Neural network decoder adaptation in a wireless communications system involving data augmentation

Abstract

Various aspects of the present disclosure relate to decoder adaptation in a wireless communications system. A decoder neural network receives at least one encoded message from a transmitter (e.g., encoded using an encoder neural network at the transmitter). One or more modified versions of the at least one encoded message are generated. One or more layers of the decoder neural network are updated based at least in part on the at least one encoded message, the one or more modified versions of the at least one encoded message, and a loss function.

Description

Lenovo Ref. No. SMM920230253-WO-PCT 1 DECODER ADAPTATION IN A WIRELESS COMMUNICATIONS SYSTEM RELATED APPLICATION

[0001] This application claims priority to U.S. Patent Application Serial No. 63 / 555,389 filed February 19, 2024 entitled “DECODER ADAPTATION IN A WIRELESS COMMUNICATIONS SYSTEM,” the disclosure of which is incorporated by reference herein in its entirety. TECHNICAL FIELD

[0002] The present disclosure relates to wireless communications, and more specifically to decoder adaptation in a wireless communications system. BACKGROUND

[0003] A wireless communications system may include one or multiple network communication devices, which may be otherwise known as network equipment (NE), supporting wireless communications for one or multiple user communication devices, which may be otherwise known as user equipment (UE), or other suitable terminology. The wireless communications system may support wireless communications with one or multiple user communication devices by utilizing resources of the wireless communication system (e.g., time resources (e.g., symbols, slots, subframes, frames, or the like) or frequency resources (e.g., subcarriers, carriers, or the like). Additionally, the wireless communications system may support wireless communications across various radio access technologies including third generation (3G) radio access technology, fourth generation (4G) radio access technology, fifth generation (5G) radio access technology, among other suitable radio access technologies beyond 5G (e.g., sixth generation (6G)). SUMMARY

[0004] An article “a” before an element is unrestricted and understood to refer to “at least one” of those elements or “one or more” of those elements. The terms “a,” “at least one,” “one or more,” and “at least one of one or more” may be interchangeable. As used herein, including in the claims, “or” as used in a list of items (e.g., a list of items prefaced by a phrase such as “at least one of” or “one or more of” or “one or both of”) indicates an inclusive list such that, for example, a list of at least one of A, B, or C means A or B or C or AB or AC or BC or ABC (i.e., A and B and C). By Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 2 way of another example, a list of at least one of A; B; or C means A or B or C or AB or AC or BC or ABC (i.e., A and B and C). Also, as used herein, the phrase “based on” shall not be construed as a reference to a closed set of conditions. For example, an example step that is described as “based on condition A” may be based on both a condition A and a condition B without departing from the scope of the present disclosure. In other words, as used herein, the phrase “based on” shall be construed in the same manner as the phrase “based at least in part on”. Further, as used herein, including in the claims, a “set” may include one or more elements.

[0005] An apparatus (e.g., a UE or a NE), which may also be referred to as a first apparatus, for wireless communication is described. The apparatus may be configured to, capable of, or operable to perform one or more operations as described herein. For example, the apparatus may be configured to, capable of, or operable to receive at least one encoded message; generate one or more modified versions of the at least one encoded message; and update one or more layers of a decoder neural network based at least in part on the at least one encoded message, the one or more modified versions of the at least one encoded message, one or more tuning parameters, and a loss function.

[0006] A processor (e.g., a standalone processor chipset, a component of a UE, or a component of a NE) for wireless communication is described. The processor may be configured to, capable of, or operable to perform one or more operations as described herein. For example, the processor may be configured to, capable of, or operable to receive at least one encoded message; generate one or more modified versions of the at least one encoded message; and update one or more layers of a decoder neural network based at least in part on the at least one encoded message, the one or more modified versions of the at least one encoded message, one or more tuning parameters, and a loss function.

[0007] A method performed or performable by an apparatus (e.g., a UE or a NE) for wireless communication is described. The method may include receiving at least one encoded message; generating one or more modified versions of the at least one encoded message; and updating one or more layers of a decoder neural network based at least in part on the at least one encoded message, the one or more modified versions of the at least one encoded message, one or more tuning parameters, and a loss function. Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 3

[0008] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to update the one or more layers of the decoder neural network by minimizing the loss function.

[0009] In some implementations of the apparatus, processor, and method described herein, minimizing the loss function comprises minimizing maximum entropy in a region through min-max optimization.

[0010] In some implementations of the apparatus, processor, and method described herein, the at least one encoded message includes multiple encoded messages and the one or more modified versions of the at least one encoded message includes one or more modified versions of each of the multiple encoded messages, and the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to update the one or more layers of the decoder neural network based at least in part on the multiple encoded messages and the one or more modified versions of each of the multiple encoded messages.

[0011] In some implementations of the apparatus, processor, and method described herein, the multiple encoded messages are received sequentially in time domain, frequency domain, or spatial domain.

[0012] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to determine to update the one or more layers of the decoder neural network based at least in part on periodicity information indicating the one or more layers of the decoder neural network are to be updated at regular intervals of time.

[0013] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to determine to update the one or more layers of the decoder neural network based at least in part on a change in one or more characteristics of one or more received encoded messages.

[0014] In some implementations of the apparatus, processor, and method described herein, the one or more characteristics are statistical characteristics of the one or more received encoded messages. Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 4

[0015] In some implementations of the apparatus, processor, and method described herein, the statistical characteristics include at least one of a mean, a median, a standard deviation, or a probability distribution.

[0016] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to determine to update the one or more layers of the decoder neural network based at least in part on an indication, received from a second apparatus, to update the one or more layers of the decoder neural network.

[0017] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to determine to update the one or more layers of the decoder neural network based at least in part on a change in a value of at least one parameter corresponding to a channel, wherein the channel corresponds to a communication medium between the first apparatus and a second apparatus.

[0018] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to monitor a quality of the at least one encoded message, wherein the quality of the at least one encoded message comprises one or more of an indication of decoding success of the at least one encoded message, an indication of cyclical redundancy check status of the at least one encoded message, an acknowledgement indication for the at least one encoded message, a retransmission indication for the at least one encoded message, or a decoding error probability for the at least one encoded message above a threshold; and determine to update the one or more layers of the decoder neural network based at least in part on the monitoring.

[0019] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to receive the at least one encoded message from a second apparatus, and to determine to update the one or more layers of the decoder neural network based at least in part on an indication or a configuration, received from a third apparatus, to update the one or more layers of the decoder neural network. Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 5

[0020] In some implementations of the apparatus, processor, and method described herein, the first apparatus comprises a user equipment UE, the second apparatus comprises a base station, and the third apparatus comprises the base station or another network equipment NE.

[0021] In some implementations of the apparatus, processor, and method described herein, each encoded message of the at least one encoded message comprises one of multiple different information messages.

[0022] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to determine the one or more layers of the decoder neural network.

[0023] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to receive the at least one encoded message from a second apparatus, and to receive, from the second apparatus or a third apparatus, an indication of the one or more layers of the decoder neural network.

[0024] In some implementations of the apparatus, processor, and method described herein, the first apparatus comprises a user equipment UE, the second apparatus comprises a base station, and the third apparatus comprises another network equipment NE.

[0025] In some implementations of the apparatus, processor, and method described herein, the one or more tuning parameters comprises one or both of a first tuning parameter that is a number of the one or more modified versions for each of the at least one encoded message, and a second tuning parameter that is a setting of the loss function.

[0026] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to determine the one or more tuning parameters; or receive, from a second apparatus or a third apparatus, the one or more tuning parameters.

[0027] In some implementations of the apparatus, processor, and method described herein, the first apparatus comprises a user equipment UE, the second apparatus comprises a base station, and the third apparatus comprises another network equipment NE. Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 6

[0028] In some implementations of the apparatus, processor, and method described herein, to update values of a number N of parameters of the one or more layers of the decoder neural network based at least in part on the at least one encoded message, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to generate a number Q of different augmented received encoded messages where each augmented received encoded message is a modified version of at least one encoded message, and Q is a first tuning parameter of the one or more tuning parameters; obtain a set of Q+1 different data samples, wherein a first data sample is the at least one encoded message and remaining Q data samples are the Q number of different augmented received encoded messages generated from the at least one encoded message; for each of the Q+1 different data samples, use the decoder neural network to compute a vector of length M where an ^^^element of the vector is a probability that an information message contained in the datasample is ^^, where ^^ ∈ {1, … , ^} and the at least one encoded message comprises one of Mmultiple different messages; add all the Q+1 number of vectors, each of length M,using element-wise addition to generate a temporary vector; divide each element of the temporary vector by Q+1 to generate a mean probability vector; compute an entropy of the mean probability vector, wherein the entropy is a function of parameters of the decoder neural network; and update the one or more layers of the decoder neural network by determining new values of the N parameters that minimizes a maximum entropy within an N-dimensional Euclidean ball, with the N- dimensional Euclidean ball centered at the new values of the N parameters with radius R, where R is a second tuning parameter of the one or more tuning parameters.

[0029] In some implementations of the apparatus, processor, and method described herein, the at least one encoded message is based at least in part on an encoder neural network.

[0030] In some implementations of the apparatus, processor, and method described herein, the encoder neural network is trained based on a first set of training data and the decoder neural network is trained based on a second set of training data, and wherein the first set of training data is different than the second set of training data.

[0031] In some implementations of the apparatus, processor, and method described herein, the encoder neural network and the decoder neural network are trained on a same set of training data. Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 7

[0032] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to transmit, to a second apparatus, an assistance information request message for updating one or more parameters of the decoder neural network.

[0033] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to receive, from the second apparatus based at least in part on the assistance information request message, at least one of a signal or an additional encoded message.

[0034] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to receive, from a second apparatus, an indication to determine whether to update the one or more layers of the decoder neural network.

[0035] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to update, based at least in part on successfully decoding the at least one encoded message, the one or more layers of the decoder neural network based at least in part on the at least one encoded message.

[0036] An apparatus (e.g., a UE or a NE), which may also be referred to as a first apparatus, for wireless communication is described. The apparatus may be configured to, capable of, or operable to perform one or more operations as described herein. For example, the apparatus may be configured to, capable of, or operable to generate, using an encoder neural network, at least one encoded message; transmit the at least one encoded message; and transmit an indication for a second apparatus to update one or more layers of a decoder neural network used by the second apparatus to decode the at least one encoded message.

[0037] A processor (e.g., a standalone processor chipset, a component of a UE, or a component of a NE) for wireless communication is described. The processor may be configured to, capable of, or operable to perform one or more operations as described herein. For example, the processor may be configured to, capable of, or operable to generate, using an encoder neural network, at least one encoded message; transmit the at least one encoded message; and transmit an indication for an Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 8 apparatus to update one or more layers of a decoder neural network used by the apparatus to decode the at least one encoded message.

[0038] A method performed or performable by an apparatus (e.g., a UE or a NE) for wireless communication is described. The method may include generating, using an encoder neural network, at least one encoded message; transmitting the at least one encoded message; and transmitting an indication for an apparatus to update one or more layers of a decoder neural network used by the apparatus to decode the at least one encoded message.

[0039] In some implementations of the apparatus, processor, and method described herein, the at least one encoded message includes multiple encoded messages transmitted sequentially in time domain, frequency domain, or spatial domain.

[0040] In some implementations of the apparatus, processor, and method described herein, each encoded message of the at least one encoded message comprises one of multiple different information messages.

[0041] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to transmit one or more tuning parameters used by the second apparatus to update one or more layers of the decoder neural network.

[0042] In some implementations of the apparatus, processor, and method described herein, the encoder neural network is trained based on a first set of training data and the decoder neural network is trained based on a second set of training data, and wherein the first set of training data is different than the second set of training data.

[0043] In some implementations of the apparatus, processor, and method described herein, the encoder neural network and the decoder neural network are trained on a same set of training data.

[0044] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to receive an assistance information request message for updating one or more parameters of the decoder neural network. Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 9

[0045] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to transmit, based at least in part on the assistance information request message, at least one of a signal or an additional encoded message.

[0046] In some implementations of the apparatus, processor, and method described herein, the apparatus, processor, and method may further be configured to, capable of, performed, performable, or operable to transmit an indication for the second apparatus to determine whether to update the one or more layers of the decoder neural network.

[0047] In some implementations of the apparatus, processor, and method described herein, the first apparatus comprises a base station and the second apparatus comprises a user equipment UE. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 illustrates an example of a wireless communications system in accordance with aspects of the present disclosure.

[0049] Figure 2 illustrates an example of end-to-end communication using a two-sided model in accordance with aspects of the present disclosure.

[0050] Figure 3 illustrates an example of decoder generalization or adaptation process in accordance with aspects of the present disclosure.

[0051] Figure 4 illustrates an example of a UE in accordance with aspects of the present disclosure.

[0052] Figure 5 illustrates an example of a processor in accordance with aspects of the present disclosure.

[0053] Figure 6 illustrates an example of a NE in accordance with aspects of the present disclosure.

[0054] Figures 7 and 8 illustrate flowcharts of method performed by a UE in accordance with aspects of the present disclosure.

[0055] Figure 9 and 10 illustrate flowchart of method performed by a NE in accordance with aspects of the present disclosure. Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 10 DETAILED DESCRIPTION

[0056] Recent advances in artificial intelligence / machine learning (AI / ML) have shown excellent performance across a variety of domains. Many of the AI / ML models work well when the input data (at time of model inference) and training data share the same statistical characteristics, and their performance degrades when the statistics of the input data deviates from that of the training data. It is desirable that the AI / ML works well under possible shifts in the statistical properties of the input data – this is especially the case with AI / ML models for wireless networks, as the wireless environment and the network conditions are dynamic and non-stationary in nature.

[0057] The techniques discussed herein adapt or generalize a decoder AI / ML model of an AI / ML model based receiver to a new domain. This new domain can be a new, previously unseen, domain. The techniques work with any number of unlabeled samples (including a single unlabeled sample) from the target domain. An unlabeled sample refers to a sample having an unknown value (the decoder does not know what the value of the sample is expected to be). The decoder receives encoded messages from a transmitter and can be adapted based on, for example, every single received encoded message from the transmitter without knowing what the information message is contained in the encoded message. In other words, the decoder can be adapted without requiring pilot or reference (e.g., known) symbols.

[0058] In one or more implementations, a decoder neural network receives at least one encoded message from a transmitter (e.g., encoded using an encoder neural network at the transmitter). One or more modified versions (also referred to as augmented versions) of the at least one encoded message are generated. Any of a variety of augmentation methods can be used, such as adding noise to the encoded message, multiplying the encoded message by a phaser, and so forth. One or more layers of the decoder neural network are then updated based at least in part on the at least one encoded message, the one or more modified versions of the at least one encoded message, and a loss function.

[0059] Accordingly, the techniques discussed herein adapt the decoder neural network to changes in the wireless communications system. These changes can include changes at the encoder side, such as a change in the encoder (e.g., a change in serving cell or an encoder neural network being adapted or re-trained), or a change in the communication channel (e.g., an increase in noise in Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 11 the communication channel). The decoder neural network is adapted based on the received encoded messages without requiring pilot or reference (e.g., known) encoded messages, thus avoiding expending energy and network resources on transmitting pilot or reference encoded messages.

[0060] Reference is made herein to communicating data or information, such as signaling communication resources and / or communications that are transmitted or received between devices. It is to be appreciated that other terms may be used interchangeably with communicating, such as signaling, transmitting, receiving, outputting, forwarding, retrieving, obtaining, and so forth.

[0061] Aspects of the present disclosure are described in the context of a wireless communications system.

[0062] Figure 1 illustrates an example of a wireless communications system 100 in accordance with aspects of the present disclosure. The wireless communications system 100 may include one or more NE 102, one or more UE 104, and a core network (CN) 106. The wireless communications system 100 may support various radio access technologies. In some implementations, the wireless communications system 100 may be a 4G network, such as an LTE network or an LTE-Advanced (LTE-A) network. In some other implementations, the wireless communications system 100 may be a new radio (NR) network, such as a 5G network, a 5G-Advanced (5G-A) network, or a 5G ultrawideband (5G-UWB) network. In other implementations, the wireless communications system 100 may be a combination of a 4G network and a 5G network, or other suitable radio access technology including Institute of Electrical and Electronics Engineers (IEEE) 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802.20. The wireless communications system 100 may support radio access technologies beyond 5G, for example, 6G. Additionally, the wireless communications system 100 may support technologies, such as time division multiple access (TDMA), frequency division multiple access (FDMA), or code division multiple access (CDMA), etc.

[0063] The one or more NE 102 may be dispersed throughout a geographic region to form the wireless communications system 100. One or more of the NE 102 described herein may be or include or may be referred to as a network node, a base station, a network element, a network function, a network entity, a radio access network (RAN), a NodeB, an eNodeB (eNB), a next- generation NodeB (gNB), or other suitable terminology. An NE 102 and a UE 104 may communicate via a communication link, which may be a wireless or wired connection. For example, Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 12 an NE 102 and a UE 104 may perform wireless communication (e.g., receive signaling, transmit signaling) over a Uu interface.

[0064] An NE 102 may provide a geographic coverage area for which the NE 102 may support services for one or more UEs 104 within the geographic coverage area. For example, an NE 102 and a UE 104 may support wireless communication of signals related to services (e.g., voice, video, packet data, messaging, broadcast, etc.) according to one or multiple radio access technologies. In some implementations, an NE 102 may be moveable, for example, a satellite associated with a non- terrestrial network (NTN). In some implementations, different geographic coverage areas associated with the same or different radio access technologies may overlap, but the different geographic coverage areas may be associated with different NE 102.

[0065] The one or more UE 104 may be dispersed throughout a geographic region of the wireless communications system 100. A UE 104 may include or may be referred to as a remote unit, a mobile device, a wireless device, a remote device, a subscriber device, a transmitter device, a receiver device, or some other suitable terminology. In some implementations, the UE 104 may be referred to as a unit, a station, a terminal, or a client, among other examples. Additionally, or alternatively, the UE 104 may be referred to as an Internet-of-Things (IoT) device, an Internet-of- Everything (IoE) device, or machine-type communication (MTC) device, among other examples.

[0066] A UE 104 may be able to support wireless communication directly with other UEs 104 over a communication link. For example, a UE 104 may support wireless communication directly with another UE 104 over a device-to-device (D2D) communication link. In some implementations, such as vehicle-to-vehicle (V2V) deployments, vehicle-to-everything (V2X) deployments, or cellular-V2X deployments, the communication link may be referred to as a sidelink. For example, a UE 104 may support wireless communication directly with another UE 104 over a PC5 interface.

[0067] An NE 102 may support communications with the CN 106, or with another NE 102, or both. For example, an NE 102 may interface with other NE 102 or the CN 106 through one or more backhaul links (e.g., S1, N2, N6, or other network interface). In some implementations, the NE 102 may communicate with each other directly. In some other implementations, the NE 102 may communicate with each other indirectly (e.g., via the CN 106). In some implementations, one or more NE 102 may include subcomponents, such as an access network entity, which may be an Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 13 example of an access node controller (ANC). An ANC may communicate with the one or more UEs 104 through one or more other access network transmission entities, which may be referred to as a radio heads, smart radio heads, or transmission-reception points (TRPs).

[0068] The CN 106 may support user authentication, access authorization, tracking, connectivity, and other access, routing, or mobility functions. The CN 106 may be an evolved packet core (EPC), or a 5G core (5GC), which may include a control plane entity that manages access and mobility (e.g., a mobility management entity (MME), an access and mobility management functions (AMF)) and a user plane entity that routes packets or interconnects to external networks (e.g., a serving gateway (S-GW), a packet data network (PDN) gateway (P-GW), or a user plane function (UPF)). In some implementations, the control plane entity may manage non-access stratum (NAS) functions, such as mobility, authentication, and bearer management (e.g., data bearers, signal bearers, etc.) for the one or more UEs 104 served by the one or more NE 102 associated with the CN 106.

[0069] The CN 106 may communicate with a packet data network over one or more backhaul links (e.g., via an S1, N2, N6, or other network interface). The packet data network may include an application server. In some implementations, one or more UEs 104 may communicate with the application server. A UE 104 may establish a session (e.g., a protocol data unit (PDU) session, or the like) with the CN 106 via an NE 102. The CN 106 may route traffic (e.g., control information, data, and the like) between the UE 104 and the application server using the established session (e.g., the established PDU session). The PDU session may be an example of a logical connection between the UE 104 and the CN 106 (e.g., one or more network functions of the CN 106).

[0070] In the wireless communications system 100, the NEs 102 and the UEs 104 may use resources of the wireless communications system 100 (e.g., time resources (e.g., symbols, slots, subframes, frames, or the like) or frequency resources (e.g., subcarriers, carriers)) to perform various operations (e.g., wireless communications). In some implementations, the NEs 102 and the UEs 104 may support different resource structures. For example, the NEs 102 and the UEs 104 may support different frame structures. In some implementations, such as in 4G, the NEs 102 and the UEs 104 may support a single frame structure. In some other implementations, such as in 5G and among other suitable radio access technologies, the NEs 102 and the UEs 104 may support various Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 14 frame structures (i.e., multiple frame structures). The NEs 102 and the UEs 104 may support various frame structures based on one or more numerologies.

[0071] One or more numerologies may be supported in the wireless communications system 100, and a numerology may include a subcarrier spacing and a cyclic prefix. A first numerology (e.g., ^=0) may be associated with a first subcarrier spacing (e.g., 15 kHz) and a normal cyclic prefix. In some implementations, the first numerology (e.g., ^=0) associated with the first subcarrier spacing (e.g., 15 kHz) may utilize one slot per subframe. A second numerology (e.g., ^=1) may be associated with a second subcarrier spacing (e.g., 30 kHz) and a normal cyclic prefix. A third numerology (e.g., ^=2) may be associated with a third subcarrier spacing (e.g., 60 kHz) and a normal cyclic prefix or an extended cyclic prefix. A fourth numerology (e.g., ^=3) may be associated with a fourth subcarrier spacing (e.g., 120 kHz) and a normal cyclic prefix. A fifth numerology (e.g., ^=4) may be associated with a fifth subcarrier spacing (e.g., 240 kHz) and a normal cyclic prefix.

[0072] A time interval of a resource (e.g., a communication resource) may be organized according to frames (also referred to as radio frames). Each frame may have a duration, for example, a 10 millisecond (ms) duration. In some implementations, each frame may include multiple subframes. For example, each frame may include 10 subframes, and each subframe may have a duration, for example, a 1 ms duration. In some implementations, each frame may have the same duration. In some implementations, each subframe of a frame may have the same duration.

[0073] Additionally or alternatively, a time interval of a resource (e.g., a communication resource) may be organized according to slots. For example, a subframe may include a number (e.g., quantity) of slots. The number of slots in each subframe may also depend on the one or more numerologies supported in the wireless communications system 100. For instance, the first, second, third, fourth, and fifth numerologies (i.e., ^=0, ^=1, ^=2, ^=3, ^=4) associated with respective subcarrier spacings of 15 kHz, 30 kHz, 60 kHz, 120 kHz, and 240 kHz may utilize a single slot per subframe, two slots per subframe, four slots per subframe, eight slots per subframe, and 16 slots per subframe, respectively. Each slot may include a number (e.g., quantity) of symbols (e.g., OFDM symbols). In some implementations, the number (e.g., quantity) of slots for a subframe may depend on a numerology. For a normal cyclic prefix, a slot may include 14 symbols. For an extended cyclic prefix (e.g., applicable for 60 kHz subcarrier spacing), a slot may include 12 symbols. The Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 15 relationship between the number of symbols per slot, the number of slots per subframe, and the number of slots per frame for a normal cyclic prefix and an extended cyclic prefix may depend on a numerology. It should be understood that reference to a first numerology (e.g., ^=0) associated with a first subcarrier spacing (e.g., 15 kHz) may be used interchangeably between subframes and slots.

[0074] In the wireless communications system 100, an electromagnetic (EM) spectrum may be split, based on frequency or wavelength, into various classes, frequency bands, frequency channels, etc. By way of example, the wireless communications system 100 may support one or multiple operating frequency bands, such as frequency range designations FR1 (410 MHz – 7.125 GHz), FR2 (24.25 GHz – 52.6 GHz), FR3 (7.125 GHz – 24.25 GHz), FR4 (52.6 GHz – 114.25 GHz), FR4a or FR4-1 (52.6 GHz – 71 GHz), and FR5 (114.25 GHz – 300 GHz). In some implementations, the NEs 102 and the UEs 104 may perform wireless communications over one or more of the operating frequency bands. In some implementations, FR1 may be used by the NEs 102 and the UEs 104, among other equipment or devices for cellular communications traffic (e.g., control information, data). In some implementations, FR2 may be used by the NEs 102 and the UEs 104, among other equipment or devices for short-range, high data rate capabilities.

[0075] FR1 may be associated with one or multiple numerologies (e.g., at least three numerologies). For example, FR1 may be associated with a first numerology (e.g., ^=0), which includes 15 kHz subcarrier spacing; a second numerology (e.g., ^=1), which includes 30 kHz subcarrier spacing; and a third numerology (e.g., ^=2), which includes 60 kHz subcarrier spacing. FR2 may be associated with one or multiple numerologies (e.g., at least 2 numerologies). For example, FR2 may be associated with a third numerology (e.g., ^=2), which includes 60 kHz subcarrier spacing; and a fourth numerology (e.g., ^=3), which includes 120 kHz subcarrier spacing.

[0076] In some cases, a cell refers to a radio access node in communication with a base station or including a base station. A cell typically has a coverage area, which is a geographic area in which the cell provides wireless connectivity to devices within. Different cells may operate on defined frequencies or frequency bands, referred to as subcarriers. In some examples, a UE 104 establishes a wireless connection with a cell, and subsequently that cell may be referred to as a serving cell of the UE 104. Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 16

[0077] In one or more implementations, a decoder neural network (e.g., at a UE 104) receives at least one encoded message from a transmitter (e.g., a NE 102). The decoder neural network is one side of a two-sided model, with the other side of the two-sided model being an encoder neural network (e.g., at the transmitter). One or more modified versions of the at least one encoded message are generated. Any of a variety of conventional augmentation methods can be used, such as adding noise to the encoded message, multiplying the encoded massage by a phaser, and so forth. One or more layers of the decoder neural network are then updated based at least in part on the at least one encoded message, the one or more modified versions of the at least one encoded message, one or more tuning parameters, and a loss function.

[0078] An AI / ML based end-to-end (E2E) wireless communication system, is considered and may be used for the next generation AI-native wireless networks. For implementing E2E communication through AI / ML methods, the focus is on two-sided AI / ML models, where one model employed at the transmitting end works in tandem with another model employed at the receiving end of the communication link. An autoencoder, a popular deep learning (DL) model, is a natural choice for such an application. An autoencoder is a two-sided AI / ML model that consists of an encoder neural network ^^, a deep neural network (DNN) with learnable or trainable parameters denoted by ^, and aneural network ^^, another DNN with ^ as its set of trainable or learnable parameters. The encoder neural network (NN) and the decoder NN can be trained and developed together such that the signal or data at the input to the encoder NN is reconstructed, as faithfully as possible, at the output of the decoder NN. Thus, the two neural networks, or the twomodels, ^^ and ^^, together constitute an autoencoder and are denoted as ^ = {^^, ^^}. It shouldbe the techniques discussed herein are not limited to autoencoders, and are applicable, in general, to an E2E communication system that is based on neural network-based encoder at the transmitting end and a neural network-based decoder at the receiving end.

[0079] In the discussions herein, “encoder” and “encoder neural network” (encoder NN) are used synonymously. Similarly, “decoder” and “decoder neural network” (decoder NN) are used synonymously.

[0080] Reference is made herein to a single-input single-output (SISO) wireless communication system. However, the techniques discussed herein are equally applicable to wireless communication Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 17 systems that are equipped with multiple antennas at the transmitter and / or the receiver ((e.g., multiple-input single-output (MISO) and / or multiple input and multiple output (MIMO) wireless communication systems).

[0081] Figure 2 illustrates an example of end-to-end communication using a two-sided model 200 in accordance with aspects of the present disclosure. The E2E communication is based on an autoencoder. Figure 2 illustrates a typical E2E communication system based on a two-sided AI / MLmodel, or an autoencoder, for communicating messages 208 that belong to the set ℳ = {1, 2, … , ^}over a wireless channel from one node to another node, for example, from a base station (e.g., a gNB) to a UE or in the other direction (from a UE to a gNB). The two-sided model 200 includes an encoder NN 202, a communication channel 204, and a decoder NN 206.

[0082] The encoder NN 202, ^^, encodes message ^^ ∈ ℳ into a symbol 210 (symbol ^^ ∈ℝ^^), suitable for the givenwhere ^ ≥ 1 is the number of channel uses. The encodedvector can equivalently be considered as ^^ ∈ ℂ^ with ^ as the number of channel uses, as RFcommunications over a SISO channelone complex valued symbol per channel use. Thus, the encoder NN 202 is a one-to-one function from input messages to channel symbols and can beexpressed as ^^: ℳ → ℂ^ . The value of ^ depends on the channel distributions or the channelconditionsautoencoder is getting trained. For example, when the autoencoder is trained over a channel that causes more degradation of the symbols sent over it, then the encoder NN 202 may choose to encode the message to send it over many channel uses, resulting in a higher value of ^, for achieving a certain reliability of the communication link. On the other hand, if the autoencoder is trained for communication over a benign or better channel, such as an additive white Gaussian noise (AWGN) channel, then the encoder NN 202 may choose to encode the message for transmitting it over only one channel use for achieving the same level of reliability as in the previous case.

[0083] The encoded symbols ^^are sent over the channel 204. The channel 204 corrupts ortransforms the symbols and let ! ∈ ℂ^ denote the channel o ^^ utput 212 for input ^^ ∈ ℂ . Thedecoder NN 206 ^^is trainedit decodes the channel output !^tomessage 214,message ^^carried by or contained in !^with high reliability. In other words, the decoder is trained such that it classifies the channel output into one of the ^ messages Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 18 such that the average probability of error in classification is reduced (e.g., minimized). Thus, the decoder NN 206 is a function that maps channel output 212 into messages 214, and can beexpressed as, ^ ^^: ℂ → ℳ.

[0084] the channel 204, the complex valued wireless channel gain is stochastic in nature and is time varying. The channel can be characterized through the probability densityfunction (pdf) of channel gains. For example, a SISO Rayleigh fading channel corresponds to ℎ ∼$%&0,1(, e.g., ℎ is a zero mean unit variance complex Gaussian random variable. Thus, a wireless channel 204 can be characterized in terms of its probability density or distribution function (pdf). Another equivalent way is to characterize the channel is through the channel transition probability )&!|^(, where ^ is the channel input and ! is the channel output. It should be noted that due to theone-to-one mapping between the messages and the symbols to be sent over the channel, )&^^( =)&^^(, ^ = 1, … , ^.

[0085] Further, the channel 204 is considered to include additive noise at the receiver input and the impairments or imperfections of the hardware at the transmitter and at the frontend of the receiver. Any hardware impairments or imperfections would also distort the signal sent or received over the communication link (for example, the oscillator drift leading to Carrier Frequency Offset, phase jitter, phase noise, etc.). Thus, overall, the channel 204 collectively accounts for the distortion and noise introduced into the channel input symbols (which are the encoded symbols at the output of the encoder NN 202).

[0086] In an autoencoder based E2E communications system, the autoencoder is trained to communicate a given set of messages ℳ over a given wireless channel with a small (e.g., minimum) probability of error at the decoder output. Developing or training an autoencoder can be summarized as follows.

[0087] Inputs used for training an autoencoder are: a sufficiently large number of data samples that represent the channel gains, e.g., that are obtained through simulations or from a generative model, or from field measurements, or a differentiable channel model; and a set of messages that are to be sent over the channel.

[0088] Messages ^^ ∈ ℳ = {1, … , ^}, can be given to the encoder NN as one-hot vectors,meaning that each message ^^is encoded into a length-^ binary vector, whose ^^^^element isAttorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 19 equal to 1 and all other elements are equal to zero. As the mapping from set of messages ℳ to the set of length-^ one-hot vectors is a one-to-one mapping, an input message can be denoted as an integer ^^or as length-^ one-hot vector having a zero value at all positions or elements except at the position or element ^^.

[0089] The encoder NN and the decoder NN of the autoencoder are trained together by reducing (e.g., minimizing) the error probability )+at the output of the decoder NN, where )+ = , )-&^. ^ ≠ ^^()&^^(,and )&^^(, ^^ ∈ ℳ = {1, … , ^} are messages.

[0090] Thus, the loss function considered for training an autoencoder is the decoding error probability at the output of the decoder NN. Such a method of training an autoencoder jointly trains the encoder NN and the decoder NN thereby making the encoder “paired” with the decoder or making the encoder and the decoder “matched pairs”. The encoder NN and the decoder NN deliver desired performance over the communication link only if the encoder at the Tx end and the decoder at the Rx end belong to the matched pairs, or when the encoder is made to work with its paired decoder and vice versa.

[0091] One or more points to be noted about E2E communication using autoencoder are the following. Channel, and hence, the transition probabilities )&!|^(are determined by the propagation medium which is typically not in control during the model design or training process. The encoder, or mapping performed by the encoder from the set of messages to the set of symbols,depends on |ℳ| (the cardinality of the set of messages), )&^^(, ^ = 1, … , ^, prior probabilities ofthe messages, and )&!|^(. The decoder is a classifier that classifies the channel output into one of the ^ possible transmitted messages. Input data at the decoder’s input depends on input symbols to the channel and the channel characteristics. Note that the probability density function of the data atthe input of the decoder depends on )&^^(, ^ = 1, … , ^, and )&!|^(.

[0092] When |ℳ| and / or )&^^(, ^ = 1, … , ^ and / or )&!|^( changes, then, for providing goodperformance the encoder ^^is expected to be adapted or adjusted to produce optimal encoding ofthe messages under theor changed values of |ℳ|, )&^^(, )&!|^(. Once the encoder haschanged its mapping function, the decoder ^^is expected to be adapted or adjusted to faithfully Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 20 decode the symbols encoded at the transmitter and transformed by the channel. Even if the encoder is adapted or adjusted, not changing the decoder could result in significantly high error probability in decoding the transmitted messages, resulting in a very unreliable communication link.

[0093] Wireless channel characteristics are dynamic in nature and vary on a shorter time scale making the transition probabilities change over a wide range. It is reasonable to assume that |ℳ|and )&^^(, ^ = 1, … , ^, changes less frequently (especially, when compared to the time scale atwhich )&!|^(varies).

[0094] Considering the decoder and the points mentioned above, the decoder is adapted or re- trained to work well whenever any one of the following happens: when the encoder changes (e.g., when the mapping from messages to symbols changes), or when the channel changes.

[0095] The techniques discussed herein discuss adapting the decoder in a pre-trained and deployed autoencoder such that it performs well even with possible changes with the encoder and / or the channel.

[0096] With respect to interoperability in AI-native wireless based on two-sided AI / ML models, the techniques for decoder adaptation discussed herein enable a signal encoded by an encoder to be decoded by a decoder that is not jointly developed or trained along with the encoder.

[0097] A wireless network, in practice, is a multi-user network in the sense that at any given time, multiple users are served by a base station (e.g., a gNB) or an access point (AP). In a typicalwireless cellular network, multiple UEs, say 2^3, 2^^, … , 2^4 are served by or connected to asingle gNB, say 5^at a given time and uses downlink communication (from gNB to UEs) in such anetwork. Let ^6 denote the decoder deployed at 7^^ UE, 7 = 1, … , 8 and let ^^ denote the encoder atthe gNB 5 . It may not be practical to ensure that all the decoders 6^ ^ , 7 = 1, … , 8 are trained jointlywith the encoder ^^. This problem would be further accentuated by the fact that the UEs, being mobile, will move out from the coverage area of 5^and get connected to another gNB, say 5^9,where ^ ≠ ^:. Thus, one solution is to have all the gNBs and all the UEs in the network haveencoders and decoders that are jointly trained and developed together. However, it would be very difficult, if not practically in-feasible, to ensure such a situation where the decoders employed at different UEs are trained jointly with, and perfectly matched to, the encoders employed at the gNBs. Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 21

[0098] One solution to the above problem is the following. A gNB employs its own encoder for transmitting data across the wireless channel and different UEs in the network employs their own decoders, that are not necessarily jointly developed with the encoder deployed at the gNB, or the encoders and decoders are developed and supplied by different vendors. The method of adjusting or adapting the decoder in accordance with the encoder and the wireless channel, discussed in more detail below, is helpful in such scenarios. The method efficiently adapts a decoder to work well with an encoder even when the encoder and the decoder are not trained and developed jointly. Thus, it allows UEs to employ an off-the-shelf pre-trained decoder, supplied by any vendor, and still makes it useful to decode signals encoded by an encoder supplied by a different vendor. Thus, the AI / ML model adaption method discussed in more detail below solves the problem of how to make use of a single decoder to decode signals encoded by an encoder that is not jointly trained with the decoder.

[0099] The techniques discussed herein are an inference or test time adaptation method that works even on a single unlabeled sample from the target domain. Thus, for the decoder NN adaptation being considered here, the decoder is adapted with, e.g., every single received symbol from the transmitter, without knowing what the information message is contained in the received symbol. In other words, the decoder can be adapted using every received symbol without requiring those received symbols to be pilot or reference symbols. In some examples, the decoder is adapted based on at least a subset of received symbols (e.g., data, information symbols) when the receiver decodes correctly (e.g., cyclic redundancy check (CRC) check pass) the information message contained in the set of received symbols (e.g., corresponding to a codeblock (CB) with the information bits and parity bits from a CRC check error-detection encoding of the information message and possibly encoding by a error-correction code (e.g., low-density parity-check (LDPC), Turbo code)). The decoded information bits can be re-encoded (via channel coding) to generate a set of messages at encoder input (which are the ground-truth or labels at the decoder output) which together with the corresponding received symbols can be used to create a labeled dataset (e.g., batch-mode labeled data at inference time) for adaptation of the decoder.

[0100] Though recent advances in AI / ML have shown good performance across a variety of domains, many of the AI / ML models work well when the input data (at time of model inference) and training data share the same statistical characteristics, and their performance degrades when the statistics of the input data deviates from that of the training data. It is desirable that the AI / ML Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 22 works well under possible shifts in the statistical properties of the input data – this is especially the case with AI / ML models for wireless networks, as the wireless environment and the network conditions are dynamic and non-stationary in nature.

[0101] In AI / ML, making the models adaptable or generalizable across multiple domains is an active area of research and is of interest to the wireless community, as the AI / ML models developed for wireless applications are expected to work across widely varying network settings and channel conditions.

[0102] The following definitions and notation are provided to assist with discussions of the generalization and adaptation of AI / ML models.

[0103] A domain is composed of data samples ; = {&<^, =^(}?^>3 that are sampled from adistribution )@A&<, =(. Here, < CB ∈ ℝ is an input sample from a ^ − dimensional input space E, andlG ∈ ℝ is its label from an output space ℒ and )@A is the joint distribution of input samples and theircorresponding labels. Note that <^and lGare realizations of random variable I and J, respectively, with )@&<(representing marginal distribution of input samples and )A&=(denoting the marginal distribution of output labels. Data samples in a domain may correspond to a certain set of transmitter / receiver conditions / additional conditions such as environment / scenario (e.g., urban, indoor, rural etc.), line-of-sight (LoS) or non-line-of-sight (NLoS) channel, mobility (e.g., speed, Doppler), antenna configuration (e.g., at transmitter, receiver) etc.

[0104] In the discussions herein, the argument is skipped while denoting a probabilitydistribution function. E.g., )@A&< , =( is written as )@A, )@&<( is written as )@ and )A&=( is writtenas )A.

[0105] The source domain refers to the one or more domains over which an AI / ML model istrained. While developing an AI / ML model, the model is trained over data from K ≥ 1 domains,6 6 N 6 9;6 = {L< , = M}? ∼ ) , 7 = 1, … , K, whe 6 6 : 6^ ^ ^>3 re ) ≠ ) , for 7 ≠ 7 . The domains ; , 7 = 1, … , K,one or more than one and there may be multiple source domains for achieving a better generalizability for an AI / ML model. Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 23

[0106] With ;6 ∼ )6 , 7 = ^ & ^ ^( ?O ^@A 1, … , K, as the source domains, domain ; = { <^, =^ }^>3 ∼ )@A ,where )^@A ≠ )6@A , for some or all 7 = 1, … , K, is called a target a domainwhose distribution is different from some or all the source domains of an AI / ML model (and over which the AI / ML model will have to make inferences) is called a target domain for the AI / ML model being considered.

[0107] A data set that contains both input samples and the corresponding output samples (e.g., labels) is referred to as a labeled data set. A data set that contains only the input samples without the corresponding output samples or labels is referred to as an unlabeled data set. Thus, ;6= {L<6 , =6M}?N∼ )6 is a 6 6 ?N 6^ ^ ^>3 @A labeled data set from domain 7 and ; = {<^}^>3 ∼ )@ is an unlabeled data7.

[0108] The task of domain generalization (DG) is to learn a predictive function P: E → J toachieve a small (e.g., minimum) prediction error on the data over source domains as well as the unseen target domain.

[0109] The task of domain adaptation (DA) is the same as that of the domain generalization, however, the problem set up of DA and DG have a difference. In DA the model has access to target domain data during training (at least a few samples of labelled or unlabeled data from the target domain). In DG, there is no access to target domain data during training and data samples are seen from the target domain only at the time of inference or testing. The techniques discussed herein consider availability of target domain data at the time of inference or testing and not at the time of training the AI / ML model. Accordingly, the techniques discussed herein correspond to domain generalization. However, in some situations “domain generalization” and “domain adaptation” may be synonymous.

[0110] The techniques discussed herein consider a case where data from the target domain is not seen until the testing or inference phase is entered. The AI / ML model is adapted such that it improves its inferences over a domain that remains unseen so far.

[0111] There are various methods or techniques to make AI / ML models work across multiple domains (e.g., over data having widely different statistical characteristics). Some of these techniques are discussed in the following. Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 24

[0112] Depending on when / at what time we apply the technique to generalize / adapt an AI / ML model, we can classify the DG techniques as follows.

[0113] One type of DG is training time DG. In DG, data from the target domain is not seen during training. However, the network can be trained to make it generalizable to an unseen target domain through some training techniques such as domain-invariant feature learning, data mix-up over multiple source domains during training, data-augmentation, and so forth. A basic premise behind all training-based methods is that training a model over many source domains (and thus, trained over many data sets that are rich in diversity) makes it generalizable to unseen target domains. Such a premise may not always hold good and may not result in the best possible generalizability as no direct knowledge about the target domain is utilized in such methods. Further, such training techniques typically use large amounts of data from multiple source domains and significant amounts of computational resources for training.

[0114] Another type of DG is inference or test time DG. These methods adapt or generalize an AI / ML model after the model gets trained and deployed in the field and hence, they are called inference or test time adaptation or generalization methods. The adaptation or generalization methods differ further, based on whether they work on labeled or unlabeled data from the target domain and whether the model gets to observe data from target domain in a “batch-mode” (e.g., a bunch of data samples all available at the same time) or “sequential-mode” (e.g., data samples arriving one at a time, also referred to as an online mode).

[0115] Another type of DG is training and test or inference time DG. Such methods attempt to make the model generalizable during training as well as inference time. They have a higher potential to achieve a superior generalizability but at the cost of higher complexity and training time.

[0116] Inference or test time DG may be used because a readily available off-the-shelf (e.g., vendor specific) AI / ML model can be taken and the DG method applied to make the AI / ML model work for the desired target domain. Such a method is also helpful when the encoder and the decoder of the autoencoder come from two different vendors or suppliers. For example, a gNB can employ its own encoder for transmitting data across the wireless channel and different UEs in the network can employ their own decoders, that are not necessarily jointly developed with the encoder Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 25 deployed at the gNB. In such a case, if the encoder at the gNB and the decoder at the UE have the same |ℳ|, then the test time DG method (discussed in more detail below) can be applied and the decoder at the UE made to work well with the encoder at the gNB. Thus, the only thing in common between the encoder and the decoder may be |ℳ|. This helps us realize a “single decoder for multiple encoders” policy.

[0117] The DG or DA technique described below is an inference or test time technique that works even on a single unlabeled sample from the target domain. Thus, for the decoder NN adaptation being, the decoder can be adapted with every single received symbol from the transmitter, without knowing what the information message is contained in the received symbol. Further, in the techniques discussed blow, it is sufficient to have access to the unlabeled samples from the target domain in a sequential manner, one sample at a time, and it does not require all the samples to be available in a batch-mode. Thus, for the decoder NN adaptation techniques discussed herein, the decoder can be adapted with every single received symbol from the transmitter, without requiring the symbols to be pilot or reference symbols.

[0118] In some examples, the decoder NN adaptation may be based on at least a subset of received symbols (e.g., data, information symbols) when the receiver decodes correctly (e.g., CRC check pass) the information message contained in the set of received symbols. The received encoded symbols may correspond to a codeblock with the information bits and parity bits from a CRC check error-detection encoding of the information message and possibly encoding by an error- correction code (e.g., LDPC, Turbo code)). The decoded information bits can be re-encoded (via channel coding - CRC check error-detection code, error-correction code) to generate a set of messages at encoder input (which are the ground-truth or labels at the decoder output) which together with the corresponding received symbols can be used to create a labelled dataset (e.g., labelled samples from the target domain in a batch-mode at inference time) for adaptation of the decoder at inference time based on the labelled dataset created from the target domain.

[0119] Techniques for adapting or generalizing an AI / ML model based receiver, or a decoder AI / ML model, to new, previously unseen, domain is discussed herein. The techniques work with even a single unlabeled sample from the target domain. The decoder can be adapted with, e.g., every single received symbol from the transmitter, without knowing what the information message Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 26 is contained in the received symbol. In other words, the decoder can be adapted without requiring pilot or reference symbols.

[0120] Consider the decoder ^^ part of an autoencoder {^^, ^^} trained on one or more sourcedomains (in one or more the model is trained over multiple source domains ratherthan on only one source . Multiple source domains can correspond to different types of channels (e.g., LoS vs. NLoS channels, Indoor vs. outdoor, etc.), scenarios (e.g., urban macrocells (UMa), urban microcells (UMi), indoor hot spot, etc.) and changing prior probabilities of the input messages, and so forth. The changes in distribution at the decoder input could be due to changes in the channel, or different encoder mappings (originating from encoders trained differently and not jointly with the decoder).

[0121] A message ^^is encoded into symbol ^^and then ^^is sent over the channel. The channel corrupts or ^the symbols. Let !^ ∈ ℂ denote the channel output for input ^^ ∈ℂ^. The decoder ^^is trained such that it decodeschannel output !^(also referred to as encoded messages) tothe input message ^^carried by or contained in !^with a high reliability. In other words, the decoder is trained suchit classifies the channel output into one of the ^ messages by reducing (e.g., minimizing) the average probability of error in the classification. Thus, the decoder NN is a function that maps channel outputs into messages and can be expressed as ^^: ℂ^ → ℳ, where ^ = {Q3, Q^, … , QR} denotes the set of learnable or trainable parameters withor trainable parameters of the decoder NN model. Completion of training implies that a set of parameters ^ have been learned for the decoder neural network. It should be noted that the learned mapping ^^, such as when the underlying AI / ML model is a neural network, produces a conditional distribution )&^|!; ^( that is differentiable in ^. It should also benoted that )&^|!; ^( is the distribution of ^ conditioned on !, parametrized by ^.

[0122] As ^ is a discrete variable with ^ ∈ ℳ = {1, … , ^}, )&^|!; ^( is a probability massfunction, e.g., )&^|!; ^( is a scalar value, denoting the probability that ^ is the predicted label for! under the model parameters ^. In other words, )L^6T!U; ^M is the probability that the channeloutput V^contained the message ^6, as per the prediction or inference made by the decoder NN with ^ as its model parameters and ∑X6>3 )L^6T!U; ^M = 1, or, simply, ∑ZN∈ℳ )L^6T!^; YM = 1.Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 27

[0123] When a decoder is trained, it is trained with data from one or more source domains, 66 N 6 6 9;6 = {L! , ^ M}? ∼ ) , 7 6 :^ ^ ^>3 [X = 1, … , K, where )[X ≠ )[X , for 7 ≠ 7 and K ≥ 1. When the^ = {&!^ ^( ?O^ , ^^ }^>3 ∼)[X , where )[X ≠ )[X , for some or all 7 = 1, … , K. In other words,the inference phase,the decoder might !^^sequentially one after the other and it has to infer orpredict the label ^^for each !^. However, as the decoder may not be trained over the target domain, it is bound to make more errors in decoding the channel output samples !^^into thetransmitted messages ^^^.

[0124] Withto decoder adaptation or generalization, the following describes how the decoder of an autoencoder can be adapted to an unseen target domain with a single unlabeled sample from the target domain. It should be noted that the described decoder adaptation can also be applied to the case of a batch or set of samples from the target domain and the samples may be one or both of unlabeled or labeled. Further, it should be noted that the discussed adaptation techniques can be used to adapt all the parameters of the decoder neural network or only some of the parameters of the decoder neural network. Accordingly, the discussed techniques can be employed to adapt parameters of one or more layers of the decoder NN, and adapting the entire decoder NN corresponds to adapting all the layers of the decoder NN.

[0125] The process of adapting the decoder is triggered or initiated by the decoder or another entity or node at the receiving end of the communication link, or by the encoder or a node or entity at the transmitting end of the communication link, or by a different entity that monitors the communication between the transmitting end and the receiving end. When to initiate or trigger the decoder adaptation process can be decided by observing decoding error probability or transport block (TB) / CB successful / unsuccessful decoding (e.g., based on CRC pass / fail) or acknowledgement (ACK) / negative-acknowledgement (NACK) feedback indication or retransmission indications or some other measure of the quality of the decoded messages and is a part of the model monitoring procedure.

[0126] When the decoder is to be adapted, at least one sample from the target domain is used, where the sample can be an unlabeled sample and its corresponding label need not be known. In other words, the decoder NN receives at least one corrupted symbol y from the wireless channel, Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 28 where the symbol belongs to a target domain that may have a different distribution than the one or more source distributions over which the decoder NN was trained, and the decoder NN does not know what the transmitted message ^ contained in the channel output sample V is. Thus, the sample !, based on which the decoder is adapted, need not correspond to a pilot symbol.

[0127] Adaptation of the decoder NN involves data augmentation of the received one or more samples from the test domain. Data augmentation is a concept being employed by AI / ML community. Data augmentation of a given data sample can be stated as “generating slightly modified versions of the given data sample”. A data augmentation function is a function that generates a slightly modified version of the input data sample. Any of a variety of augmentation methods can be used, such as adding noise to the encoded message, multiplying the encoded message by a phaser, and so forth. In some examples, a data augmentation function may depend on one or more parameters such as training domain statistics or distribution, channel statistics or distribution at training time, received signal to noise ratio (SNR) or signal to interference noise ratio (SINR), a measure of the difference between the training domain statistics and target domain statistics, etc. In some examples, the device or node performing the decoder NN adaptation may receive assistance information from a second device or node (e.g., the device or node performing the encoder operation or another device or node) including an indication of one or more parameters for the data augmentation function.

[0128] The following is a technique for generalization or adaptation of decoder NN ^]to the target domain.

[0129] Using data augmentation, multiple data samples are generated using the available at least a single target sample !, and multiple data augmentation functions. A number Q of different dataaugmentation functions α3&⋅(, … , αa&⋅( are selected, and a number Q of augmented data samples!b^, ^ = 1, ... , Q, where !b^ = α^&!(, ^ = 1, … , c. Let !b^>d = !bd = ! are generated.

[0130] As discussed above, the mapping ^^ produces a conditional distribution )&^|!; ^( thatis differentiable in ^. )&^|!b^; ^( for ^ = 0,… , Q, where !bd = ! is computed.

[0131] The average conditional distribution )e&^|!; ^( is computed whereAttorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 29 a )e&^|!; ^( = 1+ 1 , )&^|!b^; ^(

[0132] The AI / ML model is for the original target sample !and each of the augmented = = … , from ! through theaugmentation functions α3&⋅(, … , αa&⋅(. Further, the model is expected to have a high confidence inpredicting the same label for all the samples, !, !b3, ... , !ba. Hence, the decoder NN can be adaptedby employing g h)e&^|!; ^(i, the entropy of the average conditional distribution as the lossfunction. Thus, the loss function is given by, ℓ&^; !( = g h)e&^|!; ^(i = − , k ) &^|!; ^( =lP )e&^|!; ^(

[0133] Fora != = ^ = −1 =lP ^ = ^^ ^(

[0134] Asis differentiablein ^, loss function minimization, or, equivalently, entropy minimization, canperformed through gradient based methods. However, instead of straightforward minimization of the entropy, we employ minimizing the entropy as well as the sharpness of the entropy as explained in the following.

[0135] For the decoder NN to achieve a better generalization, the decoder NN weights are adapted through a min-max optimization ^^ = ^^^m ‖ ^q‖nso ℓ&^ + q; !(r t

[0136] Here, ^^denotes the NN. Note that ^^={Q^,3, … , Q^,R}, ^ = {Q3, Q^, … , QR}, and q = {u3, u^, … , uR}. Also note that K is the number oflearnable or trainable parameters of the decoder NN model.

[0137] Finding ^^ through ^^ = ^^^m ℓ&^; !(, a straightforward minimization of the lossfunction, gives parameter values that have low loss. By determining ^^through the min-max optimization as stated by the above equation with v as a tuning or hyper-parameter, the parameter Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 30 values that have low loss value and whose entire neighborhoods have uniformly low loss value are determined. This makes the model adaptation insensitive to large gradients possibly contributed by the current sample being used for adaptation and enables the model to reach a flat area of the entropy loss surface. Thus, the model adaptation becomes robust to noisy or large gradients.

[0138] Figure 3 illustrates an example of decoder generalization or adaptation process in accordance with aspects of the present disclosure. The decoder generalization or adaptation process of Figure 3 is using a single unlabeled sample from the target domain. The input to the decoder generalization or adaptation process is a decoder NN model ^^, a single unlabeled sample from target domain !, and a parameter η. The output from the decoder generalization or adaptation process is an adapted decoder NN model ^^O.

[0139] At 302, a number of different augmentation functions is chosen. A number c of differentaugmentation functions, α3&⋅(, … , αa&⋅( is chosen.

[0140] At 304, using each of the selected augmentation function, generate a single augmented data sample. For example, using each of the selected augmentation function α^&⋅(, a singleaugmented data sample !bG = α^&!(, ^ = 1, … , Q. Set !bd = ! is generated.

[0141] At 306, an average conditional distribution is computed. For example, )e&^|!; ^( =3ax3 ∑a ^>d )&^|!bG; ^(is computed.function is computed. For example, the following loss function is computed: ℓ&^; !( = g h)e&^|!; ^(i = − ∑0∈ℳ )e&^|!; ^( =lP )e&^|!; ^( .

[0143] Atoptimization problem. For example, the adapted decoder NN parameters are determined by solving the optimization problem, ^^ = ^^^m ^ ℓ&^ + q; !(.

[0144] One advantage of theabove, is the following. The decoder can continuously adapt itself based on the received symbols. Such a continuously self- Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 31 adaptive decoder is of great value for any practical communication system that encounters continuously changing channel conditions and other hardware related impairments such as transmit filter, receive filter, and so forth. Such a continuous adaptation also adapts the decoder even for minor changes in its input data statistics, thereby avoiding a sudden degradation in its performance. It should be noted that the continuous adaptation of the decoder does not need pilot or reference symbols, which helps greatly in saving the valuable network resources, such as time, bandwidth and energy that would need to be spent on transmitting reference symbols.

[0145] It should be noted that the decoder NN can also become more tuned for giving better performance over the target domain with multiple successive adaptations based on unlabeled samples received from the target domain sequentially, or by performing model adaptation based on a set or batch of samples (unlabeled and / or labeled).

[0146] The techniques discussed herein consider the problem of adapting or generalizing the decoder in an autoencoder based E2E wireless communication system or, more generally, E2E communication based on two sided AI / ML models.

[0147] The techniques discussed herein are applicable to any off-the-shelf decoder. As the techniques are an inference or test time generalization method, they can be applied to any readily available pre-trained off-the-shelf decoder, irrespective of how it has been trained.

[0148] The techniques discussed herein are a truly online method. The decoder neural network can adapt itself to new, unseen data distributions based on every received symbol from the transmitter (e.g., also referred to as online), sequentially, one at a time, without requiring the symbol to be a pilot or reference symbol. The techniques do not require the samples from the target domain to be available in a batch-mode.

[0149] The techniques discussed herein are resource efficient. Network resources (e.g., time, bandwidth) for model adaptation need not be spent as the techniques discussed herein do not require reference symbol transmissions to adapt the decoder NN.

[0150] The techniques discussed herein are provide for the scenario of single decoder, multiple encoders in wireless networks, where the encoder and the decoder of autoencoder comes from two different vendors or suppliers. For example, a gNB can employ its own encoder for transmitting Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 32 data across the wireless channel and different UEs in the network can employ their own decoders that are not necessarily jointly developed with the encoder deployed at the gNB.

[0151] The techniques discussed herein are a source-free method. Some of the DG methods require access to data from the one or more source domains over which the model was trained. Getting access to source data over which the model has been trained may not always be possible. For example, the vendor who supplied the model may not be willing to share the source data (or, even the distribution of the source domain). The techniques discussed herein do not require source data, not even the distribution of the source data.

[0152] The techniques discussed herein allow the decoder to adapt itself continuously, without the need of pilot signals, which helps to avoid the possibility of sudden degradation in decoder performance.

[0153] Figure 4 illustrates an example of a UE 400 in accordance with aspects of the present disclosure. The UE 400 may include a processor 402, a memory 404, a controller 406, and a transceiver 408. The processor 402, the memory 404, the controller 406, or the transceiver 408, or various combinations thereof or various components thereof may be examples of means for performing various aspects of the present disclosure as described herein. These components may be coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more interfaces.

[0154] The processor 402, the memory 404, the controller 406, or the transceiver 408, or various combinations or components thereof may be implemented in hardware (e.g., circuitry). The hardware may include a processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or other programmable logic device, or any combination thereof configured as or otherwise supporting a means for performing the functions described in the present disclosure.

[0155] The processor 402 may include an intelligent hardware device (e.g., a general-purpose processor, a DSP, a CPU, an ASIC, an FPGA, or any combination thereof). In some implementations, the processor 402 may be configured to operate the memory 404. In some other implementations, the memory 404 may be integrated into the processor 402. The processor 402 may Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 33 be configured to execute computer-readable instructions stored in the memory 404 to cause the UE 400 to perform various functions of the present disclosure.

[0156] The memory 404 may include volatile or non-volatile memory. The memory 404 may store computer-readable, computer-executable code including instructions when executed by the processor 402 cause the UE 400 to perform various functions described herein. The code may be stored in a non-transitory computer-readable medium such as the memory 404 or another type of memory. Computer-readable media includes both non-transitory computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A non-transitory storage medium may be any available medium that may be accessed by a general-purpose or special-purpose computer.

[0157] In some implementations, the processor 402 and the memory 404 coupled with the processor 402 may be configured to cause the UE 400 to perform one or more of the functions described herein (e.g., executing, by the processor 402, instructions stored in the memory 404). For example, the processor 402 may support wireless communication at the UE 400 in accordance with examples as disclosed herein. The UE 400 may be configured to or operable to support a means for receiving at least one encoded message; generating one or more modified versions of the at least one encoded message; and updating one or more layers of a decoder neural network based at least in part on the at least one encoded message, the one or more modified versions of the at least one encoded message, one or more tuning parameters, and a loss function.

[0158] Additionally, the UE 400 may be configured to support any one or combination of updating the one or more layers of the decoder neural network by minimizing the loss function; where minimizing the loss function comprises minimizing maximum entropy in a region through min-max optimization; where the at least one encoded message includes multiple encoded messages and the one or more modified versions of the at least one encoded message includes one or more modified versions of each of the multiple encoded messages, and further including updating the one or more layers of the decoder neural network based at least in part on the multiple encoded messages and the one or more modified versions of each of the multiple encoded messages; where the multiple encoded messages are received sequentially in time domain, frequency domain, or spatial domain; determining to update the one or more layers of the decoder neural network based at least in part on periodicity information indicating the one or more layers of the decoder neural Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 34 network are to be updated at regular intervals of time; determining to update the one or more layers of the decoder neural network based at least in part on a change in one or more characteristics of one or more received encoded messages; where the one or more characteristics are statistical characteristics of the one or more received encoded messages; where the statistical characteristics include at least one of a mean, a median, a standard deviation, or a probability distribution; determining to update the one or more layers of the decoder neural network based at least in part on an indication, received from an apparatus, to update the one or more layers of the decoder neural network; determining to update the one or more layers of the decoder neural network based at least in part on a change in a value of at least one parameter corresponding to a channel, where the channel corresponds to a communication medium between the UE and an apparatus monitoring a quality of the at least one encoded message, where the quality of the at least one encoded message comprises one or more of an indication of decoding success of the at least one encoded message, an indication of cyclical redundancy check status of the at least one encoded message, an acknowledgement indication for the at least one encoded message, a retransmission indication for the at least one encoded message, or a decoding error probability for the at least one encoded message above a threshold; and determining to update the one or more layers of the decoder neural network based at least in part on the monitoring; where the receiving comprises receiving the at least one encoded message from a first apparatus, and determining to update the one or more layers of the decoder neural network based at least in part on an indication or a configuration, received from a second apparatus, to update the one or more layers of the decoder neural network; where the first apparatus comprises a base station, and the second apparatus comprises the base station or another NE; where each encoded message of the at least one encoded message comprises one of multiple different information messages; determining the one or more layers of the decoder neural network; where the receiving comprises receiving the at least one encoded message from a first apparatus, and receiving, from the first apparatus or a second apparatus, an indication of the one or more layers of the decoder neural network; where the first apparatus comprises a base station, and the second apparatus comprises another NE; where the one or more tuning parameters comprises one or both of a first tuning parameter that is a number of the one or more modified versions for each of the at least one encoded message, and a second tuning parameter that is a setting of the loss function; determining the one or more tuning parameters; or receiving, from a first apparatus or a second apparatus, the one or more tuning parameters; where the first apparatus comprises a base Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 35 station, and the second apparatus comprises another NE; the updating including updating values of a number N of parameters of the one or more layers by: generating a number Q of different augmented received encoded messages where each augmented received encoded message is a modified version of at least one encoded message, and Q is a first tuning parameter of the one or more tuning parameters; obtaining a set of Q+1 different data samples, where a first data sample is the at least one encoded message and remaining Q data samples are the Q number of different augmented received encoded messages generated from the at least one encoded message; for each of the Q+1 different data samples, using the decoder neural network to compute a vector of length M where an ^^^element of the vector is a probability that an information message contained in thedata sample is ^^, where ^^ ∈ {1, … , ^} and the at least one encoded message comprises one of Mmultiple different information messages; adding all the Q+1 number of vectors, each of length M, using element-wise addition to generate a temporary vector; dividing each element of the temporary vector by Q+1 to generate a mean probability vector; computing an entropy of the mean probability vector, where the entropy is a function of parameters of the decoder neural network; and updating the one or more layers of the decoder neural network by determining new values of the N parameters that minimizes a maximum entropy within an N-dimensional Euclidean ball, with the N- dimensional Euclidean ball centered at the new values of the N parameters with radius R, where R is a second tuning parameter of the one or more tuning parameters; where the at least one encoded message is based at least in part on an encoder neural network; where the encoder neural network is trained based on a first set of training data and the decoder neural network is trained based on a second set of training data, and where the first set of training data is different than the second set of training data; where the encoder neural network and the decoder neural network are trained on a same set of training data; transmitting, to a first apparatus, an assistance information request message for updating one or more parameters of the decoder neural network; receiving, from the first apparatus in response to the assistance information request message, at least one of a signal or an additional encoded message; receiving, from the first apparatus based at least in part on the assistance information request message, at least one of a signal or an additional encoded message; receiving, from the first apparatus, an indication to determine whether to update the one or more layers of the decoder neural network; updating, in response to successfully decoding the at least one encoded message, the one or more layers of the decoder neural network based at least in part on the at least one encoded message; updating, based at least in part on successfully decoding the at least Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 36 one encoded message, the one or more layers of the decoder neural network based at least in part on the at least one encoded message.

[0159] In some implementations, the processor 402 and the memory 404 coupled with the processor 402 may be configured to cause the UE 400 to perform one or more of the functions described herein (e.g., executing, by the processor 402, instructions stored in the memory 404). For example, the processor 402 may support wireless communication at the UE 400 in accordance with examples as disclosed herein. The UE 400 may be configured to or operable to support a means for generating, using an encoder neural network, at least one encoded message; transmitting the at least one encoded message; and transmitting an indication for an apparatus to update one or more layers of a decoder neural network used by the apparatus to decode the at least one encoded message.

[0160] Additionally, the UE 400 may be configured to support any one or combination of where the at least one encoded message includes multiple encoded messages transmitted sequentially in time domain, frequency domain, or spatial domain; where each encoded message of the at least one encoded message comprises one of multiple different information messages; transmitting one or more tuning parameters used by the apparatus to update one or more layers of the decoder neural network; where the encoder neural network is trained based on a first set of training data and the decoder neural network is trained based on a second set of training data, and where the first set of training data is different than the second set of training data; where the encoder neural network and the decoder neural network are trained on a same set of training data; receiving an assistance information request message for updating one or more parameters of the decoder neural network; transmitting, in response to the assistance information request message, at least one of a signal or an additional encoded message; transmitting, based at least in part on the assistance information request message, at least one of a signal or an additional encoded message; transmitting an indication for the apparatus to determine whether to update the one or more layers of the decoder neural network; transmitting an indication for the apparatus to determine whether to update the one or more layers of the decoder neural network; where the apparatus comprises a UE.

[0161] Additionally, or alternatively, the UE 400 may support at least one memory (e.g., the memory 404) and at least one processor (e.g., the processor 402) coupled with the at least one memory and configured to cause the UE to: receive at least one encoded message; generate one or more modified versions of the at least one encoded message; and update one or more layers of a Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 37 decoder neural network based at least in part on the at least one encoded message, the one or more modified versions of the at least one encoded message, one or more tuning parameters, and a loss function.

[0162] Additionally, the UE 400 may be configured to support any one or combination of where the at least one processor is further configured to cause the first apparatus to update the one or more layers of the decoder neural network by minimizing the loss function; where minimizing the loss function comprises minimizing maximum entropy in a region through min-max optimization; where the at least one encoded message includes multiple encoded messages and the one or more modified versions of the at least one encoded message includes one or more modified versions of each of the multiple encoded messages, and where the at least one processor is further configured to cause the first apparatus to update the one or more layers of the decoder neural network based at least in part on the multiple encoded messages and the one or more modified versions of each of the multiple encoded messages; where the multiple encoded messages are received sequentially in time domain, frequency domain, or spatial domain; to cause the first apparatus to determine to update the one or more layers of the decoder neural network based at least in part on periodicity information indicating the one or more layers of the decoder neural network are to be updated at regular intervals of time; where the at least one processor is further configured to cause the first apparatus to determine to update the one or more layers of the decoder neural network based at least in part on a change in one or more characteristics of one or more received encoded messages; where the one or more characteristics are statistical characteristics of the one or more received encoded messages; where the statistical characteristics include at least one of a mean, a median, a standard deviation, or a probability distribution; where the at least one processor is further configured to cause the first apparatus to determine to update the one or more layers of the decoder neural network based at least in part on an indication, received from a second apparatus, to update the one or more layers of the decoder neural network; where the at least one processor is further configured to cause the first apparatus to determine to update the one or more layers of the decoder neural network based at least in part on a change in a value of at least one parameter corresponding to a channel, where the channel corresponds to a communication medium between the first apparatus and a second apparatus; where the at least one processor is further configured to cause the first apparatus to: monitor a quality of the at least one encoded message, where the quality of the at least one encoded Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 38 message comprises one or more of an indication of decoding success of the at least one encoded message, an indication of cyclical redundancy check status of the at least one encoded message, an acknowledgement indication for the at least one encoded message, a retransmission indication for the at least one encoded message, or a decoding error probability for the at least one encoded message above a threshold; and determine to update the one or more layers of the decoder neural network based at least in part on the monitoring; where the at least one processor is further configured to cause the first apparatus to receive the at least one encoded message from a second apparatus, and to determine to update the one or more layers of the decoder neural network based at least in part on an indication or a configuration, received from a third apparatus, to update the one or more layers of the decoder neural network; where the first apparatus comprises a UE, the second apparatus comprises a base station, and the third apparatus comprises the base station or another NE; where each encoded message of the at least one encoded message comprises one of multiple different information messages; where the at least one processor is further configured to cause the first apparatus to determine the one or more layers of the decoder neural network; where the at least one processor is further configured to cause the first apparatus to receive the at least one encoded message from a second apparatus, and to receive, from the second apparatus or a third apparatus, an indication of the one or more layers of the decoder neural network; where the first apparatus comprises a UE, the second apparatus comprises a base station, and the third apparatus comprises another NE; where the one or more tuning parameters comprises one or both of a first tuning parameter that is a number of the one or more modified versions for each of the at least one encoded message, and a second tuning parameter that is a setting of the loss function; where the at least one processor is further configured to cause the first apparatus to: determine the one or more tuning parameters; or receive, from a second apparatus or a third apparatus, the one or more tuning parameters; where the first apparatus comprises a UE, the second apparatus comprises a base station, and the third apparatus comprises another NE; where the at least one processor is further configured, to update values of a number N of parameters of the one or more layers of the decoder neural network based at least in part on the at least one encoded message, to cause the first apparatus to: generate a number Q of different augmented received encoded messages where each augmented received encoded message is a modified version of at least one encoded message, and Q is a first tuning parameter of the one or more tuning parameters; obtain a set of Q+1 different data samples, where a first data sample is the at least one encoded message and remaining Q data Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 39 samples are the Q number of different augmented received encoded messages generated from the at least one encoded message; for each of the Q+1 different data samples, use the decoder neural network to compute a vector of length M where an ^^^element of the vector is a probability that aninformation message contained in the data sample is ^^, where ^^ ∈ {1, … , ^} and the at least oneencoded message comprises one of M multiple different information messages; add all the Q+1 number of vectors, each of length M, using element-wise addition to generate a temporary vector; divide each element of the temporary vector by Q+1 to generate a mean probability vector; compute an entropy of the mean probability vector, where the entropy is a function of parameters of the decoder neural network; and update the one or more layers of the decoder neural network by determining new values of the N parameters that minimizes a maximum entropy within an N- dimensional Euclidean ball, with the N-dimensional Euclidean ball centered at the new values of the N parameters with radius R, where R is a second tuning parameter of the one or more tuning parameters; where the at least one encoded message is based at least in part on an encoder neural network; where the encoder neural network is trained based on a first set of training data and the decoder neural network is trained based on a second set of training data, and where the first set of training data is different than the second set of training data; where the encoder neural network and the decoder neural network are trained on a same set of training data; where the at least one processor is further configured to cause the first apparatus to transmit, to a second apparatus, an assistance information request message for updating one or more parameters of the decoder neural network; where the at least one processor is further configured to cause the first apparatus to receive, from the second apparatus in response to the assistance information request message, at least one of a signal or an additional encoded message; where the at least one processor is further configured to cause the first apparatus to receive, from the second apparatus based at least in part on the assistance information request message, at least one of a signal or an additional encoded message; where the at least one processor is further configured to cause the first apparatus to receive, from a second apparatus, an indication to determine whether to update the one or more layers of the decoder neural network; where the at least one processor is further configured to cause the first apparatus to update, in response to successfully decoding the at least one encoded message, the one or more layers of the decoder neural network based at least in part on the at least one encoded message; where the at least one processor is further configured to cause the first apparatus to update, based at least in part on successfully decoding the at least one encoded message, the one Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 40 or more layers of the decoder neural network based at least in part on the at least one encoded message; where the first apparatus comprises a UE.

[0163] Additionally, or alternatively, the UE 400 may support at least one memory (e.g., the memory 404) and at least one processor (e.g., the processor 402) coupled with the at least one memory and configured to cause the UE to: generate, using an encoder neural network, at least one encoded message; transmit the at least one encoded message; and transmit an indication for a second apparatus to update one or more layers of a decoder neural network used by the second apparatus to decode the at least one encoded message.

[0164] Additionally, the UE 400 may be configured to support any one or combination of where the at least one encoded message includes multiple encoded messages transmitted sequentially in time domain, frequency domain, or spatial domain; where each encoded message of the at least one encoded message comprises one of multiple different information messages; where the at least one processor is further configured to cause the first apparatus to transmit one or more tuning parameters used by the second apparatus to update one or more layers of the decoder neural network; where the encoder neural network is trained based on a first set of training data and the decoder neural network is trained based on a second set of training data, and where the first set of training data is different than the second set of training data; where the encoder neural network and the decoder neural network are trained on a same set of training data; where the at least one processor is further configured to cause the first apparatus to receive an assistance information request message for updating one or more parameters of the decoder neural network; where the at least one processor is further configured to cause the first apparatus to transmit, in response to the assistance information request message, at least one of a signal or an additional encoded message; where the at least one processor is further configured to cause the first apparatus to transmit, based at least in part on the assistance information request message, at least one of a signal or an additional encoded message; where the at least one processor is further configured to cause the first apparatus to transmit an indication for the second apparatus to determine whether to update the one or more layers of the decoder neural network; where the first apparatus comprises a base station and the second apparatus comprises a UE; where the at least one processor is further configured to cause the first apparatus to transmit an indication for the second apparatus to determine whether to update the one or more layers of the decoder neural network. Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 41

[0165] The controller 406 may manage input and output signals for the UE 400. The controller 406 may also manage peripherals not integrated into the UE 400. In some implementations, the controller 406 may utilize an operating system such as iOS®, ANDROID®, WINDOWS®, or other operating systems. In some implementations, the controller 406 may be implemented as part of the processor 402.

[0166] In some implementations, the UE 400 may include at least one transceiver 408. In some other implementations, the UE 400 may have more than one transceiver 408. The transceiver 408 may represent a wireless transceiver. The transceiver 408 may include one or more receiver chains 410, one or more transmitter chains 412, or a combination thereof.

[0167] A receiver chain 410 may be configured to receive signals (e.g., control information, data, packets) over a wireless medium. For example, the receiver chain 410 may include one or more antennas to receive a signal over the air or wireless medium. The receiver chain 410 may include at least one amplifier (e.g., a low-noise amplifier (LNA)) configured to amplify the received signal. The receiver chain 410 may include at least one demodulator configured to demodulate the receive signal and obtain the transmitted data by reversing the modulation technique applied during transmission of the signal. The receiver chain 410 may include at least one decoder for decoding the demodulated signal to receive the transmitted data.

[0168] A transmitter chain 412 may be configured to generate and transmit signals (e.g., control information, data, packets). The transmitter chain 412 may include at least one modulator for modulating data onto a carrier signal, preparing the signal for transmission over a wireless medium. The at least one modulator may be configured to support one or more techniques such as amplitude modulation (AM), frequency modulation (FM), or digital modulation schemes like phase-shift keying (PSK) or quadrature amplitude modulation (QAM). The transmitter chain 412 may also include at least one power amplifier configured to amplify the modulated signal to an appropriate power level suitable for transmission over the wireless medium. The transmitter chain 412 may also include one or more antennas for transmitting the amplified signal into the air or wireless medium.

[0169] Figure 5 illustrates an example of a processor 500 in accordance with aspects of the present disclosure. The processor 500 may be an example of a processor configured to perform various operations in accordance with examples as described herein. The processor 500 may include Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 42 a controller 502 configured to perform various operations in accordance with examples as described herein. The processor 500 may optionally include at least one memory 504, which may be, for example, an L1 / L2 / L3 cache. Additionally, or alternatively, the processor 500 may optionally include one or more arithmetic-logic units (ALUs) 506. One or more of these components may be in electronic communication or otherwise coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more interfaces (e.g., buses).

[0170] The processor 500 may be a processor chipset and include a protocol stack (e.g., a software stack) executed by the processor chipset to perform various operations (e.g., receiving, obtaining, retrieving, transmitting, outputting, forwarding, storing, determining, identifying, accessing, writing, reading) in accordance with examples as described herein. The processor chipset may include one or more cores, one or more caches (e.g., memory local to or included in the processor chipset (e.g., the processor 500) or other memory (e.g., random access memory (RAM), read-only memory (ROM), dynamic RAM (DRAM), synchronous dynamic RAM (SDRAM), static RAM (SRAM), ferroelectric RAM (FeRAM), magnetic RAM (MRAM), resistive RAM (RRAM), flash memory, phase change memory (PCM), and others).

[0171] The controller 502 may be configured to manage and coordinate various operations (e.g., signaling, receiving, obtaining, retrieving, transmitting, outputting, forwarding, storing, determining, identifying, accessing, writing, reading) of the processor 500 to cause the processor 500 to support various operations in accordance with examples as described herein. For example, the controller 502 may operate as a control unit of the processor 500, generating control signals that manage the operation of various components of the processor 500. These control signals include enabling or disabling functional units, selecting data paths, initiating memory access, and coordinating timing of operations.

[0172] The controller 502 may be configured to fetch (e.g., obtain, retrieve, receive) instructions from the memory 504 and determine subsequent instruction(s) to be executed to cause the processor 500 to support various operations in accordance with examples as described herein. The controller 502 may be configured to track memory addresses of instructions associated with the memory 504. The controller 502 may be configured to decode instructions to determine the operation to be performed and the operands involved. For example, the controller 502 may be configured to interpret the instruction and determine control signals to be output to other components of the Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 43 processor 500 to cause the processor 500 to support various operations in accordance with examples as described herein. Additionally, or alternatively, the controller 502 may be configured to manage flow of data within the processor 500. The controller 502 may be configured to control transfer of data between registers, ALUs 506, and other functional units of the processor 500.

[0173] The memory 504 may include one or more caches (e.g., memory local to or included in the processor 500 or other memory, such as RAM, ROM, DRAM, SDRAM, SRAM, MRAM, flash memory, etc. In some implementations, the memory 504 may reside within or on a processor chipset (e.g., local to the processor 500). In some other implementations, the memory 504 may reside external to the processor chipset (e.g., remote to the processor 500).

[0174] The memory 504 may store computer-readable, computer-executable code including instructions that, when executed by the processor 500, cause the processor 500 to perform various functions described herein. The code may be stored in a non-transitory computer-readable medium such as system memory or another type of memory. The controller 502 and / or the processor 500 may be configured to execute computer-readable instructions stored in the memory 504 to cause the processor 500 to perform various functions. For example, the processor 500 and / or the controller 502 may be coupled with or to the memory 504, the processor 500, and the controller 502, and may be configured to perform various functions described herein. In some examples, the processor 500 may include multiple processors and the memory 504 may include multiple memories. One or more of the multiple processors may be coupled with one or more of the multiple memories, which may, individually or collectively, be configured to perform various functions herein.

[0175] The one or more ALUs 506 may be configured to support various operations in accordance with examples as described herein. In some implementations, the one or more ALUs 506 may reside within or on a processor chipset (e.g., the processor 500). In some other implementations, the one or more ALUs 506 may reside external to the processor chipset (e.g., the processor 500). One or more ALUs 506 may perform one or more computations such as addition, subtraction, multiplication, and division on data. For example, one or more ALUs 506 may receive input operands and an operation code, which determines an operation to be executed. One or more ALUs 506 may be configured with a variety of logical and arithmetic circuits, including adders, subtractors, shifters, and logic gates, to process and manipulate the data according to the operation. Additionally, or alternatively, the one or more ALUs 506 may support logical operations such as Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 44 AND, OR, exclusive-OR (XOR), not-OR (NOR), and not-AND (NAND), enabling the one or more ALUs 506 to handle conditional operations, comparisons, and bitwise operations.

[0176] The processor 500 may support wireless communication in accordance with examples as disclosed herein. The processor 500 may be configured to or operable to support at least one controller (e.g., the controller 502) coupled with at least one memory (e.g., the memory 504) and configured to cause the processor to: receive at least one encoded message; generate one or more modified versions of the at least one encoded message; and update one or more layers of a decoder neural network based at least in part on the at least one encoded message, the one or more modified versions of the at least one encoded message, one or more tuning parameters, and a loss function.

[0177] Additionally, the processor 500 may be configured to or operable to support any one or combination of where the at least one controller is further configured to cause the processor to update the one or more layers of the decoder neural network by minimizing the loss function; where minimizing the loss function comprises minimizing maximum entropy in a region through min-max optimization; where the at least one encoded message includes multiple encoded messages and the one or more modified versions of the at least one encoded message includes one or more modified versions of each of the multiple encoded messages, and where the at least one controller is further configured to cause the processor to update the one or more layers of the decoder neural network based at least in part on the multiple encoded messages and the one or more modified versions of each of the multiple encoded messages; where the multiple encoded messages are received sequentially in time domain, frequency domain, or spatial domain; where the at least one controller is further configured to cause the processor to determine to update the one or more layers of the decoder neural network based at least in part on periodicity information indicating the one or more layers of the decoder neural network are to be updated at regular intervals of time; where the at least one controller is further configured to cause the processor to determine to update the one or more layers of the decoder neural network based at least in part on a change in one or more characteristics of one or more received encoded messages; where the one or more characteristics are statistical characteristics of the one or more received encoded messages; where the statistical characteristics include at least one of a mean, a median, a standard deviation, or a probability distribution; where the at least one controller is further configured to cause the processor to determine to update the one or more layers of the decoder neural network based at least in part on an indication, received from Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 45 an apparatus, to update the one or more layers of the decoder neural network; where the at least one controller is further configured to cause the processor to determine to update the one or more layers of the decoder neural network based at least in part on a change in a value of at least one parameter corresponding to a channel, where the channel corresponds to a communication medium between the processor and an apparatus; where the at least one controller is further configured to cause the processor to: monitor a quality of the at least one encoded message, where the quality of the at least one encoded message comprises one or more of an indication of decoding success of the at least one encoded message, an indication of cyclical redundancy check status of the at least one encoded message, an acknowledgement indication for the at least one encoded message, a retransmission indication for the at least one encoded message, or a decoding error probability for the at least one encoded message above a threshold; and determine to update the one or more layers of the decoder neural network based at least in part on the monitoring; where the at least one controller is further configured to cause the processor to receive the at least one encoded message from a first apparatus, and to determine to update the one or more layers of the decoder neural network based at least in part on an indication or a configuration, received from a second apparatus, to update the one or more layers of the decoder neural network; where the processor is included in a UE, the first apparatus comprises a base station, and the second apparatus comprises the base station or another NE; where each encoded message of the at least one encoded message comprises one of multiple different information messages; where the at least one controller is further configured to cause the processor to determine the one or more layers of the decoder neural network; where the at least one controller is further configured to cause the processor to receive the at least one encoded message from a first apparatus, and to receive, from the first apparatus or a second apparatus, an indication of the one or more layers of the decoder neural network; where the processor is included in a UE, the first apparatus comprises a base station, and the second apparatus comprises another NE; where the one or more tuning parameters comprises one or both of a first tuning parameter that is a number of the one or more modified versions for each of the at least one encoded message, and a second tuning parameter that is a setting of the loss function; where the at least one controller is further configured to cause the processor to: determine the one or more tuning parameters; or receive, from a first apparatus or a second apparatus, the one or more tuning parameters; where the processor is included in a UE, the first apparatus comprises a base station, and the second apparatus comprises another NE; where the at least one controller is further configured, to update values of a Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 46 number N of parameters of the one or more layers of the decoder neural network based at least in part on the at least one encoded message, to cause the processor to: generate a number Q of different augmented received encoded messages where each augmented received encoded message is a modified version of at least one encoded message, and Q is a first tuning parameter of the one or more tuning parameters; obtain a set of Q+1 different data samples, where a first data sample is the at least one encoded message and remaining Q data samples are the Q number of different augmented received encoded messages generated from the at least one encoded message; for each of the Q+1 different data samples, use the decoder neural network to compute a vector of length M where an ^^^element of the vector is a probability that an information message contained in the datasample is ^^, where ^^ ∈ {1, … , ^} and the at least one encoded message comprises one of Mmultiple different information messages; add all the Q+1 number of vectors, each of length M, using element-wise addition to generate a temporary vector; divide each element of the temporary vector by Q+1 to generate a mean probability vector; compute an entropy of the mean probability vector, where the entropy is a function of parameters of the decoder neural network; and update the one or more layers of the decoder neural network by determining new values of the N parameters that minimizes a maximum entropy within an N-dimensional Euclidean ball, with the N- dimensional Euclidean ball centered at the new values of the N parameters with radius R, where R is a second tuning parameter of the one or more tuning parameters; where the at least one encoded message is based at least in part on an encoder neural network; where the encoder neural network is trained based on a first set of training data and the decoder neural network is trained based on a second set of training data, and where the first set of training data is different than the second set of training data; where the encoder neural network and the decoder neural network are trained on a same set of training data; where the at least one controller is further configured to cause the processor to transmit, to a first apparatus, an assistance information request message for updating one or more parameters of the decoder neural network; where the at least one controller is further configured to cause the processor to receive, from the first apparatus in response to the assistance information request message, at least one of a signal or an additional encoded message; where the at least one controller is further configured to cause the processor to receive, from the first apparatus based at least in part on the assistance information request message, at least one of a signal or an additional encoded message; where the at least one controller is further configured to cause the processor to receive, from the first apparatus, an indication to determine whether to update the one Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 47 or more layers of the decoder neural network; where the at least one controller is further configured to cause the processor to update, in response to successfully decoding the at least one encoded message, the one or more layers of the decoder neural network based at least in part on the at least one encoded message; where the at least one controller is further configured to cause the processor to update, based at least in part on successfully decoding the at least one encoded message, the one or more layers of the decoder neural network based at least in part on the at least one encoded message; where the processor is included in a UE.

[0178] The processor 500 may support wireless communication in accordance with examples as disclosed herein. The processor 500 may be configured to or operable to support at least one controller (e.g., the controller 502) coupled with at least one memory (e.g., the memory 504) and configured to cause the processor to: generate, using an encoder neural network, at least one encoded message; transmit the at least one encoded message; and transmit an indication for an apparatus to update one or more layers of a decoder neural network used by the apparatus to decode the at least one encoded message.

[0179] Additionally, the processor 500 may be configured to or operable to support any one or combination of where the at least one encoded message includes multiple encoded messages transmitted sequentially in time domain, frequency domain, or spatial domain; where each encoded message of the at least one encoded message comprises one of multiple different information messages; where the at least one controller is further configured to cause the processor to transmit one or more tuning parameters used by the apparatus to update one or more layers of the decoder neural network; where the encoder neural network is trained based on a first set of training data and the decoder neural network is trained based on a second set of training data, and where the first set of training data is different than the second set of training data; where the encoder neural network and the decoder neural network are trained on a same set of training data; where the at least one controller is further configured to cause the processor to receive an assistance information request message for updating one or more parameters of the decoder neural network; where the at least one controller is further configured to cause the processor to transmit, to the apparatus in response to the assistance information request message, at least one of a signal or an additional encoded message; where the at least one controller is further configured to cause the processor to transmit, based at least in part on the assistance information request message, at least one of a signal or an additional Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 48 encoded message; where the at least one controller is further configured to cause the processor to transmit, to the apparatus, an indication for the apparatus to determine whether to update the one or more layers of the decoder neural network; where the at least one controller is further configured to cause the processor to transmit an indication for the apparatus to determine whether to update the one or more layers of the decoder neural network; where the apparatus comprises a UE.

[0180] Figure 6 illustrates an example of a NE 600 in accordance with aspects of the present disclosure. The NE 600 may include a processor 602, a memory 604, a controller 606, and a transceiver 608. The processor 602, the memory 604, the controller 606, or the transceiver 608, or various combinations thereof or various components thereof may be examples of means for performing various aspects of the present disclosure as described herein. These components may be coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more interfaces.

[0181] The processor 602, the memory 604, the controller 606, or the transceiver 608, or various combinations or components thereof may be implemented in hardware (e.g., circuitry). The hardware may include a processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or other programmable logic device, or any combination thereof configured as or otherwise supporting a means for performing the functions described in the present disclosure.

[0182] The processor 602 may include an intelligent hardware device (e.g., a general-purpose processor, a DSP, a CPU, an ASIC, an FPGA, or any combination thereof). In some implementations, the processor 602 may be configured to operate the memory 604. In some other implementations, the memory 604 may be integrated into the processor 602. The processor 602 may be configured to execute computer-readable instructions stored in the memory 604 to cause the NE 600 to perform various functions of the present disclosure.

[0183] The memory 604 may include volatile or non-volatile memory. The memory 604 may store computer-readable, computer-executable code including instructions when executed by the processor 602 cause the NE 600 to perform various functions described herein. The code may be stored in a non-transitory computer-readable medium such as the memory 604 or another type of memory. Computer-readable media includes both non-transitory computer storage media and Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 49 communication media including any medium that facilitates transfer of a computer program from one place to another. A non-transitory storage medium may be any available medium that may be accessed by a general-purpose or special-purpose computer.

[0184] In some implementations, the processor 602 and the memory 604 coupled with the processor 602 may be configured to cause the NE 600 to perform one or more of the functions described herein (e.g., executing, by the processor 602, instructions stored in the memory 604). For example, the processor 602 may support wireless communication at the NE 600 in accordance with examples as disclosed herein. The NE 600 may be configured to support a means for at least one encoded message; generating one or more modified versions of the at least one encoded message; and updating one or more layers of a decoder neural network based at least in part on the at least one encoded message, the one or more modified versions of the at least one encoded message, one or more tuning parameters, and a loss function.

[0185] Additionally, the NE 600 may be configured to support any one or combination of updating the one or more layers of the decoder neural network by minimizing the loss function; where minimizing the loss function comprises minimizing maximum entropy in a region through min-max optimization; where the at least one encoded message includes multiple encoded messages and the one or more modified versions of the at least one encoded message includes one or more modified versions of each of the multiple encoded messages, and further including updating the one or more layers of the decoder neural network based at least in part on the multiple encoded messages and the one or more modified versions of each of the multiple encoded messages; where the multiple encoded messages are received sequentially in time domain, frequency domain, or spatial domain; determining to update the one or more layers of the decoder neural network based at least in part on periodicity information indicating the one or more layers of the decoder neural network are to be updated at regular intervals of time; determining to update the one or more layers of the decoder neural network based at least in part on a change in one or more characteristics of one or more received encoded messages; where the one or more characteristics are statistical characteristics of the one or more received encoded messages; where the statistical characteristics include at least one of a mean, a median, a standard deviation, or a probability distribution; determining to update the one or more layers of the decoder neural network based at least in part on an indication, received from an apparatus, to update the one or more layers of the decoder neural Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 50 network; determining to update the one or more layers of the decoder neural network based at least in part on a change in a value of at least one parameter corresponding to a channel, where the channel corresponds to a communication medium between the UE and an apparatus monitoring a quality of the at least one encoded message, where the quality of the at least one encoded message comprises one or more of an indication of decoding success of the at least one encoded message, an indication of cyclical redundancy check status of the at least one encoded message, an acknowledgement indication for the at least one encoded message, a retransmission indication for the at least one encoded message, or a decoding error probability for the at least one encoded message above a threshold; and determining to update the one or more layers of the decoder neural network based at least in part on the monitoring; where the receiving comprises receiving the at least one encoded message from a first apparatus, and determining to update the one or more layers of the decoder neural network based at least in part on an indication or a configuration, received from a second apparatus, to update the one or more layers of the decoder neural network; where the first apparatus comprises a base station, and the second apparatus comprises the base station or another NE; where each encoded message of the at least one encoded message comprises one of multiple different information messages; determining the one or more layers of the decoder neural network; where the receiving comprises receiving the at least one encoded message from a first apparatus, and receiving, from the first apparatus or a second apparatus, an indication of the one or more layers of the decoder neural network; where the first apparatus comprises a base station, and the second apparatus comprises another NE; where the one or more tuning parameters comprises one or both of a first tuning parameter that is a number of the one or more modified versions for each of the at least one encoded message, and a second tuning parameter that is a setting of the loss function; determining the one or more tuning parameters; or receiving, from a first apparatus or a second apparatus, the one or more tuning parameters; where the first apparatus comprises a base station, and the second apparatus comprises another NE; the updating including updating values of a number N of parameters of the one or more layers by: generating a number Q of different augmented received encoded messages where each augmented received encoded message is a modified version of at least one encoded message, and Q is a first tuning parameter of the one or more tuning parameters; obtaining a set of Q+1 different data samples, where a first data sample is the at least one encoded message and remaining Q data samples are the Q number of different augmented received encoded messages generated from the at least one encoded message; for each Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 51 of the Q+1 different data samples, using the decoder neural network to compute a vector of length M where an ^^^element of the vector is a probability that an information message contained in thedata sample is ^^, where ^^ ∈ {1, … , ^} and the at least one encoded message comprises one of Mmultiple different information messages; adding all the Q+1 number of vectors, each of length M,using element- to generate a temporary vector; dividing each element of the temporary vector by Q+1 to generate a mean probability vector; computing an entropy of the mean probability vector, where the entropy is a function of parameters of the decoder neural network; and updating the one or more layers of the decoder neural network by determining new values of the N parameters that minimizes a maximum entropy within an N-dimensional Euclidean ball, with the N- dimensional Euclidean ball centered at the new values of the N parameters with radius R, where R is a second tuning parameter of the one or more tuning parameters; where the at least one encoded message is based at least in part on an encoder neural network; where the encoder neural network is trained based on a first set of training data and the decoder neural network is trained based on a second set of training data, and where the first set of training data is different than the second set of training data; where the encoder neural network and the decoder neural network are trained on a same set of training data; transmitting, to a first apparatus, an assistance information request message for updating one or more parameters of the decoder neural network; receiving, from the first apparatus in response to the assistance information request message, at least one of a signal or an additional encoded message; receiving, from the first apparatus based at least in part on the assistance information request message, at least one of a signal or an additional encoded message; receiving, from the first apparatus, an indication to determine whether to update the one or more layers of the decoder neural network; updating, in response to successfully decoding the at least one encoded message, the one or more layers of the decoder neural network based at least in part on the at least one encoded message; updating, based at least in part on successfully decoding the at least one encoded message, the one or more layers of the decoder neural network based at least in part on the at least one encoded message.

[0186] In some implementations, the processor 602 and the memory 604 coupled with the processor 602 may be configured to cause the NE 600 to perform one or more of the functions described herein (e.g., executing, by the processor 602, instructions stored in the memory 604). For example, the processor 602 may support wireless communication at the NE 600 in accordance with Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 52 examples as disclosed herein. The NE 600 may be configured to support a means for generating, using an encoder neural network, at least one encoded message; transmitting the at least one encoded message; and transmitting an indication for an apparatus to update one or more layers of a decoder neural network used by the apparatus to decode the at least one encoded message.

[0187] Additionally, the NE 600 may be configured to support any one or combination of where the at least one encoded message includes multiple encoded messages transmitted sequentially in time domain, frequency domain, or spatial domain; where each encoded message of the at least one encoded message comprises one of multiple different information messages; transmitting one or more tuning parameters used by the apparatus to update one or more layers of the decoder neural network; where the encoder neural network is trained based on a first set of training data and the decoder neural network is trained based on a second set of training data, and where the first set of training data is different than the second set of training data; where the encoder neural network and the decoder neural network are trained on a same set of training data; receiving an assistance information request message for updating one or more parameters of the decoder neural network; transmitting, in response to the assistance information request message, at least one of a signal or an additional encoded message; transmitting, based at least in part on the assistance information request message, at least one of a signal or an additional encoded message; transmitting an indication for the apparatus to determine whether to update the one or more layers of the decoder neural network; transmitting an indication for the apparatus to determine whether to update the one or more layers of the decoder neural network; where the apparatus comprises a UE.

[0188] Additionally, or alternatively, the NE 600 may support at least one memory (e.g., the memory 604) and at least one processor (e.g., the processor 602) coupled with the at least one memory and configured to cause the NE to: receive at least one encoded message; generate one or more modified versions of the at least one encoded message; and update one or more layers of a decoder neural network based at least in part on the at least one encoded message, the one or more modified versions of the at least one encoded message, one or more tuning parameters, and a loss function.

[0189] Additionally, the NE 600 may be configured to support any one or combination of where the at least one processor is further configured to cause the first apparatus to update the one or more layers of the decoder neural network by minimizing the loss function; where minimizing the loss Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 53 function comprises minimizing maximum entropy in a region through min-max optimization; where the at least one encoded message includes multiple encoded messages and the one or more modified versions of the at least one encoded message includes one or more modified versions of each of the multiple encoded messages, and where the at least one processor is further configured to cause the first apparatus to update the one or more layers of the decoder neural network based at least in part on the multiple encoded messages and the one or more modified versions of each of the multiple encoded messages; where the multiple encoded messages are received sequentially in time domain, frequency domain, or spatial domain; to cause the first apparatus to determine to update the one or more layers of the decoder neural network based at least in part on periodicity information indicating the one or more layers of the decoder neural network are to be updated at regular intervals of time; where the at least one processor is further configured to cause the first apparatus to determine to update the one or more layers of the decoder neural network based at least in part on a change in one or more characteristics of one or more received encoded messages; where the one or more characteristics are statistical characteristics of the one or more received encoded messages; where the statistical characteristics include at least one of a mean, a median, a standard deviation, or a probability distribution; where the at least one processor is further configured to cause the first apparatus to determine to update the one or more layers of the decoder neural network based at least in part on an indication, received from a second apparatus, to update the one or more layers of the decoder neural network; where the at least one processor is further configured to cause the first apparatus to determine to update the one or more layers of the decoder neural network based at least in part on a change in a value of at least one parameter corresponding to a channel, where the channel corresponds to a communication medium between the first apparatus and a second apparatus; where the at least one processor is further configured to cause the first apparatus to: monitor a quality of the at least one encoded message, where the quality of the at least one encoded message comprises one or more of an indication of decoding success of the at least one encoded message, an indication of cyclical redundancy check status of the at least one encoded message, an acknowledgement indication for the at least one encoded message, a retransmission indication for the at least one encoded message, or a decoding error probability for the at least one encoded message above a threshold; and determine to update the one or more layers of the decoder neural network based at least in part on the monitoring; where the at least one processor is further configured to cause the first apparatus to receive the at least one encoded message from a second Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 54 apparatus, and to determine to update the one or more layers of the decoder neural network based at least in part on an indication or a configuration, received from a third apparatus, to update the one or more layers of the decoder neural network; where the first apparatus comprises a UE, the second apparatus comprises a base station, and the third apparatus comprises the base station or another NE; where each encoded message of the at least one encoded message comprises one of multiple different information messages; where the at least one processor is further configured to cause the first apparatus to determine the one or more layers of the decoder neural network; where the at least one processor is further configured to cause the first apparatus to receive the at least one encoded message from a second apparatus, and to receive, from the second apparatus or a third apparatus, an indication of the one or more layers of the decoder neural network; where the first apparatus comprises a UE, the second apparatus comprises a base station, and the third apparatus comprises another NE; where the one or more tuning parameters comprises one or both of a first tuning parameter that is a number of the one or more modified versions for each of the at least one encoded message, and a second tuning parameter that is a setting of the loss function; where the at least one processor is further configured to cause the first apparatus to: determine the one or more tuning parameters; or receive, from a second apparatus or a third apparatus, the one or more tuning parameters; where the first apparatus comprises a UE, the second apparatus comprises a base station, and the third apparatus comprises another NE; where the at least one processor is further configured, to update values of a number N of parameters of the one or more layers of the decoder neural network based at least in part on the at least one encoded message, to cause the first apparatus to: generate a number Q of different augmented received encoded messages where each augmented received encoded message is a modified version of at least one encoded message, and Q is a first tuning parameter of the one or more tuning parameters; obtain a set of Q+1 different data samples, where a first data sample is the at least one encoded message and remaining Q data samples are the Q number of different augmented received encoded messages generated from the at least one encoded message; for each of the Q+1 different data samples, use the decoder neural network to compute a vector of length M where an ^^^element of the vector is a probability that aninformation message contained in the data sample is ^^, where ^^ ∈ {1, … , ^} and the at least oneencoded message comprises one of M multiple differentmessages; add all the Q+1 number of vectors, each of length M, using element-wise addition to generate a temporary vector; divide each element of the temporary vector by Q+1 to generate a mean probability vector; compute Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 55 an entropy of the mean probability vector, where the entropy is a function of parameters of the decoder neural network; and update the one or more layers of the decoder neural network by determining new values of the N parameters that minimizes a maximum entropy within an N- dimensional Euclidean ball, with the N-dimensional Euclidean ball centered at the new values of the N parameters with radius R, where R is a second tuning parameter of the one or more tuning parameters; where the at least one encoded message is based at least in part on an encoder neural network; where the encoder neural network is trained based on a first set of training data and the decoder neural network is trained based on a second set of training data, and where the first set of training data is different than the second set of training data; where the encoder neural network and the decoder neural network are trained on a same set of training data; where the at least one processor is further configured to cause the first apparatus to transmit, to a second apparatus, an assistance information request message for updating one or more parameters of the decoder neural network; where the at least one processor is further configured to cause the first apparatus to receive, from the second apparatus in response to the assistance information request message, at least one of a signal or an additional encoded message; where the at least one processor is further configured to cause the first apparatus to receive, from the second apparatus based at least in part on the assistance information request message, at least one of a signal or an additional encoded message; where the at least one processor is further configured to cause the first apparatus to receive, from a second apparatus, an indication to determine whether to update the one or more layers of the decoder neural network; where the at least one processor is further configured to cause the first apparatus to update, in response to successfully decoding the at least one encoded message, the one or more layers of the decoder neural network based at least in part on the at least one encoded message; where the at least one processor is further configured to cause the first apparatus to update, based at least in part on successfully decoding the at least one encoded message, the one or more layers of the decoder neural network based at least in part on the at least one encoded message; where the first apparatus comprises a UE.

[0190] Additionally, or alternatively, the NE 600 may support at least one memory (e.g., the memory 604) and at least one processor (e.g., the processor 602) coupled with the at least one memory and configured to cause the NE to: generate, using an encoder neural network, at least one encoded message; transmit the at least one encoded message; and transmit an indication for a Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 56 second apparatus to update one or more layers of a decoder neural network used by the second apparatus to decode the at least one encoded message.

[0191] Additionally, the NE 600 may be configured to support any one or combination of where the at least one encoded message includes multiple encoded messages transmitted sequentially in time domain, frequency domain, or spatial domain; where each encoded message of the at least one encoded message comprises one of multiple different information messages; where the at least one processor is further configured to cause the first apparatus to transmit one or more tuning parameters used by the second apparatus to update one or more layers of the decoder neural network; where the encoder neural network is trained based on a first set of training data and the decoder neural network is trained based on a second set of training data, and where the first set of training data is different than the second set of training data; where the encoder neural network and the decoder neural network are trained on a same set of training data; where the at least one processor is further configured to cause the first apparatus to receive an assistance information request message for updating one or more parameters of the decoder neural network; where the at least one processor is further configured to cause the first apparatus to transmit, in response to the assistance information request message, at least one of a signal or an additional encoded message; where the at least one processor is further configured to cause the first apparatus to transmit, based at least in part on the assistance information request message, at least one of a signal or an additional encoded message; where the at least one processor is further configured to cause the first apparatus to transmit an indication for the second apparatus to determine whether to update the one or more layers of the decoder neural network; where the first apparatus comprises a base station and the second apparatus comprises a UE; where the at least one processor is further configured to cause the first apparatus to transmit an indication for the second apparatus to determine whether to update the one or more layers of the decoder neural network.

[0192] The controller 606 may manage input and output signals for the NE 600. The controller 606 may also manage peripherals not integrated into the NE 600. In some implementations, the controller 606 may utilize an operating system such as iOS®, ANDROID®, WINDOWS®, or other operating systems. In some implementations, the controller 606 may be implemented as part of the processor 602. Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 57

[0193] In some implementations, the NE 600 may include at least one transceiver 608. In some other implementations, the NE 600 may have more than one transceiver 608. The transceiver 608 may represent a wireless transceiver. The transceiver 608 may include one or more receiver chains 610, one or more transmitter chains 612, or a combination thereof.

[0194] A receiver chain 610 may be configured to receive signals (e.g., control information, data, packets) over a wireless medium. For example, the receiver chain 610 may include one or more antennas to receive a signal over the air or wireless medium. The receiver chain 610 may include at least one amplifier (e.g., a low-noise amplifier (LNA)) configured to amplify the received signal. The receiver chain 610 may include at least one demodulator configured to demodulate the receive signal and obtain the transmitted data by reversing the modulation technique applied during transmission of the signal. The receiver chain 610 may include at least one decoder for decoding the demodulated signal to receive the transmitted data.

[0195] A transmitter chain 612 may be configured to generate and transmit signals (e.g., control information, data, packets). The transmitter chain 612 may include at least one modulator for modulating data onto a carrier signal, preparing the signal for transmission over a wireless medium. The at least one modulator may be configured to support one or more techniques such as amplitude modulation (AM), frequency modulation (FM), or digital modulation schemes like phase-shift keying (PSK) or quadrature amplitude modulation (QAM). The transmitter chain 612 may also include at least one power amplifier configured to amplify the modulated signal to an appropriate power level suitable for transmission over the wireless medium. The transmitter chain 612 may also include one or more antennas for transmitting the amplified signal into the air or wireless medium.

[0196] Figure 7 illustrates a flowchart of a method in accordance with aspects of the present disclosure. The operations of the method may be implemented by a UE as described herein. In some implementations, the UE may execute a set of instructions to control the function elements of the UE to perform the described functions.

[0197] At 702, the method may include receiving at least one encoded message. The operations of 702 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 702 may be performed by a UE as described with reference to Figure 4. Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 58

[0198] At 704, the method may include generating one or more modified versions of the at least one encoded message. The operations of 704 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 704 may be performed by a UE as described with reference to Figure 4.

[0199] At 706, the method may include updating one or more layers of a decoder neural network based at least in part on the at least one encoded message, the one or more modified versions of the at least one encoded message, one or more tuning parameters, and a loss function. The operations of 706 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 706 may be performed a UE as described with reference to Figure 4.

[0200] It should be noted that the method described herein describes a possible implementation, and that the operations and the steps may be rearranged or otherwise modified and that other implementations are possible.

[0201] Figure 8 illustrates a flowchart of a method in accordance with aspects of the present disclosure. The operations of the method may be implemented by a UE as described herein. In some implementations, the UE may execute a set of instructions to control the function elements of the UE to perform the described functions.

[0202] At 802, the method may include generating, using an encoder neural network, at least one encoded message. The operations of 802 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 802 may be performed by a UE as described with reference to Figure 4.

[0203] At 804, the method may include transmitting the at least one encoded message. The operations of 804 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 804 may be performed by a UE as described with reference to Figure 4.

[0204] At 806, the method may include transmitting an indication for an apparatus to update one or more layers of a decoder neural network used by the apparatus to decode the at least one encoded message. The operations of 806 may be performed in accordance with examples as Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 59 described herein. In some implementations, aspects of the operations of 806 may be performed a UE as described with reference to Figure 4.

[0205] It should be noted that the method described herein describes a possible implementation, and that the operations and the steps may be rearranged or otherwise modified and that other implementations are possible.

[0206] Figure 9 illustrates a flowchart of a method in accordance with aspects of the present disclosure. The operations of the method may be implemented by a NE as described herein. In some implementations, the NE may execute a set of instructions to control the function elements of the NE to perform the described functions.

[0207] At 902, the method may include receiving at least one encoded message. The operations of 902 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 902 may be performed by a NE as described with reference to Figure 6.

[0208] At 904, the method may include generating one or more modified versions of the at least one encoded message. The operations of 904 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 904 may be performed by a NE as described with reference to Figure 6.

[0209] At 906, the method may include updating one or more layers of a decoder neural network based at least in part on the at least one encoded message, the one or more modified versions of the at least one encoded message, one or more tuning parameters, and a loss function. The operations of 906 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 906 may be performed a NE as described with reference to Figure 6.

[0210] It should be noted that the method described herein describes a possible implementation, and that the operations and the steps may be rearranged or otherwise modified and that other implementations are possible.

[0211] Figure 10 illustrates a flowchart of a method in accordance with aspects of the present disclosure. The operations of the method may be implemented by a NE as described herein. In some Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 60 implementations, the NE may execute a set of instructions to control the function elements of the NE to perform the described functions.

[0212] At 1002, the method may include generating, using an encoder neural network, at least one encoded message. The operations of 1002 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 1002 may be performed by a NE as described with reference to Figure 6.

[0213] At 1004, the method may include transmitting the at least one encoded message. The operations of 1004 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 1004 may be performed by a NE as described with reference to Figure 6.

[0214] At 1006, the method may include transmitting an indication for an apparatus to update one or more layers of a decoder neural network used by the apparatus to decode the at least one encoded message. The operations of 1006 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 1006 may be performed a NE as described with reference to Figure 6.

[0215] It should be noted that the method described herein describes a possible implementation, and that the operations and the steps may be rearranged or otherwise modified and that other implementations are possible.

[0216] The description herein is provided to enable a person having ordinary skill in the art to make or use the disclosure. Various modifications to the disclosure will be apparent to a person having ordinary skill in the art, and the generic principles defined herein may be applied to other variations without departing from the scope of the disclosure. Thus, the disclosure is not limited to the examples and designs described herein but is to be accorded the broadest scope consistent with the principles and novel features disclosed herein. Attorney Ref. No. SMM920230253-WO-PCT

Claims

Lenovo Ref. No. SMM920230253-WO-PCT 61 CLAIMS What is claimed is:

1. A first apparatus for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the first apparatus to: receive at least one encoded message; generate one or more modified versions of the at least one encoded message; and update one or more layers of a decoder neural network based at least in part on the at least one encoded message, the one or more modified versions of the at least one encoded message, one or more tuning parameters, and a loss function.

2. The first apparatus of claim 1, wherein the at least one processor is further configured to cause the first apparatus to update the one or more layers of the decoder neural network by minimizing the loss function.

3. The first apparatus of claim 1, wherein the at least one encoded message includes multiple encoded messages and the one or more modified versions of the at least one encoded message includes one or more modified versions of each of the multiple encoded messages, and wherein the at least one processor is further configured to cause the first apparatus to update the one or more layers of the decoder neural network based at least in part on the multiple encoded messages and the one or more modified versions of each of the multiple encoded messages, and wherein the multiple encoded messages are received sequentially in time domain, frequency domain, or spatial domain.

4. The first apparatus of claim 1, wherein the at least one processor is further configured to cause the first apparatus to determine to update the one or more layers of the decoder neural network based at least in part on periodicity information indicating the one or more layers of the decoder neural network are to be updated at regular intervals of time.

5. The first apparatus of claim 1, wherein the at least one processor is further configured to cause the first apparatus to determine to update the one or more layers of the decoder neural Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 62 network based at least in part on a change in one or more characteristics of one or more received encoded messages.

6. The first apparatus of claim 5, wherein the one or more characteristics are statistical characteristics of the one or more received encoded messages, and wherein the statistical characteristics include at least one of a mean, a median, a standard deviation, or a probability distribution.

7. The first apparatus of claim 1, wherein the at least one processor is further configured to cause the first apparatus to determine to update the one or more layers of the decoder neural network based at least in part on an indication, received from a second apparatus, to update the one or more layers of the decoder neural network.

8. The first apparatus of claim 1, wherein the at least one processor is further configured to cause the first apparatus to determine to update the one or more layers of the decoder neural network based at least in part on a change in a value of at least one parameter corresponding to a channel, wherein the channel corresponds to a communication medium between the first apparatus and a second apparatus.

9. The first apparatus of claim 5, wherein the at least one processor is further configured to cause the first apparatus to: monitor a quality of the at least one encoded message, wherein the quality of the at least one encoded message comprises one or more of an indication of decoding success of the at least one encoded message, an indication of cyclical redundancy check status of the at least one encoded message, an acknowledgement indication for the at least one encoded message, a retransmission indication for the at least one encoded message, or a decoding error probability for the at least one encoded message above a threshold; and determine to update the one or more layers of the decoder neural network based at least in part on the monitoring.

10. The first apparatus of claim 1, wherein the at least one processor is further configured to cause the first apparatus to receive the at least one encoded message from a second apparatus, and to determine to update the one or more layers of the decoder neural network based at least in part on Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 63 an indication or a configuration, received from a third apparatus, to update the one or more layers of the decoder neural network.

11. The first apparatus of claim 1, wherein the at least one processor is further configured to cause the first apparatus to determine the one or more layers of the decoder neural network.

12. The first apparatus of claim 1, wherein the at least one processor is further configured to cause the first apparatus to receive the at least one encoded message from a second apparatus, and to receive, from the second apparatus or a third apparatus, an indication of the one or more layers of the decoder neural network.

13. The first apparatus of claim 1, wherein the one or more tuning parameters comprises one or both of a first tuning parameter that is a number of the one or more modified versions for each of the at least one encoded message, and a second tuning parameter that is a setting of the loss function.

14. A first apparatus for wireless communication, comprising: at least one memory; and at least one processor coupled with the at least one memory and configured to cause the first apparatus to: generate, using an encoder neural network, at least one encoded message; transmit the at least one encoded message; and transmit an indication for a second apparatus to update one or more layers of a decoder neural network used by the second apparatus to decode the at least one encoded message.

15. The first apparatus of claim 14, wherein the at least one encoded message includes multiple encoded messages transmitted sequentially in time domain, frequency domain, or spatial domain.

16. The first apparatus of claim 14, wherein the at least one processor is further configured to cause the first apparatus to transmit one or more tuning parameters used by the second apparatus to update one or more layers of the decoder neural network. Attorney Ref. No. SMM920230253-WO-PCTLenovo Ref. No. SMM920230253-WO-PCT 64 17. The first apparatus of claim 14, wherein the encoder neural network is trained based on a first set of training data and the decoder neural network is trained based on a second set of training data, and wherein the first set of training data is different than the second set of training data.

18. The first apparatus of claim 14, wherein the at least one processor is further configured to cause the first apparatus to receive an assistance information request message for updating one or more parameters of the decoder neural network.

19. A method performed by a user equipment (UE), the method comprising: receiving at least one encoded message; generating one or more modified versions of the at least one encoded message; and updating one or more layers of a decoder neural network based at least in part on the at least one encoded message, the one or more modified versions of the at least one encoded message, one or more tuning parameters, and a loss function.

20. A method performed by a base station, the method comprising: generating, using an encoder neural network, at least one encoded message; transmitting the at least one encoded message; and transmitting an indication for an apparatus to update one or more layers of a decoder neural network used by the apparatus to decode the at least one encoded message. Attorney Ref. No. SMM920230253-WO-PCT

Images