CN105915476A - Bayes-based phase noise compensation method - Google Patents

Abstract

The invention, which belongs to the technical field of wireless communication, especially relates to the field of phase noise estimation realization by using an iterative method in a wireless communication system. According to the invention, on the basis of a channel estimation sequence, an equivalent discrete time-domain channel impulse response is estimated; a common phase error (CPE) of the phase noises is estimated by interpolation; and then phase noise compensation is realized by using an iterative method. Therefore, the system reliability is improved and the error rate is reduced.

Description

A kind of make an uproar compensation method based on Bayesian phase

Technical field

The invention belongs to wireless communication technology field, particularly relate to that wireless communication system realizes phase noise by the method for iteration and estimate.

Background technology

Development of Wireless Communications by now, comes into new epoch.Along with the explosive growth of mobile multimedia application, indicating that communication creates tremendous influence to people's life, its importance is self-evident.The communication technology such as 802.11n standard that currently everybody commonly uses and ultra broadband (Ultra Wideband, UWB) although being capable of the data transmission of up to 300M/s, but can not meet people's demand to higher rate real-time Transmission, and relatively low wireless communication frequency band is the most crowded to capacity, to this end, a new generation's high-speed high frequency section Radio Transmission Technology is studied by everybody in succession.

For a long time, 60GHz wireless communication technology has huge exempting from by it and permits continuous bandwidth, make it can realize the high data transfer rates of Gb, simultaneously, it is not strict with along with its through-put power, cause 60GHz correlation technique to be fallen over each other research, the just like star of the communications field by everybody, more likely become one of topmost technology in future wireless system technology.The most numerousCountryContinuously opening the license of exempting from of continuous 5GHz-7GHz near 60GHz uses frequency domain resource for everybody research and development.Such as, the U.S. takes the lead in having divided 57-64GHz frequency range, and the most and then Canada, Japan, Europe and Australia have divided the 60GHz of oneself this country and exempted from license use frequency range..Along with the countries in the world attention increasingly to 60GHz technology, open 60GHz frequency range, start Ge great scientific & technical corporation of the world and the corresponding research institution tide to 60GHz technical research.

In the signals transmission of 60GHz communication system, in addition to the decline that experienced by channel, also to be affected by radio-frequency devices non-linear factor, the two factor makes the performance at receiving terminal system reduce.In 60GHz Millimeter-wave Wireless Communication System, the non-ideal part of radio-frequency front-end mainly includes phase noise, and IQ amplitude-phase is uneven, non-linearity of power amplifier distortion etc., phase noise, actually a kind of sign to frequency source frequency stability.Under normal circumstances, frequency stability is divided into long-term frequency stability and short-term frequency stability.So-called short-term frequency stability, refers to phase fluctuation or the frequency fluctuation caused by random noise.As for the frequency slow drift caused because of temperature, aging etc., the most referred to as long-term frequency stability.Generally primary concern is that short-term stability problem, it is believed that phase noise is exactly short-term frequency stability, only two kinds of different representations of a physical phenomenon.For oscillator, frequency stability is that it produces the one of same frequency in the time range of whole regulation and measures.If signal frequency exists instantaneous change, it is impossible to keep constant, then signal source exists for unstability, cause is exactly phase noise.

In a communications system, transmitting terminal is required for receiving terminal producing corresponding carrier wave with the frequency spectrum conversion completing between corresponding radio frequency and base band.But the crystal oscillator producing carrier wave exists certain otherness with phaselocked loop, cause carrier frequency and there is random difference in short-term with target frequency, in turn result in produced sine wave signal generation random phase saltus step, show as phase noise.For OFDM (Orthogonal Frequency Division Multiplex, OFDM) communication system phase noise can produce common phase error (Common Phase Error, and inter-carrier interference (Inter Carrier Interference, ICI) CPE)；And single carrier frequency domain equalization (Single Carrier with Frequency Domain Equalization, SC-FED) system phase noise can produce common phase error CPE and intersymbol interference (Inter Symbol Interference, ISI).

Summary of the invention

For the deficiencies in the prior art, the present invention provides a kind of and makes an uproar compensation suppressing method based on Bayesian phase, and the method can provide the reliability that signal transmit, the reduction bit error rate.

In order to describe present disclosure easily, first the definition that belongs to used in the present invention is illustrated:

Special word (Unique Word, UW): in order to carry out synchronizing or parameter Estimation etc. at receiving terminal, transmitting terminal send have some particular characteristics, to special sequence known to receiving terminal.

A kind of make an uproar compensation method based on Bayesian phase, specifically comprise the following steps that

S1, receiving terminal utilize channel estimation sequence carry out channel estimate obtain the transmitting terminal impulse response estimate to the equivalent time domain channel of receiving terminal

S2, obtain the signal of time domain at receiving terminal and affected by phase noise and white Gaussian noise, phase noise is compensated by receiving terminal by iteration, described phase noise phase noise power spectrum density (Phase noise power spectrum density, PSD) characterize, i.e.Wherein, f represents that the frequency at offset carrier center, PSD (0) are constants, f_p=1MHz is pole frequency, f_Z=100MHz is zero frequency, described PSD (f) is " pole/zero " model about phase noise PSD that communication standard IEEE802.15.3c and IEEE802.11ad is given, PSD (0)=-87dbc/Hz in IEEE802.15.3c, PSD (0)=-90dbc/Hz in IEEE802.11ad standard.

Further, phase noise is compensated by iteration and specifically comprises the following steps that by receiving terminal described in S2

S21, UW is added in data sequence to be transmitted, one data block contains two UW wherein, the length of described UW is more than the length of equivalent time domain channel described in S1, and UW is made up of Gray's known array of 64, and the data to be transferred in data block is 448 bit sign sequences；

S22, the method for employing interpolation estimate phase noise constant and the ratio of a of the i-th data block of transmission between i-th UW and i+1 UWWherein, described i-th UW is expressed as by reception signal during channel:Wherein a is phase noise constant, y_uw_(i)It is that i-th UW is by reception signal during channel, a_uw_(i)It is i-th UW by the constant of making an uproar mutually during channel,It is equivalent time domain channel impulse response, w_(i)Be i-th UW by channel time the noise that is subject to,Being the length of channel, i is the natural number being not zero；

S23, removal Cyclic Prefix (Cyclic Prefix, CP) are followed by receiving signal frequency domain matrix form and are expressed as: Y_{N × 1}=A_{N × N}H_{N × N}X_{N × 1}+W_{N × 1}, wherein, Y_{N × 1}For receiving the matrix that the frequency domain of signal is constituted, A_{N × N}The Toeplitz matrix constituted for phase noise frequency domain, H_{N × N}For estimating the diagonal matrix that the frequency domain of channel is constituted, X_{N × 1}The matrix constituted for transmission data frequency domain, W_{N × 1}Matrix for white Gaussian noise frequency domain structure；

S24, receive signal time-domain representation be: y_{N × 1}=diag (p_{N × 1})h_{N × N}x_{N × 1}+w_{N × 1}, wherein, y_{N × 1}Receive the matrix that the time domain of signal is constituted, diag (p_{N × N}) be phase noise time domain constitute diagonal matrix, h_{N × N}The Toeplitz matrix constituted for the time domain of channel, x_{N × 1}The matrix constituted for transmission data time domain, w_{N × 1}The matrix being configured to for noise time domain；

S25, structure interpolating matrix P_{N × N}, orderWhereinφ_kThe noise being subject to for kth data, k=1,2 ... N.Described in S22A is constructed as primary condition_{1,N × N}, utilize Y described in S23_{N × 1}=A_{N × N}H_{N × N}X_{N × 1}+W_{N × 1}, y described in S24_{N × 1}=diag (p_{N × 1})h_{N × N}x_{N × 1}+w_{N × 1}And p_{N × 1}The modulus value of element is 1 these three condition, and continuous iterative estimate goes outObtained by interpolationRealize the estimation of phase noise, i.e. common phase error (CPE) part；

S26, carry out CPE compensation and channel equalization obtains new data module by receiving signal, the most remaining is made an uproar mutually, i.e. inter-carrier interference (ICI) is estimated, as a example by a data block, general thought is the estimation made an uproar mutually in conjunction with Bayesian Estimation algorithm by data block piecemeal.If inter-carrier interference (ICI) part is made an uproar mutually as u=[φ₁,φ₂,…φ_N], in n iteration, it is M=2 by u cutting^(n-1)Part, such as,WhereinA length of N/M.The phase noise PSD model be given by IEEE802.11ad standard can calculate the auto-correlation function drawing time domain phase noise φ:

$c\left(\mathrm{\τ}\right)=E\[\mathrm{\φ}\left(t\right)\mathrm{\φ}(t+\mathrm{\τ})\]=\frac{K_{\mathrm{\φ}}{f}_p^2}{f_z^2}\mathrm{\δ}\left(\mathrm{\τ}\right)+K_{\mathrm{\φ}}{\mathrm{\πf}}_p(1-\frac{f_p^2}{f_z^2})e^{-2{\mathrm{\πf}}_p\left|\mathrm{\τ}\right|}$

In formula, f_p=1MHz, f_z=100MHz, K_φ=PSD (0).U is understood by formula_mAutocorrelation matrix C be C=V Λ V by Eigenvalues Decomposition^T, wherein Λ=diag{ [λ₁,…λ_L]^T, eigenvalue λ₁＞ λ₂＞ ... ＞ λ_L, V=[v₁,v₂,…v_L] it is each characteristic value characteristic of correspondence vector, according to the smooth change of process of making an uproar mutually, it is known that only fraction of characteristic value is effective, and we are now concentrated at a part of u_mOn be analyzed, u_mΝ (0, C), about V, u_mU can be expressed as_m=Vx, wherein x=[x₁,x₂,…x_L]^T, x_iBe average be zero, variance is λ_iGaussian variable, andIndependentIn x_{i ′}(i≠i′).Because λ_iMajor part is close to zero, so x_iMajor part is also close to zero, say, that vector x adds up sparse, it is contemplated that the impact of noise, can be by u_mWrite as furtherEach element of w be zero-mean variance be δ²'sIndependentGaussian random variable.Our target be exactly fromMiddle estimation x, uses sparse Bayesian algorithm, and x can be recovered as the Gaussian vectors containing average and variance, and wherein average and variance are expressed as

$\stackrel{\→}{\mathrm{\μ}}={\mathrm{\α}}_0{\mathrm{\ΣV}}^T{\tilde{u}}_m$

Σ=(α₀V^TV+A)^-1

In formula, α₀=1/ δ², A=diag{ [α₁,α₂,…α_L]},α_i=1/ λ_i.VectorValue be exactly the value of the required x estimated.Being multiplied by x with V estimating gained and be required noise, followed by making an uproar mutually, compensation suppression can recover accurately data with channel equalization.

The invention has the beneficial effects as follows:

The present invention is to first pass through channel estimation sequence, estimating the response of equivalent dispersion time domain channel impulse, then gone out the common phase error (CPE) of phase noise by Interpolate estimation, the method finally by iteration realizes phase noise compensation, the reliability of raising system, reduces the bit error rate.

Accompanying drawing explanation

Figure 1It it is the single-carrier frequency domain equalization system signal under the effect of phase noise that uses of the present inventionFigure

Figure 2It is that the channel relevant based on sequence that the present invention uses estimates signalFigure；

Figure 3It it is characteristic value normalization signal in theory analysis of the present inventionFigure；

Figure 4Be the present invention realize phase noise estimation compensation suppression flow processFigure；

Figure 5It it is inventive algorithm bit error rate (BER) performance curveFigure。

Detailed description of the invention

Below in conjunction with embodiment andAccompanying drawing, describe technical scheme in detail.

S1, utilize channel estimation sequence realize channel estimate, including:

S11, channel estimation sequence are the sequences being made up of some known symbols, such as in 802.11.ad standard, single carrier channel estimated sequence is [-Gb₁₂₈,-Ga₁₂₈,Gb₁₂₈,-Ga₁₂₈,-Gb₁₂₈,Ga₁₂₈,-Gb₁₂₈,-Ga₁₂₈,-Gb₁₂₈], wherein Ga₁₂₈And Gb₁₂₈It is that Golay sequence is constituted.

S12, the form of matrix: y can be expressed as to obtain the signal of time domain at receiving terminal_{N × 1}=A_{N × N}h_{N × N}x_{N × 1}+w_{N × 1}, wherein, y_{N × 1}It is the form of N × 1 column vector, A_{N × N}It is that the diagonal matrix of a N × N is made up of phase noise, h_{N × N}It is Teoplitz (Toeplitz) matrix being made up of equivalent time domain channel impulse response, x_{N × 1}It is N × 1 column vector being made up of transmission data, w_{N × 1}It it is the noise vector of N × 1.

S13, the conventional channel estimation technique is utilized to estimate, such as: the channel relevant based on sequence is estimated, least square method (LS) channel is estimated, orthogonal matching pursuit algorithm (OMP, Orthogonal Matching Pursuit).

Figure 1It it is the single-carrier frequency domain equalization system signal under the effect of phase noise that uses of the present inventionFigure。

Figure 2It is that the channel relevant based on sequence that the present invention uses estimates signalFigure。

S14, as a example by the channel that sequence is relevant is estimated.If channel estimation sequence meets strong autocorrelation i.e. sequence autocorrelation peak zero secondary lobe occurs, then this character can be utilized to carry out channel estimation.Such as, sequence P=of a length of N [P (0), P (1) ..., P (N-1)]^TThere is strong autocorrelation, meet:

$\mathrm{\ρ}\left(m\right)=\underset{k=0}{\overset{N-1}{\Σ}}P{(k+m)}_{\mathrm{mod}N}{P}^{*}\left(k\right)=\{\begin{array}{cc}N& m\mathrm{mod}N=0\\ 0& m\mathrm{mod}N\≠0\end{array}$

In formula, what mod represented is complementation.As a example by channel estimation sequence in IEEE 802.11ad standard, this sequence is to be made up of Gu512, Gv512 and Gv128, do not consider under influence of noise, receiving terminal receives the sequence that 1152 symbols are constituted, slip is done to the sequence received relevant with Gv512 sequence, owing to the channel estimation sequence in IEEE 802.11ad standard is actually Golay sequence, it meets strong autocorrelation, so have the null value sequence of a section longer at slip autocorrelation peak the right and left, extract the data of correlation peak and a segment length afterwards accordingly as the estimation to channel.The advantage estimated based on sequence correlated channels is to utilize autocorrelation, directly can extract channel response from autocorrelation sequence.

Phase noise is compensated by S2, receiving terminal by iteration, including:

S21, estimated the impulse response of equivalent time domain channel that obtained estimating by channelAs a example by block data transmission, wherein, UW is already known sequence, and UW is added in data sequence to be transmitted, it is achieved remove intersymbol interference.

S22, supposition UW length are more than equivalent time domain channel length, the data sequence of transmission there is UW, UW is equivalent time domain channel impulse response h and UW convolution by channel essence, but it is affected by phase noise and white Gaussian noise simultaneously, the phase noise being subject in UW sequence is that (the phase noise constant that different UW are subject to is different for a constant, phase noise constant is really the common phase error CPE of phase noise), i-th UW by reception signal during channel can be with approximate representation:By the UW in transmission data sequence, the method for interpolation is taked to estimate the phase noise constant of i-th data block and the ratio of a transmitted between i-th UW and i+1 UW:

S23, for receiving for signal, remove CP and be followed by receiving signal frequency domain and can be expressed as with matrix form: Y_{N × 1}=A_{N × N}H_{N × N}X_{N × 1}+W_{N × 1}, wherein A_{N × N}The Toeplitz matrix constituted for phase noise frequency domain, H_{N × N}For estimating the diagonal matrix that the frequency domain of channel is constituted, X_{N × 1}The matrix constituted for transmission data frequency domain, Y_{N × 1}For receiving the matrix that the frequency domain of signal is constituted, W_{N × 1}Matrix for white Gaussian noise frequency domain structure.Correspondingly, receiving signal time-domain representation is: y_{N × 1}=diag (p_{N × 1})h_{N × N}x_{N × 1}+w_{N × 1}, wherein, y_{N × 1}Receive the matrix that the time domain of signal is constituted, diag (p_{N × N}) be phase noise time domain constitute diagonal matrix, h_{N × N}The Toeplitz matrix constituted for the time domain of channel, x_{N × 1}The matrix constituted for transmission data time domain, w_{N × 1}The matrix being configured to for noise time domain.In order to reduce complexity, by structure interpolating matrix P_{N × N}, make p_{N × 1}=P_{N × N}c_{s × 1}.WeA is constructed as primary condition_{1,N × N}, utilize Y_{N × 1}=A_{N × N}H_{N × N}X_{N × 1}+W_{N × 1}And y_{N × 1}=diag (p_{N × 1})h_{N × N}x_{N × 1}+w_{N × 1}And p_{N × 1}The modulus value of element is 1 these three condition.Gone out by continuous iterative estimateObtained by interpolationThus make use of the method for iteration to achieve the estimation of phase noise.

S24, making an uproar mutually of now obtaining are CPE part, reception signal is carried out CPE compensation and channel equalization obtains new data module, the most remaining make an uproar mutually (i.e. ICI) is estimated, as a example by a data block, general thought is the estimation made an uproar mutually in conjunction with Bayesian Estimation algorithm by data block piecemeal.If ICI part is made an uproar mutually as u=[φ₁,φ₂,…φ_N], in n iteration, it is M=2 by u cutting^(n-1)Part, such as,WhereinA length of N/M.The phase noise PSD model be given by IEEE802.11ad standard can calculate the auto-correlation function drawing time domain phase noise φ:

$c\left(\mathrm{\τ}\right)=E\[\mathrm{\φ}\left(t\right)\mathrm{\φ}(t+\mathrm{\τ})\]=\frac{K_{\mathrm{\φ}}{f}_p^2}{f_z^2}\mathrm{\δ}\left(\mathrm{\τ}\right)+K_{\mathrm{\φ}}{\mathrm{\πf}}_p(1-\frac{f_p^2}{f_z^2})e^{-2{\mathrm{\πf}}_p\left|\mathrm{\τ}\right|}$

In formula, f_p=1MHz, f_z=100MHz, K_φ=PSD (0).

Figure 3It it is characteristic value normalization signal in theory analysis of the present inventionFigure；

U is understood by formula_mAutocorrelation matrix C be C=V Λ V by Eigenvalues Decomposition^T, wherein Λ=diag{ [λ₁,…λ_L]^T, eigenvalue λ₁＞ λ₂＞ ... ＞ λ_L, V=[v₁,v₂,…v_L] it is each characteristic value characteristic of correspondence vector, according to the smooth change of process of making an uproar mutually, it is known that only fraction of characteristic value is effective, and we are now concentrated at a part of u_mOn be analyzed, about V, u_mU can be expressed as_m=Vx, wherein x=[x₁,x₂,…x_L]^T, x_iBe average be zero, variance is λ_iGaussian variable, andIndependentIn x_{i ′}(i≠i′).Because λ_iMajor part is close to zero, so x_iMajor part is also close to zero, say, that vector x adds up sparse, it is contemplated that the impact of noise, can be by u_mWrite as furtherEach element of w be zero-mean variance be δ²'sIndependentGaussian random variable.Our target be exactly fromMiddle estimation x, uses sparse Bayesian algorithm, and x can be recovered as the Gaussian vectors containing average and variance, and wherein average and variance are expressed as

$\stackrel{\→}{\mathrm{\μ}}={\mathrm{\α}}_0{\mathrm{\ΣV}}^T{\tilde{u}}_m$

∑=(α₀V^TV+A)^-1

In formula, α₀=1/ δ², A=diag{ [α₁,α₂,…α_L]},α_i=1/ λ_i.VectorValue be exactly the value of the required x estimated.Being multiplied by x with V estimating gained and be required noise, followed by making an uproar mutually, compensation suppression can recover accurately data with channel equalization.

Figure 4Be the present invention realize phase noise estimation compensation suppression flow processFigure；

Figure 5It is to useFigure 1SystemFigure,Figure 2Channel estimateFigure,Figure 3Characteristic value normalizationFigure,Figure 4Algorithm flow, be applied in concrete communication system, choose single-carrier frequency domain equalization system here as an example, what emulation obtained is inventive algorithm bit error rate (BER) performance curve in single carrier frequency domain systemFigure.The analogue system of this example is belonging to high-frequency high-speed ultra-wideband communication system, its main simulation parameter is: carrier frequency is 60GHz, character rate is 1.76Gbps, 16QAM modulates, phase noise is-86dbc/Hz@1MHz, it is 1 iteration and 2 iteration respectively, the most i.e. can be seen that described algorithm greatly to achieve mutually to make an uproar compensation and suppression.

Claims (2)

1. make an uproar compensation method based on Bayesian phase for one kind, it is characterised in that specifically comprise the following steps that

S1, receiving terminal utilize channel estimation sequence carry out channel estimate obtain the transmitting terminal impulse to the equivalent time domain channel of receiving terminal Response estimation value

S2, obtaining the signal of time domain at receiving terminal and affected by phase noise and white Gaussian noise, receiving terminal passes through iteration pair Phase noise compensates, described phase noise phase noise power spectrum density (Phase noise power spectrum Density, PSD) characterize, i.e.Wherein, f represents the frequency at offset carrier center Rate, PSD (0) is a constant, f_p=1MHz is pole frequency, f_Z=100MHz is zero frequency, and described PSD (f) is " pole/zero " model about phase noise PSD that communication standard IEEE802.15.3c and IEEE802.11ad is given, PSD (0)=-87dbc/Hz in IEEE802.15.3c, PSD (0)=-90dbc/Hz in IEEE802.11ad standard.

2. make an uproar compensation method based on Bayesian phase according to the one described in claim, it is characterised in that: receiving terminal described in S2 By iteration phase noise compensated and specifically comprise the following steps that

S21, UW is added in data sequence to be transmitted, a data block contains two UW wherein, described UW's Length is more than the length of equivalent time domain channel described in S1, and UW is made up of Gray's known array of 64, treating in data block Passing data is 448 bit sign sequences；

S22, the method for employing interpolation estimate the phase place of the i-th data block of transmission between i-th UW and i+1 UW Noise constant and the ratio of aWherein, described i-th UW is by reception during channel Signal is expressed as:Wherein a It is phase noise constant, y_uw_(i)It is that i-th UW is by reception signal during channel, a_uw_(i)It is that i-th UW is passed through Constant of making an uproar mutually during channel,It is equivalent time domain channel impulse response, w_(i)Be i-th UW by channel time be subject to make an uproar Sound,Being the length of channel, i is the natural number being not zero；

S23, removal Cyclic Prefix (Cyclic Prefix, CP) are followed by receiving signal frequency domain matrix form and are expressed as:

Y_N×1=A_N×NH_N×NX_N×1+W_N×1, wherein, Y_N×1For receiving the matrix that the frequency domain of signal is constituted, A_N×NFor phase noise frequency domain The Toeplitz matrix constituted, H_N×NFor estimating the diagonal matrix that the frequency domain of channel is constituted, X_N×1Constitute for transmission data frequency domain Matrix, W_N×1Matrix for white Gaussian noise frequency domain structure；

S24, receive signal time-domain representation be: y_N×1=diag (p_N×1)h_N×Nx_N×1+w_N×1, wherein, y_N×1Receive the time domain of signal The matrix constituted, diag (p_N×N) be phase noise time domain constitute diagonal matrix, h_N×NThe Toeplitz constituted for the time domain of channel Matrix, x_N×1The matrix constituted for transmission data time domain, w_N×1The matrix being configured to for noise time domain；

S25, structure interpolating matrix P_N×N, orderDescribed in S22Construct as primary condition A₁,_N×N, utilize Y described in S23_N×1=A_N×NH_N×NX_N×1+W_N×1, y described in S24_N×1=diag (p_N×1)h_N×Nx_N×1+w_N×1With p_N×1The modulus value of element is 1 these three condition, and continuous iterative estimate goes outObtained by interpolationRealize phase place to make an uproar The estimation of sound, i.e. common phase error (CPE) part, whereinφ_kThe noise being subject to for kth data, K=1,2 ... N；

S26, carry out CPE compensation and channel equalization obtains new data module by receiving signal, remaining is made an uproar mutually, i.e. carries (ICI) is disturbed to estimate between ripple.