CN103685124B - Compressed domain frequency shift estimation method - Google Patents

Abstract

The invention discloses a kind of compression domain frequency offset estimation methods, the carrier frequency synchronizations of each distinguishable path signal for wireless communication system, comprising: construction perception matrix Θ, and extract the reception signal sequence x on distinguishable path; Using the offset estimation pretreatment for receiving signal sequence x and carrying out compression domain on the path on perception matrix Θ and the distinguishable path, the pretreatment estimated value is obtained According to pretreatment estimated value Map out offset estimation value of the distinguishable path signal in uncompressed domain Complete Carrier frequency offset estimation when wireless system Frequency Synchronization. The present invention can greatly save processor resource.

Description

Compressed domain frequency offset estimation method

Technical Field

The present invention relates to a frequency synchronization technique in a wireless communication system, and more particularly, to a frequency offset (carrier frequency offset is referred to as frequency offset) estimation method using a compressed sensing technique.

Background

In a wireless communication system, carrier frequency offset (frequency offset for short) is caused by two aspects: on one hand, due to the instability of the oscillator, frequency deviation exists between local carriers of a transmitter and a receiver; on the other hand, relative motion between the transmitter and the receiver produces a doppler shift. The presence of frequency offset can significantly degrade the performance of the wireless communication system, such as parameter estimation performance, bit error rate performance, and the like. Therefore, it is generally necessary to compensate for the frequency offset prior to demodulation of the received signal.

As a primary task of frequency offset compensation, frequency offset estimation is a technical hotspot and research difficulty in the past. The existing frequency offset estimation methods mainly include two main types, namely a data-aided method and a non-data-aided method, wherein the data-aided method utilizes a pilot frequency sequence or a training sequence to carry out frequency offset estimation, and the non-data-aided method utilizes the self characteristics of a received signal to carry out frequency offset estimation. In practical wireless communication systems, non-data aided frequency offset estimation methods are less adopted due to their excessive computational load. Although the data-assisted frequency offset estimation method has certain feasibility, the data-assisted frequency offset estimation method such as the maximum likelihood method adopted for obtaining good estimation performance still has larger computational complexity and occupies more processor resource overhead.

On the other hand, the arrival direction of different radio path signals at the receiving end is usually different. Therefore, the frequency offsets of the different path signals may be different. Thus, the frequency offset of each resolvable path signal needs to be estimated and compensated separately. This further increases processor resource overhead in frequency offset estimation.

The theory of compressed sensing appearing in recent years indicates that, as long as the signal is sparse in a certain transform domain, the compressed sensing method can be used for projecting a high-dimensional signal onto a low-dimensional space to realize the compression processing of the signal. The original signal is then reconstructed with high probability from the compressed small amount of information. In frequency offset estimation in a wireless communication system, the amplitude of the estimated observation quantity usually has a larger amplitude value only at a few frequency points relative to the whole observation space. The observed quantity can be regarded as approximate sparsity in an observation space, and frequency offset estimation by utilizing a compressed sensing technology becomes possible. Therefore, in order to reduce the resource overhead of the wireless communication system in the frequency offset estimation process, the invention provides a frequency offset estimation method based on a compressed sensing technology.

Disclosure of Invention

The invention mainly aims to provide a novel frequency offset estimation method, which aims to solve the technical problem of greatly reducing the calculation amount during frequency offset estimation based on a compressed sensing technology, and reducing the processor resource overhead of a wireless communication system in the frequency offset estimation process, so that the method is more suitable for practical application.

The frequency offset estimation method is performed when the time synchronization has been completed and a set of starting points S of distinguishable paths has been formed. For example, the known training sequences of both the transmitter and the receiver can be used to perform sliding correlation operation on the received signal to obtain a correlation peak position with an amplitude value greater than a predetermined detection threshold, where the correlation peak position is the starting position of the distinguishable path signal.

The method comprises the following steps:

a compressed domain frequency offset estimation method for carrier frequency synchronization of each resolvable path signal of a wireless communication system comprises the following steps:

a1, extracting a received signal sequence x with length of Nx 1 on a distinguishable path according to the starting position of the distinguishable path;

a2, the receiving end uses the constructed sensing matrix theta and the received signal sequence x on the path on the distinguishable path to perform frequency offset estimation preprocessing of the compressed domain to obtain the preprocessed estimation value；

A3, estimating value according to pretreatmentMapping out frequency offset estimates of the resolvable path signals in the uncompressed domain；

A4, deleting the element corresponding to the currently processed path in the starting point set S of the resolvable path; then, processing the frequency offset estimation of the next resolvable path under the condition that S is not a null set; otherwise, ending the frequency offset estimation process.

In the method, the sensing matrix theta used for the compressed domain frequency offset estimation preprocessing is constructed in advance and stored in a designated storage space.

In the method, the construction step of the perceptual matrix Θ used for preprocessing the frequency offset estimation in the compressed domain includes:

b1, using said known training sequence a ═ a₁,a₂,…,a_N]^TThe length N of the target is obtained, and a sparse level K is set by combining engineering experience;

b2, setting a frequency search step length delta f according to the frequency offset estimation precision requirement and the maximum possible frequency offset value;

b3, calculating the search length of the frequency offset estimation in the uncompressed domainAnd length of observed quantity in compressed domainWherein, the symbolDenotes the rounding-up operation on x, said f_maxTaking the value of the maximum possible frequency offset; typically, M < Z;

b4, constructing an M multiplied by Z observation matrix phi;

b5, constructing Z uncompressed domain frequency offset estimation search matrixes of N × N

B6, estimating and searching matrix by using each element of the observation matrix phi and frequency offset of the uncompressed domainAnd a known training sequence A, constructing a sub-matrix theta of the perception matrix theta_iI is 1,2, …, M, hasWherein, the superscript "H" represents taking conjugate transpose operation;

b7, using said Θ_iI-1, 2, …, M forming said perceptual matrix Θ, having

The method is characterized in that the construction method of the observation matrix phi of M multiplied by Z comprises the following steps: get

The method, the Z uncompressed domain frequency offset estimation search matrixes of N × NThe structure of, saidIs a diagonal matrix and hasWherein the frequency offset attempt valuei＝1,2,…,Z。

In the method, the frequency offset estimation preprocessing process includes:

utilizing the submatrix theta of the perception matrix theta_iI is 1,2, …, M and the received signal sequence x on the resolvable path, and form a set y { | Θ of frequency offset estimation preprocessing observations in the compressed domain₁x|²,|Θ₂x|²,…,|Θ_Mx|²For convenience of description, the compressed domain frequency offset estimation preprocessing observation quantity set y is expressed as y { | y (0) | y²,|y(1)|²,…|y(ρ)|²,|y(ρ+1)|²,…,|y(M-1)|²}；

Finding the position of the maximum value from the set of preprocessed observations yNamely, it is

And finishing the processing flow of the frequency offset estimation preprocessing of the compressed domain.

The method, the pre-processing of the estimateMapping out frequency offset estimates in the uncompressed domainThe treatment process comprises the following steps:

according to the pre-processing estimated valueFrom the collectionMiddle truncated element, forming a truncated set of frequency offset attempt valuesWherein the frequency offset attempt valuei＝1,2,…,Z；

Constructing a set of observations y from said truncated set β₁Is provided with

Wherein,

finding out the frequency deviation value corresponding to the maximum value element from the observation quantity set y1, namely the frequency deviation estimation valueNamely:

$\begin{array}{c}\hat{v}=\arg \underset{{\tilde{v}}_i\&Element;\mathrm{\&beta;}}{\max }\left\{y_1\right\}\\ =\arg \underset{{\tilde{v}}_i\&Element;\mathrm{\&beta;}}{\max }\left\{\right|A^H{\mathrm{\&Gamma;}}^H\left({\tilde{v}}_{2K(\widehat{\mathrm{\&rho;}}-1)+1}\right)x{|}^2,|A^H{\mathrm{\&Gamma;}}^H\left({\tilde{v}}_{2K(\widehat{\mathrm{\&rho;}}-1)+2}\right)x{|}^2,...,|A^H{\mathrm{\&Gamma;}}^H\left({\tilde{v}}_{\min (2K\widehat{\mathrm{\&rho;}},Z)}\right)x{|}^2\}\end{array}.$

from the above, it can be seen that the frequency offset estimation method based on the compressed sensing technology provided by the invention has the following characteristics and advantages:

1) the perception matrix theta is constructed in advance before the communication process is initiated, and real-time construction is not needed. Therefore, the processing time and the processor resources of the wireless communication system in the communication process can be not occupied.

2) In frequency offset estimation, a compressed sensing technology is adopted for processing, and the operation amount in processing is greatly reduced. Thereby reducing the processor resource overhead of the wireless communication system in the frequency offset estimation process.

Drawings

Fig. 1 is a frame format of a radio frame in which a training sequence is transmitted by a transmitter, which is applicable to the present invention;

FIG. 2 is a flow diagram of a method of frequency offset estimation according to an embodiment of the present invention;

FIG. 3 is a flow chart illustrating the construction of a sensing matrix in a frequency offset estimation method according to an embodiment of the present invention;

fig. 4 is a flow chart of a frequency offset estimation preprocessing in the compressed domain by the frequency offset estimation method according to an embodiment of the invention.

Fig. 5 is a flow diagram of a process for mapping a compressed domain estimate to a frequency offset estimate in an uncompressed domain by a frequency offset estimation method according to an embodiment of the invention.

Detailed Description

The present invention will be described in detail with reference to specific examples.

In the embodiment of the present invention, a schematic diagram of a signal frame format of a wireless communication system is shown in fig. 1. When the frequency offset estimation is performed, a training sequence-based method is adopted, and the position of the training sequence can be from the frame header or other positions in the frame.

The receiving antenna of the receiving end receives the data frame transmitted by the transmitting end, the length of the training sequence is set to be N, namely N sampling points, and the length of the received signal is greater than N (2N data can be taken for observation).

Before the frequency offset estimation processing process, the compressed sensing matrix theta is constructed in advance and stored in the designated sensing matrix storage area, so that the processing time and the processing resources of a processor in the actual communication process are not occupied. As shown in fig. 2, the construction process of the perception matrix Θ is as follows:

and 2.1, reading the length N of the training sequence, and setting the sparse level K and the frequency search step length delta f.

The training sequence A ═ a₁,a₂,…,a_N]^TAnd the length N is a sequence and length value known by both the transmitting and receiving parties, wherein the superscript T represents the transposition operation.

The sparse level K is set according to the training sequence length N and by combining engineering experience. For example, for the case of training sequence length N being 512, the sparsity level K being 20 may be set.

The frequency searching step length delta f is set according to the frequency offset estimation precision requirement and the maximum possible frequency offset value f_maxAnd (4) setting. For example, the maximum possible frequency offset value f_maxThe required estimation accuracy is 1Hz at 1000Hz, and the frequency search step Δ f may be set at 1 Hz.

Step 2.2, calculating the frequency offset estimation search length in the non-compression domain according to the search step length delta fCalculating the length of the observed quantity in the compressed domain according to the search length Z and the sparse level KWherein, the symbolDenotes the rounding-up operation on x, f_maxTaking the value of the maximum possible frequency offset; typically, M < Z.

Step 2.3, according to the search length Z and the length M of the observed quantity in the compressed domain, constructing an M multiplied by Z observation matrix phi, namely taking

Step 2.4, Z uncompressed domain frequency offset estimation search matrixes of N × N are constructed according to the sequence length N and the frequency search step delta fWhereini＝1,2,…,Z，As a diagonal matrix, i.e.

Wherein,representing a complex number.

Step 2.5, using each element of the constructed observation matrix phi to estimate and search the matrix by the frequency deviation of the uncompressed domainAnd a known training sequence a ═ a₁,a₂,…,a_N]^TConstructing a sub-matrix theta of the perception matrix theta_iI is 1,2, …, M, has

${\mathrm{\&Theta;}}_i={\mathrm{\&phi;}}_{i1}{A}^H{\mathrm{\&Gamma;}}^H\left({\tilde{v}}_1\right)+{\mathrm{\&phi;}}_{i2}{A}^H{\mathrm{\&Gamma;}}^H\left({\tilde{v}}_2\right)+...+{\mathrm{\&phi;}}_{iZ}{A}^H{\mathrm{\&Gamma;}}^H\left({\tilde{v}}_Z\right),i=1,2,...,M$

Wherein, the superscript "H" represents the conjugate transpose operation. Using theta_iI-1, 2, …, M forms the perceptual matrix Θ, havingAnd (4) putting the constructed sensing matrix into a specified storage area, thereby completing the processing flow of constructing the sensing matrix theta.

The following describes the frequency offset estimation process in detail, as shown in fig. 3.

The frequency offset estimation method is performed when the time synchronization has been completed and a set of starting points S of distinguishable paths has been formed. For example, the known training sequences of both the transmitter and the receiver can be used to perform sliding correlation operation on the received signal to obtain a correlation peak position with an amplitude value greater than a predetermined detection threshold, where the correlation peak position is the starting position of the distinguishable path signal. The invention mainly focuses on the frequency offset estimation technology, and the processing process is as follows:

and 3.1, extracting a receiving signal sequence x with the length of Nx 1 on one distinguishable path according to the initial position of the distinguishable path.

Example (c): if N is 8, the received signal sequence is { x }₁,x₂,…,x₈,x₉,…,x₁₆The starting point set of resolvable paths S ═ 2,3, 5. Thus, it can be extracted that the received signal sequence whose start position is "2" is x ═ x₂,x₃,…,x₉}。

Step 3.2, the frequency deviation estimation pretreatment of the compression domain is carried out by utilizing the perception matrix theta and the received signal sequence x on the path to obtain the pretreatment estimation value. The specific frequency offset estimation preprocessing flow is shown in fig. 4 and is described as follows:

and 3.2.1, extracting the perception matrix theta from the perception matrix storage space.

Step 3.2.2, utilize the submatrix theta of theta_iI is 1,2, …, M is multiplied by x one by one, and constitutes the frequency offset estimation preprocessing observation set y { | Θ of the compressed domain₁x|²,|Θ₂x|²,…,|Θ_Mx|²}。

For convenience of description, the set y of preprocessed observations is expressed as:

y＝{|y(0)|²,|y(1)|²,…,|y(M-1)|²}。

step 3.2.3, finding out the position of the maximum value from the observation quantity set y, namely the position of the maximum value is the preprocessing estimation valueHere, the positions of the elements in the set are expressed as ρ, ρ ∈ {0,1, …, M-1}, having

Thus, the frequency offset estimation preprocessing flow of the compression domain is completed.

Step 3.3, frequency offset estimation using preprocessingMapping the frequency offset estimate of the resolved path signal in the uncompressed domainThe specific mapping process flow is shown in fig. 5, and is described as follows:

step 3.3.1, estimate based on pretreatmentFrom the collectionThe truncated elements form a truncated frequency setWherein,i＝1,2,…,Z。

step 3.3.2, receiving signal sequence x and known training sequence a ═ a according to said truncated set of frequencies β₁,a₂,…,a_N]^TConstructing a set of observations y₁：

$y_1=\left\{\right|A^H{\mathrm{\&Gamma;}}^H\left({\tilde{v}}_{2K(\widehat{\mathrm{\&rho;}}-1)+1}\right)x{|}^2,|A^H{\mathrm{\&Gamma;}}^H\left({\tilde{v}}_{2K(\widehat{\mathrm{\&rho;}}-1)+2}\right)x{|}^2,...,|A^H{\mathrm{\&Gamma;}}^H\left({\tilde{v}}_{min(2K\widehat{\mathrm{\&rho;}},Z)}\right)x{|}^2\}.$

Wherein, the superscript "H" represents the conjugate transpose operation,and is

Step 3.3.3, from the set of observations y₁Finding out the frequency deviation value corresponding to the maximum value element as the frequency deviation estimated valueThat is to say are

$\begin{array}{c}\hat{v}=\arg \underset{{\tilde{v}}_i\&Element;\mathrm{\&beta;}}{\max }\left\{y_1\right\}\\ =\arg \underset{{\tilde{v}}_i\&Element;\mathrm{\&beta;}}{\max }\left\{\right|A^H{\mathrm{\&Gamma;}}^H\left({\tilde{v}}_{2K(\widehat{\mathrm{\&rho;}}-1)+1}\right)x{|}^2,|A^H{\mathrm{\&Gamma;}}^H\left({\tilde{v}}_{2K(\widehat{\mathrm{\&rho;}}-1)+2}\right)x{|}^2,...,|A^H{\mathrm{\&Gamma;}}^H\left({\tilde{v}}_{\min (2K\widehat{\mathrm{\&rho;}},Z)}\right)x{|}^2\}\end{array}$

And 3.3.4, deleting the element corresponding to the currently processed path in the starting point set S of the resolvable path.

Example (c): for example, if the starting point set S of the resolvable path is S ═ {2,3,5,7}, and the starting point corresponding to the currently processed path is 3, the element "3" in S should be deleted, and the updated set S is S ═ {2,5,7 }.

Step 3.4, judging whether all the distinguishable paths are processed completely, namely judging whether the starting point set S of the distinguishable paths is an empty set, if so, ending the frequency offset estimation process; otherwise, step 3.5 is executed.

And 3.5, reading an element from the starting position set S of the resolvable path, and returning to the step 3.1 to process the next resolvable path.

The next resolvable path is the path corresponding to the starting point of the path, which is the element read from the set S in step 3.5.

In the following, comparison is made in terms of reducing the amount of computation of the present invention as a whole.

In the traditional method, complex multiplication NZ times, complex addition (N-1) Z times, complex modular operation Z times and real number comparison times Z-1 times when searching the maximum value are required to be calculated.

In the method, in the compressed domain preprocessing, the complex multiplication NM times, the complex addition (N-1) M, the complex modular operation M times and the real number comparison times M-1 times when the maximum value is searched are required to be calculated. In the frequency offset estimation mapping of the uncompressed domain, complex multiplication is required to be performed for 2KN times, complex addition is required to be performed for 2K (N-1), modular operation of the complex is required to be performed for 2K times, and real number comparison times in searching for the maximum value is required to be performed for 2K-1 times.

Therefore, the invention needs total complex multiplication MN +2KN times, the number of complex addition is (N-1) M +2K (N-1), the number of modular operation M +2K of complex, and the number of real comparison M +2K-2 when searching the maximum value.

For example, taking N512, from engineering experience, K20; in order to ensure the estimation accuracy, the value of Z is usually a positive integer multiple of N, and Z is 4N 2048 without loss of generality.The control ratios are shown in table 1.

TABLE 1

According to the specific example, the method greatly reduces the operation amount and saves the processor resources because of utilizing the compressed sensing technology.

It will be understood that modifications and variations can be made by persons skilled in the art in light of the above teachings and all such modifications and variations are intended to be included within the scope of the invention as defined in the appended claims.

Claims (2)

1. A compressed domain frequency offset estimation method for carrier frequency synchronization of each resolvable path signal in a wireless communication system, comprising the steps of:

a1, extracting a received signal sequence x with length of Nx 1 on a distinguishable path according to the starting position of the distinguishable path;

a2, the receiving end uses the constructed sensing matrix theta and the received signal sequence x on the distinguishable path to perform frequency offset estimation preprocessing of the compressed domain to obtain the preprocessed estimation valueA sensing matrix theta used for compressed domain frequency offset estimation preprocessing is constructed in advance and stored in a designated storage space;

the construction step of the sensing matrix theta used for the compressed domain frequency offset estimation preprocessing comprises the following steps:

b1, using a known training sequence a ═ a₁,a₂,…,a_N]^TThe length N of the target is obtained, and a sparse level K is set by combining engineering experience;

b2, setting a frequency search step length delta f according to the frequency offset estimation precision requirement and the maximum possible frequency offset value;

b3, calculating the search length of the frequency offset estimation in the uncompressed domainAnd length of observed quantity in compressed domainWherein, the symbolDenotes the rounding-up operation on x, f_maxTaking the value of the maximum possible frequency offset; m<<Z；

B4, constructing an M multiplied by Z observation matrix phi;

b5, constructing Z uncompressed domain frequency offset estimation search matrixes of N × NThe Z uncompressed domain frequency offset estimation search matrixes of N × NIn the above-described manner, the structure of (1),is a diagonal matrix and has

Wherein the frequency offset trial valuei＝1,2,…,Z；

B6, estimating and searching matrix by using each element of the observation matrix phi and frequency offset of the uncompressed domainAnd a known training sequence A, constructing a sub-matrix theta of the perception matrix theta_iI is 1,2, …, M, hasWherein, the superscript "H" represents taking conjugate transpose operation;

b7, using said Θ_iI-1, 2, …, M forming said perceptual matrix Θ, having

The frequency offset estimation preprocessing process comprises the following steps:

utilizing the submatrix theta of the perception matrix theta_iI is 1,2, …, M and the received signal sequence x on the resolvable path, and form a set y { | Θ of frequency offset estimation preprocessing observations in the compressed domain₁x|²,|Θ₂x|²,…,|Θ_Mx|²Expressing the compressed domain frequency offset estimation preprocessing observation quantity set y as

y＝{|y(0)|²,|y(1)|²,…|y(ρ)|²,|y(ρ+1)|²,…,|y(M-1)|²}; finding the position of the maximum value from the set of preprocessed observations yNamely, it isCompleting the processing flow of the frequency offset estimation preprocessing of the compressed domain;

a3, estimating value according to pretreatmentMapping out frequency offset estimates of the resolvable path signals in the uncompressed domainBy preprocessing the estimatesMapping out frequency offset estimates in the uncompressed domainThe treatment process comprises the following steps:

according to the pre-processing estimated valueFrom the collectionMiddle truncated element, forming a truncated set of frequency offset attempt valuesWherein the frequency offset trial valuei＝1,2,…,Z；

Constructing a set of observations y from said truncated set β₁Is provided with

Wherein,

from said set of observations y₁Finding out the frequency deviation value corresponding to the maximum value element as the frequency deviation estimated valueNamely:

$\begin{array}{c}\hat{v}=\arg \underset{{\widetilde{\mathrm{\&nu;}}}_i\&Element;\mathrm{\&beta;}}{\max }\left\{y_1\right\}\\ =\arg \underset{{\widetilde{\mathrm{\&nu;}}}_i\&Element;\mathrm{\&beta;}}{\max }\{{\left|A^H{\mathrm{\&Gamma;}}^H\left({\widetilde{\mathrm{\&nu;}}}_{2K(\widehat{\mathrm{\&rho;}}-1)+1}\right)x\right|}^2,{\left|A^H{\mathrm{\&Gamma;}}^H\left({\widetilde{\mathrm{\&nu;}}}_{2K(\widehat{\mathrm{\&rho;}}-1)+2}\right)x\right|}^2,...,{\left|A^H{\mathrm{\&Gamma;}}^H\left({\widetilde{\mathrm{\&nu;}}}_{\min (2K\widehat{\mathrm{\&rho;}},Z)}\right)x\right|}^2\};\end{array}$

a4, deleting the element corresponding to the currently processed path in the starting point set S of the resolvable path; then, processing the frequency offset estimation of the next resolvable path under the condition that S is not a null set; otherwise, ending the frequency offset estimation process.

2. The method of claim 1, wherein the M x Z observation matrix Φ is constructed by: get