DETAILED ACTION
Claims 1-5 are pending.
Priority
Receipt is acknowledged of certified copies of papers required by 37 CFR 1.55.
The examiner notes that the foreign priority application (IN202341054172, filed 8/11/2023) is contained within the file wrapper (dated 12/6/2024). The examiner further notes that the foreign priority application is in English. The examiner has reviewed this subject matter and does not find support under 35 USC 112(a) for the current claims. See MPEP 216. In particular, the examiner fails to find support for the channel Ht and reconstructed channel Ĥt within the context of the claimed functions. Therefore, for the purposes of applying prior art, the examiner considers the effective filing date of the claims to be 8/9/2024. If the applicant disagrees with this conclusion, they are invited to provide support for 112(a) on the record.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 7/29/2025 was filed. The submission is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Drawings
The drawings were received on 8/9/2024. These drawings are accepted.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-4 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
As per claim 1, the claim uses the symbol Ht for both “a first channel” (see line 3) and “a compressed channel” (see line 7). Therefore, each of these terms are indefinite, as they rely on the same symbol, but were introduced as the same term.
Claims 2-3 are rejected as being dependent upon a rejected parent claim.
As per claim 4, the claim uses the symbol Ht for both “a first channel” (see line 6) and “a compressed channel” (see line 10). Therefore, each of these terms are indefinite, as they rely on the same symbol, but were introduced as the same term.
Claim Objections
Claim 1 is objected to because of the following informalities: “a second CQIt+1” (see line 12) and “second CQI (CQIt+1)” (see line 14) are two mismatched wordings for the same term. The chosen language should also be used in dependent claim 2. Appropriate correction is required.
Claim 2 is objected to because of the following informalities: the phrasing for first CQIt should be harmonized with claim 1. Appropriate correction is required.
Claim 4 is objected to because of the following informalities: line 10 recites “the CQIt” which should be amended to “the first CQIt”. Appropriate correction is required.
Claim 4 is objected to because of the following informalities: “a second CQIt+1” (see line 15) and “second CQI (CQIt+1)” (see line 16) are two mismatched wordings for the same term. The chosen language should also be used in dependent claim 2. Appropriate correction is required.
Claim Rejections - 35 USC § 103
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, 3, and 4 are rejected under 35 U.S.C. 103 as being unpatentable over Tan et al. (WO 2024/173366) in view of Panasonic (R1-2307004, NPL W on pg. 2 of PTO-892).
The examiner notes that the cited portions of TAN find support in Provisional Application No. 63/445,594 (filed 2/14/2023), see fig. 9 and its associated disclosure. In addition, Panasonic was uploaded to the public 3GPP FTP server on 8/10/2023. See MPEP 2128.
As per claim 1, Tan et al. a method, comprising:
receiving, by a UE (104), a first reference signal in a first time slot (t) [Tan, ¶ 0140, “The base station determines and sends 502 configuration information and CSI-RS(s) to the WTRU”, The WTRU (or UE) receives a reference signal (CSI-RS) during a first time period (t). See also fig. 5, which is a signalling diagram for a basic AI/ML CSI feedback framework (see ¶s 0020, 0021, 0105, and 0106).];
estimating, by the UE (104), a first channel Ht and a first Channel Quality Indicator (CQIt) at a first time instance [Tan, ¶ 0140, “After receiving the traditional CSI-RS(s), the WTRU may estimate 505 CSI for channel H and determines a set of initial RI/CQI for a set of precoders according to configuration by the NW. The WTRU may compute RI/CQI based on the configuration of precoders”, The WTRU estimates channel H (see also fig. 5, step 505) and computes CQI of the CSI-RS for the time t based on the estimated precoders.];
compressing, by the UE (104), the first channel (Ht) [Tan, ¶ 0140, “In method 500, the WTRU sends a compressed channel matrix (H) and the determined set of RI/CQI as part of CSI feedback for the base station to reconstruct 506 the precoders, for example in line with the WTRU configuration”, The channel is compressed for transmission to the base station.]…
transmitting, by the UE (104), a compressed channel (Ht) along with the first CQIt to a Base Station (BS) (102) [Tan, ¶ 0140, “In method 500, the WTRU sends a compressed channel matrix (H) and the determined set of RI/CQI as part of CSI feedback for the base station to reconstruct 506 the precoders, for example in line with the WTRU configuration”, CSI feedback may include the CQI and compressed channel matrix (H). See also ¶ 0141.];
receiving, by the UE (104), a second reference signal precoded with a reconstructed channel (Ĥt) in a second time slot (t+1), wherein the reconstructed channel (Ĥt) is generated from the compressed channel (Ht) [Tan, ¶ 0140, “The WTRU may then receive the DL scheduling configuration and precoded reference signals to calculate 515 effective precoder gains, update the RI/CQI and select the precoder with optimal effective precoder gain. The WTRU may compute new RI/CQI based on the precoded reference signals and report updated RI/CQI and preferably, an indication of the selected precoder”, Step 510 occurs at a time later (t+1) than step 502 (t). The precoded RS represents reference signals that were precoded with reconstructed precoders. The reconstructed precoders at the base station side are adjusted based on the compressed channel matrix (H) received by the WTRU in step 506.]…
estimating, by the UE (104), a second channel Ht+1 and a second CQIt+1 at a second time instance [Tan, ¶ 0140, “The WTRU may then receive the DL scheduling configuration and precoded reference signals to calculate 515 effective precoder gains, update the RI/CQI and select the precoder with optimal effective precoder gain. The WTRU may compute new RI/CQI based on the precoded reference signals and report updated RI/CQI and preferably, an indication of the selected precoder”, Step 515 includes estimating/updating a new CQI based on the precoded RS signals, which are sent over a new channel, which is based on the precoders at the base station. The UE may determine a CQI difference from a prior CQI value is greater than an offset (see ¶ 0138).].
Tan et al. do not explicitly teach using an encoder of a two-sided model for Channel State Information (CSI) compression... by a decoder of the two-sided model of the BS (102)… comparing, by the UE (104), the second CQI (CQIt+1) and the first CQI (CQIt) for determining performance of the encoder of the two-sided model.
However, in an analogous art, Panasonic teaches using an encoder of a two-sided model for Channel State Information (CSI) compression [Panasonic, section 2.1, pg. 1, “CSI compression is realized by autoencoder. Autoencoder is typically trained end to end with a loss function to minimize the difference between input and reconstructed output. An auto-encoder has two main parts: an encoder that maps the original input into the “internal representation”, and a decoder that maps the “internal representation” to a reconstruction of the original input. Typically, the dimension of the “internal representation” can be smaller than the original input, and then, auto-encoder can realize the compression. An “encoder” of autoencoder corresponds to the UE processing in which the original input could be raw data (e.g., received CSI-RS value) or something after pre-computation (e.g., channel coefficient measured from CSI-RS, eigen vector, or coefficients before calculating Type II codebook). The output of UE encoder, which corresponds to “internal representation” in autoencoder, is transmitted from UE and received at gNB. An “decoder” of autoencoder corresponds to the gNB’s processing in which the original input is reconstructed from the “internal representation”, The paper pertains to CSI compression with a two-sided model for CSI feedback enhancement (see section 1). The UE computes CSI, then encodes the CSI for CSI feedback to the NW (or base station). See section 2.1, pg. 3, Type 2, First Bullet, particularly the CSI feedback step.]... by a decoder of the two-sided model of the BS (102) [Panasonic, section 2.1, pg. 1, “CSI compression is realized by autoencoder. Autoencoder is typically trained end to end with a loss function to minimize the difference between input and reconstructed output. An auto-encoder has two main parts: an encoder that maps the original input into the “internal representation”, and a decoder that maps the “internal representation” to a reconstruction of the original input. Typically, the dimension of the “internal representation” can be smaller than the original input, and then, auto-encoder can realize the compression. An “encoder” of autoencoder corresponds to the UE processing in which the original input could be raw data (e.g., received CSI-RS value) or something after pre-computation (e.g., channel coefficient measured from CSI-RS, eigen vector, or coefficients before calculating Type II codebook). The output of UE encoder, which corresponds to “internal representation” in autoencoder, is transmitted from UE and received at gNB. An “decoder” of autoencoder corresponds to the gNB’s processing in which the original input is reconstructed from the “internal representation”, The paper pertains to CSI compression with a two-sided model for CSI feedback enhancement (see section 1). The UE computes CSI, then encodes the CSI for CSI feedback to the NW (or base station). See section 2.1, pg. 3, Type 2, First Bullet, particularly the back propagation with gradients, which is used to indicate signals back to the UE (step 2).]… comparing, by the UE (104), the second CQI (CQIt+1) and the first CQI (CQIt) for determining performance of the encoder of the two-sided model [Panasonic, section 2.3, pg. 9, ¶ 3, “On Direction 2, if the output of the CSI reconstruction model is indicated by the network, this includes three steps. In the first step, UE feeds back the CSI feedback to the network. In the second step, the network recovers the CSI using the CSI reconstruction model, and indicates the recovery CSI to the UE afterwards. In the third step, the UE calculates the intermediate KPI (such as SGCS) with the measured ground-truth CSI and the received recovery CSI. Direction 2 requires much overhead to indicate the recovery CSI to the UE. Direction 2 may be useful for the case that network side of CSI reconstruction part is trained by UE (i.e., UE-side Type 1 joint training)”, Direction 2 AI/ML monitoring includes monitoring at the UE based on reconstructed CSI (or CSI based on a second transmission), see pg. 9, bullet 1. The UE may calculate KPI (SGCS) to determine performance of the two-sided model.].
Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to explicitly adopt the training collaboration and monitoring techniques of CSI feedback for AI/ML as taught by Panasonic into Tan et al. One would have been motivated to do this because the similar implementations between Tan et al. and Panasonic may be merged to further CSI feedback enhancement via a common two-sided model (see Tan, ¶ 0105 and Panasonic, section 1) with a reasonable expectation of success.
As per claim 3, Tan et al. in view of Panasonic teach the method as claimed in claim 1. Tan et al. do not explicitly teach further comprising receiving, from the BS (102), by the UE (104), a ground-truth CSI report in a periodic, aperiodic, or semi-persistent manner.
However, in an analogous art, Panasonic teaches receiving, from the BS (102), by the UE (104), a ground-truth CSI report in a periodic, aperiodic, or semi-persistent manner [Panasonic, section 2.3, pg. 9, ¶ 3, “On Direction 2, if the output of the CSI reconstruction model is indicated by the network, this includes three steps. In the first step, UE feeds back the CSI feedback to the network. In the second step, the network recovers the CSI using the CSI reconstruction model, and indicates the recovery CSI to the UE afterwards. In the third step, the UE calculates the intermediate KPI (such as SGCS) with the measured ground-truth CSI and the received recovery CSI. Direction 2 requires much overhead to indicate the recovery CSI to the UE. Direction 2 may be useful for the case that network side of CSI reconstruction part is trained by UE (i.e., UE-side Type 1 joint training)”, Direction 2 AI/ML monitoring includes monitoring at the UE based on reconstructed CSI (or CSI based on a second transmission), see pg. 9, bullet 1. Periodic, aperiodic, and semi-persistent cover all possibilities of ground-truth CSI, so therefore a mere recitation of ground truth CSI covers at least one of these possibilities.].
Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to explicitly adopt the training collaboration and monitoring techniques of CSI feedback for AI/ML as taught by Panasonic into Tan et al. One would have been motivated to do this because the similar implementations between Tan et al. and Panasonic may be merged to further CSI feedback enhancement via a common two-sided model (see Tan, ¶ 0105 and Panasonic, section 1) with a reasonable expectation of success.
As per claim 4, Tan et al. teach a UE (104) [Tan, fig. 5, WTRU/UE] comprising:
a processor [Tan, fig. 1B, element 118]; and a memory [Tan, fig. 1B, element 130] coupled to the processor, wherein the memory comprises programmed [Tan, ¶ 0043] instructions to:
receive a first reference signal in a first time slot (t) [Tan, ¶ 0140, “The base station determines and sends 502 configuration information and CSI-RS(s) to the WTRU”, The WTRU (or UE) receives a reference signal (CSI-RS) during a first time period (t). See also fig. 5, which is a signalling diagram for a basic AI/ML CSI feedback framework (see ¶s 0020, 0021, 0105, and 0106).];
estimate a first channel Ht and a first Channel Quality Indicator (CQIt) at a first time instance [Tan, ¶ 0140, “After receiving the traditional CSI-RS(s), the WTRU may estimate 505 CSI for channel H and determines a set of initial RI/CQI for a set of precoders according to configuration by the NW. The WTRU may compute RI/CQI based on the configuration of precoders”, The WTRU estimates channel H (see also fig. 5, step 505) and computes CQI of the CSI-RS for the time t based on the estimated precoders.];
compress the first channel (Ht) [Tan, ¶ 0140, “In method 500, the WTRU sends a compressed channel matrix (H) and the determined set of RI/CQI as part of CSI feedback for the base station to reconstruct 506 the precoders, for example in line with the WTRU configuration”, The channel is compressed for transmission to the base station.] …
transmit a compressed channel (Ht) along with the CQIt to a Base Station (BS) (102) [Tan, ¶ 0140, “In method 500, the WTRU sends a compressed channel matrix (H) and the determined set of RI/CQI as part of CSI feedback for the base station to reconstruct 506 the precoders, for example in line with the WTRU configuration”, CSI feedback may include the CQI and compressed channel matrix (H). See also ¶ 0141.];
receive a second reference signal precoded with a reconstructed channel (Ĥt) in a second time slot (t+1), wherein the reconstructed channel (Ĥt) is generated from the compressed channel (Ht) [Tan, ¶ 0140, “The WTRU may then receive the DL scheduling configuration and precoded reference signals to calculate 515 effective precoder gains, update the RI/CQI and select the precoder with optimal effective precoder gain. The WTRU may compute new RI/CQI based on the precoded reference signals and report updated RI/CQI and preferably, an indication of the selected precoder”, Step 510 occurs at a time later (t+1) than step 502 (t). The precoded RS represents reference signals that were precoded with reconstructed precoders. The reconstructed precoders at the base station side are adjusted based on the compressed channel matrix (H) received by the WTRU in step 506.] …
estimate a second channel Ht+1 and a second CQIt+1 at a second time instance [Tan, ¶ 0140, “The WTRU may then receive the DL scheduling configuration and precoded reference signals to calculate 515 effective precoder gains, update the RI/CQI and select the precoder with optimal effective precoder gain. The WTRU may compute new RI/CQI based on the precoded reference signals and report updated RI/CQI and preferably, an indication of the selected precoder”, Step 515 includes estimating/updating a new CQI based on the precoded RS signals, which are sent over a new channel, which is based on the precoders at the base station. The UE may determine a CQI difference from a prior CQI value is greater than an offset (see ¶ 0138).].
Tan et al. do not explicitly teach using an encoder of a two-sided model for Channel State Information (CSI) compression... by a decoder of the two-sided model of the BS (102)… comparing, by the UE (104), the second CQI (CQIt+1) and the first CQI (CQIt) for determining performance of the encoder of the two-sided model.
However, in an analogous art, Panasonic teaches using an encoder of a two-sided model for Channel State Information (CSI) compression [Panasonic, section 2.1, pg. 1, “CSI compression is realized by autoencoder. Autoencoder is typically trained end to end with a loss function to minimize the difference between input and reconstructed output. An auto-encoder has two main parts: an encoder that maps the original input into the “internal representation”, and a decoder that maps the “internal representation” to a reconstruction of the original input. Typically, the dimension of the “internal representation” can be smaller than the original input, and then, auto-encoder can realize the compression. An “encoder” of autoencoder corresponds to the UE processing in which the original input could be raw data (e.g., received CSI-RS value) or something after pre-computation (e.g., channel coefficient measured from CSI-RS, eigen vector, or coefficients before calculating Type II codebook). The output of UE encoder, which corresponds to “internal representation” in autoencoder, is transmitted from UE and received at gNB. An “decoder” of autoencoder corresponds to the gNB’s processing in which the original input is reconstructed from the “internal representation”, The paper pertains to CSI compression with a two-sided model for CSI feedback enhancement (see section 1). The UE computes CSI, then encodes the CSI for CSI feedback to the NW (or base station). See section 2.1, pg. 3, Type 2, First Bullet, particularly the CSI feedback step.]... by a decoder of the two-sided model of the BS (102) [Panasonic, section 2.1, pg. 1, “CSI compression is realized by autoencoder. Autoencoder is typically trained end to end with a loss function to minimize the difference between input and reconstructed output. An auto-encoder has two main parts: an encoder that maps the original input into the “internal representation”, and a decoder that maps the “internal representation” to a reconstruction of the original input. Typically, the dimension of the “internal representation” can be smaller than the original input, and then, auto-encoder can realize the compression. An “encoder” of autoencoder corresponds to the UE processing in which the original input could be raw data (e.g., received CSI-RS value) or something after pre-computation (e.g., channel coefficient measured from CSI-RS, eigen vector, or coefficients before calculating Type II codebook). The output of UE encoder, which corresponds to “internal representation” in autoencoder, is transmitted from UE and received at gNB. An “decoder” of autoencoder corresponds to the gNB’s processing in which the original input is reconstructed from the “internal representation”, The paper pertains to CSI compression with a two-sided model for CSI feedback enhancement (see section 1). The UE computes CSI, then encodes the CSI for CSI feedback to the NW (or base station). See section 2.1, pg. 3, Type 2, First Bullet, particularly the back propagation with gradients, which is used to indicate signals back to the UE (step 2).]… comparing, by the UE (104), the second CQI (CQIt+1) and the first CQI (CQIt) for determining performance of the encoder of the two-sided model [Panasonic, section 2.3, pg. 9, ¶ 3, “On Direction 2, if the output of the CSI reconstruction model is indicated by the network, this includes three steps. In the first step, UE feeds back the CSI feedback to the network. In the second step, the network recovers the CSI using the CSI reconstruction model, and indicates the recovery CSI to the UE afterwards. In the third step, the UE calculates the intermediate KPI (such as SGCS) with the measured ground-truth CSI and the received recovery CSI. Direction 2 requires much overhead to indicate the recovery CSI to the UE. Direction 2 may be useful for the case that network side of CSI reconstruction part is trained by UE (i.e., UE-side Type 1 joint training)”, Direction 2 AI/ML monitoring includes monitoring at the UE based on reconstructed CSI (or CSI based on a second transmission), see pg. 9, bullet 1. The UE may calculate KPI (SGCS) to determine performance of the two-sided model].
Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to explicitly adopt the training collaboration and monitoring techniques of CSI feedback for AI/ML as taught by Panasonic into Tan et al. One would have been motivated to do this because the similar implementations between Tan et al. and Panasonic may be merged to further CSI feedback enhancement via a common two-sided model (see Tan, ¶ 0105 and Panasonic, section 1) with a reasonable expectation of success.
Claim 2 is rejected under 35 U.S.C. 103 as being unpatentable over Tan et al. (WO 2024/173366) in view of Panasonic (R1-2307004, NPL W on pg. 2 of PTO-892) and NVIDIA (R1- 2305160, NPL V on pg. 2 of PTO-892).
As per claim 2, Tan et al. in view of Panasonic teaches the method as claimed in claim 1. Tan et al. do not explicitly teach further comprising transmitting the second CQI (CQIt+1) to the BS (102).
However, in an analogous art, NVIDIA teaches transmitting the second CQI (CQIt+1) to the BS (102) [NVIDIA, section 2.1, Type-3, “Step 2: After UE side training is finished, UE side shares network side with a set of information (e.g., a training dataset consisting of (encoder output, target CSI)) that is used by the network side to be able to train the CSI reconstruction part. Step3: Network side trains the network side CSI reconstruction part based on the received set of information”, The UE performs its training, then reports CQI (or encoder output) to the NW side for further training. This is considered joint training, see Fig. 3, where the UE and base station exchange gradients (forward and backward), which are used for output comparison and training (see pg. 4).], wherein the BS (102) compares the second CQI (CQIt+1) and the first CQI (CQIt) for determining the performance of the encoder of the two-sided model [Panasonic section 2.1, pg. 5, ¶ 2, “For the evaluation of an example of Type 3, several evaluation cases for sequential training can be considered for multi-vendors. In the baseline Case 1, Type 3 training is carried out between one NW part model and one UE part model. This case can be naturally applied to the NW-first training case with 1 NW part model to M>1 separate UE part models”, The NW side and UE side models may be trained together, where the NW (or BS side) model is trained and the accuracy is measured according to KPIs (see section 3, pg. 7, ¶ 5).].
Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to explicitly adopt the training collaboration and monitoring techniques of CSI feedback for AI/ML as taught by NVIDIA into Tan et al. One would have been motivated to do this because the similar implementations between Tan et al. and Panasonic may be merged to further CSI feedback enhancement via a common two-sided model (see Tan, ¶ 0105 and NVIDIA, section 2.1, ¶ 3) with a reasonable expectation of success.
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over NVIDIA (R1- 2305160, NPL V on pg. 2 of PTO-892) in view of Panasonic (R1-2307004, NPL W on pg. 2 of PTO-892).
As per claim 5, NVIDIA teaches a BS (102) [NVIDIA, section 2.1, pg. 5, Fig. 3] comprising:
a processor; and a memory coupled to the processor, wherein the memory comprises programmed instructions [NVIDIA, section 2.1, pg. 5, Fig. 3, The base station of the network (NW) side is structure that is well recognized as including at least a transceiver, a processor, and a memory to perform its functions.] to:
estimate a first channel Ht and a first Channel Quality Indicator (CQIt) at a first time instance [NVIDIA, section 2.1, Type-2, “Figure 3 illustrates Type-2 training. In this case, joint training is performed at two sides in a single training session. The encoder model and decoder model are trained at UE side and at network side, respectively. But the models are jointly trained in the same loop for forward propagation and backward propagation, by exchanging forward activation and backward gradient between UE-side entity and network-side entity”, Fig. 2 shows joint training (text says fig. 3). Joint training is where both the UE and NW (or base station) train their models using CSI feedback (or CQI, see section 2.1). Gradients represent the change of one value from another. Therefore, training with backward gradients, indicates that a base station receives at least two CSI feedbacks (or CQIs) for eventual comparison and KPI generation.];
receive, from a UE (104), a second CQI (CQIt+1) estimated by the UE (104) [NVIDIA, section 2.1, Type-2, “Figure 3 illustrates Type-2 training. In this case, joint training is performed at two sides in a single training session. The encoder model and decoder model are trained at UE side and at network side, respectively. But the models are jointly trained in the same loop for forward propagation and backward propagation, by exchanging forward activation and backward gradient between UE-side entity and network-side entity”, Fig. 2 shows joint training (text says fig. 3). Joint training is where both the UE and NW (or base station) train their models using CSI feedback (or CQI, see section 2.1). Gradients represent the change of one value from another. Therefore, training with backward gradients, indicates that a base station receives at least two CSI feedbacks (or CQIs) for eventual comparison and KPI generation.].
NVIDIA does not explicitly teach compare the second CQI (CQIt+1) and the first CQI (CQIt) for determining performance of an encoder of a two-sided model for CSI compression.
However, in an analogous art, Panasonic teaches compare the second CQI (CQIt+1) and the first CQI (CQIt) for determining performance of an encoder of a two-sided model for CSI compression [Panasonic, section 2.3, pg. 9, “In the second step, the network recovers the CSI using the CSI reconstruction part, and calculate the intermediate KPI (such as SGCS) with the recovery and the reported ground-truth CSI. For Direction 1, the potential specification impact is how to obtain/report target CSI from UE to network, and it can be one of discussion points in data collection”, Direction 2 AI/ML monitoring includes monitoring at the UE based on reconstructed CSI (or CSI based on a second transmission), see pg. 9, bullet 1. The UE may calculate KPI (SGCS) to determine performance of the two-sided model.].
Thus, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to explicitly adopt the training of CSI feedback for AI/ML as taught by NVIDIA into Panasonic. One would have been motivated to do this because the similar implementations between Panasonic and NVIDIA may be merged to further CSI feedback enhancement via a common two-sided model (see NVIDIA, section 2.1 and Panasonic, section 1) with a reasonable expectation of success.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
The reference, Than et al. (WO 2026021302), teaches combining CQI and channel information (see fig. 10).
The reference, Tan et al. (WO 2024173359), teaches sending CQI and a compressed channel H to a base station (see fig. 9).
The reference, Guo et al. (NPL), teaches channel prediction based on CSI feedback using an autoencoder (see pg. 175, “Joint Design with Channel Prediction”).
The reference, Huawei (R1-2304653), teaches CSI compression for CSI feedback (see section 3).
The reference, CATT (R1-2304722), teaches two-sided AI/ML models for CSI feedback (see section 2.1.4).
The reference, Intel (R1-2304813), teaches SRS-based model performance monitoring (see section 2.7).
The reference, DOCOMO (R1-2307467), teaches performance monitoring of a two-sided model (see section 2.4).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Paul H. Masur whose telephone number is (571)270-7297. The examiner can normally be reached Monday to Friday, 4:30 AM to 5PM.
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, Rebecca Song can be reached at (571) 270-3667. 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.
/Paul H. Masur/
Primary Examiner
Art Unit 2417
Prosecution Insights