Iterative procedure for separate reception / transmission training enhancement for precoder channel state information feedback

Abstract

A user equipment (UE) computes an original downlink (DL) channel estimate between a base station and the UE based on a first channel state information reference signal (CSI-RS); determines precoder information based on the original DL channel estimate, the precoder information representing a precoder that would cause the UE to experience an expected effective DL channel; encodes the precoder information into a bitstream using an encoder of a machine learning (ML) model at the UE; transmits the bitstream to the base station; computes an actual effective DL channel experienced at the UE based on a second CSI-RS; and computes an error between the expected effective DL channel and the actual effective DL channel. The UE then performs an adjustment to the encoder at the UE based on the error and / or transmits error information corresponding to the error to the base station. Related network-side functionality is also disclosed.

Description

Technical Field

[0001] This application relates in general to wireless communication systems, including systems that implement artificial intelligence (AI) / machine learning (ML) models for transmitting and / or receiving channel state information (CSI) feedback. Background Technology

[0002] Wireless mobile communication technologies use various standards and protocols to transmit data between base stations and wireless communication devices. For example, wireless communication system standards and protocols may include, for instance, 3GPP Long Term Evolution (LTE) (e.g., 4G), 3GPP New Radio (NR) (e.g., 5G), and the Institute of Electrical and Electronics Engineers (IEEE) 802.11 standard for Wireless Local Area Networks (WLANs) (often referred to as Wi-Fi within the industry organization). ® ).

[0003] As envisioned by 3GPP, different wireless communication system standards and protocols can use various radio access networks (RANs) for communication between RAN base stations (sometimes referred to as RAN nodes, network nodes, or simply nodes) and wireless communication equipment called user equipment (UEs). 3GPP RANs can include, for example, Global System for Mobile Communications (GSM), Enhanced Data Rate GSM Evolution (EDGE) RAN (GERAN), Universal Terrestrial Radio Access Network (UTRAN), Evolved Universal Terrestrial Radio Access Network (E-UTRAN), and / or Next Generation Radio Access Network (NG-RAN).

[0004] Each RAN can use one or more Radio Access Technologies (RATs) to perform communication between the base station and the UE. For example, GERAN implements the GSM and / or EDGE RAT, UTRAN implements the Universal Mobile Telecommunications System (UMTS) RAT or other 3GPP RATs, E-UTRAN implements the LTE RAT (sometimes simply referred to as LTE), and NG-RAN implements the NR RAT (this NR RAT is sometimes referred to herein as the 5G RAT, 5G NR RAT, or simply NR). In some deployments, E-UTRAN may also implement the NR RAT. In some deployments, NG-RAN may also implement the LTE RAT.

[0005] The base stations used by a RAN can correspond to that RAN. An example of an E-UTRAN base station is an Evolved Universal Terrestrial Radio Access Network (E-UTRAN) Node B (also commonly referred to as Evolved Node B, Enhanced Node B, eNodeB, or eNB). An example of an NG-RAN base station is a Next Generation Node B (sometimes also called gNode B or gNB).

[0006] The RAN provides communication services to external entities through its connection with the core network (CN). For example, E-UTRAN can utilize the evolved packet core (EPC), while NG-RAN can utilize the 5G core network (5GC). Attached Figure Description

[0007] To facilitate the identification of any particular element or action in the discussion, one or more of the most significant digits in the figure reference numerals refer to the figure number in which the element was first introduced.

[0008] Figure 1 An illustration is provided showing an example of a two-sided AI / ML model for CSI compression and decompression according to the implementation scheme discussed herein.

[0009] Figure 2 An example diagram illustrates training performed jointly at either the UE or the network.

[0010] Figure 3 An example is shown illustrating training conducted in a joint manner and across each of the UE and the network (each of the UE and the network is an active participant in the training).

[0011] Figure 4 The flowchart illustrates a process where the first training of the encoder on the UE side of the model can be performed at the UE, while the second training of the decoder on the network side of the model can be performed at the network.

[0012] Figure 5 A flowchart illustrating an example process between the network and the UE is provided, which is used to train an AI / ML model for CSI feedback at the network using joint training.

[0013] Figure 6 A flowchart illustrating the training and use of an AI / ML model for CSI compression according to the implementation scheme discussed herein is provided, wherein the AI / ML model is trained using a collaborative process that employs separate training between the network (e.g., a base station of the network) and the UE.

[0014] Figure 7 A flowchart illustrating the training and use of an AI / ML model for CSI compression according to the implementation scheme discussed herein is provided, wherein the AI / ML model is trained using a collaborative process that employs separate training between the network (e.g., a base station of the network) and the UE.

[0015] Figure 8A flowchart illustrating the training and use of an AI / ML model for CSI compression according to the implementation scheme discussed herein is provided, wherein the AI / ML model is trained using a collaborative process that employs separate training between the network (e.g., a base station of the network) and the UE.

[0016] Figure 9 A flowchart illustrating the training and use of an AI / ML model for CSI compression according to the implementation scheme discussed herein is provided, wherein the AI / ML model is trained using a collaborative process that employs separate training between the network (e.g., a base station of the network) and the UE.

[0017] Figure 10 A flowchart illustrating the training and use of an AI / ML model for CSI compression according to the implementation scheme discussed herein is provided, wherein the AI / ML model is trained using a collaborative process that employs separate training between the network (e.g., a base station of the network) and the UE.

[0018] Figure 11 A flowchart illustrating the training and use of an AI / ML model for CSI compression according to the implementation scheme discussed herein is provided, wherein the AI / ML model is trained using a collaborative process that employs separate training between the network (e.g., a base station of the network) and the UE.

[0019] Figure 12 An illustration is shown demonstrating an explicit calculation of CQI / RI that can be performed by the UE.

[0020] Figure 13 An example is given of using the encoder of the AI / ML model at the UE to generate a diagram corresponding to the input CQI / RI according to the implementation of this paper.

[0021] Figure 14 An illustration is provided showing an example of reporting CSI / CQI / RI after AI / ML model alignment between the network and the UE according to the implementation described herein.

[0022] Figure 15 A method for a UE according to the implementation scheme of this document is illustrated.

[0023] Figure 16 A method for a UE according to the implementation scheme of this document is illustrated.

[0024] Figure 17 A method for a base station according to the implementation scheme described herein is illustrated.

[0025] Figure 18 A method for a UE according to the implementation scheme of this document is illustrated.

[0026] Figure 19 A method for a UE according to the implementation scheme of this document is illustrated.

[0027] Figure 20 A method for a base station according to the implementation scheme described herein is illustrated.

[0028] Figure 21 An example architecture of a wireless communication system according to the implementation scheme disclosed herein is illustrated.

[0029] Figure 22 A system for performing signaling transmission between a wireless device and a network device according to an embodiment disclosed herein is illustrated. Detailed Implementation

[0030] Various implementations are described with respect to the UE. However, references to the UE are provided for illustrative purposes only. The example implementations can be used with any electronic component capable of establishing a connection to a network and configured with hardware, software, and / or firmware for exchanging information and data with the network. Therefore, the UE as described herein is used to represent any suitable electronic component.

[0031] Figure 1 Figure 100 illustrates an example of a two-sided artificial intelligence (AI) / machine learning (ML) model 102 for channel state information (CSI) compression and decompression according to the implementation scheme discussed herein. As illustrated, model 102 includes a (logical) UE side 104 (the portion of model 102 present at UE 106) and a (logical) network side 108 (the portion of model 102 present at a base station, such as network 110).

[0032] The UE side 104 illustrates that UE 106 includes an encoder 112. Encoder 112 is configured to accept input 116 and provide a bitstream 118 based on that input 116 as output. As illustrated, in some cases, input 116 may include downlink (DL) channel information, such as raw DL channel estimates. In some cases, input 116 may include pre-decoder information, such as a pre-decoder W and / or a set of feature vectors corresponding to the pre-decoder W.

[0033] Then bit stream 118 is sent from UE 106 to network 110.

[0034] Network side 108 illustrates network 110 (e.g., a base station of network 110) including decoder 114. Decoder 114 is configured to accept bitstream 118 as input and decode the information in that bitstream into output 120 for further processing at network 110. In this way, the raw information from input 116 is known to network 110. Therefore, in some cases, output 120 may include downlink (DL) channel information, such as raw DL channel estimates. In some cases, output 120 may include pre-decoder information, such as a pre-decoder W and / or a set of feature vectors corresponding to the pre-decoder W.

[0035] Based on the encoding mechanism used by encoder 112, bitstream 118 can be smaller than the original data presented in input 116. Therefore, encoder 112 can be understood as “compressing” input 116 into a bitstream, which is correspondingly understood as representing “compressed” information. As a result, transmitting bitstream 118 from UE 106 to network 110 uses fewer radio resources compared to an alternative where the original data (e.g., as present in input 116) is transmitted from UE 106 to network 110 without such encoding / compression.

[0036] Model 102 (including UE-side 104 and network-side 108) can be a trained AI / ML model. Various AI / ML model training collaborations can be considered for the purpose of training a two-sided model for use cases such as CSI compression (as in model 102).

[0037] In this paper, the term "joint training" may refer to the generative model (e.g., encoder 112) and the reconstruction model (e.g., decoder 114) being trained in the same loops used for forward and backward propagation, respectively. Joint training can be performed at a single network entity or across multiple network entities. For example, joint training can be performed through gradient exchange between network entities.

[0038] In this paper, the term "independent training" may refer to performing sequential and / or parallel training relative to the generative model (e.g., encoder 112) and the reconstruction model (e.g., decoder 114). Sequential training may begin with training at the UE and then proceed to training at the network side, or it may begin with training at the network side and then proceed to training at the UE side. In parallel training, training at the UE side and training at the network side are performed simultaneously.

[0039] The first type of AI / ML model training collaboration may involve joint training of bilateral models at a single-sided / network entity. In this case, the training can be on the UE side or the network side. Figure 2A diagram 200 illustrates a training 202 performed in a manner that is combined 204 and executed at either UE 106 or network 110.

[0040] As another example, the second type of AI / ML model training collaboration may include joint training of the two-sided model at both the network side and the UE side. Figure 3 A diagram 300 illustrates training 302 performed in a manner that is joint 304 and across each of UE 106 and network 110 (each of UE 106 and network 110 is an active participant in training 302).

[0041] As another example, a third type of AI / ML model training collaboration may include the individual training of each of UE 106 and Network 110. Figure 4 Figure 400 illustrates that the first training 402 of the encoder 112 on the UE side 104 of model 102 can be performed at the UE 106, while the second training 404 of the decoder 114 on the network side 108 of model 102 can be performed at the network 110.

[0042] Figure 5 A flowchart 500 illustrates an example process between network 502 and UE 504, which is used to train an AI / ML model for CSI feedback at the network using joint training.

[0043] Network 502 transmits 506 a first channel state information reference signal (CSI-RS) to UE 504. UE 504 receives the first CSI-RS and uses it to calculate pre-decoder information (e.g., calculate the eigenvector of the channel).

[0044] As illustrated in the figure, the UE transmits pre-decoder information as training data to the network 502. The network 502 uses the training data from the UE (and its own knowledge of the characteristics of the transmitted CSI-RS) to perform joint model training / generation 510, where both the UE side (including, for example, the encoder) and the network side (including, for example, the decoder) of the AI / ML model are trained.

[0045] After training the AI / ML model at network 502, network 502 transmits the AI / ML model to the UE side at 512 to UE 504.

[0046] On the UE side utilizing the AI / ML model at the UE, the inference phase is now available. Network 502 transmits 514 a second CSI-RS to UE 504. UE 504 measures the channel and calculates pre-decoder information. UE 504 then performs encoder inference 516 by feeding the pre-decoder information to the encoder on the UE side of the AI / ML model. As shown, UE 504 then reports 518 the encoder results to network 502 in the form of compressed CSI.

[0047] Network 502 then applies the compressed CSI to the decoder on the network side of the AI / ML model to generate decoder inference 520, thereby recovering the pre-decoder information encoded at UE 504 during encoder inference 516.

[0048] In some cases, when joint training of a two-sided model is performed at a single-sided / network entity as described herein, a mechanism for exchanging AI / ML models between the network and the UE is used. It may be necessary to specify a new data format / representation / procedure for signaling the AI / ML model between the UE and the network. Furthermore, the transfer of the AI / ML model between entities will utilize air interface resources. Moreover, when the model becomes outdated and needs retraining, a new corresponding transfer will be required (a process that may become undesirably expensive in terms of resources). Additionally, there may be intellectual property issues related to the AI / ML model, which may preclude allowing this transfer between the UE and the network. Furthermore, it may be necessary to specify a new data format / representation / procedure for signaling the training data between the UE and the network.

[0049] In other cases, when joint training of the two-sided model is performed at both the network and UE sides, there may be extensive forward and backward propagation information exchange over the air interface to facilitate joint training across network entities. Furthermore, it may be necessary to specify new data formats / representations / procedures for signaling this information between the UE and the network.

[0050] The implementation schemes disclosed herein relate to collaborative training processes using AI / ML models trained individually and in parallel. In such processes, compared to joint training of bilateral models at a single-sided / network entity, no training data transfer or model transfer is required (these are indicated by “X” in flowchart 500 to show that they will not be required in such cases).

[0051] Furthermore, in such processes, compared to joint training of the two-sided model at both the network side and the UE side, there is no need for backpropagation information exchange, thus saving air interface resources.

[0052] The collaborative training process using AI / ML models trained individually and in parallel allows for iterative training to calibrate the AI / ML models.

[0053] Figure 6 A flowchart 600 illustrating the training and use of an AI / ML model for CSI compression according to the implementation scheme discussed herein is provided, wherein the AI / ML model is trained using a collaborative process that employs separate training between network 602 (e.g., a base station of network 602) and UE 604.

[0054] Flowchart 600 begins with training phase 606. As part of training phase 606, network 602 transmits 608 CSI-RS to UE 604. This CSI-RS is used by UE 604 to generate DL channel estimation 612 at the UE. In the implementation of flowchart 600, DL channel estimation 612 is used to determine pre-decoder information in the form of feature vector 614 corresponding to the pre-decoder of the channel. Feature vector 614 is used to train UE AI / ML model 616, which includes UE encoder 618 and UE decoder 620 (note that, in this case, both are located at the UE). As illustrated, UE encoder 618 is configured to encode feature vector 614 into a bitstream 622, which can be decoded back to feature vector 614 by UE decoder 620. Bitstream 622 can be smaller / compressed compared to the original representation of feature vector 614.

[0055] Continuing with training phase 606: UE 604 transmits an uplink (UL) sounding reference signal (SRS) 610 to network 602. This UL SRS is used by network 602 to generate a UL channel estimate 624 at the network. In the implementation of flowchart 600, the UL channel estimate 624 is used to determine pre-decoder information in the form of a feature vector 626 corresponding to the pre-decoder of the channel. The feature vector 626 is used to train a network AI / ML model 628, which includes a network encoder 630 and a network decoder 632 (note that, in this case, both are located at network 602). As illustrated, the network encoder 630 is configured to encode the feature vector 626 into a bitstream 634, which can be decoded back into the feature vector 626 by the network decoder 632. The bitstream 634 can be smaller / compressed compared to the original representation of the feature vector 626.

[0056] Flowchart 600 then proceeds to the model alignment stage 636. The purpose of model alignment stage 636 is to align each of the UE AI / ML model 616 and the network AI / ML model 628 so that (at least) pre-decoder information (feature vectors) encoded into a compressed bitstream by the UE encoder 618 of the UE AI / ML model 616 can be correctly (or at least sufficiently accurately) decoded from the compressed bitstream by the network decoder 632 of the network AI / ML model 628.

[0057] As illustrated in the figure, network 602 transmits 638 CSI-RS for DL ​​channel estimation to UE 604. UE 604 can use this CSI-RS to generate the raw DL channel estimate (represented as H in flowchart 600). DL Then, the original DL channel estimation is used to determine the pre-decoder information (represented as W in flowchart 600). UE The pre-decoder information indicates that, if used by the base station, the pre-decoder should cause UE 604 to experience the desired valid DL channel (represented as H in flowchart 600) on DL transmissions using the pre-decoder. eff (For example, H) eff = H DL W UE In the example of flowchart 600, the pre-decoder information may be or include a feature vector corresponding to the pre-decoder used for the channel.

[0058] The UE encoder 618 then encodes the pre-decoder information into a compressed CSI bitstream (this bitstream is also represented as PMI in flowchart 600). UE UE 604 then transmits a 640-bit stream to network 602.

[0059] Network 602 then uses network decoder 632 to decode the pre-decoder information from the bitstream (as shown as W in flowchart 600 for the pre-decoder information decoded at network 602). NW It should be noted that due to the differences between the (separately trained) UE AI / ML model 616 and the network AI / ML model 628, such as the pre-decoder information (W) decoded from the bitstream at network 602, NW This may be related to the pre-decoder information (W) encoded into a bitstream at UE 604. UE )different.

[0060] Network 602 then uses the pre-decoder information (W) decoded from the bitstream. NWThe UE 604 generates a beamformed CSI-RS and transmits this beamformed CSI-RS to the UE 604. Upon receiving the beamformed CSI-RS, the UE 604 uses it to calculate the perceived effective DL channel (“actual effective DL channel”, denoted as K in flowchart 600). eff UE 604 then calculates the previously determined expected effective DL channel (H). eff ) and the actual effective DL channel (Ĥ) determined by using the UE encoder 618 and network decoder 632 and the transmission 642 of beamforming CSI-RS as already described. eff A similarity measure between the two (e.g., generalized cosine similarity (GCS)) is used to determine the error 644 (e.g., “model mismatch” as represented in flowchart 600).

[0061] If the error 644 is higher than the threshold, the error 644 can be used to drive the update of the UE AI / ML model 616 on the UE side (e.g., update 646 of the UE encoder 618 of the UE AI / ML model 616, as illustrated in the figure) to reduce the error 644, thereby improving the model alignment between the UE AI / ML model 616 and the network AI / ML model 628.

[0062] After the UE AI / ML model 616 has been updated at UE 604, the various functions previously described can be repeated. For example, the UE uses the UE encoder 618 to encode pre-decoder information into a bitstream. Because the UE encoder 618 of the UE AI / ML model 616 has changed from its previous state to compensate for previous errors 644 (as just described), this bitstream may be different from the previously transmitted bitstream. UE 604 then transmits this (new) bitstream 648 to network 602. The network then (again) decodes the bitstream into (new) network-side pre-decoder information W in the manner already described. NW The UE then uses this new pre-decoder information to transmit another beamforming CSI-RS to UE 604. The UE can then calculate the (new) actual effective DL channel K. eff and compare it with the (original) expected effective DL channel H eff The comparison was performed to calculate the (new) error of 644.

[0063] When the error 644 remains above the threshold (e.g., a small value), another update 646 corresponding to the error 644 can be performed relative to the UE encoder 618 of the UE AI / ML model 616, and further repetitions such as those just described can be performed to try to eventually drive the error 644 within the threshold.

[0064] If the error 644 falls within the threshold at any point, UE 604 can determine that the UE encoder 618 of the UE AI / ML model 616 is accurate (e.g., aligned with the network decoder 632 of the network AI / ML model 628, such that the bitstream encoded by the UE encoder 618 at UE 604 can be decoded by the network decoder 632 of the network AI / ML model 628 with acceptable accuracy), and discard any upcoming repetitions. Furthermore, UE 604 transmits 650 to network 602 an indication that the UE encoder 618 of the UE AI / ML model 616 is accurate, so that the network does not expect such repetitions to be used for model calibration purposes in the future.

[0065] Flowchart 600 then proceeds to operation phase 652. In operation phase, network 602 transmits 654 CSI-RS to UE 604. The transmission of this CSI-RS can be used for substantive channel sounding (i.e., for purposes other than model alignment).

[0066] UE 604 uses the CSI-RS to perform channel estimation and calculates the corresponding pre-decoder information. This pre-decoder information is then encoded into a bitstream using the UE encoder 618 of the UE AI / ML model 616. UE 604 then transmits the 656-bit stream to network 602, which decodes the bitstream back to the pre-decoder information using the network decoder 632 of the network AI / ML model 628. Due to the preceding operations of the model alignment phase 636 as described, the pre-decoder information processed by each of UE 604 and network 602 is aligned (e.g., identical, or at least similar within acceptable accuracy), making the actual channel detection itself accurate.

[0067] Figure 7 A flowchart 700 illustrating the training and use of an AI / ML model for CSI compression according to the implementation scheme discussed herein is illustrated, wherein the AI / ML model is trained using a collaborative process that employs separate training between network 702 (e.g., a base station of network 702) and UE 704.

[0068] Flowchart 700 begins with training phase 606. Training phase 606 and its related elements, as illustrated in flowchart 700, can be explained as per the description of flowchart 600.

[0069] Flowchart 700 then proceeds to the model alignment stage 706. The purpose of the model alignment stage 706 is to align each of the UE AI / ML model 616 and the network AI / ML model 628 so that the pre-decoder information (feature vector) encoded into a compressed bitstream by the UE encoder 618 of the UE AI / ML model 616 can be correctly decoded from the compressed bitstream by the network decoder 632 of the network AI / ML model 628.

[0070] As illustrated in the figure, network 702 transmits 708 CSI-RS for DL ​​channel estimation to UE 704. The result can be the raw DL channel estimate (represented as H in flowchart 700). DL Then, the original DL channel estimation is used to determine the pre-decoder information (represented as W in flowchart 700). UE The pre-decoder information indicates that, if used by the base station, the pre-decoder should cause UE 704 to experience the desired valid DL channel (represented as H in flowchart 700) on DL transmissions using the pre-decoder. eff (For example, H) eff = H DL W UE In the example of flowchart 700, the pre-decoder information may be or include a feature vector corresponding to the pre-decoder used for the channel.

[0071] The pre-decoder information is then encoded into a compressed CSI bitstream by the UE encoder 618 (this bitstream is also represented as PMI in flowchart 700). UE ). UE 704 then transmits a 710-bit stream to network 702.

[0072] Network 702 then uses network decoder 632 to decode the pre-decoder information from the bitstream (as shown in flowchart 700, the pre-decoder information decoded at network 702 is represented as W). NW It should be noted that due to the differences between the (separately trained) UE AI / ML model 616 and the network AI / ML model 628, such as the pre-decoder information (W) decoded from the bitstream at network 702, NW This may be related to the pre-decoder information (W) encoded into a bitstream at UE 704. UE )different.

[0073] Network 602 then generates a beamformed CSI-RS based on the pre-decoder information decoded from the bitstream and transmits the beamformed CSI-RS to UE 704 712. Upon receiving the beamformed CSI-RS, UE 704 uses it to calculate the perceived actual effective DL channel (K in flowchart 700). effUE 704 then calculates the previously determined expected effective DL channel (H). eff ) and the actual effective DL channel (Ĥ) determined by using the UE encoder 618 and network decoder 632 and the transmission 712 of beamforming CSI-RS as already described. eff A similarity measure (e.g., GCS) between them is used to determine the error 714.

[0074] If the error exceeds a threshold (e.g., a small value), the error encoder 716 can be used to encode the error 714 to generate an error code 718 corresponding to the error 714. It should be noted that the error encoder 716 can be an encoder other than the UE encoder 618 of the UE AI / ML model 616. It is anticipated that in some cases, the error encoder 716 can be an encoder of another AI / ML model used at the UE (in addition to the UE AI / ML model 616).

[0075] UE 704 then transmits error code 720 718 to network 702. Error code 718 can be used to drive an update 722 to the network AI / ML model 628 at the network side (e.g., an update 722 to the network decoder 632 of the network AI / ML model 628, as illustrated) to reduce error 714, thereby improving model alignment between UE AI / ML model 616 and network AI / ML model 628.

[0076] After the network AI / ML model 628 has been updated at network 702, various functions can be repeated. For example, network 702 can use network decoder 632 to decode the (originally received) bitstream again. Since the network decoder 632 of the network AI / ML model 628 changes from its previous state (e.g., driven by error code 718) (as just described), the resulting pre-decoder information W NW Possibly related to the previous W NW Different. Network 702 then according to the new W NW Generate and transmit the (new) beamformed CSI-RS. Using the new beamformed CSI-RS, the UE calculates the (new) actual effective DL channel K. eff and compare it with the (original) expected effective DL channel H eff A comparison is made to calculate the (new) error 714. If the error 714 remains above a threshold, the error encoder 716 can be used to generate another error code 718, which is transmitted to the network 702 to drive another update 722 to the network AI / ML model 628.

[0077] It should be noted that when the error 714 remains above the threshold, further repetitions, such as those just described, can be performed to try to eventually drive the error 714 into the threshold.

[0078] If the error 714 falls within the threshold at any point, UE 704 can determine that the network decoder 632 of the network AI / ML model 628 is accurate (e.g., aligned with the UE encoder 618 of the UE AI / ML model 616, such that the bitstream encoded by the UE encoder 618 at UE 704 can be decoded by the network decoder 632 of the network AI / ML model 628 with acceptable accuracy), and discard any upcoming repetitions. Furthermore, UE 704 transmits 724 to network 702 an indication that the UE encoder 618 of the UE AI / ML model 616 is accurate, so that the network does not expect such repetitions to be used for model calibration purposes in the future.

[0079] Flowchart 700 then proceeds to operation phase 652. Operation phase 652 may proceed as explained with respect to flowchart 600.

[0080] Figure 8 A flowchart 800 illustrating the training and use of an AI / ML model for CSI compression according to the implementation scheme discussed herein is illustrated, wherein the AI / ML model is trained using a collaborative process that employs separate training between network 802 (e.g., a base station of network 802) and UE 804.

[0081] Flowchart 800 begins with training phase 606. Training phase 606 and its related elements, as illustrated in flowchart 800, can be explained as per the description of flowchart 600.

[0082] Flowchart 800 then proceeds to the model alignment stage 806. The purpose of model alignment stage 806 is to align each of the UE AI / ML model 616 and the network AI / ML model 628 so that the pre-decoder information (feature vector) encoded into a compressed bitstream by the UE encoder 618 of the UE AI / ML model 616 can be correctly decoded from the compressed bitstream by the network decoder 632 of the network AI / ML model 628.

[0083] As illustrated in the figure, network 802 transmits 808 CSI-RS for DL ​​channel estimation to UE 804. The result can be the raw DL channel estimate (represented as H in flowchart 800). DL Then, the original DL channel estimation is used to determine the pre-decoder information (represented as W in flowchart 800). UEThe pre-decoder information indicates that, if used by the base station, the pre-decoder should cause UE 804 to experience the desired valid DL channel (represented as H in flowchart 800) on DL transmissions using the pre-decoder. eff (For example, H) eff = H DL W UE In the example of flowchart 800, the pre-decoder information may be or include a feature vector corresponding to the pre-decoder used for the channel.

[0084] The pre-decoder information is then encoded into a compressed CSI bitstream by the UE encoder 618 (this bitstream is also represented as PMI in flowchart 800). UE ). UE 804 then transmits an 810-bit stream to network 802.

[0085] Network 802 then uses network decoder 632 to decode the pre-decoder information from the bitstream (as shown in flowchart 800, the pre-decoder information decoded at network 802 is represented as W). NW It should be noted that due to the differences between the (separately trained) UE AI / ML model 616 and the network AI / ML model 628, such as the pre-decoder information (W) decoded from the bitstream at network 802, NW This may be related to the pre-decoder information (W) encoded into a bitstream at UE 804. UE )different.

[0086] Network 602 then generates a beamformed CSI-RS based on the pre-decoder information decoded from the bitstream and transmits the beamformed CSI-RS to UE 804. Upon receiving the beamformed CSI-RS, UE 804 uses it to calculate the perceived actual effective DL channel (K in flowchart 800). eff UE 804 then calculates the previously determined expected effective DL channel (H). eff ) and the actual effective DL channel (Ĥ) determined by using the UE encoder 618 and network decoder 632 and the transmission 812 of beamforming CSI-RS as already described. eff A similarity measure (e.g., GCS) between them is used to determine the error 814.

[0087] If the error 814 is higher than a threshold (e.g., a small value), the error 814 can be used to drive an update of the UE AI / ML model 616 on the UE side (e.g., an update 816 of the UE encoder 618 of the UE AI / ML model 616, as illustrated in the figure) to reduce the error 814, thereby improving the model alignment between the UE AI / ML model 616 and the network AI / ML model 628.

[0088] Furthermore, error 814 can be encoded using error encoder 818 to generate error code 820 corresponding to error 814. It should be noted that error encoder 818 can be an encoder other than the UE encoder 618 of UE AI / ML model 616. It is anticipated that in some cases, error encoder 818 can be an encoder of another AI / ML model (other than UE AI / ML model 616) used at UE 804.

[0089] UE 804 then transmits error code 822 to network 802. Error code 820 can be used to drive an update 824 to the network AI / ML model 628 at the network side (e.g., an update 824 to the network decoder 632 of the network AI / ML model 628, as illustrated) to reduce error 814, thereby improving model alignment between UE AI / ML model 616 and network AI / ML model 628.

[0090] After the UE encoder 618 has been updated at the UE AI / ML model 616 and the network AI / ML model 628 has been updated at the network AI / ML model 628, various functions can be repeated. For example, UE 804 uses the UE encoder 618 to encode pre-decoder information into a bitstream. Because the UE encoder 618 of the UE AI / ML model 616 has changed from its previous state to compensate for previous errors 814 (as just described), this bitstream may be different from the previously transmitted bitstream. UE 804 then transmits 826 this (new) bitstream to the network 802.

[0091] The network then decodes the (new) bitstream into (new) network-side pre-decoder information W in the manner already described (using its updated network decoder 632). NW The new pre-decoder information is then used to transmit another beamformed CSI-RS to UE 804. UE 804 can then calculate the (new) actual effective DL channel K. eff and compare it with the (original) expected effective DL channel H eff The comparison was performed to calculate the (new) error of 814.

[0092] While error 814 remains above the threshold, another update 816 corresponding to error 814 can be performed relative to the UE encoder 618 of the UE AI / ML model 616, and another error code 820 can be generated by the error encoder 818 using error 814 and transmitted to the network 802 to drive another update 824 to the network decoder 632 of the network AI / ML model 628. Further repetitions, such as those just described, can then be performed in an attempt to eventually drive error 814 within the threshold.

[0093] If the error 814 falls within a threshold at any point, UE 804 can determine that each of the UE encoder 618 of the UE AI / ML model 616 and the network decoder 632 of the network AI / ML model 628 is accurate (e.g., aligned such that the bitstream encoded at UE 804 by the UE encoder 618 of the UE AI / ML model 616 can be decoded at network 802 with acceptable accuracy by the network decoder 632 of the network AI / ML model 628), and discard any upcoming repetitions. Furthermore, UE 804 transmits 828 to network 802 an indication that the UE encoder 618 of the UE AI / ML model 616 and / or the network decoder 632 of the network AI / ML model 628 are accurate, such that the network does not expect such repetitions to be used for model calibration purposes in the future.

[0094] Flowchart 800 then proceeds to operation phase 652. Operation phase 652 may proceed as explained with respect to flowchart 600.

[0095] Figure 9 A flowchart 900 illustrating the training and use of an AI / ML model for CSI compression according to the implementation scheme discussed herein is provided, wherein the AI / ML model is trained using a collaborative process that employs separate training between network 902 (e.g., a base station of network 902) and UE 904.

[0096] Flowchart 900 begins with training phase 906. As part of training phase 906, network 902 transmits 908 CSI-RS to UE 904. This CSI-RS is used by UE 904 to generate DL channel estimate 912 at UE 904. DL channel estimate 912 is used to train UE AI / ML model 914, which includes UE encoder 916 and UE decoder 918 (note that in this case, both are located at the UE). As illustrated, UE encoder 916 is configured to encode DL channel estimate 912 into a bitstream 920, which can be decoded back into DL channel estimate 912 by UE decoder 918. Bitstream 920 can be smaller / compressed compared to the original representation of DL channel estimate 912.

[0097] Continuing with training phase 906: UE 904 transmits 910 UL SRS to network 902. This UL SRS is used by network 902 to generate UL channel estimate 922 at the network. In the implementation of flowchart 900, UL channel estimate 922 is used to train network AI / ML model 924, which includes network encoder 926 and network decoder 928 (note that in this case, both are located at the network). As illustrated, network encoder 926 is configured to encode UL channel estimate 922 into bitstream 920, which can be decoded back to UL channel estimate 922 by network decoder 928. Bitstream 930 can be smaller / compressed compared to the original representation of UL channel estimate 922.

[0098] Flowchart 900 then proceeds to the model alignment stage 932. The purpose of model alignment stage 932 is to align each of the UE AI / ML model 914 and the network AI / ML model 924 so that the original DL channel estimate, as encoded into a compressed bitstream by the UE encoder 916 of the UE AI / ML model 914, can be correctly decoded from the compressed bitstream by the network decoder 928 of the network AI / ML model 924.

[0099] As illustrated in the diagram, network 902 transmits 934 CSI-RS for DL ​​channel estimation to UE 904. UE 904 can use this CSI-RS to generate the raw DL channel estimate (represented as H in flowchart 900). DL Then, the original DL channel estimation is used to determine the pre-decoder information (represented as W in flowchart 900). UE This pre-decoder information indicates that, if used by the base station, the pre-decoder should cause the UE to experience the desired effective DL channel (represented as H in flowchart 900) on DL transmissions using this pre-decoder. eff (For example, H) eff = H DL W UE In the example of flowchart 900, the pre-decoder information may be or include a feature vector corresponding to the pre-decoder used for the channel.

[0100] The original DL channel estimate is then encoded into a compressed CSI bitstream by the UE encoder 916 (this bitstream is also represented as PMI in flowchart 900). UE UE 904 then transmits a 936-bit stream to network 902.

[0101] Network 902 then uses network decoder 928 to decode the raw DL channel estimate from the bitstream (in flowchart 900, the raw DL channel estimate decoded at network 902 is denoted as H). NWIt should be noted that due to the differences between the (separately trained) UE AI / ML model 914 and the network AI / ML model 924, such as the raw DL channel estimation (H) decoded from the bitstream at network 902, NW This may be related to the original DL channel estimate (H) encoded into a bitstream at UE 904. DL )different.

[0102] Raw DL channel estimation at network 902 (H NW Then it is used by network 902 to determine the pre-decoder information (represented as W in flowchart 900). NW This pre-decoder information indicates that, if used by the base station, the pre-decoder should cause UE 904 to experience the desired effective DL channel on DL transmissions using this pre-decoder, assuming H... NW Accurately reflects the channel conditions from the UE's perspective (e.g., assuming H... NW Equal to or at least close to H DL In the example of flowchart 900, the pre-decoder information may be or include a feature vector corresponding to the pre-decoder.

[0103] Network 902 then based on pre-decoder information (W NW The UE 904 generates a beamformed CSI-RS and transmits this beamformed CSI-RS to the UE 904. Upon receiving the beamformed CSI-RS, the UE 904 uses it to calculate the actual effective DL channel (denoted as K in flowchart 900). eff UE 904 then calculates the previously determined expected effective DL channel (H). eff ) and the actual effective DL channel (Ĥ) determined by using the UE encoder 916 and network decoder 928 and the transmission 938 of beamforming CSI-RS as already described. eff Similarity measures (e.g., GCS) between the models are used to determine the error 940 (e.g., “model mismatch” as represented in flowchart 900).

[0104] If the error 940 is higher than the threshold, the error 940 can be used to drive the update of the UE AI / ML model 914 on the UE side (e.g., update 942 of the UE encoder 916 of the UE AI / ML model 914, as illustrated in the figure) to reduce the error 940, thereby improving the model alignment between the UE AI / ML model 914 and the network AI / ML model 924.

[0105] After the UE AI / ML model 914 has been updated at UE 904, the various functions previously described can be repeated. For example, the UE uses the UE encoder 916 to encode the raw DL channel estimate into a bitstream. Because the UE encoder 916 of the UE AI / ML model 914 has changed from its previous state to compensate for the previous error 940 (as just described), this bitstream may be different from the previously transmitted bitstream. UE 904 then transmits this (new) bitstream to network 602. The network then (again) decodes the bitstream into the (new) network-side raw DL channel estimate H. NW And use it to derive the (new) network-side pre-decoder information W NW The new network-side pre-decoder information is used to transmit another beamforming CSI-RS to UE 904 in the manner already described. The UE can then calculate the (new) actual effective DL channel K. eff and compare it with the (original) expected effective DL channel H eff The comparison was performed to calculate the (new) error of 940.

[0106] When the error 940 remains above the threshold (e.g., a small value), another update 942 corresponding to the error 940 can be performed relative to the UE encoder 916 of the UE AI / ML model 914, and further repetitions such as those just described can be performed to try to eventually drive the error 940 within the threshold.

[0107] If the error 940 falls within the threshold at any point, UE 904 can determine that the UE encoder 916 of the UE AI / ML model 914 is accurate (e.g., aligned with the network decoder 928 of the network AI / ML model 924, such that the bitstream encoded by the UE encoder 916 at UE 904 can be decoded by the network decoder 928 of the network AI / ML model 924 with acceptable accuracy), and discard any upcoming repetitions. Furthermore, UE 904 transmits 946 to network 902 an indication that the UE encoder 916 of the UE AI / ML model 914 is accurate, so that the network does not expect such repetitions to be used for model calibration purposes in the future.

[0108] Flowchart 900 then proceeds to operation phase 948. During operation phase, network 902 transmits 950 CSI-RS to UE 904. This transmission of CSI-RS can be used for substantive channel sensing (i.e., for purposes other than model alignment).

[0109] UE 904 uses the CSI-RS to perform a raw DL channel estimate and encodes it into a bitstream using the UE encoder 916 of the UE AI / ML model 914. UE 904 then transmits the 952 bitstream to network 902, which decodes the bitstream back to the raw DL channel estimate using the network decoder 928 of the network AI / ML model 924 (which can then be used to derive pre-decoder information for DL ​​transmission). Due to the prior operations of the model alignment phase 932 as described, the raw DL channel estimates processed by each of UE 904 and network 902 are aligned (e.g., identical, or at least similar within acceptable accuracy), making the actual channel detection itself accurate.

[0110] Figure 10 A flowchart 1000 illustrating the training and use of an AI / ML model for CSI compression according to the implementation scheme discussed herein is illustrated, wherein the AI / ML model is trained using a collaborative process that employs separate training between network 1002 (e.g., a base station of network 1002) and UE 1004.

[0111] Flowchart 1000 begins with training phase 906. Training phase 906 and its related elements, as illustrated in flowchart 1000, can be explained as per flowchart 900.

[0112] Flowchart 1000 then proceeds to the model alignment stage 1006. The purpose of model alignment stage 1006 is to align each of the UE AI / ML model 914 and the network AI / ML model 924 so that the original DL channel estimate, as encoded into a compressed bitstream by the UE encoder 916 of the UE AI / ML model 914, can be correctly decoded from the compressed bitstream by the network decoder 928 of the network AI / ML model 924.

[0113] As illustrated in the figure, network 1002 transmits 1008 CSI-RS for DL ​​channel estimation to UE 704. The result can be the raw DL channel estimate (represented as H in flowchart 1000). DL Then, the original DL channel estimation is used to determine the pre-decoder information (represented as W in flowchart 1000). UE The pre-decoder information indicates that, if used by the base station, the pre-decoder should enable UE 1004 to experience the desired valid DL channel (represented as H in flowchart 1000) on DL transmissions using the pre-decoder. eff (For example, H) eff = H DL W UEIn the example of flowchart 1000, the pre-decoder information may be or include a feature vector corresponding to the pre-decoder used for the channel.

[0114] The original DL channel estimate is then encoded into a compressed CSI bitstream by the UE encoder 916 (this bitstream is also represented as PMI in flowchart 1000). UE UE 1004 then transmits a 1010 bit stream to network 1002.

[0115] Network 1002 then uses network decoder 928 to decode the raw DL channel estimate from the bitstream (in flowchart 1000, the raw DL channel estimate decoded at network 1002 is denoted as H). NW It should be noted that due to the differences between the (separately trained) UE AI / ML model 914 and the network AI / ML model 924, such as the raw DL channel estimation (H) decoded from the bitstream at network 1002, NW This may be related to the original DL channel estimate (H) encoded into a bitstream at UE 1004. DL )different.

[0116] The original DL channel estimation at network 1002 (H NW Then, at network 1002, it is used to determine the pre-decoder information (represented as W in flowchart 1000). NW This pre-decoder information indicates that, if used by the base station, the pre-decoder should cause UE 1004 to experience the desired effective DL channel on DL transmissions using this pre-decoder, assuming H... NW Accurately reflects the channel conditions from the UE's perspective (e.g., assuming H... NW Equal to or at least close to H DL In the example of flowchart 1000, the pre-decoder information may be or include a feature vector corresponding to the pre-decoder.

[0117] Network 1002 then based on pre-decoder information (W) NW The system generates a beamformed CSI-RS and transmits it to the UE 1004. Upon receiving the beamformed CSI-RS, the UE 1004 uses it to calculate the perceived effective DL channel (“actual effective DL channel”, denoted as K in flowchart 1000). eff UE 1004 then calculates the previously determined expected effective DL channel (H). eff ) and the actual effective DL channel (Ĥ) determined by using the UE encoder 916 and network decoder 928 and the transmission 1012 of beamforming CSI-RS as already described. effA similarity measure (e.g., GCS) between them is used to determine the error 1014.

[0118] If the error 1014 is higher than a threshold (e.g., a small value), the error encoder 1016 can be used to encode the error 1014 to generate an error code 1018 corresponding to the error 1014. It should be noted that the error encoder 1016 can be an encoder other than the UE encoder 916 of the UE AI / ML model 914. It is anticipated that in some cases, the error encoder 1016 can be an encoder of another AI / ML model used at the UE (other than the UE AI / ML model 914).

[0119] UE 1004 then transmits error code 1020 1018 to network 1002. Error code 1018 can be used to drive an update 1022 to the network AI / ML model 924 at the network side (e.g., an update 1022 to the network decoder 928 of the network AI / ML model 924, as illustrated) to reduce error 1014, thereby improving model alignment between UE AI / ML model 914 and network AI / ML model 924.

[0120] After the network AI / ML model 924 has been updated at network 1002, various functions can be repeated. For example, network 1002 can use network decoder 928 to decode the (originally received) bitstream again. Since the network decoder 928 of the network AI / ML model 924 changes from its previous state (e.g., driven by error code 1018) (as just described), the resulting original DL channel estimate H NW Possibly related to the previous H NW Different. Using the (new) H NW Network 1002 determines new network-side pre-decoder information W NW Network 1002 then, based on the new W... NW Generate and transmit the (new) beamformed CSI-RS. Using the new beamformed CSI-RS, the UE calculates the (new) actual effective DL channel K. eff and compare it with the (original) expected effective DL channel H eff The comparison was performed to calculate the (new) error of 1014.

[0121] It should be noted that when the error 1014 remains above the threshold, further repetitions, such as those just described, can be performed to try to eventually drive the error 1014 into the threshold.

[0122] If the error 1014 falls within the threshold at any point, UE 1004 can determine that the network decoder 928 of the network AI / ML model 924 is accurate (e.g., aligned with the UE encoder 916 of the UE AI / ML model 914, such that the bitstream encoded by the UE encoder 916 at UE 1004 can be decoded by the network decoder 928 of the network AI / ML model 924 with acceptable accuracy), and discard any upcoming repetitions. Furthermore, UE 1004 transmits 1024 to network 1002 an indication that the UE encoder 916 of the UE AI / ML model 914 is accurate, so that the network does not expect such repetitions to be used for model calibration purposes in the future.

[0123] Flowchart 1000 then proceeds to operation phase 948. Operation phase 652 can be performed as explained with respect to flowchart 900.

[0124] Figure 11 A flowchart 1100 illustrating the training and use of an AI / ML model for CSI compression according to the implementation scheme discussed herein is illustrated, wherein the AI / ML model is trained using a collaborative process that employs separate training between network 1102 (e.g., a base station of network 1102) and UE 1104.

[0125] Flowchart 1100 begins with training phase 906. Training phase 906 and its related elements, as illustrated in flowchart 1100, can be explained as per flowchart 900.

[0126] Flowchart 1100 then proceeds to the model alignment stage 1106. The purpose of the model alignment stage 1106 is to align each of the UE AI / ML model 914 and the network AI / ML model 924 so that the original DL channel estimate, as encoded into a compressed bitstream by the UE encoder 916 of the UE AI / ML model 914, can be correctly decoded from the compressed bitstream by the network decoder 928 of the network AI / ML model 924.

[0127] As illustrated in the figure, network 1102 transmits 1108 CSI-RS for DL ​​channel estimation to UE 1104. The result can be the raw DL channel estimate (represented as H in flowchart 1100). DL Then, the original DL channel estimation is used to determine the pre-decoder information (represented as W in flowchart 1100). UE The pre-decoder information indicates that, if used by the base station, the pre-decoder should enable UE 1104 to experience the desired effective DL channel (represented as H in flowchart 1100) on DL transmissions using the pre-decoder. eff (For example, H) eff = H DLW UE In the example of flowchart 1100, the pre-decoder information may be or include a feature vector corresponding to the pre-decoder used for the channel.

[0128] The original DL channel estimate is then encoded into a compressed CSI bitstream by the UE encoder 916 (this bitstream is also represented as PMI in flowchart 1100). UE ). UE 1104 then transmits a 1110 bit stream to network 902.

[0129] Network 1102 then uses network decoder 928 to decode the raw DL channel estimate from the bitstream (in flowchart 1100, the raw DL channel estimate decoded at network 1102 is denoted as H). NW It should be noted that due to the differences between the (separately trained) UE AI / ML model 914 and the network AI / ML model 924, such as the raw DL channel estimation (H) decoded from the bitstream at network 1102, NW This may be related to the original DL channel estimate (H) encoded into a bitstream at UE 1104. DL )different.

[0130] Raw DL channel estimation at the network (H NW Then, at network 1102, it is used to determine the pre-decoder information (represented as W in flowchart 1100). NW This pre-decoder information indicates a pre-decoder that, if used by the base station, should cause UE 1104 to experience the desired effective DL channel on DL transmissions using this pre-decoder, assuming H... NW Accurately reflects the channel conditions from the perspective of UE1104 (e.g., assuming H... NW Equal to or at least close to H DL In the example of flowchart 1100, the pre-decoder information may be or include a feature vector corresponding to the pre-decoder.

[0131] Network 1102 then based on the pre-decoder information (W) NW The UE 1104 generates a beamformed CSI-RS and transmits the beamformed CSI-RS to the UE 1104. Upon receiving the beamformed CSI-RS, the UE 1104 uses it to calculate the perceived effective DL channel (“actual effective DL channel”, denoted as K in flowchart 1100). eff UE 1104 then calculates the previously determined expected effective DL channel (H). eff ) and the actual effective DL channel (Ĥ) determined by using the UE encoder 916 and network decoder 928 and the transmission 1112 of CSI-RS with beamforming as already described. effA similarity measure (e.g., GCS) between them is used to determine the error 1114.

[0132] If the error 1114 is higher than a threshold (e.g., a small value), the error 1114 can be used to drive an update of the UE AI / ML model 914 on the UE side (e.g., an update 1116 of the UE encoder 916 of the UE AI / ML model 914, as illustrated in the figure) to reduce the error 1114, thereby improving the model alignment between the UE AI / ML model 914 and the network AI / ML model 924.

[0133] Furthermore, error encoder 1118 can be used to encode error 1114 to generate error code 1120 corresponding to error 1114. It should be noted that error encoder 1118 can be an encoder other than the UE encoder 916 of the UE AI / ML model 914. It is anticipated that in some cases, error encoder 1118 can be an encoder for another AI / ML model (other than the UE AI / ML model 914) used at UE 1104.

[0134] UE 1104 then transmits error code 1122 to network 1102. Error code 1120 can be used to drive an update 1124 to the network AI / ML model 924 at the network side (e.g., an update 1124 to the network decoder 928 of the network AI / ML model 924, as illustrated) to reduce error 1114, thereby improving model alignment between UE AI / ML model 914 and network AI / ML model 924.

[0135] After the UE encoder 916 has been updated at the UE AI / ML model 914 and the network AI / ML model 924 has been updated at the network AI / ML model 924, various functions can be repeated. For example, UE 1104 uses the UE encoder 916 to encode pre-decoder information into a bitstream. Because the UE encoder 916 of the UE AI / ML model 914 changes from its previous state to compensate for previous errors 1114 (as just described), this bitstream may be different from the previously transmitted bitstream. UE 1104 then transmits this (new) bitstream to the network 1102.

[0136] The network then decodes the (new) bitstream into (new) network-side pre-decoder information W in the manner already described (using its updated network decoder 928). NW The new pre-decoder information is then used to transmit another beamformed CSI-RS to UE 1104. UE 1104 can then calculate the (new) actual effective DL channel K. eff and compare it with the (original) expected effective DL channel H effThe comparison was performed to calculate the (new) error 1114.

[0137] While error 1114 remains above the threshold, another update 1116 corresponding to error 1114 can be performed relative to the UE encoder 916 of the UE AI / ML model 914. Another error code 1120 can be generated by the error encoder 1118 using error 1114 and transmitted to the network 1102 to drive another update 1124 to the network decoder 928 of the network AI / ML model 924. Further repetitions, such as those just described, can then be performed to attempt to eventually drive error 1114 within the threshold.

[0138] If the error 1114 falls within the threshold at any point, UE 1104 can determine that each of the UE encoder 916 of the UE AI / ML model 914 and the network decoder 928 of the network AI / ML model 924 is accurate (e.g., aligned such that the bitstream encoded by the UE encoder 916 at UE 1104 can be decoded by the network decoder 928 of the network AI / ML model 924 at network 1102 with acceptable accuracy), and discard any upcoming repetitions. Furthermore, UE 1104 transmits 1128 to network 1102 an indication that the UE encoder 916 of the UE AI / ML model 914 and / or the network decoder 928 of the network AI / ML model 924 are accurate, such that network 1102 does not expect such repetitions to be used for future model calibration purposes.

[0139] Flowchart 1100 then proceeds to operation phase 948. Operation phase 948 may proceed as explained with respect to flowchart 900.

[0140] In some wireless communication systems, the output of the CSI generation model can be explicit feedback (e.g., a compressed version of the original DL channel information). Another output of the CSI generation model can be implicit feedback (e.g., reusing or modifying the Rank Indicator (RI) / Channel Quality Index (CQI) / Pre-decoding Matrix Indicator (PMI) framework). In some cases, the output of the CSI generation model can be a combination of both. Feeding a compressed version of the original DL channel information achieves optimal feedback reporting quality.

[0141] In some cases, it is beneficial to send RI and CQI feedback in addition to the raw DL channel information. For example, to provide the base station with a more complete understanding of the channel conditions at the UE, it may be helpful to report both the channel and some measure of interference and noise.

[0142] CQI calculations utilize knowledge of the interference and / or noise experienced on the channel and may not be derived solely from raw DL channel information. For example, a CQI value can be obtained from a combination of channel measurements of CSI-RS and interference measurements of Channel State Information Interference Measurement (CSI-IM) resources. In some cases, the output of the network-side AI / ML model (e.g., the decoder) may not perfectly match the input of the UE-side AI / ML model (e.g., the encoder) (e.g., as discussed herein). In such cases, the reported CQI value may be irrelevant and / or misinterpreted by the base station, potentially leading to performance degradation. Where the UE possesses AI / ML decoder information, the UE can detect the mismatch and mitigate it by correcting the reported CQI value.

[0143] To provide RI and / or CQI information to the UE, further enhancements to the CSI-based AI / ML model discussed herein are anticipated. In such cases, CQI and / or RI information is expected to be encoded into a bitstream along with compressed CSI. This encoding can be performed when the bilateral AI / ML models at both the UE and the network have been aligned or calibrated (e.g., as discussed herein). In some cases, a model alignment flag (accuracy indicator) will mark the beginning of the CSI / CQI / RI report in the bitstream.

[0144] In some implementations, a relative reduction in UE computation can be achieved for CSI / CQI / RI reports based on AI / ML models using compressed CSI.

[0145] For example, consider the first scenario, where the UE performs channel estimation and uses it as input to an AI / ML model to encode / feed back CSI based on the channel estimation. The UE can then perform multiple calculations to estimate channel capacity and / or mutual information to find the optimal CQI / RI corresponding to the channel estimation. The UE can then report the CQI / RI to the network separately from the encoded CSI.

[0146] Figure 12 Figure 1200 illustrates an explicit calculation of CQI / RI that can be performed by the UE. First, the UE calculates the CQI / RI based on the (actual) effective DL channel H experienced at the UE. eff The Gaussian capacity 1202 and the substream capacity 1204 are calculated for each. The UE then performs a nonlinear mapping 1206 to generate mutual information 1208 (e.g., capacity and / or spectral efficiency information in bits per second per hertz (Hz)). The mutual information 1208 is fed into a lookup table (LUT) 1210, which is used to ultimately identify the CQI / RI 1212.

[0147] In contrast, if the UE alternatively performs the original DL channel estimation and / or pre-decoder information, and uses the original DL channel estimation and / or pre-decoder information as input to an AI / ML model trained to co-encode / feedback CSI / CQI / RI, then UE computation can be significantly reduced at the UE. This use avoids the need for explicit UE computation for estimating channel capacity and / or mutual information (e.g., as described with respect to Figure 1200).

[0148] In some implementations, the raw DL channel estimate and / or pre-decoder information at the UE is whitened to remove interference. The raw DL channel estimate and / or pre-decoder information can also be scaled by the signal-to-noise ratio (SNR) to more comprehensively reflect noise conditions. In some cases, either whitening or scaling of the channel estimate can be performed, while in others, both whitening and scaling of the channel estimate can be performed.

[0149] The AI / ML model at the UE can be enhanced (e.g., as discussed herein) to additionally estimate the CQI / RI based on training that utilizes the CQI / RI from the training data (e.g., as calculated based on the original DL channel estimates also used for training). For example, complex capacity expressions can be directly learned by the AI / ML based on the input channel estimation information. This training enables the encoder of such a model to generate the CQI / RI based on the original DL channel estimates (and / or by extending pre-decoder information corresponding to such original DL channel estimates).

[0150] Figure 13 A diagram 1300 illustrates an encoder 1308, according to an embodiment of this document, for generating a CQI / RI corresponding to an input using an AI / ML model at UE 1302. First, UE 1302 computes a DL channel estimate 1306. The UE may then perform whitening and / or SNR scaling 1310 on the DL channel estimate 1306 (or, in other embodiments, pre-decoder information computed at UE based on the DL channel estimate 1306). The result is then fed to an AI / ML model trained to use encoder 1308 to generate a compressed CSI 1312 (e.g., as discussed herein) and to predict the corresponding CQI / RI indication 1314 based on the training. Encoder 1308 may encode the compressed CSI 1312 into a bitstream 1316 (e.g., as already described herein). Encoder 1308 may also add the corresponding CQI / RI indication 1314 to the bitstream 1316, as illustrated (e.g., in uncompressed form).

[0151] Once bitstream 1316 reaches network 1304, decoder 1318 can be used to decode the compressed CSI 1312 back to DL channel estimation (or decode back to the corresponding pre-decoder information, as appropriate), according to the process discussed elsewhere herein. Furthermore, network 1304 can also receive (e.g., uncompressed) CQI / RI indication 1314 from bitstream 1316, thereby notifying the network of the corresponding CQI / RI.

[0152] Figure 14 Figure 1400 illustrates an example of reporting CSI / CQI / RI after AI / ML model alignment between network 1402 and UE 1404 according to the implementation described herein. The specific AI / ML models used at each of network 1402 and UE 1404, and the alignment process for aligning those models, may be described in any of flowcharts 600, 700, 800, 900, 1000, and / or 1100.

[0153] Figure 1400 illustrates operation phase 1406 (e.g., performed after the model alignment phase). Operation phase 1406 may correspond to, for example, either operation phase 652 and / or operation phase 948, as described with respect to the different embodiments disclosed herein.

[0154] During operation phase 1406, network 1402 transmits 1408 CSI-RS. UE 1404 uses CSI-RS to generate a DL channel estimate (or alternatively, a corresponding pre-decoder), which is encoded by an encoder of an AI / ML model for compressed CSI to generate compressed CSI, as described herein. Furthermore, based on the DL channel estimate (or alternatively, based on the corresponding pre-decoder), the encoder determines the CQI / RI value corresponding to the compressed CSI. UE 1404 then transmits 1410 a bitstream containing the compressed CSI and CQI / RI values ​​to network 1402, thereby informing network 1402 of the CQI / RI and the compressed CSI.

[0155] Figure 15A method 1500 for a UE according to an embodiment of this document is illustrated. Method 1500 includes calculating 1502 an initial DL channel estimate between the base station and the UE based on a first CSI-RS received from a base station. Method 1500 further includes determining 1504 pre-decoder information based on the initial DL channel estimate, the pre-decoder information representing a pre-decoder for use by the base station to cause the UE to experience a desired effective DL channel corresponding to a DL transmission from the base station. Method 1500 further includes encoding 1506 the pre-decoder information into a first bit stream using an encoder of an ML model at the UE. Method 1500 further includes transmitting 1508 the first bit stream to the base station. Method 1500 further includes calculating 1510 a first actual effective DL channel experienced at the UE based on a second CSI-RS received from the base station in response to the first bit stream. Method 1500 further includes calculating 1512 a first error between the desired effective DL channel and the first actual effective DL channel. Method 1500 further includes performing 1514 a first adjustment to the encoder of the ML model at the UE based on the first error.

[0156] In some implementations of method 1500, the predecoder information includes a set of feature vectors corresponding to the predecoder.

[0157] In some embodiments, method 1500 further includes: after performing a first adjustment to the encoder of the ML model at the UE, encoding pre-decoder information into a second bitstream using the encoder of the ML model at the UE; and transmitting the second bitstream to the base station. Some such embodiments further include: calculating a second actually effective DL channel experienced at the UE using a third CSI-RS received from the base station in response to the second bitstream; calculating a second error between the expected effective DL channel and the second actually effective DL channel; and performing a second adjustment to the encoder of the ML model at the UE based on the second error. Some other such embodiments further include: calculating a second actually effective DL channel experienced at the UE using a third CSI-RS received from the base station in response to the second bitstream; determining that the second error between the expected effective DL channel and the second actually effective DL channel is within a threshold; and based on determining that the second error is within the threshold, transmitting to the base station an indication that the encoder of the ML model at the UE is accurate.

[0158] In some implementations, method 1500 further includes performing initial training of the ML model at the UE using a third CSI-RS received from the base station prior to the first CSI-RS. Some such implementations also include transmitting a UL SRS to the base station in response to receiving the third CSI-RS.

[0159] In some implementations, method 1500 further includes: determining one or more of CQI and RI based on pre-decoder information at the encoder of the ML model at the UE; and including one or more of CQI and RI in the first bit stream before transmitting the first bit stream to the base station.

[0160] In some implementations, method 1500 further includes transmitting first error information corresponding to the first error to the base station.

[0161] Figure 16 A method 1600 for a UE according to an embodiment of this document is illustrated. Method 1600 includes calculating 1602 an initial DL channel estimate between the base station and the UE based on a first CSI-RS received from a base station. Method 1600 further includes determining 1604 pre-decoder information based on the initial DL channel estimate, the pre-decoder information representing a pre-decoder for use by the base station to cause the UE to experience a desired effective DL channel corresponding to a DL transmission from the base station. Method 1600 further includes encoding 1606 the pre-decoder information into a first bit stream using an encoder of an ML model at the UE. Method 1600 further includes transmitting 1608 the first bit stream to the base station. Method 1600 further includes calculating 1610 a first actual effective DL channel experienced at the UE based on a second CSI-RS received from the base station in response to the first bit stream. Method 1600 further includes calculating 1612 a first error between the desired effective DL channel and the first actual effective DL channel. Method 1600 further includes transmitting 1614 first error information corresponding to the first error to the base station.

[0162] In some implementations of method 1600, the predecoder information includes a set of feature vectors corresponding to the predecoder.

[0163] In some implementations, method 1600 further includes generating first error information by encoding the first error using a second ML model at the UE.

[0164] In some implementations, method 1600 further includes: using a third CSI-RS received from the base station to calculate a second actual effective DL channel experienced at the UE after transmitting the first error information; calculating a second error between the expected effective DL channel and the second actual effective DL channel; and transmitting second error information corresponding to the second error to the base station.

[0165] In some implementations, method 1600 further includes: using a third CSI-RS received from the base station to calculate a second actual effective DL channel experienced at the UE after transmitting the first error information; determining that a second error between the expected effective DL channel and the second actual effective DL channel is within a threshold; and based on determining that the second error is within the threshold, transmitting to the base station an indication that the decoder of the second ML model at the base station is accurate.

[0166] In some implementations, method 1600 further includes performing initial training of the ML model at the UE using a third CSI-RS received from the base station prior to the first CSI-RS. Some such implementations also include transmitting a UL SRS to the base station in response to receiving the third CSI-RS.

[0167] In some implementations, method 1600 further includes: determining one or more of CQI and RI based on pre-decoder information at the encoder of the ML model at the UE; and including one or more of CQI and RI in the first bit stream before transmitting the first bit stream to the base station.

[0168] Figure 17 A method 1700 for a base station according to an embodiment of this document is illustrated. Method 1700 includes transmitting 1702 a first CSI-RS to a UE. Method 1700 further includes receiving 1704 a first bit stream from the UE in response to the first CSI-RS. Method 1700 further includes decoding 1706 first pre-decoder information from the first bit stream using a decoder of a first ML model at the base station. Method 1700 further includes transmitting 1708 a second CSI-RS to the UE for beamforming according to the first pre-decoder represented by the first pre-decoder information.

[0169] In some implementations of method 1700, the first predecoder information includes a set of feature vectors corresponding to the predecoder.

[0170] In some implementations, method 1700 further includes: receiving a second bit stream from the UE in response to a second CSI-RS; decoding second pre-decoder information from the second bit stream using a decoder of the ML model at the base station; and transmitting a third CSI-RS to the UE for beamforming according to the second pre-decoder represented by the second pre-decoder information.

[0171] In some implementations, method 1700 also includes receiving from the UE an accurate indication of the encoder of the second ML model at the UE.

[0172] In some implementations, method 1700 further includes using ULSRS received from the UE prior to transmitting the first CSI-RS to perform initial training of the ML model at the base station.

[0173] In some implementations of method 1700, the bitstream also includes one or more of CQI and RI.

[0174] In some embodiments, method 1700 further includes: receiving first error information from the UE, the first error information corresponding to a first error between the expected effective DL channel and the first actually effective DL channel experienced by the UE corresponding to the second CSI-RS; and performing adjustment of the decoder of the first ML model at the base station based on the first error information. In some such embodiments, the first error information includes encoding of the first error. Some other such embodiments further include: after adjusting the decoder of the first ML model at the base station, decoding the first bitstream into second pre-decoder information using the decoder of the ML model at the base station; and transmitting to the UE a third CSI-RS beamforming according to the second pre-decoder represented by the second pre-decoder information. Some such embodiments further include receiving from the UE, after transmitting the third CSI-RS, an indication that the decoder of the ML model at the base station is accurate.

[0175] Figure 18 Method 1800 for a UE according to an embodiment of this document is illustrated. Method 1800 includes calculating 1802 an initial DL channel estimate between the base station and the UE based on a first CSI-RS received from a base station. Method 1800 further includes determining 1804 a desired effective DL channel corresponding to a DL transmission from the base station based on the initial DL channel estimate. Method 1800 further includes encoding 1806 the initial DL channel estimate into a first bit stream using an encoder of an ML model at the UE. Method 1800 further includes transmitting 1808 the first bit stream to the base station. Method 1800 further includes calculating 1810 a first actual effective DL channel experienced at the UE based on a second CSI-RS received from the base station in response to the first bit stream. Method 1800 further includes calculating 1812 a first error between the desired effective DL channel and the first actual effective DL channel. Method 1800 further includes performing 1814 a first adjustment to the encoder of the ML model at the UE based on the first error.

[0176] In some implementations, method 1800 further includes: after performing a first adjustment to the encoder of the ML model at the UE, encoding the original DL channel estimate into a second bitstream using the encoder of the ML model at the UE; and transmitting the second bitstream to the base station. Some such implementations also include: calculating a second actual effective DL channel experienced at the UE using a third CSI-RS received from the base station in response to the second bitstream; calculating a second error between the expected effective DL channel and the second actual effective DL channel; and performing a second adjustment to the encoder of the ML model at the UE based on the second error. Some other such implementations further include: calculating the second actual effective DL channel experienced at the UE using a third CSI-RS received from the base station in response to the second bitstream; determining that the second error between the expected effective DL channel and the second actual effective DL channel is within a threshold; and based on determining that the second error is within the threshold, transmitting to the base station an indication that the encoder of the ML model at the UE is accurate.

[0177] In some implementations, method 1800 further includes performing initial training of the ML model at the UE using a third CSI-RS received from the base station prior to the first CSI-RS. Some such implementations also include transmitting a UL SRS to the base station in response to receiving the third CSI-RS.

[0178] In some implementations, method 1800 further includes: determining one or more of CQI and RI based on the original DL channel estimation at the encoder of the ML model at the UE; and including one or more of CQI and RI in the first bit stream before transmitting the first bit stream to the base station.

[0179] In some implementations, method 1800 further includes transmitting first error information corresponding to the first error to the base station.

[0180] Figure 19A method 1900 for a UE according to an embodiment of this document is illustrated. Method 1900 includes calculating 1902 an initial DL channel estimate between the base station and the UE based on a first CSI-RS received from a base station. Method 1900 further includes determining 1904 a desired effective DL channel corresponding to a DL transmission from the base station based on the initial DL channel estimate. Method 1900 further includes encoding 1906 the initial DL channel estimate into a first bit stream using an encoder of an ML model at the UE. Method 1900 further includes transmitting 1908 the first bit stream to the base station. Method 1900 further includes calculating 1910 a first actual effective DL channel experienced at the UE based on a second CSI-RS received from the base station in response to the first bit stream. Method 1900 further includes calculating 1912 a first error between the desired effective DL channel and the first actual effective DL channel. Method 1900 further includes transmitting 1914 first error information corresponding to the first error to the base station.

[0181] In some implementations, method 1900 further includes generating first error information by encoding the first error using a second ML model at the UE.

[0182] In some implementations, method 1900 further includes: using a third CSI-RS received from the base station to calculate a second actual effective DL channel experienced at the UE after transmitting the first error information; calculating a second error between the expected effective DL channel and the second actual effective DL channel; and transmitting second error information corresponding to the second error to the base station.

[0183] In some implementations, method 1900 further includes: using a third CSI-RS received from the base station to calculate a second actual effective DL channel experienced at the UE after transmitting the first error information; determining that a second error between the expected effective DL channel and the second actual effective DL channel is within a threshold; and based on determining that the second error is within the threshold, transmitting to the base station an indication that the decoder of the second ML model at the base station is accurate.

[0184] In some implementations, method 1900 further includes performing initial training of the ML model at the UE using a third CSI-RS received from the base station prior to the first CSI-RS. Some such implementations also include transmitting a UL SRS to the base station in response to receiving the third CSI-RS.

[0185] In some implementations, method 1900 further includes: determining one or more of CQI and RI based on the original DL channel estimation at the encoder of the ML model at the UE; and including one or more of CQI and RI in the first bit stream before transmitting the first bit stream to the base station.

[0186] Figure 20 A method 2000 for a base station according to an embodiment of this document is illustrated. Method 2000 includes transmitting 2002 a first CSI-RS to a UE. Method 2000 also includes receiving 2004 a first bitstream from the UE in response to the first CSI-RS. Method 2000 further includes decoding 2006 a first raw DL channel estimate from the first bitstream using a decoder of a first ML model at the base station. Method 2000 further includes determining 2008 first pre-decoder information based on the first raw DL channel estimate, the first pre-decoder information representing a pre-decoder for use such that the UE experiences a desired effective DL channel corresponding to a DL transmission from the base station, under the conditions of the first raw DL channel estimate. Method 2000 further includes transmitting 2010 a second CSI-RS for beamforming according to the first pre-decoder to the UE.

[0187] In some implementations of method 2000, the predecoder information includes a set of feature vectors corresponding to the predecoder.

[0188] In some implementations, method 2000 further includes: receiving a second bitstream from the UE in response to a second CSI-RS; decoding a second raw DL channel estimate from the second bitstream using a decoder of an ML model at the base station; determining second pre-decoder information based on the second raw DL channel estimate, the second pre-decoder information representing a second pre-decoder for use by the base station such that the UE experiences a desired effective DL channel corresponding to a DL transmission from the base station under the conditions of the second raw DL channel estimate; and transmitting a third CSI-RS to the UE based on the second pre-decoder for beamforming.

[0189] In some implementations, method 2000 also includes receiving from the UE an accurate indication of the encoder of the second ML model at the UE.

[0190] In some implementations, method 2000 also includes using ULSRS received from the UE prior to transmitting the first CSI-RS to perform initial training of the ML model at the base station.

[0191] In some implementations of method 2000, the bitstream also includes one or more of CQI and RI.

[0192] In some embodiments, method 2000 further includes: receiving first error information from the UE, the first error information corresponding to a first error between the expected effective DL channel and the first actually effective DL channel experienced by the UE corresponding to the second CSI-RS; and performing adjustment of the decoder of the first ML model at the base station based on the first error information. In some such embodiments, the first error information includes encoding of the first error. Some other such embodiments further include: after adjusting the decoder of the first ML model at the base station, decoding the first bitstream into a second raw DL channel estimate using the decoder of the ML model at the base station; determining second pre-decoder information based on the second raw DL channel estimate, the second pre-decoder information representing a second pre-decoder for use by the base station such that the UE experiences the expected effective DL channel corresponding to the DL transmission from the base station under the conditions of the second raw DL channel estimate; and transmitting a third CSI-RS beamforming according to the second pre-decoder to the UE. Some such embodiments further include receiving from the UE, after transmitting the third CSI-RS, an indication that the decoder of the ML model at the base station is accurate.

[0193] Figure 21 An example architecture of a wireless communication system 2100 according to an embodiment disclosed herein is illustrated. The following description is for an example wireless communication system 2100 operating in conjunction with LTE system standards and / or 5G or NR system standards provided by 3GPP technical specifications.

[0194] like Figure 21 As shown, the wireless communication system 2100 includes UE 2102 and UE 2104 (but any number of UEs may be used). In this example, UE 2102 and UE 2104 are exemplified as smartphones (e.g., handheld touchscreen mobile computing devices capable of connecting to one or more cellular networks), but may also include any mobile or non-mobile computing device configured for wireless communication.

[0195] UE 2102 and UE 2104 can be configured to communicatively couple with RAN 2106. In an implementation, RAN 2106 can be NG-RAN, E-UTRAN, etc. UE 2102 and UE 2104 utilize connections (or channels) with RAN 2106 (shown as connection 2108 and connection 2110, respectively), where each connection includes a physical communication interface. RAN 2106 may include one or more base stations (such as base station 2112 and base station 2114) implementing connection 2108 and connection 2110.

[0196] In this example, connection 2108 and connection 2110 are air interfaces that enable this type of communication coupling and are compliant with the RAT used by RAN 2106, such as LTE and / or NR.

[0197] In some implementations, UE 2102 and UE 2104 can also directly exchange communication data via sidelink interface 2116. UE 2104 is shown configured to access an access point (shown as AP 2118) via connection 2120. For example, connection 2120 may include a local wireless connection, such as a connection conforming to any IEEE 802.11 protocol, while AP 2118 may include Wi-Fi. ® Router. In this example, AP 2118 may connect to another network (e.g., the Internet) without using CN 2124.

[0198] In the implementation, UE 2102 and UE 2104 may be configured to communicate with each other or with base station 2112 and / or base station 2114 on a multi-carrier communication channel using orthogonal frequency division multiplexing (OFDM) communication signals according to various communication technologies, such as, but not limited to, orthogonal frequency division multiple access (OFDMA) communication technology (e.g., for downlink communication) or single-carrier frequency division multiple access (SC-FDMA) communication technology (e.g., for uplink and ProSe or sidelink communication), but the scope of the implementation is not limited in this respect. The OFDM signal may include multiple orthogonal subcarriers.

[0199] In some implementations, all or some of the base stations in base station 2112 or base station 2114 may be implemented as one or more software entities running on a server computer as part of a virtual network. Furthermore, or in other implementations, base station 2112 or base station 2114 may be configured to communicate with each other via interface 2122. In implementations where the wireless communication system 2100 is an LTE system (e.g., when CN 2124 is an EPC), interface 2122 may be an X2 interface. This X2 interface may be defined between two or more base stations (e.g., two or more eNBs, etc.) connected to the EPC and / or between two eNBs connected to the EPC. In implementations where the wireless communication system 2100 is an NR system (e.g., when CN 2124 is a 5GC), interface 2122 may be an Xn interface. The Xn interface is defined between two or more base stations (e.g., two or more gNBs, etc.) connected to the 5GC, between a base station 2112 (e.g., a gNB) connected to the 5GC and an eNB, and / or between two eNBs connected to the 5GC (e.g., CN 2124).

[0200] RAN 2106 is shown communicatively coupled to CN 2124. CN 2124 may include one or more network elements 2126 configured to provide various data and telecommunications services to customers / subscribers (e.g., users of UE 2102 and UE 2104) connected to CN 2124 via RAN 2106. Components of CN 2124 may be implemented in a single physical device or a separate physical device, including components for reading and executing instructions from machine-readable or computer-readable media (e.g., non-transitory machine-readable storage media).

[0201] In the implementation scheme, CN 2124 may be an EPC, and RAN 2106 may be connected to CN 2124 via S1 interface 2128. In the implementation scheme, S1 interface 2128 may be divided into two parts: an S1 user plane (S1-U) interface, which carries service data between base station 2112 or base station 2114 and the serving gateway (S-GW); and an S1-MME interface, which is the signaling interface between base station 2112 or base station 2114 and the mobility management entity (MME).

[0202] In the implementation scheme, CN 2124 may be a 5GC, and RAN 2106 may be connected to CN 2124 via NG interface 2128. In the implementation scheme, NG interface 2128 may be divided into two parts: an NG user plane (NG-U) interface, which carries service data between base station 2112 or base station 2114 and user plane function (UPF); and an S1 control plane (NG-C) interface, which is the signaling interface between base station 2112 or base station 2114 and access and mobility management function (AMF).

[0203] Generally, application server 2130 may be an element that provides Internet Protocol (IP) bearer resources (e.g., packet-switched data services) for use with CN 2124. Application server 2130 may also be configured to support one or more communication services (e.g., VoIP sessions, group communication sessions, etc.) for UE 2102 and UE 2104 via CN 2124. Application server 2130 can communicate with CN 2124 via IP communication interface 2132.

[0204] Figure 22A system 2200 for performing signaling transmission 2234 between a wireless device 2202 and a network device 2218, according to an embodiment disclosed herein, is illustrated. System 2200 may be part of a wireless communication system as described herein. Wireless device 2202 may be, for example, a UE (User Equipment) of a wireless communication system. Network device 2218 may be, for example, a base station (e.g., an eNB or gNB) of a wireless communication system.

[0205] Wireless device 2202 may include one or more processors 2204. Processor 2204 is executable instructions that enable various operations of wireless device 2202 to be performed as described herein. Processor 2204 may include one or more baseband processors, which are implemented using, for example, a central processing unit (CPU), digital signal processor (DSP), application-specific integrated circuit (ASIC), controller, field-programmable gate array (FPGA) device, another hardware device, firmware device, or any combination thereof configured to perform the operations described herein.

[0206] Wireless device 2202 may include memory 2206. Memory 2206 may be a non-transitory computer-readable storage medium that stores instructions 2208, which may include, for example, instructions executed by processor 2204. Instructions 2208 may also be referred to as program code or a computer program. Memory 2206 may also store data used by processor 2204 and results calculated by the processor.

[0207] Wireless device 2202 may include one or more transceivers 2210, which may include radio frequency (RF) transmitter circuitry and / or receiver circuitry, which use antenna 2212 of wireless device 2202 to facilitate signaling transmissions to and / or from wireless device 2202 and other devices (e.g., network device 2218) according to the corresponding RAT (e.g., signaling transmission 2234).

[0208] Wireless device 2202 may include one or more antennas 2212 (e.g., one, two, four or more). In embodiments with multiple antennas 2212, wireless device 2202 may fully utilize the spatial diversity of such multiple antennas 2212 to transmit and / or receive multiple different data streams on the same time-frequency resource. This behavior may be referred to as, for example, multiple-input multiple-output (MIMO) behavior (referring to multiple antennas used at each of the transmitting and receiving devices to implement this aspect). MIMO transmission by wireless device 2202 may be achieved according to pre-decoding (or digital beamforming) applied at wireless device 2202, which multiplexes data streams across antennas 2212 based on known or assumed channel characteristics, such that each data stream is received with appropriate signal strength relative to the other streams at a desired location in the spatial domain (e.g., the location of the receiver associated with that data stream). Some implementations may use a single-user MIMO (SU-MIMO) approach (where all data streams are directed to a single receiver) and / or a multi-user MIMO (MU-MIMO) approach (where individual data streams may be directed to individual (different) receivers at different locations in the airspace).

[0209] In some implementations with multiple antennas, wireless device 2202 can implement analog beamforming technology, whereby the phase of the signal transmitted by antenna 2212 is relatively adjusted so that the (joint) transmission of antenna 2212 can be directed (this is sometimes referred to as beam control).

[0210] Wireless device 2202 may include one or more interfaces 2214. Interface 2214 can be used to provide input to or output to wireless device 2202. For example, wireless device 2202 as a UE may include interface 2214, such as a microphone, speaker, touchscreen, and buttons, to allow a user of the UE to make inputs and / or outputs to the UE. Other interfaces of such a UE may consist of transmitters, receivers, and other circuitry that allow the UE to communicate with other devices (e.g., in addition to the transceiver 2210 / antenna 2212 already described), and may be based on known protocols (e.g., Wi-Fi). ® and Bluetooth ® (etc.) to perform the operation.

[0211] Wireless device 2202 may include a CSI compression module 2216. The CSI compression module 2216 may be implemented via hardware, software, or a combination thereof. For example, the CSI compression module 2216 may be implemented as a processor, circuitry, and / or instructions 2208 stored in memory 2206 and executed by processor 2204. In some examples, the CSI compression module 2216 may be integrated within processor 2204 and / or transceiver 2210. For example, the CSI compression module 2216 may be implemented by a combination of software components (e.g., executed by a DSP or general-purpose processor) and hardware components (e.g., logic gates and circuitry) within processor 2204 or transceiver 2210.

[0212] The CSI compression module 2216 can be used in various aspects of this disclosure, such as Figure 6 , Figure 7 , Figure 8 , Figure 9 , Figure 10 , Figure 11 , Figure 13 , Figure 15 , Figure 16 , Figure 18 and / or Figure 19 In various aspects. For example, the CSI compression module 2216 can configure the wireless device 2202 to: calculate the raw DL channel estimate; determine pre-decoder information representing the pre-decoder used for the desired effective DL channel; encode the raw DL channel estimate into a bit stream using an encoder using an AI / ML model; encode the pre-decoder into a bit stream using an encoder using an AI / ML model; calculate the actual effective DL channel experienced at the UE based on the CSI-RS received in response to the transmitted bit stream; calculate the error between the desired effective DL channel and the actual effective DL channel; adjust the AI / ML model based on the error; and / or encode the error into an error code and transmit the error code to the base station, as described herein.

[0213] Network device 2218 may include one or more processors 2220. Processor 2220 is executable instructions to perform various operations of network device 2218 as described herein. Processor 2220 may include one or more baseband processors, which are implemented using, for example, a CPU, DSP, ASIC, controller, FPGA device, another hardware device, firmware device, or any combination thereof configured to perform the operations described herein.

[0214] Network device 2218 may include memory 2222. Memory 2222 may be a non-transitory computer-readable storage medium that stores instructions 2224, which may include, for example, instructions executed by processor 2220. Instructions 2224 may also be referred to as program code or a computer program. Memory 2222 may also store data used by processor 2220 and results calculated by the processor.

[0215] Network device 2218 may include one or more transceivers 2226, which may include RF transmitter circuitry and / or receiver circuitry that uses the antenna 2228 of network device 2218 to facilitate signaling transmissions (e.g., signaling transmission 2234) to and / or from network device 2218 and other devices (e.g., wireless device 2202) according to the corresponding RAT.

[0216] Network device 2218 may include one or more antennas 2228 (e.g., one, two, four or more). In embodiments having multiple antennas 2228, network device 2218 may perform MIMO, digital beamforming, analog beamforming, beam control, etc., as described.

[0217] Network device 2218 may include one or more interfaces 2230. Interfaces 2230 may be used to provide input to or output to network device 2218. For example, network device 2218 as a base station may include interfaces 2230 consisting of transmitters, receivers, and other circuitry (e.g., in addition to the transceiver 2226 / antenna 2228 already described), which enable the base station to communicate with other equipment in the core network and / or enable the base station to communicate with external networks, computers, databases, etc., for the purpose of performing operations, management, and maintenance of the base station or other equipment operatively connected to the base station.

[0218] Network device 2218 may include a CSI compression module 2232. The CSI compression module 2232 may be implemented via hardware, software, or a combination thereof. For example, the CSI compression module 2232 may be implemented as a processor, circuitry, and / or instructions 2224 stored in memory 2222 and executed by processor 2220. In some examples, the CSI compression module 2232 may be integrated within processor 2220 and / or transceiver 2226. For example, the CSI compression module 2232 may be implemented by a combination of software components (e.g., executed by a DSP or general-purpose processor) and hardware components (e.g., logic gates and circuitry) within processor 2220 or transceiver 2226.

[0219] The CSI compression module 2232 can be used in various aspects of this disclosure, such as Figure 6 , Figure 7 , Figure 8 , Figure 9 , Figure 10 , Figure 11 , Figure 13 , Figure 17 and / or Figure 20 In various aspects. For example, the CSI compression module 2232 can configure the network device 2218 to: decode the raw DL channel estimate from the bitstream received from the UE using the AI / ML model decoder; decode the pre-decoder information representing the pre-decoder for the desired effective DL channel from the bitstream received from the UE using the AI / ML model decoder; determine the pre-decoder information representing the pre-decoder for the desired effective DL channel based on the raw DL channel estimate; generate CSI-RS based on the pre-decoder information and transmit CSI-RS to the UE; and / or receive error codes from the UE corresponding to the error between the desired effective DL channel and the actual effective DL channel and use the error codes to adjust the AI / ML model decoder, as described herein.

[0220] The embodiments contemplated herein include an apparatus comprising components for performing one or more elements of any of method 1500, method 1600, method 1800, and / or method 1900. The apparatus may be, for example, a UE (such as wireless device 2202 as a UE, as described herein).

[0221] The embodiments contemplated herein include one or more non-transitory computer-readable media comprising instructions that, when executed by one or more processors of an electronic device, cause the electronic device to perform one or more elements of any of method 1500, method 1600, method 1800, and / or method 1900. The non-transitory computer-readable medium may be, for example, a memory of a UE (such as memory 2206 of a wireless device 2202 serving as a UE, as described herein).

[0222] The embodiments contemplated herein include an apparatus comprising logic components, modules, or circuitry for performing one or more elements of any of method 1500, method 1600, method 1800, and / or method 1900. The apparatus may be, for example, a UE (such as wireless device 2202 as a UE, as described herein).

[0223] The embodiments contemplated herein include an apparatus comprising: one or more processors; and one or more computer-readable media including instructions that, when executed by the one or more processors, cause the one or more processors to perform one or more elements of any of method 1500, method 1600, method 1800, and / or method 1900. The apparatus may be, for example, an apparatus of a UE (such as wireless device 2202 as a UE, as described herein).

[0224] The implementation scheme envisioned herein includes a signal as described in or related to one or more elements of any of method 1500, method 1600, method 1800, and / or method 1900.

[0225] The embodiments contemplated herein include a computer program or computer program product comprising instructions, wherein execution by a processor causes the processor to perform one or more elements of any of method 1500, method 1600, method 1800, and / or method 1900. The processor may be a processor of the UE (such as processor 2204 as a wireless device 2202 of the UE, as described herein). These instructions may, for example, reside in the processor and / or in the memory of the UE (such as memory 2206 as a wireless device 2202 of the UE, as described herein).

[0226] The embodiments contemplated herein include an apparatus comprising components for performing one or more elements of either method 1700 and / or method 2000. The apparatus may be, for example, an apparatus for a base station (such as network device 2218 as a base station, as described herein).

[0227] The embodiments contemplated herein include one or more non-transitory computer-readable media, which include instructions for causing the electronic device to perform one or more elements of either method 1700 and / or method 2000 when executed by one or more processors of the electronic device. The non-transitory computer-readable medium may be, for example, the memory of a base station (such as memory 2222 of a network device 2218 serving as a base station, as described herein).

[0228] The embodiments contemplated herein include an apparatus comprising logic components, modules, or circuitry for performing one or more elements of either method 1700 and / or method 2000. The apparatus may be, for example, an apparatus for a base station (such as network device 2218 as a base station, as described herein).

[0229] The embodiments contemplated herein include an apparatus comprising: one or more processors and one or more computer-readable media including instructions that, when executed by the one or more processors, cause the one or more processors to perform one or more elements of either method 1700 and / or method 2000. The apparatus may be, for example, an apparatus for a base station (such as network device 2218 as a base station, as described herein).

[0230] The implementation scheme envisioned herein includes a signal as described in or related to one or more elements of either method 1700 or method 2000.

[0231] The embodiments contemplated herein include a computer program or computer program product comprising instructions, wherein execution of the program by a processing element causes the processing element to perform one or more elements of either method 1700 and / or method 2000. The processor may be a processor of a base station (such as processor 2220 of network device 2218 as a base station, as described herein). These instructions may, for example, reside in the processor and / or in the memory of the base station (such as memory 2222 of network device 2218 as a base station, as described herein).

[0232] For one or more embodiments, at least one of the components illustrated in one or more of the foregoing figures may be configured to perform one or more operations, techniques, processes, and / or methods as described herein. For example, a baseband processor as described herein in conjunction with one or more of the foregoing figures may be configured to operate according to one or more of the examples illustrated herein. Similarly, circuitry associated with a UE, base station, network element, etc., as described above in conjunction with one or more of the foregoing figures may be configured to operate according to one or more of the examples illustrated herein.

[0233] Unless otherwise expressly stated, any of the embodiments described above may be combined with any other embodiment (or combination of embodiments). The foregoing description of one or more specific embodiments provides illustrative and descriptive information, but is not intended to be exhaustive or to limit the scope of the embodiments to the precise form disclosed. In light of the teachings above, modifications and variations are possible, or modifications and variations may be derived from practice with various embodiments.

[0234] Implementations and specific embodiments of the systems and methods described herein may include various operations embodied in machine-executable instructions to be executed by a computer system. The computer system may include one or more general-purpose or special-purpose computers (or other electronic devices). The computer system may include hardware components, including specific logical parts for performing the operations; or may include a combination of hardware, software, and / or firmware.

[0235] It should be recognized that the systems described herein include descriptions of specific implementations. These implementations may be combined into a single system, partially integrated into other systems, divided into multiple systems, or otherwise partitioned or combined. Furthermore, it is conceivable to use parameters, attributes, aspects, etc., of one implementation in one implementation. For clarity, these parameters, attributes, aspects, etc., are described only in one or more implementations, and it should be recognized that, unless expressly stated herein, these parameters, attributes, aspects, etc., may be combined with or substituted for parameters, attributes, aspects, etc., of another implementation.

[0236] As is widely recognized, the use of personally identifiable information should comply with privacy policies and practices that are generally accepted to meet or exceed industry or governmental requirements for protecting user privacy. Specifically, personally identifiable information data should be managed and processed to minimize the risk of unintentional or unauthorized access or use, and the nature of authorized use should be clearly explained to users.

[0237] Although the foregoing has been described in considerable detail for clarity, it will be apparent that certain changes and modifications can be made without departing from the principles of the invention. It should be noted that there are many alternative ways to implement both the processes and apparatus described herein. Therefore, embodiments of the invention should be considered illustrative rather than restrictive, and this specification is not limited to the details given herein, but can be modified within the scope and equivalents of the appended claims.

Claims

1. A method for a user equipment (UE), the method comprising: The raw downlink (DL) channel estimate between the base station and the UE is calculated based on the first channel state information reference signal (CSI-RS) received from the base station; Predecoder information is determined based on the original DL channel estimation. The predecoder information represents a predecoder for use by the base station to enable the UE to experience the expected valid DL channel corresponding to the DL transmission from the base station. The encoder of the machine learning (ML) model at the UE encodes the pre-decoder information into a first bit stream; The first bit stream is transmitted to the base station; The first actual effective DL channel experienced at the UE is calculated based on the second CSI-RS received from the base station in response to the first bit stream; Calculate the first error between the expected effective DL channel and the first actual effective DL channel; and The encoder of the ML model at the UE is first adjusted based on the first error.

2. The method according to claim 1, wherein the predecoder information includes a set of feature vectors corresponding to the predecoder.

3. The method according to claim 1, further comprising: After performing the first adjustment to the encoder of the ML model at the UE, the pre-decoder information is encoded into a second bit stream using the encoder of the ML model at the UE; as well as The second bit stream is transmitted to the base station.

4. The method according to claim 3, further comprising: The second actual effective DL channel experienced at the UE is calculated using the third CSI-RS received from the base station in response to the second bit stream; Calculate the second error between the expected effective DL channel and the second actual effective DL channel; and A second adjustment to the encoder of the ML model at the UE is performed based on the second error.

5. The method according to claim 3, further comprising: The second actual effective DL channel experienced at the UE is calculated using the third CSI-RS received from the base station in response to the second bit stream; The second error between the expected effective DL channel and the second actual effective DL channel is determined to be within a threshold. as well as Based on the determination that the second error is within the threshold, the encoder transmitting information about the ML model at the UE to the base station is an accurate indication.

6. The method according to claim 1, further comprising: The initial training of the ML model at the UE is performed using a third CSI-RS received from the base station prior to the first CSI-RS.

7. The method according to claim 6, further comprising: In response to receiving the third CSI-RS, the uplink (UL) probe reference signal (SRS) is transmitted to the base station.

8. The method according to claim 1, further comprising: At the encoder of the ML model at the UE, one or more of the Channel Quality Index (CQI) and Rank Indicator (RI) are determined based on the pre-decoder information; and One or more of the CQI and the RI are included in the first bit stream before being transmitted to the base station.

9. The method according to claim 1, further comprising: The first error information corresponding to the first error is transmitted to the base station.

10. A method for a user equipment (UE), the method comprising: The raw downlink (DL) channel estimate between the base station and the UE is calculated based on the first channel state information reference signal (CSI-RS) received from the base station; Predecoder information is determined based on the original DL channel estimation. The predecoder information represents a predecoder for use by the base station to enable the UE to experience the expected valid DL channel corresponding to the DL transmission from the base station. The encoder of the machine learning (ML) model at the UE encodes the pre-decoder information into a first bit stream; The first bit stream is transmitted to the base station; The first actual effective DL channel experienced at the UE is calculated based on the second CSI-RS received from the base station in response to the first bit stream; Calculate the first error between the expected effective DL channel and the first actual effective DL channel; and The first error information corresponding to the first error is transmitted to the base station.

11. The method of claim 10, wherein the predecoder information includes a set of feature vectors corresponding to the predecoder.

12. The method according to claim 10, further comprising: The first error information is generated by encoding the first error using a second ML model at the UE.

13. The method according to claim 10, further comprising: After transmitting the first error information, the third CSI-RS received from the base station is used to calculate the second actual effective DL channel experienced at the UE; Calculate the second error between the expected effective DL channel and the second actual effective DL channel; and The second error information corresponding to the second error is transmitted to the base station.

14. The method according to claim 10, further comprising: After transmitting the first error information, the third CSI-RS received from the base station is used to calculate the second actual effective DL channel experienced at the UE; The second error between the expected effective DL channel and the second actual effective DL channel is determined to be within a threshold. as well as Based on the determination that the second error is within the threshold, transmitting the decoder of the second ML model at the base station to the base station is an accurate indication.

15. The method according to claim 10, further comprising: The initial training of the ML model at the UE is performed using a third CSI-RS received from the base station prior to the first CSI-RS.

16. The method according to claim 15, further comprising: In response to receiving the third CSI-RS, the uplink (UL) probe reference signal (SRS) is transmitted to the base station.

17. The method of claim 10, further comprising: At the encoder of the ML model at the UE, one or more of the Channel Quality Index (CQI) and Rank Indicator (RI) are determined based on the pre-decoder information; and One or more of the CQI and the RI are included in the first bit stream before being transmitted to the base station.

18. A method for using a base station, the method comprising: Transmit the first channel state information reference signal (CSI-RS) to the user equipment (UE); In response to the first CSI-RS, a first bit stream is received from the UE; The decoder of the first machine learning (ML) model at the base station decodes the first pre-decoder information from the first bitstream; as well as The second CSI-RS, which is beamformed according to the first predecoder represented by the first predecoder information, is transmitted to the UE.

19. The method of claim 18, wherein the first predecoder information includes a set of feature vectors corresponding to the first predecoder.

20. The method according to claim 18, further comprising: In response to the second CSI-RS, a second bit stream is received from the UE; The decoder of the ML model at the base station decodes the second pre-decoder information from the second bitstream; as well as The third CSI-RS, which is beamformed according to the second predecoder represented by the second predecoder information, is transmitted to the UE.

21. The method according to claim 18, further comprising: The encoder received from the UE regarding the second ML model at the UE is an accurate indication.

22. The method according to claim 18, further comprising: Initial training of the ML model at the base station is performed using the uplink (UL) probe reference signal (SRS) received from the UE before transmitting the first CSI-RS.

23. The method of claim 18, wherein the bit stream further comprises one or more of a channel quality index (CQI) and a rank indicator (RI).

24. The method of claim 18, further comprising: The UE receives first error information, which corresponds to a first error between the expected effective DL channel and the first actual effective DL channel experienced by the UE and corresponding to the second CSI-RS. as well as The decoder of the first ML model at the base station is adjusted based on the first error information.

25. The method of claim 24, wherein the first error information includes encoding of the first error.

26. The method of claim 24, further comprising: After adjusting the decoder of the first ML model at the base station, the first bit stream is decoded into second pre-decoder information using the decoder of the ML model at the base station; as well as The third CSI-RS, which is beamformed according to the second predecoder represented by the second predecoder information, is transmitted to the UE.

27. The method according to claim 26, further comprising: The decoder received from the UE after transmitting the third CSI-RS is an accurate indication of the ML model at the base station.

28. An apparatus comprising components for performing the method according to any one of claims 1 to 27.

29. A computer-readable medium comprising instructions that, when executed by one or more processors of an electronic device, cause the electronic device to perform the method according to any one of claims 1 to 27.

30. An apparatus comprising a logic component, module, or circuitry for performing the method according to any one of claims 1 to 27.

Images