CN110071881B

CN110071881B - Active user detection and channel estimation method with adaptive overhead

Info

Publication number: CN110071881B
Application number: CN201910345235.7A
Authority: CN
Inventors: 高镇; 柯玛龙; 肖振宇; 吴泳澎
Original assignee: Beijing Institute of Technology BIT
Current assignee: Beijing Institute of Technology BIT
Priority date: 2019-04-26
Filing date: 2019-04-26
Publication date: 2020-12-15
Anticipated expiration: 2039-04-26
Also published as: CN110071881A

Abstract

The invention discloses a method for detecting active users and estimating channels with self-adaptive spending, which utilizes the estimated active user set and the channels thereof to estimate the estimated quality by calculating the error of received signals, can self-adaptively adjust the time slot spending of pilot frequency in a frame structure according to the number of active users and the channel environment in a real system, ensures the service quality with the lowest possible access time delay, and realizes the ultra-reliable active user detection and channel estimation. The invention also aims at the enhanced mobile broadband scene in the future cellular network and the macro address communication scene such as large-scale machine type communication, and utilizes the sporadic characteristic of the uplink transmission flow of the user and the sparsity of the large-scale MIMO virtual angle domain channel to reasonably design the pilot frequency signal for broadband uplink authorization-free multiple access, and introduces the advanced compressive sensing technology, thereby greatly reducing the user access time delay.

Description

Active user detection and channel estimation method with adaptive overhead

Technical Field

The invention relates to the field of active users and channel estimation in wireless communication, in particular to active user detection and channel estimation in a massive user uplink multiple access scene in macro-site communication supporting enhanced mobile broadband.

Background

With the rapid development of video streaming, social networking and emerging internet of things technologies, the number of user equipment and service traffic will increase explosively in future cellular networks, and therefore, a base station is required to support the formation of a huge connection capable of carrying huge service traffic with a large number of users. The current cellular network design mainly aims at mobile phone users, the number of supportable users is limited, and the reliable support of the macro-dimensional uplink multiple access is very challenging for the current network.

The traditional random access protocol based on authorization needs additional control signaling and resource scheduling, and the user equipment needs to be authorized by the base station to access the network. Typical physical random access channel protocols in 4G-LTE systems are classified into contention-free random access and contention random access. In contention-free random multiple access, a base station allocates a pilot frequency specific to each user so that the base station can distinguish different users; in the contention random multiple access, an active user selects a pilot frequency from a predefined pilot frequency set and transmits the pilot frequency, and when two users select the same pilot frequency, additional request collision processing is required. Unfortunately, in macro address communication, due to the large number of users, the access mode of such competing resources may cause serious access request collision, thereby causing a large number of users to be unable to access the network quickly. If each user is allocated orthogonal multiple access resources, a huge overhead is incurred. In addition, after the user accesses the network successfully, the pilot frequency needs to be sent additionally to realize uplink channel estimation. Thus, for macro-address communication, the conventional grant-based random multiple access scheme may result in higher access latency and complex request collision handling. On the other hand, an important feature of future macro address communication is that a large number of user equipments are low-power consumption sensors with limited battery, etc., are in a dormant state for most of the time, and only when external things trigger, the user equipments will wake up and send data. A typical feature of macrodimensional multiple access is that a large number of devices do not continuously transmit data upstream to the communication network, but rather long silent and burst transmissions are interleaved. By utilizing the sparsity of user traffic, at present, researchers propose an authorization-free random multiple access scheme based on compressed sensing, and active equipment directly transmits pilot frequency and data to a base station in an uplink manner without authorization of the base station. This approach undoubtedly reduces access delay, but also brings new challenges, i.e. the user needs to detect active users according to the received signal, and perform channel estimation and data detection simultaneously.

By utilizing sporadic uplink transmission flow and a compressed sensing technology of users, various active user detection schemes are proposed at present to realize effective and reliable authorization-free random multiple access. The main idea of these schemes is to model the active user detection problem as a sparse signal reconstruction problem. Specifically, wang showa schoolteaching et al, university of qinghua, proposes a multi-user detection scheme based on compressed sensing to implement joint active user detection and data decoding. To further improve performance, professor l.wu at Oregon State University (algorithm State University) in the united states, using a priori information to send discrete symbols, proposes an active user detection scheme based on AMP (Approximate messaging) algorithm and EM (Expectation Maximization) algorithm. Professor x.wang, University of Columbia (Columbia University) indicates that Block Sparsity exists in an uplink random multiple access signal, and proposes a TA-BSASP (Threshold Aided Block sparse Adaptive Subspace tracking) algorithm to improve compressed sensing reconstruction performance. In addition, professor v.k.n.lau of hong kong science and technology university, etc., expands the work of large-scale random multiple Access to C-RAN (Cloud Radio Access Network) architecture, and proposes a reliable multiple Access scheme under the architecture. However, the above schemes all rely on perfect channel state information, and the acquisition of the channel state information, especially for macro address communication, is a very challenging task.

In order to jointly implement active user detection and channel estimation, a blind detection scheme of SCMA (Sparse Code Multiple Access) is proposed by a.bayeseh et al, hua technology limited company (canada), to support unlicensed random Multiple Access in macro-address communication, but the scheme only considers the scenario of a single antenna of a base station. For multi-antenna systems, professor v.k.n.lau of hong kong science and technology university proposes a multiple access scheme based on an improved BCS (Bayesian compressed Sensing) algorithm for C-RAN networks, which improves active user detection and channel estimation performance by exploiting the structured sparsity between different receiving antennas of the uplink channel matrix. Professor w.yu of University of Toronto, canada further reduces the computational complexity in massive random multiple access by proposing an active user detection and channel estimation algorithm based on AMP algorithm, while analyzing the superiority of multi-antenna system in macro-site communication, however, this solution needs channel specific distribution and noise power as prior information, and there is still a barrier in the application of practical communication system.

In addition, all of the above large-scale random multiple access schemes consider narrow-band macro communication of single carrier transmission, and it is still a very challenging problem for reliable support of large-scale random multiple access in an Enhanced Mobile Broadband (eMBB) scenario. Moreover, for an actual system, the number of active users and the channel environment change constantly, so that the sparsity of an uplink channel matrix is time-varying, and when the sparsity is smaller, a reliable active user set and a channel thereof can be obtained as long as the pilot frequency time slot overhead is smaller; on the contrary, when the sparsity is large, a large overhead is required. All of the above schemes use fixed-length pilot overhead, which undoubtedly results in pilot waste or insufficiency, and therefore a more flexible way is needed to reasonably control the pilot overhead.

Disclosure of Invention

In view of this, the present invention provides an adaptive overhead active user detection and channel estimation method for macro-dimensional random multiple access in a broadband scenario, which can adaptively adjust multiple access pilot overhead according to the number of actual active users and the channel environment, so as to ensure reliable user detection and channel estimation performance, and is independent of prior information such as channel and noise;

furthermore, the invention can greatly reduce the access time delay by utilizing the structural sparsity of the space-frequency domain and the angle-frequency domain of the uplink channel matrix.

In order to solve the technical problem, the invention is realized as follows:

an adaptive overhead active user detection and channel estimation method, comprising:

step 1, in each time slot, an active user adopts an authorization-free multiple access protocol to transmit signals in an uplink manner;

step 2, according to the initial time slot overhead G₀Collecting the active users received by the base station end in G₀Carrying out active user detection and channel estimation by using the received signals to obtain an active user set and a corresponding large-scale MIMO channel by using the uplink pilot signals sent in each time slot;

step 3, reconstructing a channel matrix by using the active user set and the corresponding channel obtained by estimation, and calculating a received signal error; evaluating the detection quality of the active users and the channel estimation quality according to the received signal errors, broadcasting the evaluation result to all users, if the result meets the preset standard, stopping the active users from sending pilot frequency, and only sending data in the next time slot, thereby obtaining a reliable active user set and a channel thereof; otherwise, the active user continues to send the pilot frequency, the base station end collects the received signal of one more time slot to perform active user detection and channel estimation again, and executes the step 3 again until the collected received signal is enough to obtain a reliable active user set and the channel thereof.

Preferably, the frame structure corresponding to the multiple access protocol without authorization adopted in step 1 includes two parts, namely a time domain and a frequency domain; the frequency domain is composed of N subcarriers, and the time domain is composed of T time slots; the first G time slots are used for simultaneously transmitting pilot frequency and data, and P pilot frequency sub-carriers are used for transmitting pilot frequency and (N-P) data sub-carriers are used for transmitting data in N sub-carriers of the G time slots; the last (T-G) slots are only used for transmitting data, so all N carriers are data subcarriers.

Preferably, the pilot signals sent by the active users are non-orthogonal pilot signals, and for each pilot subcarrier, the pilot signals of all the users in the pilot overhead are generated by independent and identically distributed standard complex gaussians; the pilots on different pilot subcarriers are different from each other.

Preferably, the active user detection and channel estimation performed in step 2 are:

the method comprises the steps that a subchannel matrix formed by MIMO channel vectors of all users has column sparsity, sparsity observed on different antennas is completely the same, and subchannel matrices corresponding to different pilot frequency subcarriers have the same sparse pattern, so that the problem of detection of active users is modeled into a spatial domain compressed sensing problem to form a spatial-frequency domain model;

based on the fact that cluster sparsity exists in a virtual angle domain of a large-scale MIMO channel and different sub-channel matrixes also have the same sparse pattern in a frequency domain, the problem of channel estimation of active users is modeled to be a compressed sensing problem in the angle domain to form an angle-frequency domain model;

obtaining an active user set by using a space-frequency domain model, and further obtaining two active user subsets with different reliability in the active user set through different confidence thresholds; and substituting the active user set into an angle-frequency domain model to carry out channel estimation, subtracting signals corresponding to users in an active user subset with high reliability from received signals to obtain received signal residual errors, substituting the residual errors into a space-frequency domain model to carry out residual active user detection, and repeating the steps to alternately carry out active user detection and channel estimation until a given maximum iteration number is reached or the received signal residual errors are smaller than a threshold given in advance.

Preferably, in step 2, the modeling process of the space-frequency domain is as follows:

for the p pilot sub-carrier, defining the signal of the kth user in the t time slot received by the base station as:

wherein

For the p-th pilot subchannel for the k-th user,

is the uplink multiple access pilot transmitted by the kth user on the p-th pilot subchannel,

is gaussian white noise of the p-th pilot subchannel; the base station is provided with large-scale antennas, and M is the number of the base station antennas; the user uses a single antenna; p pilot frequency subcarriers in total;

further defining the activity factor of the k user as alpha_kIf the user is active and silent, it takes 1 and 0, then the total received signal at the base station can be written as:

wherein

(·)^TIs a transposed symbol; k is the total number of users;

then, in G consecutive time slots, the received signal collected by the base station can be written as:

wherein

Is a spatial domain uplink multiple access channel matrix,

the model is a space-frequency domain compressed sensing model because G & lt K is expected, the channel matrix has column sparsity, and different columns have a common support set; knowing the received signal

And a pilot matrix

Estimating X_pTo obtain an estimated active user set

Thereby enabling active user detection.

Preferably, in step 2, the modeling process of the angle-frequency domain is as follows:

considering the transformation of the channel from the spatial domain to the virtual angular domain, the transformation relationship is:

wherein w_p,kIs virtualAn angle domain channel;

is a transform matrix, a discrete Fourier matrix in the case of a uniform linear array with antennas spaced at one-half wavelength, (-)^HIs a conjugate transposed symbol; thus, an angle-frequency domain model can be obtained:

wherein the content of the first and second substances,

is a multiple access channel matrix in the angular domain,

(·)^*is a conjugate symbol; the model is used for channel estimation; knowing the received signal

Equivalent received signals can be obtained through the transformation relation between the spatial domain and the angular domain

By using

Pilot matrix

And estimated active user set

The virtual angle domain channel corresponding to the active user can be estimated

And then the channel is transformed back to the space domain, thereby realizing the channel estimation of the active users.

Preferably, the space-frequency domain model is adopted for active user detection, the angle-frequency domain model is adopted for channel estimation, the improved AMP algorithm is adopted for resolving, the improved AMP algorithm utilizes the EM algorithm to learn the hyper-parameters, and the hyper-parameter updating rule is optimized according to the sparse structure of the channel matrix; wherein the content of the first and second substances,

sparseness ratio when the improved AMP algorithm is applied to a space-frequency domain model

The update rule of (1) is:

in formula (I), sparse rate

Representing the probability that the (p, k, m) th element of the channel matrix is nonzero, p is the index of a pilot frequency subcarrier, k is the user index, m is the index of a base station antenna, and i represents the ith iteration and the set of the algorithm

Where (q, l, u) denotes the index of the elements of the three-dimensional channel matrix, the confidence factor

Is an intermediate variable defined in the AMP algorithm; l. capillary_cRepresenting the number of elements of the set;

when the improved AMP algorithm is applied to the angle-frequency domain model, the update rule of the sparsity rate is:

in the formula (II), a

Has the advantages that:

(1) the invention utilizes the estimated active user set and the channel thereof, estimates the estimation quality by calculating the error of the received signal, can self-adaptively adjust the time slot overhead of the pilot frequency in the frame structure according to the number of active users and the channel environment in a real system, ensures the service quality by the lowest possible access time delay, and ensures the ultra-reliable active user detection and channel estimation.

(2) The invention provides a self-adaptive frame structure of an authorization-free multiple access protocol suitable for multiple carriers aiming at a broadband scene, wherein N sub-carriers in the first G time slots can simultaneously bear pilot frequency and data, so that a user can simultaneously send the pilot frequency and the data without authorization of a base station, the base station can decode the data after carrying out active user detection and channel estimation by using the pilot frequency, pilot frequency diversity is introduced among different pilot frequency sub-carriers, complex scheduling can be avoided, and access delay is greatly reduced.

(3) The invention combines the sporadic characteristic of the user uplink flow and the sparsity of the large-scale MIMO channel virtual angle domain, models the active user detection and the channel estimation as a distributed multi-measurement vector compressed sensing model of a space domain and an angle domain respectively, and alternately carries out the active user detection and the channel estimation, and can obtain an active user set and a corresponding channel thereof by utilizing the respective advantages of the two models.

(4) The invention can further reduce the access time delay and improve the active user detection and channel estimation performance by equipping a large-scale antenna on the base station and utilizing the sparsity of the large-scale MIMO virtual angle domain.

(5) The invention learns the hyper-parameters and the noise variance of the channel distribution through the EM algorithm, and is convenient to be applied in an actual system.

Drawings

Fig. 1 is a diagram illustrating a macro-dimensional multiple access scenario and a one-ring channel model in a massive MIMO system.

Fig. 2 is a schematic diagram of the structural sparsity of the space-frequency domain and angle-frequency domain uplink channel matrix in the macrodimensional multiple access.

Fig. 3 is a flow chart of the adaptive overhead active user detection and channel estimation method of the present invention.

Fig. 4 is a frame structure of the broadband unlicensed multiple access protocol of the present invention.

Fig. 5 is a comparison of the active user detection performance of the spatial domain DMMV-AMP algorithm-based macrodimensional multiple access scheme and three conventional unlicensed multiple access schemes.

Fig. 6 is a comparison of channel estimation performance of a macrodimensional multiple access scheme based on a spatial domain DMMV-AMP algorithm and three conventional unlicensed multiple access schemes.

Fig. 7 is an active user detection performance comparison of the present invention with three comparison schemes.

Fig. 8 is a comparison of channel estimation performance of the present invention with three comparison schemes.

Fig. 9 shows the performance of active user detection and channel estimation for different numbers of active users according to the multiple access scheme based on the Turbo-DMMV-AMP algorithm.

Detailed Description

The characteristics of the invention include the following aspects:

firstly, evaluating user detection and channel estimation quality by calculating received signal errors, broadcasting evaluation results to all users, and determining whether to continue to send pilot frequency or not by active users according to the evaluation results, thereby adaptively adjusting pilot frequency overhead; thanks to the scheme of adaptive pilot overhead, the invention can ensure that all active users reliably access the network with the smallest possible access delay.

Secondly, the invention utilizes the sporadic characteristic of the uplink transmission flow of the user and the sparsity of a large-scale MIMO virtual angle domain channel, and models the problems of active user detection and channel estimation at the base station end into the problems of distributed multi-measurement vector compressed sensing in a space-frequency domain and an angle-frequency domain respectively by designing a multi-access pilot signal and a frame structure of a broadband authorization-free multi-access protocol; and then, the provided Turbo-DMMV-AMP algorithm is used for alternately carrying out active user detection and channel estimation to obtain an active user set and a channel vector thereof. By the aid of the alternative estimation mode, sparse structures of a space-frequency domain and an angle-frequency domain can be fully utilized, and pilot frequency overhead can be greatly reduced.

In addition, the EM algorithm is introduced to learn the parameters of the real channel distribution and the noise power, so that the EM algorithm is more beneficial to the realization of an actual system.

The invention is explained in detail below with reference to the drawings and examples.

The present invention considers a typical uplink macro-dimensional multiple access scenario in a massive MIMO system, as shown in fig. 1. The system consists of a base station and K user equipments, usually K is large, the base station end is equipped with a uniform linear array of M antennas and the user end only considers a single antenna without loss of generality. In order to support a large amount of traffic load, an OFDM (Orthogonal Frequency Division Multiplexing) transmission technique of N subcarriers is adopted, and P pilots are uniformly inserted into the N subcarriers for active user detection and channel estimation. For the p pilot sub-carrier, the signal transmitted in the t time slot (t OFDM symbol) by the k user received at the base station end is

Here, the

Represents the subchannel corresponding to the p pilot subcarrier of the k user,

is the uplink multiple access pilot signal transmitted by the kth user on the pth pilot subchannel,

is additive white gaussian noise for the p-th pilot subchannel. For a typical macro-dimensional multiple access scenario, only a small fraction of the user equipments are active and require uplink access to the network within a given channel coherence time. We define the activity factor of user k as α_kWhen the user is active, 1 is taken, otherwise 0 is taken; at the same time, define the set of active users as

The number of active users is

·_cRepresenting the number of collection elements. Then, at the p pilot sub-carrier and t time slot, the base station receives signals from all active users as

Wherein

Modeling a channel as

Where ρ is_kLarge scale fading due to path fading and shadow fading,

is a small scale fading. For the p pilot sub-carrier, the sub-channel of the k user is

Where L is the number of multipath components, β_k,lIs the complex gain of the path(s),

in order to be a time delay of the path,

is the double sideband bandwidth. The base station antenna array response vector is

Here, the

Is the angle of arrival of the ith path of the kth user, λ is the propagation wavelength,

is the critical antenna spacing.

Due to the sporadic nature of the user's upstream traffic, only a small fraction of users are active, i.e., K_aK is smaller than K. Thus, a channel matrix is defined

The channel vector corresponding to the p pilot sub-carrier and the m antenna

Is to be of a sparse nature and,

|supp{[X_p]_:,m}|_c＝K_a＜＜K (4)

furthermore, the sparsity is the same for all antennas,

supp{[X_p]_:,1}＝supp{[X_p］_:,2}＝…＝supp{［X_p］_:,M}, (5)

supp {. denotes a supporting set of vectors or matrices, and we refer to this sparsity as the spatial domain structured sparsity of the macrodimensional multiple access channel matrix. Because of the fact that

Is the same for all sub-channels and is therefore

Having a common set of supports in the frequency domain,

supp{X₁}＝supp{X₂}＝…＝supp{X_P} (6)

we call the combined structured sparseness of (5) and (6) as

To illustrate this sparsity, we assume a total of 10 users, K being K _a3 users are active and require uplink access to the network, the base station is equipped with M16 antennas, then,

the structural sparsity of (a) is shown in fig. 2 (a). On the other hand, the base station is generally erected at a high position, the surrounding scatterers are few, the user equipment is generally positioned at a low position far away from the base station, the surrounding scatterers are rich, and the characteristic can be modeled by using a one-ring channel model. Considering a user at a distance R from the base station, with scatterers mainly distributed in a circle with radius R around it, the angular spread Δ ≈ arctan (R/R) observed at the base station will be very small, since R > R. This will result in sparsity of the massive MIMO virtual angle domain channel. In particular, the user virtual angle domain channel

And spatial domain channel

Can be expressed as

Where A is_RIs a transformation matrix depending on the base station antenna array geometry, a when a uniform linear array of critical antenna spacings is employed_RIs a DFT matrix. Due to the large antenna array and the extremely small angular spread, i.e., Δ < M, the channel vectors are sparse,

and such sparseness is clustered as shown in fig. 2 (b). Moreover, because the spatial propagation characteristics of the channels are very similar across the system bandwidth, the subchannels on different subcarriers share similar scatterers, resulting in

Having the same sparse pattern in the frequency domain

We call the combined structured sparseness of (8) and (9) as

Angle-frequency domain structuring sparsity. Further, a virtual angle domain channel matrix is defined

Wherein

Considering the sparseness of the space-frequency domain structuring in (4) and (6), one can obtain | suppp { [ W { [_p]_:,m}|_c< K, and

supp{W₁}＝supp{W₂}＝…＝supp{W_P} (10)

fig. 2(b) considers both the structured sparsity of the space-frequency domain and the angle-frequency domain with respect to the example of the virtual angle domain channel matrix. These sparse characteristics described above will be used in the present invention to achieve ultra-reliable low-latency active user detection and channel estimation.

Based on the above analysis, the active user detection and channel estimation process of the present invention is described in detail in steps with reference to fig. 3.

Step 1, active user sends multiple address pilot signal in uplink

In each time slot, the active users that need to access the network send non-orthogonal multiple access pilot signals directly to the base station, which are designed in advance based on DCS theory and are known to the base station side. The system adopts an authorization-free multiple access protocol. The invention further designs an authorization-free multiple access protocol aiming at the broadband system. The existing multiple access schemes all consider narrow-band macro address communication of single carrier transmission, cannot support macro connection borne by massive service flow, and is a more challenging problem for reliable support of macro-dimensional multiple access in a broadband system. To solve this problem and to take advantage of some of the characteristics in broadband systems, the present invention performs the following steps.

Step 1.1, design the frame structure of the broadband authorization-free multiple access protocol

For a massive MIMO-OFDM system, the frame structure of the broadband unlicensed multiple access protocol designed by the present invention includes two parts, namely a time domain and a frequency domain, as shown in fig. 4. The frequency domain is composed of N subcarriers, and the time domain is composed of T time slots. The first G slots are used for transmitting pilot and data simultaneously, and of the N subcarriers, P pilot subcarriers are used for transmitting pilot, and (N-P) data subcarriers are used for transmitting data. The last (T-G) slots are only used for transmitting data, so all N carriers are data subcarriers. Because N sub-carriers in the first G time slots can simultaneously bear pilot frequency and data, a user can simultaneously send the pilot frequency and the data without authorization of the base station, the base station can decode the data after performing active user detection and channel estimation by using the pilot frequency, and the frame structure can avoid complex scheduling and greatly reduce access delay.

Meanwhile, the time slot overhead G used by the pilot frequency is adaptively adjusted according to the evaluation result of the base station on the detection of the active users and the channel estimation quality. See step 3 for the adjustment process.

Step 1.2, designing broadband multiple access pilot signal

All user designed multiple access pilot signal

Will be used as a pressure when modeling at the base station endAnd (4) shrinking the perceived observation matrix. The observation matrix is very important to the performance of compressed sensing reconstruction, so that the design of multiple access pilot frequency is needed. In order to satisfy the RIP (Restricted Isometry Property) condition of the observation matrix, for the p-th pilot subcarrier, the pilots transmitted in multiple time slots of all users are generated by the complex gaussian distribution with the same distribution independently. At the same time, the pilots on different pilot subcarriers must be different from each other to introduce diversity between subcarriers, i.e. the observation matrix

Different from each other for different p. DCS theory states that this diversity can further improve compressed sensing reconstruction performance.

Step 2, active user detection and channel estimation

This step is based on the initial slot overhead G₀Collecting the active users received by the base station end in G₀And performing active user detection and channel estimation on the uplink signals sent in each time slot according to the received signals and the known pilot frequencies of all users to obtain an active user set and a corresponding large-scale MIMO channel.

In the prior art, a space domain model is adopted to simultaneously perform active user detection and channel estimation. In the invention, the problems of active user detection and channel estimation are modeled into DMMV (Distributed Multiple Measurement Vector) compressed sensing problems in a space-frequency domain and an angle-frequency domain respectively at a base station. Then, obtaining an active user set according to the space-frequency domain model (according to the set, two active user subsets with different reliability can be further obtained by setting different confidence thresholds, wherein the activity reliability of one subset user is high, see the description of the module A in step 2.3); substituting the active user set into an angle-frequency domain model to estimate to obtain a corresponding channel vector, subtracting a receiving signal corresponding to an active user with high reliability from the receiving signal according to the estimated channel to obtain a receiving signal residual error, and substituting the residual error into a space-frequency domain model to estimate the rest active users, so that the number of active users to be detected by an active user detection part is less and less; by analogy, active user detection and channel estimation are alternately performed to obtain an active user set and a channel vector thereof.

Step 2.1, active user detection and channel estimation modeling

At the base station, the signals received in G consecutive time slots can be written as

Wherein the content of the first and second substances,

for unlicensed random multiple access, in order to avoid complex scheduling and resulting access delay, the base station needs to receive signals according to the received signals

And a pilot matrix

Fast and accurate estimation of active user set

And its corresponding channel vector

The problem can be equated with estimating the channel matrix based on the model (11)

In order to solve the above problem, conventional estimation methods, such as MMSE linear estimator, require G ≧ K, and,

must be a unitary matrix. This means that orthogonal pilots need to be assigned to the usersThe base station can conveniently identify different users, the number of users in the macro connection is huge, the channel coherence time is limited, and orthogonal pilot frequencies cannot be distributed to all users, namely G is larger than or equal to K and cannot be realized. To solve this problem, the present invention makes use of

In the case of G < K, (11) becomes a spatial domain DMMV compression perception problem. Then, the received signal is known

And a pilot matrix

Estimating X_pTo obtain an estimated active user set

Thereby enabling active user detection.

Meanwhile, in order to further utilize the angle-frequency domain structural sparsity of the uplink channel matrix, the space domain model (11) can be further written as a relation (7) between the space domain and the virtual angle domain according to the massive MIMO channel

Wherein the content of the first and second substances,

is a virtual angle domain channel matrix. The model is a DMMV compression perception problem corresponding to a virtual angle domain. Knowing the received signal

By using

Pilot matrix

And estimated active user set

Because the model (12) simultaneously considers the space-frequency domain and angle-frequency domain structural sparsity of the uplink channel matrix and the channel X in the space domain model (11)_pIn contrast, channel W_pMore sparse but the common sparsity observed between different antennas is destroyed. The existing compressive sensing-based macrodimension multiple access scheme only considers X_pAnd (3) sparse spatial domain and utilizes (11) implementation for joint active user detection and channel estimation. To make full use of X_pEnhanced sparse structure and W_pThe gain from the increased sparsity is achieved by using (11) for active user detection and (12) for corresponding channel estimation. Specifically, the invention firstly improves the AMP-NNSPL algorithm, proposes the DMMV-AMP algorithm for jointly estimating the three-dimensional sparse matrix, the algorithm is the model expressed by (11) and (12), can be used for solving the compressed sensing problem (11) or (12) and realizing simultaneous active user and channel estimation (for the AMP-NNSPL algorithm, the specific reference is the translation name: the sparse massive MIMO-OFDM channel estimation is realized by using the approximate common support set, and the author, English name and place are Lin X, Wu S, et al]IEEE Communication Letters,2017,21(5): 1179-. Then, based on the DMMV-AMP algorithm, the invention provides a method for alternately carrying out active user and channel estimationThe algorithm is called Turbo-DMMV-AMP algorithm. As an enhanced version of DMMV-AMP, the Turbo-DMMV-AMP algorithm improves performance by alternating active users and channel estimation.

Step 2.2 DMMV-AMP Algorithm

The invention adopts a space-frequency domain model to detect active users and adopts an angle-frequency domain model to estimate channels, and adopts an improved AMP algorithm to solve, the improved AMP algorithm utilizes an EM algorithm to learn hyper-parameters, and the hyper-parameters are calculated according to a channel matrix (referring to a space domain channel X)_pOr angular domain channel W_p) The sparse structure of (3) optimizes the hyper-parameters.

For convenience of description, the DMMV-AMP algorithm of the present invention is described herein by taking the model (11) as an example. First we define x_p,k,m＝[X_p]_k,mThen X_pThe minimum mean square error estimate of (a posteriori) can be expressed as

To simplify the expression, here we ignore x_p,k,m，X_pAnd

subscript index p and superscript G in (1), wherein the edge posterior probability can be written as

p(x_k,m|Y)＝∫p(X|Y)dX_\k,m, (14)

Where X is_\k,mRepresentation collection

And the joint posterior probability is calculated by the Bayes formula

Wherein

Is a normalized constant factor. Under the assumption of white Gaussian noise, the likelihood function is

Where σ is the noise power. In the present invention, we assume a flexible pin plate prior model,

the model matches well with the true distribution of the channel matrix, here 0<γ_k,m<1 denotes the sparsity ratio, i.e. x_k,mProbability not equal to 0, (. cndot.) denotes the Dirichlet function, f (x)_k,m) Is the distribution of non-zero elements in the channel matrix.

The above estimator is difficult to implement because of the huge number of users in macrodimension multiple access, and the multi-dimensional integral with dimension proportional to the number of users is involved in equation (14). To solve the above challenges, the main idea of the DMMV-AMP algorithm in the present invention is to obtain the edge posterior probability p (x) using the AMP algorithm_k,mY) is determined. Specifically, the algorithm consists of the following five parts:

1. factor node message update

The invention expresses the factor relation in the formula (15) by a factor graph, and after AMP algorithm approximation, the message is updated at the factor node

And

can be expressed as

Where, i denotes the number of iterations of the algorithm,

is the posterior variance in equation (28), the initialization involved in the first iteration being

2. Variable node message update

Correspondingly, the message updated at the variable node can be expressed as

3. Posterior mean and posterior variance calculation

According to the update results of the factor node and the variable node, x_k,mCan be approximated as

In the present invention we further assume that the channel gains all follow a complex gaussian distribution, i.e. f (x)_k,m)＝CN(x_k,m(ii) a μ, τ), the posterior probability in equation (22) can be further simplified to

Wherein the content of the first and second substances,

finally, x_k,mCan be calculated as

Here, we refer to the intermediate variables

Called confidence factor, when DMMV-AMP achieves a reliable compressed perceptual reconstruction, we conclude that

Thus, the confidence factor may be used to indicate whether the (k, m) -th element of the channel matrix is zero or not.

4. Learning hyper-parameters using EM algorithm

The above equations (18) - (28) are basic iteration steps of the DMMV-AMP algorithm, but specific parameters and noise power of the real channel distribution need to be known, which is also a disadvantage of the current macro-dimensional multiple access scheme based on the AMP algorithm. In order to facilitate the application in the actual system, the invention utilizes EM algorithm to carry out the hyper-parameter pair

Learning, EM algorithm is composed ofThe method comprises the following two steps:

in the formula (30), the first and second groups,

expressed in known Y and thetaⁱConditional expectations in the case that the expectation is with respect to the joint posterior probability p (X | Y; theta)ⁱ) In (1). The above EM algorithm has two problems in the practical application process: in macrodimensional multiple access, p (X | Y; theta)ⁱ) The computational complexity of (a) is unacceptable; the joint optimization of the four parameters in θ is very complex. In order to solve the problems, the invention obtains the result by an AMP algorithm

Meanwhile, for the updating of the parameters in the theta, only one parameter is updated at a time, and other parameters are fixed and regarded as constants. By taking the derivative of equation (30) for a particular parameter and taking the reciprocal to zero, an update rule for the hyperparameter can be obtained as,

5. optimizing hyper-parameter update rules according to sparse structures of channel matrices

Notably, the sparsity ratio γ_p,k,mRepresenting the channel matrix X_pOr W_pThe (k, m) th element of (a) is non-zero. In the formula (34), for γ_p,k,mThe updating of (1) is performed independently for all p, k, m, without considering the structured sparsity of the uplink channel matrix. From here, we re-add the index of the pilot subcarrier for convenience of description. Due to the structural sparsity of the uplink channel matrix, sparsity rates corresponding to partial elements are highly correlated rather than completely independent. To this end, the invention introduces a set of indices

To correct the update rule of the sparse rate, consider it to be in the set

Sparsity ratio of channel elements in (1) and x_p,k,mThe same applies.

For the DMMV compressed sensing model (11), only the space-frequency domain structured sparsity of the channel matrix is considered, as shown in equation (6) and fig. 2 (a). In order to exploit this sparsity, the invention proposes that all channel elements corresponding to the same user have the same sparsity rate, i.e. for a particular k, γ, the same channel element has the same sparsity rate_p,k,mThe same for any p and m. Therefore, when the DMMV-AMP algorithm is applied to the model (10), the update rule of the sparsity ratio is modified to

Wherein

In contrast, the DMMV compressed sensing model (12) considers both the space-frequency domain and angle-frequency domain structured sparsity of the channel matrix, as shown in fig. 2 (b). According to the characteristic of sparse large-scale MIMO channel cluster in virtual angle domain and the channel matrix W_pAs shown in equations (8) - (10), the present invention considers the channel matrix element w_p,k,mIf the nearby element is non-zero, the element probability is also non-zero, otherwise, the element probability is zero. Thus, w_p,k,mWith the same sparseness rate as its neighbors. Define the neighbors of the element as

When the DMMV-AMP algorithm is applied to the model (12), the update rule of the sparsity ratio is modified to

Step 2.3 Turbo-DMMV-AMP algorithm

To obtain a channel matrix

Is defined as

The DMMV-AMP algorithm can be applied directly to the model (11). Alternatively, the DMMV-AMP algorithm is applied to the model (12), and the estimation is carried out

Then obtained by transformation in formula (7)

Is obtained by

Thereafter, the active user set and its channel vector can be acquired simultaneously, i.e.

The supporting set of (c) and the corresponding element value. But neither of the above two methods can fully utilize the sparse structure of the channel matrix. The invention adopts two modules to alternatively use the models (11) and (12) for iteration, and respectively carries out active user detection and channel estimation based on the DMMV-AMP algorithm, thereby further reducing pilot frequency overhead, and the alternative estimation mode is named as Turbo-DMMV-AMP algorithm. The algorithm consists of the following two modules:

1. a module A: active user detector based on DMMV-AMP algorithm

In each Turbo iteration, the module A applies the DMMV-AMP algorithm to the spatial domain model to obtain a confidence factor

(j means the jth Turbo iteration), two active user sets with different reliability are obtained according to the confidence factor, and the two sets are transmitted to the module B as messages.

Specifically, at the first Turbo iteration (j ═ 1), module a applies the DMMV-AMP algorithm to model (11), and obtains the confidence factors corresponding to the channel elements. Because when x_p,k,mConfidence factor when not equal to 0

Taking 1, otherwise taking 0, i can design the active user detector according to this property of the confidence factor. First, a threshold function r (x;) is defined when | x;,/Y;)>And r (x;) is 1, otherwise r (x;) is 0. The invention then designs an active user detector based on the confidence factor obtained,

where p is_thIs set to 0.9 and 0<<1 is reasonably set according to different requirements, and lower false alarm probability can be caused, and higher false alarm probability can be caused. Then, the present inventionBy setting different thresholds, two active user sets with different reliability are obtained: using thresholds_detGet coarse active user set omega 0.4, use threshold_rel0.9 obtaining a reliable set of active users xi^jAs shown in the following equation:

wherein

Is calculated according to the formula (39),

may find xi^jIs a subset of Ω. Module a then passes these two sets as messages to module B.

2. And a module B: channel estimator based on DMMV-AMP algorithm

Using the rough active user set Ω from module A and module B to estimate the channels of the users in this set based on the model (12), in particular, we should solve the compressed sensing problem as follows

Wherein the content of the first and second substances,

and

are respectively

And W_pThe sub-matrix of (a) is,

while

Is the set of all users. The channel estimator of the module only considers the users in the set omega, so the dimension of a channel matrix corresponding to the uplink channel estimation of the users is reduced. At the same time, due to the structured sparsity of the large-scale MIMO channel angle-frequency domain, the low-dimensional channel matrix [ W ]_p]_Ω,I.e., still sparse. The invention solves the compressed sensing problem (41) by using the proposed DMMV-AMP algorithm, estimates the low dimensional channel matrix, removes the received signal corresponding to some reliable active users, i.e. from the xi^j-1Receiving signals of partial users and residual error of the received signals

Passed to module a as a message.

Is calculated by the following formula

Wherein the elements in the set are selected from the set xi^jChinese character is randomly selected, and | luminance_c/|Ξ^j|_cAt 0.8, only the signal of an active user is removed here to avoid algorithm divergence.

In the next Turbo iteration (j >1), the active user detection problem in module A becomes

Wherein the content of the first and second substances,

is the received signal residual for the jth Turbo iteration,

is the residual of the channel matrix and,

is defined as

While

Here, the

Is the channel matrix estimated by the last Turbo iteration module B. The invention is iteratively performed on both modules in an alternating manner, since as the iteration progresses,

becoming increasingly sparse and the channels of the users in the set Ω are also constantly re-estimated, so this alternating approach significantly reduces the pilot overhead. When the iteration terminates, an estimated active user set may be obtained

And its corresponding channel vector

Step 3, evaluating the quality of active user detection and channel estimation

Reconstructing a channel matrix by using the active user set and the corresponding channel estimated in the step 2, and calculating a received signal error; evaluating the detection quality of the active users and the channel estimation quality according to the received signal errors, broadcasting the evaluation result to all users, if the result meets the preset standard, stopping the active users from sending pilot frequency, and only sending data in the next time slot, thereby obtaining a reliable active user set and a channel thereof; otherwise, the active user continues to send the pilot frequency, the base station end collects the received signal of one more time slot to perform active user detection and channel estimation again, and executes the step 3 again until the collected received signal is enough to obtain a reliable active user set and the channel thereof.

The method comprises the following three steps:

step 3.1, reestablish and estimate the matrix of the uplink multiple access channel

Reconstructing an estimated uplink multiple access channel matrix using an estimated set of active users and corresponding channels

The specific process is as follows:

while

Step 3.2, calculating the estimation error of the received signal

Calculating the estimation error of the received signal by using the estimated channel matrix, wherein the process comprises the following steps:

step 3.3, evaluating the estimated quality

And evaluating the quality of active user detection and channel estimation according to the estimation error of the received signal, and broadcasting the evaluation result to all users. If it is

And (3) the estimation is unreliable, the active users continue to send the pilot frequency, the base station collects the received signals of one more time slot to carry out active user detection and channel estimation again, and the step 3 is repeated until a reliable active user set and channel estimation are obtained. Where e is a criterion given in advance for evaluating the quality of the estimate.

The above is the adaptive overhead active user detection and channel estimation method disclosed in the present invention.

Compared with the existing grant-free-based macrodimension multiple access scheme (here, the joint active user and channel estimation methods based on the GSP algorithm, the SOMP algorithm, and the DSAMP algorithm are considered, and these schemes only consider the sparsity of the spatial domain structure), in order to embody the advantage of the present invention in reducing the access delay, the effect of the present invention is illustrated with fig. 5 to fig. 9. Meanwhile, in order to embody an estimation mode of alternately performing active user detection and channel estimation by using the sparsity of the virtual angle domain and a performance gain brought by an adaptive pilot overhead scheme, three contrast schemes are additionally proposed herein, which are summarized as follows:

(1) active user detection and channel estimation based on the spatial domain DMMV-AMP algorithm: applying the DMMV-AMP algorithm to a spatial domain compressed sensing model (11) to obtain an estimate of a spatial domain channel matrix

And its confidence factor, and utilizes the detector shown in formula (31) to perform active user detection. The space-frequency domain structural sparsity of the channel matrix caused by sporadic characteristics of user uplink traffic is only utilized.

(2) Active user detection and channel estimation based on the angle domain DMMV-AMP algorithm: applying the DMMV-AMP algorithm to an angle domain compressed sensing model (12) to obtain an estimate of an angle domain channel matrix

And its confidence factor, and utilizes the detector shown in formula (31) to perform active user detection. Here, both the space-frequency domain and angle-frequency domain structuring sparsity of the channel matrix are exploited.

(3) Active user detection and channel estimation based on Turbo-DMMV-AMP algorithm: the scheme jointly utilizes models (11) and (12) to alternately perform active user detection and channel estimation by utilizing space frequency domain and angle-frequency domain structural sparsity respectively. The only difference between this scheme and the present invention is that the pilot slot overhead in this scheme is fixed, whereas the pilot slot overhead in the present invention is adaptive according to the active user detection and channel estimation quality estimation results.

FIG. 5 compares the active user detection performance of the DMMV-AMP algorithm based scheme with the other three conventional schemes based on the spatial domain model, and the performance is determined by the detection error probability P_eMeasure, define as

As can be seen from fig. 5, the scheme based on the DMMV-AMP algorithm is significantly superior to the other three conventional schemes. Thus. When considering the same performance, the DMMV-AMP algorithm can significantly reduce the access delay, e.g., for detection error probability less than 10^-5In this case, the GSP algorithm with the best performance in the conventional scheme requires at least the pilot overhead of G-72, while the DMMV-AMP algorithm requires only the overhead of G-58, which means that the algorithm can reduce the access delay by approximately 19%. Also, by equipping the base station with more antennas or using larger antennas

(Here, the

Referring to the number of pilot subcarriers used for active user detection and channel estimation), the DMMV-AMP algorithm may achieve better performance due to a larger number of base station antennas or

The structured sparsity of the channel matrix is enhanced. Fig. 6 compares the MSE (Mean Square Error) of the channel estimation, and further illustrates the superiority of the DMMV-AMP algorithm in improving the channel estimation performance.

For the present invention and the three proposed comparison schemes, fig. 7 and 8 compare their active user detection and channel estimation performance, respectively. It can be seen that at pilot overhead G>At 18, the performance of the angle domain model (12) based scheme may be significantly better than that of the angle domain model (11) based scheme, while at G<18 is the opposite. This is because, the model (12)The structural sparsity of the space-frequency domain and the angle-frequency domain are simultaneously considered, so that the corresponding channel matrix

Is compared with that in the model (11)

More sparsely populated, but sparse structures on different antennas are destroyed. Therefore, reliability is obtained in (12)

The minimum observation (i.e. pilot overhead) required for reconstruction is much smaller than for the model (11). Under the condition of insufficient observation values, the structural sparsity of the channel matrix (11) in the model on different antennas brings performance gain to compressed sensing reconstruction, and therefore better performance can be obtained based on the spatial domain model (11) under the condition of low overhead. The above conclusion shows that the space-frequency domain and angle-frequency domain structural sparsity cannot be fully utilized by using the space domain model or the angle domain model alone, but the scheme based on the Turbo-DMMV-AMP algorithm jointly utilizes the two models, as can be seen from fig. 7 and 8, the active user detection and channel estimation performance of the scheme is better than that of the scheme using the space domain or the angle domain model alone in all the time slot overhead ranges. Additionally, compared with the three contrast schemes, the active user detection and channel estimation scheme of the adaptive overhead of the invention can adaptively adjust the pilot frequency time slot overhead, and ensure reliable active user detection and channel estimation performance. The proportion of the different pilot overheads of the present invention is marked in fig. 8.

Fig. 9 compares the user detection and channel estimation performance of the non-adaptive multiple access scheme based on the Turbo-DMMV-AMP algorithm in different numbers of active users, and labels the average pilot time slot overhead required by the present invention in each case, so that it can be seen that the detection error probability and channel estimation MSE of the non-adaptive scheme will increase with the increase of the number of active users, and the present invention adaptively adjusts the pilot time overhead to ensure that all active users can be correctly detected and obtain reliable channel estimation, which embodies the robustness of the present invention in the time-varying scene of the actual number of active users and ensures that users can quickly and reliably access the network.

In summary, the above description is only a preferred embodiment of the present invention, and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for adaptive overhead active user detection and channel estimation, comprising:

the frame structure corresponding to the authorization-free multiple access protocol comprises a time domain part and a frequency domain part; the frequency domain is composed of N subcarriers, and the time domain is composed of T time slots; the first G time slots are used for simultaneously transmitting pilot frequency and data, and P pilot frequency sub-carriers are used for transmitting pilot frequency and (N-P) data sub-carriers are used for transmitting data in N sub-carriers of the G time slots; the last (T-G) time slots are only used for transmitting data, so all N carriers are data subcarriers; the time slot overhead G used by the pilot frequency is adaptively adjusted according to the evaluation result of the base station on the detection of the active user and the channel estimation quality;

the pilot signals sent by the active users are non-orthogonal pilot signals, and for each pilot subcarrier, the pilot signals of all the users in the pilot overhead are generated by independent and identically distributed standard complex Gauss; the pilot frequencies on different pilot frequency sub-carriers are different from each other;

step 2.1, active user detection and channel estimation modeling

1) The method comprises the steps that a subchannel matrix formed by MIMO channel vectors of all users has column sparsity, sparsity observed on different antennas is completely the same, and subchannel matrices corresponding to different pilot frequency subcarriers have the same sparse pattern, so that the problem of detection of active users is modeled into a spatial domain compressed sensing problem to form a spatial-frequency domain model;

the modeling process of the space-frequency domain is as follows:

wherein

For the p-th pilot subchannel for the k-th user,

wherein

(·)^TIs a transposed symbol; k is the total number of users;

wherein

Is a spatial domain uplink multiple access channel matrix,

And a pilot matrix

Estimating X_pObtaining an estimated active user set so as to realize active user detection;

2) based on the fact that cluster sparsity exists in a virtual angle domain of a large-scale MIMO channel and different sub-channel matrixes also have the same sparse pattern in a frequency domain, the problem of channel estimation of active users is modeled to be a compressed sensing problem in the angle domain to form an angle-frequency domain model;

the modeling process of the angle-frequency domain is as follows:

wherein w_p,kIs a virtual angle domainA channel;

wherein the content of the first and second substances,

is a multiple access channel matrix in the angular domain,

By using

Pilot matrix

And estimated active user set

Then the channel is converted back to the space domain, thereby realizing the channel estimation of the active user;

3) the method comprises the following steps of adopting a space-frequency domain model to detect active users, adopting an angle-frequency domain model to estimate channels, adopting an improved AMP algorithm to solve, learning hyper-parameters by using an EM algorithm through the improved AMP algorithm, optimizing a hyper-parameter updating rule according to a sparse structure of a channel matrix, and calling the process as a DMMV-AMP algorithm; wherein the content of the first and second substances,

The update rule of (1) is:

in formula (I), sparse rate

in the formula (II), a

Defining a module A: active user detector based on DMMV-AMP algorithm

Defining a module B: channel estimator based on DMMV-AMP algorithm

In each Turbo iteration, the module A obtains an active user set by using a space-frequency domain model, further obtains two active user subsets with different reliability in the active user set through different confidence thresholds, and transmits the two active user subsets to the module B as messages; the module B substitutes the active user set into an angle-frequency domain model for channel estimation, subtracts signals corresponding to users in the active user subset with high reliability from received signals to obtain received signal residual errors, and transmits the residual errors serving as messages to the module A; substituting the module A into a space-frequency domain model to perform residual active user detection, and performing active user detection and channel estimation alternately by analogy until a given maximum iteration number is reached or a received signal residual error is smaller than a threshold given in advance;