A particle filtering method for carrier frequency offset tracking in channel state information
Through particle filtering method combined with ESPRIT algorithm, the problems of low robustness and high computational complexity of long-term tracking of carrier frequency offset are solved, and high precision and low complexity of long-term tracking of carrier frequency offset in Wi-Fi sensing systems are realized , meets the demand for stable calibration of Wi-Fi sensing systems.
Patent Information
- Application Number
- CN202311368409.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-20
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2043-10-20
AI Technical Summary
The prior art has problems of low robustness and high computational complexity in long-term tracking of carrier frequency offsets, which is difficult to meet the precise calibration requirements of Wi-Fi sensing systems for stable carrier frequency offsets.
The particle filtering method combined with the ESPRIT algorithm is used to estimate the carrier frequency offset of the training segment frame, and the state transition function of the state variable is constructed through the particle filter, and the set of prior and posterior particles is calculated to realize long-term dynamic tracking of the carrier frequency offset. .
It realizes high-precision and robust long-term tracking of carrier frequency offset, reduces the computational complexity, and only a commercial network card can complete the long-term tracking of carrier frequency offset, reducing operating costs.
Smart Images

Figure CN117640316B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of channel measurement of wireless communications, and in particular relates to a particle filtering method for tracking carrier frequency deviation in channel state information. Background Art
[0002] Wi-Fi sensing capabilities are highly dependent on the accuracy of CSI (channel state information) measurements, but the CFO (carrier frequency offset) generated by the difference in local crystal oscillator frequencies between the transmitter and receiver will cause cumulative phase errors in the CSI. In addition, the logic interface is sensitive to temperature changes. Long-term operation of the device will increase the internal temperature of the device, causing the carrier frequency offset to drift, thereby causing more phase errors in the CSI. Therefore, the carrier frequency offset has a serious impact on the development of CSI phase-based sensing technology, hindering the development of new-generation Wi-Fi technologies such as wireless phase synchronization, Wi-Fi phased array, and Wi-Fi imaging.
[0003] In the prior art, there are relatively mature estimation methods for carrier frequency offset, such as invention patent CN201310363774.6, a carrier frequency offset estimation algorithm for an orthogonal frequency division multiplexing system under a fast-changing channel. This method can effectively reduce the computational complexity, and after simulation verification, the improved algorithm performs well; invention patent CN201610147118.6, a carrier frequency offset estimation method suitable for a MIMO-OFDM system, which uses the received short training sequences and long training sequences to perform coarse and fine frequency offset estimation and compensation, thereby effectively improving the accuracy of carrier frequency offset estimation in the MIMO system; invention patent 202210328295.X, a pilot-based carrier frequency offset estimation method, which can expand the estimation range while maintaining the estimation accuracy.
[0004] The estimation methods provided above all estimate the carrier frequency offset effectively, but the method based on autocorrelation operation is sensitive to noise, and the calculation amount of the autocorrelation function is large. The estimation method based on pilot has stricter requirements on the selection and transmission of pilot. In high-speed mobile communication systems, if the transmission period of the pilot is too long, the cumulative error of the frequency offset may be large; if the transmission period of the pilot is too short, the length of the pilot sequence may not be enough to support accurate frequency offset estimation. These limitations make it difficult to apply the work of the above methods to a wide range of Wi-Fi sensing practices. More importantly, Wi-Fi sensing requires long-term stable carrier frequency offset tracking to accurately calibrate it, and short-term carrier frequency offset estimation cannot meet the needs of Wi-Fi sensing. Therefore, a highly robust and low-computational-overhead long-term tracking and estimation method for carrier frequency offset is needed. Summary of the invention
[0005] In order to solve the problem that the carrier frequency offset cannot be tracked dynamically for a long time in the background technology, resulting in low estimation accuracy and low robustness, the present invention provides a particle filtering method for tracking the carrier frequency offset in channel state information. The ESPRIT algorithm is used to estimate the carrier frequency offset of the training segment frame. On this basis, the particle filtering is performed on the Wi-Fi frame to obtain a long-term dynamic carrier frequency offset estimation value sequence of the tracking segment frame CSI. The method has the beneficial effects of high robustness, low computational overhead and high estimation accuracy.
[0006] The object of the present invention is achieved through the following technical solutions:
[0007] A particle filtering method for tracking carrier frequency offset in channel state information comprises the following steps:
[0008] S1. The transmitter sends several Wi-Fi frames to the receiver. The receiver interpolates the DC subcarrier of the channel state information of each frame and obtains the phase sequence θ containing the DC subcarrier. DC , the phase sequence Θ DC Training frame phase sequence and tracking segment frame phase sequence And construct the frame interval sequence dt;
[0009] S2, the training segment frame phase sequence is analyzed by ESPRIT algorithm Perform spectrum estimation to obtain the carrier frequency offset estimate of the training segment frame
[0010] S3, construct a state vector X of a particle filter for carrier frequency offset tracking, wherein the state vector X includes four state variables, which are the carrier frequency offset change rate f rate , carrier frequency offset value f cfo , the phase accumulation of the unwrapped unwrap And the unwrap phase accumulation θ wrap , where the value of k represents the four state variables in turn, and the value range of k is 1 to 4;
[0011] S4, through the state transfer function F k Calculate the prior particle set of the four state variables at state t respectively
[0012] S5. Through the prior particle set With the phase sequence Θ DC Calculate the particle weight set W(t);
[0013] S6. The particle weight set W(t) is used to calculate the prior particle set Resample and construct the posterior particle set of 4 state variables
[0014] S7. Through the posterior particle set The expectation of is used as the posterior prediction value of the four state variables The state variable carrier frequency offset value f cfo The posterior predictive value of This is the result of dynamic tracking of the carrier frequency offset.
[0015] Compared with the prior art, the beneficial effects of this technical solution are:
[0016] (1) The ESPRIT algorithm is used to calculate the estimated value of the carrier frequency offset of the training segment frame, which ensures the accuracy of the carrier frequency offset estimation while reducing the time complexity, and has high resolution and low computational complexity.
[0017] (2) By using the particle filter to construct the state transfer function of the corresponding state variable, and calculating the a priori particle set of the corresponding state variable, the particle weight set is calculated based on the a priori particle set and the phase sequence, and finally the a priori particle set is resampled according to the particle weight set to obtain the posterior particle set of the state variable, ensuring that the distribution of particles can reflect the posterior probability distribution of the system state. Through continuous iteration, the approximate value of the posterior probability density function of the system state can be obtained, ensuring the final carrier frequency offset value f of the state variable cfo The posterior prediction value is used as the accuracy of the long-term tracking result of the carrier frequency offset, which has the advantages of high estimation accuracy and high robustness.
[0018] (3) The particle filtering method for providing long-term tracking of carrier frequency offset in the present invention has low requirements for equipment configuration and only requires a commercial network card to complete long-term tracking of carrier frequency offset, which is more convenient for actual operation and reduces operating costs.
[0019] Preferably, step S4 comprises the following sub-steps:
[0020] S41. Construct the state vector X through 4 state variables. The specific construction method is as follows:
[0021]
[0022] The initial values are 0, The initial value X[1] of the state vector X is
[0023] S42. Construct the particle set M corresponding to the state variable k , each particle set contains PN particles, and the specific construction is as follows:
[0024]
[0025] In the formula, k represents the state variable, N represents the normal distribution, Indicates the initial value of the state variable corresponding to the state vector, Q k represents the process noise variance of the corresponding state variable;
[0026] S43, for each state variable particle set M k Initialize and obtain the initialization set of all particles corresponding to the state variables, which is the initial value set of the corresponding state variables:
[0027]
[0028] In the formula, represents the initial value of the state variable corresponding to the ith particle and is calculated using the following formula:
[0029]
[0030] In the formula, Indicates obtaining the initial value of the corresponding state variable from the initial value X[1] of the state vector X;
[0031] S44, according to the transfer function F k Calculate the prior particle set of the current t state
[0032]
[0033] In the formula, the parameters is the prior particle corresponding to the state variable t-1, and its initial value is Parameter w k is the process noise.
[0034] Further preferably, in step S44, the transfer function F k for:
[0035]
[0036] In the formula, the transfer function F k The values of subscript k correspond to the calculation formula of state variables, (k ranges from 1 to 4) represents the prior prediction value of the current t state of the corresponding state variable, represents the posterior prediction value of the t-1 state of the corresponding state variable, w k represents the process noise of the corresponding state variable, dt represents the frame interval sequence, and wrapToPi() means wrapping the phase to [-π,π].
[0037] Further preferably, in step S44, the process noise w k The calculation method is:
[0038]
[0039] In the formula, Q k is the process noise variance.
[0040] Preferably, step S5 comprises the following sub-steps:
[0041] S51, according to the phase sequence Θ DC , by observing the function G, the state variable θ is calculated wrap The observation vector Z[t] at state t;
[0042] S52, according to the state variable θ wrap The observation vector Z[t] and the prior particle set The particle weight w calculated by the weight calculation formula i , and construct a particle weight set W(t)={w1,w2,...,w i ,...w PN}.
[0043] Further preferably, the construction process of the observation vector Z[t] in step S51 is as follows:
[0044]
[0045] In the formula, represents the phase sequence under state t, and v is the observation noise random number.
[0046] Further preferably, the particle weight calculation formula in step S52 is as follows:
[0047]
[0048] In the formula, is the state variable θ of the jth particle wrap is the prior particle in state t, and z[t] is the observation vector in state t.
[0049] Preferably, the posterior prediction value in step S7 The calculation formula is as follows:
[0050]
[0051] Among them, the state variable f cfo The posterior predictive value of This is the result of dynamic tracking of the carrier frequency offset.
[0052] Further preferably, the frame interval sequence dt is constructed as follows:
[0053] dt(i)=T(i+1)-T(i)
[0054] Where i represents the i-th particle received, and T is the time of receiving the i-th particle. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] The present invention will be described in more detail below based on embodiments and with reference to the accompanying drawings, wherein:
[0056] Figure 1 is a flow chart of the present invention;
[0057] Figure 2 Schematic diagram of the comparison of the residual sequence of the carrier frequency offset estimation sequence obtained by the constant tracking method and the present method compared with the original phase sequence. DETAILED DESCRIPTION
[0058] The present invention will be further described below with reference to the accompanying drawings. It should be noted that this embodiment can work under 2.4 / 5 / 6GHz frequency bands, bandwidths bw (20, 40, 80, 160, 320MHz), protocol formats (IEEE 802.11a / g / n / ac / ax / be), number of spatial streams s (1-n) and MCS rates supported by all Wi-Fi standards.
[0059] Embodiment 1: The present invention provides a particle filtering method for tracking carrier frequency offset in channel state information. In this embodiment, the transmitting end fc is 6GHz, bw is 160MHz, the transmitting format is the IEEE 802.11ax protocol, and both the transmitting end and the receiving end operate in Wi-Fi standard channel 36 (carrier frequency 5180MHz) with a bandwidth of 20MHz.
[0060] The particle filtering method in this embodiment includes the following steps:
[0061] S1. The transmitter sends several Wi-Fi frames to the receiver. The receiver interpolates the DC subcarrier of the channel state information of each frame and obtains the phase sequence θ containing the DC subcarrier. DC , the phase sequence Θ DC Training frame phase sequence and tracking segment frame phase sequence And construct the frame interval sequence dt;
[0062] Step S1 specifically includes: according to the basic principle of wireless communication, assuming that the frequency of the signal sent by the transmitter is Z fre , the signal frequency received by the receiving end is Y, then the frequency response of the transmission channel can be expressed as H, and the mathematical relationship of the signal is Y = HX fre+N, where the noise term N represents the random interference in the environment. fre When , the channel frequency response H, also known as CSI, can be calculated through the channel estimation algorithm, which is the solution to the mathematical relationship between the three.
[0063] In OFDM-based Wi-Fi protocols (802.11 / a / g / n / ac / ax protocols), CSI can be specifically expressed as a 3D matrix Where N SC Represents the number of subcarriers, N STS To send spatial streams, N Rx is the number of Rx antennas. According to the Wi-Fi protocol standard, the subcarrier numbers are The DC subcarrier is not used; the bandwidth (bw) can be 20 / 40 / 80 / 160 / 320MHz, etc. SC The values are 64, 128, 256, 512, and 1024. For example, in the 802.11n protocol with a bandwidth of 20 MHz, the subcarrier number sequence can be expressed as K = [-28, -27, ..., -1, 1, ..., 27, 28].
[0064] Under the above conditions, the receiving end interpolates the measured CSI with the DC subcarrier. After interpolation, the subcarrier index sequence is expressed as K +DC When using Intel AX210 network card as the receiver, under the IEEE 802.11n protocol, 5.2GHz operating frequency, 40MHz bandwidth conditions, the transmitter uses a single spatial stream to transmit, and the receiver uses dual antennas to receive, then the measured size of H is 114×1×2, the first dimension of H is the subcarrier column, and its subcarrier number index is (-57,-56,…,-1,1,…,56,57), and after DC bit carrier interpolation, 115 subcarriers are obtained (DC is set to 0 in this method), and its subcarrier number index becomes (-57,-56,…,-1,0,1,…,56,57), and the H matrix becomes 115×1×2.
[0065] After interpolation, take The phase θ of any spatial stream in K+DC ,Right now Then, using θ K+DC Take out the phase value θ of the DC carrier DC Since the DC subcarrier is only affected by the carrier frequency offset error, the CSI of the DC subcarrier can be expressed as where f cfo is the carrier frequency offset, and Δt is the unit receiving time interval. After interpolation, the phase sequence of the DC subcarrier contained in each frame of CSI can be obtained. n is the total number of particles received by the receiving end. After the phase sequence is constructed, according to the arrival time T of the i-th particle in the phase sequence i , construct the frame interval sequence dt={(T2-T1),(T3-T2),...,(T n -T n-1 )}, that is, dt(i)=T(i+1)-T(i).
[0066] The phase sequence Θ DC Training frame phase sequence and tracking segment frame phase sequence The division process is:
[0067]
[0068] The ath Wi-Fi frame received by the receiver in chronological order is used as the critical point for division, that is, the ath Wi-Fi frame and the previous Wi-Fi frame segments are training segments, and the Wi-Fi frames after the ath Wi-Fi frame are tracking segments. The value range of a is 30%n <k<70%n。
[0069] S2, the training segment frame phase sequence is analyzed by ESPRIT algorithm Perform spectrum estimation to obtain the carrier frequency offset estimate of the training segment frame The ESPRIT algorithm is a high-precision algorithm for estimating signal frequency and phase. It uses the rotational invariance of signals between different sensors to decompose the signal into two mutually perpendicular subspaces to achieve high-resolution estimation. Its main advantages are high resolution and low computational complexity, and it is suitable for situations with a small number of signals.
[0070] According to the ESPRIT algorithm, the training segment frame phase sequence Divide it into two subspaces, use the rotation invariance property to estimate the signal parameters, then quantize the phase information into a matrix, and perform singular value decomposition on the matrix to obtain the basis vectors of two orthogonal subspaces. By performing eigenvalue decomposition on the two subspaces, we can further obtain the rotation matrix, which can rotate the two subspaces to the same direction; after decomposing the rotation matrix, we can determine the carrier frequency offset estimate of the training segment
[0071] Specifically, the phase sequence of the training segment frame Divide into two overlapping subsequences of equal length Θ m With Θ n , the two subsequences partially overlap, subsequence Θ m and θ n The length of of length and greater than The length of the subsequence is half, and the difference operation is performed on the two subsequences to obtain the difference sequence of the subsequences and The specific difference method is as follows:
[0072]
[0073] The difference sequence and The signal subspace matrix B is formed, the signal subspace is constructed, and the subspace matrix is subjected to singular value decomposition:
[0074] SVD(B)=UV ∧ H
[0075] Where U is the left singular vector matrix, V is the right singular vector matrix, ∧ H is the conjugate transpose. Use the first two columns of V to construct the estimation matrix E. Perform singular value decomposition on E:
[0076] SVD(E)=U s S s V s ∧ H
[0077] Where U S is the left singular vector matrix after decomposition of E. Similarly, V s is the right singular vector matrix, S s is the singular value matrix of E. s Perform eigenvalue decomposition to obtain the generalized eigenvalue, which is the estimated value of the carrier frequency offset in the training segment.
[0078] S3. Construct the state vector X of the particle filter for carrier frequency offset tracking. Particle filtering is a nonlinear filtering method based on Bayesian filtering, which is mainly used for state estimation in dynamic systems. The core idea is to fit the probability density function of the next state of the system through a group of random particles to achieve state estimation. At each moment, according to the prediction function and the observation function, the weight of the particle is updated, and resampling is performed according to the weight of the particle to ensure that the distribution of the particle can reflect the posterior probability distribution of the system state. Through continuous iteration, the approximate value of the posterior probability density function of the system state can be obtained.
[0079] The state vector X includes four state variables, which are the carrier frequency offset change rate f rate , carrier frequency offset value f cfo , the phase accumulation of the unwrapped unwrap And the unwrapped phase accumulation θ wrap, where the value of k represents four state variables in turn, and the value range of k is 1-4, that is, when the value of k is 1, the corresponding state variable is the carrier frequency offset change rate f rate ; When the value of k is 2, the corresponding state variable is the carrier frequency offset value f cfo When the value of k is 3, the corresponding state variable is the phase accumulation of the unwrapped unwrap When the value of k is 4, the corresponding state variable is the unwound phase accumulation θ wrap .
[0080] S4, through the state transfer function F k Calculate the prior particle set of the four state variables at state t respectively k represents the corresponding state variable, and step S4 includes the following sub-steps:
[0081] S41. The state vector X is constructed by four state variables. The specific construction method is as follows:
[0082]
[0083] 0, as well as As state variables f rate 、f cfo ,θ unwrap and θ wrap The initial value of It is represented by the DC subcarrier phase of the first WiFi frame received by the receiver. Represents the observed phase Initialize each state variable in the state vector X and get the initial value X[1] of the state vector X, which is At the same time, the particle set of each state variable obeys the normal distribution with a mean of 0 and a variance of the process noise of each state variable. The state vector in state t (corresponding to the tth data received by the receiving end) is represented as X[t], and the corresponding to They represent f in state t respectively. rate 、f cfo ,θ unwrap and θ wrap The prior prediction value of to They represent the corresponding posterior prediction values respectively.
[0084] S42. Construct the particle set M corresponding to the state variable k , each particle set contains PN particles, and the specific construction is as follows:
[0085]
[0086] In the formula, k represents the state variable, N represents the normal distribution, Indicates the initial value of the state variable corresponding to the state vector, Q k represents the process noise variance of the corresponding state variable.
[0087] S43, for each state variable particle set M k Initialize and obtain the initial value set M of all particles corresponding to the state variables k (1):
[0088]
[0089] In the formula, represents the initial value of the state variable corresponding to the i-th particle, PN represents the total number of particles, and the initial value of the state variable corresponding to each particle is calculated by the following formula:
[0090]
[0091] In the formula, It means to obtain the initial value of the corresponding state variable from the initial value X[1] of the state vector X, where k represents the corresponding state variable, 1 represents the initial state, x represents the value of the corresponding state variable, and RANDOM represents a random number.
[0092] S44, according to the transfer function F k Calculate the prior particle set of the current t state
[0093]
[0094] In the formula, F k Represents the state transition function corresponding to the state variable, parameter is the prior particle corresponding to the state variable t-1, and its initial value is the initial value of the state variable corresponding to the i-th particle i represents the particle number in the prior particle set, and parameter w k is the process noise, and k represents the corresponding state variable.
[0095] State transition function F k The specific construction is as follows:
[0096]
[0097] In the formula, the state transfer function F k The values of the subscript k correspond to the calculation formulas of the state variables, that is, F1 to F4 correspond to to The calculation formula is: (k ranges from 1 to 4) represents the prior prediction value of the current t state of the corresponding state variable, represents the posterior prediction value of the t-1 state of the corresponding state variable, w1 to w4 represent the process noise of the corresponding state variable, dt represents the frame interval sequence, and wrapToPi() means wrapping the phase to between [-π,π].
[0098] Process noise w k (k ranges from 1 to 4) is subject to a mean of 0 and a variance of Q k Normally distributed random numbers, process noise w k The calculation method is as follows:
[0099]
[0100] In the formula, Q k is the process noise variance, which is the control coefficient and the optimal value of fitting in the tracking process. RANDOM represents a random number that obeys the standard normal distribution, and k represents the corresponding state variable.
[0101] The particle filter uses the corresponding state transfer function F for different state variables. k And the posterior particle set of the previous state (t-1 state) The loop iterates and calculates the prior prediction value of each particle in the current state (t state) Finally, the prior particle set that constitutes the state variable
[0102] The specific calculation process is as follows: Calculate the prior particle corresponding to the current t state of the state variable When The value of corresponds to the state transfer function F k The posterior prediction value in the formula The initial value of the corresponding state variable The value as parameter The initial value of the above parameters The initial value and parameter of the process noise w k Substitute the value of into the state transfer function F k , according to the k value into the state transfer function F k Calculate the corresponding calculation formula in, that is, when k = 1, substitute Formula for calculation; when k = 2, substitute Formula for calculation; when k = 3, substitute Formula for calculation; when k = 4, substitute Formula for calculation.
[0103] The calculation result is the prior prediction value is the prior particle corresponding to the current state t of the state variable And it is used as the prior prediction value for the next calculation of the t+1 state The posterior predictive value of By looping and iterating, we can get the prior particle set of the current t state:
[0104] S5. Through the prior particle set With the phase sequence Θ DC Calculate the particle weight set W(t), step S5 includes the following sub-steps:
[0105] S51, according to the actually measured phase sequence Θ DC , by observing the function G, the state variable θ is calculated wrap The observation vector Z[t] at state t.
[0106] Since among the four state variables, only the phase accumulation θ of the state variable that is not unwrap wrap can be achieved through the phase sequence Θ DC Observe, therefore, only the unwrapped phase accumulation θ wrap The observation vector Z[t] at state t is constructed through the observation equation G. The construction process is as follows:
[0107]
[0108] In the formula, represents the measured phase sequence of the DC subcarrier in state t, v represents the observation noise random number that obeys the normal distribution with mean 0 and variance R. The construction process of the observation noise random number v is as follows:
[0109]
[0110] Where R is the noise intensity fitted by the ambient noise.
[0111] S52, particle filter according to the observation vector Z[t] in state t and the prior particle set The state variable θ is calculated by the particle weight calculation formula. wrap The particle weight w i , and construct a particle weight set W(t)={w1,w2,...,w i ,...w PN}, the particle weight calculation formula is as follows:
[0112]
[0113] In the formula is the state variable θ of the jth particle wrapis the prior particle in state t, and z[t] is the observation vector in state t.
[0114] S6. The particle weight set W(t) is used to calculate the prior particle set Resample and construct the posterior particle set of 4 state variables
[0115] Among them, only the state variable θ wrap The weight set W(t) and the prior particle set Constructing the state variable θ by systematic resampling wrap The posterior particle set The posterior particle set of the remaining state variables Equivalent to its prior particle set That is, the a posteriori particle set of the remaining state variables and are equal to the corresponding prior particle sets as well as
[0116] S7. Through the posterior particle set The expectation of is used as the posterior prediction value of the four state variables The carrier frequency offset value f cfo The posterior predictive value of This is the result of dynamic tracking of the carrier frequency offset.
[0117] The posterior particle set of each state variable The expectation of Specific posterior predictive value The calculation formula is as follows:
[0118]
[0119] State variable f cfo The posterior prediction value sequence of That is, it is an estimated value sequence for long-term dynamic tracking of the carrier frequency offset, which has the beneficial effects of high robustness, low computational overhead, and improved accuracy of the estimated value.
[0120] Finally, this method is compared with the constant tracking method, that is, the result of dynamic tracking of carrier frequency offset obtained by this method is compensated and calibrated with the original phase sequence to obtain a first residual sequence, and then the carrier frequency offset is estimated by the constant tracking method (this method is a method commonly used by technicians in this field to track carrier frequency offset). The constant tracking method compensates and calibrates the original phase sequence with a fixed carrier frequency offset estimate to obtain a second residual sequence, and the first residual sequence is compared with the second residual sequence. The comparison result is as follows: Figure 2As shown in the figure, it can be seen that the deviation of the carrier frequency offset estimate obtained by the constant tracking method compared with the original phase sequence is much greater than the deviation of the carrier frequency offset estimate obtained by this method compared with the original phase sequence. Therefore, the accuracy of the carrier frequency offset estimate in this application has been greatly improved.
Claims
1. A particle filtering method for tracking carrier frequency offset in channel state information, characterized in that: The steps include: S1. The transmitter sends several Wi-Fi frames to the receiver. The receiver interpolates the DC subcarrier of the channel state information of each frame and obtains the phase sequence θ containing the DC subcarrier. DC , extract the phase sequence Θ DC The training frame phase sequence And construct the frame interval sequence dt; S2, the training segment frame phase sequence is analyzed by ESPRIT algorithm Perform spectrum estimation to obtain the carrier frequency offset estimate of the training segment frame S3, construct a state vector X of a particle filter for carrier frequency offset tracking, wherein the state vector X includes four state variables, which are the carrier frequency offset change rate f rate , carrier frequency offset value f cfo , the phase accumulation of the unwrapped unwrap And the unwrap phase accumulation θ wrap , where the value of k represents four state variables in turn, and the value range of k is 1 to 4; S4, through the state transfer function F k Calculate the prior particle set of the four state variables at state t respectively Among them, F k represents the state transition function corresponding to the state variable k, and the t state corresponds to the tth data received by the receiving end; S5. Through the prior particle set With the phase sequence Θ DC Calculate the particle weight set W(t); S6. The particle weight set W(t) is used to calculate the prior particle set Resample and construct the posterior particle set of 4 state variables S7. Through the posterior particle set The expectation of is used as the posterior prediction value of the four state variables The state variable carrier frequency offset value f cfo The posterior predictive value of This is the result of dynamic tracking of the carrier frequency offset.
2. A particle filtering method for tracking carrier frequency offset in channel state information according to claim 1, characterized in that: Step S4 includes the following sub-steps: S41. Construct the state vector X through four state variables. The specific construction method is as follows: The initial values are 0, The initial value X[1] of the state vector X is In the formula, is the carrier frequency offset value, It is represented by the DC subcarrier phase of the first WiFi frame received by the receiver. Represents the observed phase Make the unwrapped value; S42. Construct the particle set M corresponding to the state variable k , each particle set contains PN particles, and the specific construction is as follows: In the formula, k represents the state variable, N represents the normal distribution, Indicates the initial value of the state variable corresponding to the state vector, Q k represents the process noise variance of the corresponding state variable; S43, for each state variable particle set M k Initialize and obtain the initialization set of all particles corresponding to the state variables, which is the initial value set of the corresponding state variables: In the formula, represents the initial value of the state variable corresponding to the ith particle and is calculated using the following formula: In the formula, Indicates obtaining the initial value of the corresponding state variable from the initial value X[1] of the state vector X; S44, according to the transfer function F k Calculate the prior particle set of the current t state In the formula, the parameters is the prior particle corresponding to the state variable t-1, and its initial value is Parameter w k is the process noise.
3. A particle filtering method for tracking carrier frequency offset in channel state information according to claim 2, characterized in that: Step S44: Transfer function F k for: In the formula, the transfer function F k The values of subscript k correspond to the calculation formula of state variables, (k ranges from 1 to 4) represents the prior prediction value of the current t state of the corresponding state variable, represents the posterior prediction value of the t-1 state of the corresponding state variable, w k represents the process noise of the corresponding state variable, dt represents the frame interval sequence, and wrapToPi() means wrapping the phase to [-π, π].
4. The particle filtering method for tracking carrier frequency offset in channel state information according to claim 2, characterized in that: In step S44, the process noise w k The calculation method is: In the formula, Q k is the process noise variance.
5. The particle filtering method for tracking carrier frequency offset in channel state information according to claim 1, characterized in that: Step S5 includes the following sub-steps: S51, according to the phase sequence Θ DC , by observing the function G, the state variable θ is calculated wrap The observation vector Z[t] at state t; S52, according to the state variable θ wrap The observation vector Z[t] and the prior particle set The particle weight w calculated by the weight calculation formula i , and construct a particle weight set W(t) = {w1, w2, ..., w i , ...w PN }.
6. A particle filtering method for tracking carrier frequency offset in channel state information according to claim 5, characterized in that: The construction process of the observation vector Z[t] in sub-step S51 is as follows: In the formula, represents the phase sequence under state t, and v is the observation noise random number.
7. A particle filtering method for tracking carrier frequency offset in channel state information according to claim 5, characterized in that: The particle weight calculation formula in sub-step S52 is as follows: In the formula, is the state variable θ of the jth particle wrap The prior particle in state t, z[t] is the observation vector in state t.
8. The particle filtering method for tracking carrier frequency offset in channel state information according to claim 1, characterized in that: The posterior prediction value in step S7 The calculation formula is as follows: in, is the posterior particle of the jth particle in state t corresponding to the state variable k. The value range of j is 1-PN, where PN is the total number of particles and the state variable f cfo The posterior predictive value of This is the result of dynamic tracking of the carrier frequency offset.
9. A particle filtering method for tracking carrier frequency offset in channel state information according to claim 1 or claim 3, characterized in that: The frame interval sequence dt is constructed as follows: dt(i)=T(i+1)-T(i) Where i represents the i-th particle received, and T is the time of receiving the i-th particle.
