_IMAGE027
for the received signal on the jth antenna of the base station,
Figure 541959DEST_PATH_IMAGE028
then represent
Figure 688907DEST_PATH_IMAGE029
The equivalent channel matrix of (a) is,
Figure 310381DEST_PATH_IMAGE030
regularizing the parameters while taking them as threshold factors in a soft threshold function; then, after writing ISTA into the general expression:
Figure 226385DEST_PATH_IMAGE031
Figure 416057DEST_PATH_IMAGE032
Figure 50301DEST_PATH_IMAGE033
Figure 226199DEST_PATH_IMAGE034
and using the approximate LMMSE matrix in OAMP,
Figure 996709DEST_PATH_IMAGE035
；
the method is based on the improved ISTA scheduling-free NOMA system to combine active user detection and data recovery.
3. The model-driven ISTA-Net uplink scheduling-free massive MIMO-NOMA signal detection method according to claim 1, characterized in that: in step 2, forControl residual vector factor in improved ISAT algorithm
Figure 622862DEST_PATH_IMAGE036
、
Figure 744402DEST_PATH_IMAGE037
And threshold factor in soft threshold function
Figure 583045DEST_PATH_IMAGE038
The method combines the ISTA and the deep learning DL, develops improved ISTA through DL, and learns uncertain parameters in the ISTA scheme through data, so that the ISTA scheme based on model driving is called the ISTA-Net signal detection scheme.
4. The model-driven ISTA-Net uplink scheduling-free massive MIMO-NOMA signal detection method according to claim 3, characterized in that: for control residual vector factor
Figure 598274DEST_PATH_IMAGE036
、
Figure 129750DEST_PATH_IMAGE037
Each iteration of the modified ISTA is expanded to a layer while leaving the residual factors
Figure 473006DEST_PATH_IMAGE039
Which is different in each iteration and which,
Figure 646499DEST_PATH_IMAGE040
then the DL learns continuously through back propagation, which yields:
Figure 391601DEST_PATH_IMAGE041
Figure 485457DEST_PATH_IMAGE042
。
5. the model-driven ISTA-Net uplink scheduling-free large-scale MIMO-NOMA signal detection method based on the claim 3 is characterized in that: for threshold factor
Figure 316010DEST_PATH_IMAGE038
Learning based on neural networks, making the threshold factor a learnable parameter in ISTA-Net
Figure 496455DEST_PATH_IMAGE044
Meanwhile, a common nonlinear activation function relu in the neural network is used for partially replacing the soft function of soft; wherein, the relu function formula is as follows:
Figure 361643DEST_PATH_IMAGE045
then the activation function of ISTA-Net
Figure 359555DEST_PATH_IMAGE046
Comprises the following steps:
Figure 677404DEST_PATH_IMAGE047
therefore, including a learnable threshold parameter
Figure 661541DEST_PATH_IMAGE044
The output of the ith layer of the ISTA-Net is as follows:
Figure 646814DEST_PATH_IMAGE048
。
6. the model-driven ISTA-Net uplink scheduling-free massive MIMO-NOMA signal detection method according to claim 1, characterized in that: in the step 3, the method comprises the following steps:
step 3-1, N antennas at the base station end respectively utilize an ISTA-Net detection scheme to obtain estimated values of N sending signals, namely
Figure 690993DEST_PATH_IMAGE049
；
Step 3-2, according to the judgment scheme of the active users with the first significant jump, a support set of the active users is obtained
Figure 371505DEST_PATH_IMAGE001
(ii) a N in the above steps
Figure 159332DEST_PATH_IMAGE050
The absolute values are accumulated to obtain an accumulated result
Figure 733533DEST_PATH_IMAGE051
Sorting and finding out a jump point; wherein
Figure 948614DEST_PATH_IMAGE052
Is shown for N
Figure 100109DEST_PATH_IMAGE050
Accumulating the absolute values and sequencing the absolute values in an ascending order;
Figure 691628DEST_PATH_IMAGE053
there are a total of K elements, i.e. the total number of users,
Figure 120335DEST_PATH_IMAGE054
then represent
Figure 506317DEST_PATH_IMAGE055
The kth largest component;
Figure 659955DEST_PATH_IMAGE056
indicating the jump between the inactive user and the active user, wherein L is the network layer number detected by the ISTA-Net signal; so that the index value of the active user is satisfied
Figure 789585DEST_PATH_IMAGE057
The set of subscript index values is the support set of active users
Figure 338378DEST_PATH_IMAGE001
：
Figure 426420DEST_PATH_IMAGE058
Figure 693453DEST_PATH_IMAGE059
Figure 423512DEST_PATH_IMAGE060
Figure 92391DEST_PATH_IMAGE061
Figure 85755DEST_PATH_IMAGE062
。
7. The model-driven ISTA-Net uplink scheduling-free large-scale MIMO-NOMA signal detection method based on the claim 1 is characterized in that: in step 4, the zero forcing algorithm is used to obtain signal detection, specifically, if the jth antenna of the base station end is selected, the signal is detected
Figure 184292DEST_PATH_IMAGE063
。
CN202211652733.4A 2022-12-22 2022-12-22 Model-driven ISTA-Net uplink scheduling-free large-scale MIMO-NOMA signal detection method Active CN115632726B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211652733.4A CN115632726B (en) 2022-12-22 2022-12-22 Model-driven ISTA-Net uplink scheduling-free large-scale MIMO-NOMA signal detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211652733.4A CN115632726B (en) 2022-12-22 2022-12-22 Model-driven ISTA-Net uplink scheduling-free large-scale MIMO-NOMA signal detection method

Publications (2)

Publication Number Publication Date
CN115632726A CN115632726A (en) 2023-01-20
CN115632726B true CN115632726B (en) 2023-02-28

Family

ID=84909774

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211652733.4A Active CN115632726B (en) 2022-12-22 2022-12-22 Model-driven ISTA-Net uplink scheduling-free large-scale MIMO-NOMA signal detection method

Country Status (1)

Country Link
CN (1) CN115632726B (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108650689A (en) * 2018-04-03 2018-10-12 华南理工大学 Wireless portable communications system efficiency optimization method based on NOMA downlinks
CN110677178A (en) * 2019-08-28 2020-01-10 华北电力大学（保定） Short packet transmission delay analysis method in large-scale MIMO-NOMA system

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108650689A (en) * 2018-04-03 2018-10-12 华南理工大学 Wireless portable communications system efficiency optimization method based on NOMA downlinks
CN110677178A (en) * 2019-08-28 2020-01-10 华北电力大学（保定） Short packet transmission delay analysis method in large-scale MIMO-NOMA system

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
曹倩等.《Adaptive Signal Detection Method Based on Model-Driven for Massive MIMO Systems》.2021,第1页-第5页. *
李飞等.《去蜂窝大规模多输入多输出非正交多址系统中基于量子菌群优化的接入点选择方案》.2022,第44卷(第0期),第1页-第8页. *

Also Published As

Publication number Publication date
CN115632726A (en) 2023-01-20

Similar Documents

Publication Publication Date Title
CN107483088B (en) Large-scale MIMO robust precoding transmission method
KR101381130B1 (en) Multi-user precoding and scheduling method and base station for implementing the method
CN104702390B (en) Pilot distribution method in the estimation of distributed compression channel perception
CN110233653B (en) Blind multipath identification method and system of MIMO system based on weighted integration clustering algorithm
WO2008113216A1 (en) A channel estimation method
Aref et al. Deep learning-aided successive interference cancellation for MIMO-NOMA
CN113114313A (en) Method, system and storage medium for detecting pilot auxiliary signal of MIMO-NOMA system
CN101026435A (en) Low-complexity maximum likelihood detecting method and device for communication system
CN112215335A (en) System detection method based on deep learning
CN101247373B (en) Dynamic channel equalization method for orthogonal frequency division multiplexing system based on immune network
CN108566227B (en) Multi-user detection method
Wang et al. Collision-tolerant media access control for asynchronous users over frequency-selective channels
CN104980202B (en) Uplink detection method based on MCMC technologies in a kind of extensive mimo system
CN108206795A (en) Blind authentication method and system based on the frequency selective fading channels that confidence is transmitted
CN115632726B (en) Model-driven ISTA-Net uplink scheduling-free large-scale MIMO-NOMA signal detection method
CN101258704B (en) Transmitter, receiver and method thereof
CN107171990A (en) Based on fractional spaced and iterative algorithm single-carrier signal time domain equalization method
WO2014187356A1 (en) Multiple-input multiple-output (mimo) detection method, apparatus and system for transmitting signal
CN113595680B (en) Sparsity and step length adaptive matching tracking method based on generalized Jaccard coefficient
Yuan et al. Soft-decision-aided, smoothness-constrained channel estimation over time-varying fading channels with no channel model information
US11309991B2 (en) Wireless receiver apparatus
JP2023037446A (en) Radio receiving device and method thereof
JP2006333485A (en) Partial parallel interference cancel receiver and partial parallel interference cancel method
De Souza et al. A novel signal detector in MIMO systems based on complex correntropy
CN107248876B (en) Generalized spatial modulation symbol detection method based on sparse Bayesian learning

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant