WO2007137484A1 - Procédé et dispositif d'estimation de canaux - Google Patents

Procédé et dispositif d'estimation de canaux Download PDF

Info

Publication number
WO2007137484A1
WO2007137484A1 PCT/CN2007/001531 CN2007001531W WO2007137484A1 WO 2007137484 A1 WO2007137484 A1 WO 2007137484A1 CN 2007001531 W CN2007001531 W CN 2007001531W WO 2007137484 A1 WO2007137484 A1 WO 2007137484A1
Authority
WO
WIPO (PCT)
Prior art keywords
channel
channel estimation
forgetting factor
value
pilot
Prior art date
Application number
PCT/CN2007/001531
Other languages
English (en)
French (fr)
Inventor
Yongming Liang
Feng She
Hanwen Luo
Haibin Zhang
Wu Zheng
Xiaoxun Zhao
Wenyi Dai
Original Assignee
Shanghai Jiao Tong University
Sharp Kabushiki Kaisha
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from CN 200610026428 external-priority patent/CN1866945A/zh
Priority claimed from CNB2006100264255A external-priority patent/CN100553166C/zh
Application filed by Shanghai Jiao Tong University, Sharp Kabushiki Kaisha filed Critical Shanghai Jiao Tong University
Publication of WO2007137484A1 publication Critical patent/WO2007137484A1/zh

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00Baseband systems
    • H04L25/02Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/0202Channel estimation
    • H04L25/0224Channel estimation using sounding signals
    • H04L25/0228Channel estimation using sounding signals with direct estimation from sounding signals
    • H04L25/023Channel estimation using sounding signals with direct estimation from sounding signals with extension to other symbols
    • H04L25/0232Channel estimation using sounding signals with direct estimation from sounding signals with extension to other symbols by interpolation between sounding signals
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00Baseband systems
    • H04L25/02Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/0202Channel estimation
    • H04L25/0204Channel estimation of multiple channels
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00Baseband systems
    • H04L25/02Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/0202Channel estimation
    • H04L25/022Channel estimation of frequency response
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00Baseband systems
    • H04L25/02Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/0202Channel estimation
    • H04L25/024Channel estimation channel estimation algorithms
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L27/00Modulated-carrier systems
    • H04L27/26Systems using multi-frequency codes
    • H04L27/2601Multicarrier modulation systems

Definitions

  • the invention relates to a channel estimation method for wireless communication, which is used in the technical field of wireless transmission, in particular to a channel estimation method and device based on variable forgetting factor RLS (recursive least squares) filtering, and a channel estimation based on particle filter Methods and apparatus, and a selective channel estimation method and apparatus.
  • RLS recursive least squares
  • the signal is affected by factors such as channel attenuation, multipath delay spread and Doppler frequency spread.
  • coherent demodulation is usually used to recover the transmitted signal, and coherent demodulation is required.
  • Channel parameter information which can be obtained through channel estimation, so the performance of the channel estimator directly affects system performance and becomes one of the key technologies of the receiver.
  • Orthogonal Frequency Division Multiplexing (OFDM) systems are effective against inter-symbol interference caused by multipath spreading, making it possible to transmit data over harsh wireless fading channels.
  • the channel estimation method based on two-dimensional least mean square error criterion is the best-performing channel estimation method in theoretical analysis, but such methods are not only highly complex, but also need to know some or all channels. Prior information, so the value of engineering applications is not high.
  • the channel estimation method based on the least squares criterion (LS) although the computational complexity is not high, but the performance will decrease rapidly when the signal-to-noise ratio decreases, which is not suitable for the occasion of low-to-low signal-to-noise ratio. Therefore, the study of channel estimation and the search for applicable channel estimation methods are directly related to the performance of receiver detection and decoding, and play an important role in improving the performance of broadband mobile communication systems.
  • the disadvantage is that the estimation accuracy is not very high, especially in the medium and low signal-to-noise ratio environment, which is much smaller than the minimum based on two-dimensional Channel estimation method for square error criteria.
  • the search also found that the Chinese patent application number is CN 200410021864. 8 , the name is: Two adaptive methods for estimating the MIM0-OFDM system.
  • the patent proposes to use adaptive algorithms such as LMS and RLS for channel estimation.
  • LMS and RLS for channel estimation.
  • the function of dynamically tracking channel parameters as the channel changes but in this patent, the pilot structure used is too complex and requires special subcarriers for channel estimation, and the patent gives an application example of the LMS algorithm.
  • the application example of the RLS algorithm is not given, and the modified forgetting factor scheme is not adopted. Therefore, the patent has the disadvantages that the spectral efficiency and the estimation accuracy are not very high and the forgetting factor scheme is not flexible. Therefore, it is necessary to find a flexible channel estimator with computational complexity and LMS algorithm equivalent but much higher
  • the performance of the channel estimator plays a crucial role in the performance of the broadband mobile communication system. Since the wireless channel is a time-varying channel in the frequency domain and the time domain and is nonlinear, mathematical modeling and quantitative analysis are difficult. Linear signal processing methods are often used to approximate the nonlinear time-varying wireless channel. . However, in many problems that require real-time estimation of dynamic systems, the nonlinearity of the system often becomes an important factor that plagues the optimal estimation. Recursive filtering algorithms are often used to solve such problems due to real-time processing and computational memory requirements, including extended Kalman filter-wave (EKF), modified gain extended Kalman filter (MGEKF), and so on.
  • EKF extended Kalman filter-wave
  • MGEKF modified gain extended Kalman filter
  • the object of the present invention is to overcome the deficiencies in the prior art, to provide a channel estimation method and device based on variable forgetting factor RLS filtering, and a particle filtering based channel estimation method and device, so that the channel statistical information is not known In this case, it has the characteristics of robust performance, robustness, anti-additive white Gaussian noise and non-Gaussian noise, and is easy to implement.
  • the wireless channel environment is very complicated, the wireless channel may be a linear or non-linear channel, and the environmental noise may also be additive white Gaussian noise or non-Gaussian non-stationary noise, and the mobile terminal may be at different times. Different channel environments, therefore, the present invention proposes a selective channel estimation method and device, which can adopt different channel estimation methods and devices according to different channel environments, thereby improving the anti-interference ability and performance of the communication system.
  • a channel estimation method based on a variable forgetting factor recursive least squares filtering including: an initial estimating step, initial estimating a channel according to the provided pilot information, and obtaining all pilots The initial value of the channel; the inverse Fourier transform step, performing an inverse Fourier transform on the estimated channel initial value; and the tracking and estimation step, using the recursive least squares filter based on the variable forgetting factor to track and estimate the time-varying wireless channel;
  • the transform and interpolation steps obtain the channel state information value corresponding to the data by using a Fourier transform and an interpolation algorithm, wherein the forgetting factor of the pilot symbol at the latter time is greater than or equal to the forgetting factor of the pilot symbol at the previous moment.
  • a channel estimation apparatus based on a variable forgetting factor recursive least squares filtering
  • the initial value of the channel; the inverse Fourier transform unit, used to estimate The calculated channel initial value is inverse Fourier transform;
  • the recursive least square filter has a variable forgetting factor for tracking and estimating the time-varying wireless channel, wherein the forgetting factor of the pilot symbol at the latter time is greater than or equal to the previous one a forgetting factor of the time pilot symbol;
  • a Fourier transform unit for performing a Fourier transform on the output of the recursive least squares filter; and an interpolation operation unit for performing an interpolation operation on the output of the Fourier transform unit to obtain a channel state corresponding to the data Information value.
  • a channel estimation method and device based on variable forgetting factor RLS filtering in a time/delay domain is designed based on a traditional frequency domain least square channel estimator.
  • the tracking and estimation of the time-varying channel obtains the channel state information under the action of noise, thereby obtaining a more realistic channel estimation value.
  • the channel estimation method and device of the present invention can track and estimate the channel well, and has the characteristics of good convergence and high estimation precision, and can adjust the channel length and other parameters to the channel.
  • the estimation accuracy and computational complexity are compromised and balanced to obtain better estimation performance.
  • the channel estimation method and device are convenient and flexible, and are suitable for practical applications, and can be third generation (3G) and super third generation (B3G).
  • 4G fourth-generation cellular mobile communications and channel estimation schemes for systems such as digital television, wireless local area network (WLAN), and wireless wide area network (WWAN) provide important theoretical basis and specific implementation methods and equipment.
  • a channel estimation method based on particle filtering which includes: an initial estimation step, using pilot information for initial channel estimation; and a particle filtering step, using particle filtering on the basis of initial channel values.
  • Accurate channel estimation wherein the particle filtering step comprises the following sub-steps: a posteriori probability density calculation step, the conditional probability density of the received signal conditional on the channel particle value is used as a prior probability, and the Bayesian model is constructed with the current received signal Calculating the posterior face probability density of the channel particle values subject to the current received signal; normalization and weighting steps, normalizing and weighting the weights of all the particles to obtain an accurate channel estimation value.
  • a particle filter-based channel estimation apparatus comprising: an initial estimation unit, configured to perform initial channel estimation by using pilot information; and a particle initialization setting unit, configured to set a particle range and a number of particles And assigning an initial weight to each particle; an important sampling unit for using the conditional probability density of the received signal as a condition of the channel particle value as the first probability, constructing a Bayesian model with the current received signal, and calculating the current received signal a posteriori probability density of the conditional channel particle values; and normalizing the weights of all the particles and performing a weighting operation; a resampling unit for resampling the particles according to the judgment condition, wherein the important sampling unit Performing a time recursive operation with the resampling unit to obtain an accurate channel estimate for transmitting all pilots.
  • a particle filtering based channel estimation method and apparatus which overcomes the deficiencies in the prior art, so that it has stable estimation performance in an environment where channel statistics information is unknown. Strong robustness and anti-noise ability, and easy to implement.
  • the present invention is more in line with the characteristics of the actual channel unknown, and does not need to know any channel statistical information, and proposes a better weight. Sampling algorithm.
  • the invention Compared with the recursive filtering algorithm (EKF, MGEKF, etc.) for solving such problems, the invention has the characteristics of high estimation precision, short convergence time and stable performance.
  • the present invention reduces the conditional probability problem based on Bayesian principle in a two-dimensional or even multi-dimensional space into a solution problem of conditional probability in a one-dimensional space through antenna decoupling and real and imaginary partial solutions of pilot information.
  • the resampling algorithm based on the probability distribution function value greatly reduces the computational complexity of the particle filter algorithm.
  • the invention adopts a flexible noise estimation algorithm, and has the characteristics of strong anti-noise ability. Therefore, with the decrease of computational storage cost, the channel estimation method based on particle filter is getting more and more attention.
  • the particle filter-based channel estimation method proposed by the present invention can be realized by hardware, and has certain engineering application value.
  • a selective channel estimation method including: determining a channel attribute, determining a channel of a non-linear channel or a non-Gaussian noise as a bad channel environment, and determining a channel of linear and Gaussian white noise as Non-bad channel environment; for the bad channel environment, using the particle filter based channel estimation method according to the third aspect of the present invention, performing channel estimation; for the non-bad channel environment, adopting the basis according to the first aspect of the present invention
  • a selective channel estimation apparatus including: a selective channel estimation apparatus, comprising: a channel attribute determining unit, configured to determine a channel of a non-linear channel or a non-Gaussian noise as a bad a channel environment, and a channel that linearizes and whitens white noise is determined to be a non-bad channel environment; a particle filter-based channel estimation apparatus according to a fourth aspect of the present invention, configured to perform channel estimation on a bad channel environment; The channel estimation device based on the variable forgetting factor recursive least squares filtering is used for channel estimation in a non-bad channel environment, and the interpolation and signal detecting unit is configured to perform interpolation and signal detection based on the result of the channel estimation; Bit error rate a determining unit, configured to calculate a bit error rate of the detected signal, and determine whether the bit error rate satisfies a predetermined performance requirement, wherein if the bit error rate does not satisfy the predetermined performance requirement, notifying
  • a selective channel estimation method and apparatus for a wireless channel in a complex channel environment, a linear channel, a nonlinear channel, a channel of Gaussian white noise, and a channel of non-Gaussian noise The mobility of the mobile terminal leads to changes in the channel environment.
  • the channel can be distinguished into a bad channel environment and a non-weak channel environment.
  • the particle filter based estimation method and device can be applied to the bad channel and non-bad channel environment with estimation accuracy.
  • the high characteristics also make use of the variable forgetting factor RLS filtering estimation method and equipment with the characteristics of simple estimation method and high estimation accuracy, and adopt the channel estimation method based on particle filtering or variable forgetting factor RLS filtering respectively.
  • the selective channel estimation method proposed by the present invention may adopt an appropriate channel estimation method according to changes in the wireless channel environment and changes in the location of the mobile terminal, and different channel estimation methods may pass software through the idea and method of software radio. Or hardware modules to achieve, with high engineering application value.
  • FIG. 1 is a schematic diagram of MIM0-OFDM according to a first preferred embodiment of the present invention
  • FIG. 2 is a diagram showing a pilot structure according to a first preferred embodiment of the present invention.
  • FIG. 3 is a schematic diagram of a channel estimation algorithm according to a first preferred embodiment of the present invention
  • FIG. 4 is a schematic diagram of an RLS filtering algorithm according to a first preferred embodiment of the present invention.
  • Figure 5 is a simulation diagram of MSE-M simulation according to a first preferred embodiment of the present invention.
  • Figure 6 is a simulation diagram of MSB-L simulation according to a first preferred embodiment of the present invention.
  • FIG. 7 is a graph showing MSE-SNR simulation performance according to a first preferred embodiment of the present invention.
  • Figure 8 is a BER-SNR simulation performance diagram in accordance with a first preferred embodiment of the present invention.
  • FIG. 9 is a BER-Doppler simulation performance map according to a first preferred embodiment of the present invention.
  • FIG. 10 is a comparison performance diagram of a forgetting factor scheme according to a first preferred embodiment of the present invention.
  • FIG. 11 is a single preferred embodiment according to the present invention.
  • Antenna MSE-SNR Simulation Performance Diagram FIG. 12 is a single antenna BER-SNR simulation performance diagram according to a first preferred embodiment of the present invention.
  • Figure 13 is a diagram showing the use of a variable forgetting factor scheme B according to a first preferred embodiment of the present invention.
  • Figure 14 is a diagram showing the use of a variable forgetting factor scheme B according to a first preferred embodiment of the present invention.
  • Figure 15 is a schematic block diagram of a channel estimation apparatus according to a first preferred embodiment of the present invention.
  • Figure 16 is a diagram showing a channel estimation method according to a second preferred embodiment of the present invention.
  • Figure 17 is a flow chart of a sequential important sampling method in accordance with a second preferred embodiment of the present invention.
  • Figure 18 is a schematic diagram of a sequential important sampling method in accordance with a second preferred embodiment of the present invention.
  • FIG. 19 is a schematic diagram of a probability distribution based resampling algorithm according to a second preferred embodiment of the present invention.
  • FIG. 20 is a MSE simulation performance diagram according to a second preferred embodiment of the present invention.
  • Figure 21 is a BER simulation performance diagram according to a second preferred embodiment of the present invention.
  • Figure 22 is a schematic block diagram of a channel estimation apparatus according to a second preferred embodiment of the present invention.
  • Figure 23 is a diagram showing a channel estimation method according to a third preferred embodiment of the present invention.
  • Figure 24 is a block diagram showing the principle of a channel estimating apparatus according to a third preferred embodiment of the present invention.
  • the channel estimator uses the pilot information to obtain the initial value of the channel state information at the pilot in the frequency domain by a least squares (LS) algorithm.
  • LS least squares
  • the pilot information of the present invention uses a block-shaped pilot symbol for the structure, each pilot symbol includes a time-frequency block on all sub-carriers, and the pilot frame uses a frame with several pilot symbols added in front of one frame of data.
  • the pilots of different transmit antennas at the same time in a multi-antenna system are mutually orthogonal. Since the wireless channel is often unknown and does not know any prior statistics, pilot symbols can be placed in front of each frame of data for estimation of the channel's least squares criterion in the frequency domain without any known Channel statistics.
  • the patent application number CN200410021864. 8 first uses block pilots and then uses subcarriers as pilots. This pilot scheme is too complicated to design and consumes more spectrum resources.
  • the pilot mode is actually an improved comb pilot mode, which is more suitable for fast fading time selective channels, but not for slow fading frequency selective channels such as wireless local area networks. Therefore, the pilot mode of the present invention performs better on a slow fading frequency selective channel than the prior application number.
  • the receiver of the OFDM system uses the LS channel estimation method according to the pilot information provided by the transmitting end to find the initial estimated value of the channel at each pilot in the frequency domain, which is ideal for not considering the noise effect.
  • Channel estimation method uses the LS channel estimation method according to the pilot information provided by the transmitting end to find the initial estimated value of the channel at each pilot in the frequency domain, which is ideal for not considering the noise effect.
  • the channel state information in the time/delay domain without considering noise is obtained by Fast Inverse Fourier Transform (IFFT). That is, the receiver performs a fast inverse Fourier transform on the initial value of the channel estimation estimated by the LS algorithm, and obtains a channel estimation value in the time/delay domain.
  • IFFT Fast Inverse Fourier Transform
  • the channel state information under the action of noise is taken as the expected value at the current moment, and the channel state information at the current time and the first few moments is taken as the input value, and then the RLS filter is used to track and estimate the accuracy of the current moment considering the noise.
  • Channel status information is taken as the expected value at the current moment, and the channel state information at the current time and the first few moments is taken as the input value, and then the RLS filter is used to track and estimate the accuracy of the current moment considering the noise.
  • the channel estimation value of the time/delay domain obtained by the LS algorithm and then by the IFFT transform is obtained by the RLS filtering algorithm to obtain a more accurate channel estimation value in the time I delay domain in the case of noise, in the process of using the RLS algorithm.
  • the channel expectation value is the channel estimation value obtained by the current time LS algorithm through the IFFT.
  • the current channel value is the channel estimation value obtained by the LS algorithm of the current time and the first few moments by the IFFT, and the recursive relationship of the RLS algorithm can be used to find the current
  • the accurate channel estimation value under the action of noise is considered at the moment, so as to realize real-time tracking and estimation of the time-varying wireless channel. Note that the values of the first few moments of the channel are determined by the length of the filter.
  • variable forgetting factor is used, and then the recursive operation is performed in the time/delay domain to obtain accurate channel state information considering the noise effect.
  • the variable forgetting factor scheme uses different pilot symbols.
  • the pilot symbols at different time points in the time domain have different weight coefficients in one frame, and the closer the weight symbols of the pilot symbols are, the larger the weight coefficients are.
  • This setting is actually an optimized weight coefficient scheme.
  • the weight coefficients of different pilots are set to different values to reverse Different roles of different pilot symbols on data channel information. Considering that the channel transfer functions of different pilots have different effects on the data channel transfer function, the pilot channel transfer function pairs the data as the pilot channel transfer function and the data channel transfer function are closer in time. The channel transfer function of the symbol plays an increasingly larger role, so it is reasonable to set the weight coefficient to be different.
  • the forgetting factor in the prior art uses the conventional method, which gives each pilot symbol the same forgetting factor value.
  • the forgetting factor of the present invention is changed, and the forgetting factor is set according to channel characteristics, algorithm requirements, simulation results, and delay factors, so that the forgetting factor of the pilot symbol at the latter time is greater than or equal to the forgetting factor of the pilot symbol at the previous moment. .
  • the forgetting factor of the present invention can be changed in two steps or adaptively.
  • the two-step method is to set a small forgetting factor value for the first few pilots, so that the communication system relies more on the correct channel state information estimated by the current pilot symbols; the latter pilot symbols are set high.
  • the forgetting factor value the purpose is to make the communication system rely more on the statistics of the channel provided by the previous channel estimation.
  • the adaptive forgetting factor method sets different forgetting factors for all pilot symbols: first, set an initial value for the first pilot symbol, and then set a learning rate for other pilot symbols, so that all the different pilot symbols are The respective forgetting factor values are continually updated at the learning rate so that different pilot symbols have different forgetting factors.
  • the settings of the initial forgetting factor value and the learning rate are determined by factors such as channel characteristics, algorithm requirements, human experience, computer simulation, and the like.
  • the fixed forgetting factor value given by the conventional method can be used as the initial value of the variable forgetting factor, and the learning rate can be flexibly arranged according to the number of pilot symbols.
  • the learning rate can be set to be fixed or variable, but it is guaranteed that the forgetting factor value of the pilot symbol at the last moment in a frame cannot exceed one.
  • the forgetting factor value and the learning rate can be constantly changing with time n.
  • the changing learning rate can use a linear function, an exponential function or other mapping relationship with independent variables as time.
  • the above steps obtain the exact value of the channel estimation in the time/delay domain, but the signal detection requires the state information of the time/frequency domain of the channel. Therefore, the fastest Fourier transform (FFT) is performed on the channel estimation value in the time/delay domain. Transform to get the time/frequency domain channel estimate.
  • FFT Fast Fourier transform
  • the structure of the MIMO-OFDM system is as shown in FIG. 1.
  • the present invention adopts the MIM0-0FDM system with 4 transmissions and 4 receptions, adopts the modulation method of QPSK, and the space-time coding scheme samples the orthogonal space-time block code (0-STBC), the code rate. 1 /2, space-time decoding sampling maximum likelihood (ML) decoding scheme, carrier frequency 4 GHz, bandwidth 6 M, number of subcarriers is 64, considering that the channel estimation method of the present invention is more suitable for frequency selective slow fading channels Therefore, the channel model adopts the multipath Rayleigh fading channel proposed by ITU, the number of channels multipath is 5, the delay is [0 260 520 780 1040] ns, and the power is [- 1. 78 0 -7. 47 -10 -12 . 62] dB , speed of movement 3ktn/hdonitis
  • the design of the pilot is shown in Figure 2.
  • the pilot sampling is orthogonalized.
  • the pilots transmitted by different transmitting antennas are orthogonal to each other.
  • the block orthogonal orthogonal pilot structure is adopted in the embodiment of the present invention, the purpose is to make the present invention more suitable for the slow fading frequency selective channel, but this does not affect the generality of the present invention, and increases the complexity of the calculation.
  • Sex to increase the occupancy of spectrum resources, other pilot structures, such as comb pilots, discrete pilots, etc., can also be used.
  • the LS-based channel estimation method is shown in FIG. 3.
  • Four receivers use the least-squared method (LS) in the frequency domain to find each pilot at the pilot frequency according to the orthogonalized pilot information provided by the four transmitting antennas.
  • the initial estimate of the channel which is actually an idealized channel estimation method that does not consider the effects of noise.
  • each receiver performs a fast inverse Fourier transform (IFFT) on the initial value of the channel estimation estimated by the respective LS algorithm to obtain a channel estimation value in the time/delay domain.
  • IFFT fast inverse Fourier transform
  • L L CP +1
  • L eP 15, so L is 16.
  • the RLS filtering algorithm is shown in Figure 4.
  • the channel estimation value in the time/delay domain is obtained by the RLS filtering algorithm to obtain more accurate channel estimation values in the time/delay domain in the case of noise.
  • the channel expectation value is the channel estimation value obtained by the current time LS algorithm through IFFT, input
  • the value is the channel estimation value obtained by the LS algorithm through the IFFT at the current and the first few moments.
  • the length M of the RLS filter and the maximum length L of the multipath will affect the accuracy and computational complexity of the channel estimation.
  • the RLS filtering algorithm can be replaced by the LMS filtering algorithm.
  • the computational complexity of the LMS algorithm is smaller than the RLS algorithm, the computational complexity of the LMS is [M(L ⁇ D], and the RLS computational complexity is y W ( ⁇ +D ]
  • the general sampling value is very small. Considering that the convergence speed and estimation accuracy of the LMS are far less than the RLS algorithm, the RLS algorithm is more applicable in the current era of decreasing computational cost.
  • Fig. 5 The effect of filter length on system performance is shown in Fig. 5.
  • the optimal value of the present invention is theoretically calculated by the Wiener filtering principle. Firstly, the cost function of the channel estimation error is constructed, and then the Wiener-Hop (Hener- Hopf) equation is generated by the orthogonal principle. Then, the optimal filter coefficient under the minimum mean square error ( ⁇ SE) criterion is obtained, so that the length value of the filter can be obtained.
  • the optimal length of the filter is 10 according to the above principle, but at this time The computational complexity is too high, so the filter length is generally not very large.
  • the maximum multipath extension is generally less than the cyclic prefix (CP) length.
  • the actual channel maximum extension is much smaller than the cyclic prefix length.
  • the maximum multipath extension is The characteristics of the channel itself cannot be artificially adjusted.
  • scheme ⁇ uses 6 pilot symbols with different weights
  • the constant, this setting is actually an optimized weight coefficient scheme, ⁇ 2 , cr 3 , ⁇ 4 , ⁇ 5 , are set to different values to reflect the different effects of different pilot symbols.
  • the forgetting factor of scheme B uses a two-step method or an adaptive method: the two-step method gives the first two pilots a smaller forgetting factor and gives the last four pilots a larger forgetting factor, as shown in Fig. 2 In the invention, the smaller forgetting factor is 0.
  • the larger forgetting factor is 0.9, and the first two pilot symbols are set with a low forgetting factor value, so that the communication system relies more on the current pilot symbol estimation.
  • Correct channel state information setting a high forgetting factor value for the last four pilot symbols, so that the communication system relies more on the statistical information of the channel provided by the previous channel estimation; adaptive forgetting factor method, for all pilots
  • the symbols set different forgetting factors, first setting an initial value for the first pilot symbol, and then setting a learning rate for other pilot symbols, so that all different pilot symbols have a different forgetting factor.
  • the initial forgetting factor value and the learning rate are determined by factors such as channel characteristics, algorithm requirements, computer simulation, etc.
  • the forgetting factor value and the learning rate can be continuously selected and adjusted according to the final computer simulation result.
  • the forgetting factor value and the learning rate can be changed continuously with time.
  • the learning rate can use a linear function, an exponential function or other mapping relationship with independent variables as time.
  • the present invention gives three sets of adaptive forgetting factor setting schemes based on experience:
  • the initial value in a slow fading frequency selective channel such as a wireless local area network, the initial value generally takes the value of the conventional forgetting factor, and the computational complexity based on the adaptive forgetting factor channel estimation algorithm is basically consistent with the conventional forgetting factor algorithm.
  • the fixed learning rate is set to 0.01.
  • the initial value is taken as 0.9.
  • the forgetting factor value of the sixth pilot is less than 1 and the value is 0.95, if the pilot symbol is 10 ⁇
  • the learning rate can be taken to be less than 0.01 value of 0. 009, such that the last pilot forgetting factor is not more than 1 value of 0.98.
  • the algorithm requires that the channel change can be quickly tracked, and the pilot pattern can be combed or discrete accordingly, and the initial value can take the value of the conventional forgetting factor to ensure self-based
  • the computational complexity of the channel estimation algorithm adapted to the forgetting factor is basically the same as that based on the conventional forgetting factor algorithm.
  • the learning rate of the change is set.
  • the value of the forgetting factor of the sixth pilot is less than 1.
  • the value of the forgetting factor of the sixth pilot is less than 1. If the pilot symbol is 8, then the learning rate is 0.01 (1 + 0.1)".
  • the value of the learning factor can be less than 0.01. If the pilot symbol is 10, the learning rate factor can be less than 0.01. 0. 0046, this last The forgotten factor of a pilot is no more than 1 value of 0.976. Thus, the pilot forgetting factor at different times is the rate of exponential change, which is greater than the linear rate of change in the scheme 1.
  • the algorithm requires that the channel change can be quickly tracked, and the pilot pattern can be combed or discrete accordingly, and the initial value can be taken as the value of the conventional forgetting factor to ensure
  • the computational complexity of the adaptive forgetting factor channel estimation algorithm is basically the same as that based on the conventional forgetting factor algorithm.
  • the forgotten factor of the last pilot is still no more than a value of 1.99. If the pilot symbol is 10, the forgotten of the last pilot is still a value of 0.99 which is still no more than 1.
  • the pilot forgetting factor at different times is the exponential rate of change, which is greater than the linear rate of change in the scheme 1, and the forgetting factor value of the pilot symbol at one time is always 0.99.
  • the algorithm requires that the channel change can be quickly tracked, and the pilot pattern can be combed or discrete accordingly, and the initial value can take the value of the conventional forgetting factor to ensure self-based
  • the computational complexity of the adaptive forgetting factor channel estimation algorithm is basically the same as that based on the conventional forgetting factor algorithm.
  • the initial value is taken as 0.9, so that the forgotten factor value of the sixth pilot is less than 1 and the value is 0.91.
  • the learning rate factor may be less than the value of 0.01. 0. 007, such that the forgetting factor of the last pilot is a value not exceeding 1 0. 991.
  • the learning rate factor may be less than the value of 0.01. 0. 0046, such that the forgetting factor of the last pilot is a value not exceeding 1 0. 978.
  • the rate of change of the pilot forgetting factor at different times is greater than the rate of change in the scheme 1.
  • the simulation result of the present invention adopts the adaptive forgetting factor method of the scheme 1, the initial forgetting factor value is set to 0.9, and the learning rate is set to 0.01.
  • the performance of various forgetting factor schemes is shown in Figure 7, Figure 8, and Figure 9, Scheme A (RLS-C, RLS-T', RLS-A, ) and Scheme B (RLS-C, RLS-T, RLS- The performance comparison of A) is shown in Figure 10.
  • the above steps determine the exact value of the channel estimate in the time/delay domain, but the signal What is needed for the detection is the state information of the time/frequency domain of the channel. Therefore, the channel estimation value of the time I delay domain is subjected to IFFT transformation to obtain a time/frequency domain channel estimation value.
  • the above steps obtain the channel estimation value at the time/frequency domain pilot, and then obtain the channel state information of the time/frequency domain at the data symbol by the interpolation algorithm, thereby providing the signal detection module for further processing, and the channel state information of the data is obtained by the present invention.
  • the space-time decoding module is sent to perform space-time decoding of the maximum likelihood criterion.
  • the channel estimation method of the present invention has certain versatility and can be adapted to a general OFDM communication system, and a single-antenna OFDM system using the same pilot pattern has a frequency selective slow fading signal such as an ITU channel, that is, a channel multipath number is 2,
  • the delay is [0 260] ns
  • the power is [0 -12. 77] dB
  • the MSE simulation performance of various channel estimation algorithms is shown in Figure 11
  • the BER performance is shown in Figure 12
  • the adaptive forgetting factor scheme is still adopted.
  • FIG. 14 is a schematic block diagram of a channel estimation apparatus according to a first preferred embodiment of the present invention.
  • the channel estimation apparatus based on the variable forgetting factor RLS filtering includes: an initial estimation unit (LS) for The provided pilot information initially estimates the channel to obtain an initial value of the channel at all pilots; an inverse fast Fourier transform unit (IFFT) is used to perform inverse fast Fourier transform on the estimated initial value; RLS filter, The variable forgetting factor is used for tracking and estimating the time-varying wireless channel, wherein the variable forgetting factor is set by the forgetting factor setting unit, so that the forgetting factor of the pilot symbol at the latter time is greater than or equal to the previous moment a forgetting factor of a pilot symbol; a fast Fourier transform unit (FFT) for performing a fast Fourier transform on the output of the recursive least squares filter; and an interpolation operation unit for performing an interpolation operation on the output of the Fourier transform unit to obtain a data Corresponding channel state information value.
  • LS initial estimation unit
  • IFFT inverse fast Fourier transform unit
  • RLS filter The variable forgetting factor is used for tracking and estimating
  • the signal received by the receiver is sent to the initial estimation unit (LS) through a fast Fourier transform (FFT); the initial estimation unit performs LS-based on the channel according to the provided pilot information.
  • Initial channel estimation obtaining an initial value of a channel at all pilots; initial value of the channel is obtained by an inverse fast Fourier transform unit (IFFT) to obtain initial channel state information in the time domain; and initial channel state information in the time domain is sent
  • IFFT inverse fast Fourier transform unit
  • the RLS filter of the forgetting factor is processed to obtain accurate time domain channel state information; the accurate time domain channel state information is sent to the fast Fourier transform unit (FFT) to obtain accurate channel state information in the frequency domain;
  • the accurate channel state information of the domain is sent to the interpolation unit for interpolation, and channel state information of the frequency domain corresponding to the data symbol is obtained; the channel state information of the frequency domain corresponding to the data symbol is sent to the signal detecting unit for detection, thereby being at the receiver
  • the terminal recovers the data information sent
  • a channel estimation method and device based on variable forgetting factor RLS filtering is designed based on a traditional frequency domain least square channel estimator. The tracking and estimation of the variable channel obtains the channel state information under the action of noise, thereby obtaining a more realistic channel estimation value.
  • the channel estimation method and device can track and estimate the channel well compared with the conventional channel estimation method and device, and has the characteristics of good convergence, high estimation accuracy, and the like, and can adjust the filter at the same time. Parameters such as length are used to compromise and balance the channel estimation accuracy and computational complexity to obtain better estimation performance. Therefore, the channel estimation method and device are convenient and flexible, and are suitable for practical applications, and can be used for the third generation (3G). ), super third generation (B3G), fourth generation (4G) cellular mobile communications and digital television, wireless local area network (WLAN), wireless wide area network (WWAN) and other systems channel estimation scheme provides important theoretical basis and specific implementation methods and equipment .
  • 3G third generation
  • B3G super third generation
  • 4G fourth generation
  • WLAN wireless local area network
  • WWAN wireless wide area network
  • the communication system used is a single antenna system or a multiple antenna system. If it is a single antenna system, there is no special requirement for the pilot setting. If it is a multi-antenna system, it is composed of different antennas transmitted at the transmitting end and at different times.
  • the pilot matrix must be full rank, so these pilots can be orthogonal or non-orthogonal, so that the pilot symbols of different transmit antennas can be easily decoupled and separated at the receiving antenna.
  • the invention adopts a one-dimensional space particle filter, and the signal processing methods used are based on the Bayesian principle in one-dimensional space and the sequential Monte Carlo method in one-dimensional space, therefore,
  • the invention is directed to channel information which is complex information including amplitude and phase.
  • the real and imaginary parts of the complex are separated, and the real and imaginary parts are respectively tracked and estimated based on the particle filter method, and before the end of the whole process,
  • the real and imaginary information of the channel is combined into complete channel information.
  • the method of processing the real and imaginary parts of the channel in one-dimensional space avoids the problem of solving the joint probability density problem under the Bayesian principle with extremely complicated principle and huge computational complexity, thus greatly reducing the analysis complexity and operation. the amount.
  • the initial channel estimation value based on the least squares criterion (LS) is obtained, and the channel information in the complex form obtained by the LS estimation is decomposed into channel information in the real part form and the imaginary part form.
  • the initial channel estimation value corresponding to each pilot symbol is obtained in the frequency domain; the multi-antenna system is also relatively simple, that is, the pilot is obtained under the premise of decoupling the antenna in the first step.
  • the initial channel value can be obtained by inverse Fourier transform (IFFT) to obtain time domain channel information, and the time domain channel estimation value is obtained by using a minimum mean square error algorithm (LMS) or a recursive least squares algorithm (RLS) in the time domain.
  • LMS minimum mean square error algorithm
  • RLS recursive least squares algorithm
  • the channel estimation value is then obtained by the Fourier transform (FFT) to obtain the channel value in the frequency domain, and the accuracy of the channel information is higher than the initial channel value obtained by the LS algorithm.
  • the LS method is adopted in the embodiment of the present invention instead of the more complicated LMS or RLS method.
  • the initial setting includes extracting random samples, that is, setting the number of particles, particle range, and individual particles. Corresponding weight;
  • the initial setup is to extract a series of samples (particles), including setting the number of particles, the particle range, and the weight corresponding to each particle. 1
  • Each time-frequency block that requires channel estimation requires a particle with a weight of 1 to sample.
  • the initial value setting of the channel particles of the important sampling method of the present invention is not In the same manner as the setting method in which the known channel is the AR model, in the setting method in which the known channel is the AR model, the initial value of the channel particle at the latter time is the converted value of the AR model that is the initial value of the channel particle at the previous time. Therefore, the initial value of the channel particle value of the present invention is based on the LS algorithm, which does not require any known channel statistics.
  • the core idea of the sequential important sampling method is to weight the prior probability represented by a series of random samples (particles) and the current measured value of the channel according to the Bayesian principle, and obtain the channel estimation value under the first probability.
  • the current measurement value of the channel is the initial channel estimation value of the second step estimation process
  • the obtained channel estimation value at the posterior probability is the accurate channel estimation value required by the present invention.
  • the more the number of random samples (particles) the closer the function of Monte Carlo and the function of posterior probability density are, and the closer the performance of sequential important sampling is to the optimal Bayesian estimation.
  • the number of particles is too large, the computational complexity will increase, and the number of particles will increase to a certain extent, which is close to the optimal value.
  • the reasonable number of particles is generally obtained by comparing computer simulation results and computational complexity.
  • HO represents the channel
  • ⁇ / represents the received signal
  • , ⁇ ;] represents noise
  • represents the transmission power of the N rth transmitting antenna
  • the LS estimation method is used to obtain the expression ⁇ , so To derive the noise variance ⁇ ⁇ 2
  • »] -.
  • y to ) Estimation can be expressed as P0 I y 0 ) ' complicat
  • the present invention utilizes sequential important sampling, using The particle and their corresponding weight values indicate that the required posterior probability density function is an important idea, namely (3 ⁇ 4) /'), the specific table
  • the expression is wX, ⁇ ,; ⁇ ( ⁇ , ⁇ , ,;,) represents the prior knowledge of the current system.
  • the phase is set to a value greater than the theoretical value, which is set to a value that facilitates accurate tracking after the initial tracking of the channel.
  • the noise variance is set to a value larger than the theoretical value in the slow time varying channel, and is set to a value smaller than the theoretical value in the fast time varying channel. Since the initial value setting of the channel particle value in the present invention is obtained by sampling the rough estimated value of the channel obtained by the LS algorithm, the present invention can sequentially obtain the observed value and perform online reasoning, so when the received data arrives, it is required By updating the posterior distribution, the present invention employs a sequential method that does not require the storage of all data, thereby binarizing the calculations.
  • the probability axis of the distribution is equally divided, and the particles are distributed on the re-segmented distribution axis, and the particles extended to the extension are introduced when the particles are distributed, that is, the artificially expanded particles are appropriately extracted on the basis of the sampling of the resampling algorithm. Range of values to accommodate changes in the channel.
  • the particle filter obtains the weight of the channel particle
  • the initial value of the initial channel particle is used for weighting operation
  • the channel estimation values of the real part and the imaginary part are respectively obtained.
  • the mathematical expectation value is obtained for all channel particle values and their probability density (weight)
  • the channel estimation values of the real part and the imaginary part of the current time are obtained, and the real part and the imaginary part are combined to obtain the most
  • the weighted operation is performed by using the resampled channel particle value and the weight, that is, the mathematical expectation value of the channel particle value is obtained, and then the real part and the imaginary part are combined to obtain a complex number.
  • step 5 Since particle filtering is a time recursive signal processing method, pilots at all times The exact channel estimate corresponding to the symbol should be obtained through constant iterative operations, so returning to step 5 is to prepare for the next channel estimation.
  • the channel estimation values in the form of the real part and the imaginary part are combined, so that the channel estimation value is a plural form capable of reflecting the amplitude and phase information;
  • the channel information value of the data symbol is obtained by interpolation, thereby providing more accurate channel information for data detection or decoding in the next step.
  • the channel bandwidth is 6.4 MHz and the number of subcarriers is 32.
  • the channel is a multipath Rayleigh fading channel with a channel multipath number of 2, a delay of [O ns, 260 ns], and a power delay profile subject to exponential decay.
  • the transmit antenna inserts a pilot sequence of two consecutive symbols every six OFDM symbols, and the pilot sequences on different antennas are orthogonal to each other. Since there is one channel between each transmission and each receive antenna, each channel estimation result actually contains an estimate of 4 channels.
  • pilot allocation is as follows: At time 1, antenna 1 is allocated all pilot data, and antenna 2 is also assigned a block pilot of the same 1. At time 2, pilot data of all 1 is still allocated on antenna 1, and pilot data of -1 is allocated on antenna 2. After the pilot allocation, the data of the data segment is added, and the pilot is added after 6 times, and the allocation method is the same.
  • the purpose of this is to enable the receiving end to utilize the characteristics of continuous pilot data and slow fading of the channel to perform antenna-related data separation, thereby simplifying the computational complexity of the particle filter.
  • the allocation manners of the different antenna pilots above are mutually orthogonal, and the sampling may not be orthogonal. For example, at time 1, antenna 1 is allocated all the pilot data, and antenna 2 is also assigned a block of the same 2. Pilot; At time 2, all 1 pilot data is still allocated on antenna 1, and all 1 pilot data is allocated on antenna 2, which can also meet the requirements of the present invention. Therefore, the pilot symbol for the multi-antenna system of the present invention is such that the pilot matrix formed by ⁇ ⁇ transmit antennas and ⁇ consecutive moments is a full rank matrix.
  • each receiving antenna uses the pilot data of two consecutive pilot times to perform a simple addition and subtraction operation to obtain the signals of the transmitting antenna 1 and the transmitting antenna 2.
  • the specific operation may be as follows: The data of the pilot time 1 on the receiving antenna 1 plus the data of the pilot time 2 on the receiving antenna 1 can obtain twice the transmission data of the channel H u passing through the transmitting antenna 1 to the receiving antenna 1. The data of the pilot time 1 on the receiving antenna 1 minus the data of the pilot time 2 on the receiving antenna 1 can obtain twice the transmission data passing through the transmitting antenna 2 to the receiving antenna 1 channel H 2 .
  • the receiving antenna 1 pilot time data on 1 receive antenna plus the pilot data time 2 times 2 can be obtained via the transmission antenna 2 to the reception antenna 2 to transmit data channel H 22.
  • the data of the pilot time 1 on the receiving antenna 2 minus the data of the pilot time 2 on the receiving antenna 2 can obtain twice the transmission data of the channel H 12 passing through the transmitting antenna 1 to the receiving antenna 2.
  • particle filter channel estimation can be performed for each channel (H 1 H 21 , H 12 , H 22 ). This estimation is performed on all subcarriers. Since each channel is estimated for each subcarrier, since the channel information is in the form of a complex number, the application of the particle filter directly involves two-dimensional particle filtering. The process, which is related to the joint estimation of two-dimensional random variables, has a high complexity. However, since the design of the pilot data is a real number, the data obtained in the second step is further separated by the real imaginary part at the receiving end, and the two pieces of information obtained after the separation of the real imaginary parts are respectively the pilot signals passing through one channel. The virtual channel and imaginary parameters formed by the partial parameters constitute the reception result of the virtual channel. For such a result, if the pilot data and the received noise are known, then the probability distribution of the two virtual received signals can be estimated.
  • the present invention can obtain 8 virtual received signals of 4 channels of each subcarrier of every 2 consecutive pilot blocks of the 2 transmit and receive systems. Particle filtering is performed for each virtual received signal. First, a rough estimate of these channels is made using the LS algorithm. Based on this rough estimate, the initial particles of the particle filter are initialized. Determine the particle range and particle number of the particle filter, and then set the weight of each particle to 1 / (which is the total number of particles). In this example, the number of particles can be set to 50, and the step size of the particle can be set to 0. 05, taking the particle value in both positive and negative directions, you can get 50 channel particle values; if you take 100 particle values, the particle step size can be set to 0.
  • denotes a channel
  • ⁇ ] denotes a received signal, indicating noise
  • denotes a transmission power at the second transmitting antenna
  • the SNR filter itself has a strong anti-noise capability, so in practical applications, the estimation of noise is not very accurate, and the corresponding noise variance can be set according to different situations.
  • the noise variance when the transmit power on the transmit antenna is zero, the noise variance is 0.1.
  • the noise variance can be set to a larger value such as 0.3 according to the theoretical value of the noise variance, such as 0.1. System tracking. After the initial tracking of the channel, this noise variance can be set to an appropriate value such as 0.1 for accurate tracking.
  • the noise variance value can be set larger, and the estimation algorithm of the present invention can still track and converge quickly; in a fast time-varying channel, the noise variance should be set smaller to better track Channel changes.
  • the flexible setting and use of the noise variance can adapt the particle filter based channel estimation method proposed by the present invention to channels in various situations.
  • N sff ⁇ N lh ( N lh ⁇ M)
  • N lh ⁇ M perform the resampling algorithm. Resampling is not only computationally intensive, but if used improperly, system performance is greatly reduced, so there is no need to resample at every step.
  • the resampling threshold N e# satisfies - 7 ⁇ ——, where ').
  • the invention is in the sequence ⁇ (( )) 2 )
  • a resampling step is added.
  • the algorithm flow diagram is shown in Fig. 17.
  • the principle of the sequential important sampling method is shown in Fig. 18.
  • the particles with large weight values are selected for resampling, and the resampling algorithm is based on probability.
  • the distributed resampling algorithm as shown in Figure 19, represents the number of particles before resampling, and performs integral operations on the prior probability, then obtains its probability distribution function, so it can be found that the particles with larger prior probability values The larger the value of the corresponding probability distribution function, the greater the slope of the
  • the channel estimation values of the data symbols are obtained by interpolation, thereby providing channel information for the data detection or decoding module of the next link.
  • the invention adopts a particle filter based channel estimation method, and uses the sequential important sampling method and the probability distribution function based resampling method to track and estimate the wireless channel based on the Bayesian principle and the Monte Carlo sampling principle, and the simulation result
  • This channel estimation method is shown to be better than the conventional channel estimation method, as shown in Figs. 20 and 21.
  • Figure 20 shows the minimum mean of the channel estimation method based on particle filter
  • the square error (MSE) performance is better than the extended Kalman (EKF) based channel estimation method and the least squares based channel estimation method (LS), and as the number of particles increases, the performance of the MSE also becomes better, but the number of particles To a certain extent, the performance improvement of MSE is not very obvious.
  • Figure 21 shows that the system using the particle filter based channel estimation method has better bit error rate (BER) performance than the system using the extended Kalman (EKF) channel estimation method and the least squares channel estimation method (LS).
  • the invention is also applicable to a single-issue single-receipt (SIS0), single-issue multi-receipt (SIM0), multi-issue single-receipt (MIS0) communication system, and has a wide range of applications.
  • Figure 22 is a schematic block diagram of a channel estimation apparatus according to a second preferred embodiment of the present invention.
  • the particle filter-based channel estimation apparatus includes: a pilot information setting unit (preprocessing unit;) for The pilot information is processed according to whether the communication system used is a single antenna system or a multi-antenna system; an initial estimation unit (LS) is configured to: perform least square estimation on the channel according to known pilot information, and obtain channels at the respective pilots.
  • a pilot information setting unit preprocessing unit
  • LS initial estimation unit
  • particle initialization setting unit used to set the particle range and the number of particles, and assign an initial weight to each particle
  • an important sampling unit for the conditional probability density of the received signal conditional on the channel particle value as a priori Probability, constructing a Bayesian model with the current received signal, calculating the posterior probability density of the channel particle values subject to the current received signal; and normalizing the weights of all the particles and performing a weighting operation; , used to resample the particles according to the judgment conditions, among which the important sampling unit and the re-collection
  • the sample unit performs a time recursive operation
  • an interpolation operation unit is configured to perform an interpolation operation to obtain channel estimation values of all transmitted data symbols, thereby completing estimation of channel state information.
  • the signal received by the receiver is sent to the pre-processing unit through a fast Fourier transform unit (FFT); the pre-processing unit performs a simple addition and subtraction operation according to the pilot structure of the single antenna system or the multi-antenna system.
  • FFT fast Fourier transform unit
  • the initial estimating unit performs initial LS-based channel estimation on the channel according to the provided pilot information, and uses the pilot information to obtain an initial channel estimation value based on the least squares criterion LS, Then, the channel information of the complex form obtained by the LS estimation is decomposed into channel information in the form of a real part and an imaginary part; at this time, the particle initial setting unit sets the particle range and the number of particles, and assigns an initial weight to each particle; then, it is important
  • the sampling unit takes the received signal as an observation value, and the conditional probability density of the received signal conditional on the channel particle value as a prior probability, constructs a Bayesian model in combination with the current received signal, and calculates a channel particle value conditioned on the current received signal.
  • Posterior probability density then, normalize the weights of all particles This Obtain a new weight value for each particle; use the filter degradation detection formula to judge the performance, and when the performance is lower than the threshold, perform the resampling algorithm; and obtain the accurate channel estimation of the pilot symbol at the ready time according to the above steps.
  • the value, the channel estimation value in the form of the real part and the imaginary part is combined, so that the channel estimation value is a complex form capable of reflecting the amplitude and phase information; finally, the channel information value of the data symbol is obtained by interpolation, so as to be the data of the next link. Detection or decoding provides more accurate channel information.
  • a particle filtering based channel estimation method and apparatus are provided, which overcomes the deficiencies in the prior art, so that the estimation performance is robust and unaware in an environment where channel statistics information is not known. It is characterized by strong stickiness and strong anti-noise ability, and is easy to implement. Compared with several particle filter based channel estimation methods proposed by K.
  • the second preferred embodiment of the present invention is more in line with the characteristics of the actual channel unknown, without any known channel statistical information, and A better resampling algorithm is proposed.
  • the second preferred embodiment of the present invention has the characteristics of high estimation accuracy, short convergence time, and stable performance.
  • the second preferred embodiment of the present invention solves the problem of the conditional probability based on the Bayesian principle in a two-dimensional space by decoupling the antenna and real and imaginary solutions of the pilot information.
  • the problem of solving the conditional probability, and using the resampling algorithm based on the probability distribution function value greatly simplifies the computational complexity of the particle filter algorithm.
  • the second preferred embodiment of the present invention employs a flexible noise estimation algorithm, which is characterized by strong anti-noise capability. Therefore, as the computational storage cost decreases, the particle filter-based channel estimation algorithm is gaining more and more attention.
  • the particle filter-based channel estimation algorithm proposed in the second preferred embodiment of the present invention can be implemented by hardware, and has certain engineering. Value.
  • the wireless channel may be in a linear channel, a nonlinear channel, a channel with Gaussian white noise, a channel with non-Gaussian noise, the mobile terminal has the characteristics of mobility and the motion speed may cause Doppler shift,
  • the wireless environment in which the actual receiver is located at different times is very complicated. Therefore, the present invention defines a channel with linear and Gaussian white noise as a non-bad channel environment, and a channel defining a non-linear channel or non-Gaussian noise is a bad channel environment.
  • the receiver first determines if the channel is a bad channel environment or a non-bad channel environment.
  • the selective channel estimation method is adopted according to whether the channel is in a bad channel environment, the particle channel filtering based channel estimation method is adopted for the bad channel environment, and the variable forgetting factor RLS channel estimation method is adopted for the non-bad channel environment.
  • the channel estimation method obtains channel state information which is a pilot symbol
  • the channel state information of the data symbol is obtained by an interpolation algorithm.
  • the signal detection algorithm is used, and the channel state information of the data symbols obtained by the interpolation algorithm is combined to obtain the transmission data information of the transmitter.
  • the bit error rate of the transmitted data information obtained in the above step is calculated.
  • the channel estimation method based on particle filtering is used to re-acquire the channel estimation.
  • the accuracy of the channel estimation is improved by increasing the number of samples of the particles.
  • a selective channel estimation method is adopted in consideration of the complexity of the wireless channel. 1. First determine if the wireless channel is a bad channel. If the channel is a linear and Gaussian white noise channel, the wireless channel is judged to be a non-bad channel environment; if the channel is a non-linear channel or a non-Gaussian noise channel, the channel is judged to be a bad channel environment.
  • a particle filter based channel estimation method or a channel estimation method based on a variable forgetting factor RLS is adopted.
  • the channel is a poor channel environment, and the particle filter-based channel estimation method is adopted; when the channel is a non-bad channel environment, a variable forgetting factor RLS-based channel estimation method is adopted.
  • the signal detection algorithm is used to obtain the transmitted data information of the transmitter.
  • FIG. 24 is a block diagram showing a principle of a channel estimation apparatus according to a third preferred embodiment of the present invention.
  • the selective channel estimation apparatus includes: determining a channel attribute unit, determining whether the channel is a bad channel environment; Channel estimator, a channel for estimating a bad channel environment; a channel estimator based on a variable forgetting factor RLS filtering, a channel for estimating a non-bad channel environment; a bit error rate indicator determining unit for determining whether the bit error rate is Meet performance requirements.
  • the channel attribute is judged to determine whether the channel attribute unit determines whether the channel is a bad channel environment; and the channel is a bad channel, and the particle filter is used.
  • a channel estimator wherein the channel is a non-bad channel, a channel estimator based on a variable forgetting factor RLS is used; after the channel estimator completes estimation of the channel state information, the channel state information is sent to the interpolation unit for interpolation, and the data symbol is correspondingly obtained.
  • Channel state information the channel state information corresponding to the data symbol is sent to the data detecting unit, and the signal detection algorithm is used to obtain the transmission number information of the transmitter; the error rate of the data information of one frame is calculated; the bit error rate indicator determining unit judges Whether the bit error rate satisfies the system requirements. If the bit error rate indicator meets the performance requirement, it indicates that the process of receiving a data symbol ends. If the bit error rate indicator fails to meet the performance requirements, the particle filter-based channel estimator is used to re-channel. Estimated, by increasing the sampling To increase the number of children Estimation accuracy of the channel estimator.

Description

信道估计方法和设备 技术领域
本发明涉及一种无线通信的信道估计方法, 用于无线传输技术领域, 具 体是一种基于可变遗忘因子 RLS (递归最小平方)滤波的信道估计方法和设备, 一种基于粒子滤波的信道估计方法和设备, 以及一种选择性信道估计方法和设 备。 背景技术
无线通信中信号在传播过程受到信道衰减、 多径时延扩展和多普勒频率 扩展等因素的影响, 在接收端为了能够较好地恢复出发送信号通常采用相干解 调, 而相干解调需要信道参数信息, 它可以通过信道估计来获得, 因此信道估 计器的性能直接影响系统性能, 成为接收机的关键技术之一。 正交频分复用 ( OFDM )系统能够有效抵抗多径扩展造成的符号间干扰, 使得在恶劣无线衰落 信道下进行数据传输成为可能。 在 OFDM 系统中, 基于二维最小均方误差准则 (MMSE)的信道估计方法是理论分析中性能最优的信道估计方法, 但是这类方法 不仅复杂度极高, 而且还需要知道部分或全部信道先验信息, 因此工程应用价 值不高。 另外,基于最小平方准则(LS)的信道估计方法, 虽然运算复杂度不高, 但是在信噪比下降时性能会下降很快, 不适合用于中低信噪比的场合。 因此, 开展信道估计的研究并寻找工程上适用的信道估计方法直接关系到接收机检 测和译码性能的优劣, 对提高宽带移动通信系统性能起着至关重要的作用。
经对现有技术的文献检索发现, Dieter Schafhuber 等人在 2003 年的 IEEE国际无线通信的信号处理学术会议( IEEE Workshop on Signal Proces s ing Advances in Wireless Communicat ion )上发表文章 "无线 MIM0-0FDM系统中 双选择性衰落信道的自适应辨识和跟踪 (Adapt ive Ident if icat ion and Tracking of Doubly Select ive Fading Channel s for Wireless MIM0-0FDM Sys tems )" , 该文提出一种适用于多天线 OFDM系统的基于最小均方自适应滤波 器 (LMS ) 的信道估计方法, 在无需知道信道统计信息的基础上, 利用导频符 号和 LMS滤波器, 对时变信道进行跟踪和估计, 但该方法有个明显的缺点就是 估计精度不是很高, 尤其是在中低信噪比的环境下, 远远小于基于二维最小均 方误差准则的信道估计方法。 检索中还发现, 中国专利申请号为 CN 200410021864. 8 , 名称为: 二种 MIM0- OFDM系统的自适应信道估计方法, 该专 利中提出运用自适应算法例如 LMS和 RLS来进行信道估计, 从而实现随着信道 变化而动态跟踪信道参数的功能, 但是, 在该专利中, 采用的导频结构过于复 杂且需要专门的子载波用于信道估计, 且该专利虽然给出了 LMS算法的应用实 例但并未给出 RLS算法的应用实例, 更未采用变化的遗忘因子的方案, 所以该 专利具有频谱效率和估计精度都不是很高且遗忘因子方案不灵活的缺点。 因 此, 有必要寻找一种运算复杂度和 LMS算法相当但精度远高于 LMS算法的灵活 的信道估计器。
此外, 信道估计器的性能对宽带移动通信系统性能的好坏同样起着至关 重要的作用。 位由于无线信道是频域和时域上的时变信道且是非线性的, 数学 建模和定量分析都比较困难, 往往采用线性的信号处理方法来对非线性的时变 无线信道进行近似的估计。 但在许多要求对动态系统进行实时估计的问题中, 系统的非线性往往成为困扰着最优估计的重要因素。 由于实时处理和计算存储 量的要求, 通常选用递推滤波算法来求解此类问题, 包括扩展卡尔曼滤-波 (EKF)、 修正增益的扩展卡尔曼滤波 (MGEKF)等。 它们基本思想是通过采用参数 化的解析形式对系统的非线性进行近似, 以寻求满意的估计精度。 但扩展卡尔 曼滤波算法只适用于滤波误差和预测误差很小的情况, 否则, 滤波初期估计协 方差下降太快会导致滤波不稳定甚至发散。 修正增益的扩展卡尔曼滤波算法虽 然通过改善增益矩阵, 相应改善了状态协方差的估计性能, 但该方法对测量误 差有一定限制。 若测量误差较大, 则算法在收敛精度、 收敛时间及稳定性等方 面表现得很不理想。 为解决上述问题, 随着计算存储成本的下降, 一种崭新的 基于贝叶斯原理的序贯蒙特卡罗粒子滤波器 (PF )逐渐受到重视。
经对现有技术的文献检索发现, L Huber 等在 2003年的 IEEE国际通信 大会 ( IEEE Internat ional Conference on Communica t ions )上发表文章 "粒 子滤波在 MIM0无线通信中的应用 ( Appl icat ion of Part icle Fi l ters to IMO Wireless Communicat ions )" , 该文提出一种适应于 MIM0系统的基于粒子滤波 的信道估计方法, 该方法在假设信道是 AR模型的前提上使用导频信息进行初 始信道信息的估计, 再利用信道的 AR模型, 得到信道的各个粒子, 从而进行 粒子滤波得到信道估计值,但该方法有个明显的缺点就是假设信道统计信息是 已知的,这往往不符合信道的实际情况, 因此这种方法有很大的局限性。 另外, 该方法未提出一种很好的重采样算法。 检索中还发现, W. H Chin等在 2002年 的 IEEE国际数字信号处理大会 ( IEEE Internat iona l Conference on Digi ta l Signal Proces s ing )上发表文章 "空时块编码系统中使用粒子滤波的信道估 计 ( Channel tracking for space-t ime block coded sys tems us ing par t icle f i l ter ing )" 中提出了一种适用于 MISO系统得基于粒子滤波的信道估计方法, 其核心仍旧是建立在已知信道是 AR模型的, 仍旧没有提出很好的重采样算法。 因此, 上述两篇文章提出的算法不大适用于信道未知的实际情况, 工程实际意 义不是很突出。 . 发明内容
本发明的目的在于克服现有技术中的不足, 提供一种基于可变遗忘因子 RLS 滤波的信道估计方法和设备以及一种基于粒子滤波的信道估计方法和设 备, 使其在不知道信道统计信息的情况下, 具有估计性能稳健、 鲁棒性强、 抗 加性高斯白噪声和非高斯噪声的能力强的特点, 且易于实现。 另外, 考虑到无 线信道环境是十分复杂的, 无线信道可能是线性或非线性的信道, 环境噪声也 可能是加性高斯白噪声或非高斯非平稳的噪声, 而移动终端在不同的时刻可能 处于不同的信道环境, 因此, 本发明提出了一种选择性信道估计方法和设备, 可以根据不同的信道环境采用不同的信道估计方法和设备, 从而提高了通信系 统的抗干扰能力和性能。
根据本发明的第一方案, 提出了一种基于可变遗忘因子递归最小平方滤 波的信道估计方法, 包括: 初始估计步骤, 根据提供的导频信息对信道进行初 始估计, 得到所有导频处的信道的初始值; 逆傅立叶变换步骤, 对估计出的信 道初始值进行逆傅立叶变换; 跟踪和估计步骤, 运用基于可变遗忘因子的递归 最小平方滤波, 对时变无线信道进行跟踪和估计; 傅立叶变换和插值步骤, 通 过傅立叶变换和插值算法得到数据所对应的信道状态信息值, 其中后一时刻导 频符号的遗忘因子大于或等于前一时刻导频符号的遗忘因子。
根据本发明的第二方案, 提出了一种基于可变遗忘因子递归最小平方滤 波的信道估计设备, 包括: 初始估计单元, 用于根据提供的导频信息对信道进 行初始估计, 得到所有导频处的信道的初始值; 逆傅立叶变换单元, 用于对估 计出的信道初始值进行逆傅立叶变换; 递归最小平方滤波器, 具有可变遗忘因 子, 用于对时变无线信道进行跟踪和估计, 其中后一时刻导频符号的遗忘因子 大于或等于前一时刻导频符号的遗忘因子; 傅立叶变换单元, 用于对递归最小 平方滤波器的输出进行傅立叶变换; 以及插值运算单元, 用于对傅立叶变换单 元的输出执行插值运算, 获得数据所对应的信道状态信息值。
根据本发明的第一和第二方案, 在传统的频域最小平方信道估计器的基 础上设计了一种时间 /时延域的基于可变遗忘因子 RLS 滤波的信道估计方法和 设备, 通过对时变信道的跟踪与估计, 得到噪声作用下的信道状态信息, 从而 得到更加符合实际的信道估计值。 本发明的信道估计方法和设备与传统的信道 估计方法和设备相比, 能够很好地跟踪和估计信道, 具有收敛性好、 估计精度 高等特点, 同时能够通过调整滤波器长度等参数来对信道估计精度和计算复杂 度进行折中和平衡, 从而获得更好的估计性能, 因此, 这种信道估计方法和设 备方便灵活, 很适合实际应用, 可以为第三代(3G )、 超三代(B3G )、 第四代 ( 4G )蜂窝移动通信和数字电视、 无线局域网 (WLAN )、 无线广域网 (WWAN ) 等系统的信道估计方案提供重要的理论依据和具体的实现方法和设备。
根据本发明的第三方案, 提出了一种基于粒子滤波的信道估计方法, 包 括: 初始估计步驟, 利用导频信息进行初始信道估计; 粒子滤波步骤, 在初始 信道值的基础上利用粒子滤波得到准确的信道估计, 其中所述粒子滤波步驟包 括以下子步骤: 后验概率密度计算步骤, 以信道粒子值为条件的接收信号的条 件概率密度作为先驗概率, 与当前接收信号构造贝叶斯模型, 计算出以当前接 收信号为条件的信道粒子值的后臉概率密度; 归一化和加权步骤, 对所有粒子 的权值进行归一化并进行加权运算, 从而得到准确的信道估计值。
根据本发明的第四方案, 提出了一种基于粒子滤波的信道估计设备, 包 括: 初始估计单元, 用于利用导频信息进行初始信道估计; 粒子初始化设置单 元, 用于设置粒子范围和粒子数目, 并为每个粒子分配初始权重; 重要采样单 元, 用于以信道粒子值为条件的接收信号的条件概率密度作为先猃概率, 与当 前接收信号构造贝叶斯模型, 计算出以当前接收信号为条件的信道粒子值的后 验概率密度; 以及对所有粒子的权值进行归一化, 并进行加权运算; 重采样单 元, 用于根据判断条件对粒子进行重采样, 其中所述重要采样单元和所述重采 样单元执行时间递归操作, 从而得到发送所有导频处的准确信道估计值。 根据本发明的第三和第四方案, 提供了一种基于粒子滤波的信道估计方 法和设备, 克服了现有技术中的不足, 使其在不知道信道统计信息的环境下具 有估计性能稳健、 鲁棒性强、 抗噪声的能力强的特点, 且易于实现。 相比于 K. Huber及 W. H Chin等人提出的几种基于粒子滤波的信道估计方法, 本发明更加 符合实际信道未知的特点, 无需已知任何信道统计信息, 且提出了更好的重采 样算法。 相比于求解此类问题的递推滤波算法(EKF、 MGEKF等), 本发明具有估 计精度高、 收敛时间短且性能稳定的特点。 另外, 本发明通过天线解耦、 导频 信息的实部和虚部分解, 把二维甚至多维空间中基于贝叶斯原理的条件概率求 解问题化简为一维空间中条件概率的求解问题, 且运用基于概率分布函数值的 重采样算法大大筒化了粒子滤波算法的计算复杂度。 而且本发明采用灵活的噪 声估计算法, 从具有抗噪声能力强的特点。 因此, 随着计算存储成本的下降, 基于粒子滤波的信道估计方法越来越受到重视, 本发明提出的基于粒子滤波的 信道估计方法可以通过硬件来实现, 具有一定的工程应用价值。
根据本发明的第五方案, 提供了一种选择性信道估计方法, 包括: 判断 信道属性, 将非线性信道或非高斯噪声的信道判定为恶劣信道环境, 将线性且 高斯白噪声的信道判定为非恶劣信道环境; 针对恶劣信道环境, 采用根据本发 明第三方案所述的基于粒子滤波的信道估计方法, 进行信道估计; 针对非恶劣 信道环境, 采用根据本发明第一方案所述的基于可变遗忘因子递归最小平方滤 波的信道估计方法, 进行信道估计; 基于信道估计的结果, 进行插值和信号检 测; 计算所检测出的信号的误码率, 并判断误码率是否满足预定的性能要求; 如果误码率不满足预定的性能要求, 则采用根据本发明第三方案所述的基于粒 子滤波的信道估计方法, 再次进行信道估计, 其中通过增大粒子数目来提高信 道估计的精度。
根据本发明的第六方案, 提供了一种选择性信道估计设备, 包括: 一种 选择性信道估计设备, 包括: 信道属性判断单元, 用于将非线性信道或非高斯 噪声的信道判定为恶劣信道环境, 以及将线性且高斯白噪声的信道判定为非恶 劣信道环境; 根据本发明第四方案所述的基于粒子滤波的信道估计设备, 用于 对恶劣信道环境进行信道估计; 根据本发明第二方案所述的基于可变遗忘因子 递归最小平方滤波的信道估计设备, 用于对非恶劣信道环境进行信道估计; 插 值和信号检测单元, 用于基于信道估计的结果, 进行插值和信号检测; 误码率 判断单元, 用于计算所检测出的信号的误码率, 并判断误码率是否满足预定的 性能要求, 其中如果误码率不满足预定的性能要求, 则通知根据本发明第四方 案所述的基于粒子滤波的信道估计设备再次进行信道估计, 其中通过增大粒子 数目来提高信道估计的精度。
根据本发明的第五和第六方案,提供了一种选择性信道估计方法和设备, 针对无线信道处于复杂的信道环境, 线性信道、 非线性信道、 高斯白噪声的信 道、 非高斯噪声的信道, 移动终端的可移动性导致信道环境的变化, 可以把信 道区别为恶劣信道环境和非恶劣信道环境, 利用了基于粒子滤波的估计方法和 设备能够适用于恶劣信道和非恶劣信道环境且估计精度高的特点, 也利用了基 于可变遗忘因子 RLS滤波的估计方法和设备具有估计方法简单且估计精度较高 的特点, 而分別采用基于粒子滤波或基于可变遗忘因子 RLS滤波的信道估计方 法和设备, 最后判断系统的比特误码率是否达到性能指标要求而决定是否重新 进行信道估计的过程。 因此, 本发明提出的选择性信道估计方法可以根据无线 信道环境的变化以及移动终端的位置的变化而分别采用适当的信道估计方法, 且不同的信道估计方法可以通过软件无线电的思想和方法通过软件或硬件的 模块来实现, 具有很高的工程应用价值。 附图说明
图 1是根据本发明第 1优选实施例的 MIM0- OFDM原理图
图 2 是根据本发明第 1优选实施例的导频结构图
图 3 是根据本发明第 1优选实施例的信道估计算法原理图
图 4 是根据本发明第 1优选实施例的 RLS滤波算法原理图
图 5 是根据本发明第 1优选实施例的 MSE- M仿真性能图
图 6 是根据本发明第 1优选实施例的 MSB- L仿真性能图
图 7 是根据本发明第 1优选实施例的 MSE-SNR仿真性能图
图 8 是根据本发明第 1优选实施例的 BER- SNR仿真性能图
图 9 是根据本发明第 1优选实施例的 BER- Dopp ler仿真性能图 图 10 是根据本发明第 1优选实施例的遗忘因子方案比较性能图 图 11 是根据本发明第 1优选实施例的单天线 MSE-SNR仿真性能图 图 12 是根据本发明第 1优选实施例的单天线 BER- SNR仿真性能图 图 13 是根据本发明第 1 优选实施例的采用可变遗忘因子方案 B 的
MSE- SNR仿真性能图
图 14 是根据本发明第 1 优选实施例的采用可变遗忘因子方案 B 的
BER-SNR仿真性能图
图 15是根据本发明第 1优选实施例的信道估计设备的原理方框图 图 16是根据本发明第 2优选实施例的信道估计方法示意图
图 17是根据本发明第 2优选实施例的序贯重要采样法流程图
图 18是根据本发明第 2优选实施例的序贯重要采样法原理图
图 19是根据本发明第 2优选实施例的基于概率分布的重采样算法原理图 图 20是根据本发明第 2优选实施例的 MSE仿真性能图
图 21是 _据本发明第 2优选实施例的 BER仿真性能图
图 22是根据本发明第 2优选实施例的信道估计设备的原理方框图 图 23是根据本发明第 3优选实施例的信道估计方法的示意图
图 24是根据本发明第 3优选实施例的信道估计设备的原理方框图 具体实施方式
以下结合附图对本发明的具体实施进行详细的说明: - 第 1优选实施例
基于可变遗忘因子 RLS滤波的信道估计方法和设备
首先, 对本发明第 1优选实施例的原理和设计思想进行详细说明。
1. 信道估计器使用导频信息通过最小平方(LS )算法而获取频域上导频 处的信道状态信息的初始值。
本发明的导频信息为结构采用块状的导频符号, 每个导频符号包括所有 的子载波上的时频块, 导频的帧采用在一帧数据前面加入若干个导频符号的帧 结构, 在多天线系统中不同发射天线在同一时刻的导频是互相正交的。 由于无 线信道往往是未知的且不知道任何先猃统计信息, 因此可以在每帧数据区前均 放置导频符号用于在频域内对信道进行最小平方准则的估计, 而不需要任何已 知的信道统计信息。 申请号为 CN200410021864. 8 的专利先采用块状导频再采 用子载波做导频, 这种导频方案设计过于复杂且占用的频谱资源比较多, 这种 导频方式实际上是一种改进的梳状导频方式, 比较适合于快衰落时间选择性的 信道, 而不适用于象无线局域网这类慢衰落频率选择性信道。 因此, 本发明的 导频方式在慢衰落频率选择性信道下的性能是优于在先申请号为
CN200410021864. 8 专利中的导频方式的。 在本发明中, OFDM 系统的接收机根 据发送端所提供的导频信息, 运用 LS信道估计方法, 在频域求出每个导频处 信道的初始估计值, 这是不考虑噪声作用的理想化信道估计方法。
2. 通过快速逆傅立叶变换(IFFT )得到时间 /时延域的未考虑噪声作用 下的信道状态信息。 即接收机对 LS算法估计出来的信道估计初值做快速逆傅 立叶变换, 得到时间 /时延域的信道估计值。
3. 以当前时刻未考虑噪声作用下的信道状态信息作为期望值, 以当前时 刻和前几个时刻的信道状态信息为输入值, 再利用 RLS滤波器跟踪和估计当前 时刻的考虑噪声作用下的准确的信道状态信息。
先经过 LS 算法再经过 IFFT 变换得出的时间 /时延域的信道估计值通过 RLS滤波算法得到考虑噪声情况下的时间 I时延域中的更精确的信道估计值,在 用 RLS算法的过程中, 信道期望值是当前时刻 LS算法通过 IFFT得到的信道估 计值, 信道当前值是当前时刻和前几个时刻的 LS算法通过 IFFT得到的信道估 计值, 利用 RLS算法的递归关系, 可以求出当前时刻的考虑噪声作用下准确的 信道估计值, 从而实现对时变无线信道的即时跟踪与估计。 注意, 在这里信道 的前几个时刻值是由滤波器长度决定的, 滤波器长度和系统性能之间是一种复 杂的关系。 当滤波器长度在一定的范围以内, 系统性能随着滤波器长度增大而 变好, 当滤波器长度超过某个值, 系统性能随着滤波器长度增加而变差, 因此, 系统性能在滤波器为最佳值时为最好。 滤波器最佳值是基于维纳滤波原理通过 理论计算而得到的。 但滤波器处于这个最佳值时, 计算复杂度仍然很高, 因此 滤波器长度的选取值往往小于最佳值, 一般取最佳值的一半就可以达到很好的 系统性能, 这样就取得了系统性能和计算复杂度的平衡。
4. 在跟踪和估计的过程中采用可变遗忘因子, 再在时间 /时延域进行递 归运算, 得到考虑噪声作用的准确的信道状态信息。
可变遗忘因子方案采用不同的导频符号,在一帧中时域上不同时刻点的 导频符号具有不同的权系数, 离数据符号越近的导频符号的权系数越大。 这种 设置其实是一种优化的权系数方案, 不同导频的权系数被设置为不同的值以反 映不同导频符号对数据信道信息所起的不同作用。 考虑到不同导频的信道传输 函数对数据信道传输函数所起的作用不同, 所以随着导频信道传输函数与数据 信道传输函数在时间上的距离越来越近, 导频信道传输函数对数据符号的信道 传输函数所起的作用越来越大, 因此权系数的设置为不一样是合理的。 现有技 术中的遗忘因子采用常规法, 它是给每个导频符号以相同的遗忘因子值。 而本 发明的遗忘因子是变化的, 根据信道特征、 算法需求、 仿真结果和经猃因素来 设置遗忘因子, 使后一时刻导频符号的遗忘因子大于或等于前一时刻导频符号 的遗忘因子。
本发明的遗忘因子可以两步变化, 也可以自适应变化。 两步法是给前几 个导频设置较小的遗忘因子值, 目的是使通信系统更多地依靠当前导频符号估 计得到的正确的信道状态信息; 给后几个导频符号设置高的遗忘因子值, 目的 使通信系统更多地依靠前次信道估计提供的信道的统计信息。 自适应遗忘因子 法, 是给所有导频符号设置不同的遗忘因子: 首先给第一个导频符号设置一个 初始值, 然后给其他导频符号设置一个学习速率, 这样所有不同的导频符号均 以学习速率来不断更新各自的遗忘因子值, 从而使不同的导频符号具有不同的 遗忘因子。 初始遗忘因子值和学习速率的设置由诸如信道特性、 算法需求、 人 为经验、 计算机仿真等因素决定。 在一般情况下, 常规法所给定的固定的遗忘 因子值可以作为可变遗忘因子的初始值, 可以根据导频符号的个数来灵活安排 学习速率。 学习速率可以设置为固定的或变化的, 但是要保证一桢中最后一个 时刻的导频符号的遗忘因子值不能超过 1。 遗忘因子值和学习速率可以随时刻 n不断变化, 变化的学习速率可以采用自变量为时间的线性函数、 指数函数或 其它映射关系。
5. 进行快速傅立叶变换, 得到时间 /频率域的信道传输函数。
以上步骤求得时间 /时延域的信道估计的准确值,但是信号检测需要的是 信道的时间 /频率域的状态信息, 因此, 对时间 /时延域的信道估计值做快速傅 立叶变换( FFT ) 变换得到时间 /频率域信道估计值。
6. 通过插值算法得到数据符号所对应的信道传输函数, 把数据符号所对 应的传输函数送给信号检测模块进行检测或译码。 以下, 给出了本发明第 1优选实施例的具体应用示例。 (1) MIMO-OFDM系统的构造
MIMO - OFDM系统的构造如图 1所示, 本发明采用 4发 4 收的 MIM0-0FDM 系统, 采用 QPSK的调制方式, 空时编码方案采样正交空时块码(0- STBC ), 码 率为 1 /2 , 空时解码采样最大似然(ML )译码方案, 载频 4GHz , 带宽 6M, 子载 波个数为 64 , 考虑到本发明的信道估计方法比较适合于频率选择性慢衰落信 道, 因此信道模型采用 ITU提出的多径瑞利衰落信道, 信道多径数目为 5 , 时 延为 [0 260 520 780 1040] ns ,功率为 [- 1. 78 0 -7. 47 -10 -12. 62] dB , 运动 速度 3ktn/h„
(2)导频的设计
导频的设计如图 2 所示, 导频采样正交化设计, 不同发送天线发送的导 频互相正交,发送天线 1到发送天线 4的导频分別为 =[1, 1, 1, 1] τ, P2=[l, -1,
Figure imgf000012_0001
本发明的实施例中虽然采用了块状正交的导频结构, 目的是为了使本发 明更适用于慢衰落频率选择性信道, 但这并不影响本发明的一般性, 在增加计 算的复杂性,增加频谱资源的占用下,也可以使用其他导频结构,如梳状导频、 离散导频等结构。
(3)基于最小平方法 ( LS ) 的信道估计方法
基于 LS的信道估计方法如图 3所示, 4个接收机根据 4个发送天线所提 供的正交化的导频信息, 在频域利用最小平方法(LS )求出每个导频处的信道 的初始估计值, 这实际上是一种不考虑噪声影响的理想化的信道估计方法。
(4) 快速逆傅立叶变换( IFFT )
快速逆傅立叶变换(IFFT)如图 3所示,每个接收机对各自的 LS算法估计 出来的信道估计初值做快速逆傅立叶变换(IFFT ), 得到时间 /时延域的信道估 计值。 这里, 考虑到信道的最大多径长度是未知的, 所以本发明采样最保守的 假设, 即最大多径长度为循环前缀长度加 1 ( L = LCP+1 ), 这样时间 /时延域的信 道状态值在时延域的维数最大为 L = Lcp+1 ? 在本发明中, LeP=15,所以 L为 16。
(5) RLS滤波算法
RLS滤波算法如图 4所示, 时间 /时延域的信道估计值通过 RLS滤波算法 得到考虑噪声情况下的时间 /时延域中的更精确的信道估计值, 在用 RLS算法 的过程中, 信道期望值是当前时刻 LS算法通过 IFFT得到的信道估计值, 输入 值是当前和前几个时刻 LS算法通过 IFFT得到的信道估计值, 利用 RLS算法的 递归关系,可以求出当前时刻的考虑噪声作用下准确的信道估计值, 从而实现 对时变无线信道的即时跟踪与估计。 这里, RLS滤波器的长度 M和最大多径时 延长度 L均会影响到信道估计的精度和计算复杂度, 往往精度越高, 计算复杂 度越高, 如图 5和图 6所示。 另外, RLS滤波算法可以用 LMS滤波算法替代, 尽管 LMS算法的计算复杂度小于 RLS算法, LMS的计算复杂度为 } [M(L^D ], RLS计算复杂度为 y W (^+D ] , 一般 Μ取值很小, 考虑到 LMS的收敛速度和 估计精度均远远小于 RLS算法, 所以在计算成本日益减小的当今时代, RLS算 法更具有应用价值的。
(6)滤波器长度及最大多径时延长度
滤波器长度对系统性能的影响如图 5 ,通过理论分析和计算机仿真找到最 优或次最优的滤波器长度, 从而使信道估计精度和复杂度达到折中和平衡。 本 发明的最优值是通过维纳滤波(wiener f i l ter ing )原理进行理论计算得到, 首先构造信道估计误差的代价函数, 再利用正交原理, 产生维纳 -霍夫 ( Wiener- Hopf ) 方程, 再得到最小均方误差 (匪 SE ) 准则下的最佳滤波器系 数, 这样就可以得到滤波器的长度值, 在本发明中, 根据上述原理得到滤波器 最佳长度为 10 ,但是这时计算复杂度太高,所以滤波器长度一般不能取得很大, 一般取值为最佳长度的一半即可达到很好性能且计算复杂度低很多 , 所以本发 明取 M = 4。 最大多径时延长度对系统性能的影响如图 6 , 最大多径时延长度一 般小于为循环前缀( CP )长度,实际信道最大时延长度远小于循环前缀长度, 最 大多径时延长度是信道自身特性, 是不能人为调整的。
(7) 可变遗忘因子方案
可变遗忘因子方案如图 1所示: 方案 A采用 6个导频符号全部具有相同 的权系数 1 / 6, Η= α ,Ηι+ α2Η2+ α3Η3+ α5Η5+ α6Η6, cr,+ ατ2+ α3+ α4+ as+ α6=1, 遗忘因子采样相同的值, 全部为 0. 9; 方案 Β采用 6个导频符号具有不 同的权系数, 即变化权重系数 , , 使后一个时刻导频的权系数 α ι+1和前一时刻 导频的权系数 ^满足关系 " 1+1= ^+(1, q是一个大于 0小于 1的常数, 这种设 置其实是一种优化的权系数方案, α2, cr3, α4, α5, 被设置为不同的 值以反映不同导频符号所起的不同作用。 考虑到 , ¾, ff Ά, ,#6对 所 起的作用不同,且随着 Η; 2, ... , 6)与 时间上的距离越来越短, H、, ff2, ffi, ff 对数据符号的信道传输函数 H所起的作用越来越大, 因此权系数的 设置是合理的, 所以方案 B比方案 A设计更合理。 方案 B的遗忘因子采用两步 法或自适应法: 两步法是给前两个导频以较小的遗忘因子而给后四个导频以较 大的遗忘因子, 如图 2 , 在本发明中较小遗忘因子为 0. 6 > 较大的遗忘因子为 0. 9 , 给前两个导频符号设置低的遗忘因子值, 目的是使通信系统更多地依靠 当前导频符号估计得到的正确的信道状态信息; 给后四个导频符号设置高的遗 忘因子值, 目的使通信系统更多地依靠前次信道估计提供的信道的统计信息; 自适应遗忘因子法, 给所有导频符号设置不同的遗忘因子, 首先给第一个导频 符号设置一个初始值, 然后给其他导频符号设置一个学习速率, 这样所有不同 的导频符号均具有一个不同的遗忘因子。 初始遗忘因子值和学习速率的设置由 诸如信道特性、算法需求、计算机仿真等因素决定,在算法需求允许的前提下, 可以根据最终的计算机仿真结果, 不断选择和调整遗忘因子值和学习速率。 遗 忘因子值和学习速率可以随时刻刀不断变化, 学习速率可以采用自变量为时间 的线性函数、 指数函数或其它映射关系。 本发明根据经验给出三种自适应遗忘 因子的设置方案:
在方案 1 ,在类似无线局域网等慢衰落频率选择性信道下,初始值一般取 常规法遗忘因子的值, 保证基于自适应遗忘因子信道估计算法的计算复杂度和 基于常规遗忘因子算法基本一致, 第 时刻的导频的遗忘因子满足 = ^ + 0.01π , A。为初始值, 固定的学习速率设为 0. 01, 在本例中初始值取 为 0. 9,这样第六个导频的遗忘因子值为小于 1的值 0. 95 ,如果导频符号为 10, 则学习速率可以取小于 0. 01的值 0. 009 ,这样最后一个导频的遗忘因子是不超 过 1的值 0. 98。
在方案 2 ,在快衰落频率选择性信道下, 算法要求能快速地跟踪信道的变 化, 导频图案相应地可以采用梳状或离散状, 初始值可以取常规法遗忘因子的 值以保证基于自适应遗忘因子信道估计算法的计算复杂度与基于常规遗忘因 子算法基本一致, 第 3时刻的导频的遗忘因子满足 λ„= ΐ。+0.01?7(1 + 0.1)" , 变 化的学习速率设为 0.01(1 + 0.1)" , 例如本例中初始值取为 0. 9 , 这样第六个导频 的遗忘因子值为小于 1的值 0. 98。 如果导频符号为 8 , 则学习速率因子可以取 小于 0. 01的值 0. 006, 这样最后一个导频的遗忘因子为不超过 1的值 0. 982。 如果导频符号为 10, 则学习速率因子可以取小于 0. 01的值 0. 0046 , 这样最后 一个导频的遗忘因子为不超过 1的值 0. 976。 这样不同时刻导频遗忘因子为指 数变化速率, 大于方案 1中的线性变化速率。
在方案 3 ,在快衰落频率逸择性信道下, 算法要求能快速地跟踪信道的变 化, 导频图案相应地可以采用梳状或离散状, 初始值可以取常规法遗忘因子的 值以保证基于自适应遗忘因子信道估计算法的计算复杂度与基于常规遗忘因 子算法基本一致, 第 时刻的导频的遗忘因子满足 1,, = 1。+ 0.0999 , 变化的 学习速率设为 0.0999f— 是导频符号的个数, 例如本例中初始值取为 0. 9 , 这样第六个导频的遗忘因子值为小于 1的值 0. 99。 如果导频符号为 8 , 最后一 个导频的遗忘因子仍为不超过 1的值 0. 99。 如果导频符号为 10 , 最后一个导 频的遗忘因于仍旧为不超过 1的值 0. 99。逸样不同时刻导频遗忘因子为指数变 化速率, 大于方案 1中的线性变化速率, 且最好一个时刻的导频符号的遗忘因 子值始终为 0. 99。
在方案 4 ,在快衰落频率选择性信道下,算法要求能快速地跟踪信道的变 化, 导频图案相应地可以采用梳状或离散状, 初始值可以取常规法遗忘因子的 值以保证基于自适应遗忘因子信道估计算法的计算复杂度与基于常规遗忘因 子 算 法 基 本 一 致 , 第 时 刻 的 导 频 的 遗 忘 因 子 满 足 λη = 1
Figure imgf000015_0001
本例中初始值取为 0. 9 , 这样第六个导频的遗忘因子值为小于 1的值 0. 99。 如 果导频符号为 8 , 则学习速率因子可以取小于 0. 01的值 0. 007 , 这样最后一个 导频的遗忘因子为不超过 1的值 0. 991。 如果导频符号为 10 , 则学习速率因子 可以取小于 0. 01的值 0. 0046 , 这样最后一个导频的遗忘因子为不超过 1的值 0. 978。 这样不同时刻导频遗忘因子的变化速率大于方案 1中的变化速率。
本发明的仿真结果采用方案 1 的自适应遗忘因子方法, 初始遗忘因子值 设为 0. 9 , 学习速率设为 0. 01。 各种遗忘因子方案的性能如图 7、 图 8和图 9 所示, 方案 A (RLS-C , RLS-T' , RLS - A, )和方案 B (RLS-C, RLS-T, RLS-A) 的性能比较如图 10所示。
(S)快速傅立叶 (FFT ) 变换
如图 3所示, 以上步骤求得时间 /时延域的信道估计的准确值, 但是信号 检测需要的是信道的时间 /频率域的状态信息, 因此, 对时间 I时延域的信道估 计值做 IFFT变换得到时间 /频率域信道估计值。
(9)插值算法
以上步骤得到时间 /频率域导频处的信道估计值,再通过插值算法得到数 据符号处的时间 /频率域的信道状态信息, 从而提供给信号检测模块进一步处 理, 本发明把数据的信道状态信息送給空时解码模块进行最大似然准则的空时 译码。
本发明的信道估计方法具有一定的通用性,可适应于一般 OFDM通信系统, 采用相同导频图案的单天线 OFDM系统在频率选择性慢衰落信 i 如 ITU信道, 即信道多径数目为 2 , 时延为 [0 260] ns , 功率 [0 -12. 77] dB, 各种信道估计算 法的 MSE仿真性能如图 11所示, BER性能如图 12所示, 自适应遗忘因子方案 仍旧采用方案 A (LS, , LMS' , RLS-C' , RLS-T' , RLS-A' )和方案 B (LS, LMS, RLS-C, RLS-T, RLS-A) , 仿真性能如图 13和 14所示, 可以发现方案 Β的性能 比方案 Α优越。 仿真结果表明, 基于自适应遗忘因子的 RLS算法和基于两步法 遗忘因子算法的性能明显好于基于固定遗忘因子的 RLS算法, 采用自适应遗忘 因子方案或两步法遗忘因子方案可以更好地反应信道的变化, 且具有更小的估 计误差和更好的性能。 图 15是根据本发明第 1优选实施例的信道估计设备的原理方框图 根据本发明第 1优选实施例的基于可变遗忘因子 RLS滤波的信道估计设 备包括: 初始估计单元(LS ), 用于根据提供的导频信息对信道进行初始估计, 得到所有导频处的信道的初始值; 逆快速傅立叶变换单元 (IFFT ), 用于对所 估计出的初始值进行逆快速傅立叶变换; RLS滤波器, 具有可变遗忘因子, 用 于对时变无线信道进行跟踪和估计, 其中所迷可变遗忘因子由遗忘因子设置单 元进行设置, 从而使得后一时刻导频符号的遗忘因子大于或等于前一时刻导频 符号的遗忘因子; 快速傅立叶变换单元(FFT ), 用于对递归最小平方滤波器的 输出进行快速傅立叶变换; 以及插值运算单元, 用于对傅立叶变换单元的输出 执行插值运算, 获得数据所对应的信道状态信息值。
如图 15所示, 接收机接收来的信号经过快速傅立叶变换(FFT )后送入 初始估计单元(LS ); 初始估计单元根据提供的导频信息对信道进行基于 LS的 初始信道估计, 得到所有导频处的信道的初始值; 信道的初始值经过逆快速傅 立叶变换单元( IFFT )后得到时域的初始信道状态信息; 把时域的初始信道状 态信息送入具有可变遗忘因子的 RLS滤波器进行处理, 得到准确的时域信道状 态信息; 将准确的时域信道状态信息送给快速傅立叶变换单元(FFT )后得到 频域的准确的信道状态信息; 再把频域的准确的信道状态信息送入插值单元进 行插值 , 得到数据符号所对应的频域的信道状态信息; 数据符号所对应的频域 的信道状态信息送入信号检测单元进行检测, 从而在接收机端恢复出发射机发 送的数据信息。
相应地, 上述各个步驟可以由对应的硬件单元来实现, 这里仅给出了示 例性的实例, 本领域普通技术人员完全可以根据以上描述构造出不同的硬件结 构。 可选地, 上述各个功能可以由单一的硬件单元来实现, 或者单一的功能可 以由多个硬件单元来实现。 这些具体的实施方式均应理解为包含在本发明的保 护范围内。 根据本发明的第 1优选实施例, 在传统的频域最小平方信道估计器的基 础上设计了一种时间 /时延域的基于可变遗忘因子 RLS滤波的信道估计方法和 设备, 通过对时变信道的跟踪与估计, 得到噪声作用下的信道状态信息, 从而 得到更加符合实际的信道估计值。根据本发明第 1优选实施例的信道估计方法 和设备与传统的信道估计方法和设备相比, 能够很好地跟踪和估计信道, 具有 收敛性好、 估计精度高等特点, 同时能够通过调整滤波器长度等参数来对信道 估计精度和计算复杂度进行折中和平衡, 从而获得更好的估计性能, 因此, 这 种信道估计方法和设备方便灵活, 很适合实际应用, 可以为第三代(3G )、 超 三代(B3G )、 第四代(4G )蜂窝移动通信和数字电视、 无线局域网 (WLAN )、 无线广域网 (WWAN )等系统的信道估计方案提供重要的理论依据和具体的实现 方法和设备。
- 第 2优选实施例
基于粒子滤波的信道估计方法和设备
首先, 对本发明第 2优选实施例的原理和设计思想进行详细说明。
1. 首先, 根据采用的通信系统是单天线系统还是多天线系统, 设置相应 的导频信息;
考虑所采用的通信系统是单天线系统还是多天线系统, 如果是单天线系 统, 对导频的设置没有任何特殊要求, 如果是多天线系统, 则在发送端发送的 不同天线及不同时刻构成的导频矩阵必须满秩, 因此这些导频可以是互相正交 也可以不正交, 这样在接收天线端可以很容易地对不同发送天线的导频符号进 行解耦和分离。 本发明采用的是一维空间的粒子滤波器, 所采用的信号处理方 法都是建立在一维空间的贝叶斯原理和一维空间的序贯蒙特卡罗法的基础上 的, 因此, 本发明针对信道信息是包含幅度和相位的复数信息, 把复数的实部 和虚部分离开来, 分别对实部和虚部进行各自的基于粒子滤波法的跟踪与估 计, 整个过程结束前, 再把信道实部和虚部信息合并成完整的信道信息。 采用 一维空间的信道实部和虛部分別处理的方法, 避免了原理极其复杂且运算量极 大的贝叶斯原理下联合概率密度问题的求解问题, 从而大大减小了分析复杂度 和运算量。
2.利用导频符号, 求出基于最小平方准则(LS)下的初始的信道估计值, 再把 LS估计得到的复数形式的信道信息分解成实部形式和虛部形式的信道信 息。
这个过程对单天线系统很简单, 就是在频域求出各导频符号对应的初始 信道估计值; 对多天线系统也比较简单, 就是在第 1步天线解耦的前提下求出 各导频符号对应的初始信道值。 这个初始信道值可以通过逆傅立叶变换(IFFT) 得到时域的信道信息, 在时域采用最小均方误差算法(LMS)或递归最小二乘算 法(RLS)得到时域信道估计值, 这个时域信道估计值再通过傅立叶变换 (FFT)而 得到频域的信道值, 这个信道信息的精度高于 LS算法得到的初始的信道值。 为减小计算复杂度,本发明的实施实例中采用 LS方法而不采用较为复杂的 LMS 或 RLS方法。
3. 利用得到的实部和虚部形式的初始的信道估计值, 进行粒子滤波的序 贯重要采样法的初始化设置, 初始化设置包括抽取随机样本, 即设置粒子的数 目、 粒子范围、 各个粒子所对应的权重;
初始化设置就是抽取一系列样本(粒子), 包括设置粒子的数目 、 粒子 范围、 各个粒子所对应的权重 1/ 每一个需要信道估计的时频块都需要 个 权重为 1/ 的粒子进行采样。 本发明重要采样法的信道粒子的初始值的设置不 同于已知信道为 AR模型的设置方法, 在已知信道为 AR模型的设置方法中, 后 面时刻信道粒子初始值是前面时刻信道粒子初始值的服从 AR模型的转换值。 因此本发明的信道粒子值的初始值是建立在 LS算法的基础上的, 它不需要任 何已知信道统计信息。序贯重要采样法的核心思想是根据贝叶斯原理对一系列 随机样本(粒子)所表示的先验概率和信道的当前量测值进行加权运算, 得到 先猃概率下的信道估计值。 这里信道的当前量测值就是第 2步估计处理的初始 信道估计值, 得到的后验概率下的信道估计值就是本发明所要求的准确的信道 估计值。 随机样本(粒子)的个数越多, 蒙特卡罗的特性和后验概率密度的函 数表示就越接近, 序贯重要采样法的性能就越接近于最优贝叶斯估计。 但是粒 子数目太多会导致计算复杂度增加, 而且粒子数目增加到一定的程度就逼近于 最优值, 合理的粒子数目一般通过比较计算机仿真结果和计算复杂度的情况下 得到。
4. 根据噪声方差得到以信道粒子值为条件的接收信号的条件概率密度, 即先验概率;
根 据 噪 声 估 计 得 到 噪 声 的 估 计 方 差 , 在 η{η, k]中, 表示 n时刻 :子载波的发送信号,
Figure imgf000019_0001
HO ]表示信道, η/ ]表示接收信号, ?7| ,Α;]表示噪声, ^表示第 Nr根发 送 天 线 上 的 发送 功 率 , 则 运 用 LS 估 计 方 法 得 到 式 子 τ , 因此可
Figure imgf000019_0002
以推导出噪声方差 ση 2|»] = -。
" SNR
5. 用上一步先验概率和当前的接收信号来更新每个粒子的权值, 即运用 贝叶斯公式得到以当前接收信号为条件的信道粒子值的后验概率密度, 然后再 对所有粒子的权值进行归一化 , 得到各个粒子的新的权重值;
本发明对插入导频的时频点进行信道估计,假设经过 OFDM解调之后 ί时 刻导频子载波上的已有的数据序列为 。 = 。, ... , :f; 根据噪声估计得到噪声 的估计方差,
Figure imgf000020_0001
率密度 ] h[) , 1=1, 2,…, Μ, Μ表示粒子总数, 即以信道粒子值为条件的接 收信号的条件概率密度;对于后验概率密度函数 |yto)的贝叶斯估计可以表 示为 P0 I y0 ) = '„ |yo:'- '); 为了能够递归的计算 p(h, I y0,)的贝叶斯 估计, 本发明利用序贯重要采样, 使用粒子和它们相对应的权重值表示要求的 后验概率密度函数是其中重要的思想, 即 (¾ ) /'), 具体的表
Figure imgf000020_0002
达式为 wX,^^^^,; τ(Α,μ, ,,;,)表示了当前系统的先验知识,如 果^ ^,Λ)越接近真实的后 概率函数, 粒子滤波器的性能就越好; 使用先 验重要函数的表式形式可以把 χ O'lW 1)的问题化简为 w[ = wi_]P(yi I h,) '再用接收信号和信道信息粒子构成的先验概率和当前的接收 信号来更新每个粒子的权值; 然后,再对所有粒子的权值进行归一化, 如 , 这样就得到各个粒子的新的权重值。 噪声方差在信道跟踪的初始
Figure imgf000020_0003
阶段设置为比理论值大的值, 在初步跟踪到信道之后, 设置为便于精确的跟踪 的值。 而且, 在慢时变信道中噪声方差设置为比理论值大的值, 在快时变信道 中,设置为比理论值小的值。 由于本发明中信道粒子值的初始值设置是根据 LS 算法得到的信道粗略估计值进行取样得到的, 因此本发明可以按顺序地得到观 测值, 进行在线推理, 因而当接收数据到达时, 就需要更新后验分布, 本发明 采用按顺序的方法, 不需要存储所有数据, 从而筒化了计算。
6. 根据滤波器退化检测公式, 性能低于门限值时, 进行重采样算法
M 1
利用粒子滤波器退化检测公式 Neff = ——进行检
1
Ar∑( ¾'))2) ∑(( ))2) 测, 为重采样前的初始粒子个数, 如果滤波器性能下降低于门限值, 即当 N,ff < Nlh ( Nlh < M), 进行重采样算法。 重采样不仅计算量很大, 而且如果使 用不当会导致系统性能大大降低, 因此没有必要在每一步都进行重采样, 当滤 波器性能下降至低于一个门限值时, 即 Ne r < NA时, 再进行重采样。 重采样门 限值 Neff满足 = Tir^—— = "ΛΓ-^——,其中 = Mw;]。 选取权重值 ∑( ∑(( ))2) 大的粒子进行重采样。 这里使用的重采样过程是不同于以往文献中基于信道模 型的(例如 AR模型), 而是基于接收信号的后验分布, 然后引入一定的扩展性, 对于变化的信道有不错的跟踪性能。 具体的步骤如下: 在第 5步中得到了所有 粒子的值和更新权值之后, 把这些离散的粒子看作是信道的离散概率密度分 布, 利用离散的积分方法, 把这概率密度转换成概率分布。 然后对概率分布的 概率轴进行 等分, 对重新分割的分布轴上进行分配粒子, 并且在分配粒子的 时候引入向外延拓的粒子, 即在重采样算法采样好的基础上再适当地人为扩大 粒子的取值范围, 以便适应信道的变化。
7. 进行加权运算, 即对所有信道粒子值和它们的概率密度来求出数学期 望值, 得到当前时刻的准确的信道估计值, 这些信道估计值为实部和虚部的形 式;
由于本发明 4巴信道的实部和虛部分开进行处理, 因此当粒子滤波器得到 信道粒子的权重后, 利用初信道粒子的初始值进行加权运算, 分别得到实部和 虛部的信道估计值, 即对所有信道粒子值和它们的概率密度(权重)来求出数 学期望值, 得到当前时刻实部和虚部的信道估计值, 再合并实部和虚部得到最
- M U
终复数形式的信道估计值,如式 I ,A] = Zwr/i¾I , :] + J'∑w ;'i , k , /和 w 分别表示信道实部和虚部在 ί时刻的粒子值, /表示 个粒子中的第 /个粒 子。 在权重更新过程中, 如果粒子滤波器出现退化现象, 则运用重采样后的信 道粒子值和权重进行加权运算, 即求出信道粒子值的数学期望值, 再进行实部 和虚部合并而得到复数形式的信道估计值。
8.返回到第 5步, 进行下一时刻的迭代运算。
由于粒子滤波是一种时间上递归的信号处理方法, 所以所有时刻的导频 符号对应的准确的信道估计值应通过不断的迭代运算得到, 因此返回第 5步是 为下一次信道估计做准备。
9. 根据以上步驟求出的各个时刻的导频符号的准确信道估计值,合并实 部和虚部的形式的信道估计值, 使信道估计值为能体现幅度和相位信息的复数 形式; 最后, 利用插值的方法得到数据符号的信道信息值, 从而为下一环节的 数据检测或译码提供较为准确的信道信息。 以下, 给出了本发明第 2优选实施例的具体应用示例。
如图 16 所示, 结合实例对本发明的技术方案作进一步描述:
以一个 2 发 2 收的 MIM0- OFDM 系统为例, 信道带宽为 6. 4MHz , 子载 波数为 32。 信道为多径瑞利衰落信道, 信道多径数目为 2 , 时延为 [O ns, 260 ns] , 功率延迟分布服从指数衰减。 数据传输时, 发送天线每隔 6 个 OFDM符 号插入连续两个符号的导频序列,不同天线上的导频序列互相正交。 由于在每 一个发射和每一个接收天线之间存在一条信道, 所以每个信道估计结果实际包 含对 4 个信道的估计。
1. 首先在导频上安排时间上连续排列的块状导频, 这种分配导频的方法 应用于 2发 2收的每个发送天线上。 如图 16所示, 假设有天线 1和天线 2 , 这 种导频分配具体如下: 在时刻 1 , 天线 1上面分配全为 1的导频数据, 天线 2 上也是分配同样为 1的块导频; 在时刻 2 , 天线 1上面仍旧分配全为 1的导频 数据, 天线 2上则分配全为 -1的导频数据。 导频分配之后加入数据段的数据, 相隔 6个时刻之后再加入如上的导频, 分配方法也是一样的。 这么做的目的在 于使接收端可以利用连续导频的数据和信道慢衰落的特点, 进行天线相关的数 据分离, 从而简化了粒子滤波器的计算复杂度。 上面不同天线导频的分配方式 是互相正交的, 也可采样不是正交的形式, 例如在时刻 1 , 天线 1上面分配全 为 1的导频数据, 天线 2上也是分配同样为 2的块导频; 在时刻 2 , 天线 1上 面仍旧分配全为 1的导频数据, 天线 2上则分配全为 1的导频数据, 这样也可 以满足本方明的要求。 因此, 本发明的对多天线系统导频符号的要求是在由 Ντ 个发送天线、 Λτ个连续时刻上构成的导频矩阵是满秩矩阵。
2. 在接收端, 在接收导频的时候, 每个接收天线利用连续的两个导频时 刻的导频数据, 进行简单的加减运算便可以得到发送天线 1和发送天线 2的信 道分离的数据。 具体操作可以如下: 接收天线 1上导频时刻 1的数据加上接收 天线 1上导频时刻 2的数据可以得到 2倍的经过发送天线 1到接收天线 1信道 Hu的发送数据。 接收天线 1上导频时刻 1的数据减去接收天线 1上导频时刻 2的数据可以得到 2倍的经过发送天线 2到接收天线 1信道 H2,的发送数据。 接收天线 2上导频时刻 1的数据加上接收天线 1上导频时刻 2的数据可以得到 2倍的经过发送天线 2到接收天线 2信道 H22的发送数据。 接收天线 2上导频 时刻 1的数据减去接收天线 2上导频时刻 2的数据可以得到 2倍的经过发送天 线 1到接收天线 2信道 H12的发送数据。
3 . 根据四个导频信道相关数据的分离结果, 可以对于每个信道 ( H1 H21,H12, H22 )进行粒子滤波信道估计。 这种估计是在所有子载波上进行 的, 在对每个子载波的每个信道估计的时候, 由于信道信息是复数形式的, 如 果直接进行粒子滤波器的应用将会涉及到二维的粒子滤波过程, 这和二维随机 变量的联合估计相关, 复杂度较高。 但是由于导频数据的设计是实数, 所以在 接收端只要将第 2步得到的数据进行进一步的实虚部分离, 实虚部分离之后得 到的两部分信息分别是导频信号经过一个信道的实部参数构成的虚拟信道和 虚部参数构成虚拟信道的接收结果。 对于这样一个结果, 如果知道导频的数据 和接收噪声, 那么就可以估计出这两种虚拟接收信号的概率分布情况。
4. 通过前三步, 本发明可以得到 2发 2收系统每 2个连续导频块的每个 子载波的 4个信道的 8个虚拟接收信号。对于每个虛拟接收信号进行粒子滤波。 首先利用 LS算法对这些信道有个粗略的估计, 以这个粗略的估计为基础, 初 始化粒子滤波器的初始粒子。 确定粒子滤波器的粒子范围和粒子个数, 然后把 每个粒子的权值设为 1 / (其中 是粒子总数), 本例中粒子数目可以设置为 50 , 粒子的步长可设置为 0. 05 , 正负两个方向取粒子值, 则可以得到 50个信 道粒子值; 若取 100个粒子值, 则粒子的步长可设置为 0. 025 , 正负两个方向 取粒子值, 则可以得到 100个信道粒子值。 注意, 粒子数目的增加会导致估计 精度增加, 同时也导致计算复杂度增加, 但是当粒子数目增加一定的程度, 估 计精度很难提高, 因此必须在估计精度和计算复杂度之间取平衡值, 本例中, 根据计算机仿真结果, 粒子值为 100时即可以达到很好的性能。
5 . 根 据 噪 声 估 计 得 到 噪 声 的 估 计 方 差 , 在 Y[n, k] + η[η, k] 中, : T[",/c]表示 n时刻 k子载波的发送信
Figure imgf000024_0001
号, ί ]表示信道, η ]表示接收信号, 表示噪声, ^表示第^根 发送天线 上 的 发送功 率 , 则 运 用 LS 估 计 方 法得到 式 子
H[n,k] , 因此可
Figure imgf000024_0002
以推导出噪声方差 [",Α] =」^"。 考虑到粒子滤波器的序贯统计特性, 粒子
SNR 滤波器本身就有较强的抗噪声能力, 因此在实际的应用中, 对噪声的估计并不 用非常精确, 可以根据不同的情况设置相应的噪声方差。 在(7„ η2|>7^] = ~^中,
SNR
Λ¾?=10 时, 发送天线上的发送功率为 时, 噪声方差为 0.1, 在信道跟踪的 初始阶段,可以根据噪声方差的理论值如 0.1 ,把噪声方差设置较大的值如 0.3, 以便于系统的跟踪。 在初步跟踪到信道之后, 可以把这个噪声方差设置为适当 的值如 0.1, 以便精确的跟踪。 一般在慢时变信道中, 噪声方差值可以设置得 大一些, 本发明的估计算法仍然能够快速跟踪和收敛; 在快时变信道中, 噪声 方差应该设置的小一点, 以更好地跟踪信道的变化。 噪声方差的灵活设置和使 用,可以使本发明提出的基于粒子滤波的信道估计方法适应于各种不同情况下 的信道。
6. 用噪声的估计方差得到条件为信道信息粒子值的接收信号分布的条件 分布 Ρθ, ΙΑ)。 如式 ^7^' Wff[", k [", k] + k] '如果噪声 l ,W的方 差已知,即噪声为服从高斯分布 N(o, ση 2)的白噪声, η ]为接收信号, 为已知的导频信号, 则信道频域传输函数 的条件概率密度 ρθ, 1 h,)就可 以推导出来, 因此粒子值的先验概率就可以确定。 用每个信道估计粒子的接收 信号分布来更新每个粒子的权值, 即^^^-^5!" ^^^'"))。 然后再对所有 粒子的权值进行归一化, 即 =
Figure imgf000025_0001
7. 利用更新后的粒子和对应的权值估计出这个虛拟信道的信道估计, p( t \ yt) « 同样的操作应用于所有的所有接收天线的子载波的
Figure imgf000025_0002
所有 8个虛拟信道, 得到这个时刻的所有子载波的导频符号的信道估计值。
8.利用粒子滤波器退化检测公式 Neff = N M = -γ- ——进行检
- ∑( ¾0)2) ∑( 测, 为重采样前的初始粒子个数, 如果滤波器性能下降低于门限值, 即当
Nsff < Nlh ( Nlh < M), 进行重采样算法。 重采样不仅计算量很大, 而且如果使 用不当会导致系统性能大大降低, 因此没有必要在每一步都进行重采样, 当滤 波器性能下降至低于一个门限值时, 即 N < Nrt时, 再进行重采样。 重采样门 限值 Ne#满足 - ——, 其中 ')。 本发明在序 ∑(( ))2)
Figure imgf000025_0003
贯重要采样步骤后加一个重采样步骤, 其算法流图如图 17所示,序贯重要采样 法的原理如图 18所示, 选取权重值大的粒子进行重采样,重采样算法采用基于 概率分布的重采样算法, 如图 19所示, 表示重采样前粒子的个数, 对先验概 率 )进行积分运算, 则得到它的概率分布函数, 因此可以发现先验概率 值越大的粒子所对应的概率分布函数值越大, 即图 19 中 | )函数的斜率 越大。
9. 最后, 据利用粒子滤波器估计出来的导频符号的信道估计值, 运用 插值的方法得到数据符号的信道估计值, 从而为下一环节的数据检测或译码模 块提供信道信息。
本发明采用基于粒子滤波的信道估计方法, 在贝叶斯原理和蒙特卡罗采 样原理的基础上采用序贯重要采样法和基于概率分布函数的重采样法对无线 信道进行跟踪和估计, 仿真结果表明这种信道估计方法好于传统的信道估计方 法, 如图 20和图 21表示。 图 20表明基于粒子滤波的信道估计方法的最小均 方误差 (MSE )性能好于基于扩展卡尔曼(EKF )的信道估计方法和基于最小平 方的信道估计方法(LS ), 且随着粒子数目的增加, MSE的性能也相应变好, 但 是粒子数目增加到一定的程度, MSE的性能提高不是很明显。 图 21表明采用基 于粒子滤波的信道估计方法的系统的比特误码率 ( BER )性能好于采用扩展卡 尔曼(EKF )信道估计方法和最小平方的信道估计方法(LS ) 的系统。 本发明 也适用于单发单收(SIS0 )、 单发多收(SIM0 )、 多发单收(MIS0 )的通信系统, 具有很广的适用范围。 图 22是根据本发明第 2优选实施例的信道估计设备的原理方框图 根据本发明第 2优选实施例的基于粒子滤波的信道估计设备包括: 导频 信息设置单元(预处理单元;), 用于根据采用的通信系统是单天线系统还是多 天线系统, 对导频信息进行处理; 初始估计单元(LS ), 用于 ■据已知导频信 息对信道进行最小平方估计, 得到各个导频处信道估计的初始值; 粒子初始化 设置单元, 用于设置粒子范围和粒子数目, 并为每个粒子分配初始权重; 重要 采样单元, 用于以信道粒子值为条件的接收信号的条件概率密度作为先验概 率, 与当前接收信号构造贝叶斯模型, 计算出以当前接收信号为条件的信道粒 子值的后验概率密度; 以及对所有粒子的权值进行归一化, 并进行加权运算; 重采样单元, 用于根据判断条件对粒子进行重采样, 其中重要采样单元和重采 样单元执行时间递归操作; 以及插值运算单元, 用于执行内插运算, 得到所有 发送数据符号的信道估计值, 从而完成信道状态信息的估计。
如图 22所示, 接收机接收来的信号经过快速傅立叶变换单元(FFT )后 送入预处理单元; 预处理单元根据单天线系统或多天线系统的导频结构, 进行 简单的加减运算便可以得到发送天线的信道分离的数据; 初始估计单元(LS ) 根据提供的导频信息对信道进行基于 LS 的初始信道估计, 利用导频信息求出 基于最小平方准则 LS下的初始信道估计值,再把 LS估计得到的复数形式的信 道信息分解成实部形式和虚部形式的信道信息; 此时, 粒子初始化设置单元设 置粒子范围和粒子数目, 并为每个粒子分配初始权重; 然后, 重要采样单元以 接收信号作为观测值,以信道粒子值为条件的接收信号的条件概率密度作为先 验概率, 结合当前接收信号构造贝叶斯模型, 计算出以当前接收信号为条件的 信道粒子值的后验概率密度; 然后, 再对所有粒子的权值进行归一化, 这样就 得到各个粒子的新的权重值; 运用滤波器退化检测公式对性能进行判断, 性能 低于门限值时, 进行重采样算法; 根据以上步驟求出的备个时刻的导频符号的 准确信道估计值, 合并实部和虚部的形式的信道估计值, 使信道估计值为能体 现幅度和相位信息的复数形式; 最后利用插值的方法得到数据符号的信道信息 值, 从而为下一环节的数据检测或译码提供较为准确的信道信息。
相应地, 上述各个步骤可以由对应的硬件单元来实现, 这里仅给出了示 例性的实例, 本领域普通技术人员完全可以根据以上描述构造出不同的硬件结 构。 可选地, 上述各个功能可以由单一的硬件单元来实现, 或者单一的功能可 以由多个硬件单元来实现。 这些具体的实施方式均应理解为包含在本发明的保 护范围内。 根据本发明的第 2优选实施例, 提供了一种基于粒子滤波的信道估计方 法和设备, 克服了现有技术中的不足, 使其在不知道信道统计信息的环境下具 有估计性能稳健、 鲁棒性强、 抗噪声的能力强的特点, 且易于实现。 相比于 K. Huber及 W. H Chin等人提出的几种基于粒子滤波的信道估计方法, 本发明的第 2优选实施例更加符合实际信道未知的特点 , 无需已知任何信道统计信息, 且 提出了更好的重采样算法。 相比于求解此类问题的递推滤波算法(EKF、 MGEKF 等), 本发明的第 2优选实施例具有估计精度高、 收敛时间短且性能稳定的特 点。 另外, 本发明的第 2优选实施例通过天线解耦、 导频信息的实部和虛部分 解,把二维甚至多维空间中基于贝叶斯原理的条件概率求解问题化筒为一维空 间中条件概率的求解问题, 且运用基于概率分布函数值的重采样算法大大简化 了粒子滤波算法的计算复杂度。 而且本发明的第 2优选实施例采用灵活的噪声 估计算法, 从具有抗噪声能力强的特点。 因此, 随着计算存储成本的下降, 基 于粒子滤波的信道估计算法越来越受到重视, 本发明的第 2优选实施例提出的 基于粒子滤波的信道估计算法可以通过硬件来实现, 具有一定的工程应用价 值。
- 第 3优选实施例
选择性信道估计方法和设备
首先, 对本发明第 3优选实施例的原理和设计思想进行详细说明。 1. 首先, 判断信道属性;
考虑到无线信道的复杂性, 无线信道可能处于线性信道、 非线性信道、 高斯白噪声的信道、 非高斯噪声的信道, 移动终端具有可移动性的特点且运动 速度可能造成多普勒频移, 实际接收机在不同时刻所处的无线环境是十分复杂 的, 因此, 本发明定义线性且高斯白噪声的信道为非恶劣信道环境, 定义非线 性信道或非高斯噪声的信道为恶劣信道环境。接收机首先判断信道是恶劣信道 环境还是非恶劣信道环境。
2. 根据信道环境而采用选择性信道估计方法;
根据信道是否处于恶劣信道环境而采用选择性的信道估计方法, 恶劣信 道环境采用基于粒子滤波的信道估计方法, 非恶劣信道环境采用基于可变遗忘 因子 RLS信道估计方法。
3. 进行插值算法;
考虑到信道估计方法得到是导频符号的信道状态信息, 因此通过插值算 法得到数据符号的信道状态信息。
4. 进行信号检测算法;
运用信号检测算法, 结合插值算法得到的数据符号的信道状态信息, 得 到发射机的发送数据信息。
5. 计算比特误码率;
计算上述步骤得到的发送数据信息的比特误码率。
6. 判断误码率指标。
判断误码率指标是否达到要求, 如果误码率达到系统设计要求, 则表示 接收机接收一帧符号结束, 如果误码率达不到系统设计要求, 则表示接收机接 收的一帧符号不能作为最终接收数据, 这时运用基于粒子滤波的信道估计方法 重新进行信道估计, 其中为提高基于粒子滤波信道估计的精度, 通过加大粒子 的采样数目来提高信道估计的精度。 以下, 给出了本发明第 3优选实施例的具体应用示例。
如图 23 所示, 结合实例对本发明的技术方案作进一步描述:
在结合基于粒子滤波的信道估计方法和基于可变遗忘因子 RLS信道估计 方法的基 上,考虑到无线信道的复杂性,而采用一种可选择性信道估计方法。 1.首先判断无线信道是否为恶劣信道。 如果信道是线性且高斯白噪声的 信道, 则判断无线信道为非恶劣信道环境; 如果信道是非线性信道或非高斯噪 声的信道, 则判断信道为恶劣信道环境。
2.采用基于粒子滤波的信道估计方法或基于可变遗忘因子 RLS 的信道估 计方法。信道是恶劣信道环境则采用基于粒子滤波的信道估计方法; 信道是非 恶劣信道环境则采用基于可变遗忘因子 RLS的信道估计方法。
3.运用插值算法。 运用插值算法, 得到数据符号所对应的信道状态信息。
4.运用信号检测算法。 运用信号检测算法, 得到发射机的发送数据信息。
5.计算一帧数据符号的比特误码率( BER )。
6.判断误码率指标。 如果误码率指标达到性能要求则表示接收一帧数据 符号的过程结束; 如果误码率指标达不到性能要求, 则运用基于粒子滤波的信 道估计方法重新进行信道估计, 其中信道估计的精度通过设置粒子的采样数目 来控制, 直至误码率性能达到系统性能要求。 图 24是根据本发明第 3优选实施例的信道估计设备的原理方框图 根据本发明第 3优选实施例的选择性信道估计设备包括: 判断信道属性 单元, 判断信道是否为恶劣信道环境; 基于粒子滤波的信道估计器, 用于估计 恶劣信道环境的信道; 基于可变遗忘因子 RLS滤波的信道估计器, 用于估计非 恶劣信道环境的信道; 误码率指标判断单元, 用于判断误码率是否满足性能要 求。
如图 24所示, 接收机接收来的信号经过快速傅立叶变换单元(FFT )后, 对信道属性进行判断, 判断信道属性单元判断信道是否为恶劣信道环境; 信道 是恶劣信道则采用基于粒子滤波的信道估计器, 信道是非恶劣信道则采用基于 可变遗忘因子 RLS的信道估计器; 信道估计器完成对信道状态信息的估计后, 把信道状态信息送给插值单元进行插值, 得到数据符号所对应的信道状态信 息; 将数据符号所对应的信道状态信息送给数据检测单元, 运用信号检测算法 得到发射机的发送数椐信息; 计算一帧发送数据信息的误码率; 误码率指标判 断单元判断误码率是否满足系统要求, 如果误码率指标达到性能要求则表示接 收一桢数据符号的过程结束, 如果误码率指标达不到性能要求, 则运用基于粒 子滤波的信道估计器重新进行信道估计, 其中通过增大采样的粒子数目来提高 信道估计器的估计精度。
相应地, 上述各个步骤可以由对应的硬件单元来实现, 这里仅给出了示 例性的实例, 本领域普通技术人员完全可以根据以上描述构造出不 的硬件结 构。 可选地, 上述各个功能可以由单一的硬件单元来实现, 或者单一的功能可 以由多个硬件单元来实现。 这些具体的实施方式均应理解为包含在本发明的保 护范围内。

Claims

权 利 要 求
1、 一种基于可变遗忘因子递归最小平方 (RLS ) 滤波的信道估计方法, 包括:
初始估计步骤, 根据提供的导频信息对信道进行初始估计, 得到所有导 频处的信道的初始值;
逆傅立叶变换步骤, 对估计出的信道初始值进行逆傅立叶变换; 跟踪和估计步骤, 运用基于可变遗忘因子的递归最小平方滤波, 对时变 无线信道进行跟踪和估计;
傅立叶变换和插值步驟, 通过傅立叶变换和插值算法得到数据所对应的 信道状态信息值,
其中所述可变遗忘因子是变化的, 后一时刻导频符号的遗忘因子大于或 等于前一时刻导频符号的遗忘因子。
2、根据权利要求 1所迷的基于可变遗忘因子递归最小平方滤波的信道估 计方法, 其特征在于:
在所述初始估计步骤中, 使用导频信息通过最小平方算法, 获取频域上 导频处的信道状态信息的初始值。
3、根据权利要求 1所述的基于可变遗忘因子递归最小平方滤波的信道估 计方法, 其特征在于:
在所述逆傅立叶变换步驟中, 通过快速逆傅立叶变换得到时间 I时延域的 未考虑噪声作用下的信道状态信息。
4、根据权利要求 1所述的基于可变遗忘因子递归最小平方滤波的信道估 计方法, 其特征在于:
在所述跟踪和估计步骤中, 以当前时刻未考虑噪声作用下的信道状态信 息作为期望值, 以当前时刻和前几个时刻的信道状态信息为输入值, 再利用递 归最小平方滤波, 跟踪和估计当前时刻的考虑噪声作用下的准确的信道状态信 息。
5、根据权利要求 1或 4所述的基于可变遗忘因子递归最小平方滤波的信 道估计方法, 其特征在于:
在跟踪和估计的过程中采用可变遗忘因子,再在时间 I时延域进行递归运 算, 得到考虑噪声作用的准确的信道状态信息。
6、根据权利要求 1所述的基于可变遗忘因子递归最小平方滤波的信道估 计方法, 其特征在于:
在所述傅立叶变换和插值步骤中, 执行快速傅立叶变换, 得到时间 /频率 域的信道传输函数; 以及通过插值算法得到数据符号所对应的信道传输函数, 并把数据符号所对应的传输函数送给信号检测模块进行检测或译码。
7、根据权利要求 2所述的基于可变遗忘因子递归最小平方滤波的信道估 计方法, 其特征在于:
所述导频信息为结构是块状的导频符号, 每个导频符号包括所有子载波 上的时频块, 导频的帧结构采用在一帧数据前面加入若干个导频符号的帧结 构, 在多天线系统中不同发射天线在同一时刻的导频是互相正交的。
8、根据权利要求 7所述的基于可变遗忘因子递归最小平方滤波的信道估 计方法, 其特征在于:
所述导频符号, 如果采用块状导频结构从而在一帧中时域上不同时刻点 的导频符号具有不同的权系数, 离数据符号越近的导频符号的权系数越大; 所 述导频符号如果采用梳状导频、 离散导频结构, 可设置不同的权系数, 且随着 时刻的增加, 导频符号的权系数越来越大。
9、根据权利要求 1所述的基于可变遗忘因子递归最小平方滤波的信道估 计方法, 其特征在于:
所述可变遗忘因子, 其变化采用两步法可变遗忘因子方案, 给前几个导 频设置较小的遗忘因子值而给后几个导频设置较大的遗忘因子值, 给后几个导 频符号设置高的遗忘因子值, 这样使通信系统更多地依靠前次信道估计提供的 信道的统计信息。
10、 根据权利要求 1 所述的基于可变遗忘因子递归最小平方滤波的信道 估计方法, 其特征在于:
所述可变遗忘因子, 其变化采用自适应遗忘因子方案, 具体为: 给不同 位置的导频符号自适应地分配遗忘因子值, 给处于第一个位置处的导频符号分 配一个初始遗忘因子和一个学习速率, 这样, 除第一个位置外的导频就根据遗 忘因子的学习速率来不断更新遗忘因子的值, 处于不同时刻的导频符号具有不 同的遗忘因子值, 从而更好地跟踪和估计时变信道的变化。
11、根据权利要求 10所迷的基于可变遗忘因子递归最小平方滤波的信道 估计方法, 其特征在于:
在慢衰落频率选择性信道下, 自适应遗忘因子的学习速率采用固定的学 习速率, 首先给第一个导频符号的遗忘因子设置一个初始值, 初始值取常规法 遗忘因子的值以保证计算复杂度与基于常规遗忘因子算法一致, 再给除第一个 导频符号外的导频符号设置一个固定的学习速率, 从而使不同的导频符号均具 有不同的遗忘因子值。
12、根据权利要求 10所述的基于可变遗忘因子递归最小平方滤波的信道 估计方法, 其特征在于:
在快衰落频率选择性信道下, 自适应遗忘因子的学习速率采用变化的学 习速率, 首先给第一个导频符号的遗忘因子设置一个初始值, 初始值取常规法 遗忘因子的值以保证计算复杂度与基于常规遗忘因子算法基本一致,再给除第 一个导频符号外的导频符号设置一个随时间不断变化的学习速率, 学习速率采 用自变量为时间的线性函数、 指数函数或其它映射关系, 从而使不同的导频符 号均具有不同的遗忘因子。
13、根据权利要求 10所述的基于可变遗忘因子递归最小平方滤波的信道 估计方法, 其特征在于:
遗忘因子根据导频符号的个数设置学习速率。
14、 根据权利要求 1 所述的基于可变遗忘因子递归最小平方滤波的信道 估计方法, 其特征在于:
通过调整递归最小平方滤波的.长度来控制所述信道估计方法的估计精度 和计算复杂度, 通过理论分析和计算机仿真找到最优或次最优的递归最小平方 滤波长度, 从而使所述信道估计方法的估计精度和复杂度达到折中和平衡。
15、 一种基于可变遗忘因子递归最小平方滤波的信道估计设备, 包括: 初始估计单元, 用于根据提供的导频信息对信道进行初始估计, 得到所 有导频处的信道的初始值;
逆傅立叶变换单元, 用于对估计出的信道初始值进行逆傅立叶变换; 递归最小平方滤波器, 具有可变遗忘因子, 用于对时变无线信道进行跟 踪和估计, 其中所述可变遗忘因子是变化的, 后一时刻导频符号的遗忘因子大 于或等于前一时刻导频符号的遗忘因子; 傅立叶变换单元, 用于对递归最小平方滤波器的输出进行傅立叶变换; 以及
插值运算单元, 用于对傅立叶变换单元的输出执行插值运算, 获得数据 所对应的信道状态信息值。
16、根据权利要求 15所述的基于可变遗忘因子递归最小平方滤波的信道 估计设备, 其特征在于:
所述初始估计单元使用导频信息通过最小平方算法, 获取频域上导频处 的信道状态信息的初始值。
17、根据权利要求 15所述的基于可变遗忘因子递归最小平方滤波的信道 估计设备, 其特征在于:
所述逆傅立叶变换单元通过快速逆傅立叶变换得到时间 /时延域的未考 虑噪声作用下的信道状态信息。
18、根据权利要求 15所述的基于可变遗忘因子递归最小平方滤波的信道 估计设备, 其特征在于:
所述递归最小平方滤波器以当前时刻未考虑噪声作用下的信道状态信息 作为期望值, 以当前时刻和前几个时刻的信道状态信息为输入值, 再利用递归 最小平方滤波, 跟踪和估计当前时刻的考虑噪声作用下的准确的信道状态信
19、 根据权利要求 15或 18所述的基于可变遗忘因子递归最小平方滤波 的信道估计设备, 其特征在于:
所述递归最小平方滤波器采用可变遗忘因子,再在时间 /时延域进行递归 运算, 得到考虑噪声作用的准确的信道状态信息。
20、根据权利要求 15所述的基于可变遗忘因子递归最小平方滤波的信道 估计设备, 其特征在于:
所述傅立叶变换单元执行快速傅立叶变换, 得到时间 /频率域的信道传输 函数; 以及
所述插值运算单元通过插值算法得到数据符号所对应的信道传输函数, 并把数据符号所对应的传输函数送给信号检测模块进行检测或译码。
21、根据权利要求 16所述的基于可变遗忘因子递归最小平方滤波的信道 估计设备, 其特征在于: 所述导频信息为结构是块状的导频符号, 每个导频符号包括所有子载波 上的时频块, 导频的帧结构采用在一帧数据前面加入若干个导频符号的帧结 构, 在多天线系统中不同发射天线在同一时刻的导频是互相正交的。
22、根据权利要求 21所述的基于可变遗忘因子递归最小平方滤波的信道 估计设备, 其特征在于:
所述导频符号, 如果采用块状导频结构从而在一帧中时域上不同时刻点 的导频符号具有不同的权系数, 离数据符号越近的导频符号的权系数越大; 所 述导频符号如果采用梳状导频、 离散导频结构, 可设置不同的权系数, 且随着 时刻的增加, 导频符号的权系数越来越大。
23、根据权利要求 15所述的基于可变遗忘因子递归最小平方滤波的信道 估计设备, 其特征在于:
所述可变遗忘因子, 其变化采用两步法可变遗忘因子方案, 给前几个导 频设置较小的遗忘因子值而给后几个导频设置较大的遗忘因子值, 给后几个导 频符号设置高的遗忘因子值, 这样使通信系统更多地依靠前次信道估计提供的 信道的统计信息。
24、根据权利要求 15所述的基于可变遗忘因子递归最小平方滤波的信道 估计设备, 其特征在于:
所述可变遗忘因子, 其变化采用自适应遗忘因子方案, 具体为: 给不同 位置的导频符号自适应地分配遗忘因子值, 给处于第一个位置处的导频符号分 配一个初始遗忘因子和一个学习速率, 这样, 除第一个位置外的导频就根据遗 忘因子的学习速率来不断更新遗忘因子的值, 处于不同时刻的导频符号具有不 同的遗忘因子值, 从而更好地跟踪和估计时变信道的变化。
25、根据权利要求 24所述的基于可变遗忘因子递归最小平方滤波的信道 估计设备, 其特征在于:
在慢衰落频率选择性信道下, 自适应遗忘因子的学习速率采用固定的学 习速率, 首先给第一个导频符号的遗忘因子设置一个初始值, 初始值取常规法 遗忘因子的值以保证计算复杂度与基于常规遣忘因子算法一致, 再给除第一个 导频符号外的导频符号设置一个固定的学习速率, 从而使不同的导频符号均具 有不同的遗忘因子值。
26、根据权利要求 24所述的基于可变遗忘因子递归最小平方滤波的信道 估计设备, 其特征在于:
在快衰落频率选择性信道下, 自适应遗忘因子的学习速率采用变化的学 习速率, 首先给第一个导频符号的遗忘因子设置一个初始值, 初始值取常规法 遣忘因子的值以保证计算复杂度与基于常规遗忘因子算法基本一致, 再给除第 一个导频符号外的导频符号设置一个随时间不断变化的学习速率, 学习速率采 用自变量为时间的线性函数、 指数函数或其它映射关系, 从而使不同的导频符 号均具有不同的遗忘因子。
27、根据权利要求 24所述的基于可变遗忘因子递归最小平方滤波的信道 估计设备, 其特征在于:
遗忘因子根据导频符号的个数设置学习速率。
28、根据权利要求 15所述的基于可变遗忘因子递归最小平方滤波的信道 估计设备, 其特征在于:
通过调整递归最小平方滤波器的滤波长度来控制所述信道估计设备的估 计精度和计算复杂度, 通过理论分析和计算机仿真找到最优或次最优的递归最 小平方滤波器的滤波长度, 从而使所述信道设备的估计精度和复杂度达到折中 和平衡。
29.一种基于粒子滤波的信道估计方法, 包括:
初始估计步骤, 利用导频信息进行初始信道估计;
粒子滤波步骤, 在初始信道值的基础上利用粒子滤波得到准确的信道估 计,
其中所述粒子滤波步驟包括以下子步骤:
后驗概率密度计算步骤, 以信道粒子值为奈件的接收信号的 条件概率密度作为先验概率, 与当前接收信号构造贝叶斯模型, 计 算出以当前接收信号为奈件的信道粒子值的后验概率密度;
归一化和加权步驟, 对所有粒子的权值进行归一化并进行加 权运算, 从而得到准确的信道估计值。
30.根据权利要求 29所述的基于粒子滤波的信道估计方法, 其特征在于 还包括:
导频信息设置步骤,根据采用的通信系统是单天线系统还是多天线系统, 设置相应的导频信息。
31.根据权利要求 29或 30所述的基于粒子滤波的信道估计方法, 其特征 在于:
在所述初始估计步骤中, 利用导频信息求出基于最小平方准则 LS下的初 始信道估计值, 再把 LS估计得到的复数形式的信道信息分解成实部形式和虛 部形式的信道信息。
32.根据权利要求 29或 30所述的基于粒子滤波的信道估计方法, 其特征 在于:
在所述粒子滤波步骤中, 利用得到的实部和虚部形式的初始信道估计值, 进行粒子滤波的序贯重要采样法的初始化设置, 初始化设置包括抽取随机样 本, 即设置粒子的数目、 粒子范围、 各个粒子所对应的权重。
33.根据权利要求 29或 30所述的基于粒子滤波的信道估计方法, 其特征 在于:
针对当前时刻,
在所述后验概率密度计算步骤中, 根据噪声方差得到以信道 粒子值为条件的接收信号的条件概率密度, 即先验概率; 用该先验 概率和当前的接收信号来更新每个粒子的权值, 即运用贝叶斯公式 得到以当前接收信号为条件的信道粒子值的后验概率密度; 以及 在所述归一化和加权步骤中 ,对所有粒子的权值进行归一化, 得到各个粒子的新的权重值; 根据滤波退化检测公式, 性能低于门 P艮值时, 进行重采样; 进行加权运算, 即对所有信道粒子值和它们 . 的概率密度来求出数学期望值, 得到当前时刻的准确的信道估计 值, 这些信道估计值为实部和虛部的形式,
针对下一时刻, 执行所述后验概率密度计算步骤与所述归一化和加权步 骤的迭代运算。
34.根据权利要求 33所述的基于粒子滤波的信道估计方法, 其特征在于: 根据各个时刻的导频符号的准确信道估计值, 合并实部和虚部的形式的 信道估计值, 使信道估计值为能体现幅度和相位信息的复数形式;
利用插值的方法得到数据符号的信道信息值, 从而为下一环节的数据检 测或译码提供较为准确的信道信息。
35.根据权利要求 30所述的基于粒子滤波的信道估计方法,其特征在于: 当天线为多天线系统时, 不同天线不同时刻构成的导频矩阵为满秩矩阵, 对不同发送天线的导频符号进行解耦和分离, 把多天线系统的信道估计问题简 化为单天线系统的信道估计问题。
36.根据权利要求 31所述的基于粒子滤波的信道估计方法,其特征在于: 导频符号设计为实数, 把信道信息分解为实部和虛部, 分別对实部和虛 部进行基于粒子滤波的信道估计, 在信道估计过程结束前, 再把信道信息的实 部和虚部进行合并, 从而得到完整的信道信息的估计值。
37.根据权利要求 33所述的基于粒子滤波的信道估计方法,其特征在于: 所述噪声方差在慢时变信道中在信道跟踪的初始阶段设置为比理论值大 的值, 在初步跟踪到信道之后, 设置为便于精确跟踪的值。
38.根据权利要求 33所述的基于粒子滤波的信道估计方法,其特征在于: 所述噪声方差在快时变信道中在信道跟踪的初始阶段设置为比理论值小 的值, 在初步跟踪到信道之后, 设置为便于精确跟踪的值。
39.根据权利要求 33所述的基于粒子滤波的信道估计方法,其特征在于: 所进行的重采样基于所述以当前接收信号为条件的信道粒子值的后验概 率密度, 利用离散的积分将概率密度转化为概率分布, 然后对概率分布的概率 轴进行等分, 在重新分割后的分布轴上分配粒子,并且在分配粒子的时候引入 向外延拓的粒子, 以适应信道的变化。
40.根据权利要求 39所述的基于粒子滤波的信道估计方法,其特征在于: 在粒子滤波的性能低于门限值时, 进行重采样, 并运用重采样后的信道 粒子值和权重进行加权运算。
41.一种基于粒子滤波的信道估计设备, 包括:
初始估计单元, 用于利用导频信息进行初始信道估计;
粒子初始化设置单元, 用于设置粒子范围和粒子数目, 并为每个粒子分 配初始权重;
重要采样单元, 用于以信道粒子值为条件的接收信号的条件概率密度作 为先验概率, 与当前接收信号构造贝叶斯模型, 计算出以当前接收信号为条件 的信道粒子值的后验概率密度; 以及对所有粒子的权值进行归一化, 并进行加 权运算;
重采样单元, 用于根据判断条件对粒子进行重采样, 其中所述重要采样单元和所述重采样单元执行时间递归操作, 从而得到 发送所有导频处的准确信道估计值。
42.根据权利要求 41所述的基于粒子滤波的信道估计设备, 其特征在于 还包括:
导频信息设置单元,根据采用的通信系统是单天线系统还是多天线系统, 设置相应的导频信息。
43.根据权利要求 41或 42所述的基于粒子滤波的信道估计设备, 其特征 在于:
所述初始估计单元利用导频信息求出基于最小平方准则 LS下的初始信道 估计值, 再把 LS估计得到的复数形式的信道信息分解成实部形式和虚部形式 的信道信息。
44.根据权利要求 41或 42所述的基于粒子滤波的信道估计设备, 其特征 在于:
所述粒子初始化设置单元利用得到的实部和虛部形式的初始信道估计 值, 进行粒子滤波的序贯重要采样法的初始化设置, 初始化设置包括抽取随机 样本, 即设置粒子的数目、 粒子范围、 各个粒子所对应的权重。
45.根据权利要求 41或 42所述的基于粒子滤波的信道估计设备, 其特征 在于:
针对当前时刻,
所述重要采样单元根据噪声方差得到以信道粒子值为条件的 接收信号的条件概率密度, 即先验概率; 用该先验概率和当前的接 收信号来更新每个粒子的权值, 即运用贝叶斯公式得到以当前接收 信号为条件的信道粒子值的后验概率密度; 对所有粒子的权值进行 归一化, 得到各个粒子的新的权重值; 进行加权运算, 即对所有信 道粒子值和它们的概率密度来求出数学期望值,得到当前时刻的准 确的信道估计值, 这些信道估计值为实部和虛部的形式, 所述重采样单元根据滤波退化检测公式, 判断性能低于门限 值时, 进行重采样操作;
针对下一时刻, 所述重要采样单元和所述重采样单元执行迭代运算。
46.根据权利要求 45所述的基于粒子滤波的信道估计设备, 其特征在于 还包括:
信道估计值实部与虛部合并单元, 用于根据各个时刻的导频符号的准确 信道估计值, 合并实部和虛部的形式的信道估计值, 使信道估计值为能体现幅 度和相位信息的复数形式; 以及
插值运算单元, 用于通过插值的方法得到数据符号的信道信息值, 从而 为下一环节的数据检测或译码提供较为准确的信道信息。
47.根据权利要求 42所述的基于粒子滤波的信道估计设备,其特征在于: 当天线为多天线系统时, 不同天线不同时刻构成的导频矩阵为满秩矩阵, 对不同发送天线的导频符号进行解耦和分离, 把多天线系统的信道估计问题简 化为单天线系统的信道估计问题。
48.根据权利要求 43所述的基于粒子滤波的信道估计设备,其特征在于: 导频符号设计为实数, 把信道信息分解为实部和虚部, 分别对实部和虚 部进行基于粒子滤波的信道估计, 在信道估计过程结束前, 再把信道信息的实 部和虚部进行合并, 从而得到完整的信道信息的估计值。
49.根据权利要求 45所述的基于粒子滤波的信道估计设备,其特征在于: 所述噪声方差在慢时变信道中在信道跟踪的初始阶段设置为比理论值大 的值, 在初步跟踪到信道之后, 设置为便于精确跟踪的值。
50.根据权利要求 45所述的基于粒子滤波的信道估计设备,其特征在于: 所述噪声方差在快时变信道中在信道跟踪的初始阶段设置为比理论值小 的值, 在初步跟踪到信道之后, 设置为便于精确跟踪的值。
51.根据权利要求 45所迷的基于粒子滤波的信道估计设备,其特征在于: 所述重采样单元所进行的重采样基于所述以当前接收信号为条件的信道 粒子值的后验概率密度, 利用离散的积分将概率密度转化为概率分布, 然后对 概率分布的概率轴进行等分, 在重新分割后的分布轴上分配粒子,并且在分配 粒子的时候引入向外延拓的粒子, 以适应信道的变化。
52.根据权利要求 51所述的基于粒子滤波的信道估计设备,其特征在于: 所述重采样单元在判断粒子滤波的性能低于门限值时, 进行重采样; 以 及所述重要采样单元运用重采样后的信道粒子值和权重进行加权运算。
53. 一种选择性信道估计方法, 包括:
判断信道属性, 将非线性信道或非高斯噪声的信道判定为恶劣信道环境, 将线性且高斯白噪声的信道判定为非恶劣信道环境;
针对恶劣信道环境, 采用根据权利要求 29 ~ 40之一所述的基于粒子滤波 的信道估计方法, 进行信道估计;
针对非恶劣信道环境, 采用根据权利要求 1 ~ 14之一所述的基于可变遗 忘因子递归最小平方滤波的信道估计方法, 进行信道估计;
基于信道估计的结果, 进行插值和信号检测;
计算所检测出的信号的误码率, 并判断误码率是否满足预定的性能要求; 如果误码率不满足预定的性能要求, 则采用根据权利要求 29 ~ 40之一所 述的基于粒子滤波的信道估计方法, 再次进行信道估计, 其中通过增大粒子数 目来提高信道估计的精度。
54. 根据权利要求 53所述的选择性信道估计方法, 其中
如果误码率满足预定的性能要求, 则接收机结束一帧符号的接收。
55. 一种选择性信道估计设备, 包括:
信道属性判断单元, 用于将非线性信道或非高斯噪声的信道判定为恶劣 信道环境, 以及将线性且高斯白噪声的信道判定为非恶劣信道环境;
根据权利要求 41 ~ 52之一所述的基于粒子滤波的信道估计设备, 用于对 恶劣信道环境进行信道估计;
根据权利要求 15 - 28之一所述的基于可变遗忘因子递归最小平方滤波的 信道估计设备, 用于对非恶劣信道环境进行信道估计;
插值和信号检测单元, 用于基于信道估计的结果, 进行插值和信号检测; 误码率判断单元, 用于计算所检测出的信号的误码率, 并判断误码率是 否满足预定的性能要求, 其中如果误码率不满足预定的性能要求, 则通知根据 权利要求 41 - 52之一所述的基于粒子滤波的信道估计设备再次进行信道估计, 其中通过增大粒子数目来提高信道估计的精度。
56. 根据权利要求 55所述的选择性信道估计设备, 其中
如果误码率满足预定的性能要求, 则接收机结束一帧符号的接收。
PCT/CN2007/001531 2006-05-11 2007-05-10 Procédé et dispositif d'estimation de canaux WO2007137484A1 (fr)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
CN 200610026428 CN1866945A (zh) 2006-05-11 2006-05-11 Ofdm系统中基于可变遗忘因子的rls信道估计方法
CN200610026428.9 2006-05-11
CN200610026425.5 2006-05-11
CNB2006100264255A CN100553166C (zh) 2006-05-11 2006-05-11 基于粒子滤波的信道估计方法

Publications (1)

Publication Number Publication Date
WO2007137484A1 true WO2007137484A1 (fr) 2007-12-06

Family

ID=38778102

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2007/001531 WO2007137484A1 (fr) 2006-05-11 2007-05-10 Procédé et dispositif d'estimation de canaux

Country Status (1)

Country Link
WO (1) WO2007137484A1 (zh)

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102215188A (zh) * 2011-06-20 2011-10-12 哈尔滨工程大学 改进人工蜂群粒子滤波信道估计方法
CN105634583A (zh) * 2015-12-28 2016-06-01 中国电子科技集团公司第五十四研究所 基于二进制粒子群的卫星ofdm载波优化配置方法
CN108470089A (zh) * 2018-02-12 2018-08-31 杭州电子科技大学 一种基于最小二乘样本拟合的复信号时延估计方法
CN110032968A (zh) * 2019-04-11 2019-07-19 长春精仪光电技术有限公司 基于双树复小波和自适应半软阈值法的去噪方法
CN111835365A (zh) * 2020-07-30 2020-10-27 西安烽火电子科技有限责任公司 一种适用于短波通信的梳状滤波Turbo均衡算法
CN112054975A (zh) * 2020-09-11 2020-12-08 郑州大学 一种基于bp-mf框架和vamp的联合估计与检测方法
CN112244876A (zh) * 2020-09-30 2021-01-22 北京工业大学 一种基于可变遗忘因子的自适应滤波器的脑电信号时序随机噪声去除方法
CN112637093A (zh) * 2020-12-09 2021-04-09 齐鲁工业大学 一种基于模型驱动深度学习的信号检测方法
CN112751792A (zh) * 2019-10-31 2021-05-04 华为技术有限公司 一种信道估计方法及装置
CN113037590A (zh) * 2021-03-02 2021-06-25 成都爱瑞无线科技有限公司 一种用于通信系统中的时延估计方法和装置
CN113225044A (zh) * 2021-02-22 2021-08-06 清华大学苏州汽车研究院(相城) 一种智能粒子滤波方法
CN113422745A (zh) * 2021-06-17 2021-09-21 中山大学 一种基于深度学习的空海无线信道估计方法
CN114244675A (zh) * 2021-12-29 2022-03-25 电子科技大学 一种基于深度学习的mimo-ofdm系统信道估计方法
CN114785650A (zh) * 2022-05-10 2022-07-22 北京邮电大学 一种新型盲相位搜索算法结构及实现方法
CN115022133A (zh) * 2022-05-31 2022-09-06 厦门大学 一种结构分解的多径类型适配水声信道估计方法和系统
CN115343570A (zh) * 2022-07-21 2022-11-15 国网湖北省电力有限公司电力科学研究院 基于可变遗忘因子递归最小二乘算法的电网阻抗在线辨识方法及装置
CN116962123A (zh) * 2023-09-20 2023-10-27 大尧信息科技(湖南)有限公司 软件定义框架的升余弦成型滤波带宽估计方法与系统

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1437338A (zh) * 2003-03-08 2003-08-20 华中科技大学 正交频分复用通信系统中的信道估计方法
CN1481100A (zh) * 2003-08-06 2004-03-10 北京交通大学 空时解相关码分多址多径信号波达方向估计方法和装置

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1437338A (zh) * 2003-03-08 2003-08-20 华中科技大学 正交频分复用通信系统中的信道估计方法
CN1481100A (zh) * 2003-08-06 2004-03-10 北京交通大学 空时解相关码分多址多径信号波达方向估计方法和装置

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
LIANG Y. ET AL.: "RLS channel estimation with adaptive forgetting factor in space-time coded MIMO-OFDM systems", JOURNAL OF ZHEJIANG UNIVERSITY SCIENCE A, vol. 7, no. 4, 30 April 2006 (2006-04-30), pages 507 - 515, XP019361001 *
WANG S. ET AL.: "Blind channel equalization's performance of chaos-based communication systems with particle filter", JOURNAL OF CIRCUITS AND SYSTEMS, vol. 10, no. 1, 28 February 2005 (2005-02-28), pages 98 - 102, XP008093088 *

Cited By (27)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102215188A (zh) * 2011-06-20 2011-10-12 哈尔滨工程大学 改进人工蜂群粒子滤波信道估计方法
CN105634583A (zh) * 2015-12-28 2016-06-01 中国电子科技集团公司第五十四研究所 基于二进制粒子群的卫星ofdm载波优化配置方法
CN108470089B (zh) * 2018-02-12 2021-12-21 杭州电子科技大学 一种基于最小二乘样本拟合的复信号时延估计方法
CN108470089A (zh) * 2018-02-12 2018-08-31 杭州电子科技大学 一种基于最小二乘样本拟合的复信号时延估计方法
CN110032968A (zh) * 2019-04-11 2019-07-19 长春精仪光电技术有限公司 基于双树复小波和自适应半软阈值法的去噪方法
CN110032968B (zh) * 2019-04-11 2022-12-09 长春精仪光电技术有限公司 基于双树复小波和自适应半软阈值法的去噪方法
CN112751792B (zh) * 2019-10-31 2022-06-10 华为技术有限公司 一种信道估计方法及装置
CN112751792A (zh) * 2019-10-31 2021-05-04 华为技术有限公司 一种信道估计方法及装置
CN111835365A (zh) * 2020-07-30 2020-10-27 西安烽火电子科技有限责任公司 一种适用于短波通信的梳状滤波Turbo均衡算法
CN112054975B (zh) * 2020-09-11 2022-11-01 郑州大学 一种基于bp-mf框架和vamp的联合估计与检测方法
CN112054975A (zh) * 2020-09-11 2020-12-08 郑州大学 一种基于bp-mf框架和vamp的联合估计与检测方法
CN112244876A (zh) * 2020-09-30 2021-01-22 北京工业大学 一种基于可变遗忘因子的自适应滤波器的脑电信号时序随机噪声去除方法
CN112637093B (zh) * 2020-12-09 2022-05-17 齐鲁工业大学 一种基于模型驱动深度学习的信号检测方法
CN112637093A (zh) * 2020-12-09 2021-04-09 齐鲁工业大学 一种基于模型驱动深度学习的信号检测方法
CN113225044A (zh) * 2021-02-22 2021-08-06 清华大学苏州汽车研究院(相城) 一种智能粒子滤波方法
CN113225044B (zh) * 2021-02-22 2024-04-09 清华大学苏州汽车研究院(相城) 一种智能粒子滤波方法
CN113037590A (zh) * 2021-03-02 2021-06-25 成都爱瑞无线科技有限公司 一种用于通信系统中的时延估计方法和装置
CN113422745A (zh) * 2021-06-17 2021-09-21 中山大学 一种基于深度学习的空海无线信道估计方法
CN114244675A (zh) * 2021-12-29 2022-03-25 电子科技大学 一种基于深度学习的mimo-ofdm系统信道估计方法
CN114785650A (zh) * 2022-05-10 2022-07-22 北京邮电大学 一种新型盲相位搜索算法结构及实现方法
CN114785650B (zh) * 2022-05-10 2024-02-27 北京邮电大学 一种新型盲相位搜索算法结构及实现方法
CN115022133B (zh) * 2022-05-31 2023-08-18 厦门大学 一种结构分解的多径类型适配水声信道估计方法和系统
CN115022133A (zh) * 2022-05-31 2022-09-06 厦门大学 一种结构分解的多径类型适配水声信道估计方法和系统
CN115343570B (zh) * 2022-07-21 2023-05-23 国网湖北省电力有限公司电力科学研究院 一种电网阻抗在线辨识方法及装置
CN115343570A (zh) * 2022-07-21 2022-11-15 国网湖北省电力有限公司电力科学研究院 基于可变遗忘因子递归最小二乘算法的电网阻抗在线辨识方法及装置
CN116962123A (zh) * 2023-09-20 2023-10-27 大尧信息科技(湖南)有限公司 软件定义框架的升余弦成型滤波带宽估计方法与系统
CN116962123B (zh) * 2023-09-20 2023-11-24 大尧信息科技(湖南)有限公司 软件定义框架的升余弦成型滤波带宽估计方法与系统

Similar Documents

Publication Publication Date Title
WO2007137484A1 (fr) Procédé et dispositif d&#39;estimation de canaux
Prasad et al. Joint approximately sparse channel estimation and data detection in OFDM systems using sparse Bayesian learning
Taubock et al. Compressive estimation of doubly selective channels in multicarrier systems: Leakage effects and sparsity-enhancing processing
CN101232474B (zh) 宽带无线接入系统中的接收装置和方法
CN102724147B (zh) 一种水声正交频分复用的信道估计方法
JP2009532957A (ja) 急速な分散性フェージングチャンネルのためのチャンネル推定
JP4272665B2 (ja) Ofdm伝送システムのチャネルを推定する装置、方法、及びコンピュータプログラム
JP2004159277A (ja) 無線通信システムのチャネルを介して受信した変調信号のシンボルを検出する方法およびシステム
JP4147193B2 (ja) マルチキャリヤ拡散スペクトル信号の受信
CN1937598A (zh) 正交频分复用系统中的信道估计方法以及信道估计器
CN108156101B (zh) 一种mimo-scfde系统联合迭代信道估计与迭代均衡方法
WO2013121958A1 (ja) 受信装置、受信方法および受信プログラム
CN107332797A (zh) 一种电力线ofdm通信系统中的信道估计方法
CN101018219B (zh) 一种空频信号处理方法
KR101368131B1 (ko) 다중경로 채널의 지연 확산을 측정하기 위한 방법 및 장치
WO2012106963A1 (zh) 干扰和噪声消除方法及装置
Drakshayini et al. A review of wireless channel estimation techniques: challenges and solutions
Bhoyar et al. Leaky least mean square (LLMS) algorithm for channel estimation in BPSK-QPSK-PSK MIMO-OFDM system
CN111404848A (zh) 一种渐进最优的海洋通信系统信道估计方法
Li et al. Asynchronous co-channel interference suppression in MIMO OFDM systems
CN111817990B (zh) Ofdm系统中基于最小均方误差的一种信道估计改进算法
CN102891816B (zh) 一种基于mimo-ofdm相关信道的去耦合预测方法
Hedayati et al. SAGE algorithm for semi-blind channel estimation and symbol detection for STBC MIMO OFDM systems
Chang et al. Cancellation of ICI by Doppler effect in OFDM systems
Gupta et al. Simplified semi-blind channel estimation for space–time coded MIMO-OFDM systems

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 07721104

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 07721104

Country of ref document: EP

Kind code of ref document: A1