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 .
Introduction
The claims 1-15 and 18-22 are pending in this application. This is a non-final office action in response to Application Number 18/858,874 filed on 22 October 2024. The instant application is a 371 of PCT/EP2023/061233 filed on 28 April 2023 and also claims foreign priority to Greek application 20220100356 filed on 29 April 2022.
A preliminary amendment was also filed on 22 October 2024 in which the abstract is amended, the specification is amended, claims 1-15 are amended, claims 16-17 are canceled, and claims 18-22 are added.
The applicant of record is Telefonaktiebolaget LM Ericsson (publ) in Stockholm, Sweden and the inventors of record are Roy Timo, Konstantinos Vandikas, and Henrik Rydén. The application papers are signed by a U.S. registered representative.
Priority
Receipt is acknowledged of certified copies of papers required by 37 CFR 1.55.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 7 November 2024 was filed after the filing date of the instant application on 22 October 2024 and before the mailing date of the first office action on the merits. The submission is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Claim Interpretation
The claims have been considered according to the latest Patent Eligibility Guidelines and are considered eligible.
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 1-15 and 18-22 are rejected under 35 U.S.C. 103 as being unpatentable over Yoo et al. (U.S. Patent Publication 2021/0266763) in view of Timo et al. (WO 2020/180221 A1).
Regarding claim 1, Yoo disclosed a method, performed by a first node comprising an Auto Encoder, AE, encoder, for training the AE-encoder to provide encoded Channel State Information, CSI, the method (see Yoo [0080]: UE includes auto-encoder; Fig. 6, [0081]: auto-encoder includes an encoder having a convolutional layer and a fully connected layer | Fig. 8, [0093]: a method for federated channel state information (CSI) learning; Fig. 9, [0099]: an example of federated CSI learning from a base station’s perspective | [0097]: UE transmits updated decoder coefficients and updated encoder coefficients to base station; [0098]: UE receives updated CSI decoder and updated CSI encoder from base station for further training) comprising:
providing first AE-encoder data to a second node comprising a first NN-based AE-decoder and having access to channel data representing a communications channel between a first communications node and a second communications node, the first AE-encoder data including first encoder output data computed with the AE-encoder based on the channel data (see Yoo Fig. 5, [0059]: learning via a multi-layer neural network | Fig. 8, [0093]: a method for federated channel state information (CSI) learning from UE’s perspective (see Yoo [0027]); [0095]: UE trains CSI decoder and CSI encoder base on observed channel and interference conditions to obtain updated decoder coefficients and updated encoder coefficients; [0097]: UE transmits updated decoder coefficients and updated encoder coefficients to base station | [0098]: UE receives updated CSI decoder and updated CSI encoder from base station; Fig. 9, [0099]: an example of federated CSI learning from a base station’s perspective);
providing second AE-encoder data to a third node (see Yoo Fig. 8, [0093]: federated CSI learning; examiner notes that federated learning involves learning from multiple devices, i.e. the providing and receiving steps would be repeated for each pair of devices involved in the federated learning) comprising a second NN- based AE-decoder and having access to the channel data, the second AE-encoder data including second encoder output data computed with the AE-encoder based on the same channel data (see Yoo Fig. 5, [0059]: learning via a multi-layer neural network | Fig. 8, [0093]: a method for federated channel state information (CSI) learning from UE’s perspective; [0095]: UE trains CSI decoder and CSI encoder base on observed channel and interference conditions to obtain updated decoder coefficients and updated encoder coefficients; [0097]: UE transmits updated decoder coefficients and updated encoder coefficients to base station | [0098]: UE receives updated CSI decoder and updated CSI encoder from base station; Fig. 9, [0099]: an example of federated CSI learning from a base station’s perspective);
receiving, from the second node, first training assistance information (see Yoo Fig. 8, [0093]: a method for federated channel state information (CSI) learning | [0098]: UE receives updated CSI decoder and updated CSI encoder from base station);
receiving, from the third node (see Yoo Fig. 8, [0093]: federated CSI learning; examiner notes that federated learning involves learning from multiple devices, i.e. the providing and receiving steps would be repeated for each pair of devices involved in the federated learning), second training assistance information (see Yoo Fig. 8, [0093]: a method for federated channel state information (CSI) learning | [0098]: UE receives updated CSI decoder and updated CSI encoder from base station); and
determining, based on the first and second training assistance information, whether or not to continue the training by updating encoder parameters of the AE-encoder based on the received first and second training assistance information, one or both of the first and second training assistance information comprising a gradient vector of a loss function of the respective first and second AE (see Yoo-Timo combination below).
With respect to the limitation “determining, based on the first and second training assistance information, whether or not to continue the training by updating encoder parameters of the AE-encoder based on the received first and second training assistance, one or both of the first and second training assistance information comprising a gradient vector of a loss function of the respective first and second AE”:
Yoo disclosed continuously training the NNs during measurement report and also in response to an identified threshold performance increase associated with a training version of the NN (see Yoo [0100]: “In some examples, the UE may continuously train the NNs during measurement reporting. In some such examples, the UE may maintain a current operating version of the NNs and a training version of the NNs...In yet other examples, the UE may identify a threshold performance increase associated with switching from a current operating version of the NNs to a training version of the NNs, triggering the UE to feedback information about the training version of the decoder NN to the base station and update the current operating version to the training version of the NNs.”).
Yoo also disclosed determining a gradient error vector for weights used in the learning algorithm and continuing to adjust the weights until the error rate has reached a target level (see Yoo [0066]: “To adjust the weights, a learning algorithm may compute a gradient vector for the weights. The gradient may indicate an amount that an error would increase or decrease if the weight were adjusted. At the top layer, the gradient may correspond directly to the value of a weight connecting an activated neuron in the penultimate layer and a neuron in the output layer. In lower layers, the gradient may depend on the value of the weights and on the computed error gradients of the higher layers. The weights may then be adjusted to reduce the error. This manner of adjusting the weights may be referred to as “back propagation” as it involves a “backward pass” through the neural network.”; [0067]: “In practice, the error gradient of weights may be calculated over a small number of examples, so that the calculated gradient approximates the true error gradient. This approximation method may be referred to as stochastic gradient descent. Stochastic gradient descent may be repeated until the achievable error rate of the entire system has stopped decreasing or until the error rate has reached a target level…”).
Yoo did not explicitly disclose that the encoder’s performance is evaluated based on a loss function and that the loss function is sent to the first node.
However in a related art of determining channel estimates, Timo’s Fig. 11 disclosed transmitting parameters from the network node to the terminal device in #1101 and #1103, transmitting new compressed downlink channel estimates to the network node in #1102 and that the parameters transmitted in #1103 indicate an objective function, e.g., loss function, for evaluating the performance of the compression function (see Timo Fig. 11, 26:8-27:9). Additional details about the cost function tuple and the parameters used are described in section 6.2.2 (see Timo 30:31-33:13). Encoding/decoding parameters are updated to minimize error (see Timo 36:5-28) and CSI-RS measurements are normalized to ensure the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses (see Timo 40:10-30).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Yoo and Timo to further clarify how Yoo’s performance is evaluated during training. Including Timo’s teachings would improve customizability, optimize data resources, and ensure that the encoder/decoders are configured suitably based on the intended use (see Timo 32:18-23), while also ensuring the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses (see Timo 40:10-30).
Regarding claim 2, Yoo-Timo disclosed the method according to claim 1, wherein the one or both of the first and second training assistance information further comprises one or more of:
an indication of a loss value of the loss function, an indication of whether or not the AE-encoder has achieved sufficient training performance on the shared channel data when used with the respective AE-decoder such that a pass criterion is fulfilled (see Yoo [0066]: “To adjust the weights, a learning algorithm may compute a gradient vector for the weights. The gradient may indicate an amount that an error would increase or decrease if the weight were adjusted. At the top layer, the gradient may correspond directly to the value of a weight connecting an activated neuron in the penultimate layer and a neuron in the output layer. In lower layers, the gradient may depend on the value of the weights and on the computed error gradients of the higher layers. The weights may then be adjusted to reduce the error. This manner of adjusting the weights may be referred to as “back propagation” as it involves a “backward pass” through the neural network.”; [0067]: “In practice, the error gradient of weights may be calculated over a small number of examples, so that the calculated gradient approximates the true error gradient. This approximation method may be referred to as stochastic gradient descent. Stochastic gradient descent may be repeated until the achievable error rate of the entire system has stopped decreasing or until the error rate has reached a target level…”).
Regarding claim 3, Yoo-Timo disclosed the method according to claim 1, wherein computing with the AE-encoder the first and second encoder output data comprises quantizing the first and second AE-encoder data (see Timo 1:2-3, claim 1: quantizing during compression and decompression of downlink channel estimates).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Yoo and Timo to further clarify how Yoo’s performance is evaluated during training. Including Timo’s teachings would improve customizability, optimize data resources, and ensure that the encoder/decoders are configured suitably based on the intended use (see Timo 32:18-23), while also ensuring the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses (see Timo 40:10-30).
Regarding claim 4, Yoo-Timo disclosed the method according to claim 1, wherein determining whether or not to continue the training comprises determining whether or not a first pass criterion of a first output of a first loss function of the AE is fulfilled based on the received first training assistance information and determining whether or not a second pass criterion of a second output of a second loss function of the AE is fulfilled based on the received second training assistance information (see Yoo [0100]: “In some examples, the UE may continuously train the NNs during measurement reporting. In some such examples, the UE may maintain a current operating version of the NNs and a training version of the NNs...In yet other examples, the UE may identify a threshold performance increase associated with switching from a current operating version of the NNs to a training version of the NNs, triggering the UE to feedback information about the training version of the decoder NN to the base station and update the current operating version to the training version of the NNs.” | [0066]: “To adjust the weights, a learning algorithm may compute a gradient vector for the weights. The gradient may indicate an amount that an error would increase or decrease if the weight were adjusted. At the top layer, the gradient may correspond directly to the value of a weight connecting an activated neuron in the penultimate layer and a neuron in the output layer. In lower layers, the gradient may depend on the value of the weights and on the computed error gradients of the higher layers. The weights may then be adjusted to reduce the error. This manner of adjusting the weights may be referred to as “back propagation” as it involves a “backward pass” through the neural network.”; [0067]: “In practice, the error gradient of weights may be calculated over a small number of examples, so that the calculated gradient approximates the true error gradient. This approximation method may be referred to as stochastic gradient descent. Stochastic gradient descent may be repeated until the achievable error rate of the entire system has stopped decreasing or until the error rate has reached a target level…”).
Regarding claim 5, Yoo-Timo disclosed the method according to claim 1, wherein the first AE-decoder uses the same loss function as the second AE-decoder (see Yoo Fig. 5, [0059]: learning via a multi-layer neural network | [0066]: “To adjust the weights, a learning algorithm may compute a gradient vector for the weights. The gradient may indicate an amount that an error would increase or decrease if the weight were adjusted…The weights may then be adjusted to reduce the error…”; Fig. 7, [0090]: “The base station 710 analyzes the received decoder weights and the encoder weights to extract common parts. That is, the base station 710 aggregates the information received from the UEs 720a-c and derives a new model, as seen in block 750. For example, the weights of common layers received from the various UEs 720a-c can be averaged…The base station 710 pushes the updates (e.g., new model) to the UEs 720a-c to improve their learning of the encoder and decoder coefficients.” | Fig. 8, [0093]: a method for federated channel state information (CSI) learning from UE’s perspective; [0095]: UE trains CSI decoder and CSI encoder base on observed channel and interference conditions to obtain updated decoder coefficients and updated encoder coefficients; [0097]: UE transmits updated decoder coefficients and updated encoder coefficients to base station; [0098]: UE receives updated CSI decoder and updated CSI encoder from base station; Fig. 9, [0099]: an example of federated CSI learning from a base station’s perspective).
Regarding claim 6, Yoo-Timo disclosed the method according to claim 1, further comprising receiving, from the second node an indication of any one or more of:
a loss function used by the AE-decoder (see Timo Fig. 11, 26:8-27:9: transmitting parameters from the network node to the terminal device in #1101 and #1103, transmitting new compressed downlink channel estimates to the network node in #1102 and that the parameters transmitted in #1103 indicate an objective function, e.g., loss function, for evaluating the performance of the compression function | 30:31-33:13: Additional details about the cost function tuple and the parameters used are described in section 6.2.2. | 36:5-28: Encoding and decoding parameters are updated to minimize error; 40:10-30: CSI-RS measurements are normalized to ensure the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses);
use of a same loss function as the third node;
an expected minimum performance of a combination of the AE-encoder and the AE-decoder;
a margin for an adjustment of a lambda value in a regularize; and
meta data about decoder architecture of the decoder.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Yoo and Timo to further clarify how Yoo’s performance is evaluated during training. Including Timo’s teachings would improve customizability, optimize data resources, and ensure that the encoder/decoders are configured suitably based on the intended use (see Timo 32:18-23), while also ensuring the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses (see Timo 40:10-30).
Regarding claim 7, Yoo-Timo disclosed the method according to claim 3, wherein the AE-encoder comprises multiple layers and wherein computing the first and second encoder output data comprises splitting a single encoder output from a last layer of the multiple layers into the first and second encoder output data (see Yoo Fig. 5, [0059]: learning via a multi-layer neural network | Fig. 8, [0093]: a method for federated channel state information (CSI) learning from UE’s perspective; [0095]: UE trains CSI decoder and CSI encoder base on observed channel and interference conditions to obtain updated decoder coefficients and updated encoder coefficients; [0097]: UE transmits updated decoder coefficients and updated encoder coefficients to base station | [0098]: UE receives updated CSI decoder and updated CSI encoder from base station; Fig. 9, [0099]: an example of federated CSI learning from a base station’s perspective), and then quantizing the first and second encoder output data (see Timo 1:2-3, claim 1: quantizing during compression and decompression of downlink channel estimates | Fig. 11, 26:8-27:9: transmitting parameters from the network node to the terminal device in #1101 and #1103, transmitting new compressed downlink channel estimates to the network node in #1102 and that the parameters transmitted in #1103 indicate an objective function, e.g., loss function, for evaluating the performance of the compression function | 30:31-33:13: Additional details about the cost function tuple and the parameters used are described in section 6.2.2. | 36:5-28: Encoding and decoding parameters are updated to minimize error; 40:10-30: CSI-RS measurements are normalized to ensure the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Yoo and Timo to further clarify how Yoo’s performance is evaluated during training. Including Timo’s teachings would improve customizability, optimize data resources, and ensure that the encoder/decoders are configured suitably based on the intended use (see Timo 32:18-23), while also ensuring the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses (see Timo 40:10-30).
Regarding claim 8, Yoo-Timo disclosed the method according to claim 1, further comprising computing, with the AE-encoder, the first and second encoder output data based on a same set of input channel data representing the communications channel between the first communications node and the second communications node (see Yoo Fig. 8, [0093]: a method for federated channel state information (CSI) learning | [0097]: UE transmits updated decoder coefficients and updated encoder coefficients to base station; [0098]: UE receives updated CSI decoder and updated CSI encoder from base station | Fig. 7, [0090]: “The base station 710 analyzes the received decoder weights and the encoder weights to extract common parts. That is, the base station 710 aggregates the information received from the UEs 720a-c and derives a new model, as seen in block 750. For example, the weights of common layers received from the various UEs 720a-c can be averaged…The base station 710 pushes the updates (e.g., new model) to the UEs 720a-c to improve their learning of the encoder and decoder coefficients.”).
Regarding claim 9, Yoo-Timo disclosed the method according to claim 1, wherein the AE-encoder is trained to provide encoded CSI from the first communications node to the second communications node over the communications channel in the communications network, wherein the CSI is provided in an operational phase of the AE-encoder (see Yoo Fig. 8, [0093]: a method for federated channel state information (CSI) learning | [0097]: UE transmits updated decoder coefficients and updated encoder coefficients to base station; [0098]: UE receives updated CSI decoder and updated CSI encoder from base station).
Regarding claim 10, Yoo-Timo disclosed a method, performed by a second node comprising an Auto Encoder, AE,-decoder, for assisting in training an AE-encoder, comprised in a first node, to provide encoded Channel State Information, CSI, the method comprising:
providing, to the first node, an indication of any one or more of:
a loss function used by the AE-decoder (see Timo Fig. 11, 26:8-27:9: transmitting parameters from the network node to the terminal device in #1101 and #1103, transmitting new compressed downlink channel estimates to the network node in #1102 and that the parameters transmitted in #1103 indicate an objective function, e.g., loss function, for evaluating the performance of the compression function | 30:31-33:13: Additional details about the cost function tuple and the parameters used are described in section 6.2.2. | 36:5-28: Encoding and decoding parameters are updated to minimize error; 40:10-30: CSI-RS measurements are normalized to ensure the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses);
use of a same loss function as a third node comprising a second AE-decoder;
an expected minimum performance of a combination of the AE-encoder and the AE-decoder;
a margin for an adjustment of a lambda value in a regularize; and meta data about decoder architecture of the decoder.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Yoo and Timo to further clarify how Yoo’s performance is evaluated during training. Including Timo’s teachings would improve customizability, optimize data resources, and ensure that the encoder/decoders are configured suitably based on the intended use (see Timo 32:18-23), while also ensuring the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses (see Timo 40:10-30).
Regarding claim 11, Yoo-Timo disclosed the method according to claim 10, further comprising:
normalizing the loss function (see Timo Fig. 11, 26:8-27:9: parameters transmitted in #1103 indicate an objective function, e.g., loss function, for evaluating the performance of the compression function; 36:5-28: Encoding and decoding parameters are updated to minimize error; 40:10-30: CSI-RS measurements are normalized to ensure the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses); and
providing, to the first node, first training assistance information based on the normalized loss (see Timo 1:2-3, claim 1: quantizing during compression and decompression of downlink channel estimates | Fig. 11, 26:8-27:9: transmitting parameters from the network node to the terminal device in #1101 and #1103, transmitting new compressed downlink channel estimates to the network node in #1102 and that the parameters transmitted in #1103 indicate an objective function, e.g., loss function, for evaluating the performance of the compression function | 30:31-33:13: Additional details about the cost function tuple and the parameters used are described in section 6.2.2. | 36:5-28: Encoding and decoding parameters are updated to minimize error; 40:10-30: CSI-RS measurements are normalized to ensure the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Yoo and Timo to further clarify how Yoo’s performance is evaluated during training. Including Timo’s teachings would improve customizability, optimize data resources, and ensure that the encoder/decoders are configured suitably based on the intended use (see Timo 32:18-23), while also ensuring the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses (see Timo 40:10-30).
Regarding claim 12, the claim contains the limitations, substantially as claimed, as described in claim 1 above. Examiner notes that claim 1 describes a method performed by a first node whereas claim 12 describes a first node configured to perform a method. Yoo disclosed, as recited in claim 12: A first node, comprising an Auto Encoder, AE,- encoder, configured for training the AE-encoder to provide encoded Channel State Information, CSI (see Yoo [0080]: UE includes auto-encoder; Fig. 6, [0081]: auto-encoder includes an encoder having a convolutional layer and a fully connected layer | Fig. 8, [0093]: a method for federated channel state information (CSI) learning; Fig. 9, [0099]: an example of federated CSI learning from a base station’s perspective | [0097]: UE transmits updated decoder coefficients and updated encoder coefficients to base station; [0098]: UE receives updated CSI decoder and updated CSI encoder from base station for further training), the first node being further configured to:
provide first AE-encoder data to a second node comprising a first NN-based AE- decoder and having access to channel data representing a communications channel between a first communications node and a second communications node, the first AE-encoder data including first encoder output data computed with the AE-encoder based on the channel data (see Yoo Fig. 5, [0059]: learning via a multi-layer neural network | Fig. 8, [0093]: a method for federated channel state information (CSI) learning from UE’s perspective (see Yoo [0027]); [0095]: UE trains CSI decoder and CSI encoder base on observed channel and interference conditions to obtain updated decoder coefficients and updated encoder coefficients; [0097]: UE transmits updated decoder coefficients and updated encoder coefficients to base station | [0098]: UE receives updated CSI decoder and updated CSI encoder from base station; Fig. 9, [0099]: an example of federated CSI learning from a base station’s perspective);
provide second AE-encoder data to a third node (see Yoo Fig. 8, [0093]: federated CSI learning; examiner notes that federated learning involves learning from multiple devices, i.e. the providing and receiving steps would be repeated for each pair of devices involved in the federated learning) comprising a second NN-based AE-decoder and having access to the channel data, the second AE-encoder data including second encoder output data computed with the AE-encoder based on the same channel data (see Yoo Fig. 5, [0059]: learning via a multi-layer neural network | Fig. 8, [0093]: a method for federated channel state information (CSI) learning from UE’s perspective; [0095]: UE trains CSI decoder and CSI encoder base on observed channel and interference conditions to obtain updated decoder coefficients and updated encoder coefficients; [0097]: UE transmits updated decoder coefficients and updated encoder coefficients to base station | [0098]: UE receives updated CSI decoder and updated CSI encoder from base station; Fig. 9, [0099]: an example of federated CSI learning from a base station’s perspective);
receive, from the second node, first training assistance information (see Yoo Fig. 8, [0093]: a method for federated channel state information (CSI) learning | [0098]: UE receives updated CSI decoder and updated CSI encoder from base station);
receive, from the third node (see Yoo Fig. 8, [0093]: federated CSI learning; examiner notes that federated learning involves learning from multiple devices, i.e. the providing and receiving steps would be repeated for each pair of devices involved in the federated learning), second training assistance information (see Yoo Fig. 8, [0093]: a method for federated channel state information (CSI) learning | [0098]: UE receives updated CSI decoder and updated CSI encoder from base station); and
determine, based on the first and second training assistance information, whether or not to continue the training by updating encoder parameters of the AE-encoder based on the received first and second training assistance information (see Yoo-Timo combination below).
With respect to the limitation “determine, based on the first and second training assistance information, whether or not to continue the training by updating encoder parameters of the AE-encoder based on the received first and second training assistance information”:
Yoo disclosed continuously training the NNs during measurement report and also in response to an identified threshold performance increase associated with a training version of the NN (see Yoo [0100]: “In some examples, the UE may continuously train the NNs during measurement reporting. In some such examples, the UE may maintain a current operating version of the NNs and a training version of the NNs...In yet other examples, the UE may identify a threshold performance increase associated with switching from a current operating version of the NNs to a training version of the NNs, triggering the UE to feedback information about the training version of the decoder NN to the base station and update the current operating version to the training version of the NNs.”).
Yoo also disclosed determining a gradient error vector for weights used in the learning algorithm and continuing to adjust the weights until the error rate has reached a target level (see Yoo [0066]: “To adjust the weights, a learning algorithm may compute a gradient vector for the weights. The gradient may indicate an amount that an error would increase or decrease if the weight were adjusted. At the top layer, the gradient may correspond directly to the value of a weight connecting an activated neuron in the penultimate layer and a neuron in the output layer. In lower layers, the gradient may depend on the value of the weights and on the computed error gradients of the higher layers. The weights may then be adjusted to reduce the error. This manner of adjusting the weights may be referred to as “back propagation” as it involves a “backward pass” through the neural network.”; [0067]: “In practice, the error gradient of weights may be calculated over a small number of examples, so that the calculated gradient approximates the true error gradient. This approximation method may be referred to as stochastic gradient descent. Stochastic gradient descent may be repeated until the achievable error rate of the entire system has stopped decreasing or until the error rate has reached a target level…”).
Yoo did not explicitly disclose that the encoder’s performance is evaluated based on a loss function and that the loss function is sent to the first node.
However in a related art of determining channel estimates, Timo’s Fig. 11 disclosed transmitting parameters from the network node to the terminal device in #1101 and #1103, transmitting new compressed downlink channel estimates to the network node in #1102 and that the parameters transmitted in #1103 indicate an objective function, e.g., loss function, for evaluating the performance of the compression function (see Timo Fig. 11, 26:8-27:9). Additional details about the cost function tuple and the parameters used are described in section 6.2.2 (see Timo 30:31-33:13). Encoding/decoding parameters are updated to minimize error (see Timo 36:5-28) and CSI-RS measurements are normalized to ensure the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses (see Timo 40:10-30).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Yoo and Timo to further clarify how Yoo’s performance is evaluated during training. Including Timo’s teachings would improve customizability, optimize data resources, and ensure that the encoder/decoders are configured suitably based on the intended use (see Timo 32:18-23), while also ensuring the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses (see Timo 40:10-30).
Regarding claim 13, the claim contains the limitations, substantially as claimed, as described in claim 2 above. Yoo-Timo disclosed, as recited in claim 13: The first node according to claim 12, wherein the one or both of the first and second training assistance information further comprises one or more of an indication of a loss value of the loss function, an indication of whether or not the AE-encoder has achieved sufficient training performance on the shared channel data when used with the respective AE-decoder such that a pass criterion is fulfilled (see Yoo [0066]: “To adjust the weights, a learning algorithm may compute a gradient vector for the weights. The gradient may indicate an amount that an error would increase or decrease if the weight were adjusted. At the top layer, the gradient may correspond directly to the value of a weight connecting an activated neuron in the penultimate layer and a neuron in the output layer. In lower layers, the gradient may depend on the value of the weights and on the computed error gradients of the higher layers. The weights may then be adjusted to reduce the error. This manner of adjusting the weights may be referred to as “back propagation” as it involves a “backward pass” through the neural network.”; [0067]: “In practice, the error gradient of weights may be calculated over a small number of examples, so that the calculated gradient approximates the true error gradient. This approximation method may be referred to as stochastic gradient descent. Stochastic gradient descent may be repeated until the achievable error rate of the entire system has stopped decreasing or until the error rate has reached a target level…”).
Regarding claim 14, the claim contains the limitations, substantially as claimed, as described in claim 10 above. Examiner notes that claim 10 describes a method performed by a second node whereas claim 14 describes a second node configured to perform a method. Yoo-Timo disclosed, as recited in claim 14: A second node, comprising an Auto Encoder, AE,- decoder, configured for assisting in training an AE-encoder comprised in a first node, to provide encoded Channel State Information, CSI, the second node being further configured to:
provide to the first node, an indication of any one or more of:
a loss function used by the AE-decoder (see Timo Fig. 11, 26:8-27:9: transmitting parameters from the network node to the terminal device in #1101 and #1103, transmitting new compressed downlink channel estimates to the network node in #1102 and that the parameters transmitted in #1103 indicate an objective function, e.g., loss function, for evaluating the performance of the compression function | 30:31-33:13: Additional details about the cost function tuple and the parameters used are described in section 6.2.2. | 36:5-28: Encoding and decoding parameters are updated to minimize error; 40:10-30: CSI-RS measurements are normalized to ensure the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses);
use of a same loss function as a third node comprising a second AE-decoder;
an expected minimum performance of a combination of the AE-encoder and the AE-decoder;
a margin for an adjustment of a lambda value in a regularize; and
meta data about decoder architecture of the decoder.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Yoo and Timo to further clarify how Yoo’s performance is evaluated during training. Including Timo’s teachings would improve customizability, optimize data resources, and ensure that the encoder/decoders are configured suitably based on the intended use (see Timo 32:18-23), while also ensuring the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses (see Timo 40:10-30).
Regarding claim 15, the claim contains the limitations, substantially as claimed, as described in claim 11 above. Yoo-Timo disclosed, as recited in claim 15: The second node according to claim 14, wherein the second node is further configured to:
normalize the loss function (see Timo Fig. 11, 26:8-27:9: parameters transmitted in #1103 indicate an objective function, e.g., loss function, for evaluating the performance of the compression function; 36:5-28: Encoding and decoding parameters are updated to minimize error; 40:10-30: CSI-RS measurements are normalized to ensure the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses); and
provide, to the first node, first training assistance information based on the normalized loss (see Timo 1:2-3, claim 1: quantizing during compression and decompression of downlink channel estimates | Fig. 11, 26:8-27:9: transmitting parameters from the network node to the terminal device in #1101 and #1103, transmitting new compressed downlink channel estimates to the network node in #1102 and that the parameters transmitted in #1103 indicate an objective function, e.g., loss function, for evaluating the performance of the compression function | 30:31-33:13: Additional details about the cost function tuple and the parameters used are described in section 6.2.2. | 36:5-28: Encoding and decoding parameters are updated to minimize error; 40:10-30: CSI-RS measurements are normalized to ensure the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Yoo and Timo to further clarify how Yoo’s performance is evaluated during training. Including Timo’s teachings would improve customizability, optimize data resources, and ensure that the encoder/decoders are configured suitably based on the intended use (see Timo 32:18-23), while also ensuring the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses (see Timo 40:10-30).
Regarding claim 18, the claim contains the limitations, substantially as claimed, as described in claim 3 above. Yoo-Timo disclosed, as recited in claim 18: The method according to claim 2, wherein computing with the AE-encoder the first and second encoder output data comprises quantizing the first and second AE-encoder data (see Timo 1:2-3, claim 1: quantizing during compression and decompression of downlink channel estimates).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Yoo and Timo to further clarify how Yoo’s performance is evaluated during training. Including Timo’s teachings would improve customizability, optimize data resources, and ensure that the encoder/decoders are configured suitably based on the intended use (see Timo 32:18-23), while also ensuring the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses (see Timo 40:10-30).
Regarding claim 19, the claim contains the limitations, substantially as claimed, as described in claim 4 above. Yoo-Timo disclosed, as recited in claim 19: The method according to claim 2, wherein determining whether or not to continue the training comprises determining whether or not a first pass criterion of a first output of a first loss function of the AE is fulfilled based on the received first training assistance information and determining whether or not a second pass criterion of a second output of a second loss function of the AE is fulfilled based on the received second training assistance information (see Yoo [0100]: “In some examples, the UE may continuously train the NNs during measurement reporting. In some such examples, the UE may maintain a current operating version of the NNs and a training version of the NNs...In yet other examples, the UE may identify a threshold performance increase associated with switching from a current operating version of the NNs to a training version of the NNs, triggering the UE to feedback information about the training version of the decoder NN to the base station and update the current operating version to the training version of the NNs.” | [0066]: “To adjust the weights, a learning algorithm may compute a gradient vector for the weights. The gradient may indicate an amount that an error would increase or decrease if the weight were adjusted. At the top layer, the gradient may correspond directly to the value of a weight connecting an activated neuron in the penultimate layer and a neuron in the output layer. In lower layers, the gradient may depend on the value of the weights and on the computed error gradients of the higher layers. The weights may then be adjusted to reduce the error. This manner of adjusting the weights may be referred to as “back propagation” as it involves a “backward pass” through the neural network.”; [0067]: “In practice, the error gradient of weights may be calculated over a small number of examples, so that the calculated gradient approximates the true error gradient. This approximation method may be referred to as stochastic gradient descent. Stochastic gradient descent may be repeated until the achievable error rate of the entire system has stopped decreasing or until the error rate has reached a target level…”).
Regarding claim 20, the claim contains the limitations, substantially as claimed, as described in claim 5 above. Yoo-Timo disclosed, as recited in claim 20: The method according to claim 2, wherein the first AE-decoder uses the same loss function as the second AE-decoder (see Yoo Fig. 5, [0059]: learning via a multi-layer neural network | [0066]: “To adjust the weights, a learning algorithm may compute a gradient vector for the weights. The gradient may indicate an amount that an error would increase or decrease if the weight were adjusted…The weights may then be adjusted to reduce the error…”; Fig. 7, [0090]: “The base station 710 analyzes the received decoder weights and the encoder weights to extract common parts. That is, the base station 710 aggregates the information received from the UEs 720a-c and derives a new model, as seen in block 750. For example, the weights of common layers received from the various UEs 720a-c can be averaged…The base station 710 pushes the updates (e.g., new model) to the UEs 720a-c to improve their learning of the encoder and decoder coefficients.” | Fig. 8, [0093]: a method for federated channel state information (CSI) learning from UE’s perspective; [0095]: UE trains CSI decoder and CSI encoder base on observed channel and interference conditions to obtain updated decoder coefficients and updated encoder coefficients; [0097]: UE transmits updated decoder coefficients and updated encoder coefficients to base station; [0098]: UE receives updated CSI decoder and updated CSI encoder from base station; Fig. 9, [0099]: an example of federated CSI learning from a base station’s perspective).
Regarding claim 21, the claim contains the limitations, substantially as claimed, as described in claim 6 above. Yoo-Timo disclosed, as recited in claim 21: The method according to claim 2, further comprising receiving, from the second node an indication of any one or more of:
a loss function used by the AE-decoder (see Timo Fig. 11, 26:8-27:9: transmitting parameters from the network node to the terminal device in #1101 and #1103, transmitting new compressed downlink channel estimates to the network node in #1102 and that the parameters transmitted in #1103 indicate an objective function, e.g., loss function, for evaluating the performance of the compression function | 30:31-33:13: Additional details about the cost function tuple and the parameters used are described in section 6.2.2. | 36:5-28: Encoding and decoding parameters are updated to minimize error; 40:10-30: CSI-RS measurements are normalized to ensure the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses);
use of a same loss function as the third node;
an expected minimum performance of a combination of the AE-encoder and the AE- decoder;
a margin for an adjustment of a lambda value in a regularize; and
meta data about decoder architecture of the decoder.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Yoo and Timo to further clarify how Yoo’s performance is evaluated during training. Including Timo’s teachings would improve customizability, optimize data resources, and ensure that the encoder/decoders are configured suitably based on the intended use (see Timo 32:18-23), while also ensuring the trained auto-encoder-based CSC is able to be used for several different UEs with different pathlosses (see Timo 40:10-30).
Regarding claim 22, the claim contains the limitations, substantially as claimed, as described in claim 8 above. Yoo-Timo disclosed, as recited in claim 22: The method according to claim 2, further comprising computing, with the AE-encoder, the first and second encoder output data based on a same set of input channel data representing the communications channel between the first communications node and the second communications node (see Yoo Fig. 8, [0093]: a method for federated channel state information (CSI) learning | [0097]: UE transmits updated decoder coefficients and updated encoder coefficients to base station; [0098]: UE receives updated CSI decoder and updated CSI encoder from base station | Fig. 7, [0090]: “The base station 710 analyzes the received decoder weights and the encoder weights to extract common parts. That is, the base station 710 aggregates the information received from the UEs 720a-c and derives a new model, as seen in block 750. For example, the weights of common layers received from the various UEs 720a-c can be averaged…The base station 710 pushes the updates (e.g., new model) to the UEs 720a-c to improve their learning of the encoder and decoder coefficients.”).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Angela Widhalm de Rodriguez whose telephone number is (571)272-1035. The examiner can normally be reached M-F: 6am-2:30pm 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, Nicholas Taylor can be reached at (571)272-3889. 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.
/ANGELA WIDHALM DE RODRIGUEZ/Examiner, Art Unit 2443
Prosecution Insights