Apparatus, computer program and method for CSI encoding and selection of quantization mode

Abstract

An apparatus comprising means for: receiving, from a network node, at least one ref- erence signal; determining at least one latent vector based on target Channel State Information acquired from the at least one reference signal and a history-independ- ent CSI encoder; using the at least one latent vector to provide an input into a his- tory-dependent recurrent quantizer to provide a quantized output; calculating a mean squared error between the input and the quantized output; comparing the mean squared error to a threshold level; based on the comparing, determining to op- erate in a irst mode, wherein in the irst mode, the apparatus performs: determining at least one further latent vector based on target Channel State Information acquired from the at least one reference signal and the history-independent CSI encoder; us- ing the at least one further latent vector to provide an input into a scalar quantizer to provide a full-resolution quantization output; sending the full-resolution quantiza- tion output to the network node.

Description

[0001] APPARATUS, COMPUTER PROGRAM AND METHOD

[0002] TECHNICAL FIELD

[0003] Various example embodiments relate generally to Channel State Information, (CSI) for a telecommunications system. Some examples relate to compressing CSI, or an architecture for compressing CSI.

[0004] BACKGROUND

[0005] A CSI-Reference Signal (CSI-RS) is a reference signal that can be used in the Downlink direction in 5G New Radio (NR). It can be used for for the purpose of Channel Sounding and used to measure the characteristics of a radio channel so that it can use correct modulation, code rate, beam forming etc.

[0006] BRIEF DESCRIPTION

[0007] According to an aspect of the invention, there is provided an apparatus comprising means for: receiving, from a network node, at least one reference signal; determining at least one latent vector based on target Channel State Information acquired from the at least one reference signal and a history-independent CSI encoder; using the at least one latent vector to provide an input into a historydependent recurrent quantizer to provide a quantized output; calculating a mean squared error between the input and the quantized output; comparing the mean squared error to a threshold level; based on the comparing, determining to operate in a first mode, wherein in the first mode, the apparatus performs: determining at least one further latent vector based on target Channel State Information acquired from the at least one reference signal and the history-independent CSI encoder; using the at least one further latent vector to provide an input into a scalar quantizer to provide a full-resolution quantization output; sending the full-resolution quantization output to the network node.

[0008] According to some examples, the apparatus determines to operate in the first mode when the mean squared error is above the threshold level.

[0009] According to some examples, the means are further configured for: based on the comparing, determining to operate in a second mode, wherein in the second mode the apparatus performs: sending the quantized output to the network node.

[0010] According to some examples, the apparatus determines to operate in the second mode when the mean squared error is below the threshold level and above a second threshold level.

[0011] According to some examples, the means are further configured for: determining whether the mean squared error between the input and the quantized output is below a second threshold for a predefined amount of time; when the mean squared error is below the second threshold for a predetermined time, sending a first indication to the network node.

[0012] According to some examples, the means are further configured for: after sending the first indication to the network node , receiving, from the network node, a command to operate in a third mode, wherein in the third mode the apparatus performs: i) receiving, from the network node, one or more reference signals; ii) inputting target Channel State Information acquired from channel estimates based on the one or more reference signals into the history-independent CS1 encoder to provide one or more latent vectors; hi) using the one or more latent vectors to provide an input into a history-dependent recurrent quantizer to provide a second quantized output; iv) sending the second quantized output to the network node; repeating steps i) to iv) to provide a plurality of inputs and a corresponding plurality of quantized outputs; sending the plurality of quantized outputs to the network node; sending, to the network node, a plurality of Squared Generalized Cosine Similarity, SGCS, scores and / or a plurality of Mean Squared Error, MSE, values between each of the inputs and the corresponding quantized outputs.

[0013] According to some examples, in the third mode a recurrent quantizer state machine of the apparatus and an inverse recurrent quantizer state machine of the network node are synchronized by aligning a latest step size and a latest dequantized value for the history-dependent recurrent scalar quantizer.

[0014] According to some examples, the means are further configured for: receiving, from the network node, a request for a status of the state machine of the apparatus; sending, to the network node, the status for the state machine of the apparatus.

[0015] According to some examples, at least one of the first threshold and the second threshold is at least one of: configured by the network node at the user equipment; aligned between the apparatus and the network node.

[0016] According to an aspect of the invention, there is provided a method comprising: receiving, from a network node, at least one reference signal; determining at least one latent vector based on target Channel State Information acquired from the at least one reference signal and a history-independent CS1 encoder; using the at least one latent vector to provide an input into a history-dependent recurrent quantizer to provide a quantized output; calculating a mean squared error between the input and the quantized output; comparing the mean squared error to a threshold level; based on the comparing, determining to operate in a first mode, wherein in the first mode, the apparatus performs: determining at least one further latent vector based on target Channel State Information acquired from the at least one reference signal and the history-independent CS1 encoder; using the at least one further latent vector to provide an input into a scalar quantizer to provide a full-resolution quantization output; sending the full-resolution quantization output to the network node.

[0017] According to some examples, the method comprises determining to operate in the first mode when the mean squared error is above the threshold level.

[0018] According to some examples, the method comprises based on the comparing, determining to operate in a second mode, wherein in the second mode the apparatus performs: sending the quantized output to the network node.

[0019] According to some examples, the method comprises determining to operate in the second mode when the mean squared error is below the threshold level and above a second threshold level.

[0020] According to some examples, the method comprises: determining whether the mean squared error between the input and the quantized output is below a second threshold for a predefined amount of time; when the mean squared error is below the second threshold for a predetermined time, sending a first indication to the network node.

[0021] According to some examples, the method comprises: after sending the first indication to the network node , receiving, from the network node, a command to operate in a third mode, wherein in the third mode the apparatus performs: i) receiving, from the network node, one or more reference signals; ii) inputting target Channel State Information acquired from channel estimates based on the one or more reference signals into the history-independent CS1 encoder to provide one or more latent vectors; hi) using the one or more latent vectors to provide an input into a history-dependent recurrent quantizer to provide a second quantized output; iv) sending the second quantized output to the network node; repeating steps i) to iv) to provide a plurality of inputs and a corresponding plurality of quantized outputs; sending the plurality of quantized outputs to the network node; sending, to the network node, a plurality of Squared Generalized Cosine Similarity, SGCS, scores and / or a plurality of Mean Squared Error, MSE, values between each of the inputs and the corresponding quantized outputs.

[0022] According to some examples, in the third mode a recurrent quantizer state machine of the apparatus and an inverse recurrent quantizer state machine of the network node are synchronized by aligning a latest step size and a latest dequantized value for the history-dependent recurrent scalar quantizer.

[0023] According to some examples, the method comprises: receiving, from the network node, a request for a status of the state machine of the apparatus; sending, to the network node, the status for the state machine of the apparatus.

[0024] According to some examples, at least one of the first threshold and the second threshold is at least one of: configured by the network node at the user equipment; aligned between the apparatus and the network node.

[0025] According to an aspect of the invention, there is provided an apparatus comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: receiving, from a network node, at least one reference signal; determining at least one latent vector based on target Channel State Information acquired from the at least one reference signal and a history-independent CS1 encoder; using the at least one latent vector to provide an input into a history-dependent recurrent quantizer to provide a quantized output; calculating a mean squared error between the input and the quantized output; comparing the mean squared error to a threshold level; based on the comparing, determining to operate in a first mode, wherein in the first mode, the apparatus performs: determining at least one further latent vector based on target Channel State Information acquired from the at least one reference signal and the history-independent CS1 encoder; using the at least one further latent vector to provide an input into a scalar quantizer to provide a fullresolution quantization output; sending the full-resolution quantization output to the network node.

[0026] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: determining to operate in the first mode when the mean squared error is above the threshold level.

[0027] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: based on the comparing, determining to operate in a second mode, wherein in the second mode the apparatus performs: sending the quantized output to the network node.

[0028] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: determining to operate in the second mode when the mean squared error is below the threshold level and above a second threshold level.

[0029] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: determining whether the mean squared error between the input and the quantized output is below a second threshold for a predefined amount of time; when the mean squared error is below the second threshold for a predetermined time, sending a first indication to the network node.

[0030] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: after sending the first indication to the network node , receiving, from the network node, a command to operate in a third mode, wherein in the third mode the apparatus performs: i) receiving, from the network node, one or more reference signals; ii) inputting target Channel State Information acquired from channel estimates based on the one or more reference signals into the history-independent CS1 encoder to provide one or more latent vectors; hi) using the one or more latent vectors to provide an input into a history-dependent recurrent quantizer to provide a second quantized output; iv) sending the second quantized output to the network node; repeating steps i) to iv) to provide a plurality of inputs and a corresponding plurality of quantized outputs; sending the plurality of quantized outputs to the network node; sending, to the network node, a plurality of Squared Generalized Cosine Similarity, SGCS, scores and / or a plurality of Mean Squared Error, MSE, values between each of the inputs and the corresponding quantized outputs.

[0031] According to some examples, in the third mode a recurrent quantizer state machine of the apparatus and an inverse recurrent quantizer state machine of the network node are synchronized by aligning a latest step size and a latest dequantized value for the history-dependent recurrent scalar quantizer.

[0032] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: receiving, from the network node, a request for a status of the state machine of the apparatus; sending, to the network node, the status for the state machine of the apparatus.

[0033] According to some examples, at least one of the first threshold and the second threshold is at least one of: configured by the network node at the user equipment; aligned between the apparatus and the network node.

[0034] According to an aspect of the invention, there is provided a computer program product embodied on a distribution medium readable by a computer and comprising program instructions which, when loaded into an apparatus, execute a method comprising: receiving, from a network node, at least one reference signal; determining at least one latent vector based on target Channel State Information acquired from the at least one reference signal and a history-independent CS1 encoder; using the at least one latent vector to provide an input into a history-dependent recurrent quantizer to provide a quantized output; calculating a mean squared error between the input and the quantized output; comparing the mean squared error to a threshold level; based on the comparing, determining to operate in a first mode, wherein in the first mode, the apparatus performs: determining at least one further latent vector based on target Channel State Information acquired from the at least one reference signal and the history-independent CS1 encoder; using the at least one further latent vector to provide an input into a scalar quantizer to provide a full-resolution quantization output; sending the full-resolution quantization output to the network node.

[0035] According to an aspect of the invention, there is provided an apparatus comprising means for: sending, to a user equipment, at least one reference signal; receiving a full-resolution quantization output from a user equipment, wherein the full-resolution quantization output is generated at the user equipment by inputting the at least one reference signal into a channel estimator, a history-independent CS1 encoder and then a scalar quantizer.

[0036] According to some examples, the means are further configured for: receiving a quantized output from the user equipment, wherein an input is at least one further latent vector generated based on target Channel State Information acquired from channel estimates based on the at least one reference signal and a history-independent CS1 encoder; and wherein the quantized output is generated at the user equipment by providing the input into a recurrent quantizer.

[0037] According to some examples, the means are further configured for: using a recurrent de-quantizer to de-quantize the quantized output to provide a second output; determining reconstructed CS1 of the channel between the apparatus and the user equipment based on the second output and a decoder.

[0038] According to some examples, the means are further configured for: when the mean squared error is below a second threshold, receiving an indication from the user equipment that the output of the recurrent quantizer is stable.

[0039] According to some examples, the means are further configured for: sending, to the user equipment, a command to change to a mode of operation where quantization error for the quantized output is below the second threshold level. According to some examples, in the mode of operation, a recurrent quantizer state machine of the user equipment and an inverse recurrent quantizer state machine of the apparatus are synchronized by aligning a latest step size and a latest de-quantized value for the history-dependent recurrent scalar quantizer.

[0040] According to some examples, the means are further configured for: sending, to the user equipment, a request for a status of a state machine of the user equipment; receiving, from the user equipment, the status for the state machine of the user equipment; synchronizing a state machine at the apparatus according to the status for the state machine of the user equipment.

[0041] According to some examples, the second threshold is at least one of: configured at the user equipment by the apparatus; aligned between the apparatus and the user equipment.

[0042] According to an aspect of the invention, there is provided a method comprising: receiving a quantized output from the user equipment, wherein an input is at least one further latent vector generated based on target Channel State Information acquired from channel estimates based on the at least one reference signal and a history-independent CS1 encoder; and wherein the quantized output is generated at the user equipment by providing the input into a recurrent quantizer.

[0043] According to some examples, the method comprises: using a recurrent de-quantizer to de-quantize the quantized output to provide a second output; determining reconstructed CS1 of the channel between the apparatus and the user equipment based on the second output and a decoder.

[0044] According to some examples, the method comprises, when the mean squared error is below a second threshold, receiving an indication from the user equipment that the output of the recurrent quantizer is stable.

[0045] According to some examples, the method comprises: sending, to the user equipment, a command to change to a mode of operation where quantization error for the quantized output is below the second threshold level. According to some examples, in the mode of operation, a recurrent quantizer state machine of the user equipment and an inverse recurrent quantizer state machine of the apparatus are synchronized by aligning a latest step size and a latest de-quantized value for the history-dependent recurrent scalar quantizer.

[0046] According to some examples, the method comprises: sending, to the user equipment, a request for a status of a state machine of the user equipment; receiving, from the user equipment, the status for the state machine of the user equipment; synchronizing a state machine at the apparatus according to the status for the state machine of the user equipment.

[0047] According to an aspect of the invention, there is provided an apparatus comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: receiving a quantized output from the user equipment, wherein an input is at least one further latent vector generated based on target Channel State Information acquired from channel estimates based on the at least one reference signal and a history-independent CS1 encoder; and wherein the quantized output is generated at the user equipment by providing the input into a recurrent quantizer.

[0048] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: using a recurrent de-quantizer to de-quantize the quantized output to provide a second output; determining reconstructed CS1 of the channel between the apparatus and the user equipment based on the second output and a decoder.

[0049] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: when the mean squared error is below a second threshold, receiving an indication from the user equipment that the output of the recurrent quantizer is stable.

[0050] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: sending, to the user equipment, a command to change to a mode of operation where quantization error for the quantized output is below the second threshold level. According to some examples, in the mode of operation, a recurrent quantizer state machine of the user equipment and an inverse recurrent quantizer state machine of the apparatus are synchronized by aligning a latest step size and a latest de-quantized value for the history-dependent recurrent scalar quantizer.

[0051] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: sending, to the user equipment, a request for a status of a state machine of the user equipment; receiving, from the user equipment, the status for the state machine of the user equipment; synchronizing a state machine at the apparatus according to the status for the state machine of the user equipment.

[0052] According to an aspect of the invention, there is provided a computer program product embodied on a distribution medium readable by a computer and comprising program instructions which, when loaded into an apparatus, execute a method comprising: receiving a quantized output from the user equipment, wherein an input is at least one further latent vector generated based on target Channel State Information acquired from channel estimates based on the at least one reference signal and a history-independent CS1 encoder; and wherein the quantized output is generated at the user equipment by providing the input into a recurrent quantizer.

[0053] According to an aspect of the invention, there is provided an apparatus comprising means for: i) receiving, from a network node, at least one reference signal; ii) inputting target Channel State Information acquired from channel estimates based on the at least one reference signal into a history-independent CS1 encoder to provide at least one latent vector; hi) using the at least one latent vector to provide an input into a history-dependent recurrent quantizer to provide a quantized output; iv) sending the quantized output to the network node; repeating steps i) to iv) to provide a plurality of inputs and a corresponding plurality of quantized outputs; sending the plurality of quantized outputs to the network node; sending, to the network node, a plurality of Squared Generalized Cosine Similarity, SGCS, scores and / or a plurality of Mean Squared Error, MSE, values between each of the inputs and the corresponding quantized outputs. According to some examples, a cause of performance degradation is associated with a scenario where a block error rate, BLER, of a downlink channel from the network node to the apparatus is above a threshold.

[0054] According to some examples, the apparatus sends the plurality of SGCS values and the plurality of MSE values to the network node, wherein a cause of CS1 loss due to Uplink Control Information, UC1, failure is associated with a scenario where BLER is above the threshold but the MSE values do not increase over a threshold and the SGCS values do not decrease under a threshold during a time period corresponding to the BLER increasing above the threshold.

[0055] According to some examples, the apparatus sends the plurality of SGCS values and the plurality of MSE values to the network node, and wherein the means are further configured for: receiving, from the network node, a request to report state machine status of the recurrent quantizer, wherein receiving the request from the network node is triggered based on the MSE and the SGCS values sent to the network node from the apparatus; sending the state machine status to the network node for synchronization of a state machine at the network node.

[0056] According to some examples, the apparatus sends the plurality of SGCS values and the plurality of MSE values to the network node, and wherein a cause of CS1 encoder inference degradation and / or channel drift is associated with a scenario where the MSE values do not increase over a threshold but the SCGS values do decrease under a threshold during a time corresponding the BLER increasing above the threshold.

[0057] According to some examples, the apparatus sends the plurality of SGCS values and the plurality of MSE values to the network node, and wherein a cause of recurrent quantizer tracking degradation is associated with scenario where the MSE values do increase over a threshold but the SCGS values do not decrease under a threshold during a time corresponding to the BLER increasing above the threshold.

[0058] According to some examples, the apparatus sends the plurality of SGCS values and the plurality of MSE values to the network node, the means being further configured for: when the cause is identified as increased channel variation and / or non-continuous downlink transmission, receiving a request to use a scalar quantization scheme without recurrent tracking instead of the recurrent quantizer, wherein receiving the request from the network node is triggered based on the MSE and the SGCS values sent to the network node from the apparatus; using the scalar quantization scheme without recurrent tracking instead of the recurrent quantizer.

[0059] According to some examples, the apparatus sends the plurality of SGCS values and the plurality of MSE values to the network node, and wherein the means are further configured for: logging the SGCS scores and the MSE values in logs; receiving, from the network node and when a block error rate, BLER, of a downlink channel from the network node to the apparatus is above a threshold, a request to send the logs to the network node; sending the logs to the network node.

[0060] According to some examples, the plurality of SGCS scores provide a metric of similarity between the target CS1 and a reconstructed target CS1 at the network node.

[0061] According to an aspect of the invention, there is provided a method comprising: i) receiving, from a network node, at least one reference signal; ii) inputting target Channel State Information acquired from channel estimates based on the at least one reference signal into a history-independent CS1 encoder to provide at least one latent vector; hi) using the at least one latent vector to provide an input into a history-dependent recurrent quantizer to provide a quantized output; iv) sending the quantized output to the network node; repeating steps i) to iv) to provide a plurality of inputs and a corresponding plurality of quantized outputs; sending the plurality of quantized outputs to the network node; sending, to the network node, a plurality of Squared Generalized Cosine Similarity, SGCS, scores and / or a plurality of Mean Squared Error, MSE, values between each of the inputs and the corresponding quantized outputs.

[0062] According to some examples, a cause of performance degradation is associated with a scenario where a block error rate, BLER, of a downlink channel from the network node to the apparatus is above a threshold. According to some examples, the method comprises sending the plurality of SGCS values and the plurality of MSE values to the network node, wherein a cause of CS1 loss due to Uplink Control Information, UC1, failure is associated with a scenario where BLER is above the threshold but the MSE values do not increase over a threshold and the SGCS values do not decrease under a threshold during a time period corresponding to the BLER increasing above the threshold.

[0063] According to some examples, the method comprises sending the plurality of SGCS values and the plurality of MSE values to the network node, and wherein the means are further configured for: receiving, from the network node, a request to report state machine status of the recurrent quantizer, wherein receiving the request from the network node is triggered based on the MSE and the SGCS values sent to the network node from the apparatus; sending the state machine status to the network node for synchronization of a state machine at the network node.

[0064] According to some examples, the method comprises sending the plurality of SGCS values and the plurality of MSE values to the network node, and wherein a cause of CS1 encoder inference degradation and / or channel drift is associated with a scenario where the MSE values do not increase over a threshold but the SCGS values do decrease under a threshold during a time corresponding the BLER increasing above the threshold.

[0065] According to some examples, the method comprises sending the plurality of SGCS values and the plurality of MSE values to the network node, and wherein a cause of recurrent quantizer tracking degradation is associated with scenario where the MSE values do increase over a threshold but the SCGS values do not decrease under a threshold during a time corresponding to the BLER increasing above the threshold.

[0066] According to some examples, the method comprises sending the plurality of SGCS values and the plurality of MSE values to the network node, the means being further configured for: when the cause is identified as increased channel variation and / or non-continuous downlink transmission, receiving a request to use a scalar quantization scheme without recurrent tracking instead of the recurrent quantizer, wherein receiving the request from the network node is triggered based on the MSE and the SGCS values sent to the network node from the apparatus; using the scalar quantization scheme without recurrent tracking instead of the recurrent quantizer.

[0067] According to some examples, the method comprising sending the plurality of SGCS values and the plurality of MSE values to the network node, and wherein the means are further configured for: logging the SGCS scores and the MSE values in logs; receiving, from the network node and when a block error rate, BLER, of a downlink channel from the network node to the apparatus is above a threshold, a request to send the logs to the network node; sending the logs to the network node.

[0068] According to some examples, the plurality of SGCS scores provide a metric of similarity between the target CS1 and a reconstructed target CS1 at the network node.

[0069] According to an aspect of the invention, there is provided an apparatus comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: i) receiving, from a network node, at least one reference signal; ii) inputting target Channel State Information acquired from channel estimates based on the at least one reference signal into a history-independent CS1 encoder to provide at least one latent vector; hi) using the at least one latent vector to provide an input into a history-dependent recurrent quantizer to provide a quantized output; iv) sending the quantized output to the network node; repeating steps i) to iv) to provide a plurality of inputs and a corresponding plurality of quantized outputs; sending the plurality of quantized outputs to the network node; sending, to the network node, a plurality of Squared Generalized Cosine Similarity, SGCS, scores and / or a plurality of Mean Squared Error, MSE, values between each of the inputs and the corresponding quantized outputs.

[0070] According to some examples, a cause of performance degradation is associated with a scenario where a block error rate, BLER, of a downlink channel from the network node to the apparatus is above a threshold. According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: sending the plurality of SGCS values and the plurality of MSE values to the network node, wherein a cause of CS1 loss due to Uplink Control Information, UC1, failure is associated with a scenario where BLER is above the threshold but the MSE values do not increase over a threshold and the SGCS values do not decrease under a threshold during a time period corresponding to the BLER increasing above the threshold.

[0071] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: sending the plurality of SGCS values and the plurality of MSE values to the network node, and wherein the means are further configured for: receiving, from the network node, a request to report state machine status of the recurrent quantizer, wherein receiving the request from the network node is triggered based on the MSE and the SGCS values sent to the network node from the apparatus; sending the state machine status to the network node for synchronization of a state machine at the network node.

[0072] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: sending the plurality of SGCS values and the plurality of MSE values to the network node, and wherein a cause of CS1 encoder inference degradation and / or channel drift is associated with a scenario where the MSE values do not increase over a threshold but the SCGS values do decrease under a threshold during a time corresponding the BLER increasing above the threshold.

[0073] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: sending the plurality of SGCS values and the plurality of MSE values to the network node, and wherein a cause of recurrent quantizer tracking degradation is associated with scenario where the MSE values do increase over a threshold but the SCGS values do not decrease under a threshold during a time corresponding to the BLER increasing above the threshold.

[0074] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: sending the plurality of SGCS values and the plurality of MSE values to the network node, the means being further configured for: when the cause is identified as increased channel variation and / or non-continuous downlink transmission, receiving a request to use a scalar quantization scheme without recurrent tracking instead of the recurrent quantizer, wherein receiving the request from the network node is triggered based on the MSE and the SGCS values sent to the network node from the apparatus; using the scalar quantization scheme without recurrent tracking instead of the recurrent quantizer.

[0075] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: sending the plurality of SGCS values and the plurality of MSE values to the network node, and wherein the means are further configured for: logging the SGCS scores and the MSE values in logs; receiving, from the network node and when a block error rate, BLER, of a downlink channel from the network node to the apparatus is above a threshold, a request to send the logs to the network node; sending the logs to the network node.

[0076] According to some examples, the plurality of SGCS scores provide a metric of similarity between the target CS1 and a reconstructed target CS1 at the network node.

[0077] According to an aspect of the invention, there is provided a computer program product embodied on a distribution medium readable by a computer and comprising program instructions which, when loaded into an apparatus, execute a method comprising: i) receiving, from a network node, at least one reference signal; ii) inputting target Channel State Information acquired from channel estimates based on the at least one reference signal into a history-independent CS1 encoder to provide at least one latent vector; hi) using the at least one latent vector to provide an input into a history-dependent recurrent quantizer to provide a quantized output; iv) sending the quantized output to the network node; repeating steps i) to iv) to provide a plurality of inputs and a corresponding plurality of quantized outputs; sending the plurality of quantized outputs to the network node; sending, to the network node, a plurality of Squared Generalized Cosine Similarity, SGCS, scores and / or a plurality of Mean Squared Error, MSE, values between each of the inputs and the corresponding quantized outputs.

[0078] According to an aspect of the invention, there is provided an apparatus comprising means for: sending, to a user equipment, at least one reference signal; receiving, from the user equipment, a plurality of quantized outputs determined based on the at least one reference signal; receiving, from the user equipment, a plurality of Squared Generalized Cosine Similarity, SGCS, scores associated with the quantized outputs and / or a plurality of Mean Squared Error, MSE, values between each of the plurality of quantized outputs and a corresponding input used by the user equipment to determine the corresponding quantized output.

[0079] According to some examples, a cause of performance degradation is associated with a scenario where a block error rate, BLER, of a downlink channel from the apparatus to the user equipment is above a threshold.

[0080] According to some examples, the apparatus receives the plurality of SGCS values and the plurality of MSE values from the user equipment, wherein a cause of Channel State Information, CS1, loss due to Uplink Control Information, UC1, failure is associated with a scenario where BLER is above the threshold but the MSE values do not increase over a threshold and the SGCS values do not decrease under a threshold during a time period corresponding to the BLER increasing above the threshold.

[0081] According to some examples, the apparatus receives the plurality of SGCS values and the plurality of MSE values from the user equipment, wherein the means are further configured for: sending, based on the MSE and the SGCS values and to the user equipment, a request to report state machine status of a recurrent quantizer of the user equipment, receiving the state machine status from the user equipment; synchronizing a state machine at the apparatus based on the state machine status of the user equipment.

[0082] According to some examples, the apparatus receives the plurality of SGCS values and the plurality of MSE values from the user equipment, wherein a cause of CS1 encoder inference degradation and / or channel drift is associated with a scenario where the MSE values do not increase over a threshold but the SCGS values do decrease under a threshold during a time corresponding to the BLER increasing above the threshold.

[0083] According to some examples, the apparatus receives the plurality of SGCS values and the plurality of MSE values from the user equipment, wherein a cause of recurrent quantizer tracking degradation is associated with scenario where the MSE values do increase over a threshold but the SCGS values do not decrease under a threshold during a time corresponding the BLER increasing above the threshold.

[0084] According to some examples, the means are further configured for: when the cause is identified as increased channel variation and / or non-continuous downlink transmission, sending a request to the user equipment to use a scalar quantization scheme without recurrent tracking.

[0085] According to some examples, the apparatus receives the plurality of SGCS values and the plurality of MSE values from the user equipment, wherein the means are further configured for: when the BLER of a downlink channel from the apparatus to the user equipment is above a threshold, sending a request for a log of the SGCS values and the MSE values; receiving the log from the equipment.

[0086] According to some examples, the plurality of SGCS scores provide a metric of similarity between a target CS1 and a reconstructed target CS1 at the apparatus.

[0087] According to an aspect of the invention, there is provided a method comprising: sending, to a user equipment, at least one reference signal; receiving, from the user equipment, a plurality of quantized outputs determined based on the at least one reference signal; receiving, from the user equipment, a plurality of Squared Generalized Cosine Similarity, SGCS, scores associated with the quantized outputs and / or a plurality of Mean Squared Error, MSE, values between each of the plurality of quantized outputs and a corresponding input used by the user equipment to determine the corresponding quantized output.

[0088] According to some examples, a cause of performance degradation is associated with a scenario where a block error rate, BLER, of a downlink channel from the apparatus to the user equipment is above a threshold. According to some examples, the method comprises: receiving the plurality of SGCS values and the plurality of MSE values from the user equipment, wherein a cause of Channel State Information, CS1, loss due to Uplink Control Information, UC1, failure is associated with a scenario where BLER is above the threshold but the MSE values do not increase over a threshold and the SGCS values do not decrease under a threshold during a time period corresponding to the BLER increasing above the threshold.

[0089] According to some examples, the method comprises: receiving the plurality of SGCS values and the plurality of MSE values from the user equipment, wherein the means are further configured for: sending, based on the MSE and the SGCS values and to the user equipment, a request to report state machine status of a recurrent quantizer of the user equipment, receiving the state machine status from the user equipment; synchronizing a state machine at the apparatus based on the state machine status of the user equipment.

[0090] According to some examples, the method comprises: receiving the plurality of SGCS values and the plurality of MSE values from the user equipment, wherein a cause of CS1 encoder inference degradation and / or channel drift is associated with a scenario where the MSE values do not increase over a threshold but the SCGS values do decrease under a threshold during a time corresponding to the BLER increasing above the threshold.

[0091] According to some examples, the method comprises: receiving the plurality of SGCS values and the plurality of MSE values from the user equipment, wherein a cause of recurrent quantizer tracking degradation is associated with scenario where the MSE values do increase over a threshold but the SCGS values do not decrease under a threshold during a time corresponding the BLER increasing above the threshold.

[0092] According to some examples, the method comprises: when the cause is identified as increased channel variation and / or non-continuous downlink transmission, sending a request to the user equipment to use a scalar quantization scheme without recurrent tracking. According to some examples, the method comprises: receiving the plurality of SGCS values and the plurality of MSE values from the user equipment, wherein the means are further configured for: when the BLER of a downlink channel from the apparatus to the user equipment is above a threshold, sending a request for a log of the SGCS values and the MSE values; receiving the log from the equipment.

[0093] According to some examples, the plurality of SGCS scores provide a metric of similarity between a target CS1 and a reconstructed target CS1 at the apparatus.

[0094] According to an aspect of the invention, there is provided an apparatus comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform: sending, to a user equipment, at least one reference signal; receiving, from the user equipment, a plurality of quantized outputs determined based on the at least one reference signal; receiving, from the user equipment, a plurality of Squared Generalized Cosine Similarity, SGCS, scores associated with the quantized outputs and / or a plurality of Mean Squared Error, MSE, values between each of the plurality of quantized outputs and a corresponding input used by the user equipment to determine the corresponding quantized output.

[0095] According to some examples, a cause of performance degradation is associated with a scenario where a block error rate, BLER, of a downlink channel from the apparatus to the user equipment is above a threshold.

[0096] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: receiving the plurality of SGCS values and the plurality of MSE values from the user equipment, wherein a cause of Channel State Information, CS1, loss due to Uplink Control Information, UC1, failure is associated with a scenario where BLER is above the threshold but the MSE values do not increase over a threshold and the SGCS values do not decrease under a threshold during a time period corresponding to the BLER increasing above the threshold. According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: receiving the plurality of SGCS values and the plurality of MSE values from the user equipment, wherein the means are further configured for: sending, based on the MSE and the SGCS values and to the user equipment, a request to report state machine status of a recurrent quantizer of the user equipment, receiving the state machine status from the user equipment; synchronizing a state machine at the apparatus based on the state machine status of the user equipment.

[0097] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: receiving the plurality of SGCS values and the plurality of MSE values from the user equipment, wherein a cause of CS1 encoder inference degradation and / or channel drift is associated with a scenario where the MSE values do not increase over a threshold but the SCGS values do decrease under a threshold during a time corresponding to the BLER increasing above the threshold.

[0098] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: receiving the plurality of SGCS values and the plurality of MSE values from the user equipment, wherein a cause of recurrent quantizer tracking degradation is associated with scenario where the MSE values do increase over a threshold but the SCGS values do not decrease under a threshold during a time corresponding the BLER increasing above the threshold.

[0099] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: when the cause is identified as increased channel variation and / or non-continuous downlink transmission, sending a request to the user equipment to use a scalar quantization scheme without recurrent tracking.

[0100] According to some examples, when executed by the at least one processor, the instructions cause the apparatus at least to perform: receiving the plurality of SGCS values and the plurality of MSE values from the user equipment, wherein the means are further configured for: when the BLER of a downlink channel from the apparatus to the user equipment is above a threshold, sending a request for a log of the SGCS values and the MSE values; receiving the log from the equipment.

[0101] According to some examples, the plurality of SGCS scores provide a metric of similarity between a target CSI and a reconstructed target CSI at the apparatus.

[0102] According to an aspect of the invention, there is provided a computer program product embodied on a distribution medium readable by a computer and comprising program instructions which, when loaded into an apparatus, execute a method comprising: sending, to a user equipment, at least one reference signal; receiving, from the user equipment, a plurality of quantized outputs determined based on the at least one reference signal; receiving, from the user equipment, a plurality of Squared Generalized Cosine Similarity, SGCS, scores associated with the quantized outputs and / or a plurality of Mean Squared Error, MSE, values between each of the plurality of quantized outputs and a corresponding input used by the user equipment to determine the corresponding quantized output.

[0103] Some embodiments of the invention are defined in the dependent claims.

[0104] LIST OF THE DRAWINGS

[0105] In the following, the invention will be described in greater detail with reference to the embodiments and the accompanying drawings, in which:

[0106] Fig. 1 shows an example of a communication network to which examples disclosed herein may be applied;

[0107] Fig. 2 shows an example architecture for two-sided CSI compression with recurrent quantization;

[0108] Fig. 3 shows an example graph for performance of different compression schemes;

[0109] Fig. 4 shows example performance of different quantizers over time;

[0110] Fig. 5 shows a flow diagram of a recurrent quantizer and an inverse quantizer;

[0111] Fig. 6 shows a measure of similarity between a CSI encoder and CSI decoder output as a function of CSI message drop probability;

[0112] Fig. 7A shows a first part of a signalling diagram that can be used in an initial “transient” phase of an example network;

[0113] Fig. 7B shows a second part of a signalling diagram that can be used in an initial “transient” phase of an example network;

[0114] Fig. 8 shows a signalling diagram that can be used in a second “tracking” phase of an example network;

[0115] Fig. 9 shows a signalling diagram that can be used to diagnose issues in a network;

[0116] Fig. 10 shows an example performance monitoring procedure;

[0117] Fig. 11 shows an example of a method;

[0118] Fig. 12 shows an example of a method;

[0119] Fig. 13 shows an example of a method;

[0120] Fig. 14 shows an example of a method;

[0121] Fig. 15 shows an example of an apparatus.

[0122] DESCRIPTION OF EMBODIMENTS

[0123] The following embodiments are exemplary. Although the specification may refer to “an”, “one”, or “some” embodiment(s) in several locations of the text, this does not necessarily mean that each reference is made to the same embodiment's), or that a particular feature only applies to a single embodiment. Single features of different embodiments may also be combined to provide other embodiments. Further, when a particular feature, structure, or characteristic is described in connection of an embodiment, it is within the knowledge of one skilled in the art to apply such feature, structure, or characteristic in connection with other embodiments whether or not explicitly described. It shall be understood that although the terms “first,” “second” and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another.

[0124] For the purposes of the present disclosure, the phrases “at least one of A or B”, “at least one of A and B”, and “A and / or B” means (A), (B), or (A and B). For the purposes of the present disclosure, the phrase “A, B, and / or C” means (A), (B), (C), (A and B), (A and C), (B and C), or (A, B, and C).

[0125] Embodiments described may be implemented in a communication network, such as any of the following radio access technologies (RATs): Worldwide Interoperability for Micro-wave Access (WiMAX), Global System for Mobile communications (GSM, 2G), GSM EDGE radio access Network (GERAN), General Packet Radio Service (GRPS), Universal Mobile Telecommunication System (UMTS, 3G) based on basic wideband-code division multiple access (W-CDMA), high-speed packet access (HSPA), Long Term Evolution (LTE), LTE-Advanced, and enhanced LTE (eLTE), 5G (also called NR), or any future RAT such as 6G. Moreover, communication within the communication network may utilize any proper wireless communication technology, comprising but not limited to: Code Division Multiple Access (CDMA), Frequency Division Multiple Access (FDMA), Time Division Multiple Access (TDMA), Frequency Division Duplex (FDD), Time Division Duplex (TDD), Multiple-Input Multiple-Output (M1M0), Orthogonal Frequency Division Multiple (OFDM), and / or Discrete Fourier Transform spread OFDM (DFT-s-OFDM).

[0126] As used herein, the term “network device” or “network node” refers to a node in a communication network via which user equipment may access the network and / or which is capable of controlling radio communication and managing radio resources within a cell. The network node or network device may be referred to as a base station (BS), an access point (AP) or an access node. The network device may be, depending on the applied technology, for example, a node B (NodeB or NB), an evolved NodeB (eNodeB or eNB), an NR NB (also referred to as a gNB), a Remote Radio Unit (RRU), a radio head (RH), a remote radio head (RRH), a relay, an Integrated Access and Backhaul (1AB) node, a low power node, a non-terrestrial network (NTN) or non-ground network device such as a satellite network device, a low earth orbit (LEO) satellite and a geosynchronous earth orbit (GEO) satellite, or an aircraft network device.

[0127] Moreover, in connection of split radio access network (RAN), the network device may refer to a centralised unit (CU) of a base station and / or a distributed unit (DU) of a base station. An interface between CU and DU may be referred to as an Fl interface in NR. In the split RAN architecture, node operations may be carried out, at least partly, in the central / centralized unit, CU, (e.g. server, host or node) operationally coupled to the DU, (e.g. a radio head / node). One CU may control one or more DUs, acting at least as transmit / receive (Tx / Rx) nodes. In some embodiments, the DUs may comprise e.g. a radio link control (RLC), medium access control (MAC) layer and a physical (PHY) layer, whereas the CU may comprise the layers above RLC layer, such as a packet data convergence protocol (PDCP) layer, a radio resource control (RRC) and an internet protocol (IP) layers. Other functional splits are possible too. In practice, any processing task may be performed in either the CU or the DU and the boundary where the responsibility is shifted between the CU and the DU may depend on the applied implementation.

[0128] The term “terminal device” refers to any end device that may be capable of wireless communication. By way of example, a terminal device may be referred to as a communication device, user equipment (UE), a Subscriber Station (SS), or a Mobile Station (MS). The terminal device may include a mobile phone, a cellular phone, a smart phone, voice over IP (VoIP) phones, wireless local loop phones a tablet, a wearable terminal device, a personal digital assistant (PDA), portable computers, desktop computer, image capture terminal devices such as digital cameras, gaming terminal devices, music storage and playback appliances, vehicle-mounted wireless terminal devices, USB dongles, an Internet of Things (loT) device, a watch or other wearable, a head-mounted display (HMD), a vehicle, a drone, a medical device and applications (e.g., remote surgery), an industrial device and applications (e.g., a robot and / or other wireless devices operating in an industrial and / or an automated processing chain contexts), a consumer electronics device, a device operating on commercial and / or industrial wireless networks, and the like.

[0129] A term “resource”, as used herein, may refer to radio resources in time domain, in frequency domain, in space domain, and / or in code domain. Some examples of resources include e.g. a physical resource block (PRB), a radio frame, a subframe, a time slot, a subband, a frequency region, a sub-carrier, a beam, etc. The term “transmission” and / or “reception” may refer to wirelessly transmitting and / or receiving via a wireless propagation channel on radio resources. Fig. 1 illustrates an example of a communication network to which examples disclosed herein may be applied. The communication network or a cellular communication network may comprise a network node 110 providingone or more cells, such as cell 100, and a network node 112 providing one or more other cells, such as cell 102. Each cell may be, e.g., a macro cell, a micro cell, femto, or a pico cell, for example. The cell may define a coverage area or a service area of the corresponding access node.

[0130] The network node 110 may provide a user equipment (UE) 120 (one or more UEs) with wireless access to the communication network. The wireless access may comprise downlink (DL) communication from the network node to the UE 120 and uplink (UL) communication from the UE 120 to the network node. Examples of uplink channels comprise physical uplink control channel (PUCCH) for transmitting control information and physical uplink shared channel (PUSCH) for transmitting data towards the network. Examples of downlink channels comprise physical downlink control channel (PDCCH) for transmitting control information and physical downlink shared channel (PDSCH) for transmitting data towards the user equipment.

[0131] There may be a plurality of UEs 120, 122 in the system. Each of them may be served by the same or by different network nodes 110, 112. UE may be configured with dual connectivity (DC), wherein the UE, e.g. UE 120, may be connected to multiple network nodes 110, 112. The UEs 120, 122 may communicate with each other, in case device-to-device (D2D) communication interface is established between them via a so-called sidelink (SL). Such D2D communications may be referred to as machine-to-machine, peer-to-peer (P2P) communications, or ve- hicle-to-vehicle (V2V), for example.

[0132] In the case of multiple network nodes in the communication network, the network nodes may be connected to each other via an interface. LTE specifications call such an interface as X2 interface. An interface between an LTE node and a 5G node, or between two 5G nodes may be called Xn interface.

[0133] The network nodes 110 and 112 may be further connected via another interface to a core network 116 of the communication network. The LTE specifications specify the core network as an evolved packet core (EPC), and the core network may comprise e.g. a mobility management entity (MME) and a gateway node. The MME may handle mobility of terminal devices in a tracking area encompassing a plurality of cells and handle signalling connections between the terminal devices and the core network. The gateway node may handle data routing in the core network and to / from the terminal devices. The 5G specifications specify the core network as a 5G core (5GC). The 5G core may comprise e.g. an access and mobility management function (AMF) and a user plane function / gateway (UPF) and other functions. The AMF may handle termination of non-access stratum (NAS) signalling, NAS ciphering & integrity protection, registration management, connection management, mobility management, access authentication and authorization, security context management. The UPF node may support packet routing and forwarding, packet inspection and quality of service (QoS) handling, for example.

[0134] Some examples enable energy saving in communication networks energy (e.g., mobile networks) especially in the RAN (e.g. 5G or 6G access network, such as a gNB or other base station), which typically has the largest energy consumption in the entire network (~80% for 5G or 6G). Previous 3GPP releases of 5G have addressed energy saving primarily for the devices in the network, rather than in the network itself. A mobile network can reduce its energy consumption by leveraging low power states / modes and features to control the usage of these states, which entail the switch off of certain Hardware (HW) components and may result into the reduction of the overall network capacity and offered throughput.

[0135] Rl-2406589, "Al / ML for CS1 Compression," Nokia, 3GPP TSG RAN WG1 #118, Maastricht, Netherlands, August 19th - 23rd, 2024 and Rl-2408547, "Al / ML for CS1 Compression," Nokia, 3GPP TSG RAN WG1 #118-bis, Hefei, China, October 14th - 18th, 2024 discuss the concept of a recurrent quantizer being used in a network. Fig. 2 shows an example architecture for a two-sided CS1 compression with recurrent quantization.

[0136] A Space Frequency (SF) encoder 230 and recurrent quantizer 232 can be provided at one side (e.g., downlink at a UE). A recurrent inverse quantizer 234 and SF decoder can be provided at a different side (e.g., uplink at a network node). SF encoder 230 can be considered to comprise a history-independent CSI encoder. Other history-independent CSI encoders maybe used at 230 (not just SF encoders). The input of the history-independent CSI encoder may not have temporal correlation, but may have correlation over other dimensions (e.g., space, frequency). Decoder 236 may be similar.

[0137] A target CSI or CSI matrix Vt may be mapped to a latent space representation (e.g., a latent vector zt) by history-independent CSI encoder 230. A recurrent quantizer 232 can then be used to encode a sequence of latent vectors efficiently, taking advantage of correlation in time between consecutive latent vectors. Recurrent quantizer 232 may be history-dependent. Recurrent inverse quantizer 234 can also take advantage of correlations in time between consecutive latent vectors to efficiently decode a bit vector bt output by recurrent quantizer 232. Estimated latent vector zt can then be decoded by history-independent decoder 236 to provide an estimate (reconstructed) target CSI (e.g., target CSI matrix Pt). When it comes to its performance in terms of squared generalized cosine similarity (SGCS) between input into the CSI encoder and the output from the CSI decoder, it has been shown that a recurrent quantizer-empowered scheme (“Scheme C” 302c or “Scheme D” 302d in Fig. 3) can achieve the equivalent performance of an SF encoder / decoder with static uniform quantization (“Scheme A” 302a in Fig. 3) while using only 50% of the feedback overhead. This is shown in Rl-2408547, "AI / ML for CSI Compression," Nokia, 3GPP TSG RAN WG1 #118-bis, Hefei, China, October 14th - 18th, 2024. A scheme (“Scheme B” 302b) with an SFT encoder / decoder and static uniform quantization is also shown in Fig. 3. Two kinds of recurrent quantization scheme have been proposed, an ML-based approach (ML-RQ), and explicitly designed approach (ED-RQ).

[0138] Fig. 3 shows SGCS performances versus overhead bits of Scheme A-D, showing that time domain CSI compression together with SF encoding could achieve 50% of overhead reduction compared to spatial-frequency domain compression only.

[0139] As discussed above, recurrent quantization (RQ) has a capability of exploiting temporal correlations at the latent domain. A recurrent quantizer at UE side takes each dimension (in scalar format) of the latent domain as its input (unquantized raw / floating point value) and produces its encoding outcome (typically 1 or 2 bits / dimension). Both encoder (recurrent quantizer) and decoder (recurrent inverse quantizer or recurrent de-quantizer) run its own quantizer state machine to keep track to exploit temporal correlation of input sequence. Recurrent de-quan- tizer at NW side can compute a de-quantized value by taking encoded bit (sequence) and referring to the quantizer state machine.

[0140] An example is shown in Fig. 5. A recurrent scalar quantizer comprises a quantizer state machine 240 and delay 242. A recurrent inverse scalar quantizer comprises an inverse quantizer state machine 244 and delay 246.

[0141] An RQ scheme uses a transient phase to narrow down its estimation bounds with a limited number of encoded bits at the beginning. An example method for performing the narrowing of estimation bounds is discussed in Section 2.1.3 of Rl-2408547, "Al / ML for CS1 Compression," Nokia, 3GPP TSG RAN WG1 #118-bis, Hefei, China, October 14th - 18th, 2024. This implies that its performance may be sub-optimal at the initial transient phase before it becomes stabilized for tracking, as can be observed in Fig. 4. Fig. 4 shows that performance of the explicitly designed recurrent quantization scheme (in Fig. 4, denoted as “uncertainty tracking”) can converge to that of the conventional scheme with benefit of 1 extra bit per dimension, i.e., MSE of the uncertainty tracking with 1 bit / dim (404a) converges to that of conventional uniform scalar quantization with 2 bits / dim, and uncertainty tracking with 2 bits / dim (404b) converges to conventional with 3 bits / dim. However, uncertainty tracking requires the initial transient phase of 3~4 time steps, which is 15~20 ms for the demonstrated case. During this initial transient phase, quantization error is more significant and might be considered nontrivial, and can consequently lead to possible end to end (E2E) performance degradation. Once this initial transient phase is over, the recurrent quantizer is stabilized and enters the “steady / tracking phase”.

[0142] Further, E2E performance degradation is also expected in case of lost CS1 due to Uplink Control Information (UC1) decoding failure, due to its adversary impact on synchronization between two quantizer state machines (one at UE-side, the other at NW-side; e.g., in Fig. 5 where the recurrent scalar quantizer is on the UE-side and the recurrent inverse scalar quantizer is on the NW-side). As depicted in Fig. 6, E2E SGCS performance of the explicitly designed recurrent quantizer (ED- RQ) seems more sensitive to CS1 drop probability, whereas history-independent quantization (H1Q) scheme shows slower degradation over increased CS1 drop probability. Recurrent quantization-adopted scheme shows rapid degradation over increased CS1 drop probability, which is possibly owing to increased devia- tion / error propagation between two quantizer state machines. Fig. 6 shows SGCS between CS1 encoder input and decoder output as a function of CS1 message drop probability, for history-independent quantization (dashed lines) and recurrent quantization (solid lines). Models with overheads of approximately 64 bits, 128 bits, 240 bits and 244 bits are represented. Recurrent quantization models with more bits have higher SGCS values at a given drop probability than recurrent quantization models with fewer bits. History-independent quantization models with more bits have higher SGCS values at a given drop probability than history-independent quantization models with fewer bits.

[0143] As shown in Fig. 4, in some examples a transient mode criterion MSE level (“a first threshold level”) may be provided as well as a second threshold MSE level. The transient mode criterion (the first threshold level) is greater than the second threshold MSE level. The transient mode criterion may be considered to comprise a first threshold. A UE may operate in a first phase (the transient phase) when the MSE is greater than the second threshold level. When MSE is less that the second threshold level, the UE may operate in a steady / tracking phase. Further, the UE may operate in two different modes during the transient phase. The selection of these modes can be based on the transient mode criterion level (the first threshold level), in some examples. For example, when MSE>transientmode criterion level, the UE may operate in a performance driven mode, as the expected performance loss would otherwise be too great. When second threshold<MSE<tran- sient mode criterion (in other words, when second threshold<MSE<first threshold), the UE may operate in an overhead-limited mode. In the overhead-limited mode, the performance loss is considered acceptable, so modest performance degradation can be tolerated, and overhead signalling can be reduced.

[0144] Some examples provide methods for addressing sub-optimal E2E performance in the initial phase (transient phase) when deploying the recurrent quantizer-adopted AlML-enabled CS1 compression, which are:

[0145] Some examples provide methods for addressing error propaga- tion / Synchronization deviation between quantizer state machines at UE-side and at NW-side (due to CS1 drop) when deploying the recurrent quantizer-adopted AlML-enabled CS1 compression. Some examples provide a method to detect the error propagation / synchronization deviation, and some example methods identify a cause of the error propagation / synchronization deviation such that an appropriate corrective action can be performed.

[0146] Some examples provide an operational framework and associated signalling to support recurrent quantizer adopted AlML-enabled CS1 compression.

[0147] In an initial transient phase, there is provided two possible cases: an overhead-limited case and a performance-driven case. A mechanism is also provided for detecting the end of the initial transient phase and switching to a steady / tracking phase.

[0148] In the overhead-limited case for the initial transient phase, sub-optimum E2E performance is allowed by letting the UE report a small amount of feedback which is output to the recurrent quantizer to provide efficient signalling overhead.

[0149] In the performance driven case for the initial transient phase, an achievable E2E performance is obtained by ensuring that a UE reports both the recurrent quantizer output and a conventional full-resolution scalar quantization outcome for the sake of full E2E performance (at the cost of increased feedback overhead in comparison to the overhead-limited case). The encoded outcome of the recurrent quantizer is used for recurrent de-quantizer state machine update during the transientphase, while the full-resolution feedback information is used to guarantee full E2E CS1 reconstruction performance. Alternatively, UE can report conventional full-resolution scalar quantization outcome without the encoded outcome of the recurrent quantizer to save feedback overhead, and state machine status of the recurrent quantizer can be shared by UE with NW on request from NW.

[0150] At the UE, Mean Squared Error can be determined by taking between an input into the recurrent quantizer and the output of the recurrent quantizer. This can be used for determination of quantizer stability in comparison with pre-determined threshold value (the second threshold). This pre-determined threshold value can be acquired offline via simulations and / or field test campaign. When the MSE is below the second threshold, the recurrent quantizer can determine that the quantizer is stable and that a possible impact of distortion caused by the recurrent quantizer to E2E CS1 compression-reconstruction performance can be considered negligible (a “steady / tracking phase” of the recurrent quantizer).

[0151] In the steady / tracking phase of the recurrent quantizer, some examples described herein provide a performance monitoring framework. A root cause identification scheme may also be provided to deal with CS1 loss / CSl encoder inference performance degradation and / or recurrent quantizer tracking capability degradation. Some examples provide a method for the NW to take a counter measure depending on the identified root case.

[0152] In the steady / tracking phase., the UE can compute MSE between the input and output of the recurrent quantizer. The UE can also estimate SCGS as an E2E metric which takes input to CS1 encoder (“target CS1” at UE) and output from CS1 decoder (“reconstructed target CS1” at NW). As the UE usually does not have an access to CS1 decoder model, the UE could have either a proxy CS1 decoder model or direct SGCS estimator to estimate the SCGS.

[0153] The NW can command the UE to report MSE and SGCS estimate along with CS1 feedback to NW or ask UE to archive them for possible future reporting for diagnosis at the event of E2E performance degradation (increased BLER, for example).

[0154] The NW can identify a root cause of E2E performance degradation by examining UE-reported MSE and the UE-reported SCGS estimate (discussed further below). After identifying a root cause, the NW can take a corresponding informed decision / counter measure upon identification of the root cause. For example, when the identified root cause is UC1 loss, the NW can initiate synchronization of the quantization state machines at UE-side and NW-side. In a further example, when the identified root cause is increased channel variation or sporadic (non-con- tinuous) DL transmission, the NW can switch to a non-recurrent (history-independent) quantization scheme.

[0155] Fig. 7A shows an example method for the initial transient phase discussed above.

[0156] Initially, during a preparation phase 701, NW 710 (e.g., at least one network node) may send a capability enquiry to UE 720 at 703. At 705, UE 720 may provide capability information, including support for recurrent quantization. At 707, NW 710 may provide to UE 720 a UE configuration for recurrent quantization. This may comprise a configuration for the transient phase The configuration may be based on UE capability information received at 705, for example. NW 710 may inform UE 720 of parameter “transient_mode_criterion”, which is to be used as a criterion for a determination by UE 720 of a transient phase mode of operation (e.g., a decision to operate using the overhead-limited mode when MSE is less that the first threshold (i.e., less than the transient mode criterion) but greater than the second threshold for the steady / tracking phase, or a decision to operate in a performance driven mode when MSE is greater than the transient mode criterion level). At 707, NW 710 may also provide a configuration for a full resolution quantization to UE 720.

[0157] To allow for training of the encoder model at the UE-side, NW 710 transfers to UE 720 datasets of Target CSI and CSI feedback corresponding to the input and output of the NW-side encoder model. Alternatively, NW 710 may transfer to the UE-side the NW-side encoder parameters for a given model structure, and the Target CSI used for training the NW-side encoder structure. In both cases, some additional information may be needed from NW-side to UE-side to align the two sides with the same recurrent quantizer. In some examples, the additional information may contain a quantization codebook and other parameters needed to describe the state machine used in the recurrent quantizer adopted by NW 710.

[0158] A transient phase 709 is shown in Fig. 7A where MSE of the quantization output with respect to the input to the quantizer (i.e., output of the CSI compression / encoder) is greater than a pre-defined second threshold. The MSE can be measured at UE 720. In some examples, MSE calculation (e.g.., at 715) can be performed over multiple instances to provide statistical stability. The pre-defined second threshold and / or transient mode criterion may be acquired via numerical simulation and / or lab / field test campaign for various radio propagation environments / configurations. In an example, the pre-defined second threshold and / or transient mode criterion can be defined in the specification. In another examples, the pre-defined second threshold and / or transient mode criterion can be configured by NW 710 to UE 720. In another examples, the pre-defined second threshold and / or transient mode criterion can be aligned between the UE vendor and NW vendor prior to cell deployment.

[0159] For the transient phase, the overhead limited mode is shown at 719 to 731 and a performance-driven case is shown at 733 to 747.

[0160] At 711, at least one CS1 Reference Signal (CS1-RS) is sent from NW 710 to UE 720. At 713, channel estimation is performed and target CS1 is extracted. The history-independent CS1 encoder can then be used to provide latent vectors. These latent vectors can be input into a history-dependent recurrent quantizer at 715 to provide a quantized output. The MSE can also be calculated at 715, and in some examples can be logged. At 717, UE 720 can monitor whether the MSE is above or below the transient mode criterion for selecting the overhead-limited mode or the performance driven mode for the transient phase. UE 720 can also determine whether MSE is below the second threshold (which is less than the first threshold) for transitioning to the steady / tracking phase.

[0161] The overhead limited case tolerates sub-optimum E2E performance. UE 720 may only report a small amount of feedback which is output of the recurrent quantizer. When UE 720 is configured with or only supports the overhead-limited case, the E2E CS1 reconstruction performance may be sub-optimum, as the recurrent quantizer is in a transient phase. The required feedback overhead in the air (e.g., at 729) is small. Applying the overhead limited case to Fig. 5 bt bit per dimension is small.

[0162] Overhead-limited mode: At 719, based on the determination at 717, UE 720 determines to operate in the overhead-limited mode. At 719a, UE 720 may send a notification of transient phase operation mode (the overhead-limited mode) to NW 710. CS1 feedback (the quantized output) may be sent from UE 720 to NW 710 at 721. A recurrent de-quantizer can de-quantize the quantized output sent at 723, and CS1 reconstruction can then be performed using a history-independent CS1 decoder at 725. At 729, a physical downlink channel (PDSCH) transmission can be sent from NW 710 to UE 720, and at 731 an acknowledgement (or negative acknowledgement) is returned to indicate successful (or unsuccessful) decoding of PDSCH.

[0163] Performance-driven mode: This case guarantees an achievable E2E performance even during the initial transient phase by making UE report not only the recurrent quantizer output (bt bit / dim (dim=dimension)) but also conventional history-independent full-resolution scalar quantization outcome (n bits / dim), providing full E2E performance but of increased feedback overhead. At 733, based on the determination at 717, UE 720 determines to operate in the performance- driven mode. At 734, UE 720 uses a full-resolution (scalar) quantizer to provide full-resolution feedback. At 735, UE 720 may send a notification of transient phase operation mode (the performance-driven mode) to NW 710. The encoded outcome of the recurrent quantizer bt bit / dim) is used at 741 for recurrent de-quan- tizer state machine update during the transient phase, while the full-resolution feedback information (n bits / dim) is used at 741 to guarantee full E2E CS1 reconstruction performance. The full-resolution information (full-resolution quantization output) from UE 720 may be generated at UE inputting the at least one reference signal into a channel estimator, then optional pre-processing, then inputting into a history-independent CS1 encoder and then a scalar quantizer at 734 Alternatively, UE 720 send only the full-resolution feedback information (n bits / dim) (such as at 739), and the synchronization of the recurrent quantizer can be performed explicitly (755 in FIG. 7B) when UE is determined to operate in steady / tracking phase. In this case, constant update (745) of the state-machine of the recurrent quantizer at NW 710 is not required.

[0164] At 742, a recurrent de-quantizer can de-quantize the quantized output sent at 741 vis using bt bit / dim feedback information, and update of the state machine of the recurrent de-quantizer can then be performed. At 742, a history-independent full-resolution de-quantizer performs de-quantization of the quantized output sent at 741 via using n bits / dim feedback information. CS1 reconstruction can then be performed at 743. The recurrent de-Quantizer at NW 710 can use bt bit / dim feedback information to update its state-machine to keep it synchronised with the state machine of the quantizer at UE 710. At 746, a physical downlink channel (PDSCH) transmission can be sent from NW 710 to UE 720, and an acknowledgement (or negative acknowledgement) is returned at 747 to indicate successful (or unsuccessful) decoding of PDSCH.

[0165] At 715, UE 720 calculates MSE to monitor whether and when the MSE falls below a pre-defined second threshold level. Upon detection of an event where MSE < second threshold (e.g., at 749 of Fig. 7B), UE 720 can report this to NW 710 at 751 such that NW should be aware that UE / NW's recurrent quantizer is entering “Steady / Tracking phase”.

[0166] Unlike in the overhead-limited case, in the performance driven case NW 710 may determine to proceed to the “steady / tracking phase mode of operation (at 753 of Fig. 7B) and send a corresponding command to transition to the steady phase at 763. This can ensure that both UE 720 and NW 710 are able to switch mode of quantizer operation from conventional history-independent scalar quantization to recurrent (history-dependent) quantization in a synchronized manner, as at 765.

[0167] For the performance driven case, in an alternative embodiment (“Altl.0”, 735 and 757 to 761), feedback overhead can be reduced during the initial transient phase, by reporting only the conventional history-independent full-resolution scalar quantization outcome (n bits per dimension). UE 720 monitors the MSE and upon detection of a pre-configured event (MSE<second threshold), UE 720 reports the current state of the state machine of the recurrent quantizer when reporting this event (751). The reporting of the current state of the state machine indicates to NW 710 that UE 720 has entered a “steady / tracking phase” and that subsequent reports will only contains the recurrent quantizer output (bt bits per dimension). By using the reported current state of the state machine, NW 710 is able to align the state of the recurrent de-quantizer with that of the recurrent quantizer at UE 720, for correct decoding of the recurrent quantizer output. Alternatively, synchronization of state machine status at 755 can be triggered by NW 710 upon decision to proceed to Steady Phase at 753.

[0168] In another example that may apply to either the overhead-limited case or the performance-driven case, UE 720 may report the MSE as well as the CS1 feedback at 717, 735 to 737. In this example, the detection of the event where MSE is less than a predetermined second threshold (shown at 749 in Fig. 7B) can take place at NW 710 instead ofatUE 720. In this example reporting of the event at 751 from UE 720 to NW 710 is not required.

[0169] At 765 of Fig. 7B, MSE is less than the predetermined second threshold and the method proceeds to the method of Fig. 8.

[0170] Figure 8 shows a method for a steady / tracking phase of the recurrent quantizer. At 867, at least one CS1 Reference Signal (CS1-RS) is sent from NW 810 to UE 820. At 869, channel estimation is performed and target CS1 is extracted. The history-independent CS1 encoder can then be used to provide latent vectors. These latent vectors can be input into a history-dependent recurrent quantizer at 871 to provide a quantized output. The MSE can also be calculated at 871, and in some examples can be logged.

[0171] In the example of Fig. 8, block error rate (BLER) can be used a main metric for performance monitoring, while using MSE values and SGCS estimates as parameters for diagnosis of root-cause identification. BLER can be used as a KPI which indicates whether precoded PDSCH can be decoded successfully at UE-side or not. As long as BLER is under a certain target level, the quality of the reconstructed CS1 can be considered acceptable / under control. On the other hand, MSE and SGCS (which are intermediate KPIs and are be reported by UE 820) can be useful when NW 810 tries to identify the root cause of E2E performance degradation.

[0172] In one example (“AltO”, 873 to 883), UE 820 reports its CS1 feedback with its computed MSE (calculated from input / output of the recurrent quantizer) and SGCS estimate (which can be acquired by Al / ML-assisted SGCS estimator or proxy decoder, both of which can be run at UE 820). A recurrent de-quantizer and CSI decoder can be used at 875 and 877 for the CSI feedback. PDSCH transmission may take place at 879, and acknowledgement and / or negative acknowledgement is performed at 881. Performance monitoring can be performed by NW 810 at 883. For examples BLER monitoring of the PDSCH can be performed.

[0173] In an alternative example, (“Altl”, 885 to 897), NW 810 may operate the performance monitoring procedure such that UE 820 stores its computed MSEs and SGCS estimates at 885. CSI feedback is reported at 887 to NW 810. At 889 and 891, A recurrent de-quantizer and CSI decoder can be used at 889 and 891 for the CSI feedback. PDSCH transmission may take place at 893, and acknowledgement and / or negative acknowledgement is performed at 895. Performance monitoring can be performed by NW 810 at 897. For examples BLER monitoring of the PDSCH can be performed.

[0174] Detection of E2E performance degradation can be performed by monitoring BLER of the PDSCH. CSI acquisition, reporting of CSI feedback and precoding of PDSCH may continue at 898. At 899, when NW 810 detects that BLER is above a predetermined threshold, NW 810 can trigger a diagnosis operation (shown in Fig. 9) to try to identify the root cause of degradation, and potentially take a corresponding action. In the example of Alt 1 (885 to 897) where MSE and SGCS estimates are logged with a time stamp at 885, but not sent to NW 810, NW 820 may determine to request the log for the diagnosis at 899b. At 899c, a request for the log information is sent to UE 820 from NW 810. At 899d, UE 820 fetches the history of MSE and SGCS with corresponding timestamps, and at 899e UE 820 reports the log info to NW 810. At 899f, NW 820 may diagnose the cause of root cause of the performance degradation (e.g., the root case for BLER being above a threshold). This is discussed further below with respect to Fig. 9.

[0175] In the examples method of Fig. 9, NW 920 may use MSE and / or SGCS values reported by UE 920 to determine a root cause of E2E performance degradation (e.g., a root cause of why BLER is above a threshold).

[0176] In an example, NW 920 may be able to identify a root case of performance degradation using the logic in Table 1. Note that the logic in Table can be further augmented by considering NW-side originated issues (e.g., CS1 decoder interference failure, etc.)

[0177] Table 1: Possible root causes and corresponding expected changes of the selectee metrics

[0178] In one example, NW 820 can identify UC1 loss as a root cause and then take an appropriate measure (CaseO, 62 to 71). At 62, NW 920 identifies UC1 loss as a root case, and NW 920 commands UE 910 to report state machine status (e.g., latest step size with time stamp, frame / sub frame number, etc.) of the recurrent quantizer at 63. At 64, NW 920 may synchronise its state machine with this report. An indication of the state machine synchronization can be sent by NW 920 to UE 910 at 65. At 66, 67, 68, 69 70 and 71, the synchronized recurrent quantizers at UE 920 and 910 can be used to for CS1 compression / reconstruction.

[0179] In another example (Casel, 72 to 81), NW 920 can identify increased channel variation or sporadic (non-continuous) DL transmission which can challenge reliable operation of the recurrent quantizer as a root cause. At 72, NW 920 commands UE 910 to switch the quantizer scheme to the conventional one, e.g., scalar quantization without recurrent tracking feature, etc. Confirmation of this change is sent at 73. At 74, 75, 76, 77, 78 , 79 and 81 the conventional quantizer scheme is used for CS1 compression / reconstruction.

[0180] Fig. 10 shows an example performance monitoring procedure between a NW and a UE. This is shown for NW 810 and UE 820 but could also apply to NW 710 and UE 720 or NW 910 and UE 920. Fig. 10 shows a performance monitoring procedure and associated NW / UE interaction for the Al / ML-enabled CS1 compression with recurrent quantizer (enc: CS1 encoder / compressor, dec: CS1 decoder / re- constructor, RQ: recurrent quantizer, RdQ: recurrent de-quantizer, Rx Proc: receive processing).

[0181] Fig. 11 shows an example method. The method may be performed by a UE, such as UE 720, 820 or 920, for example.

[0182] At 1100, the method comprises receiving, from a network node, at least one reference signal.

[0183] At 1101, the method comprises determining at least one latent vector based on target Channel State Information acquired from the at least one reference signal and a history-independent CS1 encoder.

[0184] At 1102, the method comprises using the at least one latent vector to provide an input into a history-dependent recurrent quantizer to provide a quantized output.

[0185] At 1103, the method comprises calculating a mean squared error between the input and the quantized output.

[0186] At 1104, the method comprises comparing the mean squared error to a threshold level.

[0187] At 1105, the method comprises based on the comparing, determining to operate in a first mode, wherein in the first mode, the apparatus performs: determining at least one further latent vector based on target Channel State Information acquired from the at least one reference signal and the history-independent CS1 encoder; using the at least one further latent vector to provide an input into a scalar quantizer to provide a full-resolution quantization output; sending the full-resolution quantization output to the network node.

[0188] Fig. 12 shows an example method. The method may be performed by a network node, such as network node 710, 810, 910, for example,.

[0189] At 1200, the method comprises sending, to a user equipment, at least one reference signal. At 1201 the method comprises receiving a full-resolution quantization output from a user equipment, wherein the full-resolution quantization output is generated at the user equipment by inputting the at least one reference signal into a channel estimator, a history-independent CS1 encoder and then a scalar quantizer.

[0190] Fig. 13 shows an example method. The method may be performed by a UE, such as UE 720, 820 or 920, for example.

[0191] At 1300, the method comprises i) receiving, from a network node, at least one reference signal.

[0192] At 1301, the method comprises ii) inputting target Channel State Information acquired from channel estimates based on the at least one reference signal into a history-independent CS1 encoder to provide at least one latent vector.

[0193] At 1302, the method comprises hi) using the at least one latent vector to provide an input into a history-dependent recurrent quantizer to provide a quantized output.

[0194] At 1303, the method comprises iv) sending the quantized output to the network node.

[0195] At 1304, the method comprises repeating i) to iv) to provide a plurality of inputs and a corresponding plurality of quantized outputs.

[0196] At 1305, the method comprises sending the plurality of quantized outputs to the network node.

[0197] At 1306, the method comprises sending, to the network node, a plurality of Squared Generalized Cosine Similarity, SGCS, scores and / or a plurality of Mean Squared Error, MSE, values between each of the inputs and the corresponding quantized outputs.

[0198] Fig. 14 shows an example method. The method may be performed by a network node, such as network node 710, 810, 910, for example,.

[0199] At 1400, the method comprises sending, to a user equipment, at least one reference signal.

[0200] At 1401, the method comprises receiving, from the user equipment, a plurality of quantized outputs determined based on the at least one reference signal.

[0201] At 1402, the method comprises .receiving, from the user equipment, a plurality of Squared Generalized Cosine Similarity, SGCS, scores associated with the quantized outputs and / or a plurality of Mean Squared Error, MSE, values between each of the plurality of quantized outputs and a corresponding input used by the user equipment to determine the corresponding quantized output.

[0202] Fig. 15 shows, by way of example, a block diagram of an apparatus 10. The apparatus 10 comprises, for example, at least one processor 12 and at least one memory 14 storing instructions 15 that, when executed by the at least one processor, cause the apparatus 10 at least to perform the method or methods as disclosed herein, and any of the embodiments thereof. In an example, the at least one memory and the instructions (e.g. a computer program code, software), are configured, with the at least one processor, to cause the apparatus 10 to perform the method or methods as disclosed herein, and any of the embodiments thereof.

[0203] A processor 12 may comprise circuitry, or be constituted as circuitry or circuitries, the circuitry or circuitries being configured to perform phases of methods in accordance with example embodiments described herein. As used in this application, the term “circuitry” may refer to one or more or all of the following: (a) hardware-only circuit implementations, such as implementations in only analog and / or digital circuitry, and (b) combinations of hardware circuits and software, such as, as applicable: (i) a combination of analog and / or digital hardware circuit's) with software / firmware and (ii) any portions of hardware processor(s) with software (including digital signal processor(s)), software, and memory(ies) that work together to cause an apparatus, such as a user equipment, to perform various functions) and (c) hardware circuit(s) and or processor(s), such as a microprocessor's) or a portion of a microprocessor(s), that requires software (e.g., firmware) for operation, but the software may not be present when it is not needed for operation. This definition of circuitry applies to all uses of this term in this application, including in any claims. As a further example, as used in this application, the term circuitry also covers an implementation of merely a hardware circuit or processor (or multiple processors) or portion of a hardware circuit or processor and its (or their) accompanying software and / or firmware. The term circuitry also covers, for example and if applicable to the particular claim element, a baseband integrated circuit or processor integrated circuit for a mobile device or a similar integrated circuit in server, a cellular network device, or other computing or network device.

[0204] The memory 14 may be implemented using any suitable data storage technology. The memory may comprise a database for storing data. The memory 14 may be at least in part external to apparatus 10 but accessible to apparatus 10.

[0205] The instructions 15 may be comprised in a computer readable medium or a non-transitory computer readable medium. A term non-transitory, as used herein, is a limitation of the medium itself (i.e. tangible, not a signal) as opposed to a limitation on data storage persistency (e.g. random access memory, RAM, vs. read only memory, ROM).

[0206] For example, the apparatus 10 is a terminal device, such as UE 720 of Fig. 7A / B, UE 820 of Fig. 8 or UE 920 of Fig. 9. As another example, the apparatus is comprised in such a terminal device, e.g. as a chipset configured to control the terminal device. The apparatus 10 may be caused or configured to perform at least the method of Fig. 5 and / or any one or more of the embodiments described.

[0207] As another example, the apparatus 10 is a network node, e.g. 710 of Fig. 7A / B, 810 of Fig. 8 or 910 of Fig. 9, In another embodiment, the apparatus is comprised in such a network node, e.g. as a chipset configured to control the network node.

[0208] The apparatus may comprise one or more entities of any of protocol layers, such as a MAC entity, an RRC entity, an RLC entity, a PDCP entity or a PHY entity. In some embodiments, the entity is configured to perform at least the method of Fig. 8 to 12, and / or any one or more of the embodiments described.

[0209] The apparatus 10 comprises a radio interface 16. The radio interface 16 may provide the apparatus 10 with communication capabilities. The radio interface 16 may comprise a receiver configured to receive information in accordance with at least one cellular or non-cellular standard. The radio interface 16 may comprise a transmitter configured to transmit information in accordance with at least one cellular or non-cellular standard. The receiver may comprise more than one receiver. The transmitter may comprise more than one transmitter. The radio interface 16 may comprise a transceiver configured to receive and transmit information in accordance with at least one cellular or non-cellular standard. The transceiver may comprise more than one transceiver.

[0210] The apparatus 10 may comprise a user interface 18 comprising, for example, at least one of a keypad, a microphone, a touch display, a display, a speaker, etc. The user interface 18 may be used to control the apparatus by the user. The user interface 18 may be external to the apparatus 10. For example, the apparatus 10 may be connected to another device, such as a computer, either via wireless or wired connection, and the apparatus 10 is controlled by the user via the computer.

[0211] In an embodiment, at least some of the processes described herein may be carried out by an apparatus comprising means for carrying out at least some of the described processes. Means for performing method steps as disclosed herein may include software and / or hardware components of the apparatus 10. For example, the at least one processor 12, the memory 14, and the computer program code form means for carrying out the method or methods as disclosed herein, and any of the embodiments thereof. As used herein the term “means” is to be construed in singular form, i.e. referring to a single element, or in plural form, i.e. referring to a combination of single elements. Therefore, terminology “means for [performing A, B, C]”, is to be interpreted to cover an apparatus in which there is only one means for performing A, B and C, or where there are separate means for performing A, B and C, or partially or fully overlapping means for performing A, B, C. Further, terminology “means for performing A, means for performing B, means for performing C” is to be interpreted to cover an apparatus in which there is only one means for performing A, B and C, or where there are separate means for performing A, B and C, or partially or fully overlapping means for performing A, B, C.

[0212] Even though the invention has been described above with reference to an example according to the accompanying drawings, it is clear that the invention is not restricted thereto but can be modified in several ways within the scope of the appended claims. Therefore, all words and expressions should be interpreted broadly and they are intended to illustrate, not to restrict, the embodiment. It will be obvious to a person skilled in the art that, as technology advances, the inventive concept can be implemented in various ways. Further, it is clear to a person skilled in the art that the described embodiments may, but are not required to, be com- bined with other embodiments in various ways.

Claims

46WE CLAIM:

1. An apparatus comprising means for: receiving, from a network node, at least one reference signal; determining at least one latent vector based on target Channel State Information acquired from the at least one reference signal and a history-independent CSI encoder; using the at least one latent vector to provide an input into a historydependent recurrent quantizer to provide a quantized output; calculating a mean squared error between the input and the quantized output; comparing the mean squared error to a threshold level; based on the comparing, determining to operate in a first mode, wherein in the first mode, the apparatus performs: determining at least one further latent vector based on target Channel State Information acquired from the at least one reference signal and the history-independent CSI encoder; using the at least one further latent vector to provide an input into a scalar quantizer to provide a full-resolution quantization output; sending the full-resolution quantization output to the network node.

2. The apparatus according to claim 1, wherein the apparatus determines to operate in the first mode when the mean squared error is above the threshold level.

3. The apparatus according to claim 1 or claim 2, wherein the means are further configured for:47 based on the comparing, determining to operate in a second mode, wherein in the second mode the apparatus performs: sending the quantized output to the network node.

4. The apparatus according to claim 3, wherein the apparatus determines to operate in the second mode when the mean squared error is below the threshold level and above a second threshold level.

5. The apparatus according to any preceding claim, wherein the means are further configured for: determining whether the mean squared error between the input and the quantized output is below a second threshold for a predefined amount of time; when the mean squared error is below the second threshold for a predetermined time, sending a first indication to the network node.

6. The apparatus of claim 5, wherein the means are further configured for: after sending the first indication to the network node , receiving, from the network node, a command to operate in a third mode, wherein in the third mode the apparatus performs: i) receiving, from the network node, one or more reference signals; ii) inputting target Channel State Information acquired from channel estimates based on the one or more reference signals into the history-independent CSI encoder to provide one or more latent vectors; iii) using the one or more latent vectors to provide an input into a history-dependent recurrent quantizer to provide a second quantized output; iv) sending the second quantized output to the network node;48 repeating steps i) to iv) to provide a plurality of inputs and a corresponding plurality of quantized outputs; sending the plurality of quantized outputs to the network node; sending, to the network node, a plurality of Squared Generalized Cosine Similarity, SGCS, scores and / or a plurality of Mean Squared Error, MSE, values between each of the inputs and the corresponding quantized outputs.

7. The apparatus according to claim 6, where in the third mode a recurrent quantizer state machine of the apparatus and an inverse recurrent quantizer state machine of the network node are synchronized by aligning a latest step size and a latest de-quantized value for the history-dependent recurrent scalar quantizer.

8. The apparatus according to claim 7, wherein the means are further configured for: receiving, from the network node, a request for a status of the state machine of the apparatus; sending, to the network node, the status for the state machine of the apparatus.

9. The apparatus according to any preceding claim, wherein at least one of the first threshold and the second threshold is at least one of: configured by the network node at the user equipment; aligned between the apparatus and the network node.

10. A method comprising: receiving, from a network node, at least one reference signal;determining at least one latent vector based on target Channel State Information acquired from the at least one reference signal and a history-independent CSI encoder; using the at least one latent vector to provide an input into a historydependent recurrent quantizer to provide a quantized output; calculating a mean squared error between the input and the quantized output; comparing the mean squared error to a threshold level; based on the comparing, determining to operate in a first mode, wherein in the first mode, the apparatus performs: determining at least one further latent vector based on target Channel State Information acquired from the at least one reference signal and the history-independent CSI encoder; using the at least one further latent vector to provide an input into a scalar quantizer to provide a full-resolution quantization output; sending the full-resolution quantization output to the network node.

11. A computer program comprising instructions stored thereon for performing at least the following: receiving, from a network node, at least one reference signal; determining at least one latent vector based on target Channel State Information acquired from the at least one reference signal and a history-independent CSI encoder; using the at least one latent vector to provide an input into a historydependent recurrent quantizer to provide a quantized output; calculating a mean squared error between the input and the quantized output;comparing the mean squared error to a threshold level; based on the comparing, determining to operate in a first mode, wherein in the first mode, the apparatus performs: determining at least one further latent vector based on target Channel State Information acquired from the at least one reference signal and the history-independent CSI encoder; using the at least one further latent vector to provide an input into a scalar quantizer to provide a full-resolution quantization output; sending the full-resolution quantization output to the network node.

12. An apparatus comprising means for: sending, to a user equipment, at least one reference signal; receiving a full-resolution quantization output from a user equipment, wherein the full-resolution quantization output is generated at the user equipment by inputting the at least one reference signal into a channel estimator, a history-in- dependent CSI encoder and then a scalar quantizer.

13. The apparatus according to claim 12, wherein the means are further configured for: receiving a quantized output from the user equipment, wherein an input is at least one further latent vector generated based on target Channel State Information acquired from channel estimates based on the at least one reference signal and a history-independent CSI encoder; and wherein the quantized output is generated at the user equipment by providing the input into a recurrent quantizer.

14. The apparatus of claim 12 or claim 13, wherein the means are further configured for:using a recurrent de-quantizer to de-quantize the quantized output to provide a second output; determine reconstructed CSI of the channel between the apparatus and the user equipment based on the second output and a decoder.

15. The apparatus of any of claims 12 to 14, wherein the means are further configured for: when the mean squared error is below a second threshold, receiving an indication from the user equipment that the output of the recurrent quantizer is stable.

16. The apparatus according to claim 15, wherein the means are further configured for: sending, to the user equipment, a command to change to a mode of operation where quantization error for the quantized output is below the second threshold level.

17. The apparatus according to claim 16, wherein in the mode of operation, a recurrent quantizer state machine of the user equipment and an inverse recurrent quantizer state machine of the apparatus are synchronized by aligning a latest step size and a latest de-quantized value for the history-dependent recurrent scalar quantizer.

18. The apparatus according to any of claims 12 to 17, wherein the means are further configured for: sending, to the user equipment, a request for a status of a state machine of the user equipment; receiving, from the user equipment, the status for the state machine of the user equipment;52 synchronizing a state machine at the apparatus according to the status for the state machine of the user equipment.

19. The apparatus according to any of claims 15 to 18, wherein the second threshold is at least one of: configured at the user equipment by the apparatus; aligned between the apparatus and the user equipment.

20. A method comprising: sending, to a user equipment, at least one reference signal; receiving a full-resolution quantization output from a user equipment, wherein the full-resolution quantization output is generated at the user equipment by inputting the at least one reference signal into a channel estimator, a history-in- dependent CSI encoder and then a scalar quantizer.

21. A computer program comprising instructions stored thereon for performing at least the following: sending, to a user equipment, at least one reference signal; receiving a full-resolution quantization output from a user equipment, wherein the full-resolution quantization output is generated at the user equipment by inputting the at least one reference signal into a channel estimator, a history-in- dependent CSI encoder and then a scalar quantizer.

Images