CN111555990A - Channel interpolation estimation method based on long-time and short-time memory residual error network - Google Patents

Channel interpolation estimation method based on long-time and short-time memory residual error network Download PDF

Info

Publication number
CN111555990A
CN111555990A CN202010344050.7A CN202010344050A CN111555990A CN 111555990 A CN111555990 A CN 111555990A CN 202010344050 A CN202010344050 A CN 202010344050A CN 111555990 A CN111555990 A CN 111555990A
Authority
CN
China
Prior art keywords
channel
time
residual error
interpolation
long
Prior art date
Legal status (The legal status 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 status listed.)
Pending
Application number
CN202010344050.7A
Other languages
Chinese (zh)
Inventor
刘杨雨
石琦
张舜卿
徐树公
曹姗
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Shanghai for Science and Technology
Original Assignee
University of Shanghai for Science and Technology
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
Application filed by University of Shanghai for Science and Technology filed Critical University of Shanghai for Science and Technology
Priority to CN202010344050.7A priority Critical patent/CN111555990A/en
Publication of CN111555990A publication Critical patent/CN111555990A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/0202—Channel estimation
    • H04L25/0224—Channel estimation using sounding signals
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/0202—Channel estimation
    • H04L25/024—Channel estimation channel estimation algorithms
    • H04L25/0256—Channel estimation using minimum mean square error criteria
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L27/00—Modulated-carrier systems
    • H04L27/26—Systems using multi-frequency codes
    • H04L27/2601—Multicarrier modulation systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Power Engineering (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

A channel interpolation estimation method based on a long-time and short-time memory residual error network is characterized in that a residual error network formed by connecting a plurality of residual error blocks in series is utilized to approximate a nonlinear interpolation relation between a reference signal and adjacent resource elements so as to improve the accuracy of channel estimation, and then a long-time and short-time memory cyclic neural network learns the slow-varying time domain correlation between continuous OFDM symbols to obtain the complex nonlinear interpolation relation of CSI of a channel at a pilot frequency position and a data position, so that complete CSI estimation is obtained through interpolation. The invention can obtain the complete CSI with the precision far higher than that of the traditional method, and the delay budget of the channel estimation is less than 1 millisecond. And the method is suitable for LOS and NLOS actual scenes on the premise of ensuring the precision, and has higher generalization capability and high feasibility of being applied to an actual system.

Description

Channel interpolation estimation method based on long-time and short-time memory residual error network
Technical Field
The invention relates to a technology in the field of wireless communication, in particular to a channel interpolation estimation method based on a long-time and short-time memory residual error network.
Background
Enhanced mobile broadband (eMBB) transmission has been identified as one of the most important scenarios in fifth generation (5G) communication systems. In a time-varying wireless environment, in order to ensure ultra-high throughput in an eMBB scenario, Orthogonal Frequency Division Multiplexing (OFDM) transmission and coherent detection are generally adopted as main transmission methods. Higher order modulation, such as 64 quadrature amplitude modulation, is widely used to improve throughput. In order to accurately detect high order modulated signals, an efficient and fast channel estimation is generally considered to be the most important step. Since in modern wireless communication only a small fraction of the resources are used for pilot transmission, channel estimation with only a small number of observations (pilots) is considered to be a challenging ill-defined reconstruction problem.
To address such complications within the channel coherence time, the standard channel estimation process typically includes a channel state recovery stage and a low complexity interpolation stage. Under the assumption of Additive White Gaussian Noise (AWGN), conventional Channel State Information (CSI) recovery schemes usually employ Least Squares (LS) or Minimum Mean Square Error (MMSE) algorithms, but such methods are complex. In the prior art, channel estimation is performed through a neural network based on deep learning, but the neural network adopted by the techniques is complex, so that channel estimation cannot be completed within the relevant time of a channel, and an over-fitting problem is easily caused.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a channel interpolation estimation method based on a long-time memory residual error network, which redefines the channel estimation problem in a wireless communication system into a super-resolution algorithm in a deep learning algorithm, and provides an algorithm aiming at channel estimation by exploring the characteristics of a channel and utilizing related domain knowledge, thereby improving the estimation precision, meeting the commercial time delay requirement and providing some design insights for other deep learning algorithms applied to the wireless communication system in the future.
The invention is realized by the following technical scheme:
the invention relates to a channel interpolation estimation method based on a long-time and short-time memory residual error network, which is characterized in that a residual error network formed by connecting a plurality of residual error blocks in series is utilized to approximate a nonlinear interpolation relation between Reference Signals (RSs) and adjacent Resource Elements (REs) so as to improve the accuracy of channel estimation, and then a long-time and short-time memory cyclic neural network learns the slowly-varying time domain correlation between continuous OFDM symbols to obtain the nonlinear interpolation relation between CSI of a channel at a pilot frequency position and a data position, so that complete CSI estimation is obtained through interpolation.
The non-linear interpolation relationship between the Reference Signals (RSs) and the neighboring Resource Elements (REs) specifically refers to: the signals transmitted over time have a certain time continuity, for unit time intervals and frequency intervals, according to the time correlation coefficient between neighborhoods
Figure BDA0002469443900000021
Plotting the correlation coefficient against the number of intervals, wherein: h isk(fRE,tRE) Is the channel response of the neighboring resource elements,
Figure BDA0002469443900000022
for the receiving end channel correlation matrix, Δ t is 0.01 s.
The complex nonlinear interpolation relationship between the CSI of the channel at the pilot position and the CSI of the channel at the data position specifically includes: currently, the OFDM mainly uses a pilot-assisted channel estimation mode, that is, a pilot signal is inserted into a data stream, a CSI at a pilot position is obtained by extracting a pilot at a receiving end and calculating, and a CSI at other data positions is estimated by using an interpolation algorithm.
The residual error network is as follows: and 16 residual blocks are connected in series to form a network.
Preferably, the residual error network is further simplified, and the number of the residual error blocks and the number of convolution filters are reduced by adjusting the original structure of the residual error blocks, so that a simplified version of the long-term memory residual error network is obtained.
The long-time and short-time memory cyclic neural network is as follows: the length of the convolution memorizes the recurrent neural network (ConvLSTM).
The long-time memory cycle residual error neural network uses a standard COST 2100 channel model to carry out offline training on the network, and uses OFDM transmission to deploy the network to a commercial Wi-Fi system on line.
The invention relates to a system for realizing the method, which comprises the following steps: the device comprises a channel estimation precision improving module, a long-time and short-time memory recurrent neural network (LSTM) module and an interpolation module, wherein: the channel estimation precision improving module approximates to a nonlinear interpolation relation between a reference signal and adjacent resource elements according to a built-in residual error network and obtains optimized channel estimation, the LSTM module learns slow varying time domain correlation between continuous OFDM symbols, the optimized channel estimation is combined to obtain the nonlinear interpolation relation of a channel CSI at a pilot frequency position and a data position, and the interpolation module estimates the transfer characteristic of the whole channel through a channel transfer function obtained through pilot frequency points.
Technical effects
The invention provides a more appropriate learning framework based on super-resolution (SR) and integrally solves the problem of wireless channel estimation; the problem of how to extract domain knowledge based on a Super Resolution (SR) learning framework to improve estimation accuracy is solved because a wireless channel has some slowly-varying characteristics in a time domain; the estimation problem of continuous wireless channels is effectively solved within the relevant time of the channels; based on the defects that the fitting of the interpolation relation of the wireless channel matrix is not accurate and the domain knowledge is not extracted comprehensively in the SR traditional learning frame method.
Compared with the prior art, the invention can improve the NMSE performance gain of 10dB to 11dB and control the processing delay within 1 millisecond. And further, offline training is performed, so that the method is suitable for line-of-sight transmission (LOS) and non-line-of-sight transmission (NLOS) actual scenes on the premise of ensuring the precision, and has high generalization capability and high feasibility of being applied to an actual system.
Drawings
FIG. 1 is a drawing ofContaining NtA time slot and NfA resource block schematic of a subcarrier;
FIG. 2 is a schematic diagram of a super-resolution convolutional neural network (SRCNN);
FIG. 3 is a schematic diagram of an enhanced depth residual error neural network (EDSR) model for a monocular super-resolution image;
fig. 4 is a schematic diagram of an LSRN network architecture according to an embodiment;
FIG. 5 is a schematic diagram of the variation of the related structure delay before and after network simplification;
FIG. 6 is a diagram illustrating a comparison of the structures of three different residual blocks;
in the figure: (a) ResB _ Leidge, (b) ResB _ Lim, (c) ResB _ deployed;
FIG. 7 is a diagram illustrating comparison of performance of three residual block structures;
FIG. 8 is a graph illustrating a comparison of performance between different network architectures;
fig. 9 is a schematic diagram of NMSE performance comparison in LOS environment based on SRCNN and conventional channel estimation algorithm;
in the figure: (a) pilot frequency configuration: 14 × 14, (b) pilot configuration: 28X 28;
FIG. 10 is a schematic diagram of NMSE performance comparison in NLOS and LOS environments based on LSRN-L channel estimation algorithm;
fig. 11 is a schematic diagram of NMSE performance comparison based on the LSRN-L channel estimation algorithm under different training samples and different pilot configurations;
fig. 12 is a schematic structural diagram of an LSRN in an embodiment;
FIG. 13 is a schematic diagram of the structure of LSRN-L in the example.
Detailed Description
The embodiment relates to a channel interpolation estimation method based on a long-time and short-time memory residual error network, which comprises the following steps:
step 1) in a wireless time-varying environment, constructing a wireless network with a single transmitting antenna and NrThe OFDM transmission system model of the root receiving antenna specifically includes:
as shown in FIG. 1, under Additive White Gaussian Noise (AWGN), the kth resource block is at (f)Rs,tRs)thNeighboring resourcesSymbol received at element
Figure BDA0002469443900000031
The method specifically comprises the following steps: y isk(fRS,tRS)＝hk(fRS,tRs)xk(fRS,tRs)+nk(fRs,tRS) Wherein: (f)Rs,tRs)∈ΩRS,ΩRSRepresenting a set of reference signals on each resource block. x is the number ofk(fRS,tRS)、hk(fRS,tRS) Respectively, a predefined RS and an equivalent channel response, nk(fRS，tRS) Is mean value of
Figure BDA0002469443900000032
Variance of
Figure BDA0002469443900000033
Additive complex gaussian noise.
Estimating channel state based on Minimum Mean Square Error (MMSE) criterion
Figure BDA0002469443900000034
Figure BDA0002469443900000035
Wherein:
Figure BDA0002469443900000036
a receiving end channel correlation matrix is obtained; in practical systems, in order to minimize the overhead of channel estimation, the whole processing procedure uses one resource block as a unit, and the estimated channel state of the reference signal is mapped to the whole resource block hkAnd expressing the CSI of the kth resource block, wherein the interpolation process comprises the following steps:
Figure BDA0002469443900000037
wherein: omegaRSRepresents a set of reference signals on each resource block,
Figure BDA0002469443900000038
is from | ΩRSI reference signal mapping to Nt×NtInterpolation function of neighboring resource elements, NtIs the number of time slots, NfIs the number of sub-carriers.
The interpolation adopts, but is not limited to, a Fourier interpolation process by jointly utilizing a frequency domain and a Gaussian approximation of a time domain. (GI + DFTI). The image SR reconstruction algorithm has stronger nonlinear function
Figure BDA0002469443900000039
The ability to do modeling, interpolation of channel estimates can be seen as a single image SR reconstruction problem.
The present embodiment estimates CSI (h) by minimization using a super-resolution techniquek SR) And channel CSI (h)k) MSE in between to find a suitable interpolation function
Figure BDA0002469443900000041
Optimal SR reconstruction in which CSI is interpolated, i.e.
Figure BDA0002469443900000042
The method is realized by the following problem modes:
Figure BDA0002469443900000043
Figure BDA0002469443900000044
the above problem is achieved using conventional MMSE estimation of CSI, i.e.
Figure BDA0002469443900000045
Fig. 2 and 3 show schematic models of conventional SR neural networks SRCNN and EDSR.
Step 2) as shown in fig. 4, feature extraction and fusion are performed through a long-time cyclic residual error network (LSRN) to obtain NfcFeatures of the dimension, then mapping N by upsampling and high dimension featurefcDimensional feature ofSpread out to Nt×NfAnd REs, obtaining the nonlinear interpolation relation of the CSI of the channel at the pilot frequency position and the data position, thereby obtaining complete CSI estimation through interpolation.
The long and short time cycle residual error network comprises: a loop learning block and a local residual learning block for extracting a time domain correlation of a channel, wherein: the long-short time cyclic residual error network adopts a convolution long-short time memory network (ConvLSTM), the local residual error learning block comprises 16 residual error blocks ResB _ Lim which are arranged in series and used for extracting features, each residual error block ResB _ Lim comprises an input convolutional layer, an activation function and an output convolutional layer, and the long-short time cyclic residual error network utilizes the local residual error learning block to approximate a nonlinear interpolation relation between Reference Signals (RSs) and adjacent Resource Elements (REs) so as to improve the accuracy of channel estimation.
The detailed parameters of the LSRN network are shown in the following table:
TABLE 1
Figure BDA0002469443900000046
Network parameters of the LSRN network
Figure BDA0002469443900000051
Which satisfies the following conditions:
Figure BDA0002469443900000052
wherein:
Figure BDA0002469443900000053
for the LSRN network framework, the network parameter is thetaLSRN，
Figure BDA0002469443900000054
Channel CSI estimated for the network.
Said NfcCharacteristics of dimension
Figure BDA0002469443900000055
Is composed of
Figure BDA0002469443900000056
Figure BDA0002469443900000057
Via an upsampling network
Figure BDA0002469443900000058
Network and feature mapping network
Figure BDA0002469443900000059
Will NfcIs extended to Nt×Nt(ii) a (REs); the final estimated CSI value is:
Figure BDA00024694439000000510
in practical applications, the long-and-short-term cyclic residual error network (LSRN) in step 2 may also be simplified to further improve the real-time performance of channel estimation, i.e., complexity reduction is achieved by removing a convolutional layer from the local residual error learning block.
The simplified LSRN is shown in fig. 13 and includes: a loop learning block and a local residual learning block for extracting a time domain correlation of a channel, wherein: the cyclic learning block adopts a convolution long-time memory network (ConvLSTM), and the local residual learning block is a simplified residual block ResB _ pro which comprises a convolution layer and an activation function.
Alternatively, the present embodiment may also achieve structural simplification by reducing the number of residual blocks in the local residual learning block and convolution filters in the residual block.
As shown in fig. 5 and fig. 6, the performance comparison result of the simplified LSRN of the present invention and the conventional two residual block schemes is shown.
As shown in fig. 7, it is seen that by simplifying the number of convolution layers in each of the 16 residual blocks in the local residual learning block of the LSRN network, the overall processing delay is reduced from 27ms to 20ms on average, approximately 25.9%, with less performance loss.
In a conventional single-scale SR network (EDSR) scheme, reducing the number of residual blocks and filters may have a great influence on SR performance of an image. However, in a particular task, the corresponding performance degradation is controllable for the following reasons. First, since in the conventional image SR task the correlation between different REs is much stronger than the correlation between pixels, the difference in extracted features between different residual blocks is much smaller, which makes it possible to reduce the number of residual blocks. Second, in the proposed LSRN architecture, part of the characteristic loss caused by reducing the residual block and filter is compensated by a loop learning branch. With the simulation results as shown in fig. 7, the time delay is reduced from 20 msec to 0.9 msec in the case of the pilot configuration of 14 × 14.
The present embodiment uses two different methods to generate the real-time channel states, namely direct "model generation" and "prototype sampling", where: the "model generated" dataset was collected directly from the well-known channel model COST 2100, while the "prototype sampled" dataset was from an actual Wi-Fi prototype type system, as shown in FIG. 4. The detailed configuration of the prototype system is shown in table 2.
TABLE 2Wi-Fi prototype System configuration and network learning configuration parameters
Figure BDA00024694439000000511
Figure BDA0002469443900000061
Then generating the original high-resolution CSI hk(fRE,tRE) And generates low-resolution CSI according to the traditional MMSE channel estimation method,
Figure BDA0002469443900000062
different data sets for training and testing were constructed based on different configurations of RS and LOS/NLOS, as shown in Table 3 below, the training and testing data sets were each 5.6 × 10 in size5And 1.12 × 105。
TABLE 3 relevant training and testing data sets in the experiment
Figure BDA0002469443900000063
The present embodiment employs two different pilot configurations, 14 × 14 and 28 × 28, in the channel estimation portion of the pilot locations. The corresponding numerical results are shown in fig. 9. It can be seen from the figure that in the pilot configuration of 14 × 14, LSRN and LSRN-L are respectively superior to the conventional GI + DFTI algorithms by 14 ~ 15dB and 10 ~ 11 dB. Meanwhile, LSRN-L performs much better than the LSRN scheme in terms of processing delay, eventually controlling the processing delay to within 1 millisecond.
Further comparing the channel estimation performance of the traditional GI + DFT scheme and the low complexity LSRN _ L scheme for the Wi-Fi prototype type system in NLOS and LOS environments, it can be seen from the experimental results shown in fig. 10 that the Normalized Mean Square Error (NMSE) of the LSRN-L scheme of the present invention is better than that of the reference scheme GI + DFTI in all cases. The test results of different data set training models are compared to see that the generated data set training model is also suitable for channel estimation scenes under LOS and NLOS in a Wi-Fi environment, the network is robust to the actual application scene, and the complex correlation characteristics among the pilots are kept through network training, so that the method is suitable for different channel fading environments in an actual communication system.
In order to see if a small amount of prototype sample data helps to improve channel estimation in practical communication systems, some modifications were made to the training data set: set a small amount of data
Figure BDA0002469443900000064
Or
Figure BDA0002469443900000065
Blending into a data set
Figure BDA0002469443900000066
In (1) simulating an actual communication system. According to the experimental results shown in fig. 9, using a mixed training data set containing 20% sampled data, most of the gain of the results was obtained, about 70%.
In addition, in the case of a fixed number of pilotsUnder the conditions, the influence of different pilot designs on the channel estimation effect is evaluated: two different pilot frequency design modes are newly added, and the corresponding training set is
Figure BDA0002469443900000071
And
Figure BDA0002469443900000072
the test set is
Figure BDA0002469443900000073
And
Figure BDA0002469443900000074
the experimental results are shown in fig. 11, it is seen that the channel estimation performance of the pilot configuration 7 × 28 varies greatly under different signal-to-noise ratios, and the overall estimation effect is not as good as that of the pilot configuration 28 × 7. more importantly, the pilot configuration 28 × 7 uses a smaller number of pilots to obtain an nmse similar to that of the pilot configuration 28 × 28.
The foregoing embodiments may be modified in many different ways by those skilled in the art without departing from the spirit and scope of the invention, which is defined by the appended claims and all changes that come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

Claims (9)

1. A channel interpolation estimation method based on a long-time and short-time memory residual error network is characterized in that a residual error network formed by connecting a plurality of residual error blocks in series is used for approximating a nonlinear interpolation relation between a reference signal and adjacent resource elements so as to improve the accuracy of channel estimation, and then a long-time and short-time memory cyclic neural network is used for learning the slowly-varying time domain correlation between continuous OFDM symbols so as to obtain the nonlinear interpolation relation between a channel at a pilot frequency position and a Channel State Information (CSI) at a data position, so that complete CSI estimation is obtained through difference values;
the residual error network formed by connecting a plurality of residual error blocks in series comprises: a loop learning block and a local residual learning block for extracting a time domain correlation of a channel, wherein: the local residual learning block comprises a first convolution module, a residual block group or simplified residual block and a second convolution module, wherein the first convolution module, the residual block group or simplified residual block group is composed of 16 residual blocks used for extracting features, and the second convolution module is arranged in series.
2. The channel interpolation estimation method based on long-and-short-term memory residual error network as claimed in claim 1, wherein the non-linear interpolation relationship between the Reference Signal (RSs) and the neighboring Resource Element (REs) is specifically: the signals transmitted over time have a certain time continuity, for unit time intervals and frequency intervals, according to the time correlation coefficient between neighborhoods
Figure FDA0002469443890000011
Plotting the correlation coefficient against the number of intervals, wherein: h isk(fRE,tRE) Is the channel response of the neighboring resource elements,
Figure FDA0002469443890000012
for the receiving end channel correlation matrix, Δ t is 0.01 s.
3. The channel interpolation estimation method based on a long-and-short-term memory residual error network as claimed in claim 1, wherein the non-linear interpolation relationship between the CSI of the channel at the pilot position and the CSI of the channel at the data position specifically includes: by inserting pilot signals into the sending data stream, the CSI at the pilot position is obtained by extracting pilot calculation at the receiving end, and the CSI at other data positions is estimated by an interpolation algorithm.
4. The channel interpolation estimation method based on the long-and-short memory residual error network as claimed in claim 1, wherein the approximation is: obtaining N by carrying out feature extraction and fusion through long-time cyclic residual error networkfcFeatures of the dimension, then mapping N by upsampling and high dimension featurefcExtend the dimensional feature of to Nt×NfREs, from which the channel response is estimated.
5. The channel interpolation estimation method based on long-time and short-time memory residual error network as claimed in claim 1, wherein, in the residual error block group, each residual error block comprises an input convolutional layer, an activation function and an output convolutional layer; the simplified residual block includes an input convolutional layer and an activation function.
6. The channel interpolation estimation method based on long-and-short-term memory residual error network as claimed in claim 1, wherein the cyclic learning block adopts a convolution long-and-short-term memory network.
7. The channel interpolation estimation method based on long-short term memory residual error network as claimed in claim 1, wherein the long-short term memory cyclic neural network for learning the slowly varying time domain correlation between consecutive OFDM symbols employs a convolution long-short term memory cyclic neural network.
8. The channel interpolation estimation method based on long-and-short-term memory residual error network as claimed in claim 1, wherein the interpolation employs a fourier interpolation process of a frequency domain and a gaussian approximation of a time domain.
9. A system for implementing the method of any preceding claim, comprising: the device comprises a channel estimation precision improving module, a long-time and short-time memory cyclic neural network module and an interpolation module, wherein: the channel estimation precision improving module approximates to a nonlinear interpolation relation between a reference signal and adjacent resource elements according to a built-in residual error network and obtains optimized channel estimation, the LSTM module learns slow varying time domain correlation between continuous OFDM symbols, the optimized channel estimation is combined to obtain the nonlinear interpolation relation of a channel CSI at a pilot frequency position and a data position, and the interpolation module estimates the transfer characteristic of the whole channel through a channel transfer function obtained through pilot frequency points.
CN202010344050.7A 2020-04-27 2020-04-27 Channel interpolation estimation method based on long-time and short-time memory residual error network Pending CN111555990A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010344050.7A CN111555990A (en) 2020-04-27 2020-04-27 Channel interpolation estimation method based on long-time and short-time memory residual error network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010344050.7A CN111555990A (en) 2020-04-27 2020-04-27 Channel interpolation estimation method based on long-time and short-time memory residual error network

Publications (1)

Publication Number Publication Date
CN111555990A true CN111555990A (en) 2020-08-18

Family

ID=72007722

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010344050.7A Pending CN111555990A (en) 2020-04-27 2020-04-27 Channel interpolation estimation method based on long-time and short-time memory residual error network

Country Status (1)

Country Link
CN (1) CN111555990A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113347559A (en) * 2021-05-14 2021-09-03 武汉大学 Strong robustness wireless positioning method based on deep learning
CN115396262A (en) * 2021-05-19 2022-11-25 维沃移动通信有限公司 Channel estimation method, device, equipment and readable storage medium
CN116132239A (en) * 2023-01-31 2023-05-16 齐鲁工业大学(山东省科学院) OFDM channel estimation method adopting pre-activation residual error unit and super-resolution network

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070115799A1 (en) * 2005-10-18 2007-05-24 Pang-An Ting MIMO-OFDM system and pre-coding and feedback method therein
CN101611580A (en) * 2007-01-19 2009-12-23 汤姆逊许可公司 The interpolation method, channel estimation methods and the device that are used for ofdm system
CN106127684A (en) * 2016-06-22 2016-11-16 中国科学院自动化研究所 Image super-resolution Enhancement Method based on forward-backward recutrnce convolutional neural networks
CN107508777A (en) * 2017-07-07 2017-12-22 广东顺德中山大学卡内基梅隆大学国际联合研究院 The channel estimation methods of adaptive polarization linear interpolation based on enhancing
CN109981498A (en) * 2019-03-12 2019-07-05 上海大学 Wi-Fi modular system channel estimation methods based on super-resolution image restoration technology

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070115799A1 (en) * 2005-10-18 2007-05-24 Pang-An Ting MIMO-OFDM system and pre-coding and feedback method therein
CN101611580A (en) * 2007-01-19 2009-12-23 汤姆逊许可公司 The interpolation method, channel estimation methods and the device that are used for ofdm system
CN106127684A (en) * 2016-06-22 2016-11-16 中国科学院自动化研究所 Image super-resolution Enhancement Method based on forward-backward recutrnce convolutional neural networks
CN107508777A (en) * 2017-07-07 2017-12-22 广东顺德中山大学卡内基梅隆大学国际联合研究院 The channel estimation methods of adaptive polarization linear interpolation based on enhancing
CN109981498A (en) * 2019-03-12 2019-07-05 上海大学 Wi-Fi modular system channel estimation methods based on super-resolution image restoration technology

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
张舜卿等: "LSRN: A Recurrent Residual Learning Framework for Continuous Wireless Channel Estimation Using Super-Resolution Concept", 《IEEE ACCESS》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113347559A (en) * 2021-05-14 2021-09-03 武汉大学 Strong robustness wireless positioning method based on deep learning
CN113347559B (en) * 2021-05-14 2022-04-29 武汉大学 Strong robustness wireless positioning method based on deep learning
CN115396262A (en) * 2021-05-19 2022-11-25 维沃移动通信有限公司 Channel estimation method, device, equipment and readable storage medium
CN115396262B (en) * 2021-05-19 2024-05-28 维沃移动通信有限公司 Channel estimation method, device, equipment and readable storage medium
CN116132239A (en) * 2023-01-31 2023-05-16 齐鲁工业大学(山东省科学院) OFDM channel estimation method adopting pre-activation residual error unit and super-resolution network

Similar Documents

Publication Publication Date Title
Soltani et al. Deep learning-based channel estimation
CN111404849B (en) OFDM channel estimation and signal detection method based on deep learning
CN111555990A (en) Channel interpolation estimation method based on long-time and short-time memory residual error network
Hanna et al. Signal processing-based deep learning for blind symbol decoding and modulation classification
CN113285899B (en) Time-varying channel estimation method and system based on deep learning
CN106961403A (en) A kind of OFDM underwater sound voice communication Doppler effect correction methods of estimation based on sparse channel model
CN113225131B (en) Blind detection method of underwater visible light communication system
CN111510402A (en) OFDM channel estimation method based on deep learning
CN113452641B (en) FBMC channel estimation method, system, computer equipment and terminal
CN110943944A (en) Channel estimation method of OFDM system based on deep learning
US20230344675A1 (en) Radio Receiver, Transmitter and System for Pilotless-OFDM Communications
CN116132239A (en) OFDM channel estimation method adopting pre-activation residual error unit and super-resolution network
Ru et al. Model-driven channel estimation for OFDM systems based on image super-resolution network
CN116405357A (en) Carrier frequency offset estimation compensation method for Fresnel domain zero pilot frequency OCDM underwater acoustic communication
EP4356574A1 (en) Transmitter and receiver for and method of transmitting and receiving symbols over an orthogonal time frequency space communication channel subject to doppler spread and multipath
KR20180032522A (en) Joint identification of merge signals in non-cooperative digital telecommunications
CN114039821A (en) Broadband mmWave MIMO-OFDM system beam space channel estimation method
CN113537025A (en) Electromagnetic modulation signal deep denoising method and system based on self-supervision learning
CN111079347A (en) Signal-to-noise ratio estimation method based on deep learning by using constellation diagram
CN116248444A (en) OTFS system channel estimation method in car networking based on improved convolutional neural network
Pradhan et al. A two-stage CNN based channel estimation for OFDM system
Zheng et al. Deep learning-aided receiver against nonlinear distortion of HPA in OFDM systems
KR102640913B1 (en) Method and apparatus estimating channel using deep learning-based data color map techniques
Hassan et al. Doppler shift compensation using an lstm-based deep neural network in underwater acoustic communication systems
CN113572709B (en) Method and system for estimating pilot frequency auxiliary channel of base station end in large-scale MIMO system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20200818