CN112215335B - System detection method based on deep learning - Google Patents
System detection method based on deep learning Download PDFInfo
- Publication number
- CN112215335B CN112215335B CN202011024526.5A CN202011024526A CN112215335B CN 112215335 B CN112215335 B CN 112215335B CN 202011024526 A CN202011024526 A CN 202011024526A CN 112215335 B CN112215335 B CN 112215335B
- Authority
- CN
- China
- Prior art keywords
- neural network
- layer
- hidden layer
- deep learning
- detection method
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
Abstract
The invention discloses a system detection method based on deep learning, which comprises the steps of determining a neural network, wherein the neural network comprises an input layer, a hidden layer and an output layer; obtaining a relation transmitted from an output layer to a hidden layer and a relation transmitted from the hidden layer to the output layer, and determining the layer number of the hidden layer; determining the weight change amount of the neural network and the threshold change amount of the neural network; performing channel estimation on the received signal based on a neural network; obtaining a set of filter coefficients and determining initial values of weights of the neural network based on the set of filter coefficients; calculating a cost function of the network training of the neural network; outputting the corrected weight of the neural network, the corrected threshold of the neural network, the number of layers of the hidden layer and the cost function of the network training of the neural network, and accelerating the convergence speed of deep learning. Under the off-line deep learning mode, the system structure is optimized, and the overall performance of the system is improved.
Description
Technical Field
The invention relates to the technical field of mobile communication, in particular to a system detection method based on deep learning.
Background
Mobile communication technology has evolved to date, and spectrum resources have been increasingly strained. Meanwhile, in order to meet the rapidly growing mobile service demands, many industry students have begun to search for a new mobile communication technology that can not only meet the user experience, but also improve the spectrum efficiency. In this context, a new multiple access scheme, non-orthogonal multiple access (Non Orthogonal Multiple Access, NOMA) technique, is proposed.
The NOMA technology receives much attention because of the fact that the NOMA technology utilizes a complex receiver design to obtain good spectrum efficiency, and becomes one of the key technologies of the 5G in china, and the core idea of NOMA is to use superposition codes (Superposition Coding, SC) at a transmitting end and serial interference cancellation (Successive Interference Cancellation, SIC) at a receiving end. Through the operation, NOMA can realize multiple access in a power domain through different power levels on the same time-frequency resource block. For users with poor channel conditions, larger power is allocated for transmitting information, and users with good channel conditions are the opposite. At the transmitting end, the base station communicates with all users through the same time-frequency resource, namely through the SC technology. At the receiving end, the SIC is used for receiving, interference cancellation is generally performed according to the power of the user, firstly, the useful signal with larger power in all signals is used as the useful signal, the other signals are used as the interference, demodulation is performed, then the demodulated useful signal is subtracted, and the like until the information of the user is demodulated. Although the SIC scheme can gradually eliminate multiple access interference in the received signal, the implementation of the performance of the SIC scheme depends heavily on the parameters of the interference signal and the estimated parameters of the channel, and if the parameter estimation is inaccurate or the channel environment is not reasonable, the SIC scheme will lose meaning. Especially for multi-user NOMA-MIMO systems, due to excessive interference between clusters and users, ideal receiver performance cannot necessarily be obtained by adopting a SIC scheme.
Disclosure of Invention
The invention aims to provide a system detection method based on deep learning, which aims to solve the problems in the background technology.
In order to achieve the above purpose, the present invention provides the following technical solutions: a system detection method based on deep learning, comprising:
determining a neural network, wherein the neural network comprises an input layer, a hidden layer and an output layer;
obtaining a relation transmitted from an output layer to a hidden layer and a relation transmitted from the hidden layer to the output layer, and determining the layer number of the hidden layer;
determining the weight change amount of the neural network and the threshold change amount of the neural network;
performing channel estimation on the received signal based on a neural network;
obtaining a set of filter coefficients and determining initial values of weights of the neural network based on the set of filter coefficients;
calculating a cost function of the network training of the neural network;
outputting the corrected weight of the neural network, the corrected threshold of the neural network, the number of layers of the hidden layer and the cost function of the network training of the neural network.
Preferably, the relation of the output layer to the hidden layer transfer
Wherein x is i Representing an ith neuron node of the input layer of the neural network; w (w) ij Representing a connection weight between an i-th neuron node of the input layer and a j-th node of the hidden layer; b j Is the threshold of the j-th node of the hidden layer.
Preferably, the output O of the jth neuron node of the hidden layer j ＝f(z j )。
Preferably, the t-th neuron of the output layer outputs
Preferably, the corrected weight of the neural network is w jt (N+1)＝w jt (N)-ηΔw jt Wherein
η represents the step size of the gradient descent;
the threshold value after correction of the neural network is b t (N+1)＝w jt (N)-ηΔb t Wherein
Preferably, the number of layers of the hidden layer
Wherein H is the number of hidden layers, I is the number of input layers, O is the number of output layers, and T is a random number from 1 to 10.
Preferably, the cost function of the network training of the neural network
Wherein s is m，k Modulated symbols transmitted for a kth user of an mth antenna; />
Estimated symbols transmitted for the kth user of the mth antenna; mse is characterized as the overall mean square minimum error for all antennas and users.
Compared with the prior art, the invention has the beneficial effects that: a system detection method based on deep learning accelerates the convergence rate of the deep learning. Under the off-line deep learning mode, the system structure is optimized, and the overall performance of the system is improved.
Drawings
FIG. 1 is a flow chart of a system detection method based on deep learning according to an embodiment of the present invention;
FIG. 2 is a block diagram of a communication system of a deep learning based system detection method according to an embodiment of the present invention;
fig. 3 is a schematic diagram of a SIC demodulation process of a deep learning-based system detection method according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1, the present invention provides a technical solution: a system detection method based on deep learning, comprising:
determining a neural network, wherein the neural network comprises an input layer, a hidden layer and an output layer;
obtaining a relation transmitted from an output layer to a hidden layer and a relation transmitted from the hidden layer to the output layer, and determining the layer number of the hidden layer;
the realization process of the neural network is divided into a learning stage and an identification process, wherein the learning process is to obtain a group of output results, namely forward transmission of information, according to an input training sample and an initial weight and a threshold value; then correcting the weight through the error and the learning rate between the calculated output result and the given true value, namely, the back propagation of the error; and repeatedly and continuously correcting the weight and the threshold of the network until the neural network model converges and the error reaches the preset requirement, thereby obtaining the optimal weight and threshold, namely the learning process. The identification process is to infer the category or bit/symbol to which any one or more test samples sent to the neural network belong by using the training obtained weight and the threshold value.
Determining the weight change amount of the neural network and the threshold change amount of the neural network;
performing channel estimation on the received signal based on a neural network;
obtaining a filter coefficient set, and determining an initial value of a weight of the neural network based on the filter coefficient set;
calculating a cost function of the network training of the neural network;
outputting the corrected weight of the neural network, the corrected threshold of the neural network, the number of layers of the hidden layer and the cost function of the network training of the neural network.
Specifically, the relationship of the output layer transfer to the hidden layer
Wherein x is i Representing an ith neuron node of the input layer of the neural network; w (w) ij Representing a connection weight between an i-th neuron node of the input layer and a j-th node of the hidden layer; b j Is the threshold of the j-th node of the hidden layer.
Specifically, the output O of the jth neuron node of the hidden layer j ＝f(z j )。
Specifically, the output layer t-th neuron outputs
Specifically, the corrected weight of the neural network is w jt (N+1)＝w jt (N)-ηΔw jt Wherein
η represents the step size of the gradient descent;
the threshold value after correction of the neural network is b t (N+1)＝w jt (N)-ηΔb t Wherein
Specifically, the number of layers of the hidden layer
Wherein H is the number of hidden layers, I is the number of input layers, O is the output layerT is a random number of 1 to 10.
Specifically, a cost function for network training of a neural network
Wherein s is m，k Modulated symbols transmitted for a kth user of an mth antenna; />
Estimated symbols transmitted for the kth user of the mth antenna; mse is characterized as the overall mean square minimum error for all antennas and users.
As shown in fig. 2, first, a neural network system model is confirmed, and it is assumed that M transmitting antennas are configured at each Base Station (BS) side, and each antenna simultaneously transmits at most K users, and the users included in each antenna form a cluster, that is, the number of clusters is assumed to be the same as the number of transmitting antennas. Each Mobile Station (MS) configures N antennas, and a BS transmitting end signal may be expressed as:
wherein P is a precoding matrix, and the dimension of the precoding matrix is M multiplied by M;
for transmitting a modulated signal, its dimension is mx1, which can be further written in the form:
wherein s is m，k A signal transmitted on behalf of a kth user in an mth cluster; alpha m，k NOMA Power configuration coefficients representing the kth user in the mth cluster, and respective Power normalization criteria, i.e
Without losing generality, the firstThe received signal of the kth user in one cluster can be expressed as: />
Wherein H is 1，k The dimension of the Rayleigh fading channel matrix from the BS to the kth user in the 1 st cluster is N multiplied by M; n is n 1，k Is an additive Gaussian noise vector, and has a dimension of N multiplied by 1. The joint equation is available in the form of a joint equation,
wherein the first term is a target signal of the first cluster; the second term is the interfering signal of other clusters; the third term is additive noise. In order to eliminate intra-cluster interference, first, a detection vector v matched with a target user needs to be designed at a receiving end i，k . In order to completely eliminate the interference in the cluster, the precoding matrix and the detection filtering matrix need to satisfy:
Can be simplified into:
wherein h is m，ik For matrix H i，k Is the m-th column of (2). The method can be obtained by expanding the materials:
optimal detection vector v i，k Can be composed of
Is obtained by the empty set of (a) and (b),
thus, the equation
Can be further simplified into:
y′ 1，k ＝v 1，k y 1，k ＝v 1，k h 1，1k (α 1，1 s 1，1 +...+α 1，K s 1，K )+n′ 1，k
wherein h is m，ik For matrix H i，k Is the m-th column of (2). The setting of the equivalent channel gain and the user power configuration coefficient has the following relation:
according to NOMA user power distribution principle, alpha can be obtained 1，1 ≤...≤α 1，k 。
The traditional NOMA demodulation algorithm is: after the inter-cluster interference is eliminated, serial Interference Cancellation (SIC) technology is then used to demodulate the intra-cluster multi-user information. Let the cluster 1 have 2 users, i.e. k=2, the receiving-side demodulation process is shown in fig. 3. The power of the user 2 is larger than that of the user 1, so that the information of the user 2 is demodulated first, then the SIC is executed to obtain the information of the user 2, and the information of the user 1 is demodulated again.
The framework and algorithm performance were verified, configured as follows: m=k=2, n=2 or 4,
the channel type is Rayleigh channel, the modulation mode is QPSK, channel coding is not adopted, and correlation between antennas does not exist, so that correlation simulation verification is performed. Offline data was generated by MATLAB software and DNN offline training and testing was run on PYTHON software. The DNN input number and the DNN output number are respectively set as follows: 26 or 58, the range of hidden numbers is obtained from the equation: 11-20. Thus, hidden layer numbers for all experiments in this sectionSet to 16. The Epoch and the target MSE are set as follows: 200 and 0.004. The learning rate R is set to be 0.01, the activation function of changing each 50 Epoch periods R into the original 1/2 hidden layer is set to be a Relu function, and the Relu activation function is simple in calculation, can well relieve the problem of gradient disappearance, can greatly accelerate the convergence rate of the neural network, and improves the performance of the neural network. The output layer activation function is set as a Softmax function that maps the output to probabilities of the communication signal corresponding to the respective symbols so that the output of the neural network can form a probability distribution vector. And the MSE cost function is used as a cost function for model training, so that the generalization capability of the model is improved in order to avoid over fitting of the model, L2 regularization is applied to the network model, and a gradient descent method is used for training, so that the optimal parameters of the model are obtained. The settings of the parameters of the neural network are shown in table 1.
TABLE 1
On the premise of the above parameter determination, it is assumed that the frame structure of the test system includes two data portions (58 symbols) and one pilot portion (26 symbols), where the pilot may be used as parameter estimates such as channel estimation, power allocation factor, and frequency offset; the channel multipath length is 7. The complexity (single antenna case) may be further specified as shown in table 2.
Table 2 complexity analysis
TABLE 2
In addition, in the simulation process, the MMSE-SIC scheme and the DL scheme have the same channel impulse response coefficient and transmission symbol. The following performance comparative analysis was performed under the same circumstances as above, and the results showed that: the DL-based multi-user NOMA-MIMO system detection algorithm has better system performance, and when n=2, there is a gain of about 1.8dB @ ber=1%; n=4, there is a gain @ ber=0.1% of about 1.3 dB. This gain is mainly due to the large number of sample learning of DL, thus avoiding the gain due to errors of the channel estimation module, but the cost of this is the delay caused by DL training. In order to alleviate the processing delay problem caused by introducing DL, a receiver similar to ZF/MF is adopted as a preprocessing algorithm of DL, and the rough result is used as an initial value of DL. The convergence speed can be increased and the system performance can be slightly increased.
Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (8)
1. A system detection method based on deep learning, comprising:
determining a neural network, wherein the neural network comprises an input layer, a hidden layer and an output layer;
obtaining a relation transmitted from an output layer to a hidden layer and a relation transmitted from the hidden layer to the output layer, and determining the layer number of the hidden layer;
determining the weight change amount of the neural network and the threshold change amount of the neural network;
performing channel estimation on the received signal based on a neural network;
obtaining a set of filter coefficients and determining initial values of weights of the neural network based on the set of filter coefficients;
calculating a cost function of the network training of the neural network;
outputting the corrected weight of the neural network, the corrected threshold of the neural network, the number of layers of the hidden layer and the cost function of the network training of the neural network.
2. The deep learning-based system detection method of claim 1, wherein: the relation of the transmission from the output layer to the hidden layer
Wherein x is i Representing an ith neuron node of the input layer of the neural network; w (w) ij Representing a connection weight between an i-th neuron node of the input layer and a j-th node of the hidden layer; b j Is the threshold of the j-th node of the hidden layer.
3. The deep learning-based system detection method of claim 2, wherein: output O of the jth neuron node of the hidden layer j ＝f(z j )。
4. A system detection method based on deep learning according to claim 3, wherein: the t-th neuron of the output layer outputs
5. The deep learning-based system detection method of claim 4, wherein: the corrected weight value of the neural network is w jt (N+1)＝w jt (N)-ηΔw jt Wherein
η represents the step size of the gradient descent, w jt (N) represents the weight values of the jth neuron to the t-th output neuron at time N;
the threshold value of the t output neuron at the (N+1) th time after the correction of the neural network is b t (N+1)＝w jt (N)-ηΔb t Wherein
6. The deep learning-based system detection method of claim 5, wherein: the number of layers of the hidden layer
Wherein H is the number of hidden layers, I is the number of input layers, O is the number of output layers, and T is a random number from 1 to 10.
7. The deep learning-based system detection method of claim 6, wherein: cost function of network training of the neural network
Wherein s is m，k Modulated symbols transmitted for a kth user of an mth antenna; />
Estimated symbols transmitted for the kth user of the mth antenna; mse is characterized as the overall mean square minimum error for all antennas and users.
8. The deep learning-based system detection method of claim 1, wherein: the set of filter coefficients
Wherein->
Channel estimation values for channel estimation of the received signals based on the neural network. />
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011024526.5A CN112215335B (en) | 2020-09-25 | 2020-09-25 | System detection method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011024526.5A CN112215335B (en) | 2020-09-25 | 2020-09-25 | System detection method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112215335A CN112215335A (en) | 2021-01-12 |
| CN112215335B true CN112215335B (en) | 2023-05-23 |
Family
ID=74051127
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011024526.5A Active CN112215335B (en) | 2020-09-25 | 2020-09-25 | System detection method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112215335B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112989508B (en) * | 2021-02-01 | 2022-05-20 | 复旦大学 | Filter optimization design method based on deep learning algorithm |
| CN113114313A (en) * | 2021-04-13 | 2021-07-13 | 南京邮电大学 | Method, system and storage medium for detecting pilot auxiliary signal of MIMO-NOMA system |
| CN114781581A (en) * | 2022-02-28 | 2022-07-22 | 陕西理工大学 | Atmospheric visibility prediction method based on DNB |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108650201A (en) * | 2018-05-10 | 2018-10-12 | 东南大学 | Channel equalization method, interpretation method based on neural network and corresponding equipment |
| CN108809839A (en) * | 2018-07-17 | 2018-11-13 | 湖南理工学院 | Wireless Mesh backbone network network flow control methods and device |
| CN110099017A (en) * | 2019-05-22 | 2019-08-06 | 东南大学 | The channel estimation methods of mixing quantization system based on deep neural network |
| CN110213814A (en) * | 2019-07-04 | 2019-09-06 | 电子科技大学 | A kind of distributed power distributing method based on deep neural network |
| CN110336593A (en) * | 2019-06-03 | 2019-10-15 | 金华航大北斗应用技术有限公司 | MIMO-NOMA signal detection system, method and terminal installation based on deep neural network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3915236A4 (en) * | 2019-01-25 | 2023-05-24 | Genghiscomm Holdings, LLC | Orthogonal multiple access and non-orthogonal multiple access |
-
2020
- 2020-09-25 CN CN202011024526.5A patent/CN112215335B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108650201A (en) * | 2018-05-10 | 2018-10-12 | 东南大学 | Channel equalization method, interpretation method based on neural network and corresponding equipment |
| CN108809839A (en) * | 2018-07-17 | 2018-11-13 | 湖南理工学院 | Wireless Mesh backbone network network flow control methods and device |
| CN110099017A (en) * | 2019-05-22 | 2019-08-06 | 东南大学 | The channel estimation methods of mixing quantization system based on deep neural network |
| CN110336593A (en) * | 2019-06-03 | 2019-10-15 | 金华航大北斗应用技术有限公司 | MIMO-NOMA signal detection system, method and terminal installation based on deep neural network |
| CN110213814A (en) * | 2019-07-04 | 2019-09-06 | 电子科技大学 | A kind of distributed power distributing method based on deep neural network |
Non-Patent Citations (1)
| Title |
|---|
| 去蜂窝大规模MIMO系统研究进展与发展趋势;章嘉懿;《重庆邮电大学学报(自然科学版)》;20190615(第03期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112215335A (en) | 2021-01-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112215335B (en) | System detection method based on deep learning | |
| CN110266617B (en) | Multipath channel estimation method of super-Nyquist system | |
| WO2016034051A1 (en) | Interference suppression method and device | |
| CN106612135A (en) | A signal transmission method, reception method and device based on multi-carrier spatial modulation | |
| CN101026435A (en) | Low-complexity maximum likelihood detecting method and device for communication system | |
| Ye et al. | Bilinear convolutional auto-encoder based pilot-free end-to-end communication systems | |
| CN114665930A (en) | Downlink blind channel estimation method of large-scale de-cellular MIMO system | |
| CN109391315B (en) | Data model dual-drive MIMO receiver | |
| Wang et al. | Compressive sensing based user activity detection and channel estimation in uplink NOMA systems | |
| CN100499611C (en) | Inspection of blank field maximum rear-proving probability in wireless communication system | |
| CN109818891B (en) | Lattice reduction assisted low-complexity greedy sphere decoding detection method | |
| CN109167748B (en) | Partial maximum likelihood detection method based on energy sorting | |
| CN101355377A (en) | Method for detecting signal of multi-input multi-output V-BALST system | |
| CN110190908B (en) | Constellation diagram design method for minimizing bit error rate of incoherent large-scale SIMO system in ISI channel | |
| Nordenvaad et al. | Iterative reception for acoustic underwater MIMO communications | |
| Khan et al. | Data aided channel estimation for MIMO-OFDM wireless systems using reliable carriers | |
| CN114337876A (en) | AMP detection algorithm based on training of NSGA2 algorithm and system for implementing same | |
| Chen et al. | Multiuser pre-coding aided quadrature spatial modulation for large-scale MIMO channels | |
| CN111769975A (en) | MIMO system signal detection method and system | |
| Sakoda et al. | Residue Effect of Parallel Interference Canceller in Belief Propagation Decoding in Massive MIMO Systems | |
| Qian et al. | OFDM Receiver Design With Learning-Driven Automatic Modulation Recognition | |
| Tiwari et al. | ASEP of MIMO system with MMSE-OSIC detection over weibull-gamma fading channel subject to AWGGN | |
| Xie et al. | Detection Algorithm Based on Deep Learning for the Multi-user MIMO-NOMA System | |
| RU2405254C2 (en) | Evaluation method of data transmission channel in wireless communication system (versions) | |
| CN113891420B (en) | Underwater sound cooperative communication relay selection method based on multi-attribute decision |
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 |