DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This office action is in responsive to RCE filed on 05. Claims 1-8 remain pending in the application. Claims 1-4 are independent.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-8 are rejected under 35 U.S.C. 103 as being unpatentable over JE et al. (US 2023/0014613 A1, filed on 11/25/2020), hereinafter JE in view of LEE (US 2025/0055739 A1, filed on 12/22/2021), hereinafter LEE and KUMAR et al. (US 2022/0377844 A1, filed on 05/18/2021), hereinafter KUMAR.
Independent Claim 1
JE discloses a terminal device (JE, ¶¶ [0017], [0050], and [0068] with 120 in FIG. 1: user equipment (UE) or mobile station (MS)) comprising:
a reception unit (JE, ¶¶ [0437]-[0442] with 1801 in FIG. 18: the communication unit 1801 performs functions for transmitting and receiving a signal through a radio channel; some of the communication unit 1801 may be referred to as a "receiver" or a "transceiver") that receives a signal for instructing start of training (JE, ¶¶ [0389]-[00391], [0050], and [0071] with 1620 in FIG. 16A: Learning for AI-Based Handover (HO); FIG. 16A illustrates an example of operations of the BS and the UE for learning; in operation 1620, the UE may receive feedback configuration information from the base station (BS); the feedback configuration information indicates a return of learning data through a feedback request value within handover indication information (HOinformationMessage); the UE may acquire learning data according to the feedback configuration information and feedback the same to the BS; ¶¶ [0409] and [0017] with 1660 in FIG. 16A: in operation 1660, the BS may learn a neural network (NN) model for the AI-based handover on the basis of the learning data acquired from the UE; ¶ [0411] with 1670 in FIG. 16B: in operation 1670, the UE may receive feedback configuration information from the BS; ¶ [0413] with 16130 in FIG. 16B: when receiving a message including the learning data from the target BS, the source BS may learn the NN and the W for the AI HO in operation 16130; i.e., UE receives a instructing signal from BS to collect training data for learning/training NN model); and
a transmission unit (JE, ¶¶ [0437]-[0442] with 1801 in FIG. 18: the communication unit 1801 performs functions for transmitting a signal through a radio channel; some of the communication unit 1801 may be referred to as a "transmitter" or a "transceiver") that transmits training data (JE, ¶¶ [0391] and [0408] with 1650 in FIG. 16A: the UE may acquire learning data according to the feedback configuration information and feedback the same to the BS; In operation 1650, the UE may feedback training data for the AI-based handover to the serving BS according to a format configured by the feedback configuration information; ¶ [0413] with 16100 and 16110 in FIG. 16B: the handover may occur while learning data is collected like in operation 1690, and thus the handover to the target BS is made; in operation 16110, the UE may transmit collected learning data to the target BS which is which is the current serving cell, not to the source BS in this case; when receiving the corresponding message, the target BS may identify an ID of the source BS (or source cell) within the message; since the ID is different from the ID of the target BS, the target BS may feedback learning data to another BS; in operation 16120, the target BS may transmit the learning data to the source BS),
wherein the signal for instructing the start of the training includes a training start timing and a training period (JE, ¶¶ [0391]-[0405] with 1620 in FIG. 16A: the UE may receive the following information in order to record the learning data on the basis of the feedback configuration information: (a) Neural Network (NN) input info; (b) NN output info; (c) Performance; (d) training data period: NN learning data collection period; (e) training data start time: NN learning data collection start time point; and (f) training data end time: NN learning data collection end time point), and
in a case where the signal for instructing the start of the training is received, the training data is generated based on the training start timing and the training period (JE, ¶¶ [0406]-[0407] with 1630 and 1640 in FIG. 16A: in operation 1630, the UE may collect training data on the basis of feedback configuration information; the UE records input info of the NN and output info of the NN in every training data period from a training data start time to a training data end time, and the output info of the NN configures the right answer based on performance; in operation 1640, the UE may detect expiration of the learning section; the BS may learn and update the structure information (NN) and the weight information (W) of the neural network model for the AI-based handover; ¶¶ [0412]-[0413] with 1680 and 16100 in FIG.16B: in operation 1680, the UE may acquire learning data on the basis of the feedback configuration information; the UE may detect expiration of the learning section in operation 16100; ¶ [0202]: a training data period means a period of recording of NN learning data (input value/output value); a training data start time means an NN learning data collection start time point and a training data end time means an NN learning data collection end time point; the UE records an input value and an output value measured in units of training data periods between the training data start time and the training data end time; the feedback format is an example of transmitting information for learning the NN, and may mean a set of data for learning the NN through the input value/output value),
wherein the training data is generated based on a reference signal transmitted from a base station apparatus (JE, ¶¶ [0164]-[0165]: a channel quality may be acquired by measuring a received signal; hereinafter, as metric indicating the channel quality, reference signal received power (RSRP) is described as an example, but beam reference signal received power (BRSRP), reference signal received quality (RSRQ), received signal strength indicator (RSSI), signal to noise ratio (SNR), signal to interference and noise ratio (SINR), carrier to interference and noise ratio (CINR), error vector magnitude (EVM), bit error rate (BER), block error rate (BLER), other terms having the technical meaning equivalent thereto, or indexes indicating a channel quality may be additionally used; ; ¶¶ [0170] and [0179]-[0201]: the UE receives handover-related reference information from the BS; the BS may transmit information related to the handover (hereinafter, referred to as handover information) to the UE in operation 620; the handover information may include NN input info: input value information of HO NN, e.g., wireless signal intensity (RSRP, RSRQ, SINR, or the like); a feedback request is a value of which the BS informs in order to allow the UE to transmit learning information related to the handover for learning; the feedback format indicates a transmission format in which the UE feeds back handover learning data; the feedback format indicates which information is fed back for AI learning; NN input info is input value information of the handover NN used for the handover; e.g., the radio signal intensity such as RSRP, RSRQ, or SINR may be used as the input value; the input value may include the radio signal intensity and the throughput; when maximization of an RSRP value for a predetermined time period is aimed, the UE may add RSRP values of each BS for a predetermined time period after the handover and generate learning data to output a maximum BS(cell); ¶¶ [0391]-[0405]: the UE may receive the following information in order to record the learning data on the basis of the feedback configuration information: NN input info: input value information of the HO NN, e.g., wireless signal intensity (RSRP, RSRQ, SINR, or the like), throughput, an SRS, and a wireless signal which can be measured by the UE such as cqi and the like), and the training data is usable by the terminal device for handover by artificial intelligence and/or machine learning (JE, ¶¶ [0153]-[0154] and [0162]-[0164]: the AI-based handover procedure may include a scheme for transmitting the configured neural network model to the UE, a procedure for making a request for a handover to a target cell identified according to the configured model to a serving cell, or a learning procedure for updating the configured neural network mode; as a better quality of data is accumulated and construction of a neural network model which is a determination reference is more consistent with reality, a handover determination based on an AI algorithm may be more accurate for a situation that the UE or the BS individually faces; the neural network for the AI-based handover may require a plurality of input values; e.g., the neural network for the AI-based handover may consider a channel quality of the current serving cell, a channel quality of the target cell, and a type of the serving cell (e.g., whether the serving cell is a small cell or an RAT type) as an input; further, the neural network for the AI-based handover may provide a plurality of output values; e.g. the neural network for the AI-based handover may indicate a plurality of target cells to which the handover can be performed; an efficient handover may be achieved by applying an AI-based handover determination method to a wireless communication system between the BS and the UE through a combination of cell information and measurement information or environment information of the UE),
.
JE fails to explicitly disclose (1) wherein the training data is usable by the terminal device for channel estimation by artificial intelligence and/or machine learning; and (2) wherein the training start timing is determined based on at least one of: (i) a slot offset from a slot in which downlink control information (DCI) transmitted via a physical downlink control channel (PDCCH) is received: (ii) a timing when a radio resource control (RRC) connection is established; or (iii) a value of a trigger counter.
LEE teaches a system and a method relating to machine learning in a wireless communication (LEE, ¶¶ [0002] and [0006]), wherein the training data is usable by the terminal device for channel estimation by artificial intelligence and/or machine learning (LEE, ¶¶ [0100]-[0106]: the 6G system will support AI for full automation; advance in machine learning will create a more intelligent network for real-time communication in 6G; when AI is introduced to communication, real-time data transmission may be simplified and improve; AI may determine a method of performing complicated target tasks using countless analysis; i.e., AI may increase efficiency and reduce processing delay; time-consuming tasks such as handover, network selection or resource scheduling may be immediately performed by using AI; deep learning have been focused on the wireless resource management and allocation field; combine deep learning in the physical layer with wireless transmission are emerging; AI-based physical layer transmission means applying a signal processing and communication mechanism based on an AI driver rather than a traditional communication framework in a fundamental signal processing and communication mechanism; e.g., channel coding and decoding based on deep learning, signal estimation and detection based on deep learning, multiple input multiple output (MIMO) mechanisms based on deep learning, resource scheduling and allocation based on AI, etc. may be included; machine learning may be used for channel estimation and channel tracking and may be used for power allocation, interference cancellation, etc. in the physical layer of DL; in addition, machine learning may be used for antenna selection, power control, symbol detection, etc. in the MIMO system; deep learning-based AI algorithms require a lot of training data in order to optimize training parameters; however, due to limitations in acquiring data in a specific channel environment as training data, a lot of training data is used offline; static training for training data in a specific channel environment may cause a contradiction between the diversity and dynamic characteristics of a radio channel; ¶¶ [0134]-[0144] with FIGS. 15A-C: it is not easy to collect a lot of data for training the neural network in a communication channel with a long-tail distribution; even if a terminal moves to experience a new channel, collects data and performs fine-tuning training for an offline-pretrained neural network again, the terminal can hardly reflect a new change sufficiently; neural network training should reflect effect of a channel between a transmitter and a receiver; a reference signal or a pilot signal may be used as training data to reflect a channel effect; e.g., a device may be subject to transmission and reception training for a channel by receiving data based on a demodulation reference signal (DM-RS) for reception; a neural network requirement technique is needed to enable a device to secure as many reference signal wireless resources as possible and to have fast learning in a propagation environment with a long-tail distribution; proposes a communication system, a communication procedure, and a signaling method, which are based on online meta learning; meta learning uses a neural network that is pretrained based on various tasks; meta learning is learning that uses such a pretrained neural network to enable a device to well perform inference like regression and classification for a new task; i.e., meta learning is a method of enhancing learning and estimation performance for a new task and is learning for learning (learn-to-learn); when a device experiences a new task, the device relearns a model parameter ϕ suitable for the new task from the meta parameter θ and performs inference based on the relearning; ¶¶ [0145]-[0149] with FIG. 16: proposes a meta learning method for a transmitting/receiving task associated with a reference signal; there are various reference signals; e.g., (a) a device may transmit a phase-tracking reference signal with a specific pattern in order to correct a phase error in a high-frequency band; (b) the device may transmit a demodulation reference signal (DM-RS) with a specific pattern in order to estimate a channel and to estimate data; (c) the device may transmit a channel state information-reference signal (CSI-RS) with a specific pattern in order to track time synchronization of a channel, to track a beam, and to find out channel quality information like a rank indicator (RI), a precoding matrix indicator (PMI), and a channel quality indicator (CQI); (d) the device may transmit a sounding reference signal (SRS) with a specific pattern p for channel sounding; and (e) the terminal may transmit a positioning reference signal (PRS) for measuring a location; the device may transmit various types of reference signals according to purposes; define transmission/reception and measurement of reference signals as tasks so that a device may use a wireless resource as efficiently as possible; applying meta learning to transmission/reception and measurement of reference signals; the task may mean transmission/reception and measurement related to all reference signals that have passed a channel at time t, wherein the reference signals may include a CSI-RS, a PTRS, a PRS, a CRS, and a DMRS; in each task, a result of passing a channel may be used as a dataset; e.g., a terminal may perform a task of measuring a CQI based on a CSI-RS, wherein in case the terminal succeeds in CQI measurement based on the CSI-RS, a result may be used as one supervised learning dataset; online meta learning proposed may be divided mainly into two parts as follows: (a) meta-training is a procedure in which a device learns an optimal meta parameter θ based on a meta-training data set of every reference signal; and (b) adaptation is a procedure in which the device learns ϕ by learning based on the optimal meta parameter θ* to succeed in a task for a specific reference signal and performs a task related to the reference signal at the same time; in case the device succeeds in adaptation of the reference signal task, the device may obtain a reference signal dataset and a parameter ϕ for the task; when the device acquires θ*, meta training may be performed by inputting the reference signal dataset and the task parameter obtained from the adaptation; in addition, the device may acquire ϕ by performing a specific task based on θ* obtained from the meta training, thereby performing adaptation).
JE and LEE are analogous art because they are from the same field of endeavor, a system and a method relating to machine learning in a wireless communication. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to apply the teaching of LEE to JE. Motivation for doing so would easily expand AI techniques in a new task for increasing .
JE in view of LEE fails to explicitly disclose wherein the training start timing is determined based on at least one of: (i) a slot offset from a slot in which downlink control information (DCI) transmitted via a physical downlink control channel (PDCCH) is received: (ii) a timing when a radio resource control (RRC) connection is established; or (iii) a value of a trigger counter.
KUMAR teaches a system and a method relating to machine learning in communication systems (KUMAR, ¶ [0001]), wherein the training start timing is determined based on at least one of: (i) a slot offset from a slot in which downlink control information (DCI) transmitted via a physical downlink control channel (PDCCH) is received: (ii) a timing when a radio resource control (RRC) connection is established; or (iii) a value of a trigger counter (KUMAR, ¶ [0089] with FIG. 6: FIG. 6 is a call flow diagram 600 for OAM-initiated ML model training activation; OAM-initiated model training may be performed based on available data to the UE 610 or other NE in an RRC connected state; ¶ [0096] with FIG.8: FIG. 8 is a call flow diagram 800 for RAN-based ML controller-initiated ML model training activation; RAN-based ML controller-initiated model training may be performed in an RRC connected state; ¶¶ [0169] and [0178]: receive a trigger to activate an ML model training based on at least one of an indication from an ML model repository or a protocol of the network entity; and transmit an ML model training request to activate the ML model training at one or more nodes; the trigger includes a resumption of an RRC connected state for a UE, and the ML model training request includes a reinitialization notification; ¶¶ [0092] and [0098]: an event-based trigger condition may include a number of trainings performed on the model) (NOTE: for the training start timing is determined based on (iii) a value of a trigger counter, see Reference Li listed in the Conclusion section).
JE in view of LEE, and KUMAR are analogous art because they are from the same field of endeavor, a system and a method relating to machine learning in communication systems. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to apply the teaching of KUMAR to JE in view of LEE. Motivation for doing so would provide further improvements in 5G NR technology as well as other multi-access technologies and the telecommunication standards (LEE, ¶ [0003]).
Independent Claim 2
JE discloses a base station apparatus (JE, ¶¶ [0016], [0050], and [0067] with 110 in FIG. 1: base station (BS)) comprising:
a transmission unit (JE, ¶¶ [0423]-[0428] with 1701 in FIG. 17: the wireless communication unit 1701 performs functions for transmitting a signal through a radio channel; all or some of the wireless communication unit 1701 may be referred to as a 'transmitter' or a 'transceiver') that transmits a signal for instructing start of training (JE, ¶¶ [0389]-[00391], [0050], and [0071] with 1620 in FIG. 16A: Learning for AI-Based Handover (HO); FIG. 16A illustrates an example of operations of the BS and the UE for learning; in operation 1620, the UE may receive feedback configuration information from the base station (BS); the feedback configuration information indicates a return of learning data through a feedback request value within handover indication information (HOinformationMessage); the UE may acquire learning data according to the feedback configuration information and feedback the same to the BS; ¶¶ [0409] and [0017] with 1660 in FIG. 16A: in operation 1660, the BS may learn a neural network (NN) model for the AI-based handover on the basis of the learning data acquired from the UE; ¶ [0411] with 1670 in FIG. 16B: in operation 1670, the UE may receive feedback configuration information from the BS; ¶ [0413] with 16130 in FIG. 16B: when receiving a message including the learning data from the target BS, the source BS may learn the NN and the W for the AI HO in operation 16130; i.e., UE receives a instructing signal from BS to collect training data for learning/training NN model); and
a reception unit (JE, ¶¶ [0423]-[0428] with 1701 in FIG. 17: the wireless communication unit 1701 performs functions for receiving a signal through a radio channel; all or some of the wireless communication unit 1701 may be referred to as a 'receiver' or a 'transceiver') that receives training data (JE, ¶¶ [0391] and [0408] with 1650 in FIG. 16A: the UE may acquire learning data according to the feedback configuration information and feedback the same to the BS; In operation 1650, the UE may feedback training data for the AI-based handover to the serving BS according to a format configured by the feedback configuration information; ¶ [0413] with 16100 and 16110 in FIG. 16B: the handover may occur while learning data is collected like in operation 1690, and thus the handover to the target BS is made; in operation 16110, the UE may transmit collected learning data to the target BS which is which is the current serving cell, not to the source BS in this case; when receiving the corresponding message, the target BS may identify an ID of the source BS (or source cell) within the message; since the ID is different from the ID of the target BS, the target BS may feedback learning data to another BS; in operation 16120, the target BS may transmit the learning data to the source BS),
wherein the signal for instructing the start of the training includes a training start timing and a training period (JE, ¶¶ [0391]-[0405] with 1620 in FIG. 16A: the UE may receive the following information in order to record the learning data on the basis of the feedback configuration information: (a) Neural Network (NN) input info; (b) NN output info; (c) Performance; (d) training data period: NN learning data collection period; (e) training data start time: NN learning data collection start time point; and (f) training data end time: NN learning data collection end time point), and
in a case where the signal for instructing the start of the training is transmitted, the training data is instructed to be generated based on the training start timing and the training period (JE, ¶¶ [0406]-[0407] with 1630 and 1640 in FIG. 16A: in operation 1630, the UE may collect training data on the basis of feedback configuration information; the UE records input info of the NN and outinfo of the NN in every training data period from a training data start time to a training data end time, and the output info of the NN configures the right answer based on performance; in operation 1640, the UE may detect expiration of the learning section; the BS may learn and update the structure information (NN) and the weight information (W) of the neural network model for the AI-based handover; ¶¶ [0412]-[0413] with 1680 and 16100 in FIG.16B: in operation 1680, the UE may acquire learning data on the basis of the feedback configuration information; the UE may detect expiration of the learning section in operation 16100; ¶ [0202]: a training data period means a period of recording of NN learning data (input value/output value); a training data start time means an NN learning data collection start time point and a training data end time means an NN learning data collection end time point; the UE records an input value and an output value measured in units of training data periods between the training data start time and the training data end time; the feedback format is an example of transmitting information for learning the NN, and may mean a set of data for learning the NN through the input value/output value),
wherein the training data is generated based on a reference signal transmitted from the base station apparatus (JE, ¶¶ [0164]-[0165]: a channel quality may be acquired by measuring a received signal; hereinafter, as metric indicating the channel quality, reference signal received power (RSRP) is described as an example, but beam reference signal received power (BRSRP), reference signal received quality (RSRQ), received signal strength indicator (RSSI), signal to noise ratio (SNR), signal to interference and noise ratio (SINR), carrier to interference and noise ratio (CINR), error vector magnitude (EVM), bit error rate (BER), block error rate (BLER), other terms having the technical meaning equivalent thereto, or indexes indicating a channel quality may be additionally used; ; ¶¶ [0170] and [0179]-[0201]: the UE receives handover-related reference information from the BS; the BS may transmit information related to the handover (hereinafter, referred to as handover information) to the UE in operation 620; the handover information may include NN input info: input value information of HO NN, e.g., wireless signal intensity (RSRP, RSRQ, SINR, or the like); a feedback request is a value of which the BS informs in order to allow the UE to transmit learning information related to the handover for learning; the feedback format indicates a transmission format in which the UE feeds back handover learning data; the feedback format indicates which information is fed back for AI learning; NN input info is input value information of the handover NN used for the handover; e.g., the radio signal intensity such as RSRP, RSRQ, or SINR may be used as the input value; the input value may include the radio signal intensity and the throughput; when maximization of an RSRP value for a predetermined time period is aimed, the UE may add RSRP values of each BS for a predetermined time period after the handover and generate learning data to output a maximum BS(cell); ¶¶ [0391]-[0405]: the UE may receive the following information in order to record the learning data on the basis of the feedback configuration information: NN input info: input value information of the HO NN, e.g., wireless signal intensity (RSRP, RSRQ, SINR, or the like), throughput, an SRS, and a wireless signal which can be measured by the UE such as cqi and the like), and the training data is usable by the terminal device for handover by artificial intelligence and/or machine learning (JE, ¶¶ [0153]-[0154] and [0162]-[0164]: the AI-based handover procedure may include a scheme for transmitting the configured neural network model to the UE, a procedure for making a request for a handover to a target cell identified according to the configured model to a serving cell, or a learning procedure for updating the configured neural network mode; as a better quality of data is accumulated and construction of a neural network model which is a determination reference is more consistent with reality, a handover determination based on an AI algorithm may be more accurate for a situation that the UE or the BS individually faces; the neural network for the AI-based handover may require a plurality of input values; e.g., the neural network for the AI-based handover may consider a channel quality of the current serving cell, a channel quality of the target cell, and a type of the serving cell (e.g., whether the serving cell is a small cell or an RAT type) as an input; further, the neural network for the AI-based handover may provide a plurality of output values; e.g. the neural network for the AI-based handover may indicate a plurality of target cells to which the handover can be performed; an efficient handover may be achieved by applying an AI-based handover determination method to a wireless communication system between the BS and the UE through a combination of cell information and measurement information or environment information of the UE),
.
JE fails to explicitly disclose (1) wherein the training data is usable by the terminal device for channel estimation by artificial intelligence and/or machine learning; and (2) wherein the training start timing is determined based on at least one of: (i) a slot offset from a slot in which downlink control information (DCI) transmitted via a physical downlink control channel (PDCCH) is transmitted; (ii) a timing when a radio resource control (RRC) connection is established; or (iii) a value of a trigger counter.
LEE teaches a system and a method relating to machine learning in a wireless communication (LEE, ¶¶ [0002] and [0006]), wherein the training data is usable by the terminal device for channel estimation by artificial intelligence and/or machine learning (LEE, ¶¶ [0100]-[0106]: the 6G system will support AI for full automation; advance in machine learning will create a more intelligent network for real-time communication in 6G; when AI is introduced to communication, real-time data transmission may be simplified and improve; AI may determine a method of performing complicated target tasks using countless analysis; i.e., AI may increase efficiency and reduce processing delay; time-consuming tasks such as handover, network selection or resource scheduling may be immediately performed by using AI; deep learning have been focused on the wireless resource management and allocation field; combine deep learning in the physical layer with wireless transmission are emerging; AI-based physical layer transmission means applying a signal processing and communication mechanism based on an AI driver rather than a traditional communication framework in a fundamental signal processing and communication mechanism; e.g., channel coding and decoding based on deep learning, signal estimation and detection based on deep learning, multiple input multiple output (MIMO) mechanisms based on deep learning, resource scheduling and allocation based on AI, etc. may be included; machine learning may be used for channel estimation and channel tracking and may be used for power allocation, interference cancellation, etc. in the physical layer of DL; in addition, machine learning may be used for antenna selection, power control, symbol detection, etc. in the MIMO system; deep learning-based AI algorithms require a lot of training data in order to optimize training parameters; however, due to limitations in acquiring data in a specific channel environment as training data, a lot of training data is used offline; static training for training data in a specific channel environment may cause a contradiction between the diversity and dynamic characteristics of a radio channel; ¶¶ [0134]-[0144] with FIGS. 15A-C: it is not easy to collect a lot of data for training the neural network in a communication channel with a long-tail distribution; even if a terminal moves to experience a new channel, collects data and performs fine-tuning training for an offline-pretrained neural network again, the terminal can hardly reflect a new change sufficiently; neural network training should reflect effect of a channel between a transmitter and a receiver; a reference signal or a pilot signal may be used as training data to reflect a channel effect; e.g., a device may be subject to transmission and reception training for a channel by receiving data based on a demodulation reference signal (DM-RS) for reception; a neural network requirement technique is needed to enable a device to secure as many reference signal wireless resources as possible and to have fast learning in a propagation environment with a long-tail distribution; proposes a communication system, a communication procedure, and a signaling method, which are based on online meta learning; meta learning uses a neural network that is pretrained based on various tasks; meta learning is learning that uses such a pretrained neural network to enable a device to well perform inference like regression and classification for a new task; i.e., meta learning is a method of enhancing learning and estimation performance for a new task and is learning for learning (learn-to-learn); when a device experiences a new task, the device relearns a model parameter ϕ suitable for the new task from the meta parameter θ and performs inference based on the relearning; ¶¶ [0145]-[0149] with FIG. 16: proposes a meta learning method for a transmitting/receiving task associated with a reference signal; there are various reference signals; e.g., (a) a device may transmit a phase-tracking reference signal with a specific pattern in order to correct a phase error in a high-frequency band; (b) the device may transmit a demodulation reference signal (DM-RS) with a specific pattern in order to estimate a channel and to estimate data; (c) the device may transmit a channel state information-reference signal (CSI-RS) with a specific pattern in order to track time synchronization of a channel, to track a beam, and to find out channel quality information like a rank indicator (RI), a precoding matrix indicator (PMI), and a channel quality indicator (CQI); (d) the device may transmit a sounding reference signal (SRS) with a specific pattern p for channel sounding; and (e) the terminal may transmit a positioning reference signal (PRS) for measuring a location; the device may transmit various types of reference signals according to purposes; define transmission/reception and measurement of reference signals as tasks so that a device may use a wireless resource as efficiently as possible; applying meta learning to transmission/reception and measurement of reference signals; the task may mean transmission/reception and measurement related to all reference signals that have passed a channel at time t, wherein the reference signals may include a CSI-RS, a PTRS, a PRS, a CRS, and a DMRS; in each task, a result of passing a channel may be used as a dataset; e.g., a terminal may perform a task of measuring a CQI based on a CSI-RS, wherein in case the terminal succeeds in CQI measurement based on the CSI-RS, a result may be used as one supervised learning dataset; online meta learning proposed may be divided mainly into two parts as follows: (a) meta-training is a procedure in which a device learns an optimal meta parameter θ based on a meta-training data set of every reference signal; and (b) adaptation is a procedure in which the device learns ϕ by learning based on the optimal meta parameter θ* to succeed in a task for a specific reference signal and performs a task related to the reference signal at the same time; in case the device succeeds in adaptation of the reference signal task, the device may obtain a reference signal dataset and a parameter ϕ for the task; when the device acquires θ*, meta training may be performed by inputting the reference signal dataset and the task parameter obtained from the adaptation; in addition, the device may acquire ϕ by performing a specific task based on θ* obtained from the meta training, thereby performing adaptation).
JE and LEE are analogous art because they are from the same field of endeavor, a system and a method relating to machine learning in a wireless communication. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to apply the teaching of LEE to JE. Motivation for doing so would easily expand AI techniques in a new task for increasing efficiency and reducing processing delay (LEE, ¶¶ [0100]-[0103], [0134]-[0137], and [0144]).
JE in view of LEE fails to explicitly disclose wherein the training start timing is determined based on at least one of: (i) a slot offset from a slot in which downlink control information (DCI) transmitted via a physical downlink control channel (PDCCH) is transmitted; (ii) a timing when a radio resource control (RRC) connection is established; or (iii) a value of a trigger counter.
KUMAR teaches a system and a method relating to machine learning in communication systems (KUMAR, ¶ [0001]), wherein the training start timing is determined based on at least one of: (i) a slot offset from a slot in which downlink control information (DCI) transmitted via a physical downlink control channel (PDCCH) is transmitted: (ii) a timing when a radio resource control (RRC) connection is established; or (iii) a value of a trigger counter (KUMAR, ¶ [0089] with FIG. 6: FIG. 6 is a call flow diagram 600 for OAM-initiated ML model training activation; OAM-initiated model training may be performed based on available data to the UE 610 or other NE in an RRC connected state; ¶ [0096] with FIG.8: FIG. 8 is a call flow diagram 800 for RAN-based ML controller-initiated ML model training activation; RAN-based ML controller-initiated model training may be performed in an RRC connected state; ¶¶ [0169] and [0178]: receive a trigger to activate an ML model training based on at least one of an indication from an ML model repository or a protocol of the network entity; and transmit an ML model training request to activate the ML model training at one or more nodes; the trigger includes a resumption of an RRC connected state for a UE, and the ML model training request includes a reinitialization notification; ¶¶ [0092] and [0098]: an event-based trigger condition may include a number of trainings performed on the model) (NOTE: for the training start timing is determined based on (iii) a value of a trigger counter, see Reference Li listed in the Conclusion section).
JE in view of LEE, and KUMAR are analogous art because they are from the same field of endeavor, a system and a method relating to machine learning in communication systems. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to apply the teaching of KUMAR to JE in view of LEE. Motivation for doing so would provide further improvements in 5G NR technology as well as other multi-access technologies and the telecommunication standards (LEE, ¶ [0003]).
Independent Claim 3
JE discloses a communication method used in a terminal device (JE, ¶¶ [0017], [0050], and [0068] with 120 in FIG. 1: a user equipment (UE) or a mobile station (MS) in a wireless communication system),
wherein a first processing unit (JE, ¶¶ [0437] and [0444] with 1805 in FIG. 18: the UE 120 includes a controller 1805; the controller 1805 may include at least one processor or microprocessor or may be a part of the processor) that processes a signal for instructing start of training (JE, ¶¶ [0389]-[00391], [0050], and [0071] with 1620 in FIG. 16A: Learning for AI-Based Handover (HO); FIG. 16A illustrates an example of operations of the BS and the UE for learning; in operation 1620, the UE may receive feedback configuration information from the base station (BS); the feedback configuration information indicates a return of learning data through a feedback request value within handover indication information (HOinformationMessage); the UE may acquire learning data according to the feedback configuration information and feedback the same to the BS; ¶¶ [0409] and [0017] with 1660 in FIG. 16A: in operation 1660, the BS may learn a neural network (NN) model for the AI-based handover on the basis of the learning data acquired from the UE; ¶ [0411] with 1670 in FIG. 16B: in operation 1670, the UE may receive feedback configuration information from the BS; ¶ [0413] with 16130 in FIG. 16B: when receiving a message including the learning data from the target BS, the source BS may learn the NN and the W for the AI HO in operation 16130; i.e., UE receives and processes a instructing signal from BS to collect training data for learning/training NN model) and a second processing unit (JE, ¶¶ [0424] and [0431]-[0433]: the BS 110 includes a controller 1707; the controller 1707 controls the overall operations of the BS 110; ¶¶ [0437] and [0444] with 1805 in FIG. 18: the UE 120 includes a controller 1805; the controller 1805 may include at least one processor or microprocessor or may be a part of the processor) that processes training data are provided (JE, ¶¶ [0391] and [0408]-[0409] with 1650 in FIG. 16A: the UE may acquire learning data according to the feedback configuration information and feedback the same to the BS; in operation 1650, the UE may feedback training data for the AI-based handover to the serving BS according to a format configured by the feedback configuration information; in operation 1660, the BS may learn a neural network model for the AI-based handover on the basis of the learning data acquired from the UE; ¶ [0413] with 16100 and 16110 in FIG. 16B: the handover may occur while learning data is collected like in operation 1690, and thus the handover to the target BS is made; in operation 16110, the UE may transmit collected learning data to the target BS which is which is the current serving cell, not to the source BS in this case; when receiving the corresponding message, the target BS may identify an ID of the source BS (or source cell) within the message; since the ID is different from the ID of the target BS, the target BS may feedback learning data to another BS; in operation 16120, the target BS may transmit the learning data to the source BS),
the signal for instructing the start of the training includes a training start timing and a training period (JE, ¶¶ [0391]-[0405] with 1620 in FIG. 16A: the UE may receive the following information in order to record the learning data on the basis of the feedback configuration information: (a) Neural Network (NN) input info; (b) NN output info; (c) Performance; (d) training data period: NN learning data collection period; (e) training data start time: NN learning data collection start time point; and (f) training data end time: NN learning data collection end time point), and
in a case where the signal for instructing the start of the training is received, the training data is generated based on the training start timing and the training period (JE, ¶¶ [0406]-[0407] with 1630 and 1640 in FIG. 16A: in operation 1630, the UE may collect training data on the basis of feedback configuration information; the UE records input info of the NN and outinfo of the NN in every training data period from a training data start time to a training data end time, and the output info of the NN configures the right answer based on performance; in operation 1640, the UE may detect expiration of the learning section; the BS may learn and update the structure information (NN) and the weight information (W) of the neural network model for the AI-based handover; ¶¶ [0412]-[0413] with 1680 and 16100 in FIG.16B: in operation 1680, the UE may acquire learning data on the basis of the feedback configuration information; the UE may detect expiration of the learning section in operation 16100; ¶ [0202]: a training data period means a period of recording of NN learning data (input value/output value); a training data start time means an NN learning data collection start time point and a training data end time means an NN learning data collection end time point; the UE records an input value and an output value measured in units of training data periods between the training data start time and the training data end time; the feedback format is an example of transmitting information for learning the NN, and may mean a set of data for learning the NN through the input value/output value),
wherein the training data is generated based on a reference signal transmitted from a base station apparatus (JE, ¶¶ [0164]-[0165]: a channel quality may be acquired by measuring a received signal; hereinafter, as metric indicating the channel quality, reference signal received power (RSRP) is described as an example, but beam reference signal received power (BRSRP), reference signal received quality (RSRQ), received signal strength indicator (RSSI), signal to noise ratio (SNR), signal to interference and noise ratio (SINR), carrier to interference and noise ratio (CINR), error vector magnitude (EVM), bit error rate (BER), block error rate (BLER), other terms having the technical meaning equivalent thereto, or indexes indicating a channel quality may be additionally used; ; ¶¶ [0170] and [0179]-[0201]: the UE receives handover-related reference information from the BS; the BS may transmit information related to the handover (hereinafter, referred to as handover information) to the UE in operation 620; the handover information may include NN input info: input value information of HO NN, e.g., wireless signal intensity (RSRP, RSRQ, SINR, or the like); a feedback request is a value of which the BS informs in order to allow the UE to transmit learning information related to the handover for learning; the feedback format indicates a transmission format in which the UE feeds back handover learning data; the feedback format indicates which information is fed back for AI learning; NN input info is input value information of the handover NN used for the handover; e.g., the radio signal intensity such as RSRP, RSRQ, or SINR may be used as the input value; the input value may include the radio signal intensity and the throughput; when maximization of an RSRP value for a predetermined time period is aimed, the UE may add RSRP values of each BS for a predetermined time period after the handover and generate learning data to output a maximum BS(cell); ¶¶ [0391]-[0405]: the UE may receive the following information in order to record the learning data on the basis of the feedback configuration information: NN input info: input value information of the HO NN, e.g., wireless signal intensity (RSRP, RSRQ, SINR, or the like), throughput, an SRS, and a wireless signal which can be measured by the UE such as cqi and the like), and the training data is usable by the terminal device for handover by artificial intelligence and/or machine learning (JE, ¶¶ [0153]-[0154] and [0162]-[0164]: the AI-based handover procedure may include a scheme for transmitting the configured neural network model to the UE, a procedure for making a request for a handover to a target cell identified according to the configured model to a serving cell, or a learning procedure for updating the configured neural network mode; as a better quality of data is accumulated and construction of a neural network model which is a determination reference is more consistent with reality, a handover determination based on an AI algorithm may be more accurate for a situation that the UE or the BS individually faces; the neural network for the AI-based handover may require a plurality of input values; e.g., the neural network for the AI-based handover may consider a channel quality of the current serving cell, a channel quality of the target cell, and a type of the serving cell (e.g., whether the serving cell is a small cell or an RAT type) as an input; further, the neural network for the AI-based handover may provide a plurality of output values; e.g. the neural network for the AI-based handover may indicate a plurality of target cells to which the handover can be performed; an efficient handover may be achieved by applying an AI-based handover determination method to a wireless communication system between the BS and the UE through a combination of cell information and measurement information or environment information of the UE),
.
JE fails to explicitly disclose (1) wherein the training data is usable by the terminal device for channel estimation by artificial intelligence and/or machine learning; and (2) wherein the training start timing is determined based on at least one of: (i) a slot offset from a slot in which downlink control information (DCI) transmitted via a physical downlink control channel (PDCCH) is received: (ii) a timing when a radio resource control (RRC) connection is established; or (iii) a value of a trigger counter.
LEE teaches a system and a method relating to machine learning in a wireless communication (LEE, ¶¶ [0002] and [0006]), wherein the training data is usable by the terminal device for channel estimation by artificial intelligence and/or machine learning (LEE, ¶¶ [0100]-[0106]: the 6G system will support AI for full automation; advance in machine learning will create a more intelligent network for real-time communication in 6G; when AI is introduced to communication, real-time data transmission may be simplified and improve; AI may determine a method of performing complicated target tasks using countless analysis; i.e., AI may increase efficiency and reduce processing delay; time-consuming tasks such as handover, network selection or resource scheduling may be immediately performed by using AI; deep learning have been focused on the wireless resource management and allocation field; combine deep learning in the physical layer with wireless transmission are emerging; AI-based physical layer transmission means applying a signal processing and communication mechanism based on an AI driver rather than a traditional communication framework in a fundamental signal processing and communication mechanism; e.g., channel coding and decoding based on deep learning, signal estimation and detection based on deep learning, multiple input multiple output (MIMO) mechanisms based on deep learning, resource scheduling and allocation based on AI, etc. may be included; machine learning may be used for channel estimation and channel tracking and may be used for power allocation, interference cancellation, etc. in the physical layer of DL; in addition, machine learning may be used for antenna selection, power control, symbol detection, etc. in the MIMO system; deep learning-based AI algorithms require a lot of training data in order to optimize training parameters; however, due to limitations in acquiring data in a specific channel environment as training data, a lot of training data is used offline; static training for training data in a specific channel environment may cause a contradiction between the diversity and dynamic characteristics of a radio channel; ¶¶ [0134]-[0144] with FIGS. 15A-C: it is not easy to collect a lot of data for training the neural network in a communication channel with a long-tail distribution; even if a terminal moves to experience a new channel, collects data and performs fine-tuning training for an offline-pretrained neural network again, the terminal can hardly reflect a new change sufficiently; neural network training should reflect effect of a channel between a transmitter and a receiver; a reference signal or a pilot signal may be used as training data to reflect a channel effect; e.g., a device may be subject to transmission and reception training for a channel by receiving data based on a demodulation reference signal (DM-RS) for reception; a neural network requirement technique is needed to enable a device to secure as many reference signal wireless resources as possible and to have fast learning in a propagation environment with a long-tail distribution; proposes a communication system, a communication procedure, and a signaling method, which are based on online meta learning; meta learning uses a neural network that is pretrained based on various tasks; meta learning is learning that uses such a pretrained neural network to enable a device to well perform inference like regression and classification for a new task; i.e., meta learning is a method of enhancing learning and estimation performance for a new task and is learning for learning (learn-to-learn); when a device experiences a new task, the device relearns a model parameter ϕ suitable for the new task from the meta parameter θ and performs inference based on the relearning; ¶¶ [0145]-[0149] with FIG. 16: proposes a meta learning method for a transmitting/receiving task associated with a reference signal; there are various reference signals; e.g., (a) a device may transmit a phase-tracking reference signal with a specific pattern in order to correct a phase error in a high-frequency band; (b) the device may transmit a demodulation reference signal (DM-RS) with a specific pattern in order to estimate a channel and to estimate data; (c) the device may transmit a channel state information-reference signal (CSI-RS) with a specific pattern in order to track time synchronization of a channel, to track a beam, and to find out channel quality information like a rank indicator (RI), a precoding matrix indicator (PMI), and a channel quality indicator (CQI); (d) the device may transmit a sounding reference signal (SRS) with a specific pattern p for channel sounding; and (e) the terminal may transmit a positioning reference signal (PRS) for measuring a location; the device may transmit various types of reference signals according to purposes; define transmission/reception and measurement of reference signals as tasks so that a device may use a wireless resource as efficiently as possible; applying meta learning to transmission/reception and measurement of reference signals; the task may mean transmission/reception and measurement related to all reference signals that have passed a channel at time t, wherein the reference signals may include a CSI-RS, a PTRS, a PRS, a CRS, and a DMRS; in each task, a result of passing a channel may be used as a dataset; e.g., a terminal may perform a task of measuring a CQI based on a CSI-RS, wherein in case the terminal succeeds in CQI measurement based on the CSI-RS, a result may be used as one supervised learning dataset; online meta learning proposed may be divided mainly into two parts as follows: (a) meta-training is a procedure in which a device learns an optimal meta parameter θ based on a meta-training data set of every reference signal; and (b) adaptation is a procedure in which the device learns ϕ by learning based on the optimal meta parameter θ* to succeed in a task for a specific reference signal and performs a task related to the reference signal at the same time; in case the device succeeds in adaptation of the reference signal task, the device may obtain a reference signal dataset and a parameter ϕ for the task; when the device acquires θ*, meta training may be performed by inputting the reference signal dataset and the task parameter obtained from the adaptation; in addition, the device may acquire ϕ by performing a specific task based on θ* obtained from the meta training, thereby performing adaptation).
JE and LEE are analogous art because they are from the same field of endeavor, a system and a method relating to machine learning in a wireless communication. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to apply the teaching of LEE to JE. Motivation for doing so would easily expand AI techniques in a new task for increasing efficiency and reducing processing delay (LEE, ¶¶ [0100]-[0103], [0134]-[0137], and [0144]).
JE in view of LEE fails to explicitly disclose wherein the training start timing is determined based on at least one of: (i) a slot offset from a slot in which downlink control information (DCI) transmitted via a physical downlink control channel (PDCCH) is received: (ii) a timing when a radio resource control (RRC) connection is established; or (iii) a value of a trigger counter.
KUMAR teaches a system and a method relating to machine learning in communication systems (KUMAR, ¶ [0001]), wherein the training start timing is determined based on at least one of: (i) a slot offset from a slot in which downlink control information (DCI) transmitted via a physical downlink control channel (PDCCH) is received: (ii) a timing when a radio resource control (RRC) connection is established; or (iii) a value of a trigger counter (KUMAR, ¶ [0089] with FIG. 6: FIG. 6 is a call flow diagram 600 for OAM-initiated ML model training activation; OAM-initiated model training may be performed based on available data to the UE 610 or other NE in an RRC connected state; ¶ [0096] with FIG.8: FIG. 8 is a call flow diagram 800 for RAN-based ML controller-initiated ML model training activation; RAN-based ML controller-initiated model training may be performed in an RRC connected state; ¶¶ [0169] and [0178]: receive a trigger to activate an ML model training based on at least one of an indication from an ML model repository or a protocol of the network entity; and transmit an ML model training request to activate the ML model training at one or more nodes; the trigger includes a resumption of an RRC connected state for a UE, and the ML model training request includes a reinitialization notification; ¶¶ [0092] and [0098]: an event-based trigger condition may include a number of trainings performed on the model) (NOTE: for the training start timing is determined based on (iii) a value of a trigger counter, see Reference Li listed in the Conclusion section).
JE in view of LEE, and KUMAR are analogous art because they are from the same field of endeavor, a system and a method relating to machine learning in communication systems. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to apply the teaching of KUMAR to JE in view of LEE. Motivation for doing so would provide further improvements in 5G NR technology as well as other multi-access technologies and the telecommunication standards (LEE, ¶ [0003]).
Independent Claim 4
JE discloses a communication method used in a base station apparatus (JE, ¶¶ [0016], [0050], and [0067] with 110 in FIG. 1: a base station (BS) of a serving cell in a wireless communication system),
wherein a first processing unit (JE, ¶¶ [0437] and [0444] with 1805 in FIG. 18: the UE 120 includes a controller 1805; the controller 1805 may include at least one processor or microprocessor or may be a part of the processor) that processes a signal for instructing start of training (JE, ¶¶ [0389]-[00391], [0050], and [0071] with 1620 in FIG. 16A: Learning for AI-Based Handover (HO); FIG. 16A illustrates an example of operations of the BS and the UE for learning; in operation 1620, the UE may receive feedback configuration information from the base station (BS); the feedback configuration information indicates a return of learning data through a feedback request value within handover indication information (HOinformationMessage); the UE may acquire learning data according to the feedback configuration information and feedback the same to the BS; ¶¶ [0409] and [0017] with 1660 in FIG. 16A: in operation 1660, the BS may learn a neural network (NN) model for the AI-based handover on the basis of the learning data acquired from the UE; ¶ [0411] with 1670 in FIG. 16B: in operation 1670, the UE may receive feedback configuration information from the BS; ¶ [0413] with 16130 in FIG. 16B: when receiving a message including the learning data from the target BS, the source BS may learn the NN and the W for the AI HO in operation 16130; i.e., UE receives and processes a instructing signal from BS to collect training data for learning/training NN model) and a second processing unit (JE, ¶¶ [0424] and [0431]-[0433]: the BS 110 includes a controller 1707; the controller 1707 controls the overall operations of the BS 110; ¶¶ [0437] and [0444] with 1805 in FIG. 18: the UE 120 includes a controller 1805; the controller 1805 may include at least one processor or microprocessor or may be a part of the processor) that processes training data are provided (JE, ¶¶ [0391] and [0408]-[0409] with 1650 in FIG. 16A: the UE may acquire learning data according to the feedback configuration information and feedback the same to the BS; In operation 1650, the UE may feedback training data for the AI-based handover to the serving BS according to a format configured by the feedback configuration information; in operation 1660, the BS may learn a neural network model for the AI-based handover on the basis of the learning data acquired from the UE; ¶ [0413] with 16100 and 16110 in FIG. 16B: the handover may occur while learning data is collected like in operation 1690, and thus the handover to the target BS is made; in operation 16110, the UE may transmit collected learning data to the target BS which is which is the current serving cell, not to the source BS in this case; when receiving the corresponding message, the target BS may identify an ID of the source BS (or source cell) within the message; since the ID is different from the ID of the target BS, the target BS may feedback learning data to another BS; in operation 16120, the target BS may transmit the learning data to the source BS),
the signal for instructing the start of the training includes a training start timing and a training period (JE, ¶¶ [0391]-[0405] with 1620 in FIG. 16A: the UE may receive the following information in order to record the learning data on the basis of the feedback configuration information: (a) Neural Network (NN) input info; (b) NN output info; (c) Performance; (d) training data period: NN learning data collection period; (e) training data start time: NN learning data collection start time point; and (f) training data end time: NN learning data collection end time point), and
in a case where the signal for instructing the start of the training is transmitted, the training data is instructed to be generated based on the training start timing and the training period (JE, ¶¶ [0406]-[0407] with 1630 and 1640 in FIG. 16A: in operation 1630, the UE may collect training data on the basis of feedback configuration information; the UE records input info of the NN and outinfo of the NN in every training data period from a training data start time to a training data end time, and the output info of the NN configures the right answer based on performance; in operation 1640, the UE may detect expiration of the learning section; the BS may learn and update the structure information (NN) and the weight information (W) of the neural network model for the AI-based handover; ¶¶ [0412]-[0413] with 1680 and 16100 in FIG.16B: in operation 1680, the UE may acquire learning data on the basis of the feedback configuration information; the UE may detect expiration of the learning section in operation 16100; ¶ [0202]: a training data period means a period of recording of NN learning data (input value/output value); a training data start time means an NN learning data collection start time point and a training data end time means an NN learning data collection end time point; the UE records an input value and an output value measured in units of training data periods between the training data start time and the training data end time; the feedback format is an example of transmitting information for learning the NN, and may mean a set of data for learning the NN through the input value/output value),
wherein the training data is generated based on a reference signal transmitted from a base station apparatus (JE, ¶¶ [0164]-[0165]: a channel quality may be acquired by measuring a received signal; hereinafter, as metric indicating the channel quality, reference signal received power (RSRP) is described as an example, but beam reference signal received power (BRSRP), reference signal received quality (RSRQ), received signal strength indicator (RSSI), signal to noise ratio (SNR), signal to interference and noise ratio (SINR), carrier to interference and noise ratio (CINR), error vector magnitude (EVM), bit error rate (BER), block error rate (BLER), other terms having the technical meaning equivalent thereto, or indexes indicating a channel quality may be additionally used; ; ¶¶ [0170] and [0179]-[0201]: the UE receives handover-related reference information from the BS; the BS may transmit information related to the handover (hereinafter, referred to as handover information) to the UE in operation 620; the handover information may include NN input info: input value information of HO NN, e.g., wireless signal intensity (RSRP, RSRQ, SINR, or the like); a feedback request is a value of which the BS informs in order to allow the UE to transmit learning information related to the handover for learning; the feedback format indicates a transmission format in which the UE feeds back handover learning data; the feedback format indicates which information is fed back for AI learning; NN input info is input value information of the handover NN used for the handover; e.g., the radio signal intensity such as RSRP, RSRQ, or SINR may be used as the input value; the input value may include the radio signal intensity and the throughput; when maximization of an RSRP value for a predetermined time period is aimed, the UE may add RSRP values of each BS for a predetermined time period after the handover and generate learning data to output a maximum BS(cell); ¶¶ [0391]-[0405]: the UE may receive the following information in order to record the learning data on the basis of the feedback configuration information: NN input info: input value information of the HO NN, e.g., wireless signal intensity (RSRP, RSRQ, SINR, or the like), throughput, an SRS, and a wireless signal which can be measured by the UE such as cqi and the like), and the training data is usable by the terminal device for handover by artificial intelligence and/or machine learning (JE, ¶¶ [0153]-[0154] and [0162]-[0164]: the AI-based handover procedure may include a scheme for transmitting the configured neural network model to the UE, a procedure for making a request for a handover to a target cell identified according to the configured model to a serving cell, or a learning procedure for updating the configured neural network mode; as a better quality of data is accumulated and construction of a neural network model which is a determination reference is more consistent with reality, a handover determination based on an AI algorithm may be more accurate for a situation that the UE or the BS individually faces; the neural network for the AI-based handover may require a plurality of input values; e.g., the neural network for the AI-based handover may consider a channel quality of the current serving cell, a channel quality of the target cell, and a type of the serving cell (e.g., whether the serving cell is a small cell or an RAT type) as an input; further, the neural network for the AI-based handover may provide a plurality of output values; e.g. the neural network for the AI-based handover may indicate a plurality of target cells to which the handover can be performed; an efficient handover may be achieved by applying an AI-based handover determination method to a wireless communication system between the BS and the UE through a combination of cell information and measurement information or environment information of the UE),
.
JE fails to explicitly disclose (1) wherein the training data is usable by the terminal device for channel estimation by artificial intelligence and/or machine learning; and (2) wherein the training start timing is determined based on at least one of: (i) a slot offset from a slot in which downlink control information (DCI) transmitted via a physical downlink control channel (PDCCH) is transmitted; (ii) a timing when a radio resource control (RRC) connection is established; or (iii) a value of a trigger counter.
LEE teaches a system and a method relating to machine learning in a wireless communication (LEE, ¶¶ [0002] and [0006]), wherein the training data is usable by the terminal device for channel estimation by artificial intelligence and/or machine learning (LEE, ¶¶ [0100]-[0106]: the 6G system will support AI for full automation; advance in machine learning will create a more intelligent network for real-time communication in 6G; when AI is introduced to communication, real-time data transmission may be simplified and improve; AI may determine a method of performing complicated target tasks using countless analysis; i.e., AI may increase efficiency and reduce processing delay; time-consuming tasks such as handover, network selection or resource scheduling may be immediately performed by using AI; deep learning have been focused on the wireless resource management and allocation field; combine deep learning in the physical layer with wireless transmission are emerging; AI-based physical layer transmission means applying a signal processing and communication mechanism based on an AI driver rather than a traditional communication framework in a fundamental signal processing and communication mechanism; e.g., channel coding and decoding based on deep learning, signal estimation and detection based on deep learning, multiple input multiple output (MIMO) mechanisms based on deep learning, resource scheduling and allocation based on AI, etc. may be included; machine learning may be used for channel estimation and channel tracking and may be used for power allocation, interference cancellation, etc. in the physical layer of DL; in addition, machine learning may be used for antenna selection, power control, symbol detection, etc. in the MIMO system; deep learning-based AI algorithms require a lot of training data in order to optimize training parameters; however, due to limitations in acquiring data in a specific channel environment as training data, a lot of training data is used offline; static training for training data in a specific channel environment may cause a contradiction between the diversity and dynamic characteristics of a radio channel; ¶¶ [0134]-[0144] with FIGS. 15A-C: it is not easy to collect a lot of data for training the neural network in a communication channel with a long-tail distribution; even if a terminal moves to experience a new channel, collects data and performs fine-tuning training for an offline-pretrained neural network again, the terminal can hardly reflect a new change sufficiently; neural network training should reflect effect of a channel between a transmitter and a receiver; a reference signal or a pilot signal may be used as training data to reflect a channel effect; e.g., a device may be subject to transmission and reception training for a channel by receiving data based on a demodulation reference signal (DM-RS) for reception; a neural network requirement technique is needed to enable a device to secure as many reference signal wireless resources as possible and to have fast learning in a propagation environment with a long-tail distribution; proposes a communication system, a communication procedure, and a signaling method, which are based on online meta learning; meta learning uses a neural network that is pretrained based on various tasks; meta learning is learning that uses such a pretrained neural network to enable a device to well perform inference like regression and classification for a new task; i.e., meta learning is a method of enhancing learning and estimation performance for a new task and is learning for learning (learn-to-learn); when a device experiences a new task, the device relearns a model parameter ϕ suitable for the new task from the meta parameter θ and performs inference based on the relearning; ¶¶ [0145]-[0149] with FIG. 16: proposes a meta learning method for a transmitting/receiving task associated with a reference signal; there are various reference signals; e.g., (a) a device may transmit a phase-tracking reference signal with a specific pattern in order to correct a phase error in a high-frequency band; (b) the device may transmit a demodulation reference signal (DM-RS) with a specific pattern in order to estimate a channel and to estimate data; (c) the device may transmit a channel state information-reference signal (CSI-RS) with a specific pattern in order to track time synchronization of a channel, to track a beam, and to find out channel quality information like a rank indicator (RI), a precoding matrix indicator (PMI), and a channel quality indicator (CQI); (d) the device may transmit a sounding reference signal (SRS) with a specific pattern p for channel sounding; and (e) the terminal may transmit a positioning reference signal (PRS) for measuring a location; the device may transmit various types of reference signals according to purposes; define transmission/reception and measurement of reference signals as tasks so that a device may use a wireless resource as efficiently as possible; applying meta learning to transmission/reception and measurement of reference signals; the task may mean transmission/reception and measurement related to all reference signals that have passed a channel at time t, wherein the reference signals may include a CSI-RS, a PTRS, a PRS, a CRS, and a DMRS; in each task, a result of passing a channel may be used as a dataset; e.g., a terminal may perform a task of measuring a CQI based on a CSI-RS, wherein in case the terminal succeeds in CQI measurement based on the CSI-RS, a result may be used as one supervised learning dataset; online meta learning proposed may be divided mainly into two parts as follows: (a) meta-training is a procedure in which a device learns an optimal meta parameter θ based on a meta-training data set of every reference signal; and (b) adaptation is a procedure in which the device learns ϕ by learning based on the optimal meta parameter θ* to succeed in a task for a specific reference signal and performs a task related to the reference signal at the same time; in case the device succeeds in adaptation of the reference signal task, the device may obtain a reference signal dataset and a parameter ϕ for the task; when the device acquires θ*, meta training may be performed by inputting the reference signal dataset and the task parameter obtained from the adaptation; in addition, the device may acquire ϕ by performing a specific task based on θ* obtained from the meta training, thereby performing adaptation).
JE and LEE are analogous art because they are from the same field of endeavor, a system and a method relating to machine learning in a wireless communication. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to apply the teaching of LEE to JE. Motivation for doing so would easily expand AI techniques in a new task for increasing efficiency and reducing processing delay (LEE, ¶¶ [0100]-[0103], [0134]-[0137], and [0144]).
JE in view of LEE fails to explicitly disclose wherein the training start timing is determined based on at least one of: (i) a slot offset from a slot in which downlink control information (DCI) transmitted via a physical downlink control channel (PDCCH) is transmitted; (ii) a timing when a radio resource control (RRC) connection is established; or (iii) a value of a trigger counter.
KUMAR teaches a system and a method relating to machine learning in communication systems (KUMAR, ¶ [0001]), wherein the training start timing is determined based on at least one of: (i) a slot offset from a slot in which downlink control information (DCI) transmitted via a physical downlink control channel (PDCCH) is transmitted: (ii) a timing when a radio resource control (RRC) connection is established; or (iii) a value of a trigger counter (KUMAR, ¶ [0089] with FIG. 6: FIG. 6 is a call flow diagram 600 for OAM-initiated ML model training activation; OAM-initiated model training may be performed based on available data to the UE 610 or other NE in an RRC connected state; ¶ [0096] with FIG.8: FIG. 8 is a call flow diagram 800 for RAN-based ML controller-initiated ML model training activation; RAN-based ML controller-initiated model training may be performed in an RRC connected state; ¶¶ [0169] and [0178]: receive a trigger to activate an ML model training based on at least one of an indication from an ML model repository or a protocol of the network entity; and transmit an ML model training request to activate the ML model training at one or more nodes; the trigger includes a resumption of an RRC connected state for a UE, and the ML model training request includes a reinitialization notification; ¶¶ [0092] and [0098]: an event-based trigger condition may include a number of trainings performed on the model) (NOTE: for the training start timing is determined based on (iii) a value of a trigger counter, see Reference Li listed in the Conclusion section).
JE in view of LEE, and KUMAR are analogous art because they are from the same field of endeavor, a system and a method relating to machine learning in communication systems. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to apply the teaching of KUMAR to JE in view of LEE. Motivation for doing so would provide further improvements in 5G NR technology as well as other multi-access technologies and the telecommunication standards (LEE, ¶ [0003]).
Claims 5-8
JE in view of LEE and KUMAR discloses all the elements as stated in Claims 1-4 respectively and further discloses wherein the reference signal includes any of a demodulation reference signal (DMRS), a phase tracking reference signal (PTRS), a channel state information reference signal (CSI-RS), or synchronization signal block (SSB) (LEE, ¶ [0136]: a reference signal or a pilot signal may be used as training data to reflect a channel effect; e.g., a device may be subject to transmission and reception training for a channel by receiving data based on a demodulation reference signal (DM-RS) for reception; a neural network requirement technique is needed to enable a device to secure as many reference signal wireless resources as possible and to have fast learning in a propagation environment with a long-tail distribution; ¶¶ [0145]-[0149] with FIG. 16: proposes a meta learning method for a transmitting/receiving task associated with a reference signal; there are various reference signals; e.g., (a) a device may transmit a phase-tracking reference signal with a specific pattern in order to correct a phase error in a high-frequency band; (b) the device may transmit a demodulation reference signal (DM-RS) with a specific pattern in order to estimate a channel and to estimate data; (c) the device may transmit a channel state information-reference signal (CSI-RS) with a specific pattern in order to track time synchronization of a channel, to track a beam, and to find out channel quality information like a rank indicator (RI), a precoding matrix indicator (PMI), and a channel quality indicator (CQI); (d) the device may transmit a sounding reference signal (SRS) with a specific pattern p for channel sounding; and (e) the terminal may transmit a positioning reference signal (PRS) for measuring a location; the device may transmit various types of reference signals according to purposes; define transmission/reception and measurement of reference signals as tasks so that a device may use a wireless resource as efficiently as possible; applying meta learning to transmission/reception and measurement of reference signals; the task may mean transmission/reception and measurement related to all reference signals that have passed a channel at time t, wherein the reference signals may include a CSI-RS, a PTRS, a PRS, a CRS, and a DMRS; in each task, a result of passing a channel may be used as a dataset; e.g., a terminal may perform a task of measuring a CQI based on a CSI-RS, wherein in case the terminal succeeds in CQI measurement based on the CSI-RS, a result may be used as one supervised learning dataset; ¶ [0153]: a preferred RS set is a set consisting of every possible combination of RSs for meta learning; e.g., the preferred RS set may include at least one of a DMRS, a PTRS, a CSI-RS, a PRS, a synchronization signal block (SSB), an SRS, a cell-specific reference signal (CRS), and a meta learning RS).
Response to Arguments
Applicant’s arguments filed on with respect to Claims 1-4 have been fully considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in some of arguments. Some of arguments are not persuasive.
Applicant argues on Pages 5-6 of the Remarks that JE is directed to AI-based handover, not channel estimation, and LEE does not cure this deficiency.
In response, examiner respectfully disagrees. LEE teaches AI-based channel estimation in ¶¶ [0103] and [0136] that (1) "Machine learning may be used for channel estimation and channel tracking …, etc."; (2) "Neural network training should reflect effect of a channel between a transmitter and a receiver"; and (3) "A reference signal or a pilot signal may be used as training data to reflect a channel effect". LEE also discloses in ¶ [0101] that "Time-consuming tasks such as handover, network selection or resource scheduling may be immediately performed by using AI" (i.e., AI-based handover taught by JE). Therefore, LEE DOES cure this deficiency of JE.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Li et al. (US 2025/0227497 A1, priority date on 03/29/2022) discloses in in ¶¶ [0139]-[0172] that (1) detecting a failure or performance problem include: a failure counter; a failure timer; failure counter and failure timer; recovery counter; failure timer b; failure timer b and failure counter; failure timer b, failure counter and recovery counter; (2) the UE is configured with a failure counter maximum value (failure_max_count) so that the UE detects a failure in the ML model if the number of ML model problems reaches the failure counter maximum value; (3) the UE monitors each instance that a ML model problem occurs and increments the failure counter, and if the failure counter reaches the failure counter maximum value the ML model failure is detected; (4) the UE is configured with a failure timer value, wherein the UE detects the failure of the ML model if the failure occurs for that time value; (5) the UE does perform a recovery action if a failure or problem in the ML model occurs in sparse time instances i.e., these would preferably occur within a relatively short time; (6) the UE receives a failure timer value and a failure counter maximum value; (7) when the UE detects an ML model problem (e.g., indication internally at the UE of a MLMP, or any other criterion), the UE starts the failure timer and increments the failure counter; (8) if the number of instances reaches the failure counter maximum value while the failure timer is running, the UE declares a failure in the ML model and performs the one or more recovery actions (various recovery action discussed below; e.g., the UE performs training (or re-training) of the ML model); (9) if the timer expires (i.e., before the number of ML model problem or problem instances (or MLMP indication(s)) reaches the failure counter maximum value), the UE resets the failure counter; (10) the advantage of using the counter and the timer is that the UE does not need to take recovery actions if ML model performance problems are sparse in time and/or there are only a few problems, especially if the recovery action(s) requires a disruption in data transmission/reception, such as the reset of one or more protocol entities and/or signaling exchange with the network; (11) the UE is configured with a recovery counter value so that the UE detects a recovery of the ML model if the number of ML model recoveries reaches the recovery counter value; (12) the UE monitors each instance that a ML model recovery occurs and increments the recovery counter, and if the recovery counter reaches the recovery counter value the ML model recovery is detected; (13) the UE is configured with a failure timer value for a failure timer b, wherein the UE detects the failure of the ML model if the failure timer b expires; (14) based both on a failure timer b and a failure counter; (15) the UE receives a failure timer value b and a failure counter maximum value; (16) when the UE detects an ML model problem (e.g., indication internally at the UE of a MLMP, or any other criterion), the UE increments the failure counter and, if the number of instances reaches the failure counter maximum value the failure timer is started.; (17) based on all three of failure timer b, failure counter and recovery counter; (18) when the UE detects an ML model problem (e.g., indication internally at the UE of a MLMP, or any other criterion), the UE increments the failure counter and, if the number of instances reaches the failure counter maximum value the failure timer b is started; (19) while the failure timer b is running, the UE has an ML model performance recovery indication and increments the recovery counter, and if the recovery counter becomes higher than the recovery counter value, the UE stops the failure timer b, and resets the failure counter; (20) one advantage of using the counter and the timer is that the UE does not need to take recovery actions if ML model performance problems are sparse in time and/or there are only a few problems, especially if the recovery action(s) requires a disruption in data transmission/reception, such as the reset of one or more protocol entities and/or signaling exchange with the network; (20) in response to detecting a failure or performance problem, performing one or more resolution actions; e.g., (a) the UE stops (or suspends) the monitoring of ML model performance; (b) the UE starts (or re-starts) using the classical non-ML-algorithm/function; (c) the UE performs training (or re-training) of the ML model; (22) the UE performing the training comprises the UE initiating or starting to perform the training; (23) the UE performing the training (or re-training) of the ML model comprises the UE obtaining one or more parameters used for the training of the ML model; e.g., if the ML model produces CSI estimates/predictions (based on one or more input) for a sub-band X, the UE starts to measure and collect the CSI for sub-band X, to feed the training/re-training function; (24) the UE performing the training (or re-training) of the ML model comprises the UE starting a timer T and after the timer T expires the UE can re-start the monitoring of the ML model performance; (25) in some situations, the UE starting the training (or re-training) of the ML model comprises the UE collecting a number of measurements (or samples) 'N' for the training/re-training, wherein 'N' is possibly configured by the network or known to the UE via some other means, e.g., retrieved from memory; (26) after the number of measurements (or samples) the UE re-start the monitoring of the ML model performance (i.e., a counter to count a number of measurements (or samples) 'N' collected, and after the counter reached the number of measurements (or samples) required, the UE starting the training (or re-training) of the ML model); (27) the UE starting the training (or re-training) of the ML model comprises the UE performing measurements during a number of measurement periods 'K' for the training/re-training, wherein 'K' is possibly configured by the network or known to the UE via some other means, e.g., retrieved from memory; and (28) after the number of measurements periods the UE re-start the monitoring of the ML model performance.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HWEI-MIN LU whose telephone number is (313)446-4913. The examiner can normally be reached Mon - Fri: 9:00 AM - 6:00 PM EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Mariela D. Reyes can be reached at (571) 270-1006. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/HWEI-MIN LU/Primary Examiner, Art Unit 2142
Prosecution Insights