Entity and method for computing a diagonal loading factor of a weight matrix used for interference rejection down link precoding
Patent Information
- Application Number
- PCT/CN2023/101174
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-06-19
- Publication Date
- 2024-12-26
Smart Images

Figure CN2023101174_26122024_PF_FP_ABST
Abstract
Description
ENTITY AND METHOD FOR COMPUTING A DIAGONAL LOADING FACTOR OF A WEIGHT MATRIX USED FOR INTERFERENCE REJECTION DOWN LINK PRECODINGTECHNICAL FIELD
[0001] The present disclosure relates to an entity and method for computing a diagonal loading factor of a weight matrix that is used for interference rejection down link precoding (IRDL precoding) .BACKGROUND
[0002] In wireless communication systems multi input multi output (MIMO) allows increasing capacity of radio transmission by using multiple antennas. For example, massive MIMO is a type of wireless communica-tion technique, where entities for communication, such as base stations, comprise a very large number of antennas for improving spectral and energy efficiency.SUMMARY
[0003] To unleash the potential of massive MIMO, estimation of channel state information (CSI) at each base station (BS) of the communication system, e.g. a communication network, may be performed. Time-divi-sion duplex (TDD) may be considered for allowing estimation of downlink (DL) CSI by means of uplink (UL) training according to channel reciprocity. Herein, downlink and uplink may be abbreviated by “DL” and “UL” , respectively. To obtain reliable channel estimates, mutually orthogonal pilot sequences may be allocated to different user equipments (UEs) . The term “user equipment” may be abbreviated by the term “user” . However, in massive connectivity networks, this may be difficult in multi-cell scenarios because the orthogonal pilot sequence length, which is limited by the coherence interval, needs to be proportional to the total number of users in the whole network. As a result, the orthogonal pilot sequences need to be reused among adjacent cells leading to pilot contamination that significantly degrades the channel estima-tion performance.
[0004] Pilot reuse planning may reduce pilot contamination, but it requires careful centralized cell cooperation, which may be difficult in massive connectivity networks. Using non-orthogonal pilots may allow scala-bility for massive connectivity, e.g. internet of things (IoT) networks, without requirement of centralized network planning.
[0005] The main features of modern and future massive connectivity multi-cell massive MIMO networks may be as follows: The number of users may be very high, but only a fraction of them may be simultaneously ac-tive. The number of base station antennas may be very high. Coherence interval for TDD operation may be very limited due to users’ mobility. Available number of pilots per coherence interval may be very limited to maintain spectral efficiency.
[0006] Therefore, a number of available training samples may be lower than a number of base station antennas and a total number of signals and interference sources leading to an undersampled scenario.
[0007] A main technical issue may be how to design interference rejection (IR) DL precoding for massive con-nectivity TDD MIMO networks in the undersampled scenario without centralized network cooperation when the base stations of the network only have the knowledge of the local pilots employed in the cells or spectrum sharing sub-networks.
[0008] In view of the above, this disclosure aims to provide an entity that allows interference rejection (IR) downlink (DL) precoding for massive connectivity MIMO networks, such as massive connectivity TDD MIMO networks. An objective of this disclosure is to provide such an entity that allows IRDL precoding in the aforementioned under sampled scenario. Another objective of this disclosure is to provide such an entity that allows IRDL precoding without needing a centralized network cooperation when base stations of the massive connectivity MIMO network only have knowledge of local pilots employed in the cells or spectrum sharing sub-networks.
[0009] These and other objectives are achieved by the solution of this disclosure as described in the independent claims. Advantageous implementations are further defined in the dependent claims.
[0010] A first aspect of this disclosure provides an entity for computing a diagonal loading factor of a weight matrix that is used for interference rejection down link precoding in an area comprising Mtot user equip-ments (UEs) , each comprising at least one antenna. The entity comprises K antennas. The entity is config-ured to wirelessly receive, during a pilot interval, via a channel from M active UEs of the Mtot UEs M pilot sequences of Tp symbols. The entity is configured to compute the diagonal loading factor by solving an equation for the diagonal loading factor. The equation comprises a monotonic function of a likelihood ratio of a regularized (Tp-M) × (Tp-M) matrix, and the monotonic function of the likelihood ratio equals an average value of the monotonic function of the likelihood ratio comprising as input a K× (Tp-M) matrix of independent Gaussian variables with zero mean and unit power. The diagonal loading factor is a real number greater than or equal to zero; K, Mtot, M and Tp are each an integer greater than zero; Mtot is greater than K; K is greater than M, and Tp is greater than M.
[0011] In other words, the entity according to the first aspect allows computing a weight matrix by computing a diagonal loading factor of the weight matrix. Since the weight matrix is used for interference rejection (IR) down link (DL) precoding, the entity allows IRDL precoding for massive connectivity MIMO net-works, such as massive connectivity TDD MIMO networks. Herein, the pilot interval may be referred to as “uplink (UL) pilot interval” .
[0012] A user equipment (UE) may be understood as any communication device that may be used directly by an end-user for communication, e.g. wireless communication. The wireless communication may be radio communication. For example, the user equipments (UEs) may comprise at least one of mobile phones, smart phones, laptop computers, tablet computers etc.
[0013] The entity may be a communication device for communication, e.g. wireless communication. The wire-less communication may be radio communication.
[0014] The term “activated user equipment” may be used as a synonym for the term “active user equipment” . The term “slot” may be used as a synonym for the term “interval” .
[0015] In an implementation form of the first aspect, the entity is configured to compute the weight matrix using the computed diagonal loading factor.
[0016] This allows IRDL precoding for massive connectivity MIMO networks. The entity may be configured to compute the weight matrix using the computed diagonal loading factor, a K×Tp matrix of the signal re-ceived by the entity during the pilot interval, and a M×Tp matrix of the M pilot sequences.
[0017] In an implementation form of the first aspect, the entity is configured to compute a K×M downlink pre-coding matrix using the weight matrix. This allows IRDL precoding for massive connectivity MIMO net-works.
[0018] In an implementation form of the first aspect, the entity is configured to solve the equation for the diago-nal loading factor by computing a matrix that is the interference plus noise covariance matrix estimate during the pilot interval.
[0019] In an implementation form of the first aspect, the entity is configured to compute the matrix that is the in-terference plus noise covariance matrix estimate during the pilot interval using a K×Tp matrix of a sig-nal received by the entity during the pilot interval, a M×Tp matrix of the M pilot sequences, and an estimated channel matrix of the channel.
[0020] In an implementation form of the first aspect, the entity is configured to compute the regularized (Tp-M) × (Tp-M) matrix using Tp-M linearly independent columns of the computed matrix; and solve the equation for the diagonal loading factor using the computed regularized (Tp-M) × (Tp-M) matrix.
[0021] The average value of the monotonic function of the likelihood ratio of the actual matrix may be computed by computing the average value of the monotonic function of the likelihood ratio comprising as input the K× (Tp-M) matrix of independent Gaussian variables with zero mean and unit power.
[0022] In an implementation form of the first aspect, K is greater than Tp.
[0023] In other words, the entity according to the first aspect may allow IRDL precoding for massive connectiv-ity MIMO networks in the undersampled scenario, i.e. when the number of antennas of the entity is greater than the number of symbols of the M pilot sequences.
[0024] In an implementation form of the first aspect, the M pilot sequences are non-orthogonal pilot sequences with controllable cross-correlation properties.
[0025] The entity may know merely the M pilot sequences of the M active UEs of the Mtot UEs without neces-sarily knowing Mtot -M pilot sequences of a rest of the Mtot UEs.
[0026] In an implementation form of the first aspect, the entity is configured to select the M active UEs of the Mtot UEs.
[0027] In an implementation form of the first aspect, the entity is a base station.
[0028] The entity may be a communication device for wireless communication. The entity may be for example a NodeB in 3G UMTS networks, eNodeB in 4G LTE networks, gNodeB (gNB) in 5G networks etc. The base station may be referred to as “radio base station” .
[0029] In an implementation form of the first aspect, the area is a cell or an overlap area of two or more overlap-ping sub-networks.
[0030] In an implementation form of the first aspect, the entity is configured to receive during an uplink data in-terval Td data symbols of a known finite alphabet via the channel, Td being an integer that is greater than K.The entity may be configured to compute an updated weight matrix by performing an iterative algo-rithm with finite alphabet projections and adaptive regularization using the weight matrix and a K× (Tp+Td) matrix of a signal received by the entity during both the pilot interval and the uplink data interval.
[0031] The iterative algorithm may be referred to as “iterative interference rejection algorithm” .
[0032] In an implementation form of the first aspect, the entity is configured to perform the iterative algorithm by computing, in an initial step of the iterative algorithm, as the updated weight matrix the weight matrix using the computed diagonal loading factor, and computing, in each iteration after the initial step, the up-dated weight matrix using the updated weight matrix computed in a directly previous iteration, the K× (Tp+Td) matrix of the signal received by the entity during both the pilot interval and the uplink data interval, and a second diagonal loading factor.
[0033] In an implementation form of the first aspect, the entity is configured to compute the second diagonal loading factor by solving an equation for the second diagonal loading factor. The equation comprises a monotonic function of a likelihood ratio of a K×K regularized covariance matrix of the received signal estimated during both the pilot interval and the uplink data interval. The monotonic function of the likeli-hood ratio equals an average value of the monotonic function of the likelihood ratio comprising as input a K× (Tp+Td) matrix of independent Gaussian variables with zero mean and unit power.
[0034] In an implementation form of the first aspect, the entity is configured to, in each iteration after the initial step, compute a M× (Tp+Td) matrix of pilot symbols and finite alphabet projected data symbol esti-mates by performing a finite alphabet projection operation using the updated weight matrix computed in the directly previous iteration and a K×Td matrix of the signal received by the entity during the data in-terval. The entity may be configured to, in each iteration after the initial step, compute the updated weight matrix using the computed M× (Tp+Td) matrix of pilot symbols and finite alphabet projected data symbol estimates, the K× (Tp+Td) matrix of the signal received by the entity during both the pilot in-terval and the uplink data interval, and the second diagonal loading factor.
[0035] In an implementation form of the first aspect, the entity is configured to determine that M0 active UEs of the M active UEs do not transmit data during an uplink data interval and M1 active UEs of the M active UEs transmit data during the uplink data interval. M0 and M1 being each an integer greater than or equal to zero and the sum of M0 and M1 equals to M. The entity may be configured to compute a second type weight matrix of the M1 active UEs by computing the weight matrix of the M1 active UEs. The entity may be configured to compute the second type weight matrix of the M0 active UEs by computing the weight matrix of the M0 active UEs, a first K×K covariance matrix of a signal received by the entity during the pilot interval, a second K×K covariance matrix of a signal received by the entity during the uplink data interval, and a K×M0 cross correlation matrix for the M0 UEs.
[0036] In an implementation form of the first aspect, the entity is configured to compute the weight matrix using the computed diagonal loading factor, select a reduced number of the M active UEs for downlink trans-mission, and use a subset of the computed weight matrix corresponding to the selected reduced number of the M active UEs for the downlink transmission.
[0037] In an implementation form of the first aspect, at least one of the M active UEs comprises two or more an-tennas.
[0038] In order to achieve the entity according to the first aspect of the disclosure, some or all of the implementa-tion forms and optional features of the first aspect, as described above, may be combined with each other.
[0039] A second aspect of this disclosure provides a method for computing, by an entity, a diagonal loading fac-tor of a weight matrix that is used for interference rejection down link precoding in an area comprising Mtot user equipments (UEs) , each comprising at least one antenna, wherein the entity comprises K anten-nas. The method comprises wirelessly receiving, during a pilot interval, via a channel from M active UEs of the Mtot UEs M pilot sequences of Tp symbols, and computing the diagonal loading factor by solving an equation for the diagonal loading factor. The equation comprises a monotonic function of a likelihood ratio of a regularized (Tp-M) × (Tp-M) matrix. The monotonic function of the likelihood ratio equals an average value of the monotonic function of the likelihood ratio comprising as input a K× (Tp-M) matrix of independent Gaussian variables with zero mean and unit power. The diagonal loading factor is a real number greater than or equal to zero; K, Mtot, M and Tp are each an integer greater than zero; Mtot is greater than K; K is greater than M, and Tp is greater than M.
[0040] In an implementation form of the second aspect, the method comprises computing the weight matrix us-ing the computed diagonal loading factor.
[0041] In an implementation form of the second aspect, the method comprises computing a K×M downlink precoding matrix using the weight matrix.
[0042] In an implementation form of the second aspect, the method comprises solving the equation for the diago-nal loading factor by computing a matrix that is the interference plus noise covariance matrix estimate during the pilot interval.
[0043] In an implementation form of the second aspect, the method comprises computing the matrix that is the interference plus noise covariance matrix estimate during the pilot interval using a K×Tp matrix of a signal received during the pilot interval, a M×Tp matrix of the M pilot sequences, and an estimated channel matrix of the channel.
[0044] In an implementation form of the second aspect, the method comprises computing the regularized (Tp-M) × (Tp-M) matrix using Tp-M linearly independent columns of the computed matrix; and solving the equation for the diagonal loading factor using the computed regularized (Tp-M) × (Tp-M) matrix.
[0045] In an implementation form of the second aspect, K is greater than Tp.
[0046] In an implementation form of the second aspect, the M pilot sequences are non-orthogonal pilot se-quences with controllable cross-correlation properties.
[0047] In an implementation form of the second aspect, the method comprises selecting the M active UEs of the Mtot UEs.
[0048] In an implementation form of the second aspect, the entity is a base station.
[0049] In an implementation form of the second aspect, the area is a cell or an overlap area of two or more over-lapping sub-networks.
[0050] In an implementation form of the second aspect, the method comprises receiving during an uplink data interval Td data symbols of a known finite alphabet via the channel, Td being an integer that is greater than K. The method may comprise computing an updated weight matrix by performing an iterative algo-rithm with finite alphabet projections and adaptive regularization using the weight matrix and a K× (Tp+Td) matrix of a signal received during both the pilot interval and the uplink data interval.
[0051] In an implementation form of the second aspect, the method comprises performing the iterative algorithm by computing, in an initial step of the iterative algorithm, as the updated weight matrix the weight matrix using the computed diagonal loading factor, and computing, in each iteration after the initial step, the up-dated weight matrix using the updated weight matrix computed in a directly previous iteration, the K× (Tp+Td) matrix of the signal received during both the pilot interval and the uplink data interval, and a second diagonal loading factor.
[0052] In an implementation form of the second aspect, the method comprises computing the second diagonal loading factor by solving an equation for the second diagonal loading factor. The equation comprises a monotonic function of a likelihood ratio of a K×K regularized covariance matrix of the received signal estimated during both the pilot interval and the uplink data interval. The monotonic function of the likeli-hood ratio equals an average value of the monotonic function of the likelihood ratio comprising as input a K× (Tp+Td) matrix of independent Gaussian variables with zero mean and unit power.
[0053] In an implementation form of the second aspect, the method comprises, in each iteration after the initial step, computing a M× (Tp+Td) matrix of pilot symbols and finite alphabet projected data symbol esti-mates by performing a finite alphabet projection operation using the updated weight matrix computed in the directly previous iteration and a K×Td matrix of the signal received during the data interval. The method may comprise, in each iteration after the initial step, computing the updated weight matrix using the computed M× (Tp+Td) matrix of pilot symbols and finite alphabet projected data symbol esti-mates, the K× (Tp+Td) matrix of the signal received during both the pilot interval and the uplink data interval, and the second diagonal loading factor.
[0054] In an implementation form of the second aspect, the method comprises determining that M0 active UEs of the M active UEs do not transmit data during an uplink data interval and M1 active UEs of the M active UEs transmit data during the uplink data interval. M0 and M1 being each an integer greater than or equal to zero and the sum of M0 and M1 equals to M. The method may comprise computing a second type weight matrix of the M1 active UEs by computing the weight matrix of the M1 active UEs. The method may comprise computing the second type weight matrix of the M0 active UEs by computing the weight matrix of the M0 active UEs, a first K×K covariance matrix of a signal received during the pilot inter-val, a second K×K covariance matrix of a signal received during the uplink data interval, and a K×M0 cross correlation matrix for the M0 UEs.
[0055] In an implementation form of the second aspect, the method comprises computing the weight matrix us-ing the computed diagonal loading factor, select a reduced number of the M active UEs for downlink transmission, and use a subset of the computed weight matrix corresponding to the selected reduced num-ber of the M active UEs for the downlink transmission.
[0056] In an implementation form of the second aspect, at least one of the M active UEs comprises two or more antennas.
[0057] One or more of the steps of the method according to the second aspect and the optional steps of the implementation forms of the method according to the second aspect may be performed by the entity.
[0058] The above description of the entity according to the first aspect is correspondingly valid for the method according to the second aspect.
[0059] The entity may be the entity according to the first aspect.
[0060] The method of the second aspect and its implementation forms and optional features achieve the same advantages as the entity of the first aspect and its respective implementation forms and respective optional features.
[0061] In order to achieve the method according to the second aspect of the disclosure, some or all of the implementation forms and optional features of the second aspect, as described above, may be combined with each other.
[0062] It has to be noted that all devices, elements, units and means described in the present application could be implemented in software or hardware elements or any kind of combination thereof. All steps which are performed by the various entities described in the present application as well as the functionalities de-scribed to be performed by the various entities are intended to mean that the respective entity is adapted to or configured to perform the respective steps and functionalities. Even if, in the following description of specific embodiments, a specific functionality or step to be performed by external entities is not re-flected in the description of a specific detailed element of that entity which performs that specific step or functionality, it should be clear for a skilled person that these methods and functionalities can be imple-mented in respective software or hardware elements, or any kind of combination thereof.BRIEF DESCRIPTION OF DRAWINGS
[0063] The above described aspects and implementation forms will be explained in the following description of specific embodiments in relation to the enclosed drawings, in which
[0064] Figure 1 shows an example of an entity according to this disclosure;
[0065] Figure 2 shows an example of an implementation form of functional blocks of the entity of Figure 1;
[0066] Figure 3 shows an example of an implementation form of a functional block of the functional blocks of Figure 2;
[0067] Figure 4 shows an example of implementing a target for the functional block of Figure 3;
[0068] Figure 5 shows an example of a method according to this disclosure;
[0069] Figure 6 shows an example of a network configuration comprising the entity according to Figure 1;
[0070] Figure 7 shows data rate cumulative distributed functions (CDF) for different use cases of the en-tity of Figure 1;
[0071] Figure 8 shows data rate cumulative distributed functions (CDF) for different use cases of the en-tity of Figure 1;
[0072] Figure 9 show an example of a scenario with regard to activated user equipments (activated UEs) in which the entity of Figure 1 may be used;
[0073] Figure 10 show an example of a scenario with regard to activated user equipments (activated UEs) in which the entity of Figure 1 may be used;
[0074] Figure 11 shows bit error rate (BER) for different use cases of the entity of Figure 1;
[0075] Figure 12 shows data rate cumulative distributed functions (CDF) for different use cases of the en-tity of Figure 1;
[0076] Figure 13 shows data rate cumulative distributed functions (CDF) for different use cases of the en-tity of Figure 1;
[0077] Figure 14 shows performances for different use cases of the entity of Figure 1;
[0078] Figure 15 shows an example of a network configuration and data rate cumulative distributed func-tions (CDF) for different use cases of the entity of Figure 1 in the network configuration; and
[0079] Figure 16 shows an example of a network configuration and data rate cumulative distributed func-tions (CDF) for different use cases of the entity of Figure 1 in the network configuration.
[0080] In the Figures, corresponding elements may be labelled by the same reference sign.DETAILED DESCRIPTION OF EMBODIMENTS
[0081] Figure 1 shows an example of an entity according to this disclosure. The entity 1 of Figure 1 is an exam-ple of the entity according to the first aspect of this disclosure. Thus, the description of the entity of the first aspect is correspondingly valid for the entity 1 of Figure 1.
[0082] The entity 1 of Figure 1 is an entity for computing a diagonal loading factor of a weight matrix that is used for interference rejection (IR) down link (DL) precoding in an area comprising Mtot user equip-ments 2 (UEs) . The diagonal loading factor is a real number greater than or equal to zero. Mtot is an in-teger greater than zero. Each UE 2 of the UEs 2 comprises at least one antenna 2a. In Figure 1, one antenna 2a is shown for each UE 2. This is only by way of example and, thus, the UEs 2 may comprise a different number of antennas 2a. The entity 1 comprises K antennas 1a, wherein K is an integer greater than zero. The number K of antennas 1a of the entity 1 shown in Figure 1 is only by way of example and, thus, may be different.
[0083] The entity 1 is configured to wirelessly receive, during a pilot interval, via a channel from M active UEs 22 of the Mtot UEs 2 M pilot sequences 3 of Tp symbols. M and Tp are each an integer greater than zero. The number Mtot of UEs 2 and the number M of active UEs 22 shown in Figure 1 is only by way of exam-ple and, thus, may be different. The entity is configured to compute the diagonal loading factor by solv-ing an equation for the diagonal loading factor wherein the equation comprises a monotonic function of a likelihood ratio of a regularized (Tp-M) × (Tp-M) matrix. The monotonic function of the likelihood ratio equals an average value of the monotonic function of the likelihood ratio comprising as input a K× (Tp-M) matrix of independent Gaussian variables with zero mean and unit power. Mtot is greater than K; K is greater than M, and Tp is greater than M. In other words, Mtot is greater than M and, thus, only a fraction of the UEs 2 may be active. Optionally, K is greater than Tp. The aforementioned optional case may be referred to as undersampled scenario.
[0084] The M pilot sequences may be non-orthogonal pilot sequences with controllable cross-correlation proper-ties. For example, each active UE 22 may be assigned with one pilot sequence of Tp symbols from a set of Tp×Mtot non-orthogonal pilot sequences with controllable cross-correlation properties, e.g. Grass-manian Line Packing. In case of a network with multiple cells, wherein each cell has an entity 1 accord-ing to Figure 1, the set of Tp×Mtot non-orthogonal pilot sequences may be reused in all cells.
[0085] The M active UEs 22 may perform uplink (UL) and / or downlink (DL) communication, such as TDD UL and / or TDD downlink communication, with regard to the entity 1. An UE 22 performing UL communica-tion may be understood as communicating data from the UE to the entity 1, i.e. the UE transmits data to the entity 1. An UE 22 performing DL communication may be understood as communicating data from the entity 1 to the UE 22, i.e. the UE receives data from the entity 1. The other UEs of the Mtot UEs 2 (i.e. the other Mtot-M UEs) remain silent, i.e. do not perform UL and DL communication with regard to the entity 1.
[0086] The active UEs 22 may transmit their UL frames, the entity 1 may recover the UL data, form one or more DL precoders and transmit the DL frame (s) to the active UEs 22. The entity 1 may know the M pilot se-quences 3 of Tp symbols. The entity 1 may select the M active UEs 22 of the Mtot UEs 2.
[0087] Optionally, the entity 1 may use a set of Tp×Mort orthogonal pilot sequences for the M active UEs 22. However, using orthogonal pilots may require a centralized pilot planning which may be difficult in multi-cell or spectrum sharing massive connectivity networks.
[0088] The entity 1 may be a device, such as a device for communication (e.g. device for wireless communica-tion, such as radio communication) . The entity 1 may comprise a processor or processing circuitry (not shown) configured to perform, conduct or initiate the various operations of the entity 1 described herein. The processing circuitry may comprise hardware and / or the processing circuitry may be controlled by software. The hardware may comprise analog circuitry or digital circuitry, or both analog and digital cir-cuitry. The digital circuitry may comprise components such as application-specific integrated circuits (ASICs) , field-programmable arrays (FPGAs) , digital signal processors (DSPs) , or multi-purpose proces-sors. The entity 1 may further comprise memory circuitry, which stores one or more instruction (s) that can be executed by the processor or by the processing circuitry, in particular under control of the soft-ware. For instance, the memory circuitry may comprise a non-transitory storage medium storing executa-ble software code which, when executed by the processor or the processing circuitry, causes the various operations of the entity 1 to be performed. In one embodiment, the processing circuitry comprises one or more processors and a non-transitory memory connected to the one or more processors. The non-transi-tory memory may carry executable program code which, when executed by the one or more processors, causes the entity 1 to perform, conduct or initiate the operations or methods described herein.
[0089] The entity 1 may be a base station. The entity may be for example a NodeB in 3G UMTS networks, eNodeB in 4G LTE networks, gNodeB (gNB) in 5G networks etc.
[0090] The weight matrix that is used for the IRDL precoding may be referred to as IR weight matrix. It may be a K×M matrix. Computing said matrix may comprise estimating said matrix. Computing the weight matrix using the UL pilot data (i.e. the M pilot sequences of Tp symbols) may be expressed as:
[0091] In the above equation (1) , Xp is aK×Tp matrix of a signal received by the entity 1 during the pilot in-terval, Sp is a M×Tp matrix of the M pilot sequences for the entity 1, is the diagonal loading factor, IA is an A×A unit matrix, and (·) *is the transposition complex conjugate operation. The term “regularization factor” may be used as a synonym for the term “diagonal loading factor” .
[0092] The right part of equation (1) leads to a simpler implementation in an undersampled scenario (i.e. when K is greater than Tp, because it involves only a Tp×Tp matrix inversion compared to K×K matrix inver-sion in the central part of equation (1) , which may be for a conventional oversampled scenario.
[0093] A K×M DL precoding matrix which may be used for IRDL precoding may be formed as:
[0094] In the above equation (2) , (·) C is the complex conjugate operation, and tr (A) is a trace of matrix A. Thus, the entity 1 may compute the DL precoding matrix by normalizing the weight matrix
[0095] The diagonal loading factor of the weight matrix may be defined as a root of the following single variable equation (3) :
[0096] In the above equation (3) , is a monotonic function of the likelihood ratio of a regularized (Tp-M) × (Tp-M) matrix wherein may be expressed as:
[0097] The (Tp-M) × (Tp-M) matrix may be expressed as follows:
[0098] In the above equation (5) , the matrix U is the interference plus noise covariance matrix estimate during the pilot interval. In other words, the entity may be configured to compute the regularized (Tp-M) × (Tp-M) matrix using Tp-M linearly independent columns of the matrix U; and solve the equation, e.g. equation (3) , for the diagonal loading factor using the computed regularized (Tp-M) × (Tp-M) matrix
[0099] The matrix U may be expressed as follows:
[0100] In the above equation (6) , Xp is the K×Tp matrix of the signal received by the entity 1 during the pilot interval, Sp is the M×Tp matrix of the M pilot sequences for the entity 1, and is an estimated chan-nel matrix of the channel. Thus, the entity 1 may be configured to compute the matrix U that is the inter-ference plus noise covariance matrix estimate during the pilot interval using the K×Tp matrix Xp of the signal received by the entity 1 during the pilot interval, the M×Tp matrix Sp of the M pilot se-quences, and the estimated channel matrix of the channel. In other words, the entity 1 may compute the matrix U that is the interference plus noise covariance matrix estimate during the pilot interval by cleaning of the received signal with the pilots using the estimated channel matrix
[0101] The estimated channel matrix of the channel may be expressed as follows:
[0102] Thus, the entity 1 may be configured to compute the estimated channel matrix of the channel using the K×Tp matrix Xp of the signal received by the entity 1 during the pilot interval, and the M×Tp matrix Sp of the M pilot sequences.
[0103] In the above equation (3) , Bp (K, Tp-M) is the average value of the same monotonic function as in the above equation (4) of the likelihood ratio of the actual matrix The average value of the monotonic function of the likelihood ratio of the actual matrix may be computed by computing the average value of the monotonic function of the likelihood ratio comprising as input the K× (Tp-M) matrix E of independent Gaussian variables with zero mean and unit power.
[0104] The above mentioned average value Bp (K, Tp-M) may be expressed as follows:
[0105] In above equation (8) , E (·) is the expectation operation and E is the K× (Tp-M) matrix of independ-ent Gaussian variables with zero mean and unit power. Bp (K, Tp-M) may be precalculated as defined in above equation (8) .
[0106] A DL precoding or DL precoder that is based on the weight matrix as defined in the above equation (1) with the diagonal loading factor as defined in the above equation (3) may be referred herein as local interference rejection (local IR) precoding or precoder.
[0107] For further information on the entity 1 of Figure 1 reference is made to the description of the entity ac-cording to the first aspect and to the description of the following Figures.
[0108] Figure 2 shows an example of an implementation form of functional blocks of the entity of Figure 1.
[0109] In Figure 2, on the left side LS the optional features of the local IR precoding or precoder, described above with regard to Figure 1, are shown. The above equations (1) , (5) , (6) and (7) are directly shown in the form of functional blocks. The “normalizer” -block 22a may correspond to the above equation (2) . That is, the “normalizer” -block 22a may implement the operation described by means of equation (2) .
[0110] In the following a local interference rejection finite alphabet (IRFA) algorithm is described, which is an iterative algorithm (i.e. an iterative interference rejection (IR) algorithm) with finite alphabet projections of a known finite alphabet (FA) and adaptive regularization initialized from the local IR weight matrix The iterative algorithm may be referred to as IRFA algorithm. For this, it may be assumed that the UL data interval (may be abbreviated as data interval) , during which the entity 1 may receive data from the active UEs 22, comprises Td data symbols from the known FA, wherein Td>>K.
[0111] The entity 1 may receive during the uplink data interval Td data symbols of the known finite alphabet via the channel, Td being an integer that is greater than K. The entity 1 may compute an updated weight ma-trix by performing the iterative algorithm with the finite alphabet projections and adaptive regulari-zation using the weight matrix and a K× (Tp+Td) matrix X of a signal received by the entity 1 during both the pilot interval and the uplink data interval.
[0112] Computing the updated weight matrix by performing the iterative algorithm may be expressed by the following three equations:
[0113] In the description of the iterative algorithm and, thus, in the equations (9) , (10) and (11) , j=1, ..., J is the iteration number of totally J iterations, and is the K×M updated weight matrix at the j-th iteration, which may be referred to as local IRFA weight matrix. X= [Xp Xd] is the K× (Tp+Td) matrix of the signal received by the entity 1 during both pilot and UL data intervals. Xd is the K×Td matrix of the signal received by the entity 1 during the UL data interval, and Xp is theK×Tp matrix of the signal received by the entity 1 during the pilot interval. FA {·} is the FA projection operation (slicer) which is the closest constellation point for the linear symbol estimate, is the M× (Tp+Td) matrix of the pilot symbols and the FA projected data symbol estimates at the j-th iteration. is a diagonal loading factor of the updated weight matrix It may be referred to as second diagonal loading fac-tor in order to be disguisable from the diagonal loading factor of the local IR weight matrix
[0114] The diagonal loading factor may be defined as a solution of the following single variable equation:
[0115] In the above equation (12) , F (γ, X) is a monotonic function of the likelihood ratio of the K×K regu-larized covariance matrix of the received signal XX* estimated over the whole UL interval, i.e. the (UL) pilot interval and the UL data interval. It may be expressed as follows:
[0116] In the above equation (12) , B (K, Tp+Td) is the average value of the same monotonic function as in equation (13) of the likelihood ratio of the actual covariance matrix E (XX*) . B (K, Tp+Td) may be computed (e.g. precalculated) as defined in the following equation (14) , where E is the K× (Tp+Td) matrix of independent Gaussian variables with zero mean and unit power:
[0117] As may be derive from above equation (11) in an initial step (i.e. j = 0) of the iterative algorithm, the up-dated weight matrix equals the local IR weight matrix
[0118] In the light of the above, the entity 1 may be configured to perform the iterative algorithm by computing, in an initial step (i.e. j = 0) of the iterative algorithm, as the updated weight matrix the weight ma-trix using the diagonal loading factor and computing, in each iteration (i.e. j = 1, …, J) after the ini-tial step (i.e. j = 0) , the updated weight matrix using the updated weight matrix computed in a directly previous iteration, the K× (Tp+Td) matrix X of the signal received by the entity 1 during both the pilot interval and the uplink data interval, and a second diagonal loading factor The entity 1 may be configured to compute the second diagonal loading factor by solving an equation (e.g. above equation (12) ) for the second diagonal loading factor The equation comprises a monotonic function of a likelihood ratio of a K×K regularized covariance matrix of the received signal XX* esti-mated during both the pilot interval and the uplink data interval. The monotonic function of the likelihood ratio equals an average value of the monotonic function of the likelihood ratio comprising as input a K× (Tp+Td) matrix E of independent Gaussian variables with zero mean and unit power.
[0119] The entity 1 may be configured to, in each iteration (i.e. j = 1, …, J) after the initial step (i.e. j = 0) , com-pute the M× (Tp+Td) matrix of pilot symbols and finite alphabet projected data symbol esti-mates by performing a finite alphabet projection operation using the updated weight matrix computed in the directly previous iteration and a K×Td matrix Xd of the signal received by the entity 1 during the data interval; and compute the updated weight matrix using the computed M× (Tp+Td) matrix of pilot symbols and finite alphabet projected data symbol estimates, the K× (Tp+Td) matrix X of the signal received by the entity 1 during both the pilot interval and the uplink data interval, and the second diagonal loading factor
[0120] In Figure 2, on the right side RS the optional features of the local IRFA precoding or precoder, i.e. of the local IRFA algorithm, described above are shown. The above equations (9) and (10) are directly shown in the form of functional blocks. The “normalizer” -block 22b may be implemented accordingly to the above equation (2) . That is, the “normalizer” -block 22b may implement the operation described by means of equation (2) .
[0121] As may be derived from Figure 2, for the local IRFA precoding using the iterative algorithm (shown on the right side RS of Figure 2) the signal received by the entity 1 during the pilot interval and the data in-terval is used. For the local IR precoding (shown on the left side LS of Figure 2) the signal received by the entity 1 during the pilot interval is used.
[0122] As shown on the left side LS of Figure 2, the function of the above equation (5) is indicated in the form of a “selector” -block 23. According thereto, the K× (Tp-M) matrix may be formed from the first Tp-M columns of the K×Tp matrix U.
[0123] Figure 3 shows an example of an implementation form of a functional block of the functional blocks of Figure 2. In particular, Figure 3 shows an example of an implementation of the “root” -block 21a or 21b.
[0124] With regard to Figure 3, the A×A matrix is defined for the “root” -block 21a for computing the local IR precoding for A=Tp-M , β=δ, and for the “root” -block 21b for computing the local IRFA pre-coding for A=K, β=γ. The estimate may be found as the lowest positive root of the single varia-ble equation using any available root calculation algorithm, e.g. as in the “fzero” MATLAB routine. The target values in Fig. 3 may be computed as exemplarily shown in Figure 4.
[0125] Figure 4 shows an example of implementing a target for the functional block of Figure 3. As shown in Figure 4, the target may be computed (e.g. precalculated) for A=Tp-M, D=K for computing the local IR precoding, and A=K, D=Tp+Td for computing the local IRFA precoding. As shown in Figure 4, for each of a high number (e.g. 100000 or more) of trials a noise matrix generator block 24a may generate a matrix of independent Gaussian variables with zero mean and unit power, for example, using the “randn” MATLAB routine. A calculator block 24b may calculate the indicated variable, and an averager block 24c may find a mean value for such variable over the number of trials, for example, using the “mean” MATLAB routine.
[0126] Figure 5 shows an example of a method according to this disclosure. The method of Figure 5 is an exam-ple of the method according to the second aspect. The description of the method of the second aspect is correspondingly valid for the method of Figure 5.
[0127] The method of Figure 5 is a method for computing, by an entity, a diagonal loading factor of a weight matrix that is used for interference rejection down link precoding in an area comprising Mtot user equip-ments (UEs) , each comprising at least one antenna, wherein the entity comprises K antennas. The method comprises, as a step S1, wirelessly receiving, during a pilot interval, via a channel from M active UEs of the Mtot UEs M pilot sequences of Tp symbols. The method comprises, as a step S2 following the step S1, computing the diagonal loading factor by solving an equation for the diagonal loading factor. The equa-tion comprises a monotonic function of a likelihood ratio of a regularized (Tp-M) × (Tp-M) matrix. The monotonic function of the likelihood ratio equals an average value of the monotonic function of the likelihood ratio comprising as input a K× (Tp-M) matrix of independent Gaussian variables with zero mean and unit power. The diagonal loading factor is a real number greater than or equal to zero; K, Mtot, M and Tp are each an integer greater than zero; Mtot is greater than K; K is greater than M, and Tp is greater than M.
[0128] For further details on the method of Figure 2, reference is made to the description of the method accord-ing to the second aspect as well as to the description of the Figures.
[0129] Figure 6 shows an example of a network configuration comprising the entity according to Figure 1. In the network of Figure 6, it is assumed that the entity 1 is a base station. This is only by way of example and, thus, the entity may be a different communication device. The following description is correspondingly valid.
[0130] The network of Figure 6 comprises L cells 4, wherein each cell 4 has one entity 1 in the form of a base station (BS) and Mtot user equipments (UEs) , wherein in Figure 6 merely the M active UEs 22 of the Mtot UEs are shown for each cell 4. L is an integer that is greater than zero. In each cell 4 there may be M ac-tive UEs 22. Thus, a total number Ntot of active UEs 22 in the network may equal to the number of cells L multiplied by the number M of active UEs 22 of each cell 4 (i.e. Ntot=LM) . The Ntot active UEs 22 of the network may synchronously transmit their uplink frames, the base stations 1 may recover the uplink data, form the downlink precoders and synchronously transmit the downlink frames to the active UEs 22. Each base station 1 may know only its own set of M pilot sequences. For the undersampled scenario it may be assumed that Tp<K<Ntot.
[0131] In the network of Figure 6, it is assumed that there are 19 cells 4 (i.e. L = 19) , and each base station 1 has 64 antennas (i.e. K = 64) . Further, it is assumed that there are 10 active UEs 22 per cell 4, each compris-ing a single antenna (i.e. M = 10) . This leads to 190 active UEs 22 in the whole network (i.e. Ntot = 190) . The aforementioned numbers are only by way of example and do not limit the present disclosure. Thus, the following description is correspondingly valid in case one or more of the aforementioned numbers is different. According to the example of Figure 6, it is assumed that each active UE 22 is assigned with one pilot sequence from a set of 1024 non-orthogonal Grassmanian pilot sequences of Tp= [24, 48, 64] sym-bols. The base station 1 of each cell 4 may randomly select a group of the M active UEs 22 from the Mtot UEs 2 of the cell 4 for TDD UL / DL sessions. It is assumed, that the channels are stationary for a whole UL / DL session, i.e. no impairments. Further the following is assumed: Pathloss model, dB: 35.3+37.6 log10 (dist) , Raileigh fast fading, narrowband channel model, e.g., for OFDM subcarrier or group of sub-carriers with negligible channel variations in the frequency domain. The cell radius of each cell 4 is as-sumed to be for example 250 m. In each cell 4 it is assumed that there are random uniformly distributed UE locations, with a 35 m minimum distance between the base station 1 and the respective UE. Further, it is assumed that in the network there is -96dBm noise power, 43dBm base station (BS) power, 30dBm UE power. Furthermore, it is assumed that for communication QPSK signaling is used and 144 data symbols may be communicated in the uplink (UL) and downlink (DL) data slot (i.e. during the UL data interval and DL data interval, respectively) . When a base station 1 performs the local IRFA algorithm it is as-sumed that there is the local IR initialization (in the initial step) and three iterations (i.e. j = 0, …, J; J =3) . The assumed performance metric is data rate per UE in seven central cells 4.
[0132] A technique called multi cell minimum mean square error (MMSE) down link (DL) precoding comprises that all base stations of a communication network need to know pilots and long-term fading (pathloss, shadowing) of all active UEs in all neighboring cells of the network. This requires a high level of base stations coordination and, thus, may be unrealistic, i.e. may be mere theoretical approach. For this tech-nique, the complexity may be high since all the active UEs in the neighboring cells need to be included in the corresponding precoding matrices at each base station. The aforementioned technique may be referred to as Full MMSE and may be used in the following as a benchmark with a global pilot and long-term sta-tistics knowledge. The same technique with the known propagation channels instead of the known pilots may be referred to as Ultimate MMSE. It may be used in the following as an ultimate benchmark with a global channel and long-term statistics knowledge. Using simple linear zero forcing (ZF) precoding based on MMSE single cell uplink (UL) channel estimate allows intracell pilot orthogonalization, but it suffers from inter cell interference. This technique (e.g. precoder) may be referred to as Local ZF and it may be used as a benchmark with only a local pilot and long-term statistics knowledge.
[0133] In the following, we compare the local IR precoding and local IRFA precoding with the local ZF, Full MMSE and Ultimate MMSE benchmarks using the network of Figure 6 and the above outlined assump-tions. Corresponding simulation results are shown in Figures 7 and 8. Figure 7 shows data rate cumula-tive distributed functions (CDF) for different use cases of the entity of Figure 1. Figure 8 shows data rate cumulative distributed functions (CDF) for different use cases of the entity of Figure 1. In particular, the above mentioned simulation results are plotted in Figure 7 in terms of data rate cumulative distributed functions (CDF) , which gives a probability that the given x-value is less than the corresponding y-value. The three graphs of Figure 7 differ from each other with regard to the number Tp of symbols of the M pi-lot sequences. For the top graph, the middle graph and the bottom graph of Figure 7 the number Tp equals 24, 48 and 64 respectively. The y-axis (vertical axis) of each graph of Figure 7 represents the cumulative distributed function (CDF) , i.e. a probability that a given value of the x-axis is less than a corresponding value of the y-axis. The x-axis (horizontal axis) of each graph of Figure 7 represents a data rate in terms of bits per symbol (bits / symbol) .
[0134] In the graphs of Figure 7, the curve 71 corresponds to the local ZF benchmark, the curve 72 corresponds to the Full MMSE benchmark, the curve 73 corresponds to local IR precoding without the diagonal load-ing factor (i.e. δ=0) , the curve 74 corresponds to local IR precoding with the diagonal loading factor δ, the curve 75 corresponds to local IRFA precoding without the second loading factor (i.e. γ=0) , the curve 76 corresponds to local IRFA precoding with the second loading factor γ, and the curve 77 corre-sponds to the ultimate MMSE benchmark.
[0135] As may be derived from Figure 7, local IR precoding may outperform local ZF. The local IR results ap-proach the Full MMSE results for growing number of pilot symbols. The local IRFA precoding outper-forms the pilot-based results including the Full MMSE ones. Adaptively regularized local IR precoding and local IRFA precoding may outperform local IR and local IRFA precoding with zero diagonal loading factor (i.e.. δ=0 or γ=0) . Local IRFA results for the shorter and longer pilot sequences are close to the Ultimate MMSE benchmark.
[0136] Figure 8 shows a comparison of the local IR precoding and local IRFA precoding with the local ZF, Full MMSE and Ultimate MMSE benchmarks for non-orthogonal and orthogonal pilot sequences with central-ized pilot planning of pilot reuse 3. The pilot reuse factor may be defined as the ratio between the number of orthogonal pilot sequences and the number of active user equipments (UEs) in a given cell. Pilot reuse 3 may mean that there are at least three times more orthogonal pilot sequences compared to the number of active UEs in a given cell. For example, in case there are 10 active UEs (M = 10) in a given cell, in case of pilot reuse 3 there may be 30 orthogonal pilot sequences. Pilot reuse 3 allows neighboring cells of a given cell using different pilot sequences compared to the given cell. For this comparison it is assumed that each base station 1 comprises 64 antennas (i.e. K = 64) , there are 10 active UEs per cell 4 (i.e. M =10) , and each pilot sequence comprises 30 pilot symbols (i.e. Tp = 30) . Further, it is assumed that 10 out of 30x30 orthogonal pilot sequences may be selected with pilot reuse 3; and 10 out of 30x1024 Grass-manian pilot sequences may be randomly selected per cell 4. The y-axis (vertical axis) of the graph of Figure 8 represents the cumulative distributed function (CDF) , i.e. a probability that a given value of the x-axis is less than a corresponding value of the y-axis. The x-axis (horizontal axis) of the graph of Figure 8 represents a data rate, such as a cell edge rate, in terms of bits per symbol (bits / symbol) .
[0137] The simulation results are plotted in Figure 8 in terms of data rate cumulative distributed functions (CDF) , which gives a probability that the given x-value is less that the corresponding y-value. In Figure 8, curve 81 corresponds to the local ZF benchmark for non-orthogonal pilot sequences, curve 82 corre-sponds to local IR precoding for non-orthogonal pilot sequences, curve 83 corresponds to the local ZF benchmark for orthogonal pilot sequences, curve 84 corresponds to the Full MMSE benchmark for non-orthogonal pilot sequences, curve 85 corresponds to local IR precoding for orthogonal pilot sequences, curve 86 corresponds to the Full MMSE benchmark for orthogonal pilot sequences, curve 87 corresponds to local IRFA precoding for non-orthogonal pilot sequences, curve 88 corresponds to local IRFA for or-thogonal pilot sequences, and curve 89 corresponds to the Ultimate MMSE benchmark.
[0138] As may be derived from Figure 8, orthogonal pilot sequences with structured pilot reuse that uses a cen-tralized pilot planning may outperform the non-orthogonal pilot sequence case for both the Full MMSE benchmark and the local IR precoding. The local IRFA precoding significantly outperforms the Full MMSE benchmark and approaches the Ultimate MMSE in both non-orthogonal and structured orthogonal pilot sequence reuse schemes. A centralized cell cooperation may be needed for structured orthogonal pi-lot reuse. However no cell cooperation is needed for the local IR precoding with non-orthogonal pilot re-use.
[0139] Figure 9 show an example of a scenario with regard to activated user equipments (activated UEs) in which the entity of Figure 1 may be used.
[0140] A typical uplink (UL) / downlink (DL) traffic demand asymmetry for wireless networks may lead to a modified massive connectivity scenario with partially activated uplink data. For example, some of the ac-tive UEs 22 may not have uplink data to transmit for some UL / DL sessions, which is exemplarily illus-trated in Figure 9 for four active UEs 221, 222, 223, and 224. The number of active UEs 22 is only by way of example and, thus, the description of Figure 9 is correspondingly valid for a different number of active UEs 22. Each row of Figure 9 shows whether a respective active UE 221, 222, 223, or 224 transmits pi-lot (s) (e.g. pilot sequence (s) ; may be referred to as uplink (UL) pilot (s) ) during a pilot interval INTp, transmits data (i.e. uplink data) during an uplink data interval INTd_ul, and receives data (i.e. downlink data) during a downlink data interval INTd_dl. That is, a UL / DL session may comprise a (UL) pilot inter-val INTp for communication of pilot (s) , an uplink data interval INTd_ul for communication of uplink data, and an downlink interval INTd_dl for communication of downlink data. As shown in Figure 9, it is assumed that all active UEs 221 to 224 communicate pilots and downlink data, wherein two active UEs 221 and 224 are assumed to communicate uplink data and two active UEs 222 and 223 are assumed to not communicate uplink data.
[0141] This assumed situation may lead to different signal / interference scenarios on the pilot interval INTp and UL data interval INTd_ul for the active UEs 221 to 224 for the current coherence interval. A current co-herence interval may comprise the pilot interval INTp, UL data interval INTd_ul, and DL data interval INTd_dl. A coherence interval may be a UL / DL session. As outlined above, it is assumed that the active UEs 221 to 224 expecting the DL data may transmit their pilots on the pilot interval INTp (may be referred to as UL pilot interval) , but the UL data may be missing on the UL data interval INTd_ul for some of them, namely the UEs 222 and 223. This means that a direct estimate of some interference components plus noise covariance matrix may be obtained from the data interval INTd_ul for the UEs 222 and 223 with no UL data (i.e. transmitting no UL data) . This allows implementing the following example of a semi-blind feature of the basic IR precoder:
[0142] The adaptively regularized local IR or local IRFA precoding may be used for M1 active UEs (e.g. UEs 221 and 224) with activated UL data (i.e. they transmit the UL data) , which may be expressed as follows:
[0143] That is, the entity 1 may compute a second type weight matrix of the M1 active UEs 221 and 224 of the M active UEs 22 that transmit data during the uplink data interval INTd_ul by computing the local IR weight matrix of the M1 active UEs 221 and 224. The number of active UEs of the M active UEs 22 that do not transmit data during the uplink data interval INTd_ul is M0. M0 and M1 are each an integer greater than or equal to zero and the sum of M0 and M1 equals to M (i.e. M0 + M1 = M) .
[0144] For the M0 active UEs (e.g. UEs 222 and 223) without activated UL data the whole UL interval, i.e. the (UL) pilot interval INTp and the UL data interval INTd_ul, may be used for the covariance matrix estima-tion and the cross-correlation vector (channel) may be estimated over the pilot interval INTp. This may be expressed as follows:
[0145] In above equations (16) and (17) , is the second type weight matrix of the M0 active UEs, are the estimated K×K covariance matrices of the signals received by the entity 1 during the (UL) pilot interval INTp and the UL data interval INTd_ul, 0≤ν≤1 is the pi-lot / data weighting factor, is the estimated K×M0 cross correlation matrix for the M0 active UEs, and μ≥1 is the power balancing coefficient.
[0146] That is, the entity 1 may be configured to compute the second type weight matrix of the M0 active UEs by computing the weight matrix of the M0 active UEs, a first K×K covariance matrix of a signal received by the entity 1 during the pilot interval INTp, a second K×K covariance matrix of a signal received by the entity 1 during the uplink data interval INTd_ul, and a K×M0 cross correla-tion matrix for the M0 active UEs.
[0147] Keeping the overall transmit power unchanged, a power weighting factor between those two groups of the precoding weight vectors may be used to balance protection of all UEs. This may lead to an increased overall efficiency compared to the basic local IR or local IRFA precoding in a partially activated uplink data massive connectivity scenario, i.e. in a scenario where not all active UEs communicate UL data.
[0148] A possibility of performance improvement from the semi-blind IR precoding allows to introduce a base station (BS) / user equipment (UE) cooperation by means of user combination for TDD UL / DL trans-missions (i.e. it is exemplarily assumed that the entity 1 is a base station) :
[0149] Each base station may have a list of UEs waiting for the TDD service. All UEs in the list may desire the DL data, but only a fraction of the UEs may have the UL data. The base station may select UEs with and without UL data for the UL / DL session, for example, all the selected UEs may have DL data to communi-cate, but only half of the selected UEs may have the UL data. The above and following description is cor-respondingly valid in case the entity 1 is not a base station but a different communication entity.
[0150] For different signal / interference scenarios during the (UL) pilot interval (INTp) and UL data interval (INTd_ul) , as exemplarily shown in Figure 10, optimization of the balancing parameters μ and ν in the above equations (16) and (17) may be desired. Figure 10 shows an example of a scenario with regard to activated user equipments (activated UEs) in which the entity 1 of Figure 1 may be used. The description of Figure 9 is correspondingly valid for Figure 10.
[0151] The aforementioned optimization may be based on field trials or simulations in typical scenarios. Such precoders may be referred to as local semi-blind interference rejection (local SBIR) precoder and local SBIR with finite alphabet (FA) projections (local SBIRFA) precoder.
[0152] This additional optimization may be avoided by introducing, for the M0 active UEs 222 and 223 that do not transmit UL data, the same scenario at both the (UL) pilot interval INTp and the UL data interval INTd_ul by means of transmission of the known pilot sequences for the M0 active UEs 222 and 223, as shown in Figure 10. That is, the M0 active UEs 222 and 223 that do not transmit UL data may transmit pi-lot sequences to the entity 1 (e.g. base station) during the pilot interval INTp and transmit pilot sequences to the entity 1 (e.g. base station) during the UL data interval INTd_ul. The aformentioned pilot sequences transmitted during the pilot interval INTp may be referred to as “initial pilots” and the aformentioned pi-lot sequences transmitted during the UL data interval INTd_ul may be referred to as “additional pilots” .
[0153] Then, local IR and local IRFA precoders may be directly implemented for different number of pilot se-quences for the M0 active UEs 222 and 223 transmitting no UL data during the UL data interval INTd_ul and for the M1 active UEs 221 and 224 transmitting UL data during the UL data interval INTd_ul. These precoders may be referred to as local IRP+ precoder and local IRFAP+ precoder, respectively. The pas-sage “local IRP+” means local interference rejection (IR) with additional pilots for the M0 active UEs without UL data. The passage “local IRFAP+” means local interference rejection (IR) with finite alphabet (FA) projections and with additional pilots for the M0 active UEs without UL data.
[0154] Figure 11 shows bit error rate (BER) for different use cases of the entity of Figure 1. The bit error rate (BER) simulation results for local SBIR in two scenarios, using the network of Figure 6, are shown in Figure 11 for different values of the balancing parameters μ and ν. Thus, the description of Figure 6 is correspondingly valid. The left side of Figure 11 shows the BER assuming that there are 19 cells (L = 19) , 64 antennas per entity 1 (K = 64) , a pilot sequence has 24 symbols (Tp = 24) , and out of 10 ac-tive UEs of each cell (M = 10) , there are 5 active UEs (of the 10 active UEs) that do not transmit UL data (M0 = 5) . The right side of Figure 11 shows the BER assuming that there are 19 cells (L = 19) , 96 anten-nas per entity 1 (K = 96) , a pilot sequence has 64 symbols (Tp = 64) , and out of 20 active UEs of each cell (M = 20) , there are 10 active UEs (of the 20 active UEs) that do not transmit UL data (M0 = 10) .
[0155] As may be derived from Figure 11, optimal values of the balancing parameters may exist in both scenar-ios.
[0156] Figure 12 shows data rate cumulative distributed functions (CDF) for different use cases of the entity of Figure 1. In particular, the simulation results for different algorithms are presented in Figure 12 in the same scenarios as in the graph on the left side of Figure 11. Thus, for the simulation results of Figure 12, it is assumed that there are 64 antennas per entity 1 (K = 64) , a pilot sequence has 24 symbols (Tp = 24) , there are 144 data symbols of a known finite alphabet (FA) during the UL data interval, and out of 10 ac-tive UEs (M = 10) , there are 5 active UEs (of the 10 active UEs) that do not transmit UL data (M0 = 5) . The optimal values of the balancing parameters for local SBIR and local SBIRFA may correspond to the ones shown in the left graph of Figure 11. It is assumed that for local IRP+ and local IRFAP+ randomly generated QPSK symbols are used as the additional pilots for the M0 active UEs that do not transmit UL data. In Figure 12, arrows indicate a performance gain compared to basic local IR and local IRFA precod-ing. The y-axis (vertical axis) of the graph of Figure 12 represents the cumulative distributed function (CDF) , i.e. a probability that a given value of the x-axis is less than a corresponding value of the y-axis. The x-axis (horizontal axis) of the graph of Figure 12 represents a data rate in terms of bits per symbol (bits / symbol) .
[0157] In Figure 12, the curve 121 corresponds to the local IR precoding, the curve 122 corresponds to the Full MMSE benchmark, the curve 123 corresponds to the local SBIR precoding, the curve 124 corresponds to the local IRFA precoding, the curve 125 corresponds to the local IRP+ precoding, the curve 126 corre-sponds to the local SBIRFA precoding, the curve 127 corresponds to the local IRFAP+ precoding, and the curve 128 corresponds to the Ultimate MMSE benchmark.
[0158] From Figure 12 it may be derived that local SBIR and local SBIRFA precoding may outperform local IR and local IRFA precoding for properly selected balancing parameters. Local IRP+ and local IRFAP+ pre-coding may outperform local SBIR and local SBIRFA precoding without any additional optimization. A transmission of additional pilots (as exemplarily shown in Figure 10) may compromise energy efficiency of such network with partially activated UL data.
[0159] Local IR and local IRFA precoding performance may be improved by means of base station (BS) / user equipment (UE) cooperation with scheduling of a reduced number of UEs initially activated during the pilot interval (the following is also valid assuming that the entity is not a base station but a different com-munication entity) :
[0160] For example, in a wireless network with L cells (such as the network of Figure 6) , all cells may activate M UEs per cell for UL pilot transmission (i.e. transmitting a pilot sequence during the pilot interval INTp) . All cells may compute (e.g. estimate) local IR weights, compute (e.g. estimate) IR efficiency for local UEs (i.e. the active cells of the respective cell) , and select Mr UEs, wherein Mr≤M, for the UL / DL session for the given coherence interval using the MMSE criterion estimated over the (UL) pilot interval INTp. All base stations of the network may use a subset of the local IR weights or the local IRFA pre-coder estimated over the reduced UL data interval for DL transmissions for the reduced set of UEs. If ef-fective IR is possible for the initial number of M active UEs in the given scenario, number of base station antennas (antennas per base station) , and number Tp of pilot symbols, then such UE selection may lead to an increased precoding efficiency compared to transmission of Mr active UEs without such scheduling.
[0161] The scheduling selection metric (MSE) may be defined as:
[0162] In the equation (18) , sm is the m-th row of the initially activated M×Tp pilot matrix S, and is the m-th column of the K×M local IR weight matrix The set of Mr UEs scheduled for UL / DL transmissions may be defined as
[0163] for Mr≤M. Then, the precoding weights may be computed using theΨ-th subset of or by using the local IRFA weights estimated over the reduced UL data interval.
[0164] Thus, the entity 1 may be configured to compute the weight matrix using the computed diagonal load-ing factor select a reduced number Mr of the M active UEs for downlink transmission, and use a subset of the computed weight matrix corresponding to the selected reduced number Mr of the M active UEs for the downlink transmission.
[0165] Figure 13 shows data rate cumulative distributed functions (CDF) for different use cases of the entity of Figure 1. In particular, a comparison of the local IR and local IRFA precoders with the base station / user equipment scheduling (BS / UE scheduling) that is described above is shown in Figure 13 for a relatively simple scenario (see the left graph of Figure 13) and a more difficult scenario (see the right graph of Fig-ure 13) . For the left graph of Figure 13 it is assumed that there are 19 cells (L = 19) , 32 antennas per base station (K = 32) , and the pilot sequence comprises 24 symbols (Tp = 24) . For the right graph of Figure 13 it is assumed that there are 19 cells (L = 19) , 64 antennas per base station (K = 64) , and the pilot sequence comprises 24 symbols (Tp = 24) . The y-axis (vertical axis) of each graph of Figure 13 represents the cu-mulative distributed function (CDF) , i.e. a probability that a given value of the x-axis is less than a corre-sponding value of the y-axis. The x-axis (horizontal axis) of each graph of Figure 13 represents a data rate in terms of bits per symbol (bits / symbol) .
[0166] In the left graph of Figure 13, the curve 131 corresponds to local IR precoding with 5 active UEs without the BS / UE scheduling (i.e. M=5) , the curve 132 corresponds to local IR precoding with 5 active UEs being selected from 7 active UEs (i.e. M=7; Mr=5) , the curve 133 corresponds to local IR precoding with 5 active UEs being selected from 10 active UEs (i.e. M=10; Mr=5) . In the left graph of Figure 13, the curve 134 corresponds to local IRFA precoding with 5 active UEs without the BS / UE scheduling (i.e. M=5) , the curve 135 corresponds to local IRFA precoding with 5 active UEs being selected from 7 active UEs (i.e. M=7; Mr=5) , and the curve 136 corresponds to local IRFA precoding with 5 active UEs being selected from 10 active UEs (i.e. M=10; Mr=5) . In the right graph of Figure 13, the curve 131 corresponds to local IR precoding with 10 active UEs without the BS / UE scheduling (i.e. M=10) , the curve 132 corresponds to local IR precoding with 10 active UEs being selected from 12 active UEs (i.e. M=12; Mr=10) , the curve 133 corresponds to local IR precoding with 10 active UEs being se-lected from 15 active UEs (i.e. M=15; Mr=10) . In the right graph of Figure 13, the curve 134 corre-sponds to local IRFA precoding with 10 active UEs without the BS / UE scheduling (i.e. M=10) , the curve 135 corresponds to local IRFA precoding with 10 active UEs being selected from 12 active UEs (i.e. M=12; Mr=10) , and the curve 136 corresponds to local IRFA precoding with 10 active UEs be-ing selected from 15 active UEs (i.e. M=15; Mr=10) .
[0167] In the left graph of Figure 13, one can see a significant gain for both local IR and local IRFA precoding with BS / UE scheduling compared to the basic solutions for Mr=M, i.e. compared to local IR and local IRFA without the BS / UE scheduling. Although, in the right graph of Figure 13 only limited gain for local IRFA precoding may be observed, it illustrates an applicability area for such scheduling.
[0168] Optionally, the UEs may comprise more than one antenna. That is, KUE>1 UE antennas may be as-sumed, wherein KUE is the number of antennas of a UE. Further it is assumed, that the UL transmission is kept unchanged, e.g. from the first antenna using local IR or local IRFA precoding. The DL m-th received signal in a given cell may be estimated as follows:
[0169] In the above equations (20) , (21) and (22) , Ym is the KUE×Tdl matrix of the received signal, Tdl is the number of symbols in the DL data interval, is the KUE×1 weight vector, and is the FA projection of the first row of matrix Ym.
[0170] Figure 14 shows performances for different use cases of the entity of Figure 1. In particular, the bit error rate (BER) of simulation results, using the network of Figure 6 and four antennas per UE (i.e. KUE = 4) , are plotted in the left graph of Figure 14 for a conventional pilot-based interference rejection combining (IRC) receiver for 12 DL pilots per data interval and in the right graph of Figure 14 for a interference re-jection UE as described in the above equations (20) , (21) and (22) .
[0171] In the left graph of Figure 14, the bar 141 corresponds to an UE with a single antenna, and the bar 142 corresponds to a pilot-based IRC UE. For the left graph of Figure 14, it is assumed that the network com-prises 19 cells (L = 19) , the base station (i.e. entity 1) comprises 64 antennas (K = 64) , there are 10 active UEs per cell (M = 10) , the number Tp of symbols of a pilot sequence equals 24 (Tp = 24) , and the number Td of data symbols equaling the number Tdl of downlink data symbols equals 144 (Td = Tdl = 144) . Fur-ther, for the left graph of Figure 14, it is assumed that the UE comprises 4 antennas (KUE = 4) , and the number Tpdl of downlink (DL) pilots per data interval equals 12 (Tdl = 12) .
[0172] In the right graph of Figure 14, the bar 141 corresponds to an UE with a single antenna, and the bar 143 corresponds to an UE according to the equations (20) , (21) and (22) . For the right graph of Figure 14, it is assumed that the network comprises 19 cells (L = 19) , the base station (i.e. entity 1) comprises 64 anten-nas (K = 64) , there are 10 active UEs per cell (M = 10) , the number Tp of symbols of a pilot sequence equals 24 (Tp = 24) , and the number Td of data symbols equaling the number Tdl of downlink data sym-bols equals 144 (Td = Tdl = 144) . Further, for the right graph of Figure 14, it is assumed that the UE com-prises 4 antennas (KUE = 4) , and there are no DL pilots.
[0173] From the left graph of Figure 14, a significant performance degradation for the conventional pilot-based IRC at UE may be derived. A performance improvement for the proposed implementation according to equations (20) , (21) and (22) may be derived from the right side of Figure 14 for the local IR precoding.
[0174] Considering that no data exchange and centralized planning are needed for the local IR precoding algo-rithm and its optional versions described above, an application for such precoding algorithms may be a rule regulated spectrum sharing: For this, it is assumed that sub-networks synchronously transmit their UL / DL signals using the same transmission protocol, and there is no data exchange and centralized plan-ning between sub-networks. Further, it is assumed that sub-networks use the local IR precoding algorithm or its optional versions described above. Optionally, it may be assumed that some sub-networks may use other local algorithms, e.g. local ZF. This may lead to some performance degradation because of partial cooperation.
[0175] For the above assumption, Figure 15 shows an example of a network configuration and data rate cumula-tive distributed functions (CDF) for different use cases of the entity of Figure 1 in the network configura-tion. In particular, an exemplary network configuration for five sub-networks, each with 10 active UEs, is illustrated in Figure 15. On the left side of Figure 15, the small black dot 22 represents an active UE, the small empty dot 1 represents a base station (BS) in the form of an entity 1 according to the present disclo-sure and the big circle 5 represents a sub-network.
[0176] The right graph of Figure 15 shows a data rate CDF for a spectrum sharing between five sub-networks 5 assuming that the base station 1 comprises 64 antennas (K = 64) , there are 24 symbols of a pilot sequence (Tp = 24) , there are 144 data symbols during the UL data interval INTd_ul (Td = 144) , and all 10 active UEs 33 of each sub-network 5 transmit UL data during the UL data interval INTd_ul. The y-axis (vertical axis) of the right graph of Figure 15 represents the cumulative distributed function (CDF) , i.e. a probabil-ity that a given value of the x-axis is less than a corresponding value of the y-axis. The x-axis (horizontal axis) of the right graph of Figure 15 represents a data rate in terms of bits per symbol (bits / symbol) . In the right graph of Figure 15, the curve 151 corresponds to the case in which all sub-networks 5 use the local ZF precoding algorithm, the curve 152 corresponds to the case in which all sub-networks 5 use the local IR precoding algorithm (according to this disclosure) , and the curve 153 corresponds to the case in which all sub-networks 5 use the local IRFA precoding algorithm (according to this disclosure) .
[0177] From the right graph of Figure 15, one may derive a significant performance gain for the local IR and lo-cal IRFA precoders compared to the Local ZF benchmark. The advantage of using the IR based precoders according to the present disclosure by all the spectrum sharing sub-networks is illustrated in the left graph of Figure 16, where only 2 out of the five sub-networks 5 use such algorithms as illustrated in the right graph of Figure 16.
[0178] Figure 16 shows an example of a network configuration and data rate cumulative distributed functions (CDF) for different use cases of the entity of Figure 1 in the network configuration. In particular, an ex-emplary network configuration for five sub-networks 5, each with 10 active UEs 22, is illustrated in Fig-ure 16. On the right side of Figure 16, the small black dot 22 represents an active UE, the small empty dot 1 represents a base station in the form of an entity 1 according to the present disclosure, and the big circle 5 represents a sub-network. As indicated in the right graph of Figure 16, two sub-networks 5 of the five sub-networks 5 use a local IR or local IRFA precoding algorithms according to this disclosure, whereas the other three sub-networks 5 of the five sub-networks 5 use a local ZF algorithm.
[0179] The left graph of Figure 16 shows a data rate CDF for a spectrum sharing between five sub-networks 5 assuming that the base station 1 comprises 64 antennas (K = 64) , there are 24 symbols of a pilot sequence (Tp = 24) , there are 144 data symbols during the UL data interval INTd_ul (Td = 144) , and all 10 active UEs 22 transmit UL data during the UL data interval INTd_ul. The y-axis (vertical axis) of the left graph of Figure 16 represents the cumulative distributed function (CDF) , i.e. a probability that a given value of the x-axis is less than a corresponding value of the y-axis. The x-axis (horizontal axis) of the left graph of Figure 16 represents a data rate in terms of bits per symbol (bits / symbol) . In the left graph of Figure 16, the curve 161 corresponds to the case in which all sub-networks 5 use the local ZF precoding algorithm, the curve 162 corresponds to the case in which only two of the five sub-networks 5 use the local IR pre-coding algorithm (according to this disclosure) , the curve 163 corresponds to the case in which only two of the five sub-networks 5 use the local IRFA precoding algorithm (according to this disclosure) , the curve 164 corresponds to the case in which all sub-networks 5 use the local IR precoding algorithm (ac-cording to this disclosure) , and the curve 165 corresponds to the case in which all sub-networks 5 use the local IRFA precoding algorithm (according to this disclosure) .
[0180] As may be derived from the left graph of Figure 16, there is a performance gain for the local IR and local IRFA precoding in case all sub-network 5 use the local IR or local IRFA precoding (cf. curves 164 and 165, respectively) compare to the case, in which only two of the five sub-networks 5 use the local IR or local IRFA precoding (cf. curves 161 and 162, respectively) . In other words, from the left graph of Figure 16 it may be derived that a partial rule regulated cooperation leads to significant performance degradation compared to a full rule regulated cooperation.
[0181] In the light of the above, some examples and implementation forms are described assuming that the entity 1 of the present disclosure is a base station. This is only by way of example and, thus, the above descrip-tion is correspondingly valid in case the entity 1 is another communication entity.
[0182] The present disclosure has been described in conjunction with various embodiments as examples as well as implementations. However, other variations can be understood and effected by those persons skilled in the art and practicing the claimed matter, from the studies of the drawings, this disclosure and the inde-pendent claims. In the claims as well as in the description the word “comprising” does not exclude other elements or steps and the indefinite article “a” or “an” does not exclude a plurality. A single element or other unit may fulfill the functions of several entities or items recited in the claims. The mere fact that certain measures are recited in the mutual different dependent claims does not indicate that a combination of these measures cannot be used in an advantageous implementation.
[0183] ABSTRACT
[0184] The present disclosure relates to an entity for computing a diagonal loading factor of a weight matrix that is used for interference rejection down link precoding in an area comprising Mtot user equipments (UEs) , each comprising at least one antenna. The entity comprises K antennas. The entity is configured to wire-lessly receive, during a pilot interval, via a channel from M active UEs of the Mtot UEs M pilot sequences of Tp symbols. The entity is configured to compute the diagonal loading factor by solving an equation for the diagonal loading factor. The equation comprises a monotonic function of a likelihood ratio of a regu-larized (Tp-M) × (Tp-M) matrix, and the monotonic function of the likelihood ratio equals an average value of the monotonic function of the likelihood ratio comprising as input a K× (Tp-M) matrix of in-dependent Gaussian variables with zero mean and unit power.
[0185] (Figure 1)
Claims
1.An entity (1) for computing a diagonal loading factor of a weight matrix that is used for interference rejection down link precoding in an area comprising Mtot user equipments (2) , UEs, each comprising at least one antenna (2a) , wherein- the entity (1) comprises K antennas (1a) ,- the entity (1) is configured to wirelessly receive, during a pilot interval, via a channel from M active UEs (22) of the Mtot UEs (2) M pilot sequences (3) of Tp symbols, and- the entity (1) is configured to compute the diagonal loading factor by solving an equation for the diagonal loading factor, wherein- the equation comprises a monotonic function of a likelihood ratio of a regularized (Tp-M) × (Tp-M) matrix, and- the monotonic function of the likelihood ratio equals an average value of the mon-otonic function of the likelihood ratio comprising as input a K× (Tp-M) matrix of independent Gaussian variables with zero mean and unit power; wherein- the diagonal loading factor is a real number greater than or equal to zero; K, Mtot, M and Tp are each an integer greater than zero; Mtot is greater than K; K is greater than M, and Tp is greater than M.2.The entity (1) according to claim 1, wherein the entity (1) is configured to compute the weight matrix using the computed diagonal loading factor.3.The entity (1) according to claim 1 or 2, wherein the entity (1) is configured to compute a K×M downlink precoding matrix using the weight matrix.4.The entity (1) according to any one of the previous claims, wherein-the entity (1) is configured to solve the equation for the diagonal loading factor by com-puting a matrix that is the interference plus noise covariance matrix estimate during the pilot interval.5.The entity (1) according to claim 4, wherein- the entity (1) is configured to compute the matrix that is the interference plus noise covar-iance matrix estimate during the pilot interval using a K×Tp matrix of a signal received by the entity during the pilot interval, a M×Tp matrix of the M pilot sequences, and an estimated channel matrix of the channel.6.The entity (1) according to claim 4 or 5, wherein the entity (1) is configured to- compute the regularized (Tp-M) × (Tp-M) matrix using Tp-M linearly independent columns of the computed matrix; and- solve the equation for the diagonal loading factor using the computed regularized (Tp-M) × (Tp-M) matrix.7.The entity (1) according to any one of the previous claims, wherein- K is greater than Tp.8.The entity (1) according to any one of the previous claims, wherein- the M pilot sequences are non-orthogonal pilot sequences with controllable cross-correla-tion properties.9.The entity (1) according to any one of the previous claims, wherein- the entity (1) is configured to select the M active UEs (22) of the Mtot UEs (2) .10.The entity (1) according to any one of the previous claims, wherein- the entity (1) is a base station.11.The entity (1) according to any one of the previous claims, wherein the area is a cell (4) or an overlap area of two or more overlapping sub-networks (5) .12.The entity (1) according to any one of the previous claims, wherein the entity is config-ured to- receive during an uplink data interval Td data symbols of a known finite alphabet via the channel, Td being an integer that is greater than K, and- compute an updated weight matrix by performing an iterative algorithm with finite alpha-bet projections and adaptive regularization using the weight matrix and a K× (Tp+Td) matrix of a signal received by the entity during both the pilot interval and the uplink data interval.13.The entity (1) according claim 12, wherein the entity (1) is configured to perform the iter-ative algorithm by- computing, in an initial step of the iterative algorithm, as the updated weight matrix the weight matrix using the computed diagonal loading factor, and- computing, in each iteration after the initial step, the updated weight matrix using the up-dated weight matrix computed in a directly previous iteration, the K× (Tp+Td) matrix of the signal received by the entity during both the pilot interval and the uplink data interval, and a second diagonal loading factor.14.The entity (1) according to claim 13, wherein the entity (1) is configured to compute the second diagonal loading factor by solving an equation for the second diagonal loading factor, wherein- the equation comprises a monotonic function of a likelihood ratio of a K×K regularized covariance matrix of the received signal estimated during both the pilot interval and the uplink data interval, and- the monotonic function of the likelihood ratio equals an average value of the monotonic function of the likelihood ratio comprising as input a K× (Tp+Td) matrix of independent Gaussian variables with zero mean and unit power.15.The entity (1) according to claim 13 or 14, wherein the entity (1) is configured to, in each iteration after the initial step,- compute a M× (Tp+Td) matrix of pilot symbols and finite alphabet projected data sym-bol estimates by performing a finite alphabet projection operation using the updated weight matrix computed in the directly previous iteration and a K×Td matrix of the sig-nal received by the entity during the data interval, and- compute the updated weight matrix using the computed M× (Tp+Td) matrix of pilot symbols and finite alphabet projected data symbol estimates, the K× (Tp+Td) matrix of the signal received by the entity during both the pilot interval and the uplink data interval, and the second diagonal loading factor.16.The entity (1) according to any one of the previous claims, wherein the entity (1) is con-figured to- determine that M0 active UEs of the M active UEs (22) do not transmit data during an up-link data interval and M1 active UEs of the M active UEs (22) transmit data during the uplink data interval, M0 and M1 being each an integer greater than or equal to zero and the sum of M0 and M1 equals to M,- compute a second type weight matrix of the M1 active UEs by computing the weight ma-trix of the M1 active UEs, and- compute the second type weight matrix of the M0 active UEs by computing the weight matrix of the M0 active UEs, a first K×K covariance matrix of a signal received by the entity during the pilot interval, a second K×K covariance matrix of a signal received by the entity during the uplink data interval, and a K×M0 cross correlation matrix for the M0 UEs.17.The entity (1) according to any one of the previous claims, wherein the entity (1) is con-figured to- compute the weight matrix using the computed diagonal loading factor,- select a reduced number of the M active UEs (22) for downlink transmission, and- use a subset of the computed weight matrix corresponding to the selected reduced number of the M active UEs (22) for the downlink transmission.18.The entity (1) according to any one of the previous claims, wherein at least one of the M active UEs (22) comprises two or more antennas (2a) .19.A method for computing, by an entity, a diagonal loading factor of a weight matrix that is used for interference rejection down link precoding in an area comprising Mtot user equip-ments, UEs, each comprising at least one antenna, wherein the entity comprises K anten-nas, the method comprising:- wirelessly receiving (S1) , during a pilot interval, via a channel from M active UEs of the Mtot UEs M pilot sequences of Tp symbols, and- computing (S2) the diagonal loading factor by solving an equation for the diagonal load-ing factor, wherein- the equation comprises a monotonic function of a likelihood ratio of a regularized (Tp-M) × (Tp-M) matrix, and- the monotonic function of the likelihood ratio equals an average value of the mon-otonic function of the likelihood ratio comprising as input a K× (Tp-M) matrix of independent Gaussian variables with zero mean and unit power; wherein- the diagonal loading factor is a real number greater than or equal to zero; K, Mtot, M and Tp are each an integer greater than zero; Mtot is greater than K; K is greater than M, and Tp is greater than M.
Citation Information
Patent Citations
Method for determining a precoding matrix and precoding module
CN107615694A
Deep learning precoding method using position information
CN111049559A
Apparatus and method for transmitting data using precoding scheme in multiple antenna system
KR1020100097412A
High-performance diagonal loading for MMSE equalization
US8913700B1
