Enhanced channel measurement reporting for network node side channel quality prediction

Abstract

A method performed by a wireless communication device (WCD) is provided. The method comprises receiving wireless signals from a network node, based on the received wireless signals, estimating a channel between the WCD and the network node, and generating first singular vector, SV, group information related to a first group of one or more SVs of the estimated channel. The method further comprises generating second SV group information related to a second group of one or more SVs of the estimated channel, compressing the first SV group information, thereby generating compressed first SV group information, compressing the second SV group information, thereby generating compressed second SV group information, and transmitting the compressed first and second SV group information to the network node.

Description

ENHANCED CHANNEL MEASUREMENT REPORTING FOR NETWORK NODE SIDE CHANNEL QUALITY PREDICTION TECHNICAL FIELD

[0001] This disclosure relates to enhanced channel measurement reporting for network node side channel quality prediction. BACKGROUND

[0002] Codebook-Based Precoding

[0003] Multi-antenna techniques can significantly increase the data rates (e.g., a data transmission rate) and reliability of a wireless communication system. The performance of the wireless communication system is particularly improved if both the transmitter and the receiver of the wireless communication system are equipped with multiple antennas, which allows providing a multiple-input multiple-output (MIMO) communication channel. Such system and / or related techniques are commonly referred to as MIMO.

[0004] A core component of the fifth Generation (5G) wireless network or New Radio (NR) is the support of MIMO antenna deployments and MIMO related techniques such as spatial multiplexing. Spatial multiplexing can be used to increase data rates in favorable channel conditions. FIG.1 shows an example of spatial multiplexing. In the spatial multiplexing shown in FIG.1, an information carrying symbol vector s is multiplied by an NTx r precoding matrix or precoder ^^, which serves to distribute the transmit energy in a subspace of the NTdimensional vector space. The precoding matrix is typically selected from a codebook of possible precoding matrices, and typically indicated by means of a precoding matrix indicator (PMI), which specifies a unique precoding matrix in the codebook for a given number of symbol streams. Each of the r symbols in ^^^^corresponds to a MIMO layer and r is referred to as the transmission rank, which equals to the number of columns of the precoder ^^. Here, spatial multiplexing is achieved since multiple symbols can be transmitted simultaneously over the same time / frequency resource element (RE). The number of symbols r is typically adapted to suit the current channel properties.

[0005] NR uses Orthogonal Division Multiplexing (OFDM) in downlink. The received NR x 1 vector yn at a user equipment (UE) on a certain RE can be expressed as ^^^^= ^^^^^^ ^^^^+ ^^^^where en is a receiver noise / interference vector. The precoder ^^ can be constant over frequency (i.e., wideband), or frequency selective (i.e., per subband). The precoder ^^ is chosen to match the characteristics of the NRxNT MIMO channel matrix ^^^^, resulting in so-called channel dependent precoding. This is also commonly referred to asclosed-loop precoding.

[0006] In the closed-loop precoding, the UE feeds back recommendations on a suitable precoder to the base station (e.g., gNB) in the form of a precoding matrix indicator (PMI) based on downlink channel measurements. For that purpose, the UE is configured with a channel state information (CSI) report configuration including CSI reference signals (CSI-RS) for channel measurements and a codebook of candidate precoders. In addition to the precoders, the feedback may also include a rank indicator (RI) and one or two channel quality indicators (CQIs). RI, PMI, and CQI are part of a CSI feedback. In NR, CSI feedback can be either wideband, where one CSI is reported for the entire channel bandwidth, or frequency-selective, where one CSI is reported for each sub-band, which is defined as a number of contiguous physical resource blocks (PRBs) ranging between 4 -32 PRBs depending on the band width part (BWP) size.

[0007] Based on the CSI feedback from the UE, the gNB determines the transmission parameters it wishes to use to transmit to the UE, including the precoding matrix, transmission rank, and modulation and coding scheme (MCS).

[0008] Two-Dimensional (2D) Antenna Arrays

[0009] 2D antenna arrays are widely used and such antenna arrays can be described by a number of antenna ports, ^^1, in a first dimension (e.g., the horizontal dimension), a number of antenna ports, ^^2, in the second dimension perpendicular to the first dimension (e.g., the vertical dimension), and a number of polarizations ^^^^. The total number of the antenna ports is thus ^^ = ^^1^^2^^^^. The concept of an antenna port is non-limiting in the sense that it can refer to any virtualization (e.g., linear mapping) to the physical antenna elements. For example, pairs of physical antenna elements could be fed the same signal, and hence share the same virtualized antenna port. An example of a 4x4 (i.e., ^^1× ^^2,) array with dual-polarized antenna elements (i.e., ^^^^= 2) is shown in FIG.2. More specifically, FIG.2 illustrates a 2D antenna array of dual-polarized antenna elements ( ^^^^= 2), with ^^1= 4 horizontal antenna elements and ^^2= 4 vertical antenna elements.

[0010] Precoding may be interpreted as multiplying the signal to be transmitted by a set of beamforming weights on the antenna ports prior to transmission of the signal. A typical approach is to tailor the precoder to the antenna form factor, i.e., taking into account ^^1, ^^2and ^^^^when designing the precoder codebook.

[0011] Channel State Information Reference Signals (CSI-RS)

[0012] For CSI measurement and feedback, CSI-RS are defined. A CSI-RS is transmitted on an antenna port at an gNB and is used by a UE to measure downlink channel between the antenna port at the gNB and each of the UE’s receive antenna ports. The transmit antenna ports are also referred to as CSI-RS ports. The supported number of CSI-RS ports in NR are {1,2,4,8,12,16,24,32}. By measuring the received CSI-RS, a UE can estimate thechannel that the CSI-RS is traversing, including the radio propagation channel and antenna gains. The CSI-RS for the above purpose is also referred to as Non-Zero Power (NZP) CSI-RS.

[0013] CSI-RS can be configured to be transmitted in certain REs in certain slots. FIG.3 shows an example of CSI-RS REs for 12 antenna ports, where 1RE per resource block (RB) per port is shown.

[0014] Interference measurement resource (IMR) is also defined in NR for a UE to measure interference. An IMR resource contains 4 REs, either 4 adjacent REs in frequency in the same OFDM symbol or 2 by 2 adjacent REs in both time and frequency in a slot. By measuring both the channel based on non-zero power (NZP) CSI-RS and the interference based on an IMR, a UE can estimate the effective channel and noise plus interference to determine the CSI. Furthermore, a UE in NR may be configured to measure interference based on one or multiple NZP CSI-RS resource.

[0015] CSI framework in NR

[0016] In NR, a UE can be configured with multiple CSI reporting settings and multiple CSI-RS resource settings. Each resource setting can contain multiple resource sets, and each resource set can contain up to 8 CSI-RS resources. For each CSI reporting setting, a UE feeds back a CSI report.

[0017] Each CSI reporting setting contains at least the following information: • A CSI-RS resource setting for channel measurement • An IMR resource set for interference measurement • Optionally, a CSI-RS resource set for interference measurement • Time-domain behavior, i.e., periodic, semi-persistent, or aperiodic reporting • Frequency granularity, i.e., wideband or subband • CSI parameters to be reported such as RI, PMI, CQI, and CSI-RS resource indicator (CRI) in case of multiple CSI-RS resources in a resource set • Codebook types, i.e., type I or II, and codebook subset restriction • Measurement restriction • Subband size. One out of two possible subband sizes is indicated, the value range depends on the bandwidth of the BWP. One CQI / PMI (if configured for subband reporting) is fed back per subband.

[0018] DFT-based precoders

[0019] A common type of precoding is to use a DFT-precoder, where the precoder vector used to precode a single-layer transmission using a single-polarized uniform linear array (ULA) with ^^ antennas is defined as:^^2 ^^⋅0⋅^^ ^^ ^^ ^^ ^^ , where ^^ = 0,1, … ^^ ^^ − 1 is thefactor. ^^^^is also referred to as an one dimension (1-D) DFT beam with beam index ^^. If ULA is along the horizontal dimension, each DFT beam points to an azimuth direction. If ULA is along the vertical dimension, each DFT beam points to an elevation direction. Each precoder corresponds to a DFT beam.

[0020] A corresponding precoder vector for a two-dimensional uniform planar array (UPA) with ^^1antenna ports in one dimension and ^^2antenna ports in another dimension can be created by taking the Kronecker product of two precoder vectors as ^^2 ^^(^^, ^^)= ^^^^, ^^= ^^^^,1^ ^^^^,2, ^^ ^^ ^^2 ^^⋅0⋅^^^2 ^^⋅0⋅ ^^^1^^1^^^^2^^2^^ ^^ inwith ^^1and ^^2, respectively. ^^^^, ^^is also referred to a two dimension (2-D) DFT beam characterized by two beam indices ( ^^, ^^), one in each dimension. Each precoder corresponds to a 2D DFT beam.

[0021] Extending the DFT precoder for a dual-polarized UPA may then be done as: ^^ ( ^^, ^^) ^^ ( ^ ) ^^2 ^^, ^^ ^^^^, ^^, ^^ = 1 ^^ ^^ ^^2 ^^^^, ^^ = [2 ^^^, ^^ ^^ ( ) [ ] ( ) ] = [2 ^^]

[0001] , ^^^^^^^ ^^^^^^^ ^^ ^^^^^^^^ ^^^^ 3 ^^∈ {0, 2 , ^^, 2 }. A precoder matrix ^^2 ^^, ^^ ^^for multi-layer transmission may be created by appending columns of DFT precoder vectors as: ^^2 ^^, ^^ ^^ = [^^2 ^^, ^^ ^^( ^^1, ^^1, ^^1) ^^2 ^^, ^^ ^^( ^^2, ^^2, ^^2) ⋯ ^^2 ^^, ^^ ^^( ^^^^, ^^^^, ^^^^)],for instance in NR Type I CSI feedback, where each layer is associated with 2D DFT beam.

[0023] NR Rel-16 enhanced Type II (eTypeII) codebook

[0024] The Rel-15 type II codebook is enhanced in NR Rel-16 in which instead of reporting separateprecoders for different subbands, the precoders for all subbands are reported together by using a so called frequency domain (FD) basis. It takes advantage of frequency domain channel correlations by representing the precoder changes in frequency domain with a set of frequency domain DFT basis vectors (which will be simply referred to as frequency domain basis vectors). Due to channel correlation in frequency, only a few DFT basis vectors may be used to represent the precoder changes over all the subbands. By doing so, the feedback overhead can be reduced or performance can be improved for the same feedback overhead.

[0025] For a given CSI-RS resource with ^^1CSI-RS antenna ports in one dimension and ^^2CSI-RS antenna ports in another dimension, and with two polarizations, the Rel-16 type II codedbook based precoding vectors for each layer ^^ ( ^^ = 1, … , ^^) and across all subbands can be expressed as: ^^^^= [ ^^(0) ( ^^^^… ^^^^3−1)] =^^ ^^× 1 precoding vector at a PMI subband with subband index ^^ ∈ {0,1, … , ^^3− 1} for layer ^^, where ^^^^ ^^ ^^− ^^ ^^= ^^ ^^^^^^^^is the number of CSI-RS ports in a configured NZP CSI-RS resource;

[0027] ^^3= ^^^^ ^^× ^^ is the number of subbands for PMI, where ^^^^ ^^is the number of CQI subbands and ^^ ∈ {1,2} is a scaling factor, both ^^^^ ^^and ^^ are RRC configured

[0028] ^^^^is the same as in Rel-15 type II codebook and contains a set of selected beams or SD basis vector

[0029] ^^f,l= [ ^^(^^0), ^^(^^1), … , ^^( ^^^^^^−1)] is a size ^^3× ^^^^frequency domain (FD) compression matrix for prising ^^ selected FD basis vectors and ( ) ( ^^) ( ^^) ( ^^ ^^ layer ^^ com ^^ ) ( ^^) ^^^^^^= [ ^^0, ^^, ^^1, ^^, … , ^^^^3−1, ^^] and ^^^^, ^^= ^^− ^^2 ^^ ^^ ^^( ^^) 3, ^^ / ^^3, ^^ = 0,1, … , ^^3− 1, ^^ ( ^^) 3, ^^∈ {0,1, … , ^^3− 1} . ^^^^=⌈^^^^^^3^^ ⌉ is the number of selected FD basis of ^^^can be found in^Error! Reference source not found..

[0030] For ^^3≤ 19, a one-step free selection is used.

[0031] ^^3− 1 For each layer, the selected FD basis vectors are indicated with a ⌈log2( ^^^^− 1)⌉ bit combinatorial indicator. In TS 38.214, the combinatorial indicator is given by the index ^^^^, which UE to the

[0032] For ^^3> 19, a two-step selection with layer-common intermediary subset (IntS) is used.

[0033] In the first step, a window-based layer-common IntS selection is used, which is parameterized by ^^^^ ^^ ^^ ^^ ^^ ^^ ^^. The IntS consists of FD basis vectors { mod( ^^^^ ^^ ^^ ^^ ^^ ^^ ^^+ ^^, ^^3), ^^ = 0, 1, … , 2 ^^^^− 1 }. In TS 38.214, the selected IntS is reported by the UE to the gNB via the parameter ^^1,5, which is reported per layer as part of the PMI reported.2M

[0034] ectors are indicated with an ⌈log2(v− 1 In the second step, the selected FD basis v ^^^^− 1 )⌉-bit combinatorial indicator for each layer. In TS 38.214, the combinatorial indicator is given by the index ^^1,6, ^^, which is reported by UE to the gNB.

[0035] ^^̃2,l= [ ^^̃^^, ^^, ^^, ^^ = 0,1, … ,2 ^^ − 1, ^^ = 0,1, … , ^^^^− 1] is a size 2 ^^ × ^^^^coefficient matrix. For layer ^^, only a subset of ^^^^^^ ^^≤ ^^ coefficients are non-reported by the UE. The remaining 2 ^^ ^^^^^^^^^^ ^^non-reported coefficients are considered zero.

[0036] ^^0= ⌈ ^^ × 2 ^^ ^^1⌉ is the maximum number of non-zero coefficients per layer, where ^^ is a RRC configured ^^ values are shown in Error! Reference source not found..

[0037] For ^^ ∈ {2, 3, 4} , the total number of non-zero coefficients summed across all layers, ^^^^^^^^^^^^ = ∑ ^^ ^^=1^^ ^^ ^^ ^^, shall satisfy ^^^^^ ^^^^^^^≤ 2 ^^0.

[0038] Selected coefficient subset for each layer is indicated with ^^^^^^ ^^1s in a size 2 ^^ ^^^^bitmap, ^^1,7, ^^.

[0039] The strongest coefficient of layer ^^ (whose amplitude and phase are not reported) is identified by ^^1,8, ^^,∈{0,1,…,2 ^^−1} .

[0040] The amplitude coefficients in ^^2, ^^are indicated by ^^2,3, ^^and ^^2,4, ^^, and the phase coefficients in ^^2, ^^are indicated by ^^2,5, ^^.

[0041] The above is described in TS38.214, section 5.2.2.2.5, where ^^( ^^) ^^is expressed as 1 ∑ ^^−1 ^^=0 ^^^^(^^), ^^( ^^) ^^(1)^^,0∑^^^^−1^^=0 ^^ ( ^^) ^^, ^^^^ (2) ^^, ^^, ^^^^^^, ^^, ^^^^^^1 2^^^^ ^^(1)^^ ( ^^) ^^ (2) ^^ | (1)where { ^^1, ^^2, ^^1, ^^2, ^^3, ^^, ^^^^,^^,2,5, ^^are aand ^^ (2) ^^, ^^, ^^is the subband amplitude of the coefficient ^^̃ (2) (2) (2) (2) (2) ^^, ^^, ^^, where ^^^^, ^^, ^^is part of ^^^^= [ ^^^^,0… ^^^^, ^^ ^^−1], ^^^^, ^^= [ ^^ (2) ^^,2 ^^−1, ^^],2^^ ^

[0045] ^^^^, ^^, ^^= ^^ ^^^^^, ^^, ^^16is phase of the coefficient ^^̃^^, ^^, ^^, where ^^^^, ^^, ^^∈{0,… ,15}is part of ^^2,5, ^^= [ ^^^^,0… ^^^^, ^^ ^^−1], ^^^^, ^^= [ ^^^^,0, ^^… ^^^^,2 ^^−1, ^^]parameter configurations for ^^, ^^ and ^^^^for Rel-16 enhanced type II codebook: paramCombination-r16 ^^^^ ^^^^^^ ∈ {1,2} ^^ ∈ {3,4} 1 2 ¼ 1 / 8 ¼ 2 2 ¼ 1 / 8 ½ 3 4 ¼ 1 / 8 ¼ 4 4 ¼ 1 / 8 ½ 5 4 ¼ ¼ ¾ 6 4 ½ ¼ ½ 7 6 ¼ - ½ 8 6 ¼ - ¾

[0047] 3GPP NR Rel-18 Type II CSI enhancement

[0048] In 3GPP NR Rel-18, the Type II CSI has been further enhanced to support high / medium mobility. In particular, the Rel-16 enhanced Type II (eType II) codebook and the Rel-17 further enhanced Type II (feType II) port- selection (PS) codebook have been enhanced to support UE reporting CSI for future time slot. In the sequel, the enhanced Rel-18 Type II codebook will be referred to as the Rel-18 evoType II codebook, or simply the Rel-18 Type II codebook, as the codebook is enhanced within the MIMO evolution work item in 3GPP NR Rel-18, and no official name is available yet. The Rel-18 evoType II codebook.

[0049] CMR enhancement for Type II CSI prediction (a.k.a. Rel-18 Type II CSI) has been introduced, see the measurement part in Error! Reference source not found.. To be more specific, a single burst of ^^ ∈ {4, 8, 12} CSI- RS resources can be configured to the UE within a single CSI-RS resource set, which can be aperiodically triggered using a single downlink control information (DCI). The CSI-RS resources are uniformly spaced in time, separated by ^^ ∈ {1, 2} slots, within the resource set.

[0050] For the Rel-18 evoType II codebook enhancement, UE can be configured by gNB to report predicted PMIs for ^^4∈ {1, 2, 4} time slots, see the Rel-18 Type II PMI part in FIG.4. Note that the prediction herein is relative to the CSI-RS reference resource. The predicted ^^4PMIs are supposed to reflect the channels with ^^ ∈ {1, 2} slotsseparation, starting from ^^ ∈ {0,1,2} slots into the future relative to the CSI-RS reference resource. The spacing ^^ between the ^^4PMIs and offset ^^ relative to the CSI-RS reference resource can be configured by the gNB via RRC signaling. The ^^4PMIs are compressed in a beam-frequency-Doppler domain, and the compressed PMI is reported to the gNB in a single CSI report.

[0051] Essentially, each of the ^^4PMI follows the 3GPP NR Rel-16 eType II codebook structure. However, the ^^4PMIs are assumed to share the same spatial domain and frequency domain basis vectors, i.e., ^^1and ^^^^, as ^^1and ^^^^reflect the distribution of angle of departure (AoD) and delay spread, which are large scale fading channel parameters that do not change quickly over time. The varying part across the ^^4PMIs is the linear combination coefficient matrix ^^2. Hence, the PMI for the ^^-th instance, for ^^ = 1,… , ^^4can be written as ^^1^^2, ^^^^^^^^, where ^^2, ^^denotes the combination coefficient matrix for the ^^-th instance. Finally, ^^2, ^^is compressed in time domain with DFT-based Doppler domain basis before being reported to the gNB as part of a CSI report.

[0052] Channel prediction with Kalman filter

[0053] Kalman filter basics

[0054] The Kalman filter is a recursive estimator, which means that only the estimated state from the previous time step and the measurement at the current time step are needed to compute the estimate for the current state. Typically, the Kalman filter can be decomposed into two steps: a prediction step and an update step, which are further explained below.

[0055] Prediction step

[0056] The Kalman filter assumes that the true state at time ^^ is evolved from time ^^ − 1 according to ^^^^= ^^ ^^^^−1+ ^^ ^^^^+ ^^^^, where ^^ is a time-invariant state transition model (for wide sense stationary processes), ^^^^is the complex Gaussian distributed process noise with zero mean and covariance ^^^^. The covariance ^^^^reflects the confidence in prediction accuracy: a larger variance means less confidence in the prediction due to higher uncertainty, and vice versa. The terms ^^ and ^^^^are the control-input model and the control vector, respectively. Typically, they can be assumed known.

[0057] The prediction step uses an old state estimate from the previous time step, ^^̂^^−1| ^^−1, to produce a new state estimate for the current time step, ^^̂^^| ^^−1. The predicted state estimate ^^̂^^| ^^−1, also known as the a priori state estimate, is given by ^^̂^^| ^^−1= ^^ ^^̂^^−1| ^^−1+ ^^ ^^^^, which is the maximum likelihood estimate, as the process noise ^^^^has zero mean.

[0058] Along with predicting the state ^^̂^^| ^^−1, its covariance, ^^^^| ^^−1= Ε( ^^̂^^| ^^−1^^̂^^^^|^^−1) , with Ε(∙)being the expectation operator, is calculated according to: ^^^^| ^^−1= ^^ ^^^^−1| ^^−1^^^^+ ^^^^, which will be neededlater for the update step.

[0059] Update step

[0060] The update step is based on a noisy measurement ^^^^, wherein the true state ^^^^underlies. The measurement model can be expressed as ^^^^= ^^ ^^^^+ ^^^^, where ^^ is the measurement model that do not change over time for wide sense stationary processes, while ^^^^is the measurement noise, which is assumed to be complex Gaussian distributed with zero mean and covariance ^^^^that is also independent of time.

[0061] Based on the measurement ^^^^, which contains the true state ^^^^, the a posteriori state estimate, ^^̂^^| ^^, is obtained by refining the a priori state estimate, ^^̂^^| ^^−1, according to: ^^̂^^| ^^= ^^̂^^| ^^−1+ ^^^^( ^^^^− ^^ ^^̂^^| ^^−1), where the term ^^^^is the so-called optimal Kalman gain, which is given by ^^^^= ^^| ^^−1^^^−1 ^^^( ^^ ^^^^| ^^−1^^^^+ ^^^^)gain minimizes the residual error ^^^^− ^^ ^^̂^^| ^^−1in the mean squared sense. In essence, it is the linear minimum mean squared error (LMMSE) estimator. Along with updating the state estimate, the covariance of the state is also updated according to: ^^^^| ^^= ( ^^ − ^^^^^^) ^^^^| ^^−1

[0063] The above two steps, i.e., theusually alternate, thereby forming a time advanced estimate / predict of the state. However, note that it is possible to skip the prediction or measurement step at times. For example, if a measurement is unavailable for some reason, the update step can be skipped, and multiple prediction procedures can be performed.

[0064] Channel prediction with Kalman filter

[0065] One application of the Kalman filter is to predict the channel evolution over time. For this purpose, the control-input model ^^ and control vector ^^^^can be dropped for simplicity, which does not affect the optimality for channel prediction problem. Then, the state transition model in the prediction step becomes an auto-regression (AR) model with order 1, i.e., ^^^^= ^^ ^^^^−1+ ^^^^

[0066] Given that the current state ^^^^−1and the state transition model ^^ are both known, it is possible to predict a future state ^^^^via the prediction step. In the context of channel prediction, ^^^^−1can be considered as the channel known up to time ^^ − 1, which can be approximated by its estimate ^^̂^^−1. Then, only the state transition model ^^ is yet to be estimated, which is well-known in literature. One popular way of doing this is to use the Yule- Walker equations.

[0067] In order to derive the state transition model ^^, first assume that an AR model with order ^^ is used for predicting the channel, i.e., ^^^^=∑^^ ^^=1^^^^^^^^− ^^+ ^^^^, where ^^^^= vec(^^^^)is the vectorized narrowband channelwith ^^^^being the narrowband channel between a gNB with ^^ antenna ports and a UE with ^^ antenna ports at time ^^ . The AR parameter matrix, ^^^^, can be estimated using the Yule-Walker equations[^^1^^2⋯ ^^^^]= [ ^^1^^2⋯ ^^^^] ^̅^, where ^^0^^1⋯ ^^^^−1^^^⋯ ^^ , ] and ^^^^= Ε[ ^^^^^^^^^^−^^] is the auto-^^the AR parameters ^^^^∈ ℂ^^ ^^× ^^ ^^canbe obtained via[^^1 ^^2⋯ ^^^^]=[^^1 ^^2⋯ ^^^^]^̅^−1.

[0068] In addition, the covariance matrix of process noise, ^^^^, can be calculated by ^^ ^^^^= − ^^^^^^^^^^.

[0069] The above estimated AR to the state transition model in the Kalman filterprediction step via the following: ^^̃^^= ^^̃ ^^̃^^−1+ ^^̃ ^^^^, where ^^̃^^= [^^ ^^^^⋯ ^^^^^^^ ^− ^^+1] ∈ ℂ^^ ^^ ^^×1is the state vector obtained by concatenating the matrices^^̃ for state and ^^̃ for processing noise are given by ^^1^^2⋯ ^^^^−1^^^^^^^^ ^^^^^^ ^^⋯ ^^^^ ^^^^^^ ^^^^, where ^^^^ ^^and ^^^^ ^^arerespectively.

[0070] To summarize, if an AR model of order ^^ is used for prediction, the ML estimate of the channel for time ^^ based on channel estimates at time ^^ − 1, ^^ − 2,… , ^^ − ^^ + 1, is given by ^^̂̃^^| ^^−1= ^^̃ ^^̂̃^^−1| ^^−1.

[0071] In a special case where an AR model of order 1 is used, the predicted channeldepends on the channel at time ^^ − 1, i.e., ^^^^= ^^1^^^^−1. SUMMARY

[0072] A wireless propagation channel between a UE and a base station (e.g., gNB) is highly time-varyingwhen there are high UE mobilities (e.g., when there are many UEs leaving the cell covered by the base station and / or when there are many UEs entering the cell). This highly time-varying nature of the wireless propagation channel is a major performance impediment in many communication systems. For example, the highly time-varying nature may cause a time delay between uplink transmission and downlink transmission between a UE and a base station, and thus the CSI the UE transmitted to the base station may already be outdated (channel aging) by the time the network node received the CSI. To solve this channel aging problem, CSI prediction may be used.

[0073] As discussed above, in the 3GPP NR CSI reporting framework, the CSI prediction can be performed based on the state-of-the-art Rel-18 Type II codebook disclosed in 3GPP R1-2212174, “On CSI enhancements for Rel-18 NR MIMO evolution,” Ericsson, 3GPP RAN WG 1 Meeting #111, Nov. 2022. Even though the 3GPP specifications do not explicitly mention, the CSI prediction based on the Rel-18 Type II codebook can be treated as a UE-sided prediction, meaning that a UE predicts the quality of a channel between the UE and a base station, and feeds back a predicted CSI report to the base station.

[0074] But predicting the channel quality at the UE side has certain drawbacks. For example, performing such prediction at the UE side may require a highly complex UE as the UE needs to perform the prediction based on several channel measurements that span over time. Also, because different UE vendors may have different proprietary implementations of prediction algorithms, performing channel quality predictions at different UEs may result in a non-uniform prediction quality. As a result, the overall system performance might be affected, as the base station which determines the transmission schemes (e.g., resource allocation, scheduling decisions) may need to handle CSI reports having different prediction qualities.

[0075] Therefore, there is a need for a method and an apparatus / system that enable performing a channel quality prediction at a network node side (e.g., a base station side). Accordingly, in one aspect of some embodiments of this disclosure, there is provided a method performed by a wireless communication device (WCD). The method comprises receiving wireless signals from a network node, based on the received wireless signals, estimating a channel between the WCD and the network node, and generating first singular vector, SV, group information related to a first group of one or more SVs of the estimated channel. The method further comprises generating second SV group information related to a second group of one or more SVs of the estimated channel, compressing the first SV group information, thereby generating compressed first SV group information, compressing the second SV group information, thereby generating compressed second SV group information, and transmitting the compressed first and second SV group information to the network node.

[0076] In another aspect, there is provided a method performed by a network node. The method comprises transmitting wireless signals to a wireless communication device, WCD; and after transmitting the wireless signals, receiving from the WCD compressed first and second singular vector, SV, group information, wherein the compressedfirst SV group information is related to a first group of one or more SVs of a channel between the WCD and the network node, and the compressed second SV group information is related to a second group of one or more SVs of the channel.

[0077] In another aspect, there is provided a computer program comprising instructions which when executed by processing circuitry cause the processing circuitry to perform the method of at least one of the embodiments described above.

[0078] In another aspect, there is provided a carrier containing the computer program of the above embodiment, wherein the carrier is one of an electronic signal, an optical signal, a radio signal, and a computer readable storage medium.

[0079] In another aspect, there is provided a wireless communication device, WCD. The WCD is configured to receive wireless signals from a network node; based on the received wireless signals, estimate a channel between the WCD and the network node; and generate first singular vector, SV, group information related to a first group of one or more SVs of the estimated channel. The WCD is further configured to generate second SV group information related to a second group of one or more SVs of the estimated channel; compress the first SV group information, thereby generating compressed first SV group information; compress the second SV group information, thereby generating compressed second SV group information; and transmit the compressed first and second SV group information to the network node.

[0080] In another aspect, there is provided a network node. The network node is configured to: transmit wireless signals to a wireless communication device, WCD; and after transmitting the wireless signals, receive from the WCD compressed first and second singular vector, SV, group information, wherein the compressed first SV group information is related to a first group of one or more SVs of a channel between the WCD and the network node, and the compressed second SV group information is related to a second group of one or more SVs of the channel.

[0081] In another aspect, there is provided an apparatus comprising: a processing circuitry; and a memory, said memory containing instructions executable by said processing circuitry, whereby the apparatus is operative to perform the method of at least one of the embodiments described above.

[0082] Some embodiments of this disclosure allow performing a channel quality prediction at a network node side with reduced overhead. More detailed explanation as to how some embodiments of this disclosure enable reducing overhead while performing a network node side channel quality prediction is provided below. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] The accompanying drawings, which are incorporated herein and form part of the specification, illustrate various embodiments.

[0084] FIG.1 shows an example of a spatial multiplexing.

[0085] FIG.2 shows an example of a 2D antenna array of dual-polarized antenna elements.

[0086] FIG.3 shows an example of CSI-RS REs for 12 antenna ports.

[0087] FIG.4 shows an example of how a UE can be configured by a base station to report predicted PMIs.

[0088] FIG.5 shows a portion of a system according to some embodiments.

[0089] FIG.6 shows a process according to some embodiments.

[0090] FIG.7 shows an example of downlink channel estimate.

[0091] FIG. 8A, 8B, 9A, and 9B illustrate how a preprocessing may improve the phase evolution over frequency subbands.

[0092] FIG.10 shows a simulation result showing different user throughputs achieved according to some embodiments.

[0093] FIG.11 shows a process according to some embodiments.

[0094] FIG.12 shows a process according to some embodiments.

[0095] FIG.13 shows an apparatus according to some embodiments.

[0096] FIG.14 shows an apparatus according to some embodiments. DETAILED DESCRIPTION

[0097] FIG.5 shows a portion of a system 500 according to some embodiments. The system 500 comprises a UE 502 (a.k.a., a wireless communication device (WCD)) and a network node (“NN”) 104. Examples of the UE 502 include a mobile phone, a tablet, a computer, an Internet of Things (IoT) device, a vehicle, or any other electronic device capable of performing a wireless communication. An example of the NN 504 is a base station such as eNB, gNB, etc.

[0098] As discussed above, in order to increase the rate of data transmission from the UE 502 to the NN 504 and / or the rate of data transmission from the NN 504 to the UE 502, a spatial multiplexing can be used. As further discussed above, in performing the spatial multiplexing, a precoding matrix or precoder W (herein after just “precoder W”) may be needed, and the precoder W may be determined based on a prediction of the quality of the channel between the UE 502 and the NN 504. One example of such channel quality prediction is CSI prediction.

[0099] The CSI prediction may be performed at the UE 502. But performing the CSI prediction at the UE 502 may require the UE 502 to have certain minimum computational power. Therefore, UEs that are capable of performing the CSI prediction may generally be complex. Furthermore, because different UE vendors may havedifferent proprietary implementations of CSI prediction algorithms, the CSI predictions performed at different UEs in the same cell may produce inconsistent results.

[0100] Accordingly, in some embodiments of this disclosure, CSI prediction is performed at the NN 504. In case the NN 504 is an gNB, the CSI prediction may also be called “gNB-sided CSI prediction.” The gNB-sided CSI prediction may be used by an gNB for predicting and / or deriving an accurate MU-MIMO precoder for a future time slot.

[0101] In order to enable the NN 504 to perform the CSI prediction, the UE 502 needs to transmit to the NN 504 data that is needed for performing the CSI prediction. Such data may be called “CSI report,” and the UE 502’s transmission of the CSI report may be called “CSI reporting.”

[0102] FIG. 6 shows a process 600 for performing CSI reporting according to some embodiments. The process 600 may be performed by the UE 502. The process 600 may begin with step s602.

[0103] The step s602 comprises the UE 502 receiving wireless signal(s) (e.g., CSI-RS) from the NN 504 via a channel between the UE 502 and the NN 504. After performing the step s602, the process 600 may proceed to step s604. The step s604 comprises the UE 502 estimating the channel based on the received wireless signal(s), thereby generating a channel estimate at time ^^ and frequency ^^. Since this channel estimate is an estimate of the channel corresponding to the transmission of the wireless signal(s) from the NN 504 to the UE 502, the channel estimate may also be called a DL channel estimate. The DL channel estimate at time ^^ and frequency ^^ may be denoted as ^^^^, ^^∈ ℂ^^ ^^× ^^ ^^, where ^^^^is the number of CSI-RS ports configured for transmitting the CSI-RS and ^^^^is the ports at the UE 502 for receiving the CSI-RS. One simplified example of theDL channel estimate is shown in FIG.7.

[0104] In FIG.7, the UE 502 includes two receive antenna ports 702 and 704, and the NN 504 includes two CSI-RS ports 706 and 708. The NN 504 transmits CSI-RS via the two CSI-RS ports 706 and 708, and the UE 502 receives the transmitted CSI-RS via the two receive antenna ports 702 and 704. The CSI-RS received at the receive antenna ports 702 and 704 may be expressed as: ^^0= ℎ00∙ ^^0+ ℎ01∙ ^^1+ ^^0^^1= ℎ10∙ ^^0+ ℎ11∙ ^^1+ ^^1where ^^0is the signal transmitted from transmitted from the CSI-RS port 708,^^0is the signal received at the receive antenna ports 702, and ^^1is the signal received at the receive antenna ports 702. The factor ℎ^^ ^^identifies the complex transmission coefficient from the j-th CSI-RS port to the i-th receive antenna port. The factor ^^^^reflects the additional noise in the i-th receive antenna port. In this disclosure, estimating a channel may meana matrix of factors ℎ^^ ^^.

[0105] After determining the channel estimate ^^^^, ^^, the process 600 may proceed to step s606. The step s606 comprises performing a singular value of the channel estimate ^^^^, ^^, thereby obtaining[ ^^^^, ^^^^^^, ^^^^^^^,^^^] = svd( ^^^^, ^^), where ^^^^, ^^∈ ℂ^^ ^^× ^^ ^^and ^^^^, ^^∈ ℂ^^ ^^× ^^ ^^are composed of UE-side and gNB-side singular vectors at time ^^ and frequency ^^, respectively, while ^^^^, ^^∈ ℝ^^ ^^× ^^ ^^is composed of the singular values of ^^^^, ^^at time ^^ and frequency ^^. In this disclosure, a UE-side singular vector may mean a vector that is related to one or more spatial properties, amplitude information of one or more wireless signals received at the UE 502, and / or phase information of one or more wireless signals received at the UE 502. Similarly, a gNB-side singular vector may mean a vector that is related to one or more spatial properties, amplitude information of one or more wireless signals transmitted from the NN 504, and / or phase information of one or more wireless signals transmitted from the NN 504.

[0106] In some embodiments, instead of using an algorithm for performing the singular value decomposition, one or more of other algorithms (e.g., an algorithm for performing an eigen value decomposition or a machine learning (ML) / artificial intelligence (AI) algorithm) may be used in order to determine the gNB -side singular vector, the UE-side singular vector, and optionally the associated singular values of the estimated DL channel.

[0107] As explained above, in the step s606, the singular value decomposition is performed over the channel estimate ^^^^, ^^. However, in some embodiments, the channel estimate ^^^^, ^^may be transformed first, and the singular value decomposition may be performed over the transformed estimate. For example, thechannel estimate ^^^^, ^^may be transformed, using a function ^^(. ), from (CSI-RS port, UE antenna port) domain to (gNB transmit UE antenna port) domain, or to (gNB transmit beam, UE receive beam) domain, possibly with dimension reduction (e.g., removing insignificant beams, ports, etc.). Then, in the step s606, the singular value decomposition may be performed over ^^( ^^^^, ^^) where ^^(. ) represents a process of transforming ^^^^, ^^.

[0108] After obtaining the singular vectors and the UE-side singular vectors in step s606, theprocess 600 may proceed to steps s608 and s610. The step s608 comprises the UE 502 generating a CSI report based on the obtained the gNB-side singular vectors and / or the obtained UE-side singular vectors, and the step s610 comprises the UE 502 transmitting to the NN 504 the generated CSI report.

[0109] Instead of generating the CSI report based on the obtained the gNB-side singular vectors and / or the obtained UE-side singular vectors, the CSI report may be generated based on the channel estimate ^^^^, ^^. More specifically, without performing the step s606, the UE 502 may report the channel estimate ^^^^, ^^to 504. Ifthe channel estimate ^^^^, ^^is reported to the NN 504 for a number of time instances, e.g., for = ^^1, ^^2, … , ^^ , then^^the NN 504 can the channel for a future time slot ^^^^, ^^based on ^^^^, ^^= ^^( ^^^^,^^^^, ^^, … , ^^^^,),^^+1 ^^+12 ^^ ^^where ^^(. ) is a function that predicts a channel based time series aclassical model-based predictor (e.g., Kalman filter), or some trained AI / ML algorithm.

[0110] However, directly reporting the channel estimate ^^^^, ^^(either in a compressed form or in an uncompressed form) may introduce excessive overhead, especially channel is low rank. Also, directly reportingthe channel estimate ^^^^, ^^does not fit the 3GPP NR CSI reporting framework in a way that the channel quality indicator (CQI) becomes ambiguous. Directly reporting to the NN 504 the channel estimate ^^^^, ^without^reporting the gNB-side singular vectors and / or the obtained UE-side singular vectors may render the calculationambiguous because the CQI is calculated based on a reported precoding matrix indicator (PMI), e.g., a Type II PMI, which is essentially a quantized version of the gNB-side singular vector ^^^^, ^^.

[0111] One way to solve the above problem is only reporting the gNB-side singular vector. However, in case the UE 502 only reports the gNB-side singular vector to the NN 504, the NN 504 lacks UE-side channel information, which makes predicting a precoder at the gNB-side very challenging. This is because the singular value decomposition is not unique with respect to the gNB-side and UE-sided singular vectors.

[0112] Accordingly, in some embodiments, the CSI report includes additional information (e.g., information related to the UE-side singular vectors) as well as information related to the gNB-side singular vectors (e.g., the 3GPP NR eType II codebook, feType II PS codebook, or any CSI reporting that contains gNB -side singular vector reporting). More specifically, in some embodiments, the CSI report may contain at least the following two pieces of information: (1) information (e.g., compressed information) related to the gNB-side singular vectors for the configured frequencies ^^ = ^^1, ^^2, … , ^^^^and time instances ^^ = ^^1, ^^2, … , ^^^^, and (2) information (e.g., compressed information) the UE-side singular vectors for configured frequencies ^^ = ^^ , ^^ , … , ^1 2^^^and time instances ^^ = ^^1, ^^2, … , ^^^^. These two pieces of information may be sufficient for the NN performan accurate CSI prediction.

[0113] Preprocessing of Singular Vectors Before Compression

[0114] The singular value decomposition is not unique with respect to the gNB-side and UE-sided singular vectors. As a result, the values of elements of the singular vector ^^^^, ^^(: , ^^) (i.e., the ^^th column of ^^^^, ^^) may exhibit sudden changes along the ^^ and ^^ directions despite the channel being correlated in the ^^ and ^^ directions. In order to achieve a larger compression gain, according to some embodiments, the singular vectors may be preprocessed.

[0115] More specifically, in some embodiments, the UE-side singular vector ^^^^, ^^(: , ^^)(i.e., the ^^th column of ^^^^, ^^) and the gNB-side singular vector ^^^^, ^^(: , ^^) (i.e., the ^^ th column of preprocessed beforeby multiplying the elements of− ^^ ^^ ^^, ^^, ^^with ^^ .

[0116] Here, the phase used for preprocessing thesingular vector for layer ^^ may be chosen as ^^^^, ^^, ^^= ∠ ^^^^, ^^( ^^, ^^), where ^^ is a chosen reference row. By multiplying each element of the UE-side singular vector for layer ^^ and the chosen reference row ^^ by ^^− ^^∠ ^^ ^^, ^^( ^^, ^^), the phase of each element of the preprocessed UE-side singular vector for layer ^^ and the chosen row ^^ would be equal to zero or each element of thepreprocessed UE-side singular vector for layer ^^ and the chosen reference row ^^ would have a constant value for all frequencies ^^ (i.e., ∠ ^^′^^, ^^( ^^, ^^) = 0 or ^^′ ^^, ^^( ^^, ^^) is a constant for all frequencies ^^, where ^^′ ^^, ^^is a set of thepreprocessed UE-side singular vectors).

[0117] Similarly, the phase used for preprocessing the gNB-side singular vector for layer ^^ may be chosen as ^^^^, ^^, ^^= ∠ ^^^^, ^^( ^^, ^^),. 8B, 9A, and 9B illustrates how the preprocessing improves the phase evolution overfrequency subbands, which is helpful for further compression of ^^ and ^^. Each grid of a different grayscale in the figure indicates the amplitude of a complex entry of ^^ (in UE antenna port, subband domains) and ^^ (in gNB beam, subband domains), and the arrow on each grid pointing to a certain direction indicates the phase of the corresponding complex entry. It can be seen that before rotation, in both ^^ and ^^ the phases of the elements rapidly change along the direction of the frequency subbands. If ^^ and ^^ were to be transformed from frequency subband domain, e.g., using DFT, it would result in higher number of significant components than optimal in the sparser domain. This disadvantage can be overcome when preprocessing is applied on ^^ (and accordingly on ^^) so that the strongest row of ^^ always has zero phase. After preprocessing, phase variations of the elements in both ^^ and ^^ become continuous along the frequency subband direction, especially for the rows (e.g., for a UE antenna port, or for a gNB beam) with stronger power.

[0119] Compression of Information Related to gNB-side Singular Vectors

[0120] As discussed above, in some embodiments, the CSI report includes information related to the gNB- side singular vectors (e.g., the information indicating / identifying the gNB-side singular vectors). However, in other embodiments, the CSI report includes a compressed version of the information related to the gNB-side singular vectors. In such embodiments, the process 600 may include an optional step s612.

[0121] In the step s612, the UE 502 may compress the information related to the gNB-side singular vectors in the frequency domain by approximating the gNB-side singular vectors in frequency domain with ^^′< ^^ basis vectors (e.g., discrete Fourier transform (DFT) vectors). In this case, the selected ^^′basis vectors may be reported to the NN 504 as a part of the CSI report. In other words, the selected ^^′basis vectors may correspond to the compressed information related to the gNB-side singular vectors.

[0122] Instead of compressing the information related to the gNB-side singular vectors in the frequency domain, the UE 502 may compress the information in the time domain by approximating the gNB-side singular vectors in the time domain with ^^′< ^^ basis vectors (e.g., discrete Fourier transform (DFT) vectors). In this case, the selected ^^′basis vectors may be reported to the NN 504 as a part of the CSI report. In other words, the selected ^^′basis vectors may correspond to the compressed information related to the gNB-side singular vectors.

[0123] In some embodiments, the compressed information related to the gNB-side singular vectors corresponds to one of the followings: 1. The 3GPP NR Rel-16 eType II codebook / PMI2. The 3GPP NR Rel-16 eType II PS codebook / PMI 3. The 3GPP NR Rel-17 feType II PS codebook / PMI 4. The 3GPP NR Rel-18 evoType II codebook / PMI 5. The 3GPP NR Rel-18 evoType II PS codebook / PMI

[0124] Compression of Information Related to UE-side Singular Vectors

[0125] As discussed above, in some embodiments, the CSI report includes information related to the UE - side singular vectors (e.g., the information indicating the UE-side singular vectors). However, in other embodiments, the CSI report includes a compressed version of the information related to the UE -side singular vectors. In such embodiments, the process 600 may include an optional step s614. The step s614 may be performed before or after the step s612. Alternatively, the steps s612 and s614 may be performed simultaneously.

[0126] In the step s614, the UE 502 may compress the information related to the UE-side singular vectors in the frequency domain. There are various ways of compressing the information related to the UE-side singular vectors in the frequency domain.

[0127] In some embodiments, the information related to the UE-side singular vectors may be compressed in the frequency domain by approximating the UE-side singular vectors in frequency domain with ^^′< ^^ basis vectors (e.g., discrete Fourier transform (DFT) vectors). In this case, the selected ^^′basis vectors may be reported to the NN 504 as a part of the CSI report. In other words, the selected ^^′basis vectors may correspond to the compressed information related to the UE-side singular vectors.

[0128] In other embodiments, the information related to the UE-side singular vectors may be compressed in the frequency domain by reporting only frequency-averaged singular vector (i.e., wideband). For example, let’s assume that the UE 502 is configured to report ^^^^, ^^∈ ℂ^^ ^^× ^^ ^^for the first two layers, for ^^ = ^^1, ^^2, … , ^^^^, ^^ = ^^0. Then, the UE 502 only needs to reportfor the first layer(: ,2) for the second layer, where ^^(: , ^^) is the ^^tha complex-^^^^. Hence, in this case, the UE 502 only needs to report two complex-valued vectors with size ^^^^to the NN 504. Assuming each complex-valued scalar is quantized with 4 bits in phase and 4 bits in amplitude, then this reporting will only introduce 8 ^^ ^^^^bits for the CSI report, where ^^ is the reported rank. More detailed example illustrating how the information related to the UE-side singular vectors may be compressed in the frequency domain is provided below.

[0129] Let’s assume that there are 2 frequency subbands, and that the UE 502 has obtained the DL channel estimates ^^^^, ^^∈ ℂ^^ ^^× ^^ ^^for frequency subband ^^ = 1, 2 at time ^^ = 0. Further assume that the UE 502 has ^^^^= 2 antenna ports and ^^^^= 2 transmit antenna ports, and the channel estimates are as follows: ^^1,0= [0.7731 − 0.5442 ^^ −0.6107 − 0.1595 ^^ ] 0.7844 + 0.2626 ^^ 0.0547 + 0.7901 ^^^^2,0= [−0.8585 − 0.7701 ^^ −0.0048 + 0.3907 ^^ ] −0.7874 + 0.0230 ^^ 1.0837 + 0.7782 ^^

[0130] In the step s606, the UE 502 calculate a singular vector decomposition of the above channel estimates for each frequency subband ^^ = 1,2 via [ ^^^^, ^^, ^^^^, ^^, ^^^^^,^^^] = svd( ^^^^, ^^), which results i the following SVD decomposition of ^^1,0and ^^2,0^^1,0= ^^1,0^^1,0^^1^,^0 ^^ −0.0372 − 0.7160 ^^] [1.2821 0 ] [−0.9507 + 0.0000 ^^ 0.3100 + 0.0000 ^^] − ^^ −0.2858 + 0.6358 ^^ 0 0.9798 0.2075 + 0.2303 ^^ 0.6365 + 0.7063 ^^ ^^2,0= ^^2,0^^2,0^^ ^^ ,0= [−0.4241 − 0.4083 ^^ 0.6389 + 0.4953 ^^ ] [1.8539 0 ] [ 0.7092 + 0.0000 ^^ −0.7050 + 0.0000 ^^] −0.8066 + 0.0540 ^^ −0.5798 + 0.1018 ^^ 0 0.6681 −0.5338 + 0.4606 ^^ −0.5369 + 0.4633 ^^

[0131] Then, in the step s614, the UE 502 averages the UE-side singular vector for the two reported layers. The averaged UE-side singular vector for the first layer denoted as ^̅^^^, ^^(: ,1) may be obtained as follows: 1 1 ^̅^^^, ^^(: ,1)= ( ^^ : ,1 ) ([−0.6435 + 0.2681 ^^] [−0.4241 − 0.4083 ^^]) 21,0( )+ ^^2,0(: ,1)= + 2 −0.7147 − 0.0570 ^^ −0.8066 + 0.0540 ^^ = [−0.5338 − 0.0701 ^^] −0.7606 − 0.0015 ^^

[0132] Similarly, the averaged UE-side singular vector for the second layer denoted as ^̅^^^, ^^(: ,2)may be obtained as follows: 1 1 ^̅^^^, ^^(: ,2)= ( ^^1,0(: ,2)+ ^^2(: ,2)) = ([−0.0372 − 0.7160 ^^] + [ 0.6389 + 0.4953 ^^ ]) 2,02 −0.2858 + 0.6358 ^^ −0.5798 + 0.1018 ^^ = [ 0.3009 − 0.1104 ^^ ] −0.4328 + 0.3688 ^^

[0133] Then, in the step s610, the UE 502 reports to the NN 504 ^̅^^^, ^^(: ,1)and ^̅^^^, ^^(: ,2). In some embodiments, the averaged UE-side singular vectors ^̅^^^, ^^(: ,1)and ^̅^^^, ^^(: ,2)may be normalized and quantized before being reported to the NN 504.

[0134] As discussed above, in some embodiments, the information related to the UE-side singular vectors may be compressed in the frequency domain by reporting only frequency-averaged singular vector (i.e., wideband). However, in other embodiments, the information related to the UE-side singular vectors may further be compressed in the frequency domain by reporting only the phase of the frequency-averaged singular vector (i.e., wideband). This 1 means that, in the example provided above, the UE 502 only needs to report the phase of ^^ ^^∑^^=1^^^^ ^^, ^^0(: , ^^) , which only introduces 4 ^^ ^^^^bits overhead if 4 bits is used for quantizing the phase.

[0135] As discussed above, in step s614, the UE 502 may compress the information related to the UE-side singular vectors in the frequency domain. However, in some embodiments, instead of compressing the information related to the UE-side singular vectors in the frequency domain, the UE 502 may compress the information in thetime domain by approximating the UE-side singular vectors in the time domain with ^^′< ^^ basis vectors (e.g., discrete Fourier transform (DFT) vectors). In this case, the selected ^^′basis vectors may be reported to the NN 504 as a part of the CSI report. In other words, the selected ^^′basis vectors may correspond to the compressed information related to the UE-side singular vectors.

[0136] Reporting Singular Values

[0137] In some embodiments, in addition to the information related to the gNB-side and the UE-side singular vectors, singular values, i.e., the elements of the diagonal matrix ^^^^, ^^, are compressed and reported by the UE 502 to the NN 504 in the CSI report.

[0138] In one example, ^^ − 1 singular values may be reported for rank ^^ CSI report for a given frequency band (e.g., a CSI reporting subband). The singular values for all layers may be normalized with the strongest singular value so that the strongest singular value is always 1 and hence not reported. The remaining ^^ − 1 singluar values may be quantized and reported to the NN 504 in a CSI report.

[0139] In some embodiments, the reported ^^ − 1 singular values are used for a group of frequency bands and / or a group of time instances for the CSI report. The group size of frequency bands and / or time instances, for which the reported singular values are valid for, may be gNB configured, UE determined, or pre-determined (e.g., predetermined in 3GPP specifications).

[0140] The ^^ − 1 singular values, for example, can be obtained by averaging the ^^ − 1 singular values within each group of frequency bands and / or time instances.

[0141] FIG. 10 is a simulation result showing different user throughputs achieved in accordance with different contents of the CSI report.

[0142] In the simulation, the NN 504 (e.g., gNB) and the UE 502 are equipped with 16 and 2 antennas respectively, the number of CSI-RS ports is the same as the number of gNB antenna ports, and no CSI-RS precoding is used. A 10 MHz channel with 15 kHz subcarrier spacing at carrier frequency 3.5 GHz is generated according to the 3GPP CDL-C channel model, and non-line-of-sight (nLoS) propagation condition is modeled in the CDL-C channel model.

[0143] Also, the Rel-16 eType II codebook is used as the method for calculating the gNB-side singular vectors (same as the Rel-18 evoType II codebook) as well as the selected spatial domain basis ^^1and frequency domain basis vectors ^^^^. It is assumed that ^^1and ^^^^are selected according to the latest channel measurement and are kept constant until a new channelis available.

[0144] In the simulation, a UE speed of 10 km / h was assumed for the channel. It was further assumed that the CSI is delayed by 4 slots, which the Kalman filter compensates for by predicting into the future. A bank size of 80 slots was used, and the AR model order was 5. The CSI reporting subband size is 4 PRB.

[0145] The ideal predictor has access to every measurement and has the highest throughput. It is usedherein as an upper bound. Meanwhile, the other predictors only have access to channel measurement every 5 slots (i.e., 5 ms in this case), meaning they must predict all the other time slots. The Kalman filter predictor based on Rel- 16 eType II applies prediction on ^^2part of a Rel-16 eType II CSI report. However, due to the loss of essential information related to the UE-side singular vectors, its performance is the worst.

[0146] Three of the proposed solutions have been simulated, all of which require reporting of ^^2based on Rel-16 eType II codebook. In addition, the three schemes respectively require reporting of: 1. wideband UE-side singular vector + subband singular value 2. phase of wideband UE-side singular vector + subband singular value 3. phase of wideband UE-side singular vector + wideband singular value.

[0147] As can be seen in FIG.10, all the three alternatives outperform the legacy Rel-16 eType II based prediction by a large margin. All the three alternatives give similar throughput but the overhead varies quite a lot. Given that the conducted simulation has 13 subbands and 2 UE antenna ports, and 4 bits are used both for quantizing the phase and amplitude of a complex scalar, then, reporting wideband UE -side singular vector requires 16 bits per layer, reporting the phase of wideband UE-side singular vector requires 8 bits per layer, reporting subband singular value requires 104 bits per layer, reporting wideband singular value requires 8 bits per layer. Hence, the three alternatives will additionally require 120 bits per layer, 112 bits per layer, and 16 bits per layer, comparing to CSI prediction based on the Rel-16 eType II CSI

[0148] A key observation from the above result is that, it is sufficient to report the wideband UE-side singular vectors, or even only the phase of the wideband UE-side singular vectors to enable gNB-side CSI prediction, which requires very limited overhead.

[0149] FIG.11 shows a process 1100 performed by the WCD 502 according to some embodiments. The process 1100 may begin with step s1102. The step s1102 comprises receiving wireless signals from the NN 504. The step s1104 comprises, based on the received wireless signals, estimating a channel between the WCD and the network node. The step s1106 comprises generating first singular vector, SV, group information related to a first group of one or more SVs of the estimated channel. The step s1108 comprises generating second SV group information related to a second group of one or more SVs of the estimated channel. The step s1110 comprises compressing the first SV group information, thereby generating compressed first SV group information. The step s1112 comprises compressing the second SV group information, thereby generating compressed second SV group information. The step s1114 comprises transmitting the compressed first and second SV group information to the network node.

[0150] In some embodiments, the first group of one or more SVs of the estimated channel is related to one or more spatial properties, amplitude information of one or more wireless signals received at the WCD, and / or phase information of one or more wireless signals received at the WCD, and the second group of one or more SVs of theestimated channel is related to one or more spatial properties, amplitude information of one or more wireless signals transmitted from the network node, and / or phase information of one or more wireless signals transmitted from the network node.

[0151] In some embodiments, the process 1100 comprises performing one or more of: (i) a singular value decomposition, SVD, of the estimated channel, (ii) an eigen value decomposition of the estimated channel, and / or (iii) providing measurement values of the estimated channel to a machine learning, ML, algorithm, thereby calculating the first group of one or more SVs of the estimated channel and the second group of one or more SVs of the estimated channel.

[0152] In some embodiments, the process 1100 comprises performing the SVD of the estimated channel, and the SVD of the estimated channel corresponds to ^^^^, ^^^^^^, ^^^^^^^,^^^, where ^^^^, ^^is a first matrix including the first group of SVs at time t and frequency f, ^^^^, ^^is a second matrix including the second group of SVs at time t and frequency f, and ^^^^, ^^is a third matrix comprising a group of singular values of the estimated channel.

[0153] In some embodiments, each SV included in the first group of SVs is associated with a different frequency and / or a different timing.

[0154] In some embodiments, compressing the second SV group information related to the second group of SVs comprises determining a group of basis vectors that approximate the second group of SVs in a frequency domain and / or a time domain, a number of the basis vectors approximating the second group of SVs is less than a number of SVs included in the second group which was approximated by the group of basis vectors in the frequency domain and / or the time domain, and the compressed second SV group information indicates the basis vectors approximating the second group of SVs.

[0155] In some embodiments, compressing the first SV group information related to the first group of SVs comprises determining a group of basis vectors that approximate the first group of SVs in either a frequency domain and / or a time domain, a number of the basis vectors approximating the first group of SVs is less than a number of SVs included in the first group which was approximated by the group of basis vectors in the frequency domain and / or the time domain, and the compressed first SV group information indicates the basis vectors approximating the first group of SVs.

[0156] In some embodiments, each SV included in the first group of SVs is associated with a different frequency, compressing the first SV group information related to the first group of SVs comprises performing an averaging operation with respect to the SVs included in the first group in a frequency domain, thereby generating an averaged SV, and the compressed first SV group information indicates one or more characteristics of the averaged SV.

[0157] In some embodiments, the averaged SV comprises a plurality of complex numbers, and said one or more characteristics of the averaged SV comprises (i) both amplitudes and phases of the complex numbers or (ii) onlythe phases of the complex numbers.

[0158] In some embodiments, the process 1100 comprises applying a preprocessing to the first group of SVs and the second group of SVs, thereby generating a first group of preprocessed SVs and a second group of preprocessed SVs, wherein the group of basis vectors that approximate the first group of SVs approximate the first group of preprocessed SVs, and the group of basis vectors that approximate the second group of SVs approximate the second group of preprocessed SVs.

[0159] In some embodiments, the process 1100 comprises applying a preprocessing to the first group of SVs, thereby generating a first group of preprocessed SVs, wherein performing the averaging operation comprises averaging SVs included in the first group of preprocessed SVs, thereby generating the averaged SV.

[0160] In some embodiments, an SV included in the first group of SVs comprises a plurality of complex numbers, and applying the preprocessing to the first group of SVs comprises multiplying each of the plurality of complex numbers of the SV included in the first group of SVs by another complex number.

[0161] In some embodiments, the process 1100 comprises normalizing the singular values included in the third matrix, thereby generating a group of normalized singular values; and transmitting the normalized signal values to the network node.

[0162] FIG. 12 shows a process 1200 performed by the NN 504 according to some embodiments. The process 1200 may begin with step s1202. The step s1202 comprises transmitting wireless signals to the WCD 502. Step s1204 comprises, after transmitting the wireless signals, receiving from the WCD compressed first and second singular vector, SV, group information. The compressed first SV group information is related to a first group of one or more SVs of a channel between the WCD and the network node, and the compressed second SV group information is related to a second group of one or more SVs of the channel.

[0163] In some embodiments, the process 1200 comprises predicting the quality of the channel between the WCD and the network node; determining a precoding matrix based on the predicted quality of the channel; and communicating with the WCD using a spatial multiplexing, wherein the spatial multiplexing is determined based on the precoding matrix.

[0164] In some embodiments, the first group of one or more SVs of the channel is related to one or more spatial properties, amplitude information of one or more wireless signals received at the WCD, and / or phase information of one or more wireless signals received at the WCD, and the second group of one or more SVs of the channel is related to one or more spatial properties, amplitude information of one or more wireless signals transmitted from the network node, and / or phase information of one or more wireless signals transmitted from the network node.

[0165] In some embodiments, the first group of one or more SVs of the channel and the second group of one or more SVs of the channel are calculated based on one or more of: (i) a singular value decomposition, SVD, of the channel, (ii) an eigen value decomposition of the channel, and / or (iii) providing measurement values of the channel toa machine learning, ML, algorithm.

[0166] In some embodiments, the first group of one or more SVs of the channel and the second group of one or more SVs of the channel are calculated based on the SVD of the channel, and the SVD of the channel corresponds to ^^^^, ^^^^^^, ^^^^^^^,^^^, where ^^^^, ^^is a first matrix including the first group of SVs at time t and frequency f, ^^^^, ^^is a second matrix including the second group of SVs at time t and frequency f, and ^^^^, ^^is a third matrix comprising a group of singular values of the estimated channel.

[0167] In some embodiments, each SV included in the first group of SVs is associated with a different frequency and / or a different timing.

[0168] In some embodiments, the compressed second SV group information indicates a group of basis vectors approximating the second group of SVs in a frequency domain and / or a time domain, and a number of the basis vectors approximating the second group of SVs is less than a number of SVs included in the second group which was approximated by the group of basis vectors in the frequency domain and / or the time domain.

[0169] In some embodiments, the compressed first SV group information indicates a group of basis vectors approximating the first group of SVs in a frequency domain and / or a time domain, and a number of the basis vectors approximating the first group of SVs is less than a number of SVs included in the first group which was approximated by the group of basis vectors in the frequency domain and / or the time domain.

[0170] In some embodiments, each SV included in the first group of SVs is associated with a different frequency, and the compressed first SV group information indicates one or more characteristics of an averaged SV that is obtained by performing an averaging operation with respect to the SVs included in the first group in a frequency domain.

[0171] In some embodiments, the averaged SV comprises a plurality of complex numbers, and said one or more characteristics of the averaged SV comprises (i) both amplitudes and phases of the complex numbers or (ii) only the phases of the complex numbers.

[0172] In some embodiments, the group of basis vectors that approximate the first group of SVs approximate a first group of preprocessed SVs that are obtained by applying a preprocessing to the first group of SVs, and the group of basis vectors that approximate the second group of SVs approximate the second group of preprocessed SVs that are obtained by applying a preprocessing to the second group of SVs.

[0173] In some embodiments, performing the averaging operation comprises averaging SVs included in a first group of preprocessed SVs that are obtained by applying a preprocessing to the first group of SVs.

[0174] In some embodiments, an SV included in the first group of SVs comprises a plurality of complex numbers, and applying the preprocessing to the first group of SVs comprises multiplying each of the plurality of complex numbers of the SV included in the first group of SVs by another complex number.

[0175] In some embodiments, the process 1200 comprises receiving from the WCD normalized signal valuesthat are generated by normalizing the singular values included in the third matrix.

[0176] FIG.13 is a block diagram of the UE 502, according to some embodiments. As shown in FIG. 13, UE 502 may comprise: processing circuitry (PC) 1302, which may include one or more processors (P) 1355 (e.g., one or more general purpose microprocessors and / or one or more other processors, such as an application specific integrated circuit (ASIC), field-programmable gate arrays (FPGAs), and the like); communication circuitry 1348, which is coupled to an antenna arrangement 1349 comprising one or more antennas and which comprises a transmitter (Tx) 1345 and a receiver (Rx) 1347 for enabling UE 502 to transmit data and receive data (e.g., wirelessly transmit / receive data); and a local storage unit (a.k.a., “data storage system”) 1308, which may include one or more non-volatile storage devices and / or one or more volatile storage devices. In embodiments where PC 1302 includes a programmable processor, a computer program product (CPP) 1341 may be provided. CPP 1341 includes a computer readable medium (CRM) 1342 storing a computer program (CP) 1343 comprising computer readable instructions (CRI) 1344. CRM 1342 may be a non-transitory computer readable medium, such as, magnetic media (e.g., a hard disk), optical media, memory devices (e.g., random access memory, flash memory), and the like. In some embodiments, the CRI 1344 of computer program 1343 is configured such that when executed by PC 1302, the CRI causes UE 502 to perform steps described herein (e.g., steps described herein with reference to the flow charts). In other embodiments, UE 502 may be configured to perform steps described herein without the need for code. That is, for example, PC 1302 may consist merely of one or more ASICs. Hence, the features of the embodiments described herein may be implemented in hardware and / or software.

[0177] FIG.14 is a block diagram of the NN 504, according to some embodiments. As shown in FIG.14, the NN 504 may comprise: processing circuitry (PC) 1402, which may include one or more processors (P) 1455 (e.g., one or more general purpose microprocessors and / or one or more other processors, such as an application specific integrated circuit (ASIC), field-programmable gate arrays (FPGAs), and the like), which processors may be co-located in a single housing or in a single data center or may be geographically distributed (i.e., apparatus 1400 may be a distributed computing apparatus); a network interface 1468 comprising a transmitter (Tx) 1465 and a receiver (Rx) 1467 for enabling apparatus 1400 to transmit data to and receive data from other nodes connected to a network 110 (e.g., an Internet Protocol (IP) network) to which network interface 1448 is connected; communication circuitry 1448, which is coupled to an antenna arrangement 1449 comprising one or more antennas and which comprises a transmitter (Tx) 1445 and a receiver (Rx) 1447 for enabling the NN 504 to transmit data and receive data (e.g., wirelessly transmit / receive data); and a local storage unit (a.k.a., “data storage system”) 1408, which may include one or more non-volatile storage devices and / or one or more volatile storage devices. In embodiments where PC 1402 includes a programmable processor, a computer program product (CPP) 1441 may be provided. CPP 1441 includes a computer readable medium (CRM) 1442 storing a computer program (CP) 1443 comprising computer readable instructions (CRI) 1444. CRM 1442 may be a non-transitory computer readable medium, such as, magnetic media (e.g., a hard disk), optical media, memory devices (e.g., random access memory, flash memory), and thelike. In some embodiments, the CRI 1444 of computer program 1443 is configured such that when executed by PC 1402, the CRI causes the NN 504 to perform steps described herein (e.g., steps described herein with reference to the flow charts). In other embodiments, the NN 504 may be configured to perform steps described herein without the need for code. That is, for example, PC 1402 may consist merely of one or more ASICs. Hence, the features of the embodiments described herein may be implemented in hardware and / or softwar e.

Claims

CLAIMS 1. A method (1100) performed by a wireless communication device, WCD (502), the method comprising: receiving (s1102) wireless signals from a network node (504); based on the received wireless signals, estimating (s1104) a channel between the WCD and the network node; generating (s1106) first singular vector, SV, group information related to a first group of one or more SVs of the estimated channel; generating (s1108) second SV group information related to a second group of one or more SVs of the estimated channel; compressing (s1110) the first SV group information, thereby generating compressed first SV group information; compressing (s1112) the second SV group information, thereby generating compressed second SV group information; and transmitting (s1114) the compressed first and second SV group information to the network node.

2. The method of claim 1, wherein the first group of one or more SVs of the estimated channel is related to one or more spatial properties, amplitude information of one or more wireless signals received at the WCD, and / or phase information of one or more wireless signals received at the WCD, and the second group of one or more SVs of the estimated channel is related to one or more spatial properties, amplitude information of one or more wireless signals transmitted from the network node, and / or phase information of one or more wireless signals transmitted from the network node.

3. The method of claim 1 or 2, comprising: performing one or more of: (i) a singular value decomposition, SVD, of the estimated channel, (ii) an eigen value decomposition of the estimated channel, and / or (iii) providing measurement values of the estimated channel to a machine learning, ML, algorithm, thereby calculating the first group of one or more SVs of the estimated channel and the second group of one or more SVs of the estimated channel.

4. The method of claim 3, wherein the method comprises performing the SVD of the estimated channel, andthe SVD of the estimated channel corresponds to ^^^^, ^^^^^^, ^^^^^^^,^^^, where ^^^^, ^^is a first matrix including the first group of SVs at time t and frequency f, ^^^^, ^^is a group of SVs at time t andfrequency f, and ^^^^, ^^is a third matrix comprising a group of singular values of the estimated channel.

5. The method of at least one of claims 1-4, wherein each SV included in the first group of SVs is associated with a different frequency and / or a different timing.

6. The method of at least one of claims 1-5, wherein compressing the second SV group information related to the second group of SVs comprises determining a group of basis vectors that approximate the second group of SVs in a frequency domain and / or a time domain, a number of the basis vectors approximating the second group of SVs is less than a number of SVs included in the second group which was approximated by the group of basis vectors in the frequency domain and / or the time domain, and the compressed second SV group information indicates the basis vectors approximating the second group of SVs.

7. The method of at least one of claims 1-6, wherein compressing the first SV group information related to the first group of SVs comprises determining a group of basis vectors that approximate the first group of SVs in either a frequency domain and / or a time domain, a number of the basis vectors approximating the first group of SVs is less than a number of SVs included in the first group which was approximated by the group of basis vectors in the frequency domain and / or the time domain, and the compressed first SV group information indicates the basis vectors approximating the first group of SVs.

8. The method of at least one of claims 1-7, wherein each SV included in the first group of SVs is associated with a different frequency, compressing the first SV group information related to the first group of SVs comprises performing an averaging operation with respect to the SVs included in the first group in a frequency domain, thereby generating an averaged SV, and the compressed first SV group information indicates one or more characteristics of the averaged SV.

9. The method of claim 8, wherein the averaged SV comprises a plurality of complex numbers, andsaid one or more characteristics of the averaged SV comprises (i) both amplitudes and phases of the complex numbers or (ii) only the phases of the complex numbers.

10. The method of claim 7, comprising: applying a preprocessing to the first group of SVs and the second group of SVs, thereby generating a first group of preprocessed SVs and a second group of preprocessed SVs, wherein the group of basis vectors that approximate the first group of SVs approximate the first group of preprocessed SVs, and the group of basis vectors that approximate the second group of SVs approximate the second group of preprocessed SVs.

11. The method of claim 8 or 9, comprising: applying a preprocessing to the first group of SVs, thereby generating a first group of preprocessed SVs, wherein performing the averaging operation comprises averaging SVs included in the first group of preprocessed SVs, thereby generating the averaged SV.

12. The method of claim 11, wherein an SV included in the first group of SVs comprises a plurality of complex numbers, and applying the preprocessing to the first group of SVs comprises multiplying each of the plurality of complex numbers of the SV included in the first group of SVs by another complex number.

13. The method of at least one of claims 4-9, comprising: normalizing the singular values included in the third matrix, thereby generating a group of normalized singular values; and transmitting the normalized signal values to the network node.

14. A method (1200) performed by a network node (504), the method comprising: transmitting (s1202) wireless signals to a wireless communication device, WCD (502); and after transmitting the wireless signals, receiving (s1204) from the WCD compressed first and second singular vector, SV, group information, wherein the compressed first SV group information is related to a first group of one or more SVs of a channel between the WCD and the network node, andthe compressed second SV group information is related to a second group of one or more SVs of the channel.

15. The method of claim 14, comprising: predicting the quality of the channel between the WCD and the network node; determining a precoding matrix based on the predicted quality of the channel; and communicating with the WCD using a spatial multiplexing, wherein the spatial multiplexing is determined based on the precoding matrix.

16. The method of claim 14 or 15, wherein the first group of one or more SVs of the channel is related to one or more spatial properties, amplitude information of one or more wireless signals received at the WCD, and / or phase information of one or more wireless signals received at the WCD, and the second group of one or more SVs of the channel is related to one or more spatial properties, amplitude information of one or more wireless signals transmitted from the network node, and / or phase information of one or more wireless signals transmitted from the network node.

17. The method of at least one of claims 14-16, wherein the first group of one or more SVs of the channel and the second group of one or more SVs of the channel are calculated based on one or more of: (i) a singular value decomposition, SVD, of the channel, (ii) an eigen value decomposition of the channel, and / or (iii) providing measurement values of the channel to a machine learning, ML, algorithm.

18. The method of claim 17, wherein the first group of one or more SVs of the channel and the second group of one or more SVs of the channel are calculated based on the SVD of the channel, and the SVD of the channel corresponds to ^^^^, ^^^^^^, ^^^^^^^,^^^, where ^^^^, ^^is a first matrix including the first group of SVs at time t and frequency f, ^^^^, ^^is a second group of SVs at time t and frequency f,and ^^^^, ^^is a third matrix comprising a group of singular values of the estimated channel.

19. The method of at least one of claims 14-18, wherein each SV included in the first group of SVs is associated with a different frequency and / or a different timing.

20. The method of at least one of claims 14-19, whereinthe compressed second SV group information indicates a group of basis vectors approximating the second group of SVs in a frequency domain and / or a time domain, and a number of the basis vectors approximating the second group of SVs is less than a number of SVs included in the second group which was approximated by the group of basis vectors in the frequency domain and / or the time domain.

21. The method of at least one of claims 14-20, wherein the compressed first SV group information indicates a group of basis vectors approximating the first group of SVs in a frequency domain and / or a time domain, and a number of the basis vectors approximating the first group of SVs is less than a number of SVs included in the first group which was approximated by the group of basis vectors in the frequency domain and / or the time domain.

22. The method of at least one of claims 14-21, wherein each SV included in the first group of SVs is associated with a different frequency, and the compressed first SV group information indicates one or more characteristics of an averaged SV that is obtained by performing an averaging operation with respect to the SVs included in the first group in a frequency domain.

23. The method of claim 22, wherein the averaged SV comprises a plurality of complex numbers, and said one or more characteristics of the averaged SV comprises (i) both amplitudes and phases of the complex numbers or (ii) only the phases of the complex numbers.

24. The method of claim 23, wherein the group of basis vectors that approximate the first group of SVs approximate a first group of preprocessed SVs that are obtained by applying a preprocessing to the first group of SVs, and the group of basis vectors that approximate the second group of SVs approximate the second group of preprocessed SVs that are obtained by applying a preprocessing to the second group of SVs.

25. The method of claim 22 or 23, wherein performing the averaging operation comprises averaging SVs included in a first group of preprocessed SVs that are obtained by applying a preprocessing to the first group of SVs.

26. The method of claim 25, whereinan SV included in the first group of SVs comprises a plurality of complex numbers, and applying the preprocessing to the first group of SVs comprises multiplying each of the plurality of complex numbers of the SV included in the first group of SVs by another complex number.

27. The method of at least one of claims 18-23, comprising: receiving from the WCD normalized signal values that are generated by normalizing the singular values included in the third matrix.

28. A computer program (1343 or 1443) comprising instructions (1344 or 1444) which when executed by processing circuitry (1302 or 1402) cause the processing circuitry to perform the method of at least one of claims 1- 27.

29. A carrier containing the computer program of claim 28, wherein the carrier is one of an electronic signal, an optical signal, a radio signal, and a computer readable storage medium.

30. A wireless communication device, WCD (502), the WCD being configured to: receive (s1102) wireless signals from a network node (504); based on the received wireless signals, estimate (s1104) a channel between the WCD and the network node; generate (s1106) first singular vector, SV, group information related to a first group of one or more SVs of the estimated channel; generate (s1108) second SV group information related to a second group of one or more SVs of the estimated channel; compress (s1110) the first SV group information, thereby generating compressed first SV group information; compress (s1112) the second SV group information, thereby generating compressed second SV group information; and transmit (s1114) the compressed first and second SV group information to the network node.

31. The WCD of claim 30, wherein the WCD is further configured to perform the method of at least one of claims 2-13.

32. A network node (504), the network node being configured to: transmit (s1202) wireless signals to a wireless communication device, WCD (502); andafter transmitting the wireless signals, receive (s1204) from the WCD compressed first and second singular vector, SV, group information, wherein the compressed first SV group information is related to a first group of one or more SVs of a channel between the WCD and the network node, and the compressed second SV group information is related to a second group of one or more SVs of the channel.

33. The network node of claim 32, wherein the network node is further configured to perform the method of at least one of claims 15-27.

34. An apparatus (1300 or 1400) comprising: a processing circuitry (1302 or 1402); and a memory (1341 or 1441), said memory containing instructions executable by said processing circuitry, whereby the apparatus is operative to perform the method of at least one of claims 1-27.

Images