阅读说明：本技术 一种基于粒子滤波器的全双工系统自干扰消除方法 (Particle filter-based self-interference elimination method for full-duplex system ) 是由 管鹏鑫 汪奕汝 毕乘乾 赵玉萍 于 2021-07-13 设计创作，主要内容包括：本发明公开了一种基于粒子滤波器的全双工系统自干扰消除方法。本方法为：1)将全双工系统工作在半双工阶段,进行自干扰信道估计,得到自干扰信道估计值2)将全双工系统工作在全双工阶段并采用粒子滤波器跟踪估计n时刻的相位噪声估计值3)基于自干扰信道估计值以及相位噪声估计值消除自干扰信号。本发明采用粒子滤波器跟踪估计相位噪声的变化,从而能够更精准的重建出每个时刻的自干扰信号,与传统方案相比具有较高的自干扰消除性能增益,为全双工技术的进一步应用奠定基础。(The invention discloses a particle filter-based self-interference elimination method for a full-duplex system. The method comprises the following steps: 1) working the full-duplex system in a half-duplex stage, performing self-interference channel estimation to obtain a self-interference channel estimation value 2) Operating the full-duplex system in a full-duplex stage and tracking and estimating a phase noise estimation value at n time by adopting a particle filter 3) Channel estimation value based on self-interference And phase noise estimate And eliminating the self-interference signal. The invention adopts the particle filter to track and estimate the change of the phase noise, thereby being capable of reconstructing the self-interference signal at each moment more accurately.)

1. A full duplex system self-interference elimination method based on particle filter includes steps:

1) working the full-duplex system in a half-duplex stage, performing self-interference channel estimation to obtain a self-interference channel estimation value

2) Operating the full-duplex system in a full-duplex stage and tracking and estimating a phase noise estimation value at n time by adopting a particle filter

3) Channel estimation value based on self-interferenceAnd phase noise estimateAnd eliminating the self-interference signal.

2. The method of claim 1, wherein self-interference channel estimateWhere x '(N) is the pilot symbols transmitted during the half-duplex phase, w' (N) is the gaussian noise during the half-duplex phase, and N is the number of samples used to estimate the channel.

3. The method of claim 1, wherein a phase noise estimate is obtainedThe method comprises the following steps:

21) using a particle filter to obtain a system state transition equation theta (n) ═ theta (n-1) + v_n-1Sampling the phase noise distribution of the full duplex stage to obtain a sampling value of each sampling particle at n moments; whereinIs the sample value of the ith sample particle, theta (n) is the phase noise at time n of the full-duplex phase, r_n-1Noise present during state transitions;

22) calculating the n-moment observation result of each sampling particle; wherein n-time observation of the ith sampled particleEstimation of useful signalsReceived signal r (n) ═ hx (n) e of full duplex stage node^jθ(n)+ u (n) + w (n) for reconstructing self-interference signal The phase noise estimation value at the time n-1, x (n) is a transmitting signal at the time n in the full-duplex stage, u (n) is a useful signal in the full-duplex stage, w (n) is Gaussian noise in the full-duplex stage, theta (n) is the phase noise in the full-duplex stage, and h is an equivalent self-interference channel in the full-duplex stage;

23) calculating the weight of each sampling particle at n moments according to the particle filter principle, and carrying out weight normalization;

24) calculating the n timeEstimation of phase noise And the weight normalization value at the moment of N of the ith sampling particle is obtained, and N is the total number of the sampling particles.

4. The method as claimed in claim 3, wherein the result is obtained according to step 24)Updating And eliminating the self-interference signal at the time of n.

5. A method according to claim 3, wherein the variable v is_n-1Obedience mean value 0, variance σ ═ σ -_t+σ_r；σ_tPhase noise theta for signal transmitting node_t(n) variance of phase noise distribution, σ_rPhase noise theta for signal receiving node_r(n) variance of phase noise distribution, θ (n) ═ θ_t(n)+θ_r(n)。

Technical Field

The invention relates to the field of communication, in particular to a particle filter-based self-interference elimination method for a full-duplex communication system.

Background

Currently, in order to alleviate the contradiction between the shortage of spectrum resources and the increasing demand for bandwidth, researchers have studied Full Duplex (FD). The full-duplex technology can theoretically realize double spectrum efficiency, solve the problem of hidden terminals, improve the relay communication efficiency and enhance the communication safety. However, since the transmission and reception operate at the same time and the same frequency, the local receiver receives a locally transmitted signal replica, i.e., a Self-Interference Signal (SI). The strong self-interference signal can drown out the far-end useful signal, thereby making the useful signal unable to be demodulated.

Currently, there are a number of self-interference cancellation mechanisms, namely, propagation domain cancellation, analog domain cancellation and digital domain cancellation. However, as the self-interference elimination algorithm is studied, the researchers found that the phase noise can reduce the elimination capability of the system, and become one of the limiting factors for the application of the full-duplex technology. The traditional method adopts a least square algorithm to complete self-interference channel estimation, so that a self-interference signal is reconstructed and eliminated based on an estimation result and a known transmitting signal, but the influence of phase noise is not eliminated. Since the phase noise is time-varying, if the variation of the phase noise can be estimated and tracked, the self-interference signal can be eliminated more accurately, and the system performance is improved.

A particle filter is a recursive filter using the monte carlo method, using a set of random samples with weights to represent the posterior probability of a random event. Sanjeev Arulampaam et al completely proposes an algorithm of an SIR (sampling impedance amplifying) particle filter, is the most original and widely used one of a plurality of particle filter algorithms, and mainly solves a plurality of problems in the field of image processing.

Disclosure of Invention

In order to overcome the problem that the self-interference performance of a system is reduced due to the insufficient consideration of phase noise of the existing algorithm, the invention provides a particle filter-based self-interference elimination method of a full-duplex system. Aiming at the scene that a full-duplex system has phase noise, a set of complete algorithm is designed based on a particle filter, and a specific state transition equation and an observation equation are provided; modifications are made to reduce estimation errors when using particle filters, as in subsequent equations 19-20; simulation results show that compared with the traditional scheme, the scheme realizes better elimination performance.

The technical scheme adopted by the invention for solving the technical problems is as follows: firstly, the system works in a half-duplex stage and carries out self-interference channel estimation. And then, a particle filter is adopted to track and estimate the change of the phase noise, and finally, the self-interference signal is eliminated based on the estimated self-interference channel and the phase noise.

Two full-duplex nodes are considered, and because the hardware structures of the two nodes have symmetry, only one node is selected for model establishment and algorithm research. The nodes adopt separate antennas, namely, different antennas are adopted by a transmitting chain and a receiving chain.

Setting x (n) as a transmitting signal, converting the discrete sequence into a continuous signal x (t) through a Digital-to-Analog Converter (DAC) module, and further performing up-conversion on the continuous signal by an oscillator to obtain a radio frequency signalNamely, it is

Wherein, theta_tAnd (t) is the phase noise of the transmitting end. Subsequently, the radio frequency signalBy self-interfering channel h_SIAnd (t) reaching a local receiving end, and considering that the node works in a weak reflection scene, namely the channel is a single-path channel. Meanwhile, since the two nodes operate at the same time and the same frequency, the received signal of the local receiving antenna also includes the useful signal u (t) transmitted from the far end. t denotes the time t of the analog signal.

The analog cancellation mainly works in the radio frequency domain, and a SI signal copy is reconstructed according to the estimated self-interference channel and the known transmitting signal and is subtracted from the receiving signal, so that an analog-cancelled signal can be obtainedIs composed of

Wherein h is a joint equivalent channel of the self-interference channel and the analog domain cancellation channel. w (t) is the system Gaussian noise.

And obtaining a baseband signal r (t) after down-conversion of the oscillator at the receiving end, and further obtaining a baseband digital signal r (n) through an ADC module.

Wherein w (n) is white Gaussian noise, θ_t(n) and theta_r(n) is the discrete sampling point of the phase noise of the transmitting end and the receiving end, and u (n) is the useful signal needing to be demodulated. The phase noise is modeled as a wiener process theta by adopting a common free oscillator model_t(n₁)-θ_t(n₂)～N(0，4πf_3dB|n₁-n₂|T_s) That is, the mean value of the two-point phase difference value is 0, and the variance is 4 pi f_3dB|n₁-n₂|T_sOf Gaussian process f_3dB3dB bandwidth, T, for phase noise_sIs the sampling time interval. Receiver phase noise theta_r(n) model and transmitting end phase noise theta_t(n) same, for simplifying the expression, we define the variance of the phase noise distribution at the transmitting end and the receiving end as σ respectively_tAnd σ_r。

Thus, we can give the corresponding state transition equation of the particle filter as

θ(n)＝θ(n-1)+v_n-1#(6)

For convenience of description, let θ (n) be θ_t(n)+θ_r(n) represents the joint phase noise at the transmitting and receiving ends. Since the phase noise at the transmitting end and the receiving end are independent of each other and obey the wiener process, the state transition noise variable v in equation (6)_n-1Obedience mean 0 and variance σ θ_t(n) and θ_rSum of (n), i.e. σ ═ σ_t+σ_r。

The system has an observation equation of

r(n)＝hx(n)e^jθ(n)+u(n)+w(n)#(7)

The particle filter completes the prediction at this time point by the estimation result at the previous time point. I.e. p (theta)_n-1|r_1：n-1) To p (theta)_n|r_1：n-1) The prediction process of (1). Then, based on the prediction result and in combination with the observation data r (n) at the current time, the updating process is completed. I.e. p (theta)_n|r_1：n-1) To p (theta)_n|r_1：n) The update process of (1).

In order to solve the problem that sampling cannot be performed from posterior probability distribution, an importance sampling algorithm is introduced into the particle filter. Assuming sampling from q (θ | r) of a known distribution, the expected value of the current state can be expressed as

Wherein W_n(θ_n) The weight representing the sampled particle can be expressed as

The degeneracy of equation (8) can be transformed by the total probability integral equation, and equation (9) can be substituted to obtain:

further, the above-mentioned desired problem can be approximated by the Monte Carlo method, i.e.

Wherein N represents the number of particles sampled,representing the weight of the ith sample particle at time n.

Further, normalizing the weight of the particle in equation (9) and substituting into equation (11) can obtain:

wherein the particles are normalized in weightIs shown as

Equation (13) shows that the particles different at n moments all have their corresponding weights, and if the weight of the particle is great, it indicates that the particle is trusted to be more. This subsection derives the recursive form of the weights at two adjacent times.

Suppose the importance distribution is only equal to the previous time state θ_n-1And the measured value r (n). Then, in order to make the calculation of the weight value more concise and without loss of generality, the importance probability function is made equal to the state transition probability q (θ)_n|θ_1：n-1，r_1：n)＝p(θ_n|θ_n-1) Substituting into equation (14) to obtain:

as can be seen from equation (7), the unknowns contained in the observed signal are the equivalent self-interference channel h, the phase noise θ (n), and the useful signal u (n). The particle filter aims at estimating phase noise theta (n), firstly carries out particle sampling according to a state equation to obtain an estimation value, then carries in an observation equation to calculate the weight corresponding to the particles, and finally calculates the final estimation result according to the weight. Therefore, before the system enters the full-duplex stage, the self-interference channel must be estimated, and in addition, in order to reduce the influence of the useful signal on the particle weight, the useful signal can be demodulated and subtracted according to the estimation result at the last moment, so that the estimation accuracy is improved. Therefore, we propose a new self-interference cancellation and symbol detection scheme for full-duplex systems.

First, in order to estimate the self-interference channel, the system operates in a half-duplex phase, i.e., no useful signal is transmitted, and the local node transmits a pilot signal to estimate the self-interference channel. The received signal may be represented as

r′(n)＝hx′(n)e^jθ(n)+w′(n)#(16)

Where x '(n) is the transmitted signal and w' (n) is gaussian noise, the self-interference channel can be estimated by

Where N is the number of samples used to estimate the channel.

After the self-interference channel estimation is completed, the system enters a full-duplex stage, namely, a useful signal exists in the system. The received signal may be represented as

r(n)＝hx(n)e^jθ(n)+u(n)+w(n)#(18)

At this stage, we complete self-interference cancellation and demodulation of the useful symbols simultaneously. The algorithm flow is described by taking the nth time as an example. First, we use the estimated value of the phase noise at time n-1Self-interfering channelAnd the known transmit signal x (n) for self-interference signal reconstruction, i.e.

The estimation value of the useful signal obtained by eliminating the self-interference signal is

Phase noise estimation at time n is then done based on the particle filter:

given that the number of sampling particles in the system is N, θ (N) + v can be obtained according to the system state transition equation θ (N)_n-1Completing the sampling of the probability q (theta | r) to obtain the sampling value of the ith sampling particle at the moment n

Computing n-time observation results based on ith sampling particle

Thirdly, the weight of each particle at the n moment is calculated according to the principle of the particle filter, namely formula (15)And performing weight normalization according to the formula (13) to obtain

Fourthly, calculating the estimated value of the phase noise at the n moment according to the formula (12)

Updating useful signals again according to the estimated value of the phase noise at the n moment,

thus, the phase noise estimation and the useful signal demodulation at the time of n are completed.

The method has the advantages that the change of the phase noise is tracked and estimated by adopting the particle filter, so that the self-interference signal at each moment can be reconstructed more accurately, the self-interference elimination performance gain is higher than that of the traditional scheme, and a foundation is laid for the further application of the full-duplex technology.

Drawings

Fig. 1 is a diagram of the full-duplex self-interference cancellation architecture of the present invention.

FIG. 2 is a graph of the results of an estimation of the 3dB bandwidth of the phase noise at 10 Hz;

(a) phase noise 3dB true phase noise at 10Hz bandwidth,

(b) phase noise 3dB estimated phase noise at 10Hz bandwidth.

FIG. 3 is a graph of the estimation results for a phase noise 3dB bandwidth of 100 Hz;

(a) phase noise 3dB true phase noise at 100Hz bandwidth,

(b) phase noise 3dB estimated phase noise at 100Hz bandwidth.

Fig. 4 is a self-interference cancellation capability versus INR.

Fig. 5 is a plot of mean square error of phase noise estimate versus INR.

Fig. 6 is a self-interference cancellation capability versus 3dB bandwidth.

Figure 7 is a plot of the mean square error of the phase noise estimate versus the 3dB bandwidth.

Detailed Description

The embodiment provides a particle filter-based self-interference elimination scheme of a full-duplex system.

Fig. 1 shows a white interference cancellation architecture diagram of a system and corresponding signals are obtained according to the present invention. Specifically, the example includes two steps.

The self-interference channel estimation is done first. The system works in a half-duplex stage, namely no useful signal is transmitted, and the local node transmits a pilot signal to estimate a self-interference channel. The received signal may be represented as

r(n)＝hx(n)e^jθ(n)+w(n)

Where x (n) is the transmitted signal and w (n) is Gaussian noise, the self-interference channel can be estimated by

Where L is the number of transmitted signals used to estimate the channel.

The system then enters the full duplex phase, i.e. the presence of a useful signal in the system. The received signal may be represented as

r(n)＝hx(n)e^jθ(n)+u(n)+w(n)#(18)

At this stage, we complete self-interference cancellation and demodulation of the useful symbols simultaneously. The algorithm flow is described by taking the nth time as an example. First, we use the estimated value of the phase noise at time n-1Self-interfering channelAnd the known transmit signal x (n) for self-interference signal reconstruction, i.e.

The estimation value of the useful signal obtained by eliminating the self-interference signal is

Phase noise estimation at time n is then done based on the particle filter:

firstly, assuming that the number of sampling particles of the system is N, the sampling value of the ith sampling particle at N moments can be obtained according to the system state transition equation

Computing n-time observation results based on ith sampling particle

Thirdly, the weight of each particle at the n moment is calculated according to the principle of the particle filter, namely formula (15)And performing weight normalization according to the formula (13) to obtain

Fourthly, calculating the estimated value of the phase noise at the n moment according to the formula (12)

Updating useful signals again according to the estimated value of the phase noise at the n moment,

thus, the phase noise estimation and the useful signal demodulation at the time of n are completed.

The simulation conditions given in this example are as follows.

INR is defined as the self-interference signal power to noise power ratio. Sampling interval of T_s＝3.3×10^-8s, the data information is modulated by QPSK, and the Signal-to-Noise Ratio (SNR) of the useful Signal is defined as the power Ratio of the Signal to the Noise. The particle number of the particle filter is set to 1000 herein. The conventional approach is to estimate the self-interference channel, reconstruct the SI signal and cancel it from the received signal.

The evaluation index herein is a system self-interference cancellation capability G.

Wherein E is_IRepresenting the energy of the self-interference signal before cancellation, E_rRepresenting the residual interference signal energy after self-interference cancellation, and N is the noise energy. And G represents the self-interference energy ratio before and after the system is eliminated, and the interference elimination capability of the system is embodied.

In addition, we also measure the mean square error of the phase noise estimated by the algorithm and the true phase noise.

Wherein T is the total number of symbols counted. A smaller MSE value indicates that the phase noise estimated by the algorithm has a smaller error from the true phase noise.

Fig. 2 and 3 show the results of the estimation of the phase noise 3dB bandwidth at 10Hz and 100Hz, respectively. It can be seen from the figure that the amplitude and the fluctuation of the phase noise with the 3dB bandwidth of 100Hz are significantly larger than the phase noise with the 3dB bandwidth of 10Hz, which also indicates that the phase noise with the larger 3dB bandwidth has a larger influence on the system. Simulation results show that the phase noise estimated by the algorithm can accurately track the change of the phase noise, thereby laying a good foundation for subsequent self-interference elimination.

Set up f_3dB＝15Hz，SNR＝20dB, the simulation results of the self-interference cancellation capability of the present scheme and the conventional scheme under different INRs are shown in fig. 4. The algorithm proposed herein has better cancellation performance than the conventional scheme and increases with increasing INR. This is because the particle filter proposed herein is able to track the phase noise variations well, thereby eliminating the effects of phase noise.

The MSE versus INR curve is shown in FIG. 5. The MSE represents the mean square error of the estimated phase noise and the real phase noise, and the smaller the MSE is, the higher the estimation precision is. The MSE decreases with increasing INR because INR characterizes the energy ratio of the self-interference signal to the noise, and when INR is higher, assuming that the self-interference energy is fixed, the energy of the noise will be smaller, and the noise will interfere less with the system, thereby reducing the estimation error.

The results of the self-interference cancellation performance simulation for different 3dB bandwidth phase noise are shown in fig. 6, with INR set to 50dB and SNR set to 20 dB. With the increase of the 3dB bandwidth, the elimination capability of the two schemes is reduced, but the elimination performance of the scheme is superior to that of the traditional scheme.

The simulation results of MSE performance under different 3dB bandwidth phase noise are shown in fig. 7. The MSE increases with increasing 3dB bandwidth because the larger the 3dB bandwidth, the larger the variance of the phase noise difference values at adjacent time instants, and hence the faster the variation, and the corresponding phase noise amplitude will also increase, resulting in an increase in the error of the estimation.

The application provides a self-interference elimination and symbol detection scheme of a full-duplex system based on a particle filter. The system model is described first, and the state transition equations and observation equations for the particle filter are given. Subsequently, the particle filter based phase noise estimation and self-interference cancellation algorithm is described in detail. Finally, simulation results show that the scheme provided by the invention can better estimate the change of the phase noise, thereby realizing better system elimination performance.

Although specific embodiments of the invention have been disclosed for purposes of illustration, and for purposes of aiding in the understanding of the contents of the invention and its implementation, those skilled in the art will appreciate that: various substitutions, changes and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, it is intended that the invention not be limited to the particular embodiment disclosed as the best mode contemplated for carrying out this invention, but that the invention will include all embodiments falling within the scope of the appended claims.