Systems and methods for handling rank adaptation for channel state information compression using two sided models

Abstract

Systems and methods for handling rank adaptation for channel state information (CSI) compression using two-sided artificial intelligence (AI) / machine learning (ML) models are discussed herein. For example, a user equipment (UE) performing AI / ML model-based CSI reporting may identify that a rank adaptation prerequisite for differing from a previous rank indication (RI) indicated to a base station by the UE is met, and may generating, using a measurement of a channel state information reference signal (CSI-RS) transmitted by the base station, one or more precoder matrix indicators (PMIs) and a corresponding updated RI that is different than the previous RI. Additionally, the UE may generate encoded PMI information by applying the one or more PMIs at one or more AI / ML models, and transmit, to the base station, a CSI report comprising the encoded PMI information and the updated RI.

Description

SYSTEMS AND METHODS FOR HANDLING RANK ADAPTATION FORCHANNEL STATE INFORMATION COMPRESSION USING TWO SIDED MODELSTECHNICAL FIELD

[0001] This application relates generally to wireless communication systems, including wireless communication systems that implement artificial intelligence (AI) / machine learning (ML) models.BACKGROUND

[0002] Wireless mobile communication technology uses various standards and protocols to transmit data between a base station and a wireless communication device. Wireless communication system standards and protocols can include, for example, 3rd Generation Partnership Project (3GPP) Long Term Evolution (LTE) (e.g., 4G), 3GPP New Radio (NR) (e.g., 5G), and Institute of Electrical and Electronics Engineers (IEEE) 802.11 standard for Wireless Local Area Networks (WLAN) (commonly known to industry groups as Wi-Fi®).

[0003] As contemplated by the 3 GPP, different wireless communication systems standards and protocols can use various radio access networks (RANs) for communicating between a base station of the RAN (which may also sometimes be referred to generally as a RAN node, a network node, or simply a node) and a wireless communication device known as a user equipment (UE). 3GPP RANs can include, for example, Global System for Mobile communications (GSM), Enhanced Data Rates for 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 may use one or more radio access technologies (RATs) to perform communication between the base station and the UE. For example, the GERAN implements GSM and / or EDGE RAT, the UTRAN implements Universal Mobile Telecommunication System (UMTS) RAT or other 3GPP RAT, the E-UTRAN implements LTE RAT (sometimes simply referred to as LTE), and NG-RAN implements NR RAT (sometimes referred to herein as 5G RAT, 5G NR RAT, or simply NR). In certain deployments, the E-UTRAN may also implement NR RAT. In certain deployments, NG-RAN may also implement LTE RAT.

[0005] A base station used by a RAN may correspond to that RAN. One example of an E-UTRAN base station is an Evolved Universal Terrestrial Radio Access Network (E- UTRAN) Node B (also commonly denoted as evolved Node B, enhanced Node B, eNodeB, or eNB). One example of an NG-RAN base station is a next generation Node B (also sometimes referred to as a g Node B or gNB).

[0006] A RAN provides its communication services with external entities through its connection to a core network (CN). For example, E-UTRAN may utilize an Evolved Packet Core (EPC) while NG-RAN may utilize a 5G Core Network (5GC).BRIEF DESCRIPTION OF THE SEVERAL VIEWS OF THE DRAWINGS

[0007] To easily identify the discussion of any particular element or act, the most significant digit or digits in a reference number refer to the figure number in which that element is first introduced.

[0008] FIG. 1 illustrates a diagram of a high-level structure for the use of an AI / ML model for time-spatial-frequency domain channel state information (CSI) compression, according to embodiments herein.

[0009] FIG. 2 illustrates a diagram of a high-level structure for the use of an AI / ML model for time-spatial-frequency domain CSI compression, according to embodiments herein.

[0010] FIG. 3 illustrates a flow diagram of a procedure for time-spatial-frequency domain AI / ML-based CSI feedback, including the use by a UE of a network-based rank restriction for rank adaptation, according to embodiments herein.

[0011] FIG. 4 illustrates a flow diagram of a procedure for time-spatial-frequency domain AI / ML-based CSI feedback, including the use by a UE of a network-based rank restriction for rank adaptation, according to embodiments herein.

[0012] FIG. 5 A illustrates an example of UE based rank adaptation where the UE uses an additional rank adaptation restriction, according to embodiments herein.

[0013] FIG. 5B illustrates an example of UE based rank adaptation where the UE uses an additional rank adaptation restriction, according to embodiments herein.

[0014] FIG. 5C illustrates an example of UE based rank adaptation where the UE reports generated encoded precoder matrix indicator (PMI) information based on a maximum rank restriction received from a base station, according to embodiments herein.

[0015] FIG. 5D illustrates an example of UE based rank adaptation where the UE may select a RI for each report, according to embodiments herein.

[0016] FIG. 6 illustrates a flow diagram between a UE and a base station for rank restriction based on a UE recommendation for rank restriction sent to the base station, according to embodiments herein.

[0017] FIG. 7 illustrates a method of a UE for AI / ML model-based CSI reporting, according to embodiments herein.

[0018] FIG. 8 illustrates a method of a UE for AI / ML model-based CSI reporting, according to embodiments herein.

[0019] FIG. 9 illustrates a method of a UE for AI / ML model-based CSI reporting, according to embodiments herein.

[0020] FIG. 10 illustrates a method of a base station for AI / ML model-based CSI reporting, according to embodiments herein.

[0021] FIG. 11 illustrates an example architecture of a wireless communication system, according to embodiments disclosed herein.

[0022] FIG. 12 illustrates a system for performing signaling between a wireless device and a network device, according to embodiments disclosed herein.DETAILED DESCRIPTION

[0023] Various embodiments are described with regard to a UE. However, reference to a UE is merely provided for illustrative purposes. The example embodiments may be utilized with any electronic component that may establish a connection to a network and is configured with the hardware, software, and / or firmware to exchange information and data with the network. Therefore, the UE as described herein is used to represent any appropriate electronic component.

[0024] In some wireless communication systems, for channel state information (CSI) feedback enhancement, use of a two-sided artificial intelligence (AI) / machine learning (ML) model for CSI compression may be considered. Various mechanisms for improving the trade-off between performance and complexity / overhead of the use of AI / ML models for CSI compression that may be considered include, for example: considering extending spatial / frequency compression to spatial / temporal / frequency compression, considering the use of cell / site specific models, and / or considering the use of CSI compression plusprediction using the AI / ML models. Additionally, rank adaptation for spatial / temporal / frequency compression may be considered.

[0025] Further, mechanisms for alleviating / resolving issues related to inter-vendor AI / ML model training collaboration may be considered.

[0026] In some wireless communication mechanisms, for CSI prediction, it may be considered that using a one-sided AI / ML model may increase performance as compared to current non-AI / ML-based approaches (including the associated complexity), while addressing other aspects that may be under further consideration.

[0027] Embodiments herein relate to the performance of AI / ML-based channel state information (CSI) reporting that uses an AI / ML model that is made up of both an encoder for encoding one or more elements of CSI feedback at a UE side prior to transmission and a decoder for decoding the one or more elements out from the encoded CSI feedback as received the base station side. Such AI / ML procedures may be referred to in places herein as “AI / ML-based CSI compression” or the like. For the evaluation of temporal domain aspects of the use of such AI / ML models for CSI compression, the following categorization may be considered.Table 1. Cases for AI / ML-based CSI compression

[0028] In some instances, at the UE, the past CSI information may include prior AI / ML model inputs and / or any information derived from them. Similarly, for the network, the past CSI information may include past CSI feedback instances and / or any information derived from them. Additionally, for case 3 and case 4 (shown in Table 1), the UE may perform AI / ML prediction as a separate step or jointly with CSI compression. Similarly, the network may perform prediction as a separate step or jointly with reconstruction. It may be reported which option is selected, the number of future slots, and whether the prediction is AI / ML-based or not. Note that the “target CSI slot(s)”, in Table 1, may refer to the slot(s) to which the CSI feedback in the report corresponds to. The “present slot” of Table 1 may refer to the slot of the most recent channel state information reference signal (CSI-RS) measurement used to generate the CSI report. The “future slot(s)” of Table 1 may include at least one slot after the present slot and may include the present slot as well.

[0029] Embodiments discussed herein relate to at least cases 2 and 4 as defined in Table 1. In case 2, the target CSI slot (the slot during which information of the CSI report generated through the AI / ML model for CSI compression applies) is a present slot (the slot used by the CSI report). In case 4, the use of “predicted CSI” is contemplated, wherein the target CSI slot(s) (the slot(s) during which information of the CSI report generated through the AI / ML model for CSI compression applies) include one or more future slots (slots later than the slot used by the CSI report). Note that at least some uses of predicted CSI may also include the slot used by the CSI report as one of the target slots (in addition to one or more future slots).

[0030] In cases 2 and 4, each of the UE (e.g., the encoder of the AI / ML model at the UE) and the network (e.g., the decoder of the AI / ML model at the network) rely on past CSI information. In case 5, the network (e.g., the decoder of the AI / ML model at the network) relies on past CSI information.

[0031] Note that, in some wireless communication mechanisms, for various cases of multiple input multiple output (MIMO) feedback, including for cases for spatial- frequency domain AI / ML-based CSI compression, the network may configure a rank restriction based on a codebook configuration, and the UE may correspondingly choose to use the implemented rank when generating feedback. In some cases, a rank indicator (RI) may be indicated in CSI part 1 (where CSI part 1 has a fixed size), and the CSI payload in CSI part 2 may correspond to the indicated RI (note that such CSI payloadsare therefore understood to be variable in size based on different RIs being indicated). Under such circumstances, there is no use of memory to retain this information after the CSI is sent, and thus each RI may be understood to be independently selected for each CSI report. As a result, the RI for one state of the encoder or decoder may not apply to a second state, a third state, and so on of the encoder or decoder.

[0032] For case 2 and case 4 of extended time-frequency-spatial domain AI / ML-based CSI compression (discussed herein), as a result of a past CSI being used, and memory being updated based on previous CSI, a CSI report may not be fully independent anymore as between a first state to a second state, the second state to a third state, and so on. Note that case 3 (as discussed herein) may also use past CSI; however, in this case, the past CSI may be used as an input to a separate CSI prediction AI / ML model. The output of the CSI prediction is quantized together and there accordingly may be no memory block. As a result, the RI may be updated per CSI report for the predicted CSI altogether (the report does not ultimately take into account previous RI values).

[0033] Embodiments herein relate to the use of time-spatial-frequency domain AI / ML models (which consider, among other things, time domain aspects).

[0034] FIG. 1 illustrates a diagram 100 of a high-level structure for the use of an AI / ML model for time-spatial-frequency domain CSI compression, according to embodiments herein. FIG. 1 illustrates that such an AI / ML model for CSI compression is made up of both an encoder 104 for encoding one or more elements of CSI feedback at the UE side prior to transmission and a decoder 106 for decoding the one or more elements of CSI feedback as encoded and as received the base station side.

[0035] Additionally, to account for time domain aspects, the use of an internal state at each of the encoder 104 and the decoder 106 corresponding to a given time may be introduced. For example, the encoder 104 (at a UE), at a first time 102 (denoted tl), uses a first precoder matrix indicator (PMI) or precoder 108 Vl t il (e.g., of layer z of the CSI feedback) as an input to generate the first encoded CSI feedback 110, which is then transmitted by the UE, as illustrated. At this first time 102, the encoder 104 is in a first encoder state 114 (denoted SLnc).

[0036] Further, as shown, corresponding to the first time 102, a decoder 106 (at a base station) receives the first encoded CSI feedback 110 and decodes it into a first reconstructed precoder 112 (denoted V'l tJ 1), which may then be used by the basestation for purposes of precoder selection. At this first time 102, the decoder 106 is in a first decoder state 116 (denoted SI dec).

[0037] Then, as illustrated, at a second time 118 (denoted t2), the encoder 104 uses a second PMI or precoder 120 (denoted Vl_t_i2) (e.g., of layer z) and further uses the first encoder state 114 (information about the encoder 104 as it was in the first encoder state 114 at the first time 102) to generate the second encoded CSI feedback 122, which is then transmitted by the UE, as illustrated. The use of the first encoder state 114 at this stage (and its corresponding effect on the generation of second encoded CSI feedback 122 at the encoder 104) represents an inclusion of time domain aspects at the encoder 104 of the time-spatial-frequency domain AI / ML model. At this second time 118, the encoder 104 is in a second encoder state 126 (denoted S2enc).

[0038] Further, as shown, corresponding to the second time 118, the decoder 106 receives the second encoded CSI feedback 122 and decodes it into a second reconstructed precoder 124 (denoted V'l_tJ2), which may then be used by the base station for precoder selection. At this second time 118, the decoder 106 is in a second decoder state 128 (denoted S2dec). During this decoding process, the decoder 106 takes into account the first decoder state 116 (information about the decoder 106 as it was in the first decoder state 116 at the first time 102). The use of the first decoder state 116 at this stage (and its corresponding effect on the generation of second reconstructed precoder 124 at the decoder 106) represents an inclusion of time domain aspects at the decoder 106 of the time-spatial -frequency domain AI / ML model.

[0039] As illustrated, at a third time 130 (denoted t3), the encoder 104 uses a third PMI or precoder 132 (denoted Vl_t_i3) (e.g., of layer z) and further uses the second encoder state 126 (information about the encoder 104 as it was in the second encoder state 126 at the second time 118) to generate the third encoded CSI feedback 134, which is then transmitted by the UE, as illustrated. The use of the second encoder state 126 at this stage (and its corresponding effect on the generation of third encoded CSI feedback 134 at the encoder 104) represents an inclusion of time domain aspects at the encoder 104 of the time-spatial-frequency domain AI / ML model. At this third time 130, the encoder 104 is in a third encoder state 138 (denoted S3enc).

[0040] As shown, corresponding to the third time 130, the decoder 106 receives the third encoded CSI feedback 134 and decodes it into a third reconstructed precoder 136 (denoted V'l_tJ3), which may then be used by the base station for precoder selection.At this third time 130, the decoder 106 is in a third decoder state 140 (denoted S3dec). During this decoding process, the decoder 106 takes into account the second decoder state 128 (information about the decoder 106 as it was in the second decoder state 128 at the second time 118). The use of the second decoder state 128 at this stage (and its corresponding effect on the generation of third reconstructed precoder 136 at the decoder 106) represents an inclusion of time domain aspects at the decoder 106 of the time- spatial-frequency domain AI / ML model.

[0041] Note that while FIG. 1 explicitly illustrates the use of a PMI or precoder of a single layer z, this is given by way of example only. The encoder 104 and the decoder 106 will be understood to be extendible to analogously encode / decode CSI feedback corresponding to PMI / PMIs or precoders of multiple layers (up to and including all layers of a PMI in some embodiments). Additionally, note that, for case 2 t_i = tj understood as a current CSI and for case 4 t_i < tj, understood as a future CSI.

[0042] Such processing can be applied to multiple CSI instances with the same spatial layer in some embodiments, and for each spatial layer, a separate copy of the AI / ML encoder model may be run to update its internal state and generate output.

[0043] In some other embodiments, such processing can be applied to multiple CSI instances with multiple spatial layers, and those multiple spatial layers may comprise all the spatial layers in a CSI feedback or a group of spatial layers in a CSI feedback (e.g., predicted CSI instances for the first spatial layer and predicted CSI instances for the second spatial layer go through the same AI / ML encoder with a schedule to feed a copy of AI / ML encoder model with {first predicted CSI instance for the first spatial layer, first predicted CSI instance for the second spatial layer, second predicted CSI instance for the first spatial layer, second predicted CSI instance for the second spatial layer, ..., the last predicted CSI instance for the first spatial layer, the last predicted CSI instance for the second spatial layer}). In one example, spatial layer 1 and spatial layer 2 can be grouped together, and spatial layer 3 and spatial layer 4 can be grouped together. If the feedbacks generated from the use of the encoder 104 are aggregated and transmitted in a single report instance to the network by the UE, then other schedules to feedback a copy of an AI / ML encoder model can be utilized such as {first predicted CSI instance for the first spatial layer, ..., the last predicted CSI instance for the first spatial layer, ..., first predicted CSI instance for the last spatial layer in a group, ..., the last predicted CSI instance for the last spatial layer in a group}. In some embodiments, inputs to theencoder 104 are for the precoders of more than one spatial layer. In some embodiments, inputs to the encoder 104 are for the PMIs of a single spatial layer, and each PMI is generated from a conventional CSI feedback scheme or another AI / ML model which essentially generates a summary, hash or latency space representation of precoder, and the options in schedules for feeding encoder 104 as disclosed above can apply. In some embodiments, inputs to the encoder 104 are for the PMIs of a group of spatial layers or of all spatial layers, and each PMI is generated from a conventional CSI feedback scheme or another AI / ML model which essentially generates a summary, hash or latency space representation of precoder, and the options in schedules for feeding encoder 104 as disclosed above can apply.

[0044] In various embodiments, the encoder state information (e.g., the first encoder state 114, the second encoder state 126, and / or the third encoder state 138) may include or represent or correspond to one or more prior PMIs or precoders (e.g., prior to the corresponding time for that state) acquired by and encoded by (at least in part) the UE.

[0045] In various embodiments, the decoder state information (e.g., the first decoder state 116, the second decoder state 128, and / or the third decoder state 140) may include or represent or correspond to one or more prior PMIs or precoders (e.g., prior to the corresponding time for that state) that were previously decoded (at least in part) at the base station.

[0046] In some embodiments, each of the encoder 104 and / or the decoder 106 may be implemented using a recursive neural network (RNN) (e.g., a long short term memory (LSTM) RNN). The information about the one or more prior PMIs or precoders (e.g., from the perspective of either the UE or the base station) may thus correspondingly be represented in the encoder 104 / the decoder 106 as weighting values within the corresponding RNN (and which may be modified as the encoder 104 / the decoder 106 is continuously operated with new inputs through time).

[0047] As each of the encoder 104 and the decoder 106 of the AI / ML model of FIG. 1 operate according to time domain aspects, it may be understood that the AI / ML model of FIG. 1 corresponds to cases 2 and 4 as defined by Table 1 (discussed elsewhere herein).

[0048] FIG. 2 illustrates a diagram 200 of a high-level structure for the use of an AI / ML model for time-spatial-frequency domain CSI compression, according to embodiments herein. FIG. 2 illustrates that an AI / ML model for CSI compression is made up of both an encoder 204 for encoding one or more elements of CSI feedback atthe UE side prior to transmission and a decoder 206 for decoding the one or more elements of CSI feedback as encoded and as received by the base station side.

[0049] The encoder 204 (at a UE), at a first time 202 (denoted tl), uses a first PMI or precoder 208 Vl t il (e.g., of layer z of the CSI feedback) as an input to generate the first encoded CSI feedback 210, which is then transmitted by the UE, as illustrated.

[0050] Additionally, to account for time domain aspects, the use of an internal state at the decoder 206 corresponding to a given time may be introduced. For example, as shown, corresponding to the first time 202, a decoder 206 (at a base station) receives the first encoded CSI feedback 210 and decodes it into a first reconstructed precoder 212 (denoted V'l tJ 1), which may then be used by the base station for purposes of precoder selection. At this first time 202, the decoder 206 is in a first decoder state 214 (denoted SI dec).

[0051] Then, as illustrated, at a second time 216 (denoted t2), the encoder 204 uses a second PMI or precoder 218 (denoted Vl_t_i2) (e.g., of layer z) to generate the second encoded CSI feedback 220, which is then transmitted by the UE, as illustrated.

[0052] Further, as shown, corresponding to the second time 216, the decoder 206 receives the second encoded CSI feedback 220 and decodes it into a second reconstructed precoder 222 (denoted V'l_tJ2), which may then be used by the base station for precoder selection. At this second time 216, the decoder 206 is in a second decoder state 224 (denoted S2dec). During this decoding process, the decoder 206 takes into account the first decoder state 214 (information about the decoder 206 as it was in the first decoder state 214 at the first time 202). The use of the first decoder state 214 at this stage (and its corresponding effect on the generation of second reconstructed precoder 222 at the decoder 206) represents an inclusion of time domain aspects at the decoder 206 of the time-spatial -frequency domain AI / ML model.

[0053] As illustrated, at a third time 226 (denoted t3), the encoder 204 uses a third PMI or precoder 228 (denoted Vl_t_i3) (e.g., layer z) to generate the third encoded CSI feedback 230, which is then transmitted by the UE, as illustrated.

[0054] As shown, corresponding to the third time 226, the decoder 206 receives the third encoded CSI feedback 230 and decodes it into a third reconstructed precoder 232 (denoted V'l_tJ3), which may then be used by the base station for precoder selection. At this third time 226, the decoder 206 is in a third decoder state 234 (denoted S3dec). During this decoding process, the decoder 206 takes into account the second decoderstate 224 (information about the decoder 206 as it was in the second decoder state 224 at the second time 216). The use of the second decoder state 224 at this stage (and its corresponding effect on the generation of third reconstructed precoder 232 at the decoder 206) represents an inclusion of time domain aspects at the decoder 206 of the time- spatial-frequency domain AI / ML model.

[0055] Note that while FIG. 2 explicitly illustrates the use of a PMI or a precoder of a single layer z, this is given by way of example only. The encoder 204 and the decoder 206 will be understood to be extendible to analogously encode / decode CSI feedback corresponding to multiple layers of a PMI (up to and including all layers of a PMI in some embodiments). Additionally, note that, for case 2 t_i = tj understood as a current CSI and for case 4 t_i < tj, understood as a future CSI.

[0056] Such processing can be applied to multiple CSI instances with the same spatial layer in some embodiments, and for each spatial layer, a separate copy of the AI / ML encoder model may be run to update its internal state and generate output.

[0057] In some other embodiments, such processing can be applied to multiple CSI instances with multiple spatial layers, and those multiple spatial layers may comprise all the spatial layers in a CSI feedback or a group of spatial layers in a CSI feedback (e.g., predicted CSI instances for the first spatial layer and predicted CSI instances for the second spatial layer go through the same AI / ML encoder with a schedule to feed a copy of AI / ML encoder model with {first predicted CSI instance for the first spatial layer, first predicted CSI instance for the second spatial layer, second predicted CSI instance for the first spatial layer, second predicted CSI instance for the second spatial layer, ..., the last predicted CSI instance for the first spatial layer, the last predicted CSI instance for the second spatial layer}). In one example, spatial layer 1 and spatial layer 2 can be grouped together, and spatial layer 3 and spatial layer 4 can be grouped together. If the feedbacks generated from the use of the encoder 204 are aggregated and transmitted in a single report instance to the network by the UE, then other schedules to feedback a copy of an AI / ML encoder model can be utilized such as {first predicted CSI instance for the first spatial layer, ..., the last predicted CSI instance for the first spatial layer, ..., first predicted CSI instance for the last spatial layer in a group, ..., the last predicted CSI instance for the last spatial layer in a group}. In some embodiments, inputs to the encoder 204 are for the precoders of more than one spatial layer. In some embodiments, inputs to the encoder 204 are for the PMIs of a single spatial layer, and each PMI isgenerated from a conventional CSI feedback scheme or another AI / ML model which essentially generates a summary, hash or latency space representation of precoder, and the options in schedules for feeding encoder 204 as disclosed above can apply. In some embodiments, inputs to the encoder 204 are for the PMIs of a group of spatial layers or of all spatial layers, and each PMI is generated from a conventional CSI feedback scheme or another AI / ML model which essentially generates a summary, hash or latency space representation of precoder, and the options in schedules for feeding encoder 204 as disclosed above can apply.

[0058] In various embodiments, the decoder state information (e.g., the first decoder state 214, the second decoder state 224, and / or the third decoder state 234) may include or represent or correspond to one or more prior PMIs or precoders (e.g., prior to the corresponding time for that state) that were previously decoded (at least in part) at the base station.

[0059] In some embodiments, the decoder 206 may be implemented using a recursive neural network (RNN) (e.g., a long short term memory (LSTM) RNN). The information about the one or more prior PMIs (e.g., from the perspective of either the UE or the base station) may thus correspondingly be represented in the decoder 206 as weighting values within the corresponding RNN (and which may be modified as the decoder 206 is continuously operated with new inputs through time).

[0060] As the decoder 206 of the AI / ML model of FIG. 2 operates according to time domain aspects, it may be understood that the AI / ML model of FIG. 2 corresponds to case 5 as defined by Table 1 (discussed elsewhere herein).

[0061] As illustrated in the FIG. 1 and FIG. 2, a UE may need to send CSI feedback repeatedly over time to allow the network to derive a precoder predication / estimate. However, due to the use in each case of time domain information as described, the behavior of the decoder (e.g., the decoder 106 and / or the decoder 206) may become inaccurate if there is a lost CSI feedback.

[0062] For example, FIG. 1 illustrates an example where the first encoded CSI feedback 110 is lost 142, which would cause an initial issue with the first decoder state 116 of the decoder 106 at the first time 102. This issue would remain in play over time as the first decoder state 116 propagates / is then used by the decoder 106 at the second time 118 corresponding to a second decoder state 128 (and so on as the second decoder state 128 propagates / is then used by the decoder 106 at the third time 130 corresponding to a thirddecoder state 140, etc.). In other words, even when the network correctly receives CSI feedback after the first encoded CSI feedback 110 that is lost 142, the network may not be able to recover an accurate CSI due to the effects of the lost CSI feedback that are now propagating through the decoder over time.

[0063] As another example, FIG. 2 illustrates an example where the second encoded CSI feedback 220 is lost 236, which would cause an initial issue with the second decoder state 224 of the decoder 106 at the second time 216. This issue would remain in play over time as the second decoder state 224 propagates / is then used by the decoder 206 at the third time 226 corresponding to a third decoder state 234 (and so on as the third decoder state 234 propagates / is then used by the decoder 106 at the fourth time subsequent to the third time 226 corresponding to a fourth decoder state, etc.). In other words, even when the network correctly receives CSI feedback after the second encoded CSI feedback 220 that is lost 236, the network may not be able to recover an accurate CSI due to the effects of the lost CSI feedback that are now propagating through the decoder over time.

[0064] In response to these losses in state, it may be that the state information used by the corresponding AI / ML model is reset. Further, as noted above, it may be in various cases that the reason that CSI feedback is lost in the above embodiments is due to a change in channel conditions. It may be therefore beneficial to allow for rank adaptation in conjunction with / near these instances of lost CSI feedback (e.g., to use a lower rank than before, as may be more appropriate for a relatively deteriorated channel). Accordingly, embodiments here relate to mechanisms for relatively quickly allowing rank adaptation.

[0065] Embodiments herein discuss different ways to handle rank adaptation in, for example, cases 2 and 4 where memory is used (as discussed herein). Note that, in existing wireless communication mechanisms, there are no procedures to handle rank adaptation in such cases where memory is used for CSI compression and / or CSI decoding purposes. Embodiments herein introduce procedures for the use of rank adaptation in such circumstances. For example, embodiments herein may relate to rank adaptation for different AI / ML model(s) / state combination(s) including, for example, cases where a layer-common AI / ML model and layer-specific state(s) are used, cases where layer-specific AI / ML model(s) and layer-specific state(s) are used, and / or cases where a layer-common AI / ML model and layer-common state(s) are used.

[0066] Additionally, embodiments herein discuss handling rank adaptation through the use of rank restriction configuration / reporting and state management. For example, embodiments herein introduce a network-configuration-based rank adaptation mechanism, a UE-deci sion-based rank adaptation mechanism, and a combined UE- decision-based and network-configuration-based rank adaptation mechanism.

[0067] With respect to rank adaptation as discussed herein, the various different AI / ML models and state handling cases for AI / ML-based CSI compression may be differentiated in various ways.

[0068] For example, for cases of the use of a layer-common AI / ML model with layerspecific state information, the same AI / ML model may be used for the layer inputs (e.g., the same structure and parameters are used for the layer inputs Vl t il, V2_t_il, etc. for each of Layer 1 and Layer 2, etc. respectively). For each layer, different state information may be maintained per layer (e.g., at the UE and at the network separately). Note that for the case of a layer-common AI / ML model, the AI / ML model may be ran multiple times depending on / according to the number of layers.

[0069] In another example, for cases of the use of a layer-common AI / ML model with layer-common state information, the same AI / ML model may be used for the layer inputs (e.g., the same structure and parameters are used for the layer input Vl t il, V2_t_il, etc. for each of Layer 1 and Layer 2, etc. respectively). The same state information may be maintained across different layers.

[0070] In yet another example, for cases of the use of layer-specific AI / ML model(s) with layer-specific state information, different AI / ML model(s) may be used for the layer inputs (e.g., different parameters are used for the layer input Vl t il, V2_t_il, etc. for each of Layer 1 and Layer 2, etc. respectively). For each layer, different state information may be maintained per layer (e.g., at UE and at the network separately). Note that such layer-specific AI / ML model(s) with layer-specific state information may achieve a high degree of performance and complexity. Additionally, note that the input for the layer-specific AI / ML model(s) is inherently a per-layer consideration; thus, the layer-specific AI / ML model may be considered relatively better tuned to the layer level of input information.

[0071] FIG. 3 illustrates a flow diagram 300 of a procedure for time-spatial -frequency domain AI / ML-based CSI feedback, including the use by a UE of a network-based rank restriction for rank adaptation, according to embodiments herein.

[0072] Embodiments herein may, in some cases, use a semi-persistent CSI report framework in relation to various rank restriction handling contexts.

[0073] In some embodiments discussed herein, a procedure (such as the procedure represented by the flow diagram 300) may be designed to optimize time-spatial- frequency domain AI / ML-based CSI feedback.

[0074] The flow diagram 300 begins with a UE 302 reporting its supported AI / ML model(s) using corresponding supported AI / ML model ID(s) 306 to the base station 304. In some examples, the supported AI / ML model ID(s) 306 may be reported in a UE capability report. In other examples, the supported AI / ML model ID(s) 306 may be reported in an RRCReconfigurationComplete message (e.g., when the UE receives a handover command to handover to a new cell).

[0075] Subsequently, the base station 304 may transmit, to the UE 302, an RRC configuration 308 for a CSI report. In some cases, a supported reportConfigType information element (IE) of the RRC configuration 308 may include a semiP er sistent IE transmission on a physical uplink control channel (PUCCH) and / or a semiPersistent IE transmission on a physical uplink shared channel (PUSCH).

[0076] Further, the RRC configuration 308 may include a csi ReportConfig IE and a csi ResourceConfig IE. The csi-ReportConfig IE may configure that the feedback type is AI / ML based CSI feedback. Further, the csi ReportConfig IE may specify that the particular csi ResourceConfig IE (e.g., from multiple such csi ReportConfig IES) is to be used for measurement. Then, the csi ResourceConfig IE indicates what type of reference signal the measurement is to be transmitted on and what type of transmissions are to be performed (e.g., as between periodic, aperiodic, semi-persistent).

[0077] Subsequently, the base station 304 may transmit a message including a triggering indication 310 with a rank restriction to the UE 302 for triggering of the AI / ML model. Note that the triggering indication 310 also includes a rank restriction for use by the triggered AI / ML model.

[0078] In some cases, the rank restriction may optionally be included in the RRC configuration 308 (the rank restriction is RRC configured). For example, an ri- Restriction-AI-case 2 parameter or an ri-Restriction-AI-case 4 parameter (or other parameters of the sort), included in the RRC configuration 308, may limit the RI value across different CSI reporting instances to be the same within a set of CSI reports. Insome embodiments, the rank restriction may be included in a csi ReportConfig IE of the RRC configuration 308.

[0079] In some embodiments, the UE may be configured with a higher layer bitmap parameter RI-Restriction-AI-case 2, which forms the bit sequence {r3, r2, rl, r0,}, where rO is the least significant bit (LSB) and r3 is the most significant bit (MSB). When ri is zero, i E {0,1,...,3}, PMI and RI reporting may not be allowed to correspond to any precoder associated with v = i + 1 layers.

[0080] In some other cases, the rank restriction included in the triggering indication 310 may be updated using a medium access control (MAC) control element (CE) (MAC CE), or alternatively, the triggering indication 310 including the rank restriction may be transmitted using a MAC CE. For example, a rank restriction pattern may be activated based on the MAC CE. Additionally, a same RI may be used for next set of CSI reports, until another MAC CE update is received with a different / updated rank restriction for rank adaptation.

[0081] In yet some other cases, downlink control information (DCI) may be used to transmit the triggering indication 310 with the rank restriction and / or update the rank restriction included in the triggering indication 310.

[0082] It should be understood that the use of DCI or MAC CE to update the rank restriction may more frequently update the rank restriction as compared to current mechanisms employed in wireless communication systems.

[0083] Then, the base station 304 may perform a CSI-RS transmission 312 to the UE 302. The UE 302 acquires CSI information (including, e.g., one or more PMI(s)) for a channel between the UE and the base station 304 by measuring the CSI-RS. The UE 302 may then formulate a CSI report using that CSI-RS information. This CSI report may include one or more encoded PMI(s) that are generated by applying the one or more PMI(s) to one or more AI / ML model(s), as described herein.

[0084] The UE 302 then transmits a UCI report 314 to the base station 304 that includes the CSI report having the encoded PMI(s). Note the CSI-RS transmission 312 and responsive UCI report 314 may correspond to a first state (i.e., “State 0”).

[0085] As shown, over time, subsequent, CSI-RS transmissions 316 and responsive transmissions of UCI reports 318 may follow suit corresponding to subsequent states (such as State 1, State 2, ..., State N). This may occur until a resetting indication 320with rank restriction information is transmitted (according to various cases similar to those discussed for the triggering indication 310 with the rank restriction discussed herein).

[0086] In some embodiments, for AI / ML model(s) (either layer-specific AI / ML model(s) or layer-common AI / ML model(s)) using layer-specific state information, when a MAC CE and / or a DCI is used to update the rank restriction included in the triggering indication 310 to be higher than previous iterations, (e.g., a previous rank restriction restricting the RI to a value of 1, now restricts the RI to a value of 2), such that a higher rank is ultimately used on a current iteration, it may be that the Layer 1 state is treated as (remaining) valid, and there may be no need to reset the state information for Layer 1, unless explicitly configured by the MAC CE or the DCI to do so (e.g., due to a UCI loss). Further, the Layer 2 state information may start from an initial state (state 0), with such state information for Layer 2 subsequently being updated separately from that of Layer 1 going forward.

[0087] After a resetting indication 320 with the rank restriction information is received by the UE 302, any state information used by the encoder / decoder of the AI / ML model is dropped and the encoder and the decoder of the AI / ML model are thus returned to an initial state, including, in some cases, resetting the RI to an initial state so that it may correspond to the current rank restriction. Note that, in some cases, the rank restriction may remain the same as indicated in the triggering indication 310. The CSI process may then be continued under the (potentially new) rank restriction, (see, e.g., the CSLRS transmission 322 and the UCI report 324). This may occur in a manner that is affected by prior PMI information generated prior to the resetting indication 320 with rank restriction information for layers that remain in use after applying the (potentially) new rank restriction, as discussed herein.

[0088] FIG. 4 illustrates a flow diagram 400 of a procedure for time-spatial-frequency domain AI / ML-based CSI feedback, including the use by a UE of a network-based rank restriction for rank adaptation, according to embodiments herein.

[0089] In some existing mechanisms, the UE may select an RI value based on rank restriction that remains constant through the CSI reporting procedure. Embodiments discussed herein, rather, relate to cases where a rank can be updated to better optimize the use of CSI reports and / or corresponding CSLRS transmissions for a CSI reporting procedure (e.g., an AI / ML-based CSI reporting procedure).

[0090] In some embodiments discussed herein, a procedure (such as the procedure represented by the flow diagram 400) may be designed to optimize time-spatial- frequency domain AI / ML-based CSI feedback. The flow diagram 400 begins with a UE 402 reporting its supported AI / ML model(s) using corresponding supported AI / ML model ID(s) 406 to the base station 404. In some examples, the supported AI / ML model ID(s) 406 may be reported in a UE capability report. In other examples, the supported AI / ML model ID(s) 406 may be reported in RRCReconfigurationComplete message (e.g., when the UE receives a handover command to handover to a new cell).

[0091] Subsequently, the base station 404 may transmit, to the UE 402, an RRC configuration 408 for a CSI report (similarly to that discussed in FIG. 3).

[0092] Subsequently, the base station 404 may transmit a message including a triggering indication 410 including additional rank restriction information to the UE 402 for triggering of the AI / ML model with an included rank restriction for UE based rank adaptation.

[0093] Corresponding to such embodiments, various cases are considered in order to organize the frequency at which an RI may be updated by a UE. Some such cases may be used to prevent, for example, the too-frequent updating of RI such that the benefit to the system relative to the complexity of implementation / operation becomes out of balance. Other such cases may be motivated by the desire to organize the UE’s use of RI in a determinable way. The triggering indication 410 including additional rank restriction information may be used to configure the UE to use one or more of these options.

[0094] In first such cases, an additional network rank restriction may be applied by the UE 402. For example, a rank-restriction-N IE may be applied that may restrict the UE’s ability to change the RI so that the RI applies to at least a number of CSI reports (e.g., “N” reports), before the UE 402 may choose and report a different RI value. For example, such rank restriction may provide a rank adaptation prerequisite where the prerequisite includes a minimum RI use threshold. Note that the rank adaptation prerequisite is met when a number of prior CSI reports based on the previous RI meets the minimum RI use threshold.

[0095] In second such cases, an additional network rank restriction may be applied corresponding to an indicated rank adaptation periodicity. For example, the UE 402 may update the RI only according to a rank adaptation periodicity. Note that, in some examples, the periodicity is indicated by the base station 404. In some instances, suchrank restriction may provide a rank adaptation prerequisite, where the CSI report occurs according to a rank adaptation period defined by an RI adaptation periodicity. Note that the rank adaptation prerequisite is met when the CSI report occurs according to the rank adaptation period.

[0096] In third such cases, the UE 402 may generate and report PMI information based on a maximum rank restriction (received from the base station 404), and may additionally, in some examples, indicate a scheduling recommended RI (a maximum possible RI), to the base station 404. For example, the UE 402 may receive, from the base station 404, a rank restriction and may generate, based on the maximum possible RI indicated in the rank restriction, one or more PMIs corresponding to the maximum possible RI. Then, the UE 402 may generate encoded PMI information by applying the PMIs at an AI / ML model and transmit, to the base station 404, a CSI report comprising the encoded PMI information and the maximum possible RI.

[0097] In fourth such cases, the UE 402 may select a RI for each CSI report (reported in, for example, the UCI report, discussed herein). In such cases, the changing RI may affect the use of layer-specific state information being used by one or more AI / ML model(s) (layer-specific AI / ML models(s) and layer-common AI / ML model(s)), as discussed herein. In such cases, when RI changes, it may be that state information is kept for Layer 1, and the treatment of any additional layers (e.g., Layers 2 and above) pertaining to the new RI depends on whether the previous state for that layer is available or not. In some such cases where one or more such previous state(s) are available for one or more layer(s), these state(s) may be reused, while an initial state may be assumed for any layer(s) for which a prior state is not available. Additionally, in at least some instances for the fourth cases, if the network does not configure a rank restriction, the UE 402 may be free to select the RI for each CSI report.

[0098] For example, according to the fourth such cases, the UE 402 may generate, a plurality of PMIs and a corresponding updated RI that is different than a previous RI indicated to the base station 404. Then, the UE 402 may generate encoded PMI information by applying the PMIs at AI / ML models, wherein the AI / ML models utilize layer-specific state information. Accordingly, the UE 402 may transmit, to the base station 404, a CSI report comprising the encoded PMI information and the updated RI.

[0099] Then, as shown, the base station 404 may perform a CSLRS transmission 412 to the UE 402. The UE 402 acquires CSI information (including, e.g., a PMI) for a channelbetween the UE and the base station 404 by measuring the CSI-RS. The UE 402 may then formulate a CSI report using that CSI-RS information. This CSI report may include an encoded PMI that is generated using the acquired PMI using one or more AI / ML models and the rank restriction information.

[0100] The UE 402 then transmits UCI report 414 to the base station 404 that includes a CSI report having the encoded PMI. Note the CSI-RS transmission 412 and responsive UCI report 314 may correspond to a first state (i.e., “State 0”). The transmission of the UCI report 414 may be according to one of the cases for limiting / organizing the UE’s use of RI indications / rank restrictions, as indicated in the triggering indication 410.

[0101] As shown, over time, subsequent, CSI-RS transmissions 416 and responsive transmissions of UCI reports 418 (corresponding to an applicable / configured one of the cases for limiting / organizing the UE’s use of RI indications) may follow suit corresponding to subsequent states (such as State 1, State 2, ..., State N). This may occur until, for example, a resetting indication 420 with rank restriction information is transmitted (according to various cases similar to those discussed for the triggering resetting indication 420 discussed herein).

[0102] After a resetting indication 420 is received by the UE 402, any state information used by the encoder / decoder of the AI / ML model is dropped and the encoder and the decoder of the AI / ML model are thus returned to an initial state, including, in some cases, resetting the RI to an initial state so that it may correspond to the current rank restriction. Note that, in some cases, the rank restriction may remain the same as indicated in the triggering indication 410. The CSI process may then be continued under the (potentially new) rank restriction, (see, e.g., the CSI-RS transmission 422 and the UCI report 424). This may occur in a manner that is affected by prior PMI information generated prior to the resetting indication 420 for layers that remain in use after applying the (potentially) new rank restriction, as discussed herein.

[0103] Details with respect to the various ones of the cases for limiting / organizing the UE’s use of RI indications as introduced above are now discussed.

[0104] FIG. 5 A illustrates an example 500 of UE based rank adaptation where the UE uses an additional rank adaptation restriction, according to embodiments herein.

[0105] In the example 500, the RI of CSI reports 510 (based on CSI-RS transmissions 508 with a periodic CSI-RS (p-CSLRS), a semi-persistent CSI-RS (sp-CSLRS), or an aperiodic CSI-RS (ap-CSLRS) 512) may be restricted to a UE reported RI 516 for atleast N CSI reports 514. As a result, as shown the UE may not report a different RI 518 until sometime after N CSI reports 514 that have the same RI value. The updated different RI 518 may then be used for the subsequent N CSI reports 520. It should be understood that the rank adaptation prerequisite may be met by the UE identifying that a number of prior CSI reports 510 based on the previous RI meets a minimum RI use threshold. Alternatively, CSI report 522 (or any subsequent CSI report to the N CSI reports 514) may have been used to report the updated different RI 518.

[0106] FIG. 5B illustrates an example 502 of UE based rank adaptation where the UE uses an additional rank adaptation restriction, according to embodiments herein.

[0107] In the example 502, the RI of CSI reports 510 (based on CSI-RS transmissions 508 with a p-CSI-RS, sp-CSI-RS, or ap-CSI-RS 512) may be restricted to a UE reported RI 524, for N CSI reports 526, where N is based on a rank adaptation periodicity (e.g., as indicated by the base station). The N CSI reports 526 may have the same RI for the periodicity of “N” CSI reports 510, however, after the N CSI reports 526 (e.g., four CSI reports), the UE may update 528 (or not) the RI that is used for the next “N” CSI reports 510, as the next “N” CSI reports 510 are scheduled to begin according to the rank adaptation period. It should be understood that the UE identifying that the rank adaptation prerequisite is met includes the UE identifying that the CSI report 510 occurs according to the rank adaptation period.

[0108] FIG. 5C illustrates an example 504 of UE based rank adaptation where the UE reports generated encoded PMI information based on a maximum rank restriction received from a base station, according to embodiments herein.

[0109] In the example 504, the RI of CSI reports 510 (based on CSI-RS transmissions 508 with a p-CSI-RS, sp-CSI-RS, or ap-CSI-RS 512) may equal a maximum rank restriction received from the base station. Accordingly, it may be understood that the UE generates PMI(s) for the CSI report 510 based on the maximum rank restriction using the CSI-RS transmissions 508, and then generates encoded PMI information by applying the PMI(s) at AI / ML model(s) in one of the manners discussed herein. The CSI reports 510 including the generated, encoded PMI information according to an RI equal to the maximum rank restriction are then sent to the base station.

[0110] Note that the maximum rank restriction from the network that controls the generation and report of the PMIs 530 as described may be updated 532 at various points throughout the transmission of the CSI reports 510. The UE may further send the basestation (e.g., in or alongside a CSI report 510, or separately therefrom) an indication of a rank that is preferred by the UE, which may be used by the base station to update the maximum rank restriction that is indicated to the UE for use.[OHl] FIG. 5D illustrates an example 506 of UE based rank adaptation where the UE may select a RI for each report, according to embodiments herein.

[0112] In example 506, the RI of each CSI report 510 (based on CSI-RS transmissions 508 with a p-CSI-RS, sp-CSI-RS, or ap-CSI-RS 512) may be updated 534 for every CSI report 510.

[0113] In such cases, the changing RI may affect the use of layer-specific state information being used by one or more AI / ML model(s) (layer-specific AI / ML models(s) and / or layer-common AI / ML model(s)), as discussed herein. In such cases, when RI changes, it may be that state information is kept for Layer 1 536, and the treatment of any additional layers (e.g., Layers 2 538 and above) pertaining to the new RI depends on whether previous state for that layer is available or not. In some such cases where one or more such previous state(s) are available for one or more layer(s), these state(s) may be reused, while an initial state may be assumed for any layer(s) for which a prior state is not available. Additionally, in at least some instances for the fourth cases, if the network does not configure a rank restriction, the UE may be free to select the RI for each CSI report.

[0114] For example, for Layer 1 536, the state information may be the same from state tO 540 through state t6, and for Layer 2 538, the state information may be the same from state tO 542 to state t2 548 (e.g., with a RI value of 2). When the RI value changes 544 (of Layer 2 538), (e.g., from an RI value of 2 to an RI value of 1), the state information changes to a second set of state information that may stay the same for each state (as a previous state is available) until the RI value changes again, to, for example, a RI value of 2 and as a result, the state resets back to state tO 546 (a previous state) having state information similar to that of the state information of the original state tO 542.

[0115] FIG. 6 illustrates a flow diagram 600 between a UE 602 and a base station 604 for rank restriction based on a UE 602 recommendation for rank restriction sent to the base station 604, according to embodiments herein.

[0116] In some embodiments, the network may take into consideration a UE rank restriction recommendation, as transmitted, to the network, in for example, an uplink (UL) MAC CE or a UE assistance information (UAI) message. The UE may recommenda rank update frequency (e.g., every N values) as discussed herein. Then, the network may configure the rank restriction based on the UE recommendation and scheduling (e.g., for multiple user (MU) MIMO availability).

[0117] For example, the flow diagram 600 begins with the UE 602 reporting 606, to the base station 604, UE 602 supported AI / ML model(s) with their corresponding model ID (similarly to that in FIG. 3 and / or FIG. 4). Then, the base station 604 may transmit 608 RRC configuration (including c si ReportConfig IE and / or csi ResourceConfig IE, as discussed herein in FIG. 3 and / or FIG. 4), to the UE 602. Subsequently, the UE 602 may send 610, to the base station 604, a recommendation for the rank restriction including, for example, a recommended RI value, and a recommended timing as to when to update the RI (based on the current implementation of the UE, such as mobility). Then, the base station 604 may configure 612 the rank restriction, based on the received recommendation from the UE 602 and may transmit 614 a triggering indication with the rank restriction (in the form of, for example, DCI or MAC CE, to the UE 602, as previously discussed herein). For example, if the UE 602 is not very mobile, the UE 602 may recommend a large RI updating periodicity to the base stations 604 as there may be no need to update the RI often due to the UE immobility. In another example, if the UE 602 is mobile, the UE 602 may recommend a small RI value to the base station 604.

[0118] In some cases, the UE 602 may transmit 614 a confirmation to the UE 602 to use the recommended rank restriction. Note that the recommendation transmitted by the UE may be understood as a recommended minimum RI use threshold where the RI is recommended to be above the minimum RI use threshold.

[0119] FIG. 7 illustrates a method 700 of a UE for AI / ML model-based CSI reporting, according to embodiments herein. The illustrated method 700 includes identifying 702 that a rank adaptation prerequisite for differing from a previous RI indicated to a base station by the UE is met. The method 700 further includes generating 704, in response to the identifying that the rank adaptation prerequisite is met, using a measurement of a CSLRS transmitted by the base station, one or more PMIs and a corresponding updated RI that is different than the previous RI. The method 700 further includes generating 706, encoded PMI information by applying the one or more PMIs at one or more AI / ML models. The method 700 further includes transmitting 708, to the base station, a CSI report comprising the encoded PMI information and the updated RI.

[0120] In some embodiments, the method 700 further comprises receiving, from the base station, configuration information defining the rank adaptation prerequisite.

[0121] In some embodiments of the method 700, the rank adaptation prerequisite comprises meeting a minimum RI use threshold, and wherein the identifying that the rank adaptation prerequisite is met comprises identifying that a number of prior CSI reports based on the previous RI meets the minimum RI use threshold.

[0122] Some such embodiments further comprise sending, to the base station, a recommended minimum RI use threshold, and receiving, from the network, the minimum RI use threshold, wherein the minimum RI use threshold is equal to the recommended minimum RI use threshold.

[0123] In some embodiments of the method 700, the rank adaptation prerequisite comprises that the CSI report occurs according to a rank adaptation period defined by a rank adaptation periodicity, and wherein the identifying that the rank adaptation prerequisite is met comprises identifying that the CSI report occurs according to the rank adaptation period.

[0124] In some embodiments of the method 700, the one or more AI / ML models comprises a first AI / ML model that applies the one or more PMIs and layer-specific state information for each of one or more layers indicated by the updated RI to a first statebased encoder of the AI / ML model to generate the encoded PMI information.

[0125] In some embodiments of the method 700, the one or more AI / ML models comprises a first AI / ML model that applies the one or more PMIs and layer-common state information for one or more layers indicated by the updated RI to a first state-based encoder of the AI / ML model to generate the encoded PMI information.

[0126] In some embodiments of the method 700, the one or more PMIs comprises a plurality of PMIs; the one or more AI / ML models comprises a plurality of AI / ML models; a first AI / ML model of the plurality of AI / ML models applies a first PMI of the plurality of PMIs and first layer-specific state information for a first layer indicated by the updated RI to a first state-based encoder of the first AI / ML model to generate a first portion of the encoded PMI information; and a second AI / ML model of the plurality of AI / ML models applies a second PMI of the plurality of PMIs and second layer-specific state information for a second layer indicated by the updated RI to a second state-based encoder of the second AI / ML model to generate a second portion of the encoded PMI information.

[0127] In some embodiments of the method 700, the one or more AI / ML models comprises a first AI / ML model using a first state-based encoder that generates at least a portion of the encoded PMI information by taking default state information as layerspecific state information.

[0128] FIG. 8 illustrates a method 800 of a UE for AI / ML model-based CSI reporting, according to embodiments herein. The illustrated method 800 includes receiving 802, from a base station, a rank restriction. The method 800 further includes generating 804, based on an identification of a maximum possible RI under the rank restriction, using a measurement of a CSLRS transmitted by the base station, one or more PMIs corresponding to the maximum possible RI. The method 800 further includes generating 806 encoded PMI information by applying the one or more PMIs at one or more AI / ML models. The method 800 further includes transmitting 808, to the base station, a CSI report comprising the encoded PMI information and the maximum possible RI.

[0129] In some embodiments of the method 800, the one or more AI / ML models comprises a first AI / ML model that applies the one or more PMIs and layer-specific state information for each of one or more layers indicated by the maximum possible RI to a first state-based encoder of the AI / ML model to generate the encoded PMI information.

[0130] In some embodiments of the method 800, the one or more AI / ML models comprises a first AI / ML model that applies the one or more PMIs and layer-common state information for one or more layers indicated by the maximum possible RI to a first state-based encoder of the AI / ML model to generate the encoded PMI information.

[0131] In some embodiments of the method 800, the rank restriction is received from the base station in one of RRC signaling, a MAC CE, and a DCI.

[0132] In some embodiments of the method 800, the one or more PMIs comprises a plurality of PMIs; the one or more AI / ML models comprises a plurality of AI / ML models; a first AI / ML model of the plurality of AI / ML models applies a first PMI of the plurality of PMIs and first layer-specific state information for a first layer indicated by the maximum possible RI to a first state-based encoder of the first AI / ML model to generate a first portion of the encoded PMI information; and a second AI / ML model of the plurality of AI / ML models applies a second PMI of the plurality of PMIs and second layer-specific state information for a second layer indicated by the maximum possible RI to a second state-based encoder of the second AI / ML model to generate a second portion of the encoded PMI information.

[0133] In some embodiments of the method 800, the one or more AI / ML models comprises a first AI / ML model using a first state-based encoder that generates at least a portion of the encoded PMI information by taking default state information as layerspecific state information.

[0134] FIG. 9 illustrates a method 900 of a UE for AI / ML model-based CSI reporting, according to embodiments herein. The illustrated method 900 includes generating 902, using a measurement of a CSLRS transmitted by a base station, a plurality of PMIs and a corresponding updated RI that is different than a previous RI indicated to the base station. The method 900 further includes generating 904 encoded PMI information by applying the plurality of PMIs at a plurality of AI / ML models, wherein the plurality of AI / ML models comprises: a first AI / ML model that applies a first PMI of the plurality of PMIs and first layer-specific state information for a first layer indicated by the updated RI to a first state-based encoder of the first AI / ML model to generate a first portion of the encoded PMI information, wherein the first layer-specific state information comprises evaluated state information previously evaluated at the first state-based encoder, and a second AI / ML model of the plurality of AI / ML models applies a second PMI of the plurality of PMIs and second layer-specific state information for a second layer indicated by the updated RI to a second state-based encoder of the second AI / ML model to generate a second portion of the encoded PMI information, wherein the second layer-specific state information comprises default state information. The method 900 further includes transmitting 906, to the base station, a CSI report comprising the encoded PMI information and the updated RI.

[0135] FIG. 10 illustrates a method 1000 of a base station for AI / ML model -based CSI reporting, according to embodiments herein. The illustrated method 1000 includes sending 1002, to a UE, configuration information defining a rank adaptation prerequisite. The method 1000 further includes receiving 1004, from the UE, a CSI report comprising encoded PMI information and a RI consistent with the rank adaptation prerequisite. The method 1000 further includes applying 1006 the encoded PMI information at one or more AI / ML models to generate one or more PMIs. The method 1000 further includes sending 1008 a transmission, to the UE, wherein the transmission is of a rank indicated by the RI and is precoded based on the one or more PMIs.

[0136] In some embodiments of the method 1000, the RI comprises an updated RI that is different than a previous RI indicated to the base station.

[0137] In some embodiments of the method 1000, the rank adaptation prerequisite comprises meeting a minimum RI use threshold. Some such embodiments further comprise receiving, from the UE, a recommended minimum RI use threshold, and wherein the configuration information defines the minimum RI use threshold as equal to the recommended minimum RI use threshold.

[0138] In some embodiments of the method 1000, the rank adaptation prerequisite comprises that the CSI report occurs during a rank adaptation period defined by a rank adaptation periodicity.

[0139] In some embodiments of the method 1000, the one or more AI / ML models comprises a first AI / ML model that applies the encoded PMI information and layerspecific state information for each of one or more layers indicated by the RI to a first state-based decoder of the AI / ML model to generate the one or more PMIs.

[0140] In some embodiments of the method 1000, the one or more AI / ML models comprises a first AI / ML model that applies the encoded PMI information and layercommon state information for one or more layers indicated by the RI to a first statebased decoder of the first AI / ML model to generate the one or more PMIs.

[0141] In some embodiments of the method 1000, the one or more PMIs comprises a plurality of PMIs; the one or more AI / ML models comprises a plurality of AI / ML models; a first AI / ML model of the plurality of AI / ML models applies a first portion of the encoded PMI information and first layer-specific state information for a first layer indicated by the RI to a first state-based decoder of the first AI / ML model to generate a first PMI of the plurality of PMIs; and a second AI / ML model of the plurality of AI / ML models applies a second portion of the encoded PMI information and second layerspecific state information for a second layer indicated by the RI to a second state-based decoder of the second AI / ML model to generate a second PMI of the plurality of PMIs.

[0142] In some embodiments of the method 1000, the one or more AI / ML models comprises a first AI / ML model using a first state-based decoder that generates at least a first PMI of the one or more PMIs by taking default state information as layer-specific state information.

[0143] FIG. 11 illustrates an example architecture of a wireless communication system 1100, according to embodiments disclosed herein. The following description is provided for an example wireless communication system 1100 that operates in conjunction withthe LTE system standards and / or 5G or NR system standards as provided by 3 GPP technical specifications.

[0144] As shown by FIG. 11, the wireless communication system 1100 includes UE 1102 and UE 1104 (although any number of UEs may be used). In this example, the UE 1102 and the UE 1104 are illustrated as smartphones (e.g., handheld touchscreen mobile computing devices connectable to one or more cellular networks), but may also comprise any mobile or non-mobile computing device configured for wireless communication.

[0145] The UE 1102 and UE 1104 may be configured to communicatively couple with a RAN 1106. In embodiments, the RAN 1106 may be NG-RAN, E-UTRAN, etc. The UE 1102 and UE 1104 utilize connections (or channels) (shown as connection 1108 and connection 1110, respectively) with the RAN 1106, each of which comprises a physical communications interface. The RAN 1106 can include one or more base stations (such as base station 1112 and base station 1114) that enable the connection 1108 and connection 1110.

[0146] In this example, the connection 1108 and connection 1110 are air interfaces to enable such communicative coupling, and may be consistent with RAT(s) used by the RAN 1106, such as, for example, an LTE and / or NR.

[0147] In some embodiments, the UE 1102 and UE 1104 may also directly exchange communication data via a sidelink interface 1116. The UE 1104 is shown to be configured to access an access point (shown as AP 1118) via connection 1120. By way of example, the connection 1120 can comprise a local wireless connection, such as a connection consistent with any IEEE 802.11 protocol, wherein the AP 1118 may comprise a Wi-Fi® router. In this example, the AP 1118 may be connected to another network (for example, the Internet) without going through a CN 1124.

[0148] In embodiments, the UE 1102 and UE 1104 can be configured to communicate using orthogonal frequency division multiplexing (OFDM) communication signals with each other or with the base station 1112 and / or the base station 1114 over a multicarrier communication channel in accordance with various communication techniques, such as, but not limited to, an orthogonal frequency division multiple access (OFDMA) communication technique (e.g., for downlink communications) or a single carrier frequency division multiple access (SC-FDMA) communication technique (e.g., for uplink and ProSe or sidelink communications), although the scope of the embodiments isnot limited in this respect. The OFDM signals can comprise a plurality of orthogonal subcarriers.

[0149] In some embodiments, all or parts of the base station 1112 or base station 1114 may be implemented as one or more software entities running on server computers as part of a virtual network. In addition, or in other embodiments, the base station 1112 or base station 1114 may be configured to communicate with one another via interface 1122. In embodiments where the wireless communication system 1100 is an LTE system (e.g., when the CN 1124 is an EPC), the interface 1122 may be an X2 interface. The X2 interface may be defined between two or more base stations (e.g., two or more eNBs and the like) that connect to an EPC, and / or between two eNBs connecting to the EPC. In embodiments where the wireless communication system 1100 is an NR system (e.g., when CN 1124 is a 5GC), the interface 1122 may be an Xn interface. The Xn interface is defined between two or more base stations (e.g., two or more gNBs and the like) that connect to 5GC, between a base station 1112 (e.g., a gNB) connecting to 5GC and an eNB, and / or between two eNBs connecting to 5GC (e.g., CN 1124).

[0150] The RAN 1106 is shown to be communicatively coupled to the CN 1124. The CN 1124 may comprise one or more network elements 1126, which are configured to offer various data and telecommunications services to customers / subscribers (e.g., users of UE 1102 and UE 1104) who are connected to the CN 1124 via the RAN 1106. The components of the CN 1124 may be implemented in one physical device or separate physical devices including components to read and execute instructions from a machine- readable or computer-readable medium (e.g., a non-transitory machine-readable storage medium).

[0151] In embodiments, the CN 1124 may be an EPC, and the RAN 1106 may be connected with the CN 1124 via an SI interface 1128. In embodiments, the SI interface 1128 may be split into two parts, an SI user plane (Sl-U) interface, which carries traffic data between the base station 1112 or base station 1114 and a serving gateway (S-GW), and the SI -MME interface, which is a signaling interface between the base station 1112 or base station 1114 and mobility management entities (MMEs).

[0152] In embodiments, the CN 1124 may be a 5GC, and the RAN 1106 may be connected with the CN 1124 via an NG interface 1128. In embodiments, the NG interface 1128 may be split into two parts, an NG user plane (NG-U) interface, which carries traffic data between the base station 1112 or base station 1114 and a user planefunction (UPF), and the SI control plane (NG-C) interface, which is a signaling interface between the base station 1112 or base station 1114 and access and mobility management functions (AMFs).

[0153] Generally, an application server 1130 may be an element offering applications that use internet protocol (IP) bearer resources with the CN 1124 (e.g., packet switched data services). The application server 1130 can also be configured to support one or more communication services (e.g., VoIP sessions, group communication sessions, etc.) for the UE 1102 and UE 1104 via the CN 1124. The application server 1130 may communicate with the CN 1124 through an IP communications interface 1132.

[0154] FIG. 12 illustrates a system 1200 for performing signaling 1234 between a wireless device 1202 and a network device 1218, according to embodiments disclosed herein. The system 1200 may be a portion of a wireless communications system as herein described. The wireless device 1202 may be, for example, a UE of a wireless communication system. The network device 1218 may be, for example, a base station (e.g., an eNB or a gNB) of a wireless communication system.

[0155] The wireless device 1202 may include one or more processor(s) 1204. The processor(s) 1204 may execute instructions such that various operations of the wireless device 1202 are performed, as described herein. The processor(s) 1204 may include one or more baseband processors implemented using, for example, a central processing unit (CPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), a controller, a field programmable gate array (FPGA) device, another hardware device, a firmware device, or any combination thereof configured to perform the operations described herein.

[0156] The wireless device 1202 may include a memory 1206. The memory 1206 may be a non-transitory computer-readable storage medium that stores instructions 1208 (which may include, for example, the instructions being executed by the processor(s) 1204). The instructions 1208 may also be referred to as program code or a computer program. The memory 1206 may also store data used by, and results computed by, the processor(s) 1204.

[0157] The wireless device 1202 may include one or more transceiver(s) 1210 that may include radio frequency (RF) transmitter circuitry and / or receiver circuitry that use the antenna(s) 1212 of the wireless device 1202 to facilitate signaling (e.g., the signaling1234) to and / or from the wireless device 1202 with other devices (e.g., the network device 1218) according to corresponding RATs.

[0158] The wireless device 1202 may include one or more antenna(s) 1212 (e.g., one, two, four, or more). For embodiments with multiple antenna(s) 1212, the wireless device 1202 may leverage the spatial diversity of such multiple antenna(s) 1212 to send and / or receive multiple different data streams on the same time and frequency resources. This behavior may be referred to as, for example, multiple input multiple output (MIMO) behavior (referring to the multiple antennas used at each of a transmitting device and a receiving device that enable this aspect). MIMO transmissions by the wireless device 1202 may be accomplished according to precoding (or digital beamforming) that is applied at the wireless device 1202 that multiplexes the data streams across the antenna(s) 1212 according to known or assumed channel characteristics such that each data stream is received with an appropriate signal strength relative to other streams and at a desired location in the spatial domain (e.g., the location of a receiver associated with that data stream). Certain embodiments may use single user MIMO (SU-MIMO) methods (where the data streams are all directed to a single receiver) and / or multi user MIMO (MU-MIMO) methods (where individual data streams may be directed to individual (different) receivers in different locations in the spatial domain).

[0159] In certain embodiments having multiple antennas, the wireless device 1202 may implement analog beamforming techniques, whereby phases of the signals sent by the antenna(s) 1212 are relatively adjusted such that the (joint) transmission of the antenna(s) 1212 can be directed (this is sometimes referred to as beam steering).

[0160] The wireless device 1202 may include one or more interface(s) 1214. The interface(s) 1214 may be used to provide input to or output from the wireless device 1202. For example, a wireless device 1202 that is a UE may include interface(s) 1214 such as microphones, speakers, a touchscreen, buttons, and the like in order to allow for input and / or output to the UE by a user of the UE. Other interfaces of such a UE may be made up of transmitters, receivers, and other circuitry (e.g., other than the transceiver(s) 1210 / antenna(s) 1212 already described) that allow for communication between the UE and other devices and may operate according to known protocols (e.g., Wi-Fi®, Bluetooth®, and the like).

[0161] The wireless device 1202 may include a rank adaptation module 1216. The rank adaptation module 1216 may be implemented via hardware, software, or combinationsthereof. For example, the rank adaptation module 1216 may be implemented as a processor, circuit, and / or instructions 1208 stored in the memory 1206 and executed by the processor(s) 1204. In some examples, the rank adaptation module 1216 may be integrated within the processor(s) 1204 and / or the transceiver(s) 1210. For example, the rank adaptation module 1216 may be implemented by a combination of software components (e.g., executed by a DSP or a general processor) and hardware components (e.g., logic gates and circuitry) within the processor(s) 1204 or the transceiver(s) 1210.

[0162] The rank adaptation module 1216 may be used for various aspects of the present disclosure, for example, aspects of any of FIG. 1, FIG. 2, FIG. 3, FIG. 4, FIG. 5A, FIG. 5B, FIG. 5C, FIG. 5D, FIG. 6, and FIG. 7. In some cases, the rank adaptation module 1216 may be configured to cause the wireless device 1202, used for AI / ML model -based CSI reporting, to identify that a rank adaptation prerequisite for differing from a previous RI indicated to a network device 1218 by the wireless device 1202 is met.Further, the rank adaptation module 1216 may be configured to cause the wireless device 1202 to generate, in response to the identifying that the rank adaptation prerequisite is met, using a measurement of a CSI-RS transmitted by the base station, one or more PMIs and a corresponding updated RI that is different than the previous RI and generate encoded PMI information by applying the one or more PMIs at one or more AI / ML models. The rank adaptation module 1216 may be further configured to cause the wireless device 1202 to transmit, to the network device 1218, a CSI report comprising the encoded PMI information and the updated RI.

[0163] In some other cases, the rank adaptation module 1216 may be configured to cause the wireless device 1202 to receive, from a network device 1218, a rank restriction, generate, based on an identification of a maximum possible RI under the rank restriction, using a measurement of a CSI-RS transmitted by the base station, one or more PMIs corresponding to the maximum possible RI, and generate encoded PMI information by applying the one or more PMIs at one or more AI / ML models. The rank adaptation module 1216 may be further configured to cause the wireless device 1202 to transmit, to the network device 1218, a CSI report comprising the encoded PMI information and the maximum possible RI.

[0164] In yet some other cases, the rank adaptation module 1216 may be configured to cause the wireless device 1202 to generate, using a measurement of a CSI-RS transmitted by a base station, a plurality of PMIs and a corresponding updated RI that isdifferent than a previous RI indicated to the base station. The rank adaptation module 1216 may be further configured to cause the wireless device 1202 to generate encoded PMI information by applying the plurality of PMIs at a plurality of AI / ML models, wherein the plurality of AI / ML models comprises: a first AI / ML model that applies a first PMI of the plurality of PMIs and first layer-specific state information for a first layer indicated by the updated RI to a first state-based encoder of the first AI / ML model to generate a first portion of the encoded PMI information, wherein the first layerspecific state information comprises evaluated state information previously evaluated at the first state-based encoder, and a second AI / ML model of the plurality of AI / ML models applies a second PMI of the plurality of PMIs and second layer-specific state information for a second layer indicated by the updated RI to a second state-based encoder of the second AI / ML model to generate a second portion of the encoded PMI information, wherein the second layer-specific state information comprises default state information. The rank adaptation module 1216 may be further configured to cause the wireless device 1202 to transmit, to the network device 1218, a CSI report comprising the encoded PMI information and the updated RI.

[0165] The network device 1218 may include one or more processor(s) 1220. The processor(s) 1220 may execute instructions such that various operations of the network device 1218 are performed, as described herein. The processor(s) 1220 may include one or more baseband processors implemented using, for example, a CPU, a DSP, an ASIC, a controller, an FPGA device, another hardware device, a firmware device, or any combination thereof configured to perform the operations described herein.

[0166] The network device 1218 may include a memory 1222. The memory 1222 may be a non-transitory computer-readable storage medium that stores instructions 1224 (which may include, for example, the instructions being executed by the processor(s) 1220). The instructions 1224 may also be referred to as program code or a computer program. The memory 1222 may also store data used by, and results computed by, the processor(s) 1220.

[0167] The network device 1218 may include one or more transceiver(s) 1226 that may include RF transmitter circuitry and / or receiver circuitry that use the antenna(s) 1228 of the network device 1218 to facilitate signaling (e.g., the signaling 1234) to and / or from the network device 1218 with other devices (e.g., the wireless device 1202) according to corresponding RATs.

[0168] The network device 1218 may include one or more antenna(s) 1228 (e.g., one, two, four, or more). In embodiments having multiple antenna(s) 1228, the network device 1218 may perform MIMO, digital beamforming, analog beamforming, beam steering, etc., as has been described.

[0169] The network device 1218 may include one or more interface(s) 1230. The interface(s) 1230 may be used to provide input to or output from the network device 1218. For example, a network device 1218 that is a base station may include interface(s) 1230 made up of transmitters, receivers, and other circuitry (e.g., other than the transceiver(s) 1226 / antenna(s) 1228 already described) that enables the base station to communicate with other equipment in a core network, and / or that enables the base station to communicate with external networks, computers, databases, and the like for purposes of operations, administration, and maintenance of the base station or other equipment operably connected thereto.

[0170] The network device 1218 may include a rank adaptation module 1232. The rank adaptation module 1232 may be implemented via hardware, software, or combinations thereof. For example, the rank adaptation module 1232 may be implemented as a processor, circuit, and / or instructions 1224 stored in the memory 1222 and executed by the processor(s) 1220. In some examples, the rank adaptation module 1232 may be integrated within the processor(s) 1220 and / or the transceiver(s) 1226. For example, the rank adaptation module 1232 may be implemented by a combination of software components (e.g., executed by a DSP or a general processor) and hardware components (e.g., logic gates and circuitry) within the processor(s) 1220 or the transceiver(s) 1226.

[0171] The rank adaptation module 1232 may be used for various aspects of the present disclosure, for example, aspects of any of FIG. 1, FIG. 2, FIG. 3, FIG. 4, FIG. 5A, FIG. 5B, FIG. 5C, FIG. 5D, FIG. 6, and FIG. 7. The rank adaptation module 1232 is configured to cause the network device 1218 to send, to a wireless device 1202, configuration information defining a rank adaptation prerequisite and receive, from the wireless device 1202, a CSI report comprising encoded PMI information and a RI consistent with the rank adaptation prerequisite. The rank adaptation module 1232 may further configure the network device 1218 to apply the encoded PMI information at one or more AI / ML models to generate one or more PMIs, and send a transmission to the wireless device 1202, wherein the transmission is of a rank indicated by the RI and is precoded based on the one or more PMIs

[0172] Embodiments contemplated herein include an apparatus comprising means to perform one or more elements of any one of the method 700, method 800, and method 900. This apparatus may be, for example, an apparatus of a UE (such as a wireless device 1202 that is a UE, as described herein).

[0173] Embodiments contemplated herein include one or more non-transitory computer-readable media comprising instructions to cause an electronic device, upon execution of the instructions by one or more processors of the electronic device, to perform one or more elements of any one of the method 700, method 800, and method 900. This non-transitory computer-readable media may be, for example, a memory of a UE (such as a memory 1206 of a wireless device 1202 that is a UE, as described herein).

[0174] Embodiments contemplated herein include an apparatus comprising logic, modules, or circuitry to perform one or more elements of any one of the method 700, method 800, and method 900. This apparatus may be, for example, an apparatus of a UE (such as a wireless device 1202 that is a UE, as described herein).

[0175] Embodiments contemplated herein include an apparatus comprising: one or more processors and one or more computer-readable media comprising instructions that, when executed by the one or more processors, cause the one or more processors to perform one or more elements of any one of the method 700, method 800, and method 900. This apparatus may be, for example, an apparatus of a UE (such as a wireless device 1202 that is a UE, as described herein).

[0176] Embodiments contemplated herein include a signal as described in or related to one or more elements of any one of the method 700, method 800, and method 900.

[0177] Embodiments contemplated herein include a computer program or computer program product comprising instructions, wherein execution of the program by a processor is to cause the processor to carry out one or more elements of any one of the method 700, method 800, and method 900. The processor may be a processor of a UE (such as a processor(s) 1204 of a wireless device 1202 that is a UE, as described herein). These instructions may be, for example, located in the processor and / or on a memory of the UE (such as a memory 1206 of a wireless device 1202 that is a UE, as described herein).

[0178] Embodiments contemplated herein include an apparatus comprising means to perform one or more elements of the method 1000. This apparatus may be, for example,an apparatus of a base station (such as a network device 1218 that is a base station, as described herein).

[0179] Embodiments contemplated herein include one or more non-transitory computer-readable media comprising instructions to cause an electronic device, upon execution of the instructions by one or more processors of the electronic device, to perform one or more elements of the method 1000. This non-transitory computer- readable media may be, for example, a memory of a base station (such as a memory 1222 of a network device 1218 that is a base station, as described herein).

[0180] Embodiments contemplated herein include an apparatus comprising logic, modules, or circuitry to perform one or more elements of the method 1000. This apparatus may be, for example, an apparatus of a base station (such as a network device 1218 that is a base station, as described herein).

[0181] Embodiments contemplated herein include an apparatus comprising: one or more processors and one or more computer-readable media comprising instructions that, when executed by the one or more processors, cause the one or more processors to perform one or more elements of the method 1000. This apparatus may be, for example, an apparatus of a base station (such as a network device 1218 that is a base station, as described herein).

[0182] Embodiments contemplated herein include a signal as described in or related to one or more elements of the method 1000.

[0183] Embodiments contemplated herein include a computer program or computer program product comprising instructions, wherein execution of the program by a processing element is to cause the processing element to carry out one or more elements of the method 1000. The processor may be a processor of a base station (such as a processor(s) 1220 of a network device 1218 that is a base station, as described herein). These instructions may be, for example, located in the processor and / or on a memory of the base station (such as a memory 1222 of a network device 1218 that is a base station, as described herein).

[0184] For one or more embodiments, at least one of the components set forth in one or more of the preceding figures may be configured to perform one or more operations, techniques, processes, and / or methods as set forth herein. For example, a baseband processor as described herein in connection with one or more of the preceding figures may be configured to operate in accordance with one or more of the examples set forthherein. For another example, circuitry associated with a UE, base station, network element, etc. as described above in connection with one or more of the preceding figures may be configured to operate in accordance with one or more of the examples set forth herein.

[0185] Any of the above described embodiments may be combined with any other embodiment (or combination of embodiments), unless explicitly stated otherwise. The foregoing description of one or more implementations provides illustration and description, but is not intended to be exhaustive or to limit the scope of embodiments to the precise form disclosed. Modifications and variations are possible in light of the above teachings or may be acquired from practice of various embodiments.

[0186] Embodiments and implementations of the systems and methods described herein may include various operations, which may be embodied in machine-executable instructions to be executed by a computer system. A computer system may include one or more general-purpose or special-purpose computers (or other electronic devices). The computer system may include hardware components that include specific logic for performing the operations or may include a combination of hardware, software, and / or firmware.

[0187] It should be recognized that the systems described herein include descriptions of specific embodiments. These embodiments can be combined into single systems, partially combined into other systems, split into multiple systems or divided or combined in other ways. In addition, it is contemplated that parameters, attributes, aspects, etc. of one embodiment can be used in another embodiment. The parameters, attributes, aspects, etc. are merely described in one or more embodiments for clarity, and it is recognized that the parameters, attributes, aspects, etc. can be combined with or substituted for parameters, attributes, aspects, etc. of another embodiment unless specifically disclaimed herein.

[0188] It is well understood that the use of personally identifiable information should follow privacy policies and practices that are generally recognized as meeting or exceeding industry or governmental requirements for maintaining the privacy of users. In particular, personally identifiable information data should be managed and handled so as to minimize risks of unintentional or unauthorized access or use, and the nature of authorized use should be clearly indicated to users.

[0189] Although the foregoing has been described in some detail for purposes of clarity, it will be apparent that certain changes and modifications may be made without departing from the principles thereof. It should be noted that there are many alternative ways of implementing both the processes and apparatuses described herein. Accordingly, the present embodiments are to be considered illustrative and not restrictive, and the description is not to be limited to the details given herein, but may be modified within the scope and equivalents of the appended claims.

Claims

CLAIMS1. A method of a user equipment (UE) for (AI) / machine learning (ML) model-based channel state information (CSI) reporting, comprising: identifying that a rank adaptation prerequisite for differing from a previous rank indication (RI) indicated to a base station by the UE is met; generating, in response to the identifying that the rank adaptation prerequisite is met, using a measurement of a channel state information reference signal (CSI-RS) transmitted by the base station, one or more precoder matrix indicators (PMIs) and a corresponding updated RI that is different than the previous RI; generating encoded PMI information by applying the one or more PMIs at one or more AI / ML models; and transmitting, to the base station, a CSI report comprising the encoded PMI information and the updated RI.

2. The method of claim 1, further comprising receiving, from the base station, configuration information defining the rank adaptation prerequisite.

3. The method of claim 1, wherein the rank adaptation prerequisite comprises meeting a minimum RI use threshold, and wherein the identifying that the rank adaptation prerequisite is met comprises identifying that a number of prior CSI reports based on the previous RI meets the minimum RI use threshold.

4. The method of claim 3, further comprising: sending, to the base station, a recommended minimum RI use threshold, and receiving, from a network, the minimum RI use threshold, wherein the minimum RI use threshold is equal to the recommended minimum RI use threshold.

5. The method of claim 1, wherein the rank adaptation prerequisite comprises that the CSI report occurs according to a rank adaptation period defined by a rank adaptation periodicity, and wherein the identifying that the rank adaptation prerequisite is met comprises identifying that the CSI report occurs according to the rank adaptation period.

6. The method of claim 1, wherein the one or more AI / ML models comprises a first AI / ML model that applies the one or more PMIs and layer-specific state information foreach of one or more layers indicated by the updated RI to a first state-based encoder of the AI / ML model to generate the encoded PMI information.

7. The method of claim 1, wherein the one or more AI / ML models comprises a first AI / ML model that applies the one or more PMIs and layer-common state information for one or more layers indicated by the updated RI to a first state-based encoder of the AI / ML model to generate the encoded PMI information.

8. The method of claim 1, wherein: the one or more PMIs comprises a plurality of PMIs; the one or more AI / ML models comprises a plurality of AI / ML models; a first AI / ML model of the plurality of AI / ML models applies a first PMI of the plurality of PMIs and first layer-specific state information for a first layer indicated by the updated RI to a first state-based encoder of the first AI / ML model to generate a first portion of the encoded PMI information; and a second AI / ML model of the plurality of AI / ML models applies a second PMI of the plurality of PMIs and second layer-specific state information for a second layer indicated by the updated RI to a second state-based encoder of the second AI / ML model to generate a second portion of the encoded PMI information.

9. The method of claim 1, wherein the one or more AI / ML models comprises a first AI / ML model using a first state-based encoder that generates at least a portion of the encoded PMI information by taking default state information as layer-specific state information.

10. A method of a user equipment (UE) for (AI) / machine learning (ML) model-based channel state information (CSI) reporting, comprising: receiving, from a base station, a rank restriction; generating, based on an identification of a maximum possible rank indicator (RI) under the rank restriction, using a measurement of a channel state information reference signal (CSLRS) transmitted by the base station, one or more precoder matrix indicators (PMIs) corresponding to the maximum possible RI; generating encoded PMI information by applying the one or more PMIs at one or more AI / ML models; andtransmitting, to the base station, a CSI report comprising the encoded PMI information and the maximum possible RI.

11. The method of claim 10, wherein the one or more AI / ML models comprises a first AI / ML model that applies the one or more PMIs and layer-specific state information for each of one or more layers indicated by the maximum possible RI to a first state-based encoder of the AI / ML model to generate the encoded PMI information.

12. The method of claim 10, wherein the one or more AI / ML models comprises a first AI / ML model that applies the one or more PMIs and layer-common state information for one or more layers indicated by the maximum possible RI to a first state-based encoder of the AI / ML model to generate the encoded PMI information.

13. The method of claim 10, wherein the rank restriction is received from the base station in one of radio resource control (RRC) signaling; a medium access control control element (MAC CE); and a downlink control information (DCI).

14. The method of claim 10, wherein: the one or more PMIs comprises a plurality of PMIs; the one or more AI / ML models comprises a plurality of AI / ML models; a first AI / ML model of the plurality of AI / ML models applies a first PMI of the plurality of PMIs and first layer-specific state information for a first layer indicated by the maximum possible RI to a first state-based encoder of the first AI / ML model to generate a first portion of the encoded PMI information; and a second AI / ML model of the plurality of AI / ML models applies a second PMI of the plurality of PMIs and second layer-specific state information for a second layer indicated by the maximum possible RI to a second state-based encoder of the second AI / ML model to generate a second portion of the encoded PMI information.

15. The method of claim 10, wherein the one or more AI / ML models comprises a first AI / ML model using a first state-based encoder that generates at least a portion of the encoded PMI information by taking default state information as layer-specific state information.

16. A method of a user equipment (UE) for (AI) / machine learning (ML) model-based channel state information (CSI) reporting, comprising: generating, using a measurement of a channel state information reference signal (CSI-RS) transmitted by a base station, a plurality of precoder matrix indicators (PMIs) and a corresponding updated RI that is different than a previous RI indicated to the base station; generating encoded PMI information by applying the plurality of PMIs at a plurality of AI / ML models, wherein the plurality of AI / ML models comprises: a first AI / ML model that applies a first PMI of the plurality of PMIs and first layer-specific state information for a first layer indicated by the updated RI to a first state-based encoder of the first AI / ML model to generate a first portion of the encoded PMI information, wherein the first layer-specific state information comprises evaluated state information previously evaluated at the first state-based encoder; and a second AI / ML model of the plurality of AI / ML models applies a second PMI of the plurality of PMIs and second layer-specific state information for a second layer indicated by the updated RI to a second state-based encoder of the second AI / ML model to generate a second portion of the encoded PMI information, wherein the second layerspecific state information comprises default state information; and transmitting, to the base station, a CSI report comprising the encoded PMI information and the updated RI.

17. A method of a base station for (AI) / machine learning (ML) model-based channel state information (CSI) reporting, comprising: sending, to a user equipment (UE), configuration information defining a rank adaptation prerequisite; receiving, from the UE, a CSI report comprising encoded PMI information and a rank indicator (RI) consistent with the rank adaptation prerequisite; applying the encoded PMI information at one or more AI / ML models to generate one or more PMIs; and sending a transmission to the UE, wherein the transmission is of a rank indicated by the RI and is precoded based on the one or more PMIs.

18. The method of claim 17, wherein the RI comprises an updated RI that is different than a previous RI indicated to the base station.

19. The method of claim 17, wherein the rank adaptation prerequisite comprises meeting a minimum RI use threshold.

20. The method of claim 19, further comprising receiving, from the UE, a recommended minimum RI use threshold, and wherein the configuration information defines the minimum RI use threshold as equal to the recommended minimum RI use threshold.

21. The method of claim 17, wherein the rank adaptation prerequisite comprises that the CSI report occurs during a rank adaptation period defined by a rank adaptation periodicity.

22. The method of claim 17, wherein the one or more AI / ML models comprises a first AI / ML model that applies the encoded PMI information and layer-specific state information for each of one or more layers indicated by the RI to a first state-based decoder of the AI / ML model to generate the one or more PMIs.

23. The method of claim 17, wherein the one or more AI / ML models comprises a first AI / ML model that applies the encoded PMI information and layer-common state information for one or more layers indicated by the RI to a first state-based decoder of the first AI / ML model to generate the one or more PMIs.

24. The method of claim 17, wherein: the one or more PMIs comprises a plurality of PMIs; the one or more AI / ML models comprises a plurality of AI / ML models; a first AI / ML model of the plurality of AI / ML models applies a first portion of the encoded PMI information and first layer-specific state information for a first layer indicated by the RI to a first state-based decoder of the first AI / ML model to generate a first PMI of the plurality of PMIs; and a second AI / ML model of the plurality of AI / ML models applies a second portion of the encoded PMI information and second layer-specific state information for a second layer indicated by the RI to a second state-based decoder of the second AI / ML model to generate a second PMI of the plurality of PMIs.

25. The method of claim 17, wherein the one or more AI / ML models comprises a first AI / ML model using a first state-based decoder that generates at least a first PMI of the one or more PMIs by taking default state information as layer-specific state information.

26. An apparatus comprising means to perform the method of any of claim 1 to claim 25.

27. A computer-readable media comprising instructions to cause an electronic device, upon execution of the instructions by one or more processors of the electronic device, to perform the method of any of claim 1 to claim 25.

28. An apparatus comprising logic, modules, or circuitry to perform the method of any of claim 1 to claim 25.

29. A baseband processor for a user equipment (UE) that is configured to cause the UE to perform one or more elements of any of claim 1 to claim 16.

30. A baseband processor for a base station that is configured to cause the base station to perform one or more elements of any of claim 17 to claim 25.

Images