A multiuser detection (MUD) algorithm based on deep learning network is proposed for the satellite mobile communication system. Due to relative motion between the satellite and users, multiple access interference (MUI) introduced by multipath fading channel reduces system performance. The proposed MUD algorithm based on deep learning network firstly establishes the CINR optimal loss function according to the multiuser access mode and then obtains the best multiuser detection weight through the steepest gradient iteration. Multilayer nonlinear learning obtains interference cancellation sharing weights to achieve maximum signal-to-noise ratio through gradient iteration, which is superior than the traditional serial interference cancellation algorithm and parallel interference cancellation algorithm. Then, the weights with multiuser detection through multilayer network forward learning iteration are obtained with traditional multiuser detecting quality characteristics. The proposed multiuser access detection based on deep learning network algorithm improves the MUD accuracy and reduces the number of traditional multiusers. The performance of the satellite multifading uplink system shows that the proposed deep learning network can provide high precision and better iteration times.
Due to high-speed relative motion between mobile users and satellites in the satellite mobile communication system, different users access with the satellite at different elevation angles and multipath channel between satellite and user links is fading. These factors are creating obstacles for multiuser detection. In the case of limited bandwidth system, multiuser access detection (MUD) is an important issue in satellite mobile communication systems.
In the early literature, Cao and Viswanathan [
Since multiusers accessed the satellite system at different elevation angles, different access carrier frequency offsets (CFOs) introduced multiuser access interference (MUI), so it was difficult to implement single-tap FDE to achieve multiuser detection. Some related research studies had been designed for multiuser detection. Tang and Heath [
Blind user detection did not require a priori information, which effectively improves transmission efficiency. Therefore, Zhang and Gao [
In recent years, literatures researched on the compressed sensing reconstruction algorithm for multiuser detection. Abebe and Kang [
For nonlinear transformation of single-layer network, the related literature also discussed and optimized in detail. For blind channel estimation of MIMO communication systems, AsadUllah et al. [
Especially for the NOMA system, in literature [
For multilayer networks analysis, the related literature also discussed about CNN architecture. Yinghao et al. [
For the feature extraction of multilayer neural networks, Chao et al. [
If the shadow fading follows the Nakagami distribution, the Abdi star fading model was formed [
Some related literatures also have carried out studies on the satellite ground link channel model fitting through the measured data. Loo et al. [
In this paper, the satellite-to-ground channel model is established using the measured results of the German Aerospace Research Center [
Channel model for satellite-to-ground link based on tapped delay line.
The specific method is to simulate the signal amplitude fading through the tapped delay line filter. Firstly, it is assumed that the scattering body is divided into several clusters, and the bandwidth of the signal transmission bandwidth is not resolved within each cluster. Then, the multicluster is used to model the satellite-to-ground link.
In the tapped delay line model, each tap represents a set of a plurality of delay paths with the same sum, but also the time delay path changes due to different flat fading amplitudes.
The tapped delay line model for satellite-to-ground link, multipath channel impulse response is composed of different delay characteristics; the channel modeling method is established for satellite-to-ground channel model following the tapped delay line.
Define
The Doppler shift caused by satellite motion is regular. At the same time, for defined mobile user, the Doppler shift is determined by the velocity of the high-speed motion and the elevation angle of the user. The Doppler shift introduced by high-speed satellite motion can be approximately equal.
Defining
After introducing frequency offset interference, the received signal can be expressed as
The receiving end performs
When the carrier frequency offset is zero, the interference coefficient is 1. This shows that the interference between carriers depends on the relative frequency deviation and the serial number distance between subcarriers. As the relative frequency offset interference factor increases,
Equation (
If the interference signal with frequency offset is DFT transformed, it can be written as follows:
Equation (
Defining
Through the elemental analysis of the matrix, in the case of smaller frequency offset interference, the energy is mainly concentrated on the diagonal. The larger the frequency offset value, the more dispersed the energy, the larger the interference term, and the more the interference of the introduced ICI. Its energy distribution diagram is shown in Figures
Matrix energy distribution introduced by small frequency offset.
Matrix energy distribution introduced by large frequency offset.
We could obtain from equation (
The number of set carriers is defined as 512, the channel bandwidth is 20 MHz, the Doppler shift is 15 kHz, and the signal mapping mode is QPSK. Figure
Linear phase introduced by frequency offset.
For the multiuser uplink access, the interference comes from frequency offsets. At the same time, the larger the frequency offset range for each user, the more serious the multiuser interference. For the satellite transmission system uplink system, the access interference cancellation of each user is the key for uplink user detection.
The process of multiuser detection is divided into three parts. First, the multiuser signal is completed to cancel the access interference and the multiuser access interference is reduced by establishing the optimal weight of the multilayer network. Secondly, through the multilayer network, the weight is iterated to obtain the optimal point. Through the network weight sharing and iteration of the first two parts, the optimal identification weight is finally obtained. Multiuser detection and identification is accomplished by optimally identifying the weight network.
The proposed algorithm is based on the goal for optimizing CINR, which is to find the optimal CINR corresponding to WIC interference cancellation algorithm weights. Thus, multiuser interference signal received by satellite can be expressed as
According to multicarrier allocation, this process can be a multiuser signal separation. Firstly, each user can be according to the traditional WIC algorithm for cancellation. For user
Idealizing it, we could obtain
Secondly, according to the traditional WPIC algorithm, each user can be for parallel cancellation interference, and it can be expressed as
Due to different user access elevation angles, as well as the satellite ground link fading channel, introducing different serious Doppler frequency shifts is for the satellite-to-ground uplink system. For a particular user, because the tangential velocity of the satellite in a symbol is the same, the relative carrier frequency is kept constant, so the frequency offset introduced in a symbol period can be regarded as constant. Therefore, the normalized frequency offset factor can be considered to be constant.
The definition of satellite ground link uplink carrier number is 1024; the satellite suburban environment model is given in Table
Suburban environment parameter.
Tap | Distribution function | Parameter | Parameter distribution | Numerical value (dB) | Time delay (ns) |
---|---|---|---|---|---|
1 | LOS : Rician | Rice factor |
|
9.7 | 0 |
2 | Rayleigh | Average multipath power |
|
−23.6 | 100 |
Multiuser detection system block diagram.
It can be obtained from formula (
The number of users is 4, and the number of subcarriers is 2048. In the condition of AWGN, SNR = 5 dB, the allocation is OFDM, and the frequency offset is 0.01, 0.05, 0.15, and 0.2. The multifading channel is shown in Table
CINR performance at different weights in the IC algorithm (including WSIC and WPIC) with three-layer network in two iterations.
The proposed multiuser detection algorithm is to optimize SINR through obtaining optimal cancellation weight; therefore, the IC algorithm can be divided into WSIC and WPIC algorithms; the WSIC algorithm is cancel interference for multiuser access according to each subcarrier, and the WPIC is the multiuser interference cancellation algorithm at the same time.
Figure
The improved algorithm is based on the WIC algorithm. Optimal weights are iterated to approximate initial weights. It is specific for obtaining the optimal weights below.
Defining
The influence of the satellite to ground link on this algorithm consists of two parts. Firstly, the influence for the proposed algorithm is also induced by the multiuser access angle differences. Due to differences in relative motion between the user and satellite, multiuser access interference has been generated, which significantly degrades the satellite system performance.
When multiusers access the same satellites, the multiuser access angle differences will introduce different carrier frequency offsets in the total number of the carrier system. Under certain conditions, the carrier frequency deviation will induce the different multiple access interference, including interference simulation as shown in Figure
Secondly, the influence for the proposed algorithm is induced by multipath fading. For urban simulation scenarios, the signal reflection effect caused by buildings is larger, and the diffraction effect caused by multipath is also larger. The more delay the received signal propagation, the more serious the signal fading is.
In the countryside scene, compared with the urban scene, the multipath number is decreased and the fading is relatively flat. This is because that the contryside scene is with a smaller number of building and weaker reflection and refraction.
For multiuser received signals to cancel interference, we use shared weights to obtain the best weights and then obtain user detection and weight update.
The cost function established can be expressed as
The cost function is to find that the accurate reconstruction, which should be realized. Then, the optimal weight detection error is made. Therefore, the sparse recovery for multiweight sharing can be obtained, and the optimal user detection for all the users can be satisfied in the following:
The first term is a nonzero regular term whose position is known and is different from the traditional mode of all cost functions.
In addition to the weight constraint cost, the regularization constraint can be established for the corresponding
Therefore, when the measurement data are very small,
The augmented Lagrangian is expressed as
The scaled problem (
Shared weight updates:
Establishing that the orbital altitude is 1100 km, the rural environment measured data proposed in [
Channel model parameter in rural environment.
Tap | Distribution function | Parameter | Parameter distribution | Numerical value (dB) | Time delay (ns) |
---|---|---|---|---|---|
1 | LOS : Rician | Rice factor |
|
6.3 | 0 |
nLOS : Rayleigh | Average multipath power |
|
−9.5 | ||
2 | Rayleigh | Average multipath power |
|
−24.1 | 100 |
3 | Rayleigh | Average multipath power |
|
−25.2 | 250 |
Set the satellite beam spot beam of number 5 with a coverage diameter of 450 km. Satellite-to-ground link model with
Figure
BER curves under different SNRs.
Figure
CINR performance of different subcarriers.
Figure
BER performance of different normalized frequency offsets.
Figure
Probability of correct classification of the proposed deep learning network for multiuser access to satellite at different length curves.
A multiuser detection algorithm based on deep learning network has been proposed. The proposed deep learning network for MUD could provide high precision and lower iteration times, which firstly establishes the CINR optimal loss function according to the multiuser access interference mode and then obtains the best multiuser detection weight through the steepest gradient iteration. The important feature of the proposed algorithm is through nonlinear optimal direction learning and to achieve maximum signal-to-noise ratio through gradient iteration, and then share weights. Through establishing a typical satellite communication system simulation platform, compared with the OMP and IORLS algorithms, the proposed deep learning network algorithm has better performance in different conditions of SNR, CINR, and carrier frequency offset interference.
The data used to support the findings of this study are available from the corresponding author upon request.
The authors declare that they have no conflicts of interest.
This work was supported by the National Natural Science Foundation of China (no. 61501306), the Natural Science Foundation of Liaoning Province of China (no. 2015020026), and the Natural Science Foundation of Liaoning Provincial Department of Education (no. L2015402).