DATA-DRIVEN END-TO-END CSI ACQUISITION USING UPLINK AND DOWNLINK MEASUREMENTS

Technical Field

The present disclosure relates to channel state information acquisition and estimation, and specifically to estimation with a jointly optimized machine learning model at both a user equipment device and a network node in a wireless communication network.

Backqround

Beamforming/Precoding

The 5G downlink beamforming and precoding capabilities require base stations (BSs) to acquire Channel State Information (CSI). The Third Generation Partnership Project (3GPP) specifies the following procedures to acquire downlink CSI between a BS and a User Equipment (UE).

1) CSI acquisition based on sounding reference signal (SRSs): As illustrated in Figure 1, the BS or network node 104 utilizes the CSI estimated from the uplink-transmitted SRSs from the User Equipment device (UE) 102 to design the precoders (a.k.a. reciprocity-based precoding). In this approach, the BS 104 leverages the uplink-downlink channel reciprocity that time division duplex (TDD) systems exhibit.

2) CSI acquisition based on downlink CSI-RSs + UE feedback: BSs may acquire CSI by first transmitting CSI-RSs in downlink and subsequently receiving quantized channel feedback from the UEs in uplink. This approach is typically used when the uplink-downlink channel reciprocity does not hold, e.g., in frequency division duplex (FDD) systems. The 3GPP specifies a number of channel compression methods for feedback purposes:

• 3GPP Rel. 15 UEs utilize a predefined set of codebooks known at both the BS 104 and the UE 102 to compress the downlink CSI.

• 3GPP Rel. 18 is currently exploring the utilization of a machine learning (ML)-based feedback. As illustrated in Figure 2, the key idea of 3GPP Rel. 18 is to compress the measured DL channel at the UE 102, and then send a compressed feedback to the BS 104 to reconstruct the downlink CSI. This can, e.g., be achieved by using an autoencoder (AE) architecture with an encoder that compresses the channel and a decoder that reconstructs the channel from the compressed state, see Figure 2.

3) CSI acquisition based on both SRSs and downlink CSI-RSs + UE feedback: As illustrated in Figure 3, 3GPP Rel. 17 BSs 104 are capable of acquiring CSI by:

• first receiving SRSs to estimate the frequency-independent parameters of the wireless channel (delays and angles of arrival/departure of each propagation path),

• subsequently transmitting UE-specific CSI-RSs to the UE 102, and

• finally receiving compressed CSI and estimating the downlink channel utilizing closed -form expressions. Machine Learning

Artificial intelligence (Al) and Machine Learning (ML) have been investigated as promising tools to optimize the design of air-interface in wireless communication networks in both academia and industry. Example use cases include using autoencoders for channel state information (CSI) compression to reduce the feedback overhead and improve channel prediction accuracy; using deep neural networks for classifying line-of-sight (LOS) and non-line-of-sight (NLOS) conditions to enhance the positioning accuracy; and using reinforcement learning for beam selection at the network side and/or the user equipment (UE) side to reduce the signaling overhead and beam alignment latency; using deep reinforcement learning to learn an optimal precoding policy for complex multiple-input multiple-output (MIMO) precoding problems.

In 3GPP NR standardization work, there is a new release 18 study item (SI) on AI/ML for NR air interface ¹ . This study item will explore the benefits of augmenting the air-interface with features enabling improved support of AI/ML based algorithms for enhanced performance and/or reduced complexity/overhead . Through studying a few selected use cases (CSI feedback, beam management and positioning), this SI aims at laying the foundation for future air-interface use cases leveraging AI/ML techniques.

Summary

Systems and methods described herein provide for a ML-based Channel State Information (CSI) estimation system at both a User Equipment (UE) and network node or base station that use jointly optimized machine learning (ML) based encoders and decoders to estimate CSI for one or more resources of the channel between the UE and the network node that were not necessarily directly measured by either the UE or the network node. In an embodiment, a distributed ML model between the UE and the network node utilizes measurements of both downlink (at the UE) and uplink (at the network node) reference signals to reconstruct parts of, or the full downlink channel and/or parts of, or the full uplink channel. The reconstruction is possible by, prior to using the model for reconstruction purposes, optimizing the distributed ML model end-to-end across the two nodes.

In an embodiment, a method performed by a network node for determining channel state information of a channel between the network node and a UE using a ML based decoder can include transmitting one or more downlink reference signals to the UE on a first set of resources. The method can also include receiving, from the UE, one or more uplink reference signals on a second set of resources from the UE. The method can also include receiving, from the UE, compressed channel state information, the compressed channel state information being a version of channel state information estimated at the UE based on measurements of the one or more downlink reference signals that is compressed via a ML-based encoder. The method can also include estimating, by using the ML-based decoder, channel state information for a third set of resources based on measurements performed on the one or more uplink reference signals on the second set of resources and the compressed channel state information received from the UE.

1 . RP-213599, “Study on Artificial Intelligence (AI)ZMachine Learning (ML) for NR Air Interface”, Dec. 2021 . New 3GPP study item for Rel.18. In another embodiment, a network node can be configured to communicate with a U E, and the network node can include a radio interface and processing circuitry configured to transmitting one or more downlink reference signals to the UE and receive, from the UE, one or more uplink reference signals on a second set of resources from the UE. The network node can also receive, from the UE, compressed channel state information, the compressed channel state information being version of channel state information estimated at the UE based on measurements of the one or more downlink reference signals that is compressed via a ML-based encoder. The network node can also estimate, by using the ML-based decoder, channel state information for a third set of resources based on measurements performed on the one or more uplink reference signals on the first set of resources and channel state information of at least a portion of the downlink channel between the network node and the UE.

In an embodiment, a method performed by a UE for compressing channel state information of a channel between the UE and a network node using a ML based encoder can include receiving one or more downlink reference signals from the network node on a first set of resources. The method can also include transmitting to the network node one or more uplink reference signals on a second set of resources. The method can also include estimating channel state information of at least a portion of a downlink channel between the network node and the UE based on measurements on the one or more downlink reference signals on the first set of resources. The method can also include compressing, using the ML-based encoder, the channel state information of at least a portion of the downlink channel into compressed downlink channel state information, wherein the ML-based encoder is jointly optimized with a ML-based decoder at the network node. The method can also include providing, to the network node, the compressed downlink channel state information.

In another embodiment, a UE can be configured to communicate with a network node, and the UE can include a radio interface and processing circuitry configured to receive one or more downlink reference signals from the network node on a first set of resources. The processing circuitry can also transmit to the network node one or more uplink reference signals on a second set of resources. The processing circuitry can also estimate channel state information of at least a portion of a downlink channel between the network node and the UE based on measurements on the one or more downlink reference signals on the first set of resources. The processing circuitry can also compress, using a Machine Learning, ML, based encoder, the channel state information of at least a portion of the downlink channel into compressed downlink channel state information, wherein the ML-based encoder is jointly optimized with a ML-based decoder at the network node. The processing circuitry can also provide to the network node the compressed downlink channel state information.

Brief Description of the Drawings

The accompanying drawing figures incorporated in and forming a part of this specification illustrate several aspects of the disclosure, and together with the description serve to explain the principles of the disclosure.

Figure 1 illustrates one example of Channel State Information (CSI) acquisition according to some embodiments of the present disclosure;

Figure 2 illustrates one example of machine learning based CSI acquisition according to some embodiments of the present disclosure; Figure 3 illustrates one example of downlink CSI acquisition according to some embodiments of the present disclosure;

Figure 4 is an illustration of the half channel problem according to some embodiments of the present disclosure;

Figures 5A-5D are illustrations of a system for downlink CSI estimation according to some embodiments of the present disclosure;

Figure 6A-6C are illustrations of a system to mitigate the half channel problem according to some embodiments of the present disclosure;

Figure 7 is a message sequence chart of a system for Machine Learning (ML) based CSI estimation according to some embodiments of the present disclosure;

Figure 8 illustrates one example of a cellular communications system according to some embodiments of the present disclosure;

Figure 9 is a schematic block diagram of a radio access node according to some embodiments of the present disclosure;

Figure 10 is a schematic block diagram that illustrates a virtualized embodiment of the radio access node of Figure 9 according to some embodiments of the present disclosure;

Figure 11 is a schematic block diagram of the radio access node of Figure 9 according to some other embodiments of the present disclosure;

Figure 12 is a schematic block diagram of a User Equipment device (UE) according to some embodiments of the present disclosure; and

Figure 13 is a schematic block diagram of the UE of Figure 12 according to some other embodiments of the present disclosure.

Detailed Description

The embodiments set forth below represent information to enable those skilled in the art to practice the embodiments and illustrate the best mode of practicing the embodiments. Upon reading the following description in light of the accompanying drawing figures, those skilled in the art will understand the concepts of the disclosure and will recognize applications of these concepts not particularly addressed herein. It should be understood that these concepts and applications fall within the scope of the disclosure.

Network Node: As used herein, a "network node” or "base station” is any node in a Radio Access Network (RAN) of a cellular communications network that operates to wirelessly transmit and/or receive signals. Some examples of a radio access node include, but are not limited to, a base station (e.g., a New Radio (NR) base station (gNB) in a Third Generation Partnership Project (3GPP) Fifth Generation (5G) NR network or an enhanced or evolved Node B (eNB) in a 3GPP Long Term Evolution (LTE) network), a high-power or macro base station, a low-power base station (e.g., a micro base station, a pico base station, a home eNB, or the like), a relay node, a network node that implements part of the functionality of a base station or a network node that implements a gNB Distributed Unit (gNB-DU)) or a network node that implements part of the functionality of some other type of radio access node. User Equipment device (UE): One type of communication device is a UE, which may be any type of wireless device that has access to (i.e., is served by) a wireless network (e.g., a cellular network). Some examples of a wireless communication device include, but are not limited to: a User Equipment device (UE) in a 3GPP network, a Machine Type Communication (MTC) device, and an Internet of Things (loT) device. Such wireless communication devices may be, or may be integrated into, a mobile phone, smart phone, sensor device, meter, vehicle, household appliance, medical appliance, media player, camera, or any type of consumer electronic, for instance, but not limited to, a television, radio, lighting arrangement, tablet computer, laptop, or PC. The wireless communication device may be a portable, hand-held, computer-comprised, or vehicle-mounted mobile device, enabled to communicate voice and/or data via a wireless connection.

Note that the description given herein focuses on a 3GPP cellular communications system and, as such, 3GPP terminology or terminology similar to 3GPP terminology is oftentimes used. However, the concepts disclosed herein are not limited to a 3GPP system.

Note that, in the description herein, reference may be made to the term "cell”; however, particularly with respect to 5G NR concepts, beams may be used instead of cells and, as such, it is important to note that the concepts described herein are equally applicable to both cells and beams.

There exist certain challenges. While the acquisition of precise Channel State Information (CSI) is crucial for precoding, the reference signals (RSs) used for CSI acquisition occupy precious resources that cannot be used for data transmission or reception purposes. Therefore, the physical resources are limited to be used by the reference signals. For instance, in a practical deployment, there always exists a shortage of Sounding Reference Signals (SRS) resources such that not all UEs can be scheduled for SRS transmission .

Furthermore, practical limitations in hardware implementations can further complicate the CSI acquisition procedure. To be concrete:

CSI acquisition methods relying on uplink SRSs: As illustrated in Figure 4, UEs 102 with a smaller number of transmit (TX) chains than receive (RX) chains cannot simultaneously transmit SRS from all antennas. This problem, here referred to as the half-channel problem is mitigated by some UEs— typically high-end models— through the incorporation of switches that allow the sequential transmission of SRSs (over the different Tx chains/antennas). The main issues with this solution are that:

• Low-end UEs may not implement a switch due to cost and complexity constraints, such that the maximum downlink transmission rank may be limited because of lack of full spatial channel information. Internal evaluations of this problem showed that the existing DL beamforming methods are not sufficient to handle this issue, which renders significant performance degradation.

• SRS switching requires at least a guard symbol. This prolongs the time for the transmission of all SRSs, which may span multiple slots affecting the network performance by e.g., using aged CSI as basis for precoding. With UE mobility as one of the key bottlenecks for massive Multiple Input Multiple Output (MIMO), the channel aging effect impacts the DL beamforming performance significantly.

CSI acquisition methods relying on downlink CSI - Reference Signals (RSs) + uplink UE feedback. The feedback length of non-Machine Learning (ML) feedback solutions (3GPP Rel. 16 & Rel. 17) typically grows with the number of base stations (BS) and UE antennas. In some cases, the UE feedback may require hundreds of bits, since the full downlink channel may have to be compressed. Moreover, CSI inaccuracies may occur due to the assumptions employed, e.g., the equality of the angles or departure/arrival in uplink/downlink for the approach defined in 3GPP Rel. 17.

The ML-based feedback solution currently being developed for 3GPP Rel. 18 is focused on Frequency Division Duplexing (FDD) and requires the compression of all the relevant CSI solely based on the measured CSI- RS.

The current CSI-feedback in Rel 16 and 17 still have not solve the UE mobility use case in a satisfactory way, which serves as a key enhancement item to be settled.

In practice, the CSI feedback quality may be compromised in the coverage limit when extremely low signal to noise ratio (SNR) is expected at the physical uplink channels such as Physical Uplink Shared Channel (PUSCH) and Physical Uplink Control Channel (PUCCH). The decoding reliability of CSI bits becomes a problem in these scenarios.

Overall, an advantage of the proposed system and method in the present disclosure, is that the system and method minimize the number of resources occupied by RSs utilized for downlink/uplink CSI acquisition purposes and/or enhances the channel estimation accuracy w.r.t. prior art (the scope of [2] and earlier 3GPP releases of the CSI framework). Improved channel estimation will lead to, e.g., increased downlink (DL) throughput (when being able to better reconstruct the DL channel and using the channel estimate for example for DL beamforming). It also has the ability to minimize the feedback from the UE to the BS, by making use of channel correlations between DL and uplink (UL) measurements and compressing the DL channel state more effectively, taking knowledge of UL channel measurements into consideration.

For the case of downlink channel estimation for massive MIMO FDD systems, the gains intuitively occur because when compared to the most advanced non-ML approach of 3GPP Rel. 17, the UE will only encode and transmit the information strictly necessary to accurately reconstruct the desired downlink/uplink channel at the BS. The amount of information transmitted by the UE will generally be less than that required by the 3GPP Rel. 17 approach because the data-driven model does not rely on any closed-form expression or channel model assumptions. When compared to the ML approach considered for 3GPP Rel. 18, the UE will not need to encode and transmit all the measured downlink channel information but, considering that the BS has previously acquired CSI through the resource-efficient SRSs, only the additional information required to reconstruct the channel .

For the case of the Time Division Duplexing (TDD) half channel problem illustrated in Figures 6A-6C, the gains intuitively occur because the compressed feedback size transmitted by the UE in c) could be seen to roughly (from a channel information perspective, ignoring practical aspects, such as noisy measurements):

• consist of zero bits when the two UE antennas are fully correlated (i.e., measuring one antenna in the UL is enough to know the CSI of the second antenna and hence no additional information is required from the UE. • be reduced when compared to that transmitted through the ML approach considered for 3GPP Rel. 18 [2] when there exists some correlation between the two UE antennas.

In its main embodiment, this present disclosure proposes a distributed ML model between a UE and a BS that utilizes measurements of both downlink (at the UE) and uplink (at the BS) RSs to reconstruct parts of, or the full downlink channel and/or parts of, or the full uplink channel. The reconstruction is possible by, prior to using the model for reconstruction purposes, optimizing the distributed ML model end-to-end across the two nodes.

The abovementioned measurements serving as input to the model need not be raw measurements of the channel but could also involve a pre-processing step, e.g., transforming the channel measurements over antenna ports to angular/beam domain.

Similarly, the reconstruction of the downlink and/or uplink channel would be dependent on the target reconstruction (also potentially involving a post-processing) and the chosen loss functions used for optimization .

The proposed CSI acquisition method is illustrated in Figures 5A-5D and can be summarized as follows: Step 1 (Figure 5A): UE 502 receives downlink RSs (e.g., CSI-RS in NR) from the network node 504 to estimate part or the full downlink channel Here, part of the downlink channel may refer to a subset of subcarriers, a subset of network node 504 antenna ports, and/or a subset of UE antenna ports.

Step 2 (Figure 5B): The UE 502 transmits uplink RSs (e.g., SRSs in NR) to the network node 504, which receives the UL RSs (Y^). In some embodiments, the network node 504 estimates part or the full uplink channel (#u ^s ) in this step. Here, part of the uplink channel may refer to a subset of subcarriers and/or a subset of UE antenna ports.

Step 3 (Figure 5C): The UE 502 compresses via a trainable ML-based encoder part or the full downlink channel estimated in Step 1, quantizes it and transmits such information to the network node 504. The network node 504 receives Y^.

Step 4 (Figure 5D): The network node 504 utilizes— directly or with additional processing— the information acquired in Step 2) or Hy ^RS ) and the received information in Step 3) as inputs to an ML-based decoder to estimate part or the full downlink (H _DL ) and/or part or the full uplink channels (H _UL ) between the UE 502 and the network node 504.

It should be noted that the steps need not be carried out in the order listed, and only require the dependencies listed to be fulfilled, e.g., Step 3 is dependent on the completion of Step 1. Step 4 is dependent on the completion of Step 2 and Step 3. Because Step 1 and Step 2 may be repeated in time, there also exists a mutual dependence between such steps, e.g., in the maximum time difference permitted among the reference signal transmissions.

Prior to the deployment of the proposed CSI acquisition method, the trainable parameters of both the ML- based encoder utilized by the UE 502 in Step 3), and the ML-based decoder utilized by the network node 504 in Step 4), can be jointly optimized in an end-to-end manner during a training stage.

During deployment, the UE may indicate to the BS that it has the capability of utilizing such ML-based encoder prior to the configuration of the reference signals in Steps 1) and 2) by the network node 504, and the network node 504's indication to the UE 502 that such ML-based encoder should be utilized for CSI feedback compression in Step 3).

In some embodiments, such training will be performed in a supervised manner, where the encoder and decoder forming the AE will be optimized to minimize a given loss function that depends on the output of the decoder.

In an embodiment, the ML model is built to compress at the UE and reconstruct at the BS part or the full downlink/uplink wireless channel by jointly optimizing end-2-end (processed) channel measurements based on RSs transmitted in DL and UL, where:

• the RSs in downlink can for example be CSI-RSs and in uplink can for example be SRSs, and

• where the channel can be measured in different points in time, over a subset of network node 504 and UE 502 antenna ports, and/or over a different set of resources in frequency.

Concretely, this entails that:

• The ML part of the model at the network node 504 (decoder) utilizes the output of the ML part of the model at the UE 502 (encoder) and the uplink RSs previously transmitted by the UE 502.

• The downlink RSs I the (processed) channel measurements input to the UE 502 part of the ML model (encoder) and the uplink RSs I channel measurements input to the network node 504 part of the ML model (decoder) may have been derived from downlink and uplink RSs not transmitted using the same resources, e.g., different network node 504 and UE 502 antenna ports and/or frequency resources and/or at different points in time.

• The channel estimated/predicted at the output of the network node 504 part of the ML model (decoder) may represent resources, e.g., network node 504 UE 502 antenna ports and/or frequency resources and/or points in time, different from the ones in the RSs I the (processed) channel measurements at the input of the network node 504 part of the ML model (decoder).

The permitted time difference among inputs to the network node 504 part of the ML model (decoder) may be smaller than the time difference between the information acquired in Step 2) ( Y SI ^ULY) ^ar| d the information received in Step 3) YCSI^DL )- In ^some embodiments, the UE will indicate the capability of estimating/predicting the compressed feedback used as an input to the network node 504 part of the ML model (decoder) at a time t ₂ different from which the downlink RSs have been measured t,— either in the future or in the past. For instance, the UE may feedback a compressed version of HCSI ^UL)— ^an estimate of the channel at the time the associated uplink RSs are transmitted— instead of a compressed version of HCSI ^DL)— ^an estimate of the channel at the time the downlink RSs were received. In some embodiments, such requirements on a UE providing channel prediction may be specified in standard text, e.g., that all UEs shall be capable of predicting the channel specified by a maximum value of 1^2 ~ ^1L

In other embodiments, different UEs will have different capabilities and such capabilities are signaled by UE capability signaling.

In some embodiments, the UE 502 may communicate to the network node 504 such capability, e.g., when responding to a BS-sent UECapability Enquiry message with a UECapabilitylnformation message.

In some embodiments, the network node 504 may explicitly indicate to the UE 502 the time for which the estimation/prediction the compressed feedback used as an input to the network node 504 part of the ML model (decoder) should be produced. This could e.g., be indicated in the CSI-RS report configuration, in a downlink control information (DCI) or by RRC setup messages.

In other embodiments, the UE 502 will implicitly select the time for which the estimation/prediction the compressed feedback used as an input to the network node 504 part of the ML model (decoder) depending on the time at which the associated uplink RSs are transmitted (t _UL ).

In an embodiment, the network node 504 may indicate that a given set of configured CSI-RSs are related to a given set of configured SRSs.

The scheduling of RSs by the network node 504 may satisfy certain requirements (for example in time domain and/or spatial domain) between, e.g. the time separation between downlink RSs (Step 1) and the uplink RSs (Step 2) , and/or the time separation between uplink RSs (Step 2) and the feedback information transmitted by the UE (Step 3), and/or the antenna ports used in Step 2.

In some embodiments, the limits/allowable values of such requirements may be specified in standard text . Figure 7 illustrates a message sequence chart of a system comprising a UE 502 and a network node 504 for ML based CSI estimation according to some embodiments of the present disclosure.

At 702, the network node 504 can send a request for capabilities to the UE 502. The message can be in the form of a UECapability Enquiry message. Based on the capabilities of the UE 502, the network node 504 can determine whether the UE 502 is capable providing estimation/prediction at a time t ₂ different from which the downlink RSs have been received/measured t In other embodiments, the UE may have the capability of providing estimation/prediction of CSI of resources that are not measured, such as frequency resources, or antenna ports.

At 704, the UE 502 can respond to the UECapability Enquiry message with a UECapabilitylnformation message that defines the capabilities of the UE 502. It is to be appreciated that when a capability report is provided, that it is in response to a capability request - e.g., step 704 is subsequent to step 702.

At 706a-c, the network node 504 can configure the UE 502 to measure one or more parameters from the downlink reference signals and estimate CSI for one or more resources that can be different than the resources measured. The configuring at 706 can also configure the UE 502 to provide one or more reference signals at a specific time, on one or more specified frequency resources, or using one more specified antenna ports of the UE 502. For example, at 706a, the network node 504 can provide a first indication to the UE 502 to estimate the channel state information at a specific time other than the time of reception of the downlink reference signals. At 706b, the network node 504 can provide a second indication to the UE 502 indicating one or more resources associated with the one or more downlink reference signals to be measured by the UE 502. At 706c, the network node 504 can configure the UE 502 to provide one or more uplink reference signals using resources different than resources used by the one or more downlink reference signals.

In some embodiments, the network node 504 may explicitly indicate to the UE 502 the time for which the estimation/prediction the compressed feedback used as an input to the network node 504 part of the ML model (decoder) should be produced. This could e.g., be indicated in the CSI-RS report configuration, in a DCI or by RRC setup messages. In the DCI, in one embodiment, a trigger state is pointing to one or more RS report configurations. This would imply both separate CSI-RS and SRS report configurations or a combined RS report configuration.

In other embodiments, the UE 502 will implicitly select the time for which the estimation/prediction the compressed feedback used as an input to the network node 504 part of the ML model (decoder) depending on the time at which the associated uplink RSs described in 710 are transmitted (t _UL ).

In these embodiments, the network node 504 may indicate that a given set of configured CSI-RSs are related to a given set of configured SRSs.

In one embodiment, this could e.g., be done via a RS report configuration, similar to the CSI-RS report configurations used today, but also including an associated SRS configuration.

In some embodiments, such requirements on a UE 502 providing estimation/prediction at a time t ₂ different from which the downlink RSs have been received/measured may be specified in standard text, e.g., that all UEs shall be capable of predicting the channel specified by a maximum value of \t ₂ - t- .

At 708, the network node 504 can transmit one or more downlink reference signals to the UE 502. The downlink reference signals can be transmitted via one or more resources such as time, antenna ports, or frequency resources.

UE 502 receives downlink RSs (e.g., CSI-RS in NR) from the BS (Fes/) t° estimate part or the full downlink channel {H^ ^S _L ‘~ ^RS ). Here, part of the downlink channel may refer to a subset of subcarriers, a subset of network node 504 antenna ports, and/or a subset of UE antenna ports.

In some embodiments and prior to the transmission of the downlink RSs, the network node 504 will explicitly indicate (via the configuration at 706) the resources to be used by RSs to be measured, e.g. the resource elements in frequency domain and the set of antenna ports where the measurements are to be realized to derive H _D ^CS _L ‘- ^RS . Notably, this may include the UE 502 ports where the downlink measurements should be realized, e.g., via an additional field/parameter to the CSI-RS report configuration, downlink control information (DCI), or radio resource control (RRC) setup messages.

In some embodiments, the network node 504 will indicate to the UE 502 that it should perform the estimation/prediction of part or the full downlink channel at a time t ₂ different from which the downlink RSs have been received/measured t-j— either in the future or in the past. This is to explicitly attempt to remove the impact of channel aging and facilitate an accurate channel estimation at the network node 504 in step 716.

In some embodiments, the transmission of CSI-RS may be designed in a way such that the CSI-RS is only transmitted to the subspace with non-intersection (e.g., nullspace) of the channel estimated based on a certain set of UL reference signals. This is to target the half channel scenarios illustrated in Fig. 4 so that the DL reference signal is focused on obtaining the missing spatial information. Intuitively, this may reduce the UE feedback overhead when compared to a solution where the overall subspace of the channel can be conveyed back.

At 710, the UE 502 can provide one or more uplink reference signals to the network node 504. The UE 502 transmits uplink RSs (e.g., SRSs in NR) to the BS, which receives the UL RSs In some embodiments, the network node 504 may estimate part or the full uplink channel (Hu ^R _L ^s ). Here, part of the uplink channel may refer to a subset of subcarriers and/or a subset of UE antenna ports.

The proposed method focuses on situations where the BS does not have means to estimate the entire DL CSI from the received uplink RSs. These situations comprise those where, e.g.:

• Channel reciprocity holds but only a part of the UL CSI can be retrieved by the BS using the SRS. This may be, e.g., because the UE cannot transmit RSs from all antennas (i.e., the TDD half channel problem illustrated in Figure 6) or because the UE can only transmit RSs in a reduced subset of time/frequency due to power limitations.

• The entire UL CSI is acquired by the BS, but channel reciprocity does not hold (e.g., FDD systems).

It should be noted that the proposed method also works in cases where the entire DL CSI can be estimated from the received UL RSs (e.g., TDD systems), in which case one could expect the resultant channel estimate and/or the resultant precoder to be more accurate than systems only relying on DL or UL RSs . This is due to the utilization of correlated uplink and downlink measurements for their computation.

In some embodiments, the network node 504 may schedule (during the configuration 706) downlink and uplink RSs not transmitted using the same resources, e.g., different network node 504 /UE 502 antenna ports and/or frequency resources and/or at different points in time. Such subsets may be specified during the configuration 706 of the RSs, e.g., via an additional field/parameter to the CSI -RS report configuration, via a combined report configuration covering both downlink and uplink RSs (e.g., a joint configuration of CSI-RS and SRS in NR), downlink control information (DCI), or RRC setup messages.

The use of distinct resources facilitates a reduction in the RS overhead necessary to generate the downlink and/or uplink channel estimates in 716, since they intuitively minimize the amount of redundant information.

In those embodiments where the network node 504 estimates part or the full uplink channel (Hu ^RS ), such estimation may be realized using prior art solutions, e.g., least squares or minimum mean squared error - interference rejection combining (MMSE-IRC).

In those embodiments where the network node 504 estimates part of the uplink channel H'l ^R _L ^s , the network node 504 may schedule uplink RSs for a subset of UE antenna ports. Such subset may be specified during the configuration of the RSs, e.g., via an additional field/parameter to the CSI -RS report configuration, in the SRS resource set configuration, downlink control information (DCI), or RRC setup messages. In the DCI, in one embodiment, a trigger state is pointing to one or more RS report configurations. This would imply both separate CSI-RS report configuration and SRS resource set configuration or a combined RS report configuration (see also below).

In an embodiment, step 708 can come after step 710, while in other embodiments, 708 precedes 710. In some embodiments, the scheduling of the downlink RSs described in 708 and the uplink RSs described in 710 by the BS will satisfy specific timing constraints. In some embodiments, such timing constraints may be specified in standard text or may be implementation-specific.

At 712, the UE 502 estimates the channel state information of at least a portion of a downlink channel between the network node 504 and the UE 502 based on the first set of resources. The UE 502 can also compress, via a trainable ML-based encoder part or the full downlink channel estimated, quantize it and transmit or provide such information to the network node 504 at 714.

It should be appreciated that when reference is made to the network node 504 and the UE 502 providing or transmitting information to a specific entity, or receiving information from an entity, that this also covers a scenario where there is an intermediate node that receives and forwards the information to/from the specific entity.

The network node 504 receives Y^si- As described earlier, such compression could be performed to reconstruct the channel at the base station at another point in time than when the CSI -RS resources were received. In different embodiments, this could be achieved by separating the channel prediction and compression/encoder into separately trained models, by training an encoder that can compress the channel, targeting a different time point in time for the reconstruction, or by transforming the latent space as output from the encoder from the measurement time to a future/past point in time.

In some embodiments 712 can be executed before providing the uplink reference signals to the network node 504 at 710.

In some embodiments, the scheduling of the uplink RSs described in Step 2), and the UE feedback will satisfy specific timing constraints. In some embodiments, such timing constraints may be specified in standard text or may be implementation-specific.

A key particularity of some embodiments of the disclosed method is that the UE 502 is aware that the network node 504 will have access to CSI from the transmission of uplink RSs, at the time of channel reconstruction. Consequently, one can expect that, as a result of the optimization of the Al algorithm, the UE 502 does not need to send redundant information already accessible to the network node 504 vi a the CSI acquired via the uplink RSs previously transmitted, i.e., the Al algorithm will leverage the knowledge that the network node 504 has already acquired CSI to further compress the information.

For instance, consider the example of Figures 6A-6C where the TDD half channel problem occurs, and the UE 502 is equipped with two antennas but only one of them {Anti) is capable of transmitting uplink SRSs. In such embodiments, the UE part of the ML model (encoder) may only encode the CSI for those antennas that did not transmit uplink SRSs, which may be intuitively compressed by leveraging the correlation existing among the channels of nearby antennas. In other words, the encoder and decoder are learned conditioned on the availability of CSI H^ ¹ 1 H ¹ .

At 716, the network node 504 can estimate via the ML-based decoder, channel state information of at least a portion of an uplink channel for a second set of resources different than the first set of resources between the network node 504 and the UE 502 based on measurements performed on the one or more uplink reference signals using the first set of resources and channel state information of at least a portion of the downlink channel between the network node 504 and the UE 502 based on the encoded information. The channel state information of at least the portion of the downlink channel can be compressed channel state information as compressed in step 714. The channel state information can also be decompressed before being utilized by the network node 504 in step 716.

The network node 504 utilizes— directly or with additional processing— the information acquired based on the uplink reference signals (F^ y) and the received information from the UE 502 at 714 (F^ y) as inputs to an ML- based decoder to estimate part or the full downlink (H _DL ) channel— previously unavailable at the network node 504 — and/or part or the full uplink tft _UL ) between the UE 502 and the network node 504.

Illustrative examples of additional processing for Y _CSI include denoising, matched filtering, domain transformations based on, e.g., the discrete Fourier transform (DFT), or, more generally, an uplink channel estimation algorithm that outputs H _UL . In general, this processing may allow the training stage to converge faster and/or to increase the overall accuracy of the channel estimation (e.g., by making it easier to eliminate measurement noise).

Intuitively, if part or the full uplink channel between the UE 502 and the network node 504 H _UL was already estimated (i.e., H'l ^R _L ^s ), attempts to enhance the estimation previously available solely based on uplink RSs by also leveraging the information available on part or the full downlink channel from F^ y.

The proposed solution admits at least two distinct implementations:

Standard approach: The standard specifies the inputs (and perhaps the outputs) at the encoders/decoders so that each vendor can appropriately train their models and facilitate inter-vendor model development. A handling of understanding between UE 502 and network node 504 is also required (as outlined in the invention above) to enable accurate model management (i.e., how should the encoder act (e.g., which model to use, or, how to adapt the model output), conditioned on the information about the CSI acquired at the network node).

The main benefit is an enhanced compression when compared to the 3GPP Rel. 18 implementations currently being investigated. That is, the compression at the UE can be optimized based on the knowledge of the CSI acquired at the base station. This should lead to a minimal overhead, at a given reconstruction accuracy. Or, alternatively an improved reconstruction accuracy, at a given overhead.

Additional approach: Using the same encoder as in 3GPP Rel. 18, BSs utilize the additional SRS input.

The main benefit is an improved channel estimate. In this case, the overhead would be defined by what is investigated in Rel. 18 for the two-sided Al use case for CSI feedback, but the channel reconstruction would be improved compared to not utilizing SRS measurements in the reconstruction.

Figure 8 illustrates one example of a cellular communications system 800 in which embodiments of the present disclosure may be implemented. In the embodiments described herein, the cellular communications system 800 is a 5G system (5GS) including a Next Generation RAN (NG-RAN) and a 5G Core (5GC) or an Evolved Packet System (EPS) including an Evolved Universal Terrestrial RAN (E-UTRAN) and an Evolved Packet Core (EPC). In this example, the RAN includes base stations 802-1 and 802-2, which in the 5GS include NR base stations (gNBs) and optionally next generation eNBs (ng-eNBs) (e.g., LTE RAN nodes connected to the 5GC) and in the EPS include eNBs, controlling corresponding (macro) cells 804-1 and 804-2. The base stations 802-1 and 802-2 are generally referred to herein collectively as base stations 802 and individually as base station 802. Likewise, the (macro) cells 804-1 and 804-2 are generally referred to herein collectively as (macro) cells 804 and individually as (macro) cell 804. The RAN may also include a number of low power nodes 806-1 through 806-4 controlling corresponding small cells 808-1 through 808-4. The low power nodes 806-1 through 806-4 can be small base stations (such as pico or femto base stations) or RRHs, or the like. Notably, while not illustrated, one or more of the small cells 808-1 through 808-4 may alternatively be provided by the base stations 802. The low power nodes 806-1 through 806-4 are generally referred to herein collectively as low power nodes 806 and individually as low power node 806. Likewise, the small cells 808-1 through 808-4 are generally referred to herein collectively as small cells 808 and individually as small cell 808. The cellular communications system 800 also includes a core network 810, which in the 5G System (5GS) is referred to as the 5GC. The base stations 802 (and optionally the low power nodes 806) are connected to the core network 810.

The base stations 802 and the low power nodes 806 provide service to wireless communication devices 812-1 through 812-5 in the corresponding cells 804 and 808. The wireless communication devices 812-1 through 812-5 are generally referred to herein collectively as wireless communication devices 812 and individually as wireless communication device 812. In the following description, the wireless communication devices 812 are oftentimes UEs, but the present disclosure is not limited thereto.

The network nodes 802 and 806 can perform the functionality of the network nodes 504 disclosed herein, and wireless communications devices 812 can perform the functionality of UE 502 described herein. Jointly, the network nodes 802 and 806 and wireless communication devices 812 can have jointly optimized/trained ML models that perform ML based CSI estimation on uplink and downlink channels at their respective nodes.

Figure 9 is a schematic block diagram of a radio access node 900 according to some embodiments of the present disclosure. Optional features are represented by dashed boxes. The radio access node 900 may be, for example, a base station 802 or 806 or a network node that implements all or part of the functionality of the base station 802 or gNB described herein, including network node 504. As illustrated, the radio access node 900 includes a control system 902 that includes one or more processors 904 (e.g., Central Processing Units (CPUs), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs), and/or the like), memory 906, and a network interface 908. The one or more processors 904 are also referred to herein as processing circuitry. In addition, the radio access node 900 may include one or more radio units 910 that each includes one or more transmitters 912 and one or more receivers 914 coupled to one or more antennas 916. The radio units 910 may be referred to or be part of radio interface circuitry. In some embodiments, the radio unit(s) 910 is external to the control system 902 and connected to the control system 902 via, e.g., a wired connection (e.g., an optical cable). However, in some other embodiments, the radio unit(s) 910 and potentially the antenna(s) 916 are integrated together with the control system 902. The one or more processors 904 operate to provide one or more functions of a radio access node 900 as described herein. In some embodiments, the function(s) are implemented in software that is stored, e.g., in the memory 906 and executed by the one or more processors 904.

Figure 10 is a schematic block diagram that illustrates a virtualized embodiment of the radio access node 900 according to some embodiments of the present disclosure. This discussion is equally applicable to other types of network nodes. Further, other types of network nodes may have similar virtualized architectures. Again, optional features are represented by dashed boxes.

As used herein, a "virtualized” radio access node is an implementation of the radio access node 900 in which at least a portion of the functionality of the radio access node 900 is implemented as a virtual component(s) (e.g., via a virtual machine(s) executing on a physical processing node(s) in a network(s)). As illustrated, in this example, the radio access node 900 may include the control system 902 and/or the one or more radio units 910, as described above. The control system 902 may be connected to the radio unit(s) 910 via, for example, an optical cable or the like. The radio access node 900 includes one or more processing nodes 1000 coupled to or included as part of a network(s) 1002. If present, the control system 902 or the radio unit(s) are connected to the processing node(s) 1000 via the network 1002. Each processing node 1000 includes one or more processors 1004 (e.g., CPUs, ASICs, FPGAs, and/or the like), memory 1006, and a network interface 1008.

In this example, functions 1010 of the radio access node 900 described herein are implemented at the one or more processing nodes 1000 or distributed across the one or more processing nodes 1000 and the control system 902 and/or the radio unit(s) 910 in any desired manner. In some particular embodiments, some or all of the functions 1010 of the radio access node 900 described herein are implemented as virtual components executed by one or more virtual machines implemented in a virtual environment(s) hosted by the processing node(s) 1000. As will be appreciated by one of ordinary skill in the art, additional signaling or communication between the processing node(s) 1000 and the control system 902 is used in order to carry out at least some of the desired functions 1010. Notably, in some embodiments, the control system 902 may not be included, in which case the radio unit(s) 910 communicate directly with the processing node(s) 1000 via an appropriate network interface(s).

In some embodiments, a computer program including instructions which, when executed by at least one processor, causes the at least one processor to carry out the functionality of radio access node 900 or a node (e.g., a processing node 1000) implementing one or more of the functions 1010 of the radio access node 900 in a virtual environment according to any of the embodiments described herein is provided. In some embodiments, a carrier comprising the aforementioned computer program product is provided. The carrier is one of an electronic signal, an optical signal, a radio signal, or a computer readable storage medium (e.g., a non-transitory computer readable medium such as memory).

Figure 11 is a schematic block diagram of the radio access node 900 according to some other embodiments of the present disclosure. The radio access node 900 includes one or more modules 1100, each of which is implemented in software. The module(s) 1100 provide the functionality of the radio access node 900 described herein. This discussion is equally applicable to the processing node 1000 of Figure 10 where the modules 1100 may be implemented at one of the processing nodes 1000 or distributed across multiple processing nodes 1000 and/or distributed across the processing node(s) 1000 and the control system 902.

Figure 12 is a schematic block diagram of a wireless communication device 1200 according to some embodiments of the present disclosure. In an embodiment, the UE 502 can be an example of a wireless communication device 1200 according to one or more embodiments. As illustrated, the wireless communication device 1200 includes one or more processors 1202 (e.g., CPUs, ASICs, FPGAs, and/or the like), memory 1204, and one or more transceivers 1206 each including one or more transmitters 1208 and one or more receivers 1210 coupled to one or more antennas 1212. The transceiver(s) 1206 includes radio-front end circuitry connected to the antenna(s) 1212 that is configured to condition signals communicated between the antenna(s) 1212 and the processor(s) 1202, as will be appreciated by on of ordinary skill in the art. The processors 1202 are also referred to herein as processing circuitry. The transceivers 1206 are also referred to herein as radio circuitry. In some embodiments, the functionality of the wireless communication device 1200 described above may be fully or partially implemented in software that is, e.g., stored in the memory 1204 and executed by the processor(s) 1202. Note that the wireless communication device 1200 may include additional components not illustrated in Figure 12 such as, e.g., one or more user interface components (e.g., an input/output interface including a display, buttons, a touch screen, a microphone, a speaker(s), and/or the like and/or any other components for allowing input of information into the wireless communication device 1200 and/or allowing output of information from the wireless communication device 1200), a power supply (e.g., a battery and associated power circuitry), etc.

In some embodiments, a computer program including instructions which, when executed by at least one processor, causes the at least one processor to carry out the functionality of the wireless communication device 1200 according to any of the embodiments described herein is provided. In some embodiments, a carrier comprising the aforementioned computer program product is provided. The carrier is one of an electronic signal, an optical signal, a radio signal, or a computer readable storage medium (e.g., a non-transitory computer readable medium such as memory).

Figure 13 is a schematic block diagram of the wireless communication device 1200 according to some other embodiments of the present disclosure. The wireless communication device 1200 includes one or more modules 1300, each of which is implemented in software. The module(s) 1300 provide the functionality of the wireless communication device 1200 described herein.

Any appropriate steps, methods, features, functions, or benefits disclosed herein may be performed through one or more functional units or modules of one or more virtual apparatuses. Each virtual apparatus may comprise a number of these functional units. These functional units may be implemented via processing circuitry, which may include one or more microprocessor or microcontrollers, as well as other digital hardware, which may include Digital Signal Processors (DSPs), special-purpose digital logic, and the like. The processing circuitry may be configured to execute program code stored in memory, which may include one or several types of memory such as Read Only Memory (ROM), Random Access Memory (RAM), cache memory, flash memory devices, optical storage devices, etc. Program code stored in memory includes program instructions for executing one or more telecommunications and/or data communications protocols as well as instructions for carrying out one or more of the techniques described herein. In some implementations, the processing circuitry may be used to cause the respective functional unit to perform corresponding functions according one or more embodiments of the present disclosure.

While processes in the figures may show a particular order of operations performed by certain embodiments of the present disclosure, it should be understood that such order is exemplary (e.g., alternative embodiments may perform the operations in a different order, combine certain operations, overlap certain operations, etc.).

Those skilled in the art will recognize improvements and modifications to the embodiments of the present disclosure. All such improvements and modifications are considered within the scope of the concepts disclosed herein.