TECHNICAL FIELD

Embodiments herein relate to a UE and methods therein. In some aspects, they relate to normalizing input data to an AE used for CSI reporting to a network node in a wireless communications network.

BACKGROUND

In a typical wireless communication network, wireless devices, also known as wireless communication devices, mobile stations, stations (STA) and/or User Equipments (UE)s, communicate via a Wide Area Network or a Local Area Network such as a Wi-Fi network or a cellular network comprising a Radio Access Network (RAN) part and a Core Network (CN) part. The RAN covers a geographical area which is divided into service areas or cell areas, which may also be referred to as a beam or a beam group, with each service area or cell area being served by a radio network node such as a radio access node e.g., a Wi-Fi access point or a radio base station (RBS), which in some networks may also be denoted, for example, a NodeB, eNodeB (eNB), or gNB as denoted in Fifth Generation (5G) telecommunications. A service area or cell area is a geographical area where radio coverage is provided by the radio network node. The radio network node communicates over an air interface operating on radio frequencies with the wireless device within range of the radio network node.

3GPP is the standardization body for specify the standards for the cellular system evolution, e.g., including 3G, 4G, 5G and the future evolutions. Specifications for the Evolved Packet System (EPS), also called a Fourth Generation (4G) network, have been completed within the 3rd Generation Partnership Project (3GPP). As a continued network evolution, the new releases of 3GPP specifies a 5G network also referred to as 5G New Radio (NR).

Frequency bands for 5G NR are being separated into two different frequency ranges, Frequency Range 1 (FR1) and Frequency Range 2 (FR2). FR1 comprises sub-6 GHz frequency bands. Some of these bands are bands traditionally used by legacy standards but have been extended to cover potential new spectrum offerings from 410 MHz to 7125 MHz FR2 comprises frequency bands from 24.25 GHz to 52.6 GHz. Bands in this millimeter wave range have shorter range but higher available bandwidth than bands in the FR1.

Multi-antenna techniques may significantly increase the data rates and reliability of a wireless communication system. For a wireless connection between a single user, such as UE, and a base station, the performance is in particular improved if both the transmitter and the receiver are equipped with multiple antennas, which results in a Multiple-Input Multiple-Output (MIMO) communication channel. This may be referred to as Single-User (SU)-MIMO. In the scenario where MIMO techniques is used for the wireless connection between multiple users and the base station, MIMO enables the users to communicate with the base station simultaneously using the same time-frequency resources by spatially separating the users, which increases further the cell capacity. This may be referred to as Multi-User (MU)-MIMO. Note that MU-MIMO may benefit when each UE only has one antenna. Such systems and/or related techniques are commonly referred to as MIMO.

The 5 ^th generation mobile wireless communication system (NR) uses Orthogonal Frequency Division. Multiplexing (OFDM) with configurable bandwidths and subcarrier spacing to efficiently support a diverse set of use-cases and deployment scenarios. With respect to the 4 ^th generation system (LTE), NR improves deployment flexibility, user throughputs, latency, and reliability. The throughput performance gains are enabled, in part, by enhanced support for MU-MIMO transmission strategies, where two or more UE receives data on the same OFDM time frequency resources, i.e. , spatially separated transmissions.

The MU-MIMO transmission strategy is illustrated in Figure 1. In this illustration, a multi-antenna base station with N _TX antenna ports transmit information to several UEs, with sequence intended for the j'th UE. Before modulation and transmission, precoding w ^J) is applied to S^ j = 1,2, ...,/, to mitigate multiplexing interference - the transmissions are spatially separated. Note the order of modulation and precoding, or demodulation and combining respectively, that may differ depending on the implementation of MU-MIMO transmission.

Each UE demodulates its received signal and combines receiver antenna signals to obtain an estimate of corresponding transmitted sequence. This estimate for the j'th UE can be expressed as

The second term represents the spatial multiplexing interference (due to MU -Ml MO transmission) seen by UE(j) and the third term represents other interference and noise sources. The goal is to construct the set of precoders j to meet a given design target resulting in that Wy ¹ - ¹ correlates well with the channel observed by UE(j) whereas it correlates poorly with the channels, = i, observed by other UEs.

To construct precoders Wy ^M , i = 1,2, ... , J that enable efficient MU-MI MO transmissions, the radio access network needs to obtain detailed information about all the users downlink channels W(i), i = 1,2, . . ,].

CSI reporting in NR

In deployments where full channel reciprocity holds, detailed channel information can be obtained from uplink Sounding Reference Signals (SRS) that are transmitted periodically, or on demand, by active UEs. The radio access network can estimate the uplink channel from SRS and by reciprocity obtain the downlink channel H^.

Full channel reciprocity may be obtained in Time Division Duplex (TDD) deployments for UEs with same number of transmitters (TX chains) as receive branches (RX chains). However, a typical scenario is that UEs have fewer TX chains than RX chains, so the radio access network might only be able to estimate part of the uplink channel using SRS, in which case only certain columns of a precoding matrix can be estimated using SRS. This situation is known as partial channel knowledge.

In Frequency Division Dduplex (FDD) deployments, full channel reciprocity cannot be expected since the uplink and downlink channels use different carriers and, therefore, the uplink channel might not provide enough information about the downlink channel to enable MU-MIMO precoding. In such deployments, active UEs need to feedback Channel State Information (CSI) to the radio access network over uplink control or data channels. In LTE and NR, this feedback is achieved by a signalling protocol that can be outlined as follows:

The radio access network configures a UE to report CSI in a certain way.

The radio access network node transmits CSI reference signals (CSI-RS). • The UE estimates the downlink channel (or important features thereof) from the transmitted CSI-RS.

• The UE reports CSI over an uplink control and/or data channel.

• The radio access network uses the UE’s feedback for downlink user scheduling and precoding.

In the above, important features of the channel may refer to a Gram matrix of the channel, one or more eigenvectors that correspond to the largest eigenvalues of an estimated channel covariance matrix, approximations of such aforementioned eigenvectors, one or more DFT base vectors, or orthogonal vectors from any other suitable and defined vector space, that best correlates with an estimated channel matrix, or an estimated channel covariance matrix, the channel delay profile.

In NR, a UE can be configured to report CSI Type I and CSI Type II, where the CSI Type II reporting protocol has been specifically designed to enable MU -Ml MO operations from uplink UE reports. The CSI Type II can be configured in a normal reporting mode or in a port selection reporting mode.

The CSI Type II normal reporting mode is based on the specification of sets of Discrete Fourier Transform DFT basis functions in a precoder codebook. The UE selects and reports the L DFT vectors from the codebook that best match its channel conditions (like the classical codebook precoding matrix indicator (PMI) from earlier 3GPP releases). The number of DFT vectors L is typically 2 or 4 and it configurable by the NW. In addition, the UE reports how the L DFT vectors should be combined in terms of relative amplitude scaling and co-phasing.

Algorithms to select L, the L DFT vectors, and co-phasing coefficients are outside the specification scope -- left UE and NW implementation. Or, put another way, the Rel 16 specification only defines signalling protocols to enable the above message exchanges.

In the following, the wording “DFT beams” will be used interchangeably with “DFT vectors”. This slight abuse of terminology is appropriate whenever the base station has a uniform planar array with antenna elements separated by half of the carrier wavelength.

The CSI type II normal reporting mode is illustrated in Figure 2, see also technical specification [1], The selection and reporting of the L DFT vectors b _n and their relative amplitudes a _n is done in a wideband manner; that is, the same beams are used for both polarizations over the entire transmission band. The selection and reporting of the DFT vector co-phasing coefficients are done in a subband manner; that is, DFT vector cophasing parameters are determined for each of multiple subsets of contiguous subcarriers. The co-phasing parameters are quantized such that e ^J6n is taken from either a Quadrature Phase Shift Keying (QPSK) or 8 phase shift keying (8PSK) signal constellation.

If the Network (NW) cannot accurately estimate the full downlink channel from uplink transmissions, then active UEs need to report channel information to the NW over the uplink control or data channels. In LTE and NR, this feedback is achieved by the following signalling protocol:

- The NW transmits CSI-RS over the downlink using N ports.

- The UE estimates the downlink channel (or important features thereof) for each of the N ports from the transmitted CSI-RS.

- The UE reports CSI, e.g., Channel Quality Index (CQI), Precoding Matrix Indicator (PMI), Rank Indicator (Rl), to the NW over an uplink control and/or data channel.

- The NW uses the UE’s feedback for downlink user scheduling and MIMO precoding.

In NR, both Type I and Type II reporting is configurable, where the CSI Type II reporting protocol has been specifically designed to enable MU-MI MO operations from uplink UE reports.

The CSI Type II normal reporting mode is based on the specification of sets of Discrete Fourier Transform DFT basis functions in a precoder codebook. The UE selects and reports the L DFT vectors from the codebook that best match its channel conditions (like the classical codebook precoding matrix indicator (PMI) from earlier 3GPP releases). The number of DFT vectors L is typically 2 or 4 and it configurable by the NW. In addition, the UE reports how the L DFT vectors should be combined in terms of relative amplitude scaling and co-phasing.

Algorithms to select L, the L DFT vectors, and co-phasing coefficients are outside the specification scope -- left UE and NW implementation. Or, put another way, the Rel 16 specification only defines signalling protocols to enable the above message exchanges.

In the following, we will use “DFT beams” interchangeably with DFT vectors. This slight abuse of terminology is appropriate whenever the base station has a uniform planar array with antenna elements separated by half of the carrier wavelength.

The CSI type II normal reporting mode is illustrated in Figure 2, see also technical specification [1], The selection and reporting of the L DFT vectors b _n and their relative amplitudes a _n is done in a wideband manner; that is, the same beams are used for both polarizations over the entire transmission band. The selection and reporting of the DFT vector co-phasing coefficients are done in a sub band manner; that is, DFT vector cophasing parameters are determined for each of multiple subsets of contiguous subcarriers. The co-phasing parameters are quantized such that e ^j9n is taken from either a QPSK or 8PSK signal constellation.

With k denoting a sub-band index, the precoder W _v [k] reported by the UE to the NW can be expressed as follows:

The Type II CSI report can be used by the NW to co-schedule multiple UEs on the same OFDM time-frequency resources. For example, the NW can select UEs that have reported different sets of DFT vectors with weak correlations. The CSI Type II report enables the UE to report a precoder hypothesis that trades CSI resolution against uplink transmission overhead.

NR 3GPP Release 15 supports Type II CSI feedback using port selection mode, in addition to the above normal reporting mode. In this case,

The base station transmits a CSI-RS port in each one of the beam directions.

The UE does not use a codebook to select a DFT vector (a beam), instead the UE selects one or multiple antenna ports from the CSI-RS resource of multiple ports.

Type II CSI feedback using port selection gives the base station some flexibility to use non-standardized precoders that are transparent to the UE. For the port-selection codebook, the precoder reported by the UE can be described as follows

Here, the vector e is a unit vector with only one non-zero feature, also referred to as element, which can be viewed a selection vector that selects a port from the set of ports in the measured CSI-RS resource. The UE thus feeds back which ports it has selected, the amplitude factors and the co-phasing factors.

Different CSI-reporting frameworks The CSI type II reporting as described above falls into the category of CSI-reporting framework that can be called precoding-vector feedback, sometimes called implicit feedback. In this framework the UE is reporting suggested precoding-vectors to the NW in different ways and different frequency granularity.

Another category of CSI reporting that can be considered, especially with the development of new powerful compression algorithms, e.g., based on AEs as described below, is full-channel feedback, sometimes called explicit feedback. In this framework the UE reports a compression or representation of the whole observed and/or estimated channel, and possibly also noise covariance estimates, in the feedback.

AE based CSI reporting

Neural Network (NN) based Autoencoders (AEs) have recently gained large interests in the wireless communications research community for its capability of compressing and decompressing, also referred to as reconstructing, MIMO radio channels accurately even at high compression ratios. The use of the AE is here in the context of CSI compressing where a UE provides CSI feedback to a radio access network node by sending a CSI report that include a compressed and encoded version of the estimated downlink channel, or of important features thereof. A summary of recent academia work on this topic can be found in [3], Furthermore, 3GPP decided to start a study item for 3GPP Release18 that includes the use case of Al-based CSI reporting in which AEs will play a central part of the study [2,4],

An AE is a NN, i.e. , a type of machine learning algorithm, that has been partitioned into one encoder and one decoder. This partitioning is illustrated in Figure 3 by considering a simple NN example with fully connected layers (a.k.a. dense NN). The encoder and decoder are separated by a bottleneck layer that holds a compressed representation, Y in Figure 3, of the input data X. The variable Y is sometimes called the latent representation of the input X. More specifically,

- The size of the bottleneck (latent representation) Y is significantly smaller than the size of the input data X. The AE encoder thus compresses the input features X to Y.

- The decoder part of the AE tries to invert the encoder’s compression and reconstruct X with minimal error, according to some predefined loss function. The terms latent representation, latent vector, and output data of the AE encoder is used interchangeably herein. Analogously the terms latent space and output space are used interchangeably herein and refer to the space of all possible latent vectors, for a given architecture. Similarly, the input space is the space of all possible inputs, for a given architecture. The word space may be understood as, e.g., a linear vector space, in the mathematical sense.

AEs may have different architectures. For example, AEs may be based on dense NNs, like Figure 3, multi-dimensional convolution NNs, recurrent NNs, transformer NNs, or any combination thereof. However, all AEs architectures possess an encoder- bottleneck-decoder structure. A characteristic of AEs is that they can be used to compress and decompress data in an unsupervised manner. Figure 3 is an illustration of a fully connected AE.

Figure 4 illustrates how an AE may be used for Al-enhanced CSI reporting in NR. Figure 4 depicts the use of AE for CSI Compression, inference phase.

In summary:

• The UE estimates the downlink channel (or important features thereof) from downlink reference signal(s), e.g., CSI-RS. For example, the UE estimates the downlink channel as a 3D complex-valued tensor, with dimensions defined by the radio access network node Tx-antenna ports, the UE’s Rx antenna ports, and frequency (the granularity of which is configurable, e.g., subcarrier or subband).

• The UE uses a trained AE encoder to compress the estimated channel [features] down to a binary codeword. The binary codework is reported to the radio access network over an uplink control and/or data channel. In practice, this codeword will likely form one part of a CSI report that might also include rank, channel quality, and interference information.

• The radio access network node uses a trained AE decoder to reconstruct the estimated channel, e.g., channel features. The decompressed output of the AE decoder is used by the radio access network in, for example, MIMO precoding, scheduling, and link adaption.

The architecture of an AE, such as e.g., structure, number of layers, nodes per layer, activation function etc., will need to be tailored for each particular use case. For example, properties of the data, such as e.g., CSI-RS channel estimates, the channel size, uplink feedback rate, and hardware limitations of the encoder and decoder all need to be considered when designing the AE’s architecture.

After the AE’s architecture is fixed, it needs to be trained on one or more datasets. To achieve good performance during live operation, the so-called inference phase, the training datasets need to be representative of the actual data the AE will encounter during live operation.

The training process involves numerically tuning the AE’s trainable parameters, e.g., the weights and biases of the underlying NN, to minimize a loss function on the training datasets. The loss function could be, for example, the MSE loss calculated as the average of the squared error between the UE’s downlink channel estimate H and the NN’s reconstruction H, i.e. , \\H - H || . The purpose of the loss function is to meaningfully quantify the reconstruction error for the particular use case at hand.

The training process is typically based on some variant of the mini-batch gradient descent algorithm, which, at its core, has three components: a feedforward step, a back propagation step, and a parameter optimization step.

Feedforward: A batch of training data, such as a mini-batch, e.g., several downlinkchannel estimates, is pushed through the AE, from the input to the output. The loss function is used to compute the reconstruction loss for all training samples in the batch. The reconstruction loss may refer to an average reconstruction loss for all training samples in the batch.

Back propagation (BP): The gradients, partial derivatives of the loss function, L, with respect to each trainable parameter in the AE are computed. The back propagation algorithm sequentially works backwards from the AE output, layer-by-layer, back through the AE to the input. The back propagation algorithm is built around the chain rule for differentiation: When computing the gradients for layer n in the AE, it uses the gradients for layer n+1.

Parameter optimization: The gradients computed in the back propagation step are used to update the AE’s trainable parameters using a gradient descent method with a learning rate hyperparameter that scales the gradients. The core idea is to make small adjustments to each parameter so that the average loss over the training batch decreases. It is commonplace to use special optimizers to update the AE’s trainable parameters using gradient information. The following optimizers are widely used to reduce training time and improving overall performance: adaptive sub-gradient methods (AdaGrad) [5], RMSProp, and Adaptive Moment Estimation (ADAM) [6],

The above process, such as feedforward pass, back propagation, parameter optimization, is repeated many times until an acceptable level of performance is achieved on the training dataset. An acceptable level of performance may refer to the AE achieving a pre-defined average reconstruction error over the training dataset, e.g., normalized MSE of the reconstruction error over the training dataset is less than, say, 0.1. Alternatively, it may refer to the AE achieving a pre-defined user data throughput gain with respect to a baseline CSI reporting method (e.g., a MIMO precoding method is selected, and user throughputs are separately estimated for the baseline and the AE CSI reporting methods).

Existing solutions

Training deep NN, including AEs for CSI reporting, may be difficult because the distribution of data can change from same to sample. For example, the estimated channel of an outdoor line-of-sight UE close to the gNB can be many orders of magnitude larger than that of an indoor UE far away from the gNB. Since the scale of the neural network’s weights and biases depends on the scale of input features, such large sample-to-sample differences can result in slow training times, poor performance, and poor generalizability. For this reason, it is commonplace to perform some type of data normalization. In the CSI use case, for example, the downlink channel estimates might be normalized by their respective pathlosses.

A second problem that plagues training deep neural networks is that the process of updating weights and biases in a layer changes the distribution of data input to other layers. For example, it is commonplace to use backpropagation algorithms to update weights and biases in a backwards manner, layer-by-layer from the output to the input, using an estimate of the loss function gradient that assumes all other weights are fixed. That is, the parameter updates for a particular layer assume that the other network parameters and data distribution do not change. Furthermore, non-linear activation functions within the network are most effective during training if the data input to those functions resides within the “nonlinear domain”, typically centred on the “x = 0” origin, of the activation function; in particular, ensuring so helps address the vanishing gradient problem that slows down training.

In Reference [13] it was shown that normalizing data in between layers (essentially, standardizing the distributions of intermediate-layer activations) across training (mini-) batches can significantly improve training time and performance. Moreover, incorporating normalization operations within the network typically improves the robustness and generalizability of the trained network when deployed in the field. The key idea proposed in Reference [13] is as follows:

- For each scalar activation function in the network, compute the sample mean and variables of the (mini-) batch of inputs,

- Normalize the activation functions scalar input x to have zero mean and unit variance,

- Apply a learnable linear transformation (i.e. , y = mx + b, where m and b are trainable) to the normalized scalar x (enabling the network to learn to adapt the normalization), and

- Apply the activation function o-(-) to y (i.e., compute o-(y)).

Deep learning has primarily focussed on real-valued neural networks. For example, the methods in Reference [13] only apply to real-valued deep neural networks.

The AEs used for CSI compression deal with complex-valued inputs and outputs (i.e., downlink channel estimates and their approximations).

Reference [14] considers batch-normalization for complex-valued deep neural networks with complex-value scalar activation functions a: C C. The author’s proposed to normalize the complex inputs so that each complex variable has a circularly symmetric zero-mean complex normal distribution (the mean and covariance being defined via sample averages over the mini-batch).

SUMMARY

As a part of developing embodiments herein the inventors identified a problem which first will be discussed. ML-based CSI reporting requires complex-valued autoencoders such as e.g. deep neural networks, to be trained and deployed on complex-valued channel features, e.g., UE estimates of the downlink CSI-RS channels.

To reduce training time, and improve compression performance and generalizability, it is important to select an appropriate data normalization method for the AE; that is, normalize the channel features before compressing them with the AE encoder.

In prior art [14] it is proposed to independently normalize each complex number so that its real and imaginary components are uncorrelated with zero mean and unit variance. However, normalizing channel features in this way destroys important amplitude and phase correlations in space (e.g., across antenna elements) and frequency (e.g., over sub-bands). For example,

1. The gNB cannot use the normalized CSI (e.g., the output of the AE decoder) without having additional information about the UE’s normalization; that is, the UE would need to signal to the gNB the normalization constants it used for each complex number as part of the CSI report.

2. Phase and amplitude correlations between elements of the complex channel matrix (in the antenna frequency domain) capture important physical information about the channel. In a Line of Sight (LoS) channel with no multipath, for example, the channel phase is linear in frequency (e.g., over subcarriers) and space (e.g., over a linear antenna array). The normalization technique proposed in [14] destroys amplitude and phase correlations between adjacent antenna elements and subbands and/or subbarriers, which removes the ability of the AE to learn (and exploit) redundant structures in the channel for compression purposes.

Another way to look at this problem is as follows:

Ideally, we would like the 3GPP CSI reporting framework to enable the UE to report the strongest (right) eigenvectors and eigenvalues of its CSI-RS based channel estimate to the gNB. However, it is not possible to signal quantized versions of these eigenvectors -- it would require excessive uplink overhead.

The Rel-16/17 CSI solutions attempt to strike a good balance between uplink overhead and approximating these eigenvectors and eigenvalues.

For Rel-18 Al-based solutions, many companies have proposed training the AEs to minimize the cosine similarity between the strongest eigenvector of the channel and the reconstructed eigenvector (e.g., for rank 1 transmissions). A problem here, however, is that these eigenvectors are uniquely defined only up to an arbitrary phase shift see, e.g., the review paper [11], Therefore, training an AE to reconstruct the strongest eigenvector (e.g., w.r.t. cosine similarity) will essentially require the AE to learn unlearnable arbitrary phase rotations of the strongest channel eigenvector.

An object of embodiments herein is to improve the performance of a wireless communications network using normalization of input data for an AE-based CSI feedback.

According to an aspect, the object is achieved by a method performed by a User Equipment, UE. The method is for normalizing input data to an Autoencoder, AE, used for Channel State Information, CSI, reporting to a network node in a wireless communications network. The UE estimates complex channel features based on downlink reference signals from the network node. The UE obtains a normalization constant A.

The UE normalizes the complex channel features to be used as input data to the AE. This is performed by: Based on the normalization constant A, the UE applies a phase rotation to all of the estimated complex channel features. This is to achieve an input data that is normalized such that a certain feature related to the input data attains a predetermined complex phase. The UE then feeds the normalized input data into the AE encoder. The output data of the AE encoder is comprised in a CSI report, to be sent to the network node.

According to another aspect, the object is achieved by a User Equipment, UE. The UE is configured to normalize input data to an Autoencoder, AE. The AE is used for Channel State Information, CSI, reporting to a network node in a wireless communications network. The UE is further configured to:

- Estimate complex channel features based on downlink reference signals from the network node,

- obtain a normalization constant A,

- normalize the complex channel features to be used as input data to the AE by: based on the normalization constant A, applying a phase rotation to all of the estimated complex channel features to achieve an input data that is normalized such that a certain feature related to the input data attains a predetermined complex phase, and - feed the normalized input data into the AE encoder, wherein the output data of the AE encoder is adapted to be comprised in a CSI report to be sent to the network node.

An advantage of embodiments herein is that they reduce training time, compression performance, and generalizability. In this way the performance of a wireless communications network using normalization of input data for an AE-based CSI feedback is improved.

BIEF DESCRIPTION OF THE DRAWINGS

Figure 1 is a schematic diagram depicting prior art.

Figure 2 is a schematic diagram depicting prior art.

Figure 3 is a schematic diagram depicting prior art.

Figure 4 is a schematic diagram depicting prior art.

Figure 5 is a schematic block diagram depicting embodiments of a wireless communications network.

Figure 6 is a flow chart depicting embodiments of a method performed by a UE. Figures 7 a and b are schematic block diagrams depicting embodiments of a UE. Figure 8 schematically illustrates a telecommunication network connected via an intermediate network to a host computer.

Figure 9 is a generalized block diagram of a host computer communicating via a base station with a user equipment over a partially wireless connection.

Figures 10 to 13 are flowcharts illustrating methods implemented in a communication system including a host computer, a base station, and a user equipment.

DETAILED DESCRIPTION

As mentioned above, the object of embodiments herein is to improve the performance of a wireless communications network using normalization of input data for an AE-based CSI feedback.

Embodiments herein provide a normalization method that e.g. applies a single common phase rotation to all features of the channel features, e.g., channel or eigenvectors. This is performed so that a fixed feature, e.g., a mean value or the first indexed feature, becomes real-valued, or attains some other predetermined complex phase.

This phase rotation simplifies training the AE because it removes arbitrary phase rotations in the complex channel features to be compressed.

Embodiments herein provide advantages such as e.g., they reduce training time, and improve compression performance and generalizability.

The provided normalization speeds up AE training, improves performance, e.g., channel reconstruction accuracy in the gNB, and improves robustness and generalizability. These gains are achieved by removing unnecessary uncertainty in the complex phase of the AE input, which reduces the difficulty of training. The normalization has no impact on downlink MU -Ml MO precoding, so it does not compromise system performance.

Figure 5 is a schematic overview depicting a wireless communications network 100 wherein embodiments herein may be implemented. The wireless communications network 100 comprises one or more RANs and one or more CNs. The wireless communications network 100 may use 5G NR but may further use a number of other different technologies, such as, Wi-Fi, (LTE), LTE-Advanced, Wideband Code Division Multiple Access (WCDMA), Global System for Mobile communications/enhanced Data rate for GSM Evolution (GSM/EDGE), or Ultra Mobile Broadband (UMB), just to mention a few possible implementations.

Network nodes, such as a network node 110, operate in the wireless communications network 100. The network node 110 may respectively e.g. provides a number of cells, and may use these cells for communicating with UEs, e.g. a UE 120. The network node 110 may respectively be a transmission and reception point e.g. a radio access network node such as a base station, e.g. a radio base station such as a NodeB, an evolved Node B (eNB, eNodeB, eNode B), an NR Node B (gNB), a base transceiver station, a radio remote unit, an Access Point Base Station, a base station router, a transmission arrangement of a radio base station, a stand-alone access point, a Wireless Local Area Network (WLAN) access point, an Access Point Station (AP STA), an access controller, a UE acting as an access point or a peer in a Device to Device (D2D) communication, or any other suitable networking unit. The network node 110, may respectively e.g. further be able to communicate with each other via one or more CN nodes in the wireless communications network 100.

User Equipments operate in the wireless communications network 100, such as a UE 120. The UE 120 may e.g. be an NR device, a mobile station, a wireless terminal, an NB-loT device, an enhanced Machine-type communication (eMTC) device, an NR RedCap device, a CAT-M device, a Wi-Fi device, an LTE device and a non-access point (non-AP) STA, a STA, that communicates via a base station such as e.g. the network node 110. It should be understood by the skilled in the art that the UE relates to a nonlimiting term which means any UE, terminal, wireless communication terminal, user equipment, (D2D) terminal, or node e.g. smart phone, laptop, mobile phone, sensor, relay, mobile tablets or even a small base station communicating within a cell.

Methods herein may be performed by network node 110. As an alternative, a Distributed Node (DN) and functionality, e.g. comprised in a cloud 140 as shown in Figure 5, may be used for performing or partly performing the methods of embodiments herein.

Some embodiments herein may be described in an example comprising the steps as follows, which steps may be taken in any suitable order:

A way to normalize input data to an AE used for CSI feedback may be formally defined by a 3GPP standard.

Step 1. The UE 120 estimates a complex channel feature using downlink reference signals (e.g., CSI-RS). For example, the UE 120 may e.g., estimate: a. The CSI-RS channel in the antenna-frequency domain, b. The Tx-covariance matrix for each sub and, c. The strongest (right) eigenvectors of the Tx-covariance matrix for each subband.

This relates to Action 601 described below.

Step 2. The UE 120 normalizes the data according to the description below, a. The data normalization is based on a normalization constant. This relates to Actions 602 and 603 described below. Step 3. The UE feeds the normalized data into an AE encoder. The output of the AE encoder is included in the UE’s 120 CSI report.

This relates to Action 604 described below.

Step 4. The UE 120 sends, also referred to as signals, the UE CSI report, also referred to as a CSI report, to the network node 110, e.g., a gNB, for example over an uplink control or data channel (PUCCH and/or PUSCH). a. Optionally: The UE 120 may signal the normalization constant, possibly quantized, to the network node 110. The normalization constant may be explicitly added to the CSI report and/or implicitly described by way of other parameters. b. From the Fourth Embodiment as mentioned below. Optionally: The UE 120 may signal the phase (possibly quantized) to the network node 110. The phase constant may be explicitly added to the CSI report and/or implicitly described by way of other parameters. c. From the Seventh embodiment as mentioned below. Optionally: The UE 120 signals which entry in the normalization-tensor codebook is used to the gNB. The index may be explicitly added to the CSI report and/or implicitly described by way of other parameters.

This relates to Actions 604 and 605 described below.

Tensors and related terminology

Some embodiments herein are described in terms of tensors. The tensor-related terminology is standard in the literature and explained/defined in, e.g., appendix A of [12],

Figure 6 shows an example method performed by the UE 120. The method is e.g. for normalizing input data to an AE used for CSI reporting to the network node 110 in the wireless communications network 100.

The method may comprise any one or more of the following actions. The actions may be executed in any suitable order. Actions that are optional are comprised in dashed boxes in Figure 6.

Action 601 The UE 120 estimates complex channel features based on downlink reference signals from the network node 110. The complex channel features are also referred to herein as features or input features, these wordings may be used interchangeably herein.

Complex channel features when used herein, e.g., means the complex channel coefficients of the radio channel, where the channel coefficients may relate to a timedomain or frequency-domain representation of the channel, and may also have been subject to some preprocessing step, e.g., conversion of per-antenna coefficients to DFT beam coefficients.

Complex channel features when used herein, may e.g., further mean one or more of the strongest (right) eigenvectors of the Tx-covariance matrix for each subband, that may or may not have been subject to pre-processing and/approximations, e.g. conversion of per-antenna coefficients to DFT beam coefficients, time-domain or frequency-domain representation, approximations using the Rayleigh-Ritz method [7], and truncation.

Action 602

The UE 120 obtains a normalization constant A. The UE 120 may use the normalization constant A later on (see below in Action 603), as a basis for normalizing the complex channel features to be used as input data to the AE. The normalization constant is referred to as A, but any symbol or letter may be used for the constant.

A normalization constant A when used herein e.g., means a computed complex number by which the features are multiplied to normalize them.

The 3GPP standard specifies a constant and/or fixed tensor C, which is referred to as a normalization tensor herein. A tensor when used herein e.g., means a multidimensional array of complex numbers, e.g., a Oth order tensor is a scalar, a 1st order is a vector, a 2nd order tensor is a matrix, and a 3rd order tensor is a set of complex numbers indexed by 3 indices. In some embodiments, the UE 120 obtains the normalization constant A by computing a tensor scalar product between the normalization tensor and the estimated complex channel features. In some other embodiments such as a fifth embodiment presented later below, the UE 120 obtains the normalization constant A by computing a tensor scalar product between the normalization tensor and a function of the estimated complex channel features. A tensor scalar product when used herein e.g., means the inner product, i.e. an elementwise multiplication followed by summation over all indices. A tensor scalar product between two tensors A and B will herein be denoted (A, B). The normalization tensor, may e.g. be defined by any one out of: - a parameter learned while training the AE, specified by the standard, taken from a set of allowed normalization tensors specified by the standard,

- computed based on the estimated complex channel features,

- computed based on a subset of the estimated complex channel features,

- configured by the network node 110, and/or

- configured by the UE 120.

The function of the estimated complex channel features e.g., means any function that takes the features as input and outputs a set of real or complex numbers, e.g. based on finding the feature or a set of features selected according to some rule that have the maximum norm.

In an example scenario, the normalization tensor is all 0 except one feature that is 1 , then it is an explicit feature of the with “complex channel features” that achieves a predetermined complex phase. But if the normalization tensor is e.g., a tensor with every element being 1/number of features, then it is actually the average of the “complex channel features” that achieves this phase and e.g. magnitude.

Action 603

The UE 120 then normalizes the complex channel features to be used as input data to the AE. The UE 120 normalizes the complex channel features based on the normalization constant A by applying a phase rotation, and possibly a magnitude scaling, to all of the estimated complex channel features. The normalization achieves an input data that is normalized such that a certain feature related to the input data attains a predetermined complex phase. In this way, the complex channel features to be fed into the AE encoder is compressed.

To apply a phase rotation to all of the estimated complex channel features e.g., means that each of the complex numbers representing the channel features is multiplied by a complex factor, such that a certain feature related to the input data attains a predetermined complex phase, e.g. with the desired phase. In some embodiments, the predetermined complex phase, is defined by any one out of:

- a parameter learned while training the AE,

- as specified by the standard,

- taken from a set of allowed phases, e.g. a codebook, specified by the standard,

- configured by the network node 110, and

- configured by the UE 120. A certain feature related to the input data means, e.g., an input feature, or a quantity calculated based on the input features according to some rule, e.g., a tensor scalar product between a normalization tensor and the “complex channel features” as described above. The certain feature related to the input data may comprise a fixed feature, of any one out of: a mean value feature, a first indexed feature, a last indexed feature, a largest norm feature, and/or a DFT beam feature.

A predetermined complex phase when used herein e.g., means an angle in the complex plane that is specified in the standard or signaled to the UE. In some embodiments, the UE 120 applies the phase rotation by any one or more out of:

- applying a single common phase rotation, and

- by further applying a magnitude rescaling to all of the estimated complex channel features.

The normalization may also affect the magnitude of the features as well. In some embodiments, the magnitude and phase normalization are combined.

The input data may be normalized such that it further achieves a unit norm. To achieve a unit norm e.g., means that the norm, e.g. the Euclidean norm, the Frobenius norm, or some other norm, is 1.0. This is an advantage since many neural networks may learn faster and generalize better if inputs are normalized to approximately unit norm.

In some embodiments, the certain feature is actually given and/or defined by the choice of the normalization tensor.

In some embodiments, the input data is normalized such that a function of the estimated complex channel features further achieves a unit norm. This embodiment may relate to a second part of the fifth embodiment presented later below. E.g. this may be an alternative to have unit norm on the input data.

Action 604

The UE 120 feeds the normalized input data into the AE encoder. The output data of the AE encoder is comprised in a CSI report, e.g. a UE CSI report, to be sent to the network node 110.

In some embodiments, the normalization constant A is part of the CSI report, e.g. the UE CSI report, to be sent to the network node 110. E.g. the network node 110 wants to know how the normalization was done, to be able to undo it.

Action 605 The UE 120 sends the output data of the AE encoder in a UE CSI report to the network node 110.

In this way, by using the above method, the training of the AE will be faster, the performance is improved, e.g., channel reconstruction accuracy in the network node 110, e.g. a gNB, and robustness as well as generalizability is improved.

The methods will now be further explained and exemplified in below embodiments. The embodiments below may be combined in any suitable manner with the embodiments described above.

Some First Embodiments

The normalization, as mentioned in Action 603 above, may be applied after any preprocessing steps are done, and before the features are fed into the AE. As mentioned above, the normalization process comprises that based on the normalization constant X, the UE 120 applies a phase rotation to all of the estimated complex channel features to achieve an input data that is normalized such that a certain feature related to the input data attains a predetermined complex phase.

The normalization process may be described as follows:

The 3GPP standard specifies a constant and/or fixed tensor C, it is referred to as a normalization tensor herein. The normalization tensor may be of the same dimension as the features of interest. The features of interest when used herein may mean e.g., the original complex channel features, or the input features after some pre-processing step applied to the complex channel features.

The CSI report may comprise different types of feedback, such as e.g., full-channel feedback and precoding- vector feedback.

In some embodiments, the UE 120 sends back, e.g. to the network node 110, a whole channel, also referred to as a full-channel feedback, as measured on CSI-RS. In such embodiment the complex channel features is the complete measured channel, and all elements of this may be normalized with the same normalization constant. According to some other embodiments, the UE 120 instead sends back explicit precoding vectors, also referred to as a precoding-vector feedback, that it suggests the network node 110 to use. This report would be one matrix of precoding vectors, Label: XYZ, wherein the first index may correspond to something like the network node’s 110 Tx- port, and the second index may be over some frequency-granularity. The processing of these features by the network node 110 is transparent to the UE 120, in the sense that the network node 110 does not explicitly tell the UE 120 how it processed the feedback or how the feedback affected decisions taken by the network node. In a request for multiple transmission layers, the UE 120 then may report multiple of these matrices in a CSI report, one for each layer. Then, there may be multiple ways to consider the normalization 603:

- In some of these embodiments the UE 120 normalizes each precoding vector independently. In these embodiments, the normalization tensor may be a vector, and the complex channel features may e.g. be each precoding vector.

- In some other of these embodiments a lower overhead is wanted, e.g. if the normalization is also feed back to the network node 110. This may be achieved by the UE 120 doing the normalization either over a few vectors, or over the whole matrix of precoding vectors (XYZ). In this embodiment the normalization tensor is a matrix of corresponding dimensions.

- In another embodiment, the normalization 603 may be applied, by the UE 120, jointly to all transmission layers, i.e. jointly over all matrices (XYZ) send back. Then the normalization tensor is an order-3 tensor of corresponding dimensions. Normalization based on only a subset of the matrices (but still applied to all), may be implemented by, e.g., filling the normalization tensor with zeros in appropriate places.

The normalization tensor C may be used by the UE 120 for obtaining 602 the normalization constant A. The normalization 603 of the complex channel features to be used as input data to the AE will be based on the normalization constant A.

The normalization tensor C may be defined as follows.

Full-channel feedback

- In the case of full-channel feedback, the normalization tensor C has the same dimensionality and size as the measured/estimated channel to be compressed by the UE 120, possibly after some pre-processing step done by the UE 120. Related to the 3GPP standard that may be a 3-dimensional tensor, also known as a 3rd order tensor and/or order-3 tensor:

If the pre-processing done by the UE 120 does not change the dimension of the measurements, then the dimensions are number of configured CSI-RS ports x number of UE antenna ports x the frequency granularity of the measured/estimated channel. Note: any permutation of the CSI-RS port, UE antenna port, and frequency granularity is possible. The frequency granularity may alternatively be time granularity.

If there are pre-processing steps defined that alters the dimensions of the measured/estimated channel, then the dimensions of the normalization tensor C may be defined by the dimensions of the output of such process. One example of such processing is described in [10],

Precoding-vector feedback

- In the case of precoding-vector feedback, the normalization tensor C may have the same dimension as the individual precoding vectors. Related to the 3GPP standard that may be a vector:

If the pre-processing does not change the dimension of the measurements, then the vector may be of dimension Tx virtual ports of the network node 110 times the number of features per Tx virtual port

If there are pre-processing steps defined that alters the dimensions of the calculated, estimated, and/or approximated precoding vector, then the dimensions of the normalization tensor C is be defined by the dimensions of the output of such process. One example of such processing is the beam-reduction described for CSI Type II above.

Examples of how to define the normalization tensor C may comprise:

- A normalization tensor C with all features equal to 1.

- A normalization tensor C with all features equal to 1/number of features in the normalization tensor C.

- A normalization tensor C with all features equal to 0, except one feature that is equal to 1.

- An Nth order normalization tensor C corresponding to the outer product of N DFT vectors (same or different).

The input complex channel features before normalizing it may be denoted U.

- In some embodiments relating to full-channel feedback U=H is the measured channel tensor or a pre-processed entity derived based on it. - In some embodiments relating to precoding-vector feedback the II is one of the suggested precoding vectors, or a pre-processed entity derived based on it.

The input complex channel features when being normalized may be denoted U, thus U is the normalized complex channel features used as input to the AE. The U, may be defined as U = AU.

The computations performed by the UE 120 may comprise:

Let II denote the complex-valued the input feature to the AE.

Compute the normalization constant A as:

U = AU. Where U is the normalized complex channel features to be used as input to the AE.

The UE 120 may compute the normalization constant, A, as referenced above, by computing the tensor scalar product, denoted with angled brackets {•,•), between the normalization tensor and the estimated features that should be used as input to the AE.

In the case of precoding-vector feedback, the above process, as described above, is applied independently for each precoding vector. Hence, there will be one normalization constant A computed for each precoding vector and these may be different for different frequencies. In NR this corresponds to different CSI-RS in different PRBs. However, the normalization may also be averaged over larger frequency blocks, or based on and applied to whole matrix of precoding vectors (XYZ), as mentioned above. This reduces the overhead if the UE 120 sends back the normalization constant as well.

Some embodiments herein are described in terms of the tensor scalar product. A person skilled in the art should also appreciate that other weightings of this inner product can also be used.

Some Second Embodiments

The complex tensor C, that is the normalization tensor C defined in the First Embodiments may be defined separately for each polarization in the channel. Some Third Embodiments

In some embodiments, an additional learnable complex linear transformation is added applied by the UE 120 after the normalization 603 in some of the First Embodiments. For example,

U = m (£7) + b

Where m e (C and b e (C are trainable complex values and U is input to the AE.

Some Fourth Embodiments

The normalization 603 defined in the First Embodiment may be performed by the UE 120 such that the value to normalize, i.e. , the inner product between the complex channel features and the normalization tensor C and thus implicitly defined by the choice of the normalization tensor C, gets a fixed complex phase 0 e R that is different from 0. It should be noted that the sixth embodiment extends and/or modifies this. The last step in the normalization 603 computation, in the First Embodiments may then be changed to:

U = (e ^ie )U.

The phase 6 may e.g. be defined by the 3GPP standard, chosen from a fixed set of allowed phases, or determined by the UE 120. In the two latter cases the phase may optionally be signaled by the UE 120 to the network node 110.

Alternatively:

This embodiment is related to the Third Embodiments described above, but the values m e (C and b e (C defined in the Third Embodiments are in this fourth embodiment fixed to b = 0 and m = e ^ie , for some value of 6 e I This will ensure that the value to normalize, i.e. the inner product between the complex channel features and the normalization tensor C and thus implicitly defined by the choice of the normalization tensor C, has a fixed complex phase 6. Note that the sixth embodiment goes beyond this, to normalize.

The phase 6 may be defined as a constant by the 3GPP standard, chosen from a fixed set of allowed phases, determined by the UE 120, or a trainable parameter when designing/training the AE. In the three latter cases the phase may optionally be signaled by the UE 120 to the network node 110.

Some Fifth Embodiments The normalization 603 may in these fifth embodiments be derived by the UE 120 based on a quantity that is computed (by the UE 120) based on the input to the neural network, and possibly other quantities, but not directly fed into the neural network. For example, the UE 120 may define where f is some function. For more concrete examples, see the Sixth Embodiment.

Some Sixth Embodiments

The complex tensor C defined in the First Embodiments is not a constant that is explicitly specified by 3GPP. However, 3GPP specifies the way to compute a normalization tensor C, which may be computed by the UE 120.

One example is that the normalization 603 is applied by the UE 120 on the complexvalued input feature U is derived to normalize to the largest entry of the complex-valued the input feature U. In other words, the complex channel features to normalize, implicitly defined by the choice of the normalization tensor C, may be the, in magnitude, largest value of the input feature U. This is achieved by the UE 120, by using a normalization tensor C with all features being zeros except for a single feature that is equal to 1 . The index of that single feature may be the same as the index of the, in magnitude, largest entry of U.

In another example (somewhat more general), the normalization 603 is derived by the UE 120, based on a the largest of a set of quantities, where each quantity is computed based on the one or more input complex channel features. For example, if the set of input complex channel features are the K channel coefficients for K subcarriers for a set of M beam directions (i.e. M x K coefficients in total), then the UE 120 may, for each beam m, compute an inner product between the K coefficients of a beam and a K-feature complex tensor C, and then normalize 603 using the inner product (out of the M inner products) that has the largest absolute value. For another example, if the input features represent the channel coefficients in time domain (a tap-delay line), the UE 120 may compute values based on a sliding window of e.g. length 2 to the set of features and then select the value with the largest norm to apply the normalization.

Some Seventh Embodiments

In these embodiments, the normalization tensor C defined in the First Embodiments is taken from a codebook of normalization tensors. The codebook is defined by 3GPP, and the normalization tensor used may be optionally signalled by the UE 120 to the network node 110. See (c) in Step 4 described above.

3GPP may also specify how the normalization tensor C should be selected from the codebook. The UE 120 may perform the selection of normalization tensor C accordingly.

Some Eighth Embodiments

The overall phase of a channel would typically not affect the choice of transmission precoder, and hence the normalization constant would not have to be communicated by the UE 120 (reported 605) to the transmitter, i.e. the network node 110. However, in case there are multiple reports of partial channel information, e.g., if one report relates to the strongest eigenbeam and another report relates to the second strongest eigenbeam, then the relative phase of those two partial report components (henceforth partial reports for brevity) does matter and may have to be made known to the transmitter, i.e., the network node 110. There are several ways to achieve this:

- The UE 120 may in each partial CSI report include 604 and information about the used phase normalization constant.

- The UE 120 may in each partial report include 604 and information about the used phase normalization constant’s phase relative to some earlier report(s).

For the first partial report in a set of partial reports, the phase normalization constant may then not have to be reported.

- The UE 120 may use 603 the same normalization constant for all partial reports in a set of partial reports.

This normalization constant may be obtained 602 by being calculated by the UE 120 jointly based on all the channel information to be reported 604 in the set of partial reports (e.g., all eigenbeams to be reported).

This normalization constant may also be obtained 602 by being calculated based on a subset of the channel information belonging to a subset of the partial reports (e.g., normalization is determined based on a few of the strongest eigenbeams but applied to more beams to be reported).

Figure 7a and 7b shows an example of arrangement in the UE 120.

The UE 120 may comprise an input and output interface configured to communicate with other networking entities in the wireless communications network 100, e.g., the network node 110. The input and output interface may comprise a receiver, e.g., wired and/or wireless, (not shown) and a transmitter, e.g. wired and/or wireless, (not shown).

The UE 120 may comprise any one or more out of: An obtaining unit, an estimating unit, a normalizing unit, an applying unit, and a feeding unit to perform the method actions as described herein.

The embodiments herein may be implemented through a processor or one or more processors, such as at least one processor of a processing circuitry in the UE 120 depicted in Figure 7a, together with computer program code for performing the functions and actions of the embodiments herein. The program code mentioned above may also be provided as a computer program product, for instance in the form of a data carrier carrying computer program code for performing the embodiments herein when being loaded into the UE 120. One such carrier may be in the form of a CD ROM disc. It is however feasible with other data carriers such as a memory stick. The computer program code may furthermore be provided as pure program code on a server and downloaded to the UE 120.

The UE 120 may further comprise respective a memory comprising one or more memory units. The memory comprises instructions executable by the processor in the UE 120. The memory is arranged to be used to store instructions, data, configurations, and applications to perform the methods herein when being executed in the UE 120.

In some embodiments, a computer program comprises instructions, which when executed by the at least one processor, cause the at least one processor of the UE 120 to perform the actions above.

In some embodiments, a respective carrier comprises the respective computer program, wherein the carrier is one of an electronic signal, an optical signal, an electromagnetic signal, a magnetic signal, an electric signal, a radio signal, a microwave signal, or a computer-readable storage medium.

Those skilled in the art will also appreciate that the functional modules in the UE 120, described below may refer to a combination of analog and digital circuits, and/or one or more processors configured with software and/or firmware, e.g. stored in the UE 120, that when executed by the respective one or more processors such as the at least one processor described above cause the respective at least one processor to perform actions according to any of the actions above. One or more of these processors, as well as the other digital hardware, may be included in a single Application-Specific Integrated Circuitry (ASIC), or several processors and various digital hardware may be distributed among several separate components, whether individually packaged or assembled into a system-on-a-chip (SoC).

When using the word "comprise" or “comprising” it shall be interpreted as nonlimiting, i.e., meaning "consist at least of".

The embodiments herein are not limited to the preferred embodiments described herein. Various alternatives, modifications and equivalents may be used.

It should be noted that the order of the First-Eighth Embodiments mentioned above is not related to the order of the embodiments 1-22 below. Each of the First-Eighth Embodiments may relate to and be combined with any suitable embodiment out of the embodiments 1-22 below.

Embodiments

Below, some example Embodiments 1-22 are shortly described. See e.g. Figures 5, 6, 7a, and 7b.

It should be noted that the Embodiments 1-22 are e.g., related to and may be combined with any suitable embodiments of the first to eighth embodiments described above.

Embodiment 1. A method performed by a User Equipment, UE, 120, e.g. for normalizing input data to an Autoencoder, AE, used for Channel State Information, CSI, reporting to a network node 110 in a wireless communications network 100, the method comprising any one or more out of: estimating 601 complex channel features based on downlink reference signals from the network node 110, obtaining 602 a normalization constant A, normalizing 603 the complex channel features to be used as input data to the AE by: based on the normalization constant A, applying a phase rotation to all of the estimated complex channel features to achieve an input data that is normalized such that a certain feature related to the input data attains a predetermined complex phase, feeding 604 the normalized input data into the AE encoder, wherein the output data of the AE encoder is comprised in a CSI report, e.g. a UE CSI report, to be sent to the network node 110.

Embodiment 2. The method according to Embodiment 1 , wherein the normalization constant A is part of the CSI report, e.g. the UE CSI report, to be sent to the network node 110. E.g. the network node 110 wants to know how the normalization was done, to be able to undo it.

Embodiment 3. The method according to any of the Embodiments 1-2, wherein: the obtaining 602 of the normalization constant A, is performed by computing a tensor scalar product between a normalization tensor and the estimated complex channel features.

Embodiment 4. The method according to any of the Embodiments 1-2, wherein: the obtaining 602 of the normalization constant A, is performed by computing a tensor scalar product between a normalization tensor and a function of the estimated complex channel features. See, e.g. Embodiment 5.

Embodiment 5. The method according to any of the Embodiments 3-4, where the normalization tensor, is be defined by any one out of:

- a parameter learned while training the AE, specified by the standard, taken from a set of allowed normalization tensors specified by the standard,

- computed based on the estimated complex channel features,

- computed based on a subset of the estimated complex channel features,

- configured by the network node 110, or

- configured by the UE 120. Embodiment 6. The method according to any of the Embodiments 1-5, where the predetermined complex phase, is defined by any one out of:

- a parameter learned while training the AE,

- as specified by the standard,

- taken from a set of allowed phases, e.g. a codebook, specified by the standard,

- configured by the network node 110,

- configured by the UE 120.

Embodiment 7. The method according to any of the Embodiments 1-6, wherein the applying of the phase rotation comprises any one or more out of:

- applying single common phase rotation, and

- by further applying a magnitude rescaling to all of the estimated complex channel features.

Embodiment 8. The method according to any of the Embodiments 1-7, wherein the input data is normalized such that it further achieves a unit norm.

Embodiment 9. The method according to Embodiment 1-7, wherein the input data is normalized such that a function of the estimated complex channel features further achieves a unit norm. See Second part” of Embodiment 5. An alternative to have unit norm on the input data.

Embodiment 10. The method according to any of the Embodiments 1-9, where the certain feature related to the input data comprises a fixed feature, of any one out of: a mean value feature, a first indexed feature, a last indexed feature, a largest norm feature, or a DFT beam feature.

Embodiment 11. A computer program comprising instructions, which when executed by a processor, causes the processor to perform actions according to any of the Embodiments 1-10.

Embodiment 12. A carrier comprising the computer program of Embodiment 11 , wherein the carrier is one of an electronic signal, an optical signal, an electromagnetic signal, a magnetic signal, an electric signal, a radio signal, a microwave signal, or a computer-readable storage medium. Embodiment 13. A User Equipment, UE, 120, e.g. configured to normalize input data to an Autoencoder, AE, used for Channel State Information, CSI, reporting to a network node 110 in a wireless communications network 100, the UE 120 e.g., being further configured to any one or more out of: estimate, e.g. by means of an estimating unit comprised in the UE 120, complex channel features based on downlink reference signals from the network node 110, obtain, e.g. by means of an obtaining unit comprised in the UE 120, a normalization constant A, normalize, e.g. by means of an normalizing unit comprised in the UE 120, the complex channel features to be used as input data to the AE by: based on the normalization constant A, applying, e.g. by means of an applying unit comprised in the UE 120, a phase rotation to all of the estimated complex channel features to achieve an input data that is normalized such that a certain feature related to the input data attains a predetermined complex phase, feed, e.g. by means of an feeding unit comprised in the UE 120, the normalized input data into the AE encoder, wherein the output data of the AE encoder is adapted to be comprised in a CSI report, e.g. a UE CSI report, to be sent to the network node 110.

Embodiment 14. The UE 120 according to Embodiment 13, wherein the normalization constant A is adapted to be part of the CSI report, e.g. a UE CSI report, to be sent to the network node 110. E.g. the network node 110 wants to know how the normalization was done, to be able to undo it.

Embodiment 15. The UE 120 according to any of the Embodiments 13-14, further configured to obtain, e.g. by means of the obtaining unit comprised in the UE 120, the normalization constant A by computing a tensor scalar product between a normalization tensor and the estimated complex channel features.

Embodiment 16. The UE 120 according to any of the Embodiments 13-15, further configured to obtain, e.g. by means of the obtaining unit comprised in the UE 120, the normalization constant A, by computing a tensor scalar product between a normalization tensor and a function of the estimated complex channel features. See, e.g. Embodiment 5. Embodiment 17. The UE 120 according to any of the Embodiments 15-16, where the normalization tensor is arranged to be defined by any one out of:

- a parameter learned while training the AE, specified by the standard, taken from a set of allowed normalization tensors specified by the standard,

- computed based on the estimated complex channel features,

- computed based on a subset of the estimated complex channel features,

- configured by the network node 110, or

- configured by the UE 120.

Embodiment 18. The UE 120 according to any of the Embodiments 13-17, where the predetermined complex phase is arranged to be defined by any one out of:

- a parameter learned while training the AE,

- as specified by the standard,

- taken from a set of allowed phases, e.g. a codebook, specified by the standard,

- configured by the network node 110,

- configured by the UE 120.

Embodiment 19. The UE 120 according to any of the Embodiments 13-18, further being configures to apply the phase rotation e.g. by means of the applying unit comprised in the UE 120, according to any one or more out of:

- by applying a single common phase rotation, and

- by further apply a magnitude rescaling to all of the estimated complex channel features.

Embodiment 20. The UE 120 according to any of the Embodiments 13-19, wherein the input data is adapted to be normalized such that it further achieves a unit norm.

Embodiment 21. The UE 120 according to Embodiment 13-19, wherein the input data is adapted to be normalized such that a function of the estimated complex channel features further achieves a unit norm. See Second part” of Embodiment 5. An alternative to have unit norm on the input data. Embodiment 22. The UE 120 according to any of the Embodiments 13-21 , where the certain feature related to the input data is adapted to comprise a fixed feature, of any one out of: a mean value feature, a first indexed feature, a last indexed feature, a largest norm feature, or a DFT beam feature.

Further Extensions and Variations

With reference to Figure 8, in accordance with an embodiment, a communication system includes a telecommunication network 3210 such as the wireless communications network 100, e.g. an loT network, or a WLAN, such as a 3GPP-type cellular network, which comprises an access network 3211 , such as a radio access network, and a core network 3214. The access network 3211 comprises a plurality of base stations 3212a, 3212b, 3212c, such as the network node 110, access nodes, AP STAs NBs, eNBs, gNBs or other types of wireless access points, each defining a corresponding coverage area 3213a, 3213b, 3213c. Each base station 3212a, 3212b, 3212c is connectable to the core network 3214 over a wired or wireless connection 3215. A first user equipment (UE) e.g. the UE 120 such as a Non-AP STA 3291 located in coverage area 3213c is configured to wirelessly connect to, or be paged by, the corresponding base station 3212c. A second UE 3292 e.g. the UE 120 such as a Non-AP STA in coverage area 3213a is wirelessly connectable to the corresponding base station 3212a. While a plurality of UEs 3291, 3292 are illustrated in this example, the disclosed embodiments are equally applicable to a situation where a sole UE is in the coverage area or where a sole UE is connecting to the corresponding base station 3212.

The telecommunication network 3210 is itself connected to a host computer 3230, which may be embodied in the hardware and/or software of a standalone server, a cloud- implemented server, a distributed server or as processing resources in a server farm. The host computer 3230 may be under the ownership or control of a service provider, or may be operated by the service provider or on behalf of the service provider. The connections 3221, 3222 between the telecommunication network 3210 and the host computer 3230 may extend directly from the core network 3214 to the host computer 3230 or may go via an optional intermediate network 3220. The intermediate network 3220 may be one of, or a combination of more than one of, a public, private or hosted network; the intermediate network 3220, if any, may be a backbone network or the Internet; in particular, the intermediate network 3220 may comprise two or more sub-networks (not shown). The communication system of Figure 8 as a whole enables connectivity between one of the connected UEs 3291, 3292 and the host computer 3230. The connectivity may be described as an over-the-top (OTT) connection 3250. The host computer 3230 and the connected UEs 3291, 3292 are configured to communicate data and/or signaling via the OTT connection 3250, using the access network 3211, the core network 3214, any intermediate network 3220 and possible further infrastructure (not shown) as intermediaries. The OTT connection 3250 may be transparent in the sense that the participating communication devices through which the OTT connection 3250 passes are unaware of routing of uplink and downlink communications. For example, a base station 3212 may not or need not be informed about the past routing of an incoming downlink communication with data originating from a host computer 3230 to be forwarded e.g., handed over to a connected UE 3291. Similarly, the base station 3212 need not be aware of the future routing of an outgoing uplink communication originating from the UE 3291 towards the host computer 3230.

Example implementations, in accordance with an embodiment, of the UE, base station and host computer discussed in the preceding paragraphs will now be described with reference to Figure 9. In a communication system 3300, a host computer 3310 comprises hardware 3315 including a communication interface 3316 configured to set up and maintain a wired or wireless connection with an interface of a different communication device of the communication system 3300. The host computer 3310 further comprises processing circuitry 3318, which may have storage and/or processing capabilities. In particular, the processing circuitry 3318 may comprise one or more programmable processors, application-specific integrated circuits, field programmable gate arrays or combinations of these (not shown) adapted to execute instructions. The host computer 3310 further comprises software 3311, which is stored in or accessible by the host computer 3310 and executable by the processing circuitry 3318. The software 3311 includes a host application 3312. The host application 3312 may be operable to provide a service to a remote user, such as a UE 3330 connecting via an OTT connection 3350 terminating at the UE 3330 and the host computer 3310. In providing the service to the remote user, the host application 3312 may provide user data which is transmitted using the OTT connection 3350.

The communication system 3300 further includes a base station 3320 provided in a telecommunication system and comprising hardware 3325 enabling it to communicate with the host computer 3310 and with the UE 3330. The hardware 3325 may include a communication interface 3326 for setting up and maintaining a wired or wireless connection with an interface of a different communication device of the communication system 3300, as well as a radio interface 3327 for setting up and maintaining at least a wireless connection 3370 with a UE 3330 located in a coverage area (not shown) served by the base station 3320. The communication interface 3326 may be configured to facilitate a connection 3360 to the host computer 3310. The connection 3360 may be direct or it may pass through a core network (not shown) in Figure 9 of the telecommunication system and/or through one or more intermediate networks outside the telecommunication system. In the embodiment shown, the hardware 3325 of the base station 3320 further includes processing circuitry 3328, which may comprise one or more programmable processors, application-specific integrated circuits, field programmable gate arrays or combinations of these (not shown) adapted to execute instructions. The base station 3320 further has software 3321 stored internally or accessible via an external connection.

The communication system 3300 further includes the UE 3330 already referred to. Its hardware 3335 may include a radio interface 3337 configured to set up and maintain a wireless connection 3370 with a base station serving a coverage area in which the UE 3330 is currently located. The hardware 3335 of the UE 3330 further includes processing circuitry 3338, which may comprise one or more programmable processors, applicationspecific integrated circuits, field programmable gate arrays or combinations of these (not shown) adapted to execute instructions. The UE 3330 further comprises software 3331, which is stored in or accessible by the UE 3330 and executable by the processing circuitry 3338. The software 3331 includes a client application 3332. The client application 3332 may be operable to provide a service to a human or non-human user via the UE 3330, with the support of the host computer 3310. In the host computer 3310, an executing host application 3312 may communicate with the executing client application 3332 via the OTT connection 3350 terminating at the UE 3330 and the host computer 3310. In providing the service to the user, the client application 3332 may receive request data from the host application 3312 and provide user data in response to the request data. The OTT connection 3350 may transfer both the request data and the user data. The client application 3332 may interact with the user to generate the user data that it provides.

It is noted that the host computer 3310, base station 3320 and UE 3330 illustrated in Figure 9 may be identical to the host computer 3230, one of the base stations 3212a, 3212b, 3212c and one of the UEs 3291, 3292 of Figure 8, respectively. This is to say, the inner workings of these entities may be as shown in Figure 9 and independently, the surrounding network topology may be that of Figure 8.

In Figure 9, the OTT connection 3350 has been drawn abstractly to illustrate the communication between the host computer 3310 and the use equipment 3330 via the base station 3320, without explicit reference to any intermediary devices and the precise routing of messages via these devices. Network infrastructure may determine the routing, which it may be configured to hide from the UE 3330 or from the service provider operating the host computer 3310, or both. While the OTT connection 3350 is active, the network infrastructure may further take decisions by which it dynamically changes the routing e.g., on the basis of load balancing consideration or reconfiguration of the network.

The wireless connection 3370 between the UE 3330 and the base station 3320 is in accordance with the teachings of the embodiments described throughout this disclosure. One or more of the various embodiments improve the performance of OTT services provided to the UE 3330 using the OTT connection 3350, in which the wireless connection 3370 forms the last segment. More precisely, the teachings of these embodiments may improve the applicable RAN effect: data rate, latency, power consumption, and thereby provide benefits such as corresponding effect on the OTT service: e.g. reduced user waiting time, relaxed restriction on file size, better responsiveness, extended battery lifetime.

A measurement procedure may be provided for the purpose of monitoring data rate, latency and other factors on which the one or more embodiments improve. There may further be an optional network functionality for reconfiguring the OTT connection 3350 between the host computer 3310 and UE 3330, in response to variations in the measurement results. The measurement procedure and/or the network functionality for reconfiguring the OTT connection 3350 may be implemented in the software 3311 of the host computer 3310 or in the software 3331 of the UE 3330, or both. In embodiments, sensors (not shown) may be deployed in or in association with communication devices through which the OTT connection 3350 passes; the sensors may participate in the measurement procedure by supplying values of the monitored quantities exemplified above, or supplying values of other physical quantities from which software 3311, 3331 may compute or estimate the monitored quantities. The reconfiguring of the OTT connection 3350 may include message format, retransmission settings, preferred routing etc.; the reconfiguring need not affect the base station 3320, and it may be unknown or imperceptible to the base station 3320. Such procedures and functionalities may be known and practiced in the art. In certain embodiments, measurements may involve proprietary UE signaling facilitating the host computer’s 3310 measurements of throughput, propagation times, latency and the like. The measurements may be implemented in that the software 3311, 3331 causes messages to be transmitted, in particular empty or ‘dummy’ messages, using the OTT connection 3350 while it monitors propagation times, errors etc.

Figure 10 is a flowchart illustrating a method implemented in a communication system, in accordance with one embodiment. The communication system includes a host computer, a base station such as the network node 110, and a UE such as the UE 120, which may be those described with reference to Figure 8 and Figure 9. For simplicity of the present disclosure, only drawing references to Figure 10 will be included in this section. In a first action 3410 of the method, the host computer provides user data. In an optional sub action 3411 of the first action 3410, the host computer provides the user data by executing a host application. In a second action 3420, the host computer initiates a transmission carrying the user data to the UE. In an optional third action 3430, the base station transmits to the UE the user data which was carried in the transmission that the host computer initiated, in accordance with the teachings of the embodiments described throughout this disclosure. In an optional fourth action 3440, the UE executes a client application associated with the host application executed by the host computer.

Figure 11 is a flowchart illustrating a method implemented in a communication system, in accordance with one embodiment. The communication system includes a host computer, a base station such as an AP STA, and a UE such as a Non-AP STA which may be those described with reference to Figure 8 and Figure 9. For simplicity of the present disclosure, only drawing references to Figure 11 will be included in this section. In a first action 3510 of the method, the host computer provides user data. In an optional sub action (not shown) the host computer provides the user data by executing a host application. In a second action 3520, the host computer initiates a transmission carrying the user data to the UE. The transmission may pass via the base station, in accordance with the teachings of the embodiments described throughout this disclosure. In an optional third action 3530, the UE receives the user data carried in the transmission.

Figure 12 is a flowchart illustrating a method implemented in a communication system, in accordance with one embodiment. The communication system includes a host computer, a base station such as an AP STA, and a UE such as a Non-AP STA which may be those described with reference to Figure 8 and Figure 9. For simplicity of the present disclosure, only drawing references to Figure 12 will be included in this section. In an optional first action 3610 of the method, the UE receives input data provided by the host computer. Additionally or alternatively, in an optional second action 3620, the UE provides user data. In an optional sub action 3621 of the second action 3620, the UE provides the user data by executing a client application. In a further optional sub action 3611 of the first action 3610, the UE executes a client application which provides the user data in reaction to the received input data provided by the host computer. In providing the user data, the executed client application may further consider user input received from the user. Regardless of the specific manner in which the user data was provided, the UE initiates, in an optional third sub action 3630, transmission of the user data to the host computer. In a fourth action 3640 of the method, the host computer receives the user data transmitted from the UE, in accordance with the teachings of the embodiments described throughout this disclosure.

Figure 13 is a flowchart illustrating a method implemented in a communication system, in accordance with one embodiment. The communication system includes a host computer, a base station such as an AP STA, and a UE such as a Non-AP STA which may be those described with reference to Figure 8 and Figure 9. For simplicity of the present disclosure, only drawing references to Figure 13 will be included in this section. In an optional first action 3710 of the method, in accordance with the teachings of the embodiments described throughout this disclosure, the base station receives user data from the UE. In an optional second action 3720, the base station initiates transmission of the received user data to the host computer. In a third action 3730, the host computer receives the user data carried in the transmission initiated by the base station.

Definitions of some abbreviations and acronyms used herein.

Abbreviation _ Explanation

AE Autoencoder (see Figure 3)

Al Artificial Intelligence

BS Base station

CSI Channel State Information

DFT Discrete Fourier Transform MIMO Multiple-input multiple-output (channel)

ML Machine learning

MU-MIMO Multi-user MIMO

NN Neural network

NW Network

PM I Precoding Matrix Indicator

RS Reference signal

Rx Receiver

Tx Transmitter

UE User equipment

References:

1. 3GPP TS 38.214 “Physical layer procedures for data (Release 16)”

2. RP-213599, “Study on Artificial Intelligence (Al)/Machine Learning (ML) for NR Air Interface", December 2021

3. Zhilin Lu, Xudong Zhang, Hongyi He, Jintao Wang, and Jian Song, “Binarized Aggregated Network with Quantization: Flexible Deep Learning Deployment for CSI Feedback in MassiveMIMO System”, arXiv, 2105.00354 v1, May, 2021.

4. RWS-210024, “Rel.18 Network AI/ML,” QUALCOMM, TSG RAN Rel-18 workshop, June 28 - July 2, 2021.

5. J. Duchi, E. Hazan, and Y. Singer, “Adaptive subgradient methods for online learning and stochastic optimization”, Journal of Machine Learning Research, 2010.

6. D. Kingma and J. Ba, “A method for stochastic optimization,” arXiv, 1412.6980, Dec 2014.

7. Trefethen, Lloyd N.; Bau, III, David (1997). Numerical Linear Algebra. SIAM.

11. Anne Greenbaum, Ren-Cang Li, and Michael L. Overton, “First-Order Perturbation Theory for Eigenvalues and Eigenvectors”, SIAM Review, 2020

12. Nelson Costa and Simon Haykin, Multiple-Input, Multiple-Output Channel Models: Theory and Practice, John Wiley & Sons, 2010 (Appendix A, which is the one referenced, is openly available from the publisher.)

13. S. Loffe and C. Szegdy, “Batch normalization: Accelerated deep network training by reducing internal covariance shift,” arXiv 1502.03167, March, 2015.

14. C. Trabelsi, O. Bilaniuk, Y. Zhang, D. Serdyuk, S. Subramanian, J. F. Santos, S. Mehri, N. Rostamzadeh, Y. Bengio, and C. J. Pa, “Deep complex networks,” arXiv 1705.09792, 2018.