GB2629863A

GB2629863A - Methods and apparatuses relating to transmission of channel state information

Info

Publication number: GB2629863A
Application number: GB2307114.5A
Authority: GB
Inventors: Hyun Kim Jee
Original assignee: Nokia Technologies Oy
Current assignee: Nokia Technologies Oy
Priority date: 2023-05-12
Filing date: 2023-05-12
Publication date: 2024-11-13
Also published as: WO2024235531A1; CN121079910A; GB202307114D0

Abstract

A terminal device 100 obtains 202 a first channel state information (CSI) sample in a CSI space based on first measurements of first reference signals received 201 from a network node 150 and transmits 203, to the network node150, first CSI based on the first CSI sample. The terminal device 100 obtains 205 a second CSI sample in the CSI space based on second measurements of second reference signals received 204 from the network node 150, the second measurements being performed after the first measurements, and transmits 206, to the network node 150, second CSI based on the second CSI sample. The second CSI is indicative of a codeword within a codebook subset and the codebook subset is determined based on the first CSI and comprises a plurality of neighbouring codewords of a codebook that are determined according to a distance metric defined over the CSI space. The first and second CSI transmitted by the terminal device 100 may be used by the network to perform 207 downlink (DL) communications, e.g. by combining the coarse and refined CSI feedback to perform high-resolution beamforming on the physical downlink shared channel (PDSCH).

Description

Methods and Apparatuses relating to Transmission of Channel State Information

Field

This specification relates generally to transmission of channel state information.

Background

Tn modern telecommunications networks, terminal devices (or user equipment, UE) are able to transmit channel state information (CSI) to network nodes. CSI indicates how signals propagate via a communications channel from the network to the terminal device and may be used by the network to support downlink (DL) transmissions on the physical downlink shared channel (PDSCH) and physical downlink control channel (PDCCH).

Summary

Tn a first aspect, this specification relates to a terminal device comprising: means for obtaining a first channel state information sample in a channel state information space based on first measurements of first reference signals received from a network node; means for transmitting, to the network node, first channel state information based on the first channel state information sample; means for obtaining a second channel state information sample in the channel state information space based on second measurements of second reference signals received from the network node, the second measurements being performed after the first measurements; and means for transmitting, to the network node, second channel state information based on the second channel state information sample, wherein the second channel state information is indicative of a codeword within a codebook subset, and wherein the codebook subset is determined based on the first channel state information and comprises a plurality of neighbouring codewords of a codebook that are determined according to a distance metric defined over the channel state information space.

In some examples, the codebook comprises a plurality of codewords organised into predefined codebook subsets of neighbouring codewords according to the distance metric, the first channel state information comprises a first identifier that identifies the codebook subset, and the second channel state information comprises a second identifier that identifies the codeword within the codebook subset. In some such examples, the codebook subset is the codebook subset of the pre-defined codebook subsets with which the first channel state information sample is associated, and the codeword is the codeword from the codebook subset that is closest to the second channel state information sample according to the distance metric. In addition or alternatively, the first measurements are wideband or sub-band measurements, and the second measurements are sub-band measurements.

In other examples, the first channel state information comprises a first identifier that identifies a first reference codeword within the codebook that is closest to the first channel state information sample according to the distance metric, the codebook subset comprises the first reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the first reference codeword according to the distance metric, and the second channel state information comprises a second identifier that identifies a second reference codeword within the codebook subset that is closest to the second channel state information sample according to the distance metric. In some such examples, the terminal device further comprises means for obtaining a third channel state information sample in the channel state information space based on third measurements of third reference signals received from the network node, the third measurements being performed after the second measurements; and means for transmitting, to the network node, third channel state information based on the third channel state information sample, wherein the third channel state information comprises a third identifier that identifies a third reference codeword within a second codebook subset comprising the second reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the second reference codeword according to the distance metric. In addition or alternatively, the first and second measurements are sub-band measurements.

In some examples, the channel state information space is a vector space spanned by sub-vector components of a latent vector, and the latent vector is obtained after AI/ML so encoding of a channel eigenvector or a channel singular vector or a channel matrix that is determined based on the first or second measurements.

In a second aspect, this specification relates to a network node comprising means for transmitting first reference signals for reception at a terminal device; means for receiving, from the terminal device, first channel state information, wherein the first channel state information is based on a first channel state information sample in a -3 -channel state information space obtained by the terminal device based on first measurements of the first reference signals; means for transmitting second reference signals for reception at the terminal device; and means for receiving, from the terminal device, second channel state information, wherein the second channel state information is based on a second channel state information sample in the channel state information space obtained by the terminal device based on second measurements of the second reference signals, the second measurements being performed after the first measurements, wherein the second channel state information is indicative of a codeword within a codebook subset, and wherein the codebook subset is determined based on the first channel state information and comprises a plurality of neighbouring codewords of a codebook that are determined according to a distance metric defined over the channel state information space.

In some examples, the codebook comprises a plurality of codewords organised into pre- /5 defined codebook subsets of neighbouring codewords according to the distance metric, the first channel state information comprises a first identifier that identifies the codebook subset, and the second channel state information comprises a second identifier that identifies the codeword within the codebook subset. In some such examples, the codebook subset is the codebook subset of the pre-defined codebook subsets with which the first channel state information sample is associated, and the codeword is the codeword from the codebook subset that is closest to the second channel state information sample according to the distance metric. In addition or alternatively, the first measurements are wideband or sub-band measurements, and the second measurements are sub-band measurements.

In other examples, the first channel state information comprises a first identifier that identifies a first reference codeword within the codebook that is closest to the first channel state information sample according to the distance metric, the codebook subset comprises the first reference codeword and a pre-defined number of neighbouring 3o codewords in the vicinity of the first reference codeword according to the distance metric, and the second channel state information comprises a second identifier that identifies a second reference codeword within the codebook subset that is closest to the second channel state information sample according to the distance metric. In some such examples, the network node further comprises means for transmitting third reference signals for reception at the terminal device; and means for receiving, from the terminal device, third channel state information, wherein the third channel state -4 -information is based on a third channel state information sample in the channel state information space obtained by the terminal device based on third measurements of the third reference signals, the third measurements being performed after the second measurements, wherein the third channel state information comprises a third identifier that identifies a third reference codeword within a second codebook subset comprising the second reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the second reference codeword according to the distance metric. In addition or alternatively, the first and second measurements are sub-band measurements.

Tn some examples, the channel state information space is a vector space spanned by sub-vector components of a latent vector, and the latent vector is obtained after AI/ML encoding of a channel eigenvector or a channel singular vector or a channel matrix that is determined based on the first or second measurements.

Tn a third aspect, this specification relates to a method comprising: obtaining a first channel state information sample in a channel state information space based on first measurements of first reference signals received from a network node; transmitting, to the network node, first channel state information based on the first channel state information sample; obtaining a second channel state information sample in the channel state information space based on second measurements of second reference signals received from the network node, the second measurements being performed after the first measurements; and transmitting, to the network node, second channel state information based on the second channel state information sample, wherein the second channel state information is indicative of a codeword within a codebook subset, and wherein the codebook subset is determined based on the first channel state information and comprises a plurality of neighbouring codewords of a codebook that are determined according to a distance metric defined over the channel state information space.

Tn some examples, the codebook comprises a plurality of codewords organised into predefined codebook subsets of neighbouring codewords according to the distance metric, the first channel state information comprises a first identifier that identifies the codebook subset, and the second channel state information comprises a second identifier that identifies the codeword within the codebook subset. In some such examples, the codebook subset is the codebook subset of the pre-defined codebook -5 -subsets with which the first channel state information sample is associated, and the codeword is the codeword from the codebook subset that is closest to the second channel state information sample according to the distance metric. Tn addition or alternatively, the first measurements are wideband or sub-band measurements, and the second measurements are sub-band measurements.

In other examples, the first channel state information comprises a first identifier that identifies a first reference codeword within the codebook that is closest to the first channel state information sample according to the distance metric, the codebook subset comprises the first reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the first reference codeword according to the distance metric, and the second channel state information comprises a second identifier that identifies a second reference codeword within the codebook subset that is closest to the second channel state information sample according to the distance metric. In some such examples, the method further comprises obtaining a third channel state information sample in the channel state information space based on third measurements of third reference signals received from the network node, the third measurements being performed after the second measurements; and transmitting, to the network node, third channel state information based on the third channel state information sample, wherein the third channel state information comprises a third identifier that identifies a third reference codeword within a second codebook subset comprising the second reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the second reference codeword according to the distance metric. In addition or alternatively, the first and second measurements are sub-band measurements.

In some examples, the channel state information space is a vector space spanned by sub-vector components of a latent vector, and the latent vector is obtained after AI/ML encoding of a channel eigenvector or a channel singular vector or a channel matrix that is determined based on the first or second measurements.

Tn a fourth aspect, this specification relates to a method comprising transmitting first reference signals for reception at a terminal device; receiving, from the terminal device, first channel state information, wherein the first channel state information is based on a first channel state information sample in a channel state information space obtained by the terminal device based on first measurements of the first reference signals; -6 -transmitting second reference signals for reception at the terminal device; and receiving, from the terminal device, second channel state information, wherein the second channel state information is based on a second channel state information sample in the channel state information space obtained by the terminal device based on second measurements of the second reference signals, the second measurements being performed after the first measurements, wherein the second channel state information is indicative of a codeword within a codebook subset, and wherein the codebook subset is determined based on the first channel state information and comprises a plurality of neighbouring codewords of a codebook that are determined according to a distance metric defined over the channel state information space.

In other examples, the first channel state information comprises a first identifier that identifies a first reference codeword within the codebook that is closest to the first channel state information sample according to the distance metric, the codebook subset comprises the first reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the first reference codeword according to the distance metric, and the second channel state information comprises a second identifier that 3o identifies a second reference codeword within the codebook subset that is closest to the second channel state information sample according to the distance metric. In some such examples, the method further comprises transmitting third reference signals for reception at the terminal device; and receiving, from the terminal device, third channel state information, wherein the third channel state information is based on a third channel state information sample in the channel state information space obtained by the terminal device based on third measurements of the third reference signals, the -7 -third measurements being performed after the second measurements, wherein the third channel state information comprises a third identifier that identifies a third reference codeword within a second codebook subset comprising the second reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the second reference codeword according to the distance metric. In addition or alternatively, the first and second measurements are sub-band measurements.

In a fifth aspect, this specification relates to an apparatus (e.g. a terminal device or a component of a terminal device) comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: obtaining a first channel state information sample in a channel state information space based on first measurements of first reference signals received from a network node; transmitting, to the network node, first channel state information based on the first channel state information sample; obtaining a second channel state information sample in the channel state information space based on second measurements of second reference signals received from the network node, the second measurements being performed after the first measurements; and transmitting, to the network node, second channel state information based on the second channel state information sample, wherein the second channel state information is indicative of a codeword within a codebook subset, and wherein the codebook subset is determined based on the first channel state information and comprises a plurality of neighbouring codewords of a codebook that are determined according to a distance metric defined over the channel state information space.

so In some examples, the codebook comprises a plurality of codewords organised into pre-defined codebook subsets of neighbouring codewords according to the distance metric, the first channel state information comprises a first identifier that identifies the codebook subset, and the second channel state information comprises a second identifier that identifies the codeword within the codebook subset. In some such examples, the codebook subset is the codebook subset of the pre-defined codebook subsets with which the first channel state information sample is associated, and -8 -the codeword is the codeword from the codebook subset that is closest to the second channel state information sample according to the distance metric. In addition or alternatively, the first measurements are wideband or sub-band measurements, and the second measurements are sub-band measurements.

In other examples, the first channel state information comprises a first identifier that identifies a first reference codeword within the codebook that is closest to the first channel state information sample according to the distance metric, the codebook subset comprises the first reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the first reference codeword according to the distance metric, and the second channel state information comprises a second identifier that identifies a second reference codeword within the codebook subset that is closest to the second channel state information sample according to the distance metric. In some such examples, the instructions, when executed by the at least one processor, may cause the apparatus at least to perform obtaining a third channel state information sample in the channel state information space based on third measurements of third reference signals received from the network node, the third measurements being performed after the second measurements; and transmitting, to the network node, third channel state information based on the third channel state information sample, wherein the third channel state information comprises a third identifier that identifies a third reference codeword within a second codebook subset comprising the second reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the second reference codeword according to the distance metric. In addition or alternatively, the first and second measurements are sub-band measurements.

In some examples, the channel state information space is a vector space spanned by sub-vector components of a latent vector, and the latent vector is obtained after AI/ML encoding of a channel eigenvector or a channel singular vector or a channel matrix that is determined based on the first or second measurements. go

In a sixth aspect, this specification relates to an apparatus (e.g. a network node or a component of a network node) comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: transmitting first reference signals for reception at a terminal device; receiving, from the terminal device, first channel state information, wherein the first channel state information is based on a first channel state information -9 -sample in a channel state information space obtained by the terminal device based on first measurements of the first reference signals; transmitting second reference signals for reception at the terminal device; and receiving, from the terminal device, second channel state information, wherein the second channel state information is based on a second channel state information sample in the channel state information space obtained by the terminal device based on second measurements of the second reference signals, the second measurements being performed after the first measurements, wherein the second channel state information is indicative of a codeword within a codebook subset, and wherein the codebook subset is determined based on the first channel state information and comprises a plurality of neighbouring codewords of a codebook that are determined according to a distance metric defined over the channel state information space.

In other examples, the first channel state information comprises a first identifier that identifies a first reference codeword within the codebook that is closest to the first channel state information sample according to the distance metric, the codebook subset comprises the first reference codeword and a pre-defined number of neighbouring 3o codewords in the vicinity of the first reference codeword according to the distance metric, and the second channel state information comprises a second identifier that identifies a second reference codeword within the codebook subset that is closest to the second channel state information sample according to the distance metric. In some such examples, the instructions, when executed by the at least one processor, may cause the apparatus at least to perform transmitting third reference signals for reception at the terminal device; and receiving, from the terminal device, third channel state information, wherein the third channel state information is based on a third channel state information sample in the channel state information space obtained by the terminal device based on third measurements of the third reference signals, the third measurements being performed after the second measurements, wherein the third channel state information comprises a third identifier that identifies a third reference codeword within a second codebook subset comprising the second reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the second reference codeword according to the distance metric. In addition or alternatively, the first and second measurements are sub-band measurements.

Tn a seventh aspect. this specification describes a non-transitory computer readable medium comprising program instructions stored thereon for performing at least any of the operations described above with reference to the first to sixth aspects.

Brief Description of the Figs.

For better understanding of the present application, reference will now be made by way of example to the accompanying drawings in which: Fig. 1 is an example illustrating communications between a terminal device and a network node; Figs. 2 and 3 are example message flow sequences; Figs. 4, 5 and 6 are schematic illustrations of codebooks in accordance with examples described herein; Figs. 7 and 8 are flowcharts illustrating various operations which may be performed in accordance with examples described herein; so Fig. 9 is a schematic illustration of an example configuration of a terminal device which may be configured to perform various operations described with reference to Figs. 1 to 8; Fig. to is a schematic illustration of an example configuration of a network node which may be configured to perform various operations described with reference to Figs. 1 to 35 8; and Fig. 11 is an illustration of a computer-readable medium upon which computer readable code may be stored.

Detailed Description

In the description and drawings below, like reference numerals refer to like elements throughout.

Fig. 1 depicts a terminal device 100 in communication with a network node 150. In particular, Fig. 1 depicts CSI feedback scheme which uses an AIML (Artificial Intelligence and Machine Learning) encoder to compress CST prior to transmission.

In block 101, terminal device 100 performs downlink channel measurement based on reference signals (e.g. CSI-RS) received from network node 150. Put another way, terminal device 100 may obtain CSI based on measurements of reference signals transmitted by network node 150 and received at terminal device too. Such measurement may be referred to as channel estimation'. In some examples, the obtained CST may comprise a channel matrix or precoding matrix indicator (PMT).

In block 102, terminal device 100 may then perform pre-processing on the obtained CSI. For instance, such pre-processing may include singular value decomposition (SVD) or discrete Fourier transform (DFI'), though other forms of pre-processing will be apparent to the skilled person. In some examples, the pre-processing of the CSI may involve determination of channel eigenvectors or singular vectors based on the obtained CSI.

In block 103, the (optionally pre-processed) CSI is encoded by the AT encoder (e.g. a trained artificial neural network, ANN). In some examples, the AI encoder may be an autoencoder. The output of the AT encoder may be referred to as a latent vector', 'latent feature vector', embedding vector' or embedding feature vector'. For instance, the latent vector may be denoted.ze ERLz.i (i.e. a Lz-dimensional real-valued vector). In examples in which the CSI comprises PMI, the latent vector may correspond to Al-compressed PMI.

In block 104, the latent vector is quantised to reduce the required signalling/feedback 35 overhead when reporting the CSI to the to the network. Put another way, prior to overthe-air transmission of the encoded CSI to network 150 (indicated by the dashed arrow in Fig. 1), terminal device 100 compresses the encoded CSI by means of a quantisation scheme.

After reception at network node 150, a reverse sequence of operations is performed at the network side in order to reconstruct the CSI. In particular, dequantization is performed at block 151, AI decoding is performed at block 152, and post-processing is optionally performed at block 153. This sequence of operations results in reconstructed CSI at block 154. Such reconstructed CSI may be referred to as 'Output CSI'. Whilst the reconstructed CSI may not exactly match the CSI as measured at the terminal device, a carefully chosen quantisation/dequantization scheme and/or codebook can ensure that the reconstruction matches sufficiently closely for network tasks (e.g. controlling and optimizing DL communications).

It will be appreciated that there are two main classes of quantisation scheme that may be used to compress the latent vector, z: scalar quantisation (SQ) and vector quantisation (VQ). SQ schemes quantise each element of the latent vector in turn (i.e. one-by-one), whilst VQ schemes quantise multiple elements of the latent vector together by use of a codebook which is known to both the UE side and network side.

In some examples, the term codebook' may refer to a plurality of codewords in a given space. Such a codebook may be used to compress a set of vectors in the given space by assigning each of the vectors to the codeword from the codebook which is closest under some distance metric. Put another way, the codewords in the codebook may divide the given space into a plurality of Voronoi regions under some metric. In this case, each vector may be assigned a codeword according to which Voronoi region (or Voronoi cell') it lies within. It will be appreciated that the distance metric d(a,b) may take a variety of forms. For instance, the distance metric may be the Euclidean metric, a Manhattan metric or angular separation metric. Other appropriate metrics will be apparent to the skilled person.

Various codebook designs are described below with respect to Figs. 2 to 8.

Since the dimension, Lz, of the latent vector may be very large (e.g. 64, or even larger than 100, depending on the input size and compression ratio, CR), segmentation of the 35 latent vector may be required to render the VQ task manageable at the UE side, where computational resources may be constrained compared to the network side. For -13 -instance, the latent vector may be quantised in segments, or sub-vectors, of predefined length Ls (e.g. 2, 3, 4). Each of these sub-vectors may then be quantised using a single codebook of dimension L. The space spanned by the sub-vectors of the latent vector may be referred to as a 'CST space'. Tt will be appreciated that the codewords used to quantise the sub-vectors are included in this CSI space.

Since the codebook is known at both the UE side and network side, CSI reporting may then be accomplished by transmission of a set of codeword indices which may be used by the network to recover the codewords corresponding to the sub-vectors. For instance, a 64-dimensional latent vector can be segmented into 16 instances of 4-dimensional sub-vectors, which may each be quantised using a single 4-dimensional codebook. In this example, the network may transmit the quantised latent vector as 16 lots of ilop131 bits, where B is the cardinality (i.e. number of codewords) of the codebook.

However, searching for the closest codewords to a given CST sample can be computationally intensive on the UE side, particularly for codebooks of high cardinality.

For instance, if we assume that 2 bits are allocated for each element of a latent vector, a 4-dimensional sub-vector can be represented by an 8 bit codeword index. This implies that a codebook of cardinality 2'8=256 bits is required.

Whilst the AI encoder can accommodate a relatively large bandwidth with sub-band granularity channel information as its input (e.g. 20MHz bandwidth), the required number of inferences performed by the encoder can increase over the number of layers, of which there may be up to four. For input with 13 sub-bands having subcarrier spacing (SCS) of 3okliz, the overall system bandwidth can be woMHz (i.e. 5 times greater than the 2oMHz bandwidth of the encoder). This means that the UE may perform up to 4 (layers) 5 (looMHz/26MHz) = 20 instances of inference using the AI encoder and up to 20 corresponding instances of vector quantization, in addition to the channel estimation and pre-processing (e.g. SVD).

In the case where the compression ratio (CR) = 13/1, the output latent vector of the AI 35 encoder has dimension 64. In such an example, if the terminal device were to perform an exhaustive codebook search over all 256 codewords in the codebook for every -14 -inference for each sub-vector, a large number of search operations would be required. For instance, the number of search operations required would be 16 (sub-vectors per latent vector) * 4 (layers per channel) * 5 (looMHz/2oMHz) = 32o. Moreover, processing the latent vector in this way would require a high CSI feedback signalling overhead. For the given example, and assuming that there are 32 transmit antenna (TxAnt) ports and 4 receive antenna (Thant) ports, with sub-band wise granularity, the CSI feedback overhead is 64 (elements per latent vector) 2 (bits per element) *4 (layers) 5 (looMHz/2oMHz) = 256o bits.

It may therefore be beneficial to reduce computational complexity on the UE-side when performing VQ of channel state information, as well as over-the-air CST feedback overhead. Various implementations of the technology described provide such reductions in both UE computational complexity and CSI feedback overhead.

In some examples, the term 'terminal device' or 'user equipment' may refer to any device employed by a user to communicate. Whilst the terminal devices of Figs. to 3 are depicted as mobile telephones, it will of course be appreciated that terminal devices may comprise various other devices, including, but not limited to laptops, smartwatches, tablet computers and vehicle-based terminal devices, such as those mounted on cars, buses, uncrewed aerial vehicles (UAVs), aeroplanes, trains, or boats.

Alternatively, mobile terminal devices may be carried by a user, or worn on their person.

Various methods and apparatuses are described in detail below, by way of example only, in the context of a cellular network, such as an Evolved Universal Terrestrial Radio Access (E-UTRA) network or a 5G network. However, it will be appreciated that the techniques may be applicable with communications networks of other types (e.g. but not limited to other types of cellular network). Cellular networks may comprise one or more base stations, sometimes referred to as transmit-receive points (TRPs) or so access points (e.g. but not limited to gNBs and/or eNBs). Whilst only one base station is depicted in each of Figs. lto 3, a radio access network (RAN, NG-RAN) may typically comprise thousands of such base stations. Together, the base stations may provide cellular network coverage to one or more terminal devices over a wide geographical area.

-15 -Although by no means limited to such an implementation, the examples of the technology described herein may readily be integrated into any New Radio (NR) terminal device which performs CST reporting (e.g. as standardised in 3GPP TS 38.214). Various implementations of the technology described herein may be applicable to the AIML-enabled CSI feedback enhancements that are expected to be included in 3GPP release 18 or in a subsequent release.

In some implementations and, for instance, depending on the characteristics of the cellular network, the network nodes and terminal devices within the network may be configured to communicate with one another, for instance, using an OFDM-based communication scheme, such as orthogonal frequency-division multiple access (OFDMA), single carrier frequency-division multiple access (SC-FDMA), and/or cyclic prefix orthogonal frequency-division multiple access (CP-OFDMA). For instance, in some non-limiting examples, OFDMA may be used for downlink (DL) communications and SC-FDMA may be used for uplink (UL) communications.

As mentioned above, it be beneficial to reduce computational complexity on the UEside when performing VQ of CSI, as well as over-the-air CSI feedback overhead.

In general terms, various aspects of the technology described herein relate to a two-step CSI reporting procedure. In the first step, network node 150 transmits first reference signals for reception at the terminal device. Based on measurements of the first signals, terminal device goo obtains a first CSI sample in a channel state information space and transmits first CSI to network node 150 based on the first CSI sample. In the second step, network node 150 transmits second reference signals for reception at terminal device goo using the first CSI (e.g. for beamforming/precoding). Based on measurements of the second signals, terminal device goo obtains a second CSI sample in the channel state information space and transmits second CSI to network node 150 based on the second CST sample.

In particular, implementations of the technology described herein relate to a terminal device obtaining a first CST sample in a CST space based on first measurements of first reference signals received from a network node; transmitting, to the network node, first CSI based on the first CSI sample; obtaining a second CSI sample in the CSI space based on second measurements of second reference signals received from the network node, the second measurements being performed after the first measurements; and -16 -transmitting, to the network node, second CSI based on the second CSI sample, wherein the second CSI is indicative of a codeword within a codebook subset, and wherein the codebook subset is determined based on the first CST and comprises a plurality of neighbouring codewords of a codebook that are determined according to a distance metric defined over the CSI space. Other aspects of the technology described herein relate to a network node transmitting first reference signals for reception at a terminal device; receiving, from the terminal device, first CSI, wherein the first CSI is based on a first CSI sample in a CSI space obtained by the terminal device based on first measurements of the first reference signals; transmitting second reference signals for reception at the terminal device; and receiving, from the terminal device, second CST, wherein the second CSI is based on a second CSI sample in a CST space obtained by the terminal device based on second measurements of the second reference signals, the second measurements being performed after the first measurements, wherein the second CSI is indicative of a codeword within a codebook subset, and wherein the codebook subset is determined based on the first CSI and comprises a plurality of neighbouring codewords of a codebook that are determined according to a distance metric defined over the CST space.

As mentioned above, the CSI may take a variety of forms. For instance, the CSI may comprise a precoding matrix. Using the convention that #(*) = "number el..", such a matrix may be dimensioned as #(IXAnt ports) #(sub-bands) [optionally, #(ranks/layers)], and may be represented by a set of channel eigenvectors/singular vectors (e.g. after pre-processing using SVD). Alternatively, the CSI may comprise a channel matrix (e.g. a full channel matrix). In such examples, the channel matrix may be dimensioned as #(TxAnt ports) * #(RxAnt ports) #(sub-bands). Other formats for the CSI will be apparent to the skilled person.

In some examples, the codebook may comprise a plurality of codewords organised into pre-defined codebook subsets of neighbouring codewords according to the distance metric. In this case, the first CST comprises a first identifier that identifies the codebook subset and the second CSI comprises a second identifier that identifies the codeword within the codebook subset. In other words, the first identifier may enable unambiguous identification of the codebook subset from amongst the pre-defined codebook subsets, whilst the second identifier may enable unambiguous identification of the codeword (i.e. a specific codeword) within (or relatively to, or with respect to) the codebook subset. In such examples, the codebook may be referred to as having a -17 -combined codebook (cCB)' design and may be said to make use of a 'hierarchical' CSI reporting scheme.

The codebook subset may be the codebook subset of the pre-defined codebook subsets with which the first CSI sample is associated, and the codeword may be the codeword from the codebook subset that is closest to the second CSI sample according to the distance metric. The first measurements may be considered to be 'coarse measurements' which enable determination of the codebook subset, which represents a coarse' region of the CSI space which the first CSI sample lies within. The second measurements may be considered to be 'fine measurements' which enable determination of a codeword within the codeword subset which is closest to the second CSI sample.

It will be appreciated that a sample may be associated with a given subset of the codebook if the distance, under the distance metric, from the sample to the centroid of the codebook subset is lower than the distance, under the distance metric, from the sample to the centroid of any other codebook subset. Whilst it occupies the same CSI space as the codewords, it will be appreciated that the centroid of a codeword subset is not necessarily a codeword itself. In other examples, a separate codebook may be used to define the coarse regions which correspond to the codebook subsets. In such examples, whether a sample is associated with a given coarse region may be determined based on its distance to the codewords in the separate codebook. Codewords in the separate codebook may be chosen to lie close (i.e. within a set distance under the metric) to the centroids of the subsets of the original codebook.

Put another way, and with reference to the two-step CSI reporting procedure mentioned above, in the first step terminal device too identifies and reports to the network the codebook subset from the pre-defined subsets of the codebook (or coarse regions of the CSI space) with which a first CSI sample is associated. In the second step, 3o terminal device too identifies and reports the codeword within the subset identified in the first step which is closest to a second CSI sample according to the distance metric.

In such examples, the first step may be referred to as the 'Coarse information delivery phase' whilst the second step may be referred to as the 'Refined information delivery 35 phase'.

-18 -Further aspects of examples employing a combined codebook design are described below with reference to Fig. 2, whilst example representations of combined codebooks are depicted in Figs. 4 and 5.

In other examples, the first CSI may comprise an identifier of a first reference codeword within the codebook. The first reference codeword is the codeword within the codebook that is closest to the first CSI sample according to the distance metric. In this case, the codebook subset comprises the first reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the first reference codeword according to the distance metric, and the second CSI comprises an identifier of a second reference codeword within the codebook subset that is closest to the second CSI sample according to the distance metric. In such examples, the codebook may be referred to as having a tracking codebook (tCB)' design and may be said to make use of a subspace-tracking CSI reporting scheme. In some examples, the tracking codebook may also be referred to as a 'recursive codebook'.

Terminal device 100 may obtain a third CSI sample in the CSI space based on third measurements of third reference signals received from network node 150, the third measurements being performed after the second measurements, and may further transmit, to network node 150, third CSI based on the third CSI sample. In this case, the third CSI comprises an identifier of a third reference codeword within a second codebook subset comprising the second reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the second reference codeword according to the distance metric.

Put another way, and with reference to the two-step CSI reporting procedure mentioned above, in the first step terminal device 100 identifies and reports to the network the codeword in the codebook (i.e. the overall cod ebook, or superset) which is closest to a first CST sample according to the distance metric. Based on this codeword, 3o which may be referred to as a 'first reference codeword', a subset comprising the first reference codeword together with a pre-defined number of neighbouring codewords may be identified. Because the number of neighbouring codewords is pre-defined (i.e. known to both the terminal device and the network), the subset may be identified independently by both terminal device 100 and the network.

-19 -In the second step, terminal device loo identifies and reports the codeword within the subset identified in the first step which is closest to a second CSI sample according to the distance metric. This codeword may be referred to as a 'second reference codeword'. Analogously to in the first step, the second reference codeword may be used to identify a new subset comprising the second reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the second reference codeword. The closest codeword in this new subset to a subsequent (i.e. third) CSI sample is then reported to the network. In this way, the tracking codebook subset is continuously updated so as to be centred on current/recent CSI samples.

In such examples, the first step may be referred to as the 'Pinpointing phase' whilst the second step may be referred to as the 'Tracking phase'.

Further aspects of examples employing a tracking codebook design are described below with reference to Fig. 3, whilst example representations of combined codebooks are depicted in Figs. 4 and 6.

For both tCB and cCB cases, the majority of over-the air signalling cases in CSI reporting are covered by 'step 2' reporting, as it relates to 'short term' CST reports. Such short-term reports require frequent updates so as to keep track of temporal evolution in the channel. As explained above, signalling overhead (e.g. at step 2) can be reduced by various implementations of the technology described herein. In particular, such reductions may be achieved in the presence of channel temporal correlation. Moreover, and as mentioned above, computational complexity at TIE side for codeword search can be greatly reduced at the same time.

As standardised in 3GPP TS 38.214, an existing 'legacy' hybrid CSI feedback reporting scheme using Type II port selection codebooks also make use of a two-step process. In the first step of such a process, a long-term precoding matrix (WE) is determined based on wideband measurements at the TIE; in the second step, a short-term precoding matrix (W2) is determined based on sub-band granularity measurements at the TIE. The overall precoding matrix may then be determined by combining the long-and short-term precoding matrices (i.e. W=W, Implementations of the technology described herein may be beneficially leveraged to provide a CSI reporting framework for AIML-enabled CSI use cases that makes use of a two-step procedure as in the existing hybrid CST reporting framework.

As mentioned above, the CSI may take various forms. For instance, the (pre-processed) input to the Al encoder on the UE side, sometimes referred to as 'input-CSI-NW' or target CSI', may be a precoding matrix in the spatial-frequency domain (i.e. a set of channel eigenvectors). Likewise, the (post-processed) output of the AI decoder on the NW side, sometimes referred to as 'output-CSI-UE' or 'output/reconstructed CSI', may to take the same form. In the event that the CSI feedback mechanism were lossless (e.g. if all sub-vectors of the latent vector were exactly equal to codewords), the output CST would equal the target CSI. However, as mentioned above, this is not the case in general so these CSI differ in practice.

/5 Whilst the target CST for an ATML-enabled CST compression scheme as described herein is not necessarily of the form W=W, IN, as used in the legacy scheme (e.g. the AT encoder may perform direct compression of the channel eigenvectors), the concept of a two-step procedure is maintained.

However, since the main target use case for the AIML-enabled CSI compression scheme is MU-MIMO, for which high resolution CSI is required, the feedback overhead can be high. As mentioned above, implementations of the technology described herein may reduce this overhead.

The latent vector output by the M encoder, and hence its segmentized sub-vectors, should exhibit temporal correlation when taken at adjacent time points, since its input (CSI such as channel eigenvectors or a full channel matrix) is temporally correlated. Various implementations of the technology described herein may exploit this temporal correlation, thereby to reduce feedback overhead in CST reporting.

A comparison of a 'legacy' two-step procedure employing Type TT port selection codebook with non-limiting example procedures employing combined and tracking codebooks is set out in Table 1.

Examples of the present disclosure

-21 -Hybrid CSI feedback reporting steps Type II Combined codebook Tracking codebook port selection codebook Step CSI-RS Non-beamformed Non-beamformed Beamformed UE Wideband Wideband or Sub-band Sub-band measurement base Major UE task Find and report long term CSI (i1) Find and report best codebook subset (coarse CB index) Find and report best codeword within a superset (adjusting/pinpoin ting codeword index) Step CSI-RS Beamformed Beamformed Beamformed UE Sub-band Sub-band Sub-band measurement base Major UE task Find and report short term CSI (i2) Find and report codeword within the chosen coarse CB (fine codeword index) Find and report codeword within the chosen tracking CB (fine codeword index)

Table

Fig. 2 is a message flow sequence, indicated generally by the reference numeral 2, in accordance with some aspects of the described technology. The message flow sequence 2 shows an example implementation within which aspects of a process, such as that described with reference to Fig. 1, may be performed (e.g. a cCB-based CSI feedback scheme). In this non-limiting example, terminal device 100 is depicted as a mobile UE in communication with network node 150, which is depicted as a gNB.

to In terms of the two-step CST reporting framework described above, the first step may correspond to operations 201 to 203 and the second step may correspond to operations 204 to 206. In general terms, the first step relates to acquiring 'coarse' channel state information using codebook 25o and the second step relates to acquiring 'fine' channel state information using codebook 26o.

In operation 201, network node 15o transmits first reference signals for reception at 5 terminal device Too. In some examples, the transmitted first reference signals may be non-precoded (i.e. non-beamformed) CSI-RS.

In operation zoz, terminal device Too obtains a first CSI sample in a CSI space based on first measurements of the first reference signals transmitted in operation 201. In some examples, the first measurements are wideband measurements. In other examples, the first measurements are sub-band measurements.

Put another way, upon reception of the first reference signals, terminal device 100 performs channel estimation using either wideband measurements or sub-band measurements of the received signals, thereby to obtain a first CST sample. As described above with reference to Fig. 1, terminal device Too may then determine which of a plurality of coarse regions in the CST space the first CSI sample is associated with or lies within. For instance, as depicted in Fig. 2, the first CSI sample may be compared to a plurality of coarse codewords stored in a coarse codebook 25o. Each of the Voronoi regions defined by coarse codebook 250 may correspond to a subset of the 'fine' regions defined by codebook 26o. For instance, each region defined by codebook 25o may correspond to four (or more, or fewer) regions defined by codebook 26o.

Alternatively, rather than including true codewords which define coarse Voronoi regions, codebook 250 may instead include a plurality of centroids of coarse regions in the CSI space corresponding to the subsets. As described with reference to Fig. 1, such centroids do not necessarily have to correspond to codewords from the fine codebook 260. Centroids which do correspond to codewords from codebook z6o may be referred to as 'reference codcwords', whilst ccntroids which do not correspond to codcwords 3o from codebook 26o may be referred to as 'hypothetical codewords'.

As explained above, the first CST sample may be determined to be associated with the region corresponding to the codeword/centroid to which it is closest, under the distance metric.

In operation 203, terminal device too transmits first CSI to the network node based on the first CSI sample. For instance, the first CSI transmitted to the network may comprise a first identifier that identifies the codebook subset. As described above, the first identifier may enable unambiguous identification of the codebook subset. For instance, the first identifier may comprise at least one of: an identifier of the codebook subset, an identifier of a reference codeword for the codeword subset (e.g. from codebook 25o), or an identifier of a hypothetical codeword for the codeword subset (i.e. a centroid of the codebook subset which is not necessarily a codeword itself) as described above. In this way, coarse CSI may be reported to the network.

Put another way, terminal device too may transmit an index (or other identifier) of a subset of codebook 26o to network node 150. Such a subset may correspond to a single region defined by codebook 25o. Reporting information that enables the subset to be identified by the network may therefore be considered as a long-term CSI report (i.e. coarse latent sub-vector information).

In operation 204, network node 150 transmits second reference signals for reception at terminal device too. The second reference signals may be transmitted using the received first channel state information. In some examples, the transmitted second reference signals may be precoded (i.e. beamformed) CSI-RS.

In the two-step framework described above, step 2 may be initiated by execution of operation 204 by network node 15o. In addition or alternatively, the transition from step 1 to step 2 (the 'step transition') may be triggered by explicit signalling between terminal device too and network node 15o. In examples in which the second reference signals are precoded, the calculation or selection of the precoding matrix on the network side may be based on the codebook subset indicated by the received first CSI.

In operation 205, terminal device too obtains a second CST sample in the CSI space so based on second measurements of the second reference signals transmitted in operation 204. In some examples, the second measurements are sub-band measurements. In this way, high-resolution CST feedback is possible.

In general terms, the aim of step 2 is for terminal device too to find the best codeword 35 (i.e. the codeword closest to the second CSI sample under the distance metric) in codebook 26o (i.e. the fine codebook) within the constraints of the subset indicated by the received first CSI. The precoding/beamforming of the second reference signals may be based on the received first CSI.

Tn operation 206, terminal device loci transmits second channel state information to the network node based on the second channel state information sample. The second channel state information is indicative of a codeword within a codebook subset, and the codebook subset is determined based on the first channel state information and comprises a plurality of neighbouring codewords of a codebook that are determined according to a distance metric defined over the channel state information space.

Put another way, terminal device loo transmits information indicative of the best codeword within the codebook subset that was determined based on the first CSI. For instance, this information may be an index of the codeword within the codebook subset. Reporting information that allows this codeword to be identified may be considered to be a short-term CSI report (i.e. refined latent sub-vector information).

Tn operation 207, the first and second CST received from terminal device mo is used by the network in performance of DL communications. For instance, network node 150 may combine the coarse and refined CSI feedback to perform high-resolution beamforming on PDSCH.

Whilst not depicted in Fig. 2, each of step 1 (i.e. operations 201 to 203) and step 2 (i.e. operations 204 to 206) may be repeated independently any number of times.

Note that step 1 does not need to be executed every time the method is repeated, and should be repeated only when coarse information is needed. When the channel exhibits temporal correlation (such as in low mobility use cases), repetition of step 2 should be enough to acquire the CSI. By skipping step 1 except when needed in this way, CSI feedback overhead for over the air signalling may be further reduced.

For example, one 4-dimensional sub-vector of the latent vector can be represented by an 8-bit superset codebook (i.e. fine codebook 26o), which contains 256 codewords in total, each corresponding to a region of the CSI space. In this example, 16 adjacent encoding regions can be grouped together to form combined regions (i.e. coarse regions). The overall superset can thus be divided into 16 subsets, each of which contains 16 codewords from codebook 26o. In this case, the long-term CSI reported in step 1 can use 4 bits, and the short-term CSI reported in step 2 can also use 4 bits. Dividing the codewords into subsets in this way may enable UE complexity at step 2 to be reduced, as, for each codeword search performed at step 2, only the 16 codewords of the subset need to be considered, instead of the 256 codewords of codebook 26o in its entirety.

In some examples, the number of subsets into which codebook 260 is divided may be adaptively determined. Put another way, the bit allocation between steps 1 and 2 may be adaptively determined (e.g. 3 and 5 bits, or 5 and 3 bits) to reflect the variation of the channel over time.

In addition or alternatively, the frequency with which step 1 is performed may be determined on the network side (e.g. at network node i5o). Such frequency may be comparative to the frequency with which step 2 is performed (e.g. perform step 1 after a predefined number of iterations of step 2) or based on a predetermined time duration since the last performance of step 1. In other examples, an external control loop may be introduced by taking block error rate (BLER) or throughput into account. For instance, in the case of BLER or throughput degradation, network node 150 should trigger acquisition of high-resolution CSI performing steps 1 and 2 in succession.

Fig. 3 is a message flow sequence, indicated generally by the reference numeral 3, in accordance with some aspects of the described technology. The message flow sequence 3 shows an example implementation within which aspects of a process, such as that described with reference to Fig. 1, may be performed (e.g. a tCB-based CSI feedback scheme). In this non-limiting example, terminal device 100 is depicted as a mobile UE in communication with network node 15o, which is depicted as a gNB.

In terms of the two-step CSI reporting framework described above, the first step may correspond to operations 301 to 303 and the second step may correspond to operations 30 304 to 306.

In general terms, the aim of step 1 is for terminal device loo to find the best codeword within the superset/fine granularity codebook 36o (i.e. the codeword closest to the first CSI sample under the distance metric). This may be referred to as a 'pinpointing phase'.

In operation 3o1, network node 150 transmits first reference signals for reception at terminal device 100. In some examples, the transmitted first reference signals may be precoded (i.e. beamformed) CSI-RS.

In operation 302, terminal device 100 obtains a first CSI sample in a CSI space based on first measurements of the first reference signals transmitted in operation 301. The first measurements may be sub-band measurements.

Put another way, a first CSI sample is obtained as described with respect to Fig. 1.

Terminal device too may then determine a first reference codeword, which is the codeword within the codebook 36o that is closest to the first channel state information sample according to the distance metric. Unlike the hypothetical codewords described with reference to operation zoz, the reference codewords in the present example are actual codewords from codebook 36o.

The codebook subset, which may be referred to as the 'tracking subset' 361, comprises the first reference codeword and a pre-defined (i.e. known on both UE and NW side) number of neighbouring codewords from codebook 36o that are in the vicinity of the first reference codeword according to the distance metric.

In operation 303, terminal device loo transmits first CSI to the network node based on the first CSI sample. For instance, the first CSI transmitted to the network may comprise an identifier of the first reference codeword within the codebook.

In operation 304, network node 150 transmits second reference signals for reception at terminal device too. The second reference signals may be transmitted using the received first channel state information. In some examples, the transmitted second reference signals may be precoded (i.e. beamformed) CSI-RS.

3() Analogously to message flow sequence 2, in the two-step framework, step 2 may be initiated by execution of operation 304 by network node or may be triggered by explicit signalling between terminal device too and network node 150. In examples in which the second reference signals are precoded, the calculation or selection of the precoding matrix on the network side may be based on the codebook subset (i.e the tracking subset 361) indicated by the received first CSI.

-27 -In operation 305, terminal device loo obtains a second CSI sample in the CSI space based on second measurements of the second reference signals transmitted in operation 304. The second measurements may be sub-band measurements.

In general terms, the aim of step 2 is for terminal device to find the best codeword within the selected tracking subset 361. Since the definition of the codebook (including distance metric, codebook size etc.) is known on both UE and NW side, the configuration of the tracking subset 361 is aligned between terminal device 100 and the network. This step may be referred to as a 'subspace tracking phase' and the codebook 36o may be said to have subspace-tracking capability.

In operation 306, terminal device 100 transmits second channel state information to the network node based on the second channel state information sample. In some examples, the second channel state information comprises an identifier of a second reference codeword within the codebook subset 361that is closest to the second channel state information sample according to the distance metric.

Put another way, terminal device loo transmits information indicative of the best codeword within the tracking subset 361 (e.g. an index within the codebook subset). As 20 with operation 206, this may be considered to be a short-term CSI report.

However, in the event that a subsequent (i.e. third) CSt sample is obtained based on further reference signals received from the network node, the second reference codeword may be used to determine a new tracking subset 362 from which the codeword corresponding to the third CSI sample is selected. Put another way, third CSI transmitted to network node 150 based on the third CSI sample may comprise an identifier of a third reference codeword within a second codebook subset 362 comprising the second reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the second reference codeword according to the distance metric. It will be appreciated that, in a corresponding manner, the third reference codeword may be used to determine a further tracking subset 363 of codebook 36o from which a codeword corresponding to a fourth CST sample may be selected, and so forth.

In operation 307, the first and second CSI information received from terminal device is used by the network in performance of DL communications. For instance, network node 150 may be use the CSI to perform DL precoding for PDSCH as part of a MIMO procedure.

Whilst not depicted in Fig. 3, each of step 1 (i.e. operations 301 to 303) and step 2 (i.e. 5 operations 304 to 306) may be repeated independently any number of times.

Note that step 1 may require high-resolution CSI feedback (e.g. 8 bits per sub-vector), whilst step 2 requires comparatively lower overhead CSI feedback (say, 4 bits per sub-vector) due to the small size of the tracking subsets 361-363 compared to overall codebook 360.

Due to the subspace-tracking capability of the neighbouring codebook described above, step 2 may be repeatedly executed without having to frequently 'adjust' the CSI via performance of step 1. The ratio of step 1 to step 2 operations may be adaptively controlled by the network (see discussion under Fig. 2). For instance, this may be achieved by monitoring the channel temporal variation and/or considering the BLER or throughput of PDSCH.

It will be appreciated, however, that while execution of step 2 without the support of step 1, may lead to lower over-the-air CSI feedback overhead, this may also lead to a higher possibility of channel tracking deviation due to abrupt channel changes. This is therefore a trade-off which needs to be balanced.

In principle, the tracking codebook-based scheme described with reference to Fig. 3 should provide better performance due to its subspace-tracking capability, meaning that step 2 can be repeated back-to-back over a longer period of time without intervention of step 1 than is possible with the combined codebook-based scheme described with reference to Fig. 2. However, the combined codebook-based scheme is more closely aligned with the legacy CSI reporting framework, delivering both coarse 3o and refined CST in turn. As such, both approaches may be beneficially leveraged in UE-network node communication scenarios.

It will be appreciated that several different tracking/combined codebooks of different sizes may be preconfigured at the UE/network in order to code with different radio link 35 conditions. For instance, a small size codebook may be used in low mobility use cases, whilst a large size codebook may be used for high mobility use cases. By taking into account BLER or throughput as described above, this scheme may be made more robust.

Recall the example described with reference to Fig. 1 in which 320 search operations over the codebook of 256 codewords and 2560 bits of CSI feedback are required to process the latent vector. Adopting a split of 4 bits -4 bits for coarse/fine CSI feedback under the combined codebook approach or a 4 bit tracking codebook (i.e. size 21'4 = 16) for the tracking codebook approach, in the presence of CSI temporal correlation the number of sub-vector VQ search operations may be reduced to 320 search operations over 16 codewords (i.e. reduction by a factor of 16), whilst CST feedback may be reduced to 1280 bits (i.e. reduction by a factor of 2). Tn this way, computational complexity of VQ operation at UE side may be reduced along with the over-the-air CSI feedback overhead, while maintaining the concept of the legacy 2-step hybrid CSI reporting framework.

Tn some examples, the codewords of the tracking codebook (i.e. the subset) may be arranged in increasing order of their distance to the reference codeword under the distance metric. In such examples, the reference codeword may be given the smallest index. This approach may be referred to as using an 'adaptive codebook'. It will be appreciated that arranging the subset in this way means that the codewords are sorted in decreasing order of their (approximate) transition probabilities.

One benefit of this approach is that it facilitates easier data analytics, as the smaller index implies the shorter distance to the previously chosen codeword. This analytical data may be used, possibly in combination with external outer-loop control parameters like BLER or throughput, for adaptive configuration of tracking codebook size by network node 150.

This may allow more flexible operation of the tracking codebook and may further 3o enable proactive addressing of channel variation. This kind of information can be collected and used for offline analysis at gNB side or network operator side to improve CST feedback operation. For example, this data may be used to determine when step 1 should be initiated so as to correct for deviations induced by repeated performance of step 2 CSI acquisition.

When using channel eigenvectors as input to the AI encoder, each layer may be encoded separately. In this case, simulation results show that the dominant eigenvector is more critical to the overall performance than the other eigenvectors. As such, when configuring tracking codebooks for each layer's corresponding eigenvector, this can be exploited to further optimize the feedback overhead. For example, a higher number of bits can be allocated for the tracking codebook of the dominant eigenvector, whereas a lower number of bits can be used for that of the non-dominant eigenvector(s). Alternatively, a per-layer CSI reporting loop can be introduced to adjust the tracking codebook CSI update rate on a per layer basis. For instance, a more frequent update rate may be used for the dominant eigenvector, and a less frequent update rate for the other vectors.

In some examples, a codeword transition probability can be used as the codebook configuration metric, rather than the distance metric. This may require extensive /5 simulation and/or field measurements over various channel profiles in a variety of rich scenarios in order to get statistically meaningful state transition statistics.

Figs. 4, 5 and 6 are schematic illustrations of codebooks, in accordance with various examples. In particular, these drawings depict how the segmentized latent vectors (i.e. the sub-vectors) can be projected into a certain CSI space. It will be appreciated that training samples used to train the Al decoder may also be projected into such a space.

As depicted in Fig. 4(a), the entire space 400 may be partitioned into a number of encoding regions equal to the number of codewords in the entire codebook. In the Fig., the individual codewords of the codebook are depicted as small, filled dots 401. The boundaries between the regions defined by the codewords are depicted as lines 402. In such examples, the distance metric used for codebook configuration should be chosen consistently with the underlying vector quantisation scheme (e.g. Euclidean distance, Chordal distance etc.). By way of example only, the chordal distance d between two neighbouring codewords 401 is depicted using reference sign 403.

It will be appreciated that subsequent CST samples may be best represented by different codewords 401 within the codebook. For instance, a first CSI sample may be closest to codeword 404, a second CSI sample (obtained after the first CSI sample) may be closest to codeword 405, and a third CSI sample (obtained after the second CSI sample) may be closest to codeword 406.

-31 -Fig. 4(b) depicts the concept of the combined codebook, as described above, in the context of the CST space 400 of Fig. 4(a). Here, the space is partitioned into combined Voronoi regions. For instance, in this example, four Voronoi regions corresponding to codewords 401 from the codebook 400 are grouped into one combined region. The boundaries between the combined regions are depicted by dashed lines 407.

As described with reference to the previous Figs, these combined regions may be identified by a centroid 408 of the combined Voronoi region (depicted by larger dots /0 408 in the Fig.). Moreover, it will be appreciated that the centroids 408 representing the combined Voronoi regions are not necessarily actual codewords 401 from codebook 400, but may be other points within the CSI space.

Fig. 4(c) depicts the concept of the tracking codebook, as described above, in the /5 context of the CST space 400 of Fig. 4(a). in this case, combined regions are still formed by combining four individual Voronoi regions, but the combined regions are not disjoint from one another as they are in the combined codebook case. By contrast, the combined regions may 'overlap' as depicted in the Fig. Put another way, the codeword subsets corresponding to the combined regions may have non-zero intersection. Such overlapping regions may be constructed by i) taking the currently selected codeword as the centre codeword and ii) forming the tracking codebook by selecting the adjacent codewords (i.e. codewords nearby under a distance metric) with respect to the centre codeword so as to construct a neighbouring codeword group. As mentioned above, the same distance metric may be used as used for vector quantization itself (e.g. distance metric used for k-means algorithm). It will be appreciated that different numbers of neighbouring codewords may be chosen to constitute the tracking subset.

In the depicted non-limiting example, a first tracking subset is centred on codeword 404 and has boundary 409, a second tracking subset is centred on codeword 405 and has boundary 410, and a third tracking subset is centred on codeword 406 and has boundary 411. In this example, codeword 404 is contained the first tracking subset, codeword 405 is contained in the first, second and third tracking subsets, and codeword 406 is contained in the second and third tracking subsets. It will be appreciated that the first, second and third tracking subsets may correspond to tracking subsets 361-363 described with reference to Figure 3.

In Figs. 5 and 6, the codebooks are projected to spherical surfaces for illustrative purposes. In the case of the combined codebook (Fig. 5) the entire CS1 space 50o to which the sub-vectors of the latent vectors belong can be divided into regions according to subsets of the codebook. Such subsets are depicted as large ellipses with centres at A, B. After a subset has been selected (i.e. A or B, 'coarse information'), an individual codeword can be determined within the selected subset (i.e. a or b, 'refined information'). In the Fig., X::y' is used to indicate codeword 'y' (where y = a, b or c) within subset 'X' (where X = A or B).

to Tn case of the tracking codebook (Fig. 6), the first codeword, a, is selected by an exhaustive search over the entire CSI space 600. As described above, this may be referred to as the 'pinpointing phase'. Once the first codeword has been selected, a neighbouring 'tracking' codebook of a smaller size than the overall 'superset' codebook can be identified. Such tracking codebooks are depicted as ellipses centred at currently/recently selected codewords, i.e., 'a', 'b', 'c', etc. Tn this tracking codebook configuration, the currently selected codeword becomes the centre codeword (i.e. the reference codeword) and a subspace-tracking neighbouring codebook 4-is defined as follows: ctr 37,1 fCiiv.i. E K( ), for j../p} Tn this example, the next-step e-neighbourhood f * with centre codeword (-11 is defined as: gi(c) "'IC (C* Arj(e)1 where J 1 2, cBT, N < NCB The current state dependent tracking codebook may thus be modelled as a first-order finite-state Markov chain. The current time is p (with the chosen codeword index j), and the most likely codeword indices for the next time step are denoted i. The remaining terms are defined as follows: jp: latent sub-vector state (quantized latent sub-vector index) at time p, N": entire (superset) codebook size, Altr: subspace-tracking codebook size, d(): distance metric in use for VQ or for measuring distance between codewords.

In short, the tracking codebook Cfr is a collection of Nu.. codewords chosen from the superset C which are closest to C j,, . The tracking codebook is centred around

C_

(and includes) jk'. The tracking codebook is dependent on the current state Jr, and is of the same size irrespective ofJp, and it has a tracking capability (i.e. it is a current state-dependent codebook).

Fig. 7 is a flowchart depicting various operations which maybe performed in ro accordance with various examples. For instance, the operations depicted in Fig. 7 may be executed by a terminal device or other suitable apparatus.

In operation S7.1, a terminal device obtains a first channel state information sample in a channel state information space based on first measurements of first reference signals received from a network node.

Tn operation S7.2, the terminal device transmits, to the network node, first channel state information based on the first channel state information sample.

Tn operation S7.3, the terminal device obtains a second channel state information sample in the channel state information space based on second measurements of second reference signals received from the network node, the second measurements being performed after the first measurements.

23 In operation S7.4, the terminal device transmits, to the network node, second channel state information based on the second channel state information sample. The second channel state information is indicative of a codeword within a codebook subset, and the codebook subset is determined based on the first channel state information and comprises a plurality of neighbouring codewords of a codebook that are determined according to a distance metric defined over the channel state information space.

As will of course be appreciated, various operations illustrated in Fig. 7 may correspond to operations already described with reference to the preceding Figs. For instance, operation S7.1 may correspond to operation 202 in Fig. 2 or operation 302 in Fig. 3; operation S7.2 may correspond to operation 203 in Fig. 2 or operation 303 in Fig. 3; operation 87.3 may correspond to operation 205 in Fig. 2 or operation 305 in Fig. 3; and operation 87.4 may correspond to operation 206 in Fig. 2 or operation 306 in Fig. 3 Fig. 8 is a flowchart depicting various operations which may be performed in accordance with various examples. For instance, the operations depicted in Fig. 8 may be executed by a network node (e.g. a base station) or other suitable apparatus.

In operation S8.1, a network node transmits first reference signals for reception at a terminal device.

In operation 88.2, the network node receives, from the terminal device, first channel state information, wherein the first channel state information is based on a first channel state information sample in a channel state information space obtained by the terminal device based on first measurements of the first reference signals.

In operation S8.3, the network node transmits second reference signals for reception at the terminal device. In some examples, the transmission of the second reference signals may be based on the received first channel state information. For instance, the second reference signals may be precoded/beamformed using the received first channel state information.

In operation S8.4, the network node receives, from the terminal device, second channel state information, wherein the second channel state information is based on a second channel state information sample in the channel state information space obtained by the terminal device based on second measurements of the second reference signals, the second measurements being performed after the first measurements. The second channel state information is indicative of a codcword within a codebook subset, and the 3o codebook subset is determined based on the first channel state information and comprises a plurality of neighbouring codewords of a codebook that are determined according to a distance metric defined over the channel state information space.

As will of course be appreciated, various operations illustrated in Figure8 may correspond to operations already described with reference to the preceding Figures. For instance, operation S8.1 may correspond to operation 201 in Fig. 2 or operation 301 in Fig. 3; and operation S8.3 may correspond to operation 204 in Fig. 2 or operation 304 in Fig. 3. Similarly, the 'receiving' operations S8.2 and S8.4 may correspond to the transmitting' operations 203/303 and 206/306 respectively.

Fig. 9 is a schematic illustration of an example configuration of a computing apparatus 9 which may be configured to perform various operations described with reference to Figs. i to 8.

Computing apparatus may comprise control apparatus 900 which is configured to to control operation of other components which form part of the computing apparatus 9 thereby to enable performance of various operations described with reference to Figs. 1 to 8. The computing apparatus 900 may comprise processing apparatus 901 and memory 902. Computer-readable code 9o2-2A may be stored on the memory 902, which when executed by the processing apparatus 901, causes the control apparatus 900 to perform any of the operations described herein.

In addition, computing apparatus may further include a display 903, user interactive interface (UII) 904, radio frequency interface 905 configured to interface radio frequency signals transmitted and received via a radio frequency antenna array 905A, and global navigation satellite system (GNSS) 906. In some examples, other satellite communications systems may be used instead of or in addition to GNSS 906.

Fig. 10 is a schematic illustration of an example configuration of a radio access node or base station 10 which may be configured to perform various operations described with reference to Figs. 1 to 8.

The base station 10, which may be referred to as a gNB, base station or access point (AP), comprises control apparatus woo which is configured to control operation of other components which form part of the base station in thereby to enable transmission of signals to and receipt of signals from UEs in its coverage area vicinity.

For example, the base station control apparatus woo is configured to cause transmission of reference signals to UEs within its coverage area. Furthermore, in some examples, the control apparatus woo may be configured to enable receipt of reference signal measurement data and/or location data from the UEs in its coverage area. The control apparatus moo may also enable communication with other base stations and/or other network nodes. The control apparatus 1000 may additionally be configured to cause performance of any other operations described herein with reference to the base station 10.

The base station to comprises a radio frequency antenna array 10005 configured to receive and transmit radio frequency signals. Although the base station to in Fig. to is shown as having an array 1oo5A of four antennas, this is illustrative only. The number of antennas may vary, for instance, from one to many hundreds.

The base station to further comprises a radio frequency interface 1005 configured to interface the radio frequency signals received and transmitted by the antenna 1oo5A and a control apparatus too. The radio frequency interface too5 may also be known as a transmitter, receiver and/or transceiver. The base station to may also comprise an interface 1007 via which, for example, it can communicate with other network elements such as other radio access network entities (such as the other base stations) and/or core network entities.

The base station control apparatus woo may be configured to process signals from the radio frequency interface 1005, to control the radio frequency interface 1005 to generate suitable RF signals to communicate information to UEs via the wireless communications link, and also to exchange information with other base stations to and core network entities via the interface 1007.

The control apparatus moo may comprise processing apparatus 1001 and memory 1002. Computer-readable code 1oo2-2A may be stored on the memory 1002, which when executed by the processing apparatus 1001, causes the control apparatus woo to perform any of the operations described herein and attributed to the base station 10.

Some further details of components and features of the above-described devices/entities/apparatuses 9, in and alternatives for them will now be described.

The control apparatuses described above 900, woo may comprise processing apparatus 901, 1001 communicatively coupled with memory 902, 1002. The memory 902, 1002 has computer readable instructions 9o2-2A, 1oo2-2A stored thereon, which when executed by the processing apparatus 901, 1001 causes the control apparatus 900, t000 to cause performance of various ones of the operations described with reference to Figs. 1 to 8. The control apparatus 900, loon may in some instances be referred to, in general terms, as "apparatus".

The processing apparatus 901, looi may be of any suitable composition and may include one or more processors 901A, 1001A of any suitable type or suitable combination of types. Indeed, the term "processing apparatus" should be understood to encompass computers having differing architectures such as single/multi-processor architectures and sequencers/parallel architectures. For example, the processing apparatus 901, 1001 may be a programmable processor that interprets computer program instructions 9o2-2A, 1oo2-2A and processes data. The processing apparatus 901, 1001 may include plural programmable processors. Alternatively, the processing apparatus 904 1001 may be, for example, programmable hardware with embedded firmware. The processing apparatus 901, 1001 may alternatively or additionally include one or more specialised circuit such as field programmable gate arrays FPGA, Application Specific Integrated Circuits (ASTCs), signal processing devices etc. in some instances, processing apparatus 901, 1001 may be referred to as computing apparatus or processing means.

The processing apparatus 901, 1001 is coupled to the memory 902, 1002 and is operable to read/write data to/from the memory 902, 1002. The memory 902, 1002 may comprise a single memory unit or a plurality of memory units, upon which the computer readable instructions (or code) 902-2A, 1002-2A is stored. For example, the memory 902, 1002 may comprise both volatile memory 902-1, 1002-1 and non-volatile memory 902-2, 1002-2. In such examples, the computer readable instructions/program code 9o2-2A, 1002-2A may be stored in the non-volatile memory 902-2, 1002-2 and may be executed by the processing apparatus 901, 1001 using the volatile memory 902-4 1002-1 for temporary storage of data or data and instructions. Examples of volatile memory include random-access memory (RAM), dynamic random-access memory (DRAM), and synchronous dynamic random-access memory so (SDRAM) etc. Examples of non-volatile memory include read-only memory (ROM), programmable read-only memory (PROM), electronically erasable programmable read-only memory (EEPROM), flash memory, optical storage, magnetic storage, etc. The memory 902, 1002 may be referred to as one or more non-transitory computer readable memory medium or one or more storage devices. Further, the term 'memory', in addition to covering memory comprising both one or more non-volatile memory and one or more volatile memory, may also cover one or more volatile memories only, one or more non-volatile memories only. In the context of this document, a "memory" or "computer-readable medium" may be any media or means that can contain, store, communicate, propagate or transport the instructions for use by or in connection with an instruction execution system, apparatus, or device, such as a computer.

The computer readable instructions/program code 902-2A, 1002-2A may be preprogrammed into the control apparatus 900, 1000. Alternatively, the computer readable instructions 9o2-2A, loo2-2A may arrive at the control apparatus via an electromagnetic carrier signal or may be copied from a physical entity n such as a computer program product, a memory device or a record medium such as a compact disc read-only memory (CD-ROM) or digital versatile disc (DVD) an example of which is illustrated in Fig. 11. The computer readable instructions 902-2A, 1002-2A may provide the logic and routines that enables the entities devices/apparatuses 9, 10 to perform the functionality described above. The combination of computer-readable instructions stored on memory (of any of the types described above) may be referred to as a computer program product. In general, references to computer program, instructions, code etc. should be understood to express software for a programmable processor firmware such as the programmable content of a hardware device as instructions for a processor or configured or configuration settings for a fixed function device, gate array, programmable logic device, etc. If desired, the different functions discussed herein may be performed in a different order and/or concurrently with each other. Furthermore, if desired, one or more of the above-described functions may be optional or may be combined. Similarly, it will also be appreciated that the flow diagrams of Figs. 4 and 7 are examples only and that various operations depicted therein may be omitted, reordered and/or combined.

Although the methods and apparatuses have been described in connection with an E-sc) UTRA or 5G network, it will be appreciated that they are not limited to such networks and are applicable to radio networks of various different types.

Although various aspects of the methods and apparatuses described herein are set out in the independent claims, other aspects may comprise other combinations of features 35 from the described embodiments and/or the dependent claims with the features of the independent claims, and not solely the combinations explicitly set out in the claims.

It is also noted herein that while various examples are described above, these descriptions should not be viewed in a limiting sense. Rather, there are several variations and modifications which may be made without departing from the scope of 5 the present invention as defined in the appended claims.

Claims

Claims 1. A terminal device comprising: means for obtaining a first channel state information sample in a channel state 5 information space based on first measurements of first reference signals received from a network node; means for transmitting, to the network node, first channel state information based on the first channel state information sample; means for obtaining a second channel state information sample in the channel state information space based on second measurements of second reference signals received from the network node, the second measurements being performed after the first measurements; and means for transmitting, to the network node, second channel state information based on the second channel state information sample, wherein the second channel state information is indicative of a codeword within a codebook subset, and wherein the codebook subset is determined based on the first channel state information and comprises a plurality of neighbouring codewords of a codebook that are determined according to a distance metric defined over the channel state 20 information space.
2. The terminal device of claim 1, wherein the codebook comprises a plurality of codewords organised into pre-defined codebook subsets of neighbouring codewords according to the distance metric, wherein the first channel state information comprises a first identifier that identifies the codebook subset, and wherein the second channel state information comprises a second identifier that identifies the codeword within the codebook subset.
3. The terminal device of claim 2, wherein the codebook subset is the codebook subset of the pre-defined codebook subsets with which the first channel state information sample is associated, and wherein the codeword is the codeword from the codebook subset that is closest to the second channel state information sample according to the distance metric.
-41 - 4. The terminal device of claim 2 or 3, wherein the first measurements are wideband or sub-band measurements, and wherein the second measurements are sub-band measurements.
5. The terminal device of claim 1, wherein the first channel state information comprises a first identifier that identifies a first reference codeword within the codebook that is closest to the first channel state information sample according to the distance metric, wherein the codebook subset comprises the first reference codeword and a pre-to defined number of neighbouring codewords in the vicinity of the first reference codeword according to the distance metric, and wherein the second channel state information comprises a second identifier that identifies a second reference codeword within the codebook subset that is closest to the second channel state information sample according to the distance metric.
6. The terminal device of claim 5, further comprising: means for obtaining a third channel state information sample in the channel state information space based on third measurements of third reference signals received from the network node, the third measurements being performed after the 20 second measurements; and means for transmitting, to the network node, third channel state information based on the third channel state information sample, wherein the third channel state information comprises a third identifier that identifies a third reference codeword within a second codebook subset comprising the 25 second reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the second reference codeword according to the distance metric.
7. The terminal device of claim 5 or claim 6, wherein the first and second measurements arc sub-band measurements.
8. The terminal device of any one of the preceding claims, wherein the channel state information space is a vector space spanned by sub-vector components of a latent vector, and wherein the latent vector is obtained after Al/ML encoding of a channel eigenvector or a channel singular vector or a channel matrix that is determined based on the first or second measurements.
-42 - 9. A network node comprising: means for transmitting first reference signals for reception at a terminal device; means for receiving, from the terminal device, first channel state information, wherein the first channel state information is based on a first channel state information sample in a channel state information space obtained by the terminal device based on first measurements of the first reference signals; means for transmitting second reference signals for reception at the terminal device; and means for receiving, from the terminal device, second channel state information, wherein the second channel state information is based on a second channel state information sample in the channel state information space obtained by the terminal device based on second measurements of the second reference signals, the second measurements being performed after the first measurements, wherein the second channel state information is indicative of a codeword within a codebook subset, and wherein the codebook subset is determined based on the first channel state information and comprises a plurality of neighbouring codewords of a codebook that are determined according to a distance metric defined over the channel state 20 information space.
10. The network node of claim 9, wherein the codebook comprises a plurality of codewords organised into pre-defined codebook subsets of neighbouring codewords according to the distance metric, wherein the first channel state information comprises a first identifier that identifies the codebook subset, and wherein the second channel state information comprises a second identifier that identifies the codeword within the codebook subset.
11. The network node of claim 10, wherein the codebook subset is the codebook subset of the pre-defined codebook subsets with which the first channel state information sample is associated, and wherein the codeword is the codeword from the codebook subset that is closest to the second channel state information sample according to the distance metric.
12. The network node of claim to or 11, wherein the first measurements are wideband or sub-band measurements, and wherein the second measurements are sub-band measurements.
13. The network node of claim 9, wherein the first channel state information comprises a first identifier that identifies a first reference codeword within the codebook that is closest to the first channel state information sample according to the distance metric, wherein the codebook subset comprises the first reference codeword and a pre-to defined number of neighbouring codewords in the vicinity of the first reference codeword according to the distance metric, and wherein the second channel state information comprises a second identifier that identifies a second reference codeword within the codebook subset that is closest to the second channel state information sample according to the distance metric.
14. The network node of claim 13, further comprising: means for transmitting third reference signals for reception at the terminal device; and means for receiving, from the terminal device, third channel state information, wherein the third channel state information is based on a third channel state information sample in the channel state information space obtained by the terminal device based on third measurements of the third reference signals, the third measurements being performed after the second measurements, wherein the third channel state information comprises a third identifier that 25 identifies a third reference codeword within a second codebook subset comprising the second reference codeword and a pre-defined number of neighbouring codewords in the vicinity of the second reference codeword according to the distance metric.
15. The network node of claim 13 or claim 14, wherein the first and second 30 measurements are sub-band measurements.
16. The network node of any one of claims 9 to 15, wherein the channel state information space is a vector space spanned by sub-vector components of a latent vector, and -44 -wherein the latent vector is obtained after AI/ML encoding of a channel eigenvector or channel singular vector or a channel matrix that is determined based on the first or second measurements.
17. A method comprising: obtaining a first channel state information sample in a channel state information space based on first measurements of first reference signals received from a network node; transmitting, to the network node, first channel state information based on the /a first channel state information sample; obtaining a second channel state information sample in the channel state information space based on second measurements of second reference signals received from the network node, the second measurements being performed after the first measurements; and transmitting, to the network node, second channel state information based on the second channel state information sample, wherein the second channel state information is indicative of a codeword within a codebook subset, and wherein the codebook subset is determined based on the first channel state 20 information and comprises a plurality of neighbouring codewords of a codebook that are determined according to a distance metric defined over the channel state information space.
18. A method comprising: transmitting first reference signals for reception at a terminal device; receiving, from the terminal device, first channel state information, wherein the first channel state information is based on a first channel state information sample in a channel state information space obtained by the terminal device based on first measurements of the first reference signals; transmitting second reference signals for reception at the terminal device; and receiving, from the terminal device, second channel state information, wherein the second channel state information is based on a second channel state information sample in the channel state information space obtained by the terminal device based on second measurements of the second reference signals, the second measurements being 35 performed after the first measurements, wherein the second channel state information is indicative of a codeword within a codebook subset, and wherein the codebook subset is determined based on the first channel state information and comprises a plurality of neighbouring codewords of a codebook that 5 are determined according to a distance metric defined over the channel state information space.