Methods of joint source and channel coding for effective and semantic communication

Abstract

Systems, methods, apparatuses, and computer program products for semantic communication involving joint source and channel coding (JSCC) over current communication networks. One method may include receiving, by a first network entity, from a second network entity, a multi-level bit error rate (BER) interface configuration comprising at least one BER based upon the request for a communication session; generating, by the first network entity, an application source encoder and decoder configured to encode and decode, respectively, a plurality of levels corresponding to the at least one BER; and training, by the first network entity, the application source encoder and decoder according to the BER.

Description

TITLE METHODS OF JOINT SOURCE AND CHANNEL CODING FOR EFFECTIVE AND SEMANTIC COMMUNICATION TECHNICAL FIELD

[0001] Some example embodiments may generally relate to mobile or wireless telecommunication systems, such as 3rdGeneration Partnership Project (3GPP) Long Term Evolution (LTE), 5thgeneration (5G) radio access technology (RAT), new radio (NR) access technology, 6thgeneration (6G), and / or other communications systems. For example, certain example embodiments may relate to systems and / or methods forsemantic communication involving joint source and channel coding (JSCC) overcurrent communication networks. BACKGROUND

[0002] Examples of mobile or wireless telecommunication systems may include radio frequency (RF) 5G RAT, the Universal Mobile Telecommunications System (UMTS) Terrestrial Radio Access Network (UTRAN), LTE Evolved UTRAN (E-UTRAN), LTE-Advanced (LTE-A), LTE-A Pro, NR access technology, and / or MulteFire Alliance. 5G wireless systems refer to the next generation (NG) of radio systems and network architecture. A 5G system is typically built on a 5G NR, but a 5G (or NG) network may also be built on E-UTRA radio. It is expected that NR can support service categories such as enhanced mobile broadband (eMBB), ultra-reliable low-latency- communication (URLLC), and massive machine-type communication (mMTC). NR is expected to deliver extreme broadband, ultra-robust, low-latency connectivity, and massive networking to support the Internet of Things (IoT). The next generation radioaccess network (NG-RAN) represents the radio access network (RAN) for 5G, whichmay provide radio access for NR, LTE, and LTE-A. It is noted that the nodes in 5G providing radio access functionality to a user equipment (e.g., similar to the Node B in UTRAN or the Evolved Node B (eNB) in LTE) may be referred to as next-generationNode B (gNB) when built on NR radio, and may be referred to as next-generation eNB (NG-eNB) when built on E-UTRA radio. SUMMARY

[0003] In accordance with some example embodiments, a method may includereceiving, by a first network entity, from a second network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the requestfor a communication session. The method may further include generating, by the firstnetwork entity, an application source encoder and decoder configured to encode anddecode, respectively, a plurality of levels corresponding to the at least one bit error rate.The method may further include training, by the first network entity, the applicationsource encoder and decoder according to the bit error rate.

[0004] In accordance with certain example embodiments, an apparatus may includemeans for receiving, from a network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for a communication session. The apparatus may further include means for generating an application source encoder and decoder configured to encode and decode, respectively,a plurality of levels corresponding to the at least one bit error rate. The apparatus mayfurther include means for training the application source encoder and decoder accordingto the bit error rate.

[0005] In accordance with various example embodiments, a non-transitory computerreadable medium may include program instructions that, when executed by an apparatus, cause the apparatus to perform at least a method. The method may include receiving, from a network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for a communication session. The method may further include generating an application source encoder and decoder configured to encode and decode, respectively, a plurality of levels corresponding to the at least one bit error rate. The method may further include trainingthe application source encoder and decoder according to the bit error rate.

[0006] In accordance with some example embodiments, a computer program productmay perform a method. The method may include receiving, from a network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for a communication session. The method may further include generating an application source encoder and decoder configured to encode and decode, respectively, a plurality of levels corresponding to the at least one bit error rate. Themethod may further include training the application source encoder and decoderaccording to the bit error rate.

[0007] In accordance with certain example embodiments, an apparatus may include atleast one processor and at least one memory storing instructions that, when executed bythe at least one processor, cause the apparatus at least to receive, from a network entity,a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for a communication session. The at least one memory andinstructions, when executed by the at least one processor, may further cause theapparatus at least to generate an application source encoder and decoder configured toencode and decode, respectively, a plurality of levels corresponding to the at least one bit error rate. The at least one memory and instructions, when executed by the at leastone processor, may further cause the apparatus at least to train the application sourceencoder and decoder according to the bit error rate.

[0008] In accordance with various example embodiments, an apparatus may includereceiving circuitry configured to perform receiving, from a network entity, a multi-levelbit error rate interface configuration comprising at least one bit error rate based upon the request for a communication session. The apparatus may further include generatingcircuitry configured to perform generating an application source encoder and decoderconfigured to encode and decode, respectively, a plurality of levels corresponding to the at least one bit error rate. The apparatus may further include training circuitry configuredto perform training the application source encoder and decoder according to the bit errorrate.

[0009] In accordance with some example embodiments, a method may includetransmitting, by a first network entity, to a second network entity, a request for asemantics communication session. The method may further include receiving, by the first network entity, from the second network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for the semantics communication session. The method may further include calculating, by thefirst network entity, a semantic loss of at least one data packet according to the receivedmulti-level bit error rate. The method may further include transmitting, by the firstnetwork entity, to the second network entity, the calculated semantic loss.

[0010] In accordance with certain example embodiments, an apparatus may includemeans for transmitting, to a network entity, a request for a semantics communicationsession. The apparatus may further include means for receiving, from the network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for the semantics communication session. The apparatusmay further include means for calculating a semantic loss of at least one data packetaccording to the received multi-level bit error rate. The apparatus may further include means for transmitting, to the network entity, the calculated semantic loss.

[0011] In accordance with various example embodiments, a non-transitory computerreadable medium may include program instructions that, when executed by an apparatus, cause the apparatus to perform at least a method. The method may include transmitting, to a network entity, a request for a semantics communication session. The method may further include receiving, from the network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for the semantics communication session. The method may further include calculating a semantic loss of at least one data packet according to the received multi-level bit errorrate. The method may further include transmitting, to the network entity, the calculatedsemantic loss.

[0012] In accordance with some example embodiments, a computer program productmay perform a method. The method may include transmitting, to a network entity, arequest for a semantics communication session. The method may further include receiving, from the network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for the semantics communication session. The method may further include calculating a semantic loss ofat least one data packet according to the received multi-level bit error rate. The methodmay further include transmitting, to the network entity, the calculated semantic loss.

[0013] In accordance with certain example embodiments, an apparatus may include atleast one processor and at least one memory storing instructions that, when executed bythe at least one processor, cause the apparatus at least to transmit a request for asemantics communication session. The at least one memory and instructions, whenexecuted by the at least one processor, may further cause the apparatus at least to receivea multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for the semantics communication session. The at least onememory and instructions, when executed by the at least one processor, may furthercause the apparatus at least to calculate a semantic loss of at least one data packetaccording to the received multi-level bit error rate. The at least one memory andinstructions, when executed by the at least one processor, may further cause theapparatus at least to transmit the calculated semantic loss.

[0014] In accordance with various example embodiments, an apparatus may includetransmitting circuitry configured to perform transmitting, to a network entity, a requestfor a semantics communication session. The apparatus may further include receivingcircuitry configured to perform receiving, from the network entity, a multi-level bit errorrate interface configuration comprising at least one bit error rate based upon the request for the semantics communication session. The apparatus may further include calculatingcircuitry configured to perform calculating a semantic loss of at least one data packetaccording to the received multi-level bit error rate. The apparatus may further includetransmitting circuitry configured to perform transmitting, to the network entity, thecalculated semantic loss.

[0015] In accordance with some example embodiments, a method may includeestablishing, by a user equipment, at least one link configured for a plurality of qualityof service levels of the link. The method may further include multiplexing, by the userequipment, a plurality of segments of a data packet into a multi-level packet with a corresponding header.

[0016] In accordance with certain example embodiments, an apparatus may includemeans for establishing at least one link configured for a plurality of quality of servicelevels of the link. The apparatus may further include means for multiplexing a plurality of segments of a data packet into a multi-level packet with a corresponding header.

[0017] In accordance with various example embodiments, a non-transitory computerreadable medium may include program instructions that, when executed by an apparatus, cause the apparatus to perform at least a method. The method may include establishing at least one link configured for a plurality of quality of service levels of the link. The method may further include multiplexing a plurality of segments of a data packet into a multi-level packet with a corresponding header.

[0018] In accordance with some example embodiments, a computer program productmay perform a method. The method may include establishing at least one link configured for a plurality of quality of service levels of the link. The method may further include multiplexing a plurality of segments of a data packet into a multi-level packet with a corresponding header.

[0019] In accordance with certain example embodiments, an apparatus may include atleast one processor and at least one memory storing instructions that, when executed bythe at least one processor, cause the apparatus at least to establish at least one linkconfigured for a plurality of quality of service levels of the link. The at least one memoryand instructions, when executed by the at least one processor, may further cause theapparatus at least to multiplex a plurality of segments of a data packet into a multi-levelpacket with a corresponding header.

[0020] In accordance with various example embodiments, an apparatus may includeestablishing circuitry configured to perform establishing at least one link configured fora plurality of quality of service levels of the link. The apparatus may further includemultiplexing circuitry configured to perform multiplexing a plurality of segments of adata packet into a multi-level packet with a corresponding header.

[0021] In accordance with some example embodiments, a method may includeestablishing, by a network entity, at least one link with an application configured for a plurality of quality of service levels of the link. The method may further include processing, by the network entity, each of the plurality of segments of the data packet based upon a corresponding quality of service in a separate subchannel. The method may further include transmitting, by the network entity, the multi-level packet to an application.

[0022] In accordance with certain example embodiments, an apparatus may includemeans for establishing at least one link with an application configured for a plurality ofquality of service levels of the link. The apparatus may further include means for processing each of the plurality of segments of the data packet based upon acorresponding quality. The apparatus may further include means for transmitting themulti-level packet to an application.

[0023] In accordance with various example embodiments, a non-transitory computerreadable medium may include program instructions that, when executed by an apparatus, cause the apparatus to perform at least a method. The method may includeestablishing at least one link with an application configured for a plurality of quality ofservice levels of the link. The method may further include processing each of the plurality of segments of the data packet based upon a corresponding quality. The method may further include transmitting the multi-level packet to an application.

[0024] In accordance with some example embodiments, a computer program productmay perform a method. The method may include establishing at least one link with anapplication configured for a plurality of quality of service levels of the link. The method may further include processing each of the plurality of segments of the data packet based upon a corresponding quality. The method may further include transmitting the multi- level packet to an application.

[0025] In accordance with certain example embodiments, an apparatus may include atleast one processor and at least one memory storing instructions that, when executed bythe at least one processor, cause the apparatus at least to establish at least one link withan application configured for a plurality of quality of service levels of the link. The atleast one memory and instructions, when executed by the at least one processor, mayfurther cause the apparatus at least to process each of the plurality of segments of thedata packet based upon a corresponding quality. The at least one memory andinstructions, when executed by the at least one processor, may further cause theapparatus at least to transmit the multi-level packet to an application.

[0026] In accordance with various example embodiments, an apparatus may includeestablishing circuitry configured to perform establishing at least one link with anapplication configured for a plurality of quality of service levels of the link. Theapparatus may further include processing circuitry configured to perform processingeach of the plurality of segments of the data packet based upon a corresponding quality.The apparatus may further include transmitting circuitry configured to performtransmitting the multi-level packet to an application.

[0027] In accordance with some example embodiments, a method may includetransforming, by a first computing device, a first bit error rate pattern of a first end-to-end channel associated with at least one of a first set of bit error rates to a second biterror rate pattern of a second end-to-end channel associated with at least one of a secondset of bit error rates. The transforming comprises changing a packet length of the firstend-to-end channel. The transforming is performed using an error rate conversion code encoder and an error rate conversion code decoder.

[0028] In accordance with certain example embodiments, an apparatus may includemeans for transforming a first bit error rate pattern of a first end-to-end channel associated with at least one of a first set of bit error rates to a second bit error rate pattern of a second end-to-end channel associated with at least one of a second set of bit errorrates. The transforming comprises changing a packet length of the first end-to-endchannel. The transforming is performed using an error rate conversion code encoder and an error rate conversion code decoder.

[0029] In accordance with various example embodiments, a non-transitory computerreadable medium may include program instructions that, when executed by an apparatus, cause the apparatus to perform at least a method. The method may include transforming a first bit error rate pattern of a first end-to-end channel associated with at least one of a first set of bit error rates to a second bit error rate pattern of a second end-to-end channel associated with at least one of a second set of bit error rates. Thetransforming comprises changing a packet length of the first end-to-end channel. The transforming is performed using an error rate conversion code encoder and an error rate conversion code decoder.

[0030] In accordance with some example embodiments, a computer program productmay perform a method. The method may include transforming a first bit error rate pattern of a first end-to-end channel associated with at least one of a first set of bit error rates to a second bit error rate pattern of a second end-to-end channel associated with atleast one of a second set of bit error rates. The transforming comprises changing a packetlength of the first end-to-end channel. The transforming is performed using an error rate conversion code encoder and an error rate conversion code decoder.

[0031] In accordance with certain example embodiments, an apparatus may include atleast one processor and at least one memory storing instructions that, when executed bythe at least one processor, cause the apparatus at least to transform a first bit error ratepattern of a first end-to-end channel associated with at least one of a first set of bit error rates to a second bit error rate pattern of a second end-to-end channel associated with atleast one of a second set of bit error rates. The transforming comprises changing a packetlength of the first end-to-end channel. The transforming is performed using an error rate conversion code encoder and an error rate conversion code decoder.

[0032] In accordance with various example embodiments, an apparatus may includetransforming circuitry configured to perform transforming a first bit error rate pattern ofa first end-to-end channel associated with at least one of a first set of bit error rates to asecond bit error rate pattern of a second end-to-end channel associated with at least oneof a second set of bit error rates. The transforming comprises changing a packet lengthof the first end-to-end channel. The transforming is performed using an error rate conversion code encoder and an error rate conversion code decoder.

[0033] In accordance with some example embodiments, a method may includereceiving, by a first computing device, from a second computing device, a first bit errorrate pattern of a first end-to-end channel associated with at least one of a first set of biterror rates, the first bit error rate pattern having been transformed from a second bit error rate pattern of a second end-to-end channel associated with at least one of a second set of bit error rates by changing a packet length of the second end-to-end channel using an error rate conversion code encoder and error rate conversion code decoder.

[0034] In accordance with certain example embodiments, an apparatus may includemeans for receiving, by a computing device, a first bit error rate pattern of a first end-to-end channel associated with at least one of a first set of bit error rates, the first biterror rate pattern having been transformed from a second bit error rate pattern of a second end-to-end channel associated with at least one of a second set of bit error rates by changing a packet length of the second end-to-end channel using an error rate conversion code encoder and error rate conversion code decoder.

[0035] In accordance with various example embodiments, a non-transitory computerreadable medium may include program instructions that, when executed by an apparatus, cause the apparatus to perform at least a method. The method may further include receiving a first bit error rate pattern of a first end-to-end channel associated with at least one of a first set of bit error rates, the first bit error rate pattern having been transformed from a second bit error rate pattern of a second end-to-end channel associated with at least one of a second set of bit error rates by changing a packet length of the second end-to-end channel using an error rate conversion code encoder and error rate conversion code decoder.

[0036] In accordance with some example embodiments, a computer program productmay perform a method. The method may include receiving a first bit error rate patternof a first end-to-end channel associated with at least one of a first set of bit error rates, the first bit error rate pattern having been transformed from a second bit error rate pattern of a second end-to-end channel associated with at least one of a second set of bit error rates by changing a packet length of the second end-to-end channel using an error rate conversion code encoder and error rate conversion code decoder.

[0037] In accordance with certain example embodiments, an apparatus may include atleast one processor and at least one memory storing instructions that, when executed bythe at least one processor, cause the apparatus at least to receive a first bit error ratepattern of a first end-to-end channel associated with at least one of a first set of bit error rates, the first bit error rate pattern having been transformed from a second bit error rate pattern of a second end-to-end channel associated with at least one of a second set of bit error rates by changing a packet length of the second end-to-end channel using an error rate conversion code encoder and error rate conversion code decoder.

[0038] In accordance with various example embodiments, an apparatus may includereceiving circuitry configured to perform receiving a first bit error rate pattern of a firstend-to-end channel associated with at least one of a first set of bit error rates, the first bit error rate pattern having been transformed from a second bit error rate pattern of a second end-to-end channel associated with at least one of a second set of bit error rates by changing a packet length of the second end-to-end channel using an error rate conversion code encoder and error rate conversion code decoder. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] For a proper understanding of example embodiments, reference should be made to the accompanying drawings, wherein:

[0040] FIG. 1A illustrates an example semantic communication system;

[0041] FIG. 1B illustrates another example semantic communication system;

[0042] FIG. 2A illustrates an example of end-to-end training semanticcommunication;

[0043] FIG. 2B illustrates an example of semantic compression;

[0044] FIG. 2C illustrates an example of semantic networking;

[0045] FIG.3 illustrates an interface;

[0046] FIG.4 illustrates an exemplary parameter setting;

[0047] FIG.5 illustrates the role of an interface from the perspective of an application,where the wireless channel is represented (e.g., by additive white Gaussian noise (AWGN)).

[0048] FIG.6 illustrates the role of an interface from the perspective of the network, where the wireless channel is represented (e.g., by AWGN), and the API to report / get loss is shown from source decoder to channel decoder.

[0049] FIG.7 illustrates an example of a signaling diagram according to certain exampleembodiments;

[0050] FIG. 8 illustrates an example of another signaling diagram according to someexample embodiments;

[0051] FIG.9 illustrates an example of how an application can use the multi-level blockerror rate (BER) interface to optimize the source code;

[0052] FIG. 10 illustrates an example of how the network layer optimizes its functionalities using the multi-level BER interface;

[0053] FIG. 11 illustrates an exemplary parameter setting of the BER levels for N=10 BER levels;

[0054] FIG. 12 illustrates a comparison of various example embodiments to some techniques that do not use a multi-level BER interface;

[0055] FIG. 13 illustrates the gain of multi-level (N=10) vs. single-level (N=1) interface, both in performance and graceful degradation;

[0056] FIG.14 illustrates the robustness of the interface against variation;

[0057] FIG. 15 illustrates an example of a flow diagram of a method according to various example embodiments;

[0058] FIG.16 illustrates an example of a multi-level packet with header content;

[0059] FIG. 17 illustrates an example of RAN packet segment quality of service (QoS)multiplexing;

[0060] FIG.18 illustrates an example of receiver packet QoS demultiplexing;

[0061] FIG. 19 illustrates an example of a flow diagram of a method according to various example embodiments;

[0062] FIG. 20 illustrates an example of a flow diagram of a method according to various example embodiments;

[0063] FIG.21 illustrates a training process for non-uniform source encoder / decoder according to certain example embodiments;

[0064] FIG. 22 illustrates an example of importance patterns for source encoder / decoder depending on the network resource availability according to some example embodiments;

[0065] FIG. 23 illustrates an example of a comparison between a segmented importance pattern and continuous importance pattern;

[0066] FIG. 24 illustrates a training process for source encoder / decoder according tovarious example embodiments;

[0067] FIG. 25 illustrates an example of a schematic representation of error rateconversion code (ERCC) according to certain example embodiments;

[0068] FIG. 26 illustrates a training process for error rate conversion autoencodercode (ERCAC) according to some example embodiments;

[0069] FIG. 27 illustrates another training process for ERCAC according to variousexample embodiments;

[0070] FIG. 28 illustrates segmented error conversion using ERCACs according to certain example embodiments;

[0071] FIG.29 illustrates a training process for segmented ERCAC according to some example embodiments;

[0072] FIG.30 illustrates an example embodiment using ERCC according to various example embodiments;

[0073] FIG. 31 illustrates another example embodiment using ERCAC according to certain example embodiments;

[0074] FIG. 32 illustrates an example of a flow diagram of a method according to various example embodiments;

[0075] FIG. 33 illustrates an example of a flow diagram of a method according to certain example embodiments;

[0076] FIG. 34 illustrates an example of a flow diagram of a method according to some example embodiments;

[0077] FIG. 35 illustrates an example of a flow diagram of a method according tovarious example embodiments;

[0078] FIG. 36 illustrates an example of a flow diagram of a method according tocertain example embodiments;

[0079] FIG. 37 illustrates an example of a flow diagram of a method according tosome example embodiments;

[0080] FIG. 38 illustrates an example of various network devices according to someexample embodiments; and

[0081] FIG. 39 illustrates an example of a 5G network and system architectureaccording to certain example embodiments. DETAILED DESCRIPTION

[0082] It will be readily understood that the components of certain exampleembodiments, as generally described and illustrated in the figures herein, may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of some example embodiments of systems, methods,apparatuses, and computer program products for semantic communication involvingJSCC over current communication networks is not intended to limit the scope ofcertain example embodiments, but is instead representative of selected example embodiments.

[0083] In the technical field of semantic communications, communication links canconvey semantics of a message, while in contrast, traditional communicationstransmit the message itself as accurately as possible. Semantic communications relyon transmitting symbols based upon context information available at the receiver,which can significantly reduce the communication resources required. FIG. 1Aillustrates an example of a semantic communications system, where the end-to-end system conveys required information for face recognition at an edge server, rather than a full video stream from a camera. Similarly, FIG. 1B illustrates an effective communication system designed to improve the effectiveness of interactions betweena cloud controller and a remote-controlled robot.

[0084] Semantic communication can be implemented in a variety of ways, such asJSCC. For example, FIGs. 2A-C illustrate three categories of semanticcommunication: FIG. 2A depicts end-to-end training owned by a single entity; FIG.2B depicts semantic compression within an application or within a network; and FIG.2C depicts semantic networking via an interface, separately but coordinately. JSCCcombines application and network level codes into a single code that maps from source to channel input and vice versa, which can directly optimize application performance. Currently, cellular technologies (e.g., 5G NR), as well as generalTCP / IP networks, do not support JSCC transmission of data packets since JSCCrequires an application to have instantaneous knowledge of the channel state information (CSI). This can be impractical since the base station / access point aresimply intermediate network nodes that may be unaware of the source applicationaddress. Although JSCC may be performed at the base station or access point, thisrequires the base station to be a destination node for the source, terminating theTCP / IP connection, and then creating a new connection directly to the device with the base station as the source node. In addition to being unscalable, this technique would require the source application to reveal the source data to the base station.

[0085] As noted above, general semantic / effective communication may involve JSCC, where the source encoding / decoding and channel encoding / decoding are performed jointly to minimize the volume, and maximize the effectiveness, of data transfers. However, this can be infeasible based on how general purpose TCP / IPcommunication networks are designed that limit the practical deployment of JSCCdriven semantic communication solutions.

[0086] Specifically, the source encoder / decoder must be separated from the channel encoder / decoder by a binary interface. This models the wireline connection that typically connects the application to the base station or access point. JSCC-based solutions directly map source to channel input symbols, which assumes that real numbers can be passed through wireline connections.

[0087] Furthermore, the application should be able to encrypt the data that it transmitsover the network independently. JSCC-based solutions require both the source and channel code to be jointly designed by either the application or the network, and when it is designed by the network, the application must transmit the source information to the network without encryption, thereby preventing encryption of transmitted data.

[0088] In addition, the application and network operators must be able toindependently design their respective coding schemes (e.g., compression and channel codes), without requiring synchronized training. As described above, JSCC-based solutions require both the source and channel code to be jointly designed by either the application or the network, thereby preventing independent design of coding schemes.

[0089] Moreover, the source only has limited, if any, knowledge of the channel condition (e.g., channel signal to noise ration (SNR)). When the JSCC is designed by the application, the application would require instantaneous knowledge of the channel condition to adapt the coding scheme in real-time, thereby preventing knowledge of channel conditions.

[0090] Furthermore, certain cellular technologies (e.g., 5G NR) transmit data packets by treating a radio bearer on the basis of a QoS / quality of experience (QoE) set of requirements. Additionally, data packets that fail decoding before the latency budget are dropped from the buffer. For a wide variety of applications, the erroneous packet can still be useful for the application.

[0091] A group of such applications may perform multiple tasks (e.g., reconstructinga video and detecting objects in the video), with each take having a different end-to-end QoS requirement. This may also apply to other low latency applications (e.g.,video streaming, extended reality (XR), audio, etc). For those applications, currentsolutions establish multiple radio links (i.e., one per QoS set), with a significantoverlap among the contents. In some case (e.g., video streaming), hierarchical codingwith multiple radio links is prevalent, with one link per coding level; thus, themessage (e.g., a video frame) may be encoded to different layers, where the bottomlayer is the critical portion to be delivered in order to maintain the video connection, while the other layers are complimentary in improving the resolution and quality of experience, and are thus not as critical as the bottom layer.

[0092] Applications that have multiple downstream tasks (e.g., reconstructing a video and detecting objects in the video, each with a different end-to-end latency requirement) may be willing to accept different levels of reliability for each task. For example, a self-driving vehicle transmitting a video stream to a cloud server that both stores the video for later examination and to detect objects in the scene may require the detection task to be done with very low latency, while the video storage task is less latency sensitive. Since the semantic information in the video frames that is useful for object detection is also sparse, the application may encode it with extra redundancy without substantially increasing the packet size. This means that the application may be transmitting packets with different QoS requirements, where some may be able to accept errors in the received packet. Although current applications, when running over a wireless technology link (e.g., a 3GPP link of NR or LTE type) may establish multiple links over the top, where each level of the hierarchical coder is mapped to one of those links with its corresponding QoS, thishas the disadvantage of increasing overhead for network to manage multiple linksinstead of one link.

[0093] Furthermore, current semantic / effective communication solutions rely on JSCC, wherein the source encoding / decoding and channel encoding / decoding areperformed jointly to minimize the volume of data transfer and maximize effectiveness of it. This can be infeasible within the way current cellular technologies are designed, which is on the basis of separating those functions between the application and the network.

[0094] For example, in applications such as a CCTV camera that transmits a video stream to a cloud server that both stores the video for later examination and detects objects in the scene, the detection task needs to be done with very low latency, while the video storage task (we refer to it as non-critical information) is less latency sensitive. Since the semantic information in the video frames that is useful for object detection (we refer to it as critical information) is also sparse, the application can encode it with extra redundancy without substantially increasing the packet size. Thus, the application may be transmitting packets with different importance, where some parts of the information may be able to accept errors in the received packet.

[0095] Similarly, for an effective communication use case such as robotic control, a control message may include direction and number of steps of movement to avoid bumping into objects, while less critical aspects may include velocity and acceleration pattern to follow, which are used to improve energy efficiency but are less critical to the control and can follow the zero-hold assumption if missed.

[0096] A JSCC coder may be effective for semantic / effective type of communications if it gives non-uniform importance to bits during source coding. The non-uniformsource coding treatment can encode smore critical parts of the information with moreredundancy and vice versa.

[0097] Semantic / effective communication with JSCC may be more efficient when the knowledge of channel state is utilized in JSCC. However, this requires to either create a feedback link from network to the app to consistently share channel state with the app so that app performs JSCC, or, for the application to share its information with network and allow for JSCC to be done at the network. However, the former creates a high overhead, while the latter breaches privacy.

[0098] Certain example embodiments described herein may have various benefits and / or advantages to overcome the disadvantages described above. For example, certainexample embodiments may be beneficial for large data traffic (e.g., video, videogaming, augmented reality (AR) / virtual reality (VR), meta-verse type of traffic) that would pass through at least one link of wireless transmission (e.g., cellular, WiFi, etc.). This interface, via the differentiated multi-level BER specifications, mayenhance the robustness of both application and network functions against channeldegradation, and / or may be adopted by systems with non-standardized interface, suchas short-distance wireless transmission between XR equipment. Some exampleembodiments also address the shortcomings of current radio technologies by providing such applications with a decent level of flexibility and efficiency in QoS handling.

[0099] Various example embodiments may provide an efficient alternative solutionthat sits between establishing one separate radio link per segment with the requiredQoS. For example, certain example embodiments may reduce the overhead ofestablishing multiple links by the application, and may provide more flexibility for the app in activating and deactivating the use of each sub-channel on a time slot basis.In an example of a multi-task video streaming, the need for a semantic link and thedata rate required by it, as well as the number of required levels, may change as theenvironment and the objects change over time. In some example embodiments, theapplication may use one established link while modifying the segmentation and corresponding QoS over time without the need for managing a time-varying number of links.

[0100] Certain example embodiments may provide an efficient alternative solutionthat sits between establishing one link for the whole packet with one level of QoS(i.e., treating the whole packet with the most critical QoS level). For example, theresource savings of in example embodiments may be significant. Transmitting videoframes over a reliable and low latency link may utilize significantly larger resources compared to transmitting object and pose detection information over reliable QoS,while transmitting the rest of the information in the frame with reliability and latency relaxed QoS.

[0101] Furthermore, various example embodiments may provide an efficient end-to- end semantic / effective communication service with minimal modifications imposed on network arrangements, and minimal load imposed on the network operation. The codes (i.e., ERCCs) may encode the source encoded bits at the application beforepassing them to the network, thereby creating an output bit string that can be treateduniformly by the network. This may minimize the overhead on network performance, and more importantly, only need a low-cost (e.g., high BER) link from the network which reduces the strain on network resources. For the network, this may cost less to provide a high BER link, and be more stable over time with least fluctuations ofservice level. This may also benefit the JSCC efficiency without the need forinstantaneous CSI. JSCC may generally be designed to a specific source distribution—accounting for channel distribution on top of that in the design results in more complex design (in case of algorithmic codes) and / or need for more complex training and dataset size (in case of ML based data-driven codes). Various example embodiments may reduce that complexity and the need for exchanging channel state from the network to the JSCC. Thus, certain example embodiments discussed below are directed to improvements in computer-related technology.

[0102] Certain example embodiments relate to the context of effective and semantic communications, where the communication link is designed to convey semantics ofa message as effectively as possible, rather than the message itself. Certain exampleembodiments relate to multi-level BER interface and techniques that bridge an application with a network, enabling the application and network to independently but coordinately optimize their behavior, while also overcoming the disadvantagesdiscussed above. For example, a multi-level bit error rate (BER) interface mayinclude a binary vector interface, as illustrated in FIG. 3, wherein different sectionsof the bits in that vector may be subjected to different level of BERs. FIG. 4 illustratesan exemplary parameter setting associated with this binary vector interface. A multi-level BER interface may also include an API that enables the application.Furthermore, various example embodiments relates to application usingcommunication service where the application can tolerate different levels of distortion for different parts of the communication packet.

[0103] FIGs. 5 and 6 illustrate the role of an interface from the perspectives of theapplication and of the network, respectively, where the wireless channel is represented (e.g., by AWGN).

[0104] As discussed in more detail below, FIGs. 7 and 8 provide examples of techniques to enable the joint design of source and channel coding through a common interface, while maintaining the separate entities of the application and network providers to be independent of each other, while satisfying the restrictions of JSCC described above. The same performance provided by JSCC-based semantic communications can be achieved using the multi-level BER interface, and thesolution provided by the multi-level BER interface may be deployed in currentTCP / IP communication networks with minimal overhead.

[0105] By using the example embodiments described herein, the resultant source and channel codes satisfy the constraints. For example, the source code may be binary, which can be easily transmitted over a typical TCP / IP link, and the network layeroptimization may be done by the loss reported by the application and does not requirethe decryption of the packet. In addition, the source and channel codes are optimizedover a two-step process, and the source code remains constant in the optimization of the network parameters, which includes the channel code optimization. The source code is also optimized with respect to predefined BER levels in the multi-level BER interface. By designing the source code to be reliable over a wide range of BER levels, the source code is reliable over a wide range of corresponding real channel conditions. Therefore, it does not need to know instantaneous channel condition.

[0106] The multi-level BER interface defines several properties, including a set of NBSC subchannels, each with bit error probability ^^, where ^ denotes an index of thesubchannel; and an API for the application to report the loss to the network (e.g., / report_loss endpoint). The multi-level BER interface may also define an API for the network to obtain the loss reported by the application associated with a particular transmission (e.g., / get_loss endpoint).

[0107] The multi-level BER interface may also define a set of subchannels, each with a specific BER. These subchannels may act as an abstraction of the underlying truechannel (e.g., wireless channel medium), which the application may use to design abinary source code that is reliable under those conditions. Multiple levels (rather thanone level) may adapt the source code to a wide range of underlying channelconditions, since the application is assumed to not have instantaneous channel information. The interface also offers an API for reporting the application loss function. This may be used to optimize the network parameters, after the source code has been optimized using the subchannels. By passing the loss observed at the application, the network can optimize its channel code, power allocation, andmodulation order with respect to the application loss directly, instead of the typicalBLER metrics used in current communication systems. The API may also provide anasynchronous interface to report and obtain the loss by the application and by the network, respectively. It could also function as an interactive query process.

[0108] FIG. 7 illustrates an example of a signaling diagram 700 depicting proceduresfor training. Application 710 and network 720 may be similar to NE 3810, as illustratedin FIG.38, according to certain example embodiments.

[0109] At operation 701, application 710 may transmit a semantics communicationssession request session request to network 720.

[0110] At operation 702, network 720 may transmit to application 710 a multi-levelBER (Ni, ɛi) API configured for loss reporting.

[0111] At operation 703, application 710 may train a source encoder and decoder basedupon (Ni, ɛi). For example, based upon the prescribed N binary symmetric channel(BSC) subchannels and the associated bit error probability of each subchannel (as defined in the multi-level BER interface), application 710 may generate n bits, which may be transmitted over the subchannels in any order application 710 chooses. Forexample, in FIG. 9, the source encoder may convert X into n bits, which may be sub-divided into ^^, each with n / N bits. ^^ may then transmitted over subchannel ^.Application 710 may choose how many bits are transmitted over each subchannel; it may choose to allocate more bits to some subchannels than others.

[0112] At the output of the subchannels, the bits may be recombined in the same order they were generated at the output of the source encoder to from the vector y, and acorresponding source decoder converts y to a target format ^ X. To optimize the sourceencoder and decoder, application 710 may minimize a loss function L(X, ^ X) byoptimizing the bits in ^^.

[0113] As an example, X may be an image, and ^ X a reconstruction of X. The lossfunction may be mean-squared error of each pixel between X and ^ X. One potentialoptimization procedure may be to transform min L(X, ^ X) into min − I(X; y) ≤, which may model the source encoder and decoder as generativeprocesses ^(^|^) and ^(^|^), respectively. As a result, the source encoder anddecoder may be designed without physical transmission of bits over the N BSC channels. The design may be a representation of a lossy wireless channel used byapplication 710 to design the encoder and decoder by assuming such presentation isthe actual channel.

[0114] At operation 704, application 710 may freeze the source encoder and decoder, and / or may generate a training dataset.

[0115] At operation 705, application 710 may transmit the training dataset to network 720, and may engage the loss reporting API.

[0116] At operation 706, network 720 may train the channel encoder and decoder tominimize the loss reported via API. For example, given a trained source code inoperation 703, network layer 720 may optimize its encoder / decoder by observing the loss incurred at the application layer. Here, channel encoder and decoder may referto a collection of network functionalities (e.g., channel coding, power allocation,modulation design) to achieve the function of bits-to-symbols and symbols-to-bitsmapping, respectively. First, application 720 may generate the source code as inoperation 703, and may transmit it over the multi-level BER interface to network 720. The BSC subchannels may be removed in this operation since it may be substituted with the real network and channel condition. After receiving the output bits from source code, network layer 720 may perform channel encoding, which may map the bits from source code to a set of complex channel symbols to be transmitted over awireless channel. Channel encoding may include forward error correction (FEC)coding, power allocation, modulation, PRB assignment, and other networkfunctionalities. FIG. 10 illustrates this process with an assumed AWGN channel. Atthe receiver, a channel decoder may convert the channel output v to bits y, which maybe passed through to application 710. Application 710 may then decode y into ^ X, asin operation 703, and compute loss ^ = L(X, ^ X).

[0117] In order to optimize the channel encoding and decoding, application 710 mayreport the loss ^ to the interface connecting and coordinating the network and theapplication using the / report_loss endpoint. Network layer 720 may then obtain ^using the / get_loss endpoint. Having ^, network layer 720 may then optimize the channel encoding and decoding process by minimizing ^. Extending the exampledescribed in operation 703, to minimize L(X, ^ X), network layer 720 may transformthe problem into min −treating the mapping from ^to ^ as a generative process ^(^|^).

[0118] Following the example of transmitting an image, as described in the operations above, the source encoder / decoder and the channel encoder / decoder may be parameterized as deep neural networks (DNNs), and perform the optimization as described. The subchannel reliabilities may be heuristically selected according to =0.4, ^^= 0.001, ^^ = , ^ = 2, … , ^ − 1, with N=10 subchannels.FIG. 11 illustrates a plot of the ^^ values.

[0119] FIG. 8 illustrates an example of a signaling diagram 800 depicting proceduresfor transmission. Application 820 and network 830 may be similar to NE 3810, asillustrated in FIG.38, according to certain example embodiments.

[0120] At operation 801, application 820 may transmit a semantics communicationssession request session request to network 830.

[0121] At operation 802, network 830 may transmit to application 820 a multi-levelBER (Ni, ɛi) API configured for loss reporting.

[0122] At operation 803, application 820 may map source to bits using a source encoder.

[0123] At operation 804, application 820 may transmit the mapped bits to network 830.

[0124] At operation 805, network 830 may map bits to symbols using a channel encoder.

[0125] At operation 806, network 830 may transmit symbols over a wireless channel.

[0126] At operation 807, network 830 may map received symbols to bits using a channeldecoder.

[0127] At operation 808, network 830 may transmit the mapped bits to application 820.

[0128] At operation 809, application 820 may map bits to source using the source decoder.

[0129] At operation 810, application 820 may calculate semantic loss.

[0130] At operation 811, application 820 may transmit the calculated semantic loss tonetwork 830 via API.

[0131] A comparison may be made to a baseline where the source code uses the state- of-the-art BPG compression codec, and the channel code uses 5G low-density parity check (LDPC) codes with 512 and 768 bits blocklength. An AWGN channel may be simulated during training and testing, and the end-to-end performance may be measured using peak signal-to-noise ratio (PSNR), which may be defined as PSNR

[0132] FIG. 12 compares the various example embodiments with baseline, whichshows that the various example embodiments perform better than the baseline acrossa wide range of channel conditions, showing the efficacy of the various example embodiments as well as the multi-level BER interface.

[0133] FIG. 13 illustrates the gain of multi-level (N=10) versus single-level (N=1)interface, both in performance and graceful degradation.

[0134] FIG.14 illustrates the robustness of the interface against variation, where theseperformance gains are achieved with more BER than specified in the interface.

[0135] Various example embodiments may establish the radio link control functionality so that it can serve a single radio link with a multitude of QoS / QoE sets, all within the same packet. This may improve the application’s control to effectively manage the resources utilized for a radio link for JSCC, multi-stream and hierarchical tasks, and effective / semantic communications. Additionally, this may enable theeffective information packet (EIP) treatment to be activated only for focused parts ofthe packet where it’s absolutely needed. Certain example embodiments may also enable the application to request for different levels of service requirement forsegments of the same packet, which can improve the efficiency of the end-to-end linkby passing the control over segmentation and dynamic change of corresponding QoSto the application. A multi-level QoS paradigm may address the disadvantages notedabove by allowing the application to select where different parts of the same packet are treated with different levels of QoS / QoE.

[0136] In certain example embodiments, a packet transmission configuration may beestablished where each packet encompasses M subchannels, each with its own distinct QoS, where the network guarantees QoS on per-subchannel basis.Furthermore, a header type may be used by the application to inform the networkabout the segmentation and corresponding QoS for each packet.

[0137] Certain example embodiments may reduce latency issues by using multipleQoS levels. For example, current cellular standards require that packets arrive withouterrors, leading to repetitions and retransmissions when errors are detected, and significant delays. However, some applications may accept some errors in the received packet if the packet can be delivered faster. Thus, the availability of many QoS levels may be beneficial for semantic / effective applications where the latency is the most important factor, rather than reliability. By lowering the reliability requirement (i.e., increasing the BLER requirement) for certain packets, the physicallayer may pass erroneous packets to the application, instead of requesting for retransmission. This may reduce the latency of end-to-end packet delivery.

[0138] In particular, various example embodiments may create M QoS levelsubchannels that can be access by the application, as well as a new packet format, where the application can specify in the header, which segments of the payload should be mapped to what QoS level. The QoS is defined by at least a combination of latency and BLER. For each QoS level subchannel, the network may guarantee that the required latency and BLER is achieved. Additionally, for some segments, EIP may be activated which may enable RAN to pass those segments to the application, evenif erroneously received. This may reduce latency as long as the APP can still extractuseful information from the erroneous segments. At the receiver, the receivedsegments may then be multiplexed together, depending on the latency requirement and maximum retransmission attempts either in one or multiple versions, and passed to the application with an appropriate header.

[0139] FIG. 15 illustrates an example of a flow diagram of a method 1500 that maybe performed by a NE, such as NE 3810 illustrated in FIG. 38, according to variousexample embodiments.

[0140] At step 1501, the method may include establishing a multi-level QoSconfiguration, with M number of QoS levels, and the required QoS for each level (e.g., two levels are assumed where the first level requires 1e-5BLER with retransmission and EIP activated for it, while the second level requires 1e-2BLERwithout retransmission and without EIP). The multi-level QoS configuration may bespecified (e.g., as an application type), and / or may be established through a handshake process between the application and the network.

[0141] At step 1502, the method may further include multiplexing desired segments of its packet into a M-level packet accompanied by a header, before handing it to the network. The header contains information about at least the size of the packetsegments corresponding to each QoS level. The semantic / effective application maycreate a multi-level QoS, with M QoS levels. FIG. 16 depicts an example of theapplication specifying QoS levels that each segment in the payload should be transmitted by in the header. The header segment is optional and may contain the EIP header.

[0142] At step 1503, the method may further include, upon receiving the multi-levelQoS packet, dividing the M-level packet into the M segments based on the headerstep 1502, and maps each segment to a corresponding QoS subchannel. Eachsubchannel is treated in scheduling and RRM based on its own QoS for that segment. Each subchannel has its own retransmissions (e.g., automatic repeat request (ARQ) and hybrid automatic repeat request (HARQ)) and EIP creation configurations, basedon the QoS. For example, this may be embodied at the RLC layer by creating Mlogical subchannels for one radio bearer, where the retransmission and EIP creationis managed by MAC and PHY layers. In case of a NR network this can be embodiedat the RLC layer by mapping a radio bearer to multiple logical channels. As shown in FIG.17, for each subchannel, RAN may perform resource allocation (e.g., physicalresource block (PRB) assignment, transmit queue ordering, channel code rate andmodulation, etc.) to meet the QoS specified for each subchannel.

[0143] At step 1504, the received segments may be multiplexed together with areceiver side header and passed to the application. For each segment, corresponding information in the header is included which includes at least information about whether the segment is included or dropped.

[0144] In certain example embodiments, one version of the received packet may becreated. This may be when the latency requirements of all subchannels are equivalent(e.g., live video stream, semantic applications).

[0145] In some example embodiments, multiple versions of the received packet maybe created and passed to the application (e.g., at the end of latency budget for eachsubchannel). This may apply when the latency requirements of subchannels aredifferent (e.g., video stream for object detection and storage).

[0146] In various example embodiments, at the receiver, upon receipt of packetsegments, each segment may be treated in terms of retransmissions and EIP contentaccording to its QoS. For example, as long as a segment is not decoded correctly and hasn’t reached max retransmissions, a new retransmission is requested for it. Additionally, after decoding is deemed unsuccessful, EIP content may be created for the segment. The multi-level packet may then be reconstructed at a link control layer(e.g., RLC layer in NR), and passed to the TCP / IP or application. Procedures 1800and 1900 are illustrated, respectively, in FIG. 18 (at lower layers (e.g., MAC and PHY)) and FIG.19 (link control layer operation).

[0147] In FIG. 18, at step 1801, a determination may be made as to whether thesegment has been decoded correctly. If the segment has been correctly decoded, at step 1802, the segment may be passed to the link control layer, and the procedure ends. However, if the segment has not been correctly decoded, at step 1803, a determination is made on whether a maximum number of retransmissions has been reached. If the maximum number of transmissions has not been reached, at step 1804, retransmission is requested, and the process returns to step 1801. However, if the maximum number of transmissions has been reached, then at step 1805, a determination is made of whether EIP is activated for the segment. If the EIP is activated for the segment, at 1806, an EIP packet may be created based on the buffer content and the requested EIP format, and the procedure ends. However, if EIP is not activated for the segment, at step 1807, the buffer is dropped and reported to the link control layer.

[0148] In FIG. 19, at step 1901, the method may include receiving the segmentsand / or the report for each segment from lower layers.

[0149] At step 1902, the method may further include creating headers according to the reports. The headers may differ from the transmitter side at least according to dropped segments and segment length (e.g., in case of EIP segment).

[0150] At step 1903, the method may include multiplexing the segments and the headers, and creating multi-level packets to pass to the TCP / IP, application, etc.

[0151] In some example embodiments, once the application receives the EIP, whichmay contain only parts of the original payload, it uses the EIP configurationinformation to makes use of the erroneous segments and identify the most likely message that was intended by the source.

[0152] Various example embodiments may use a non-uniform source coder, whichmay reduce network load and minimize reliance on the network. Particularly, a familyof novel codes may be used (i.e., error conversion codes) to encode the sourceencoded bits at the application before passing them to the network. The error conversion codes may create an output bit string that can be treated uniformly by thenetwork. This may minimize the overhead on network performance, and may needonly a low-cost (e.g., high BER) link from the network, which may reduce required network resources. Some example embodiments may also use a family of JSCCs which may be developed for device implementation and / or standardized device codecs that are tailored to a specific semantic data type.

[0153] To overcome the disadvantages noted above, in some example embodiments, the network may provide a stable and low-cost pipe for the application to transmit its JSCC encoded bits, thereby lowering the cost for network and minimizing varianceof service quality over time. In addition, the application may perform JSCC with non-uniform treatment of information bits efficient JSCC without the need forinstantaneous CSI. This may result in a device / cloud implementation solution thatcan also be standardized in form of distribution-specific multimedia codecs.

[0154] Some example embodiments may establish an error rate conversion mechanism as a JSCC solution (at the application or at the network) that takes as input the quality of the channel guaranteed by the network and the non-uniform treated soft bits / symbols from source coder, then delivers as output a stream of hard bits / symbols that are transmitted by the network over a low-cost and stable communication link. The conversion mechanism may add redundancy to the input soft bits according to the conversion rate of error from source to the network link guaranteed error rate.

[0155] Various example embodiments provide an implementation solution that can be used by a semantic / effective communication application (e.g., on a device or onthe cloud) to most efficiently utilize the network resources while avoiding strain on network for providing ultra-reliable links for the critical part of the information, utilizing an ERCC. The ERCC is implemented at the transmitter and receiver sides with a potential multiplexer and demultiplexer (e.g., when the ERCC operates on a segmented packet basis).

[0156] FIG. 20 illustrates an example of a flow diagram of a method 2000 that maybe performed by a NE, such as NE 3810 illustrated in FIG. 38, according to variousexample embodiments.

[0157] At step 2001, the method may include establishing a link with desired BERand packet length. For example, the packet length and BER may be derived based onthe source encoder output importance pattern and the rate of the error conversion code.

[0158] Specifically, the method may include establishing a stable and low-cost bit / symbol pipe by the network (e.g., a high BER pipe for a fixed length ofbits / symbols per time slot). The link may be initiated by the application requesting asemantic / effective link (e.g., EIP type). This may be done by calculating the desiredBER and the desired packet length (e.g., according to the source coder outputimportance pattern and error rate conversion mechanism in step 2003, below).

[0159] Various example embodiments may establish a stable and low-cost bit / symbolpipe by the network for the application. The reason for high BER pipe of bits is thatthe network may provide a high BER pipe, as opposed to low BER and low BLER links, where the network may utilize a significantly less relative amount of (radio) resources for the former. The resources needed to transfer a packet (per informationbit) may be significantly higher for higher reliability levels (e.g., to increase reliabilityfrom 99% to 99.99% the throughput can drop by more than 50%, depending on thelatency budget). The required attributes of such links may be determined jointly bythe application and network based on the underlying source and ERCC encoder.

[0160] Computing the desired packet length and BER for the link relates to various parameters, such as an output length from the source encoder, i.e., ^^^^, which in caseof segmented source coding it is represented as ^^^^ = ^^ + ⋯ + ^^, ^ denoting thenumber of segments. The parameters may also include a conversion rate of the ERCCand ERCAC (i.e., the rate required by the code (whether the block code embodimentor the autoencoder embodiment), represented by ^^^^ / ^^^^). In addition, the parameters may include the input BER of the ERCC / ERCAC, which may be determined by the non-uniform pattern of the source coder. The input BER may bedenoted by ^^^^^ = [^^^^, … , ^^^^] which is the tuple of the input BER for eachsegment. (in case of a continuous pattern, the length of each segment is assumed tobe one bit). Based upon these parameters, the output BER of the ERCC / ERCAC (i.e.,the desired BER for the link establishment) may be determined. For the block codeembodiment of the ERCC, the required length for each segment may be derived basedon the rate of the block ERCC code (i.e., ^^ = ^^ / ^^, where ^^ must satisfy(^^ , ⌈^^(1 − 2 ∗ ^^^^)⌉, ^^) codeSimilarly, for theERCAC embodiment, based on the rate of the autoencoder, the required length is derived as ^^ / ^^.

[0161] At step 2002, the method may further include choosing a non-uniform importance source coder. The importance pattern is chosen according to a guaranteedBER and packet length from the network. For example, the application may utilize asoft source encoder that is pre-trained and is shared between the transmitter and receiver sides of the application. The encoder may be used to source encode soft orhard information bits. The source coder may treat the information in a non-uniformway, such as according to a desired error rate pattern for the source information.

[0162] In various example embodiments, the application may choose one from amonga set of source encoders, given the link attributes guaranteed by the network. Forexample, when network is loaded and can guarantee a shorter packet, the applicationmay use a source coder with importance pattern that is more weighted towards criticalinformation. When the network can guarantee longer packets, the application may use a source coder that passes through more of the non-critical information.

[0163] As illustrated in FIG. 21, a non-uniform source encoder, similar to typicalsource encoders, may be designed by training a neural network to map an input string of information bits to an output string of coded bits, where a certain importance pattern is imposed on the output bits. This may enable the semantic source coder to encode more of the critical information in the high importance bits and vice versa,more of the less critical information in the low-importance bits. For example, duringthe training of the source encoder and decoder, the link between each segment of the encoder output to the same segment of the decoder input can be modeled by a communication link of a desired BER (corresponding inversely to the importance of that segment). Through this process, the semantic source encoder / decoder may then be trained to encode the critical parts of the message into the higher importance segments of the output, and vice versa, the non-critical parts of the information into the lower importance segments.

[0164] In certain example embodiments for source code training, the training of thesource coder may not require network usage and can be done offline by the application, and after training, the trained encoder / decoder may be shared among the transmitter and receiver sides of the application.

[0165] The application may train a set of pre-trained source encoders, eachcorresponding to a network link option, and may choose the best importance patternamong those given the link attributes from the network. This may include two cases of network link attributes: first, referring to a high-loaded network scenario, where network can guarantee ^^^^at packet length ^^, and second, referring to a low- loaded network scenario, where network can guarantee ^^^^at packet length ^^,where ^^ > ^^. This is illustrated in FIG. 22, where the solid importance patternrelates to the high-loaded case, and the dashed importance pattern relates to the low-loaded case. Both patterns provide the same source output bit string length; however,the solid pattern focuses importance on a shorter subset of the output bits to guaranteecritical information delivery in high-load network state. In contrast, the dashedpattern may allow for a large portion of the output bits to have moderate importancelevel when the network load is low and more bits of information can be communicated. As a result, the output of the application (i.e., after applying ERCC)for the dashed importance pattern will be a longer packet sizewhile it would beshorter for the solid importance pattern

[0166] Some example embodiments may include training where the output are soft bits. The encoder / decoder may be trained using a different setup as shown in FIG.24.A semantic goal-driven loss function may be used, aiming to minimize the loss overa large set of samples. The semantic loss function may depend on the application, for example, for the case of video communication for three goals of pose detection, face recognition and video storage, a combined weighted loss referring to the three goals may be created, where the weighting for each loss term projects the importance / criticality of each goal among the three goals noted. Additionally, the importance pattern can be generally seen in form of a segmented importance patternas shown in solid line in FIG. 23, along the string of bits / symbols, each segmentrepresenting a few to a few tens of bits, or a more continuous pattern as shown in green.

[0167] The soft output bit / symbols proposed may use the training framework shown in FIG. 24. FIG. 24 depicts an additive random noise to each output soft bit of the encoder before feeding it to the decoder, where the power of random noise for each bit is inversely proportional to the importance considered for that bit. For example, ^ in case of Gaussian noise, the noise power should follow ^^^= ^ ^^^^(1 / ^^), where^^^ is the error rate desired for source bit ^ determined by the desired importance,and ^^^^(^) =^ ^^^^^∫^^^^. The output of the NN may have an average power of 1.

[0168] When the desired packet length and BER combination cannot be guaranteedby the network (e.g., high load situations), the application may adjust its parametersto accommodate for that, such as by choosing a source coder that demands a shorter packet length, as discussed in step 2002. The application may choose one from among a set of source encoders, given the link attributes guaranteed by the network. TheERCC / ERCAC parameters may then be computed according to the modified source choice and the guaranteed link attributes.

[0169] At step 2003, the method may further include constructing an ERCC, or a set of those codes. The error rate conversion encoder / decoder may sit between the source encoder / decoder and the network link, and the error conversion rate may be determined based on the BER and packet length from step 2001, and the importancepattern of step 2002. For example, the application may construct an error rateconversion mechanism that converts the source coder importance pattern into the guaranteed error rate by the network link. The soft bit / symbols from the source coder may then be encoded by the error conversion coder and passed to the network.

[0170] In certain example embodiments, the error conversion coder may be realizedthrough algebraic codes (e.g., a block code), where each segment of the source coderoutput is passed to a separate error conversion code.

[0171] In some example embodiments, a neural network type encoder (e.g., anautoencoder) may be trained in advance for converting an input importance pattern to an error rate that is guaranteed by the network. Such autoencoders may be trained for specific input importance pattern and output error rate sets. Alternatively, a more general autoencoder may be trained that takes take as input the importance pattern and output error rate.

[0172] In various example embodiments, the network may guarantee the typical QoSrequirements (e.g., BLER) for a link statistically over time. Particularly, for shorterpackets and reliable links, guaranteeing a stable BLER or BER over time may be complex and resource hungry. For that reason, the network may establish a low-cost, high BER and stable link for the application. Then, the application may use a JSCC code to protect the bits over such high BER link, according to their importance to the source encoder / decoder. This may be implemented using the ERCC mechanism, as described below.

[0173] FIG. 25 schematically illustrates an ERCC code. A ERCC encoder / decodermay convert a link with a given ^^^^into a link with a desired ^^^^. FEC channelcodes may be a special case of the proposed ERCC codes, where ^^^^ = 0 (i.e.,FEC codes are targeted to correct all the errors caused by the underlying link with ^^^^, while the proposes ERCC codes accept a desired rate of error in the packet which has to be ^^^^or less).

[0174] Various example embodiments may be used together with a block code toestablish an ERCC encoder / decoder. The ERCC encoder may include a puncturingmodule which takes ^^ bits and punctures ^^(2 ∗ ^^^^) portion of the bits, andthen encodes the remaining bits using a FEC block code of (^^, ⌈^^(1 − 2 ∗^^^^)⌉, ^^) with error correcting ability of ^^^^^ bits. Thus, the block code design^ has a Hamming distance that satisfies ^^^^ > ^^ ∗ ^^^^.

[0175] For a stable link with given ^^^^and long enough strings of bits (large ^^), the receiver of the ERCC decoder may receive a string of bits with size ^^whichcontains ^^ ∗ ^^^^. The long codeword assumption and / or the assumption ofstability of the link may be compromised to some extent. For example., the networkmay provide an error rate which lies most of the time in (^^^^ − ^, ^^^^ + ^)range. Such ^ value may be approximated and included when constructing the blockchannel code mentioned above to guarantee service. This may be applied with otherFEC codes (e.g., convolutional codes such as Turbo code) to guarantee an error correction of a certain number of bits in the underlying link.

[0176] The ERCC decoder performs the following. First, it decodes the receivedpacket of ^^ using the decoder of the (^^, ⌈^^(1 − 2 ∗ ^^^^)⌉, ^^) block codementioned above. Then, to reconstruct the packet it attaches a random string of^^(2 ∗ ^^^^) bits to it, exactly to the spots punctured initially by the ERCCencoder. This guarantee may be a ^^ ∗ ^^^^ number of erroneous bits (i.e., ^^^^error rate).

[0177] The parameters of packet length and desired BER in step 2001 may be determined by the choice of channel code in the ERCC encoder / decoder. Forexample, for an input packet size of ^^ and target ^^^^ the desired packet length^^and desired error rate ^^^^for the established network link must satisfy(^^ , ⌈^^(1 − 2 ∗ ^^^^)⌉, ^^) code

[0178] In some example embodiments, a NN may be trained for performing theERCC encoder / decoder functions. Training the ERCAC may not depend on SNR (e.g., the training and inference is performed for normalized average power at inputand output), and may be for soft input bits and hard output bits. The training may beperformed for a given source encoder / decoder and over a simulated binary channel with ^^^^, as shown in FIG.26. The loss function in training the ERCAC may best selected as either a binary cross entropy function or the semantic loss function similar to FIG. 24. RNN may be used as the NN for this function approximation.

[0179] The training of the ERCAC encoder shown in FIG. 26 may be performed by using the assumed source encoder / decoder. However, the training can also be done without the source code, as shown in FIG.27, where the source encoder and decoder are not present, but in calculation of the loss at the output of the ERCAC decoder, a weighting is applied to the per bit loss which corresponds to the importance pattern desired.

[0180] In various example embodiments, a segmented version of the ERCC case, asshown in FIG.27, may be an alternative implementation, which may create a modularversion of the solution in FIG. 28 (also can be seen as ML version of the solution inFIG. 24).

[0181] The training for such modular approach can go through a training approach asdepicted in FIG. 29, where the loss function monitors maintaining (e.g., binary crossentropy) that tolerates up to ^^^^ ∗ ^^, to enforce the desired ^^^^ from the sourcepoint of view.

[0182] With respect to step 2003, FIG. 30 illustrates an example embodiment ofERCC, and FIG. 31 illustrates an example embodiment of ERCAC. The network mayprovide a link with certain BER and packet length (and latency as an optional metric).The packet length and BER may be used by the application to first derive theparameters of the source code and ECAC code. As shown in FIGs. 30 and 31, thecommunication may further involve a source coding followed by ECAC coding ofthe source information before passing the ^^^^bits to the network.

[0183] At step 2004, the method may further include, at the transmitter, encodingsource information first with the source encoder of step 2002, then with the ERCC ofstep 2003, and then transmitting to the network. Alternatively, at the receiver, the method may include decoding the bits received from the network first by the error rate conversion decoder, and then by the source decoder, and then transmitting to the destination.

[0184] For example, at the transmitter side, the source information (soft or hard bits) may be first encoded using the source encoder into soft symbols / bits, encoded by theERCC into hard bits, and transmitted to the network. At the receiver side of theapplication, the receiver may choose the error conversion decoder corresponding to the error conversion code in step 2003, based on the side information received fromthe transmitter side of the application, and may decode the received packet using theerror conversion code and source decoder.

[0185] FIG. 32 illustrates an example of a flow diagram of a method 3200 that maybe performed by a NE, such as NE 3810 illustrated in FIG. 38, according to various example embodiments.

[0186] At step 3201, the method may include receiving, by a first network entity, froma second network entity, a multi-level BER interface configuration comprising at leastone BER based upon the request for a communication session.

[0187] At step 3202, the method may further include generating, by the first network entity, an application source encoder and decoder configured to encode and decode, respectively, a plurality of levels corresponding to the at least one BER.

[0188] At step 3203, the method may further include training, by the first network entity,the application source encoder and decoder according to the BER.

[0189] In certain example embodiments, the method may further include, prior toreceiving the multi-level BER interface configuration, transmitting, by the first network entity, to the second network entity, a request for the communication session.

[0190] In some example embodiments, the method may further include, prior toreceiving the multi-level BER interface configuration, transmitting, by the first network entity, to the second network entity, a request for the multi-level BER interface configuration. The request for the multi-level BER interface configuration may be transmitted via a pre-trained source.

[0191] In various example embodiments, the method may further include transmitting,by the first network entity, to the second network entity, a training dataset associated with the application source encoder and decoder.

[0192] In certain example embodiments, the training dataset may be transmitted via anAPI configured to engage loss reporting.

[0193] In some example embodiments, the method may further include freezing, by thefirst network entity, the application source encoder and decoder, and generating, by the first network entity, the training dataset.

[0194] In various example embodiments, the method may further include receiving, bythe first network entity, from the second network entity, an approval for the requested semantics communication session.

[0195] FIG. 33 illustrates an example of a flow diagram of a method 3300 that may be performed by a NE, such as NE 3810 illustrated in FIG. 38, according to various example embodiments.

[0196] At step 3301, the method may include transmitting, by a first network entity, to a second network entity, a request for a semantics communication session.

[0197] At step 3302, the method may further include receiving, by the first network entity, from the second network entity, a multi-level BER interface configuration comprising at least one BER based upon the request for the semantics communicationsession. The multi-level BER may be received via an application programming interfaceconfigured for loss reporting.

[0198] At step 3303, the method may further include calculating, by the first network entity, a semantic loss of at least one data packet according to the received multi-level BER.

[0199] At step 3304, the method may further include transmitting, by the first networkentity, to the second network entity, the calculated semantic loss. The calculatedsemantic loss may be transmitted via an application programming interface.

[0200] FIG. 34 illustrates an example of a flow diagram of a method 3400 that maybe performed by a NE or a UE, such as NE 3810 or UE 3820 illustrated in FIG. 38,according to various example embodiments. For example, the UE may be in a cloud(e.g., a virtual function / application running on a cloud).

[0201] At step 3401, the method may include establishing at least one link configuredfor a plurality of QoS levels of the link.

[0202] At step 3402, the method may further include multiplexing a plurality of segments of a data packet into a multi-level packet with a corresponding header.

[0203] In certain example embodiments, the establishing may further includeconfiguration at least one of a range of acceptable sizes for each of the plurality of segments, a range of packet sizes for each of the plurality of segments, and a headerconfigured for the network entity to determine at least one of the plurality of QoS levelsis active within each of the plurality of packets and a corresponding segment size.

[0204] FIG. 35 illustrates an example of a flow diagram of a method 3500 that maybe performed by a NE or a UE, such as NE 3810 or UE 3820 illustrated in FIG. 38,according to various example embodiments.

[0205] At step 3501, the method may include establishing, by a network entity, at leastone link with an application configured for a plurality of QoS levels of the link.

[0206] At step 3502, the method may further include processing, by the network entity,each of the plurality of segments of the data packet based upon a corresponding QoS ina separate subchannel.

[0207] At step 3503, the method may further include transmitting, by the network entity, the multi-level packet to an application.

[0208] In certain example embodiments, the establishing may further includeconfiguration at least one of a range of acceptable sizes for each of the plurality of segments, a range of packet sizes for each of the plurality of segments, and a headerconfigured for the network entity to determine at least one of the plurality of QoS levelsis active within each of the plurality of packets and a corresponding segment size.

[0209] In some example embodiments, the establishing may be performed accordingto at least one of an application type, and a configuration established through a handshake process between the network entity and the application.

[0210] FIG. 36 illustrates an example of a flow diagram of a method 3600 that maybe performed by a NE or a UE, such as NE 3810 or UE 3820 illustrated in FIG. 38,according to various example embodiments.

[0211] At step 3601, the method may include transforming, by a first computing device, a first BER pattern of a first end-to-end channel associated with at least one of a first set of BERs to a second BER pattern of a second end-to-end channel associated with at least one of a second set of BERs.

[0212] The transforming may include changing a packet length of the first end-to-endchannel. The transforming may be performed using an ERCC encoder and an ERCCdecoder.

[0213] In certain example embodiments, the method may further include receiving,by the first computing device, from a second computing device, the first BER pattern,wherein the first BER pattern is configured to provide a multi-level BER comprising atleast one BER, wherein each BER is associated with a respective packet length;generating, by the first computing device, an application source encoder and application source decoder configured to encode and decode, respectively, a plurality of levels corresponding to the at least one BER; generating, by the first computing device, anERCC transformer based upon the application source encoder and the received firstBER pattern; and transforming the first BER pattern to the second BER patternaccording to the generated ERCC transformer.

[0214] In some example embodiments, the first computing device may include userequipment or cloud software application. In various example embodiments, thesecond computing device may include a network entity.

[0215] FIG. 37 illustrates an example of a flow diagram of a method 3700 that maybe performed by a NE or a UE, such as NE 3810 or UE 3820 illustrated in FIG. 38,according to various example embodiments.

[0216] At step 3701, the method may include receiving, by a first computing device,from a second computing device, a first BER pattern of a first end-to-end channelassociated with at least one of a first set of BERs, the first BER pattern having beentransformed from a second BER pattern of a second end-to-end channel associatedwith at least one of a second set of BERs by changing a packet length of the secondend-to-end channel using an ERCC encoder and ERCC decoder. In various exampleembodiments, the first BER pattern may be associated with at least one FEC channelcode.

[0217] In certain example embodiments, the first computing device may include anetwork entity. In some example embodiments, the second computing device may include user equipment or cloud software application.

[0218] In some example embodiments, the method may further include puncturing asubset of a predetermined number of bits; and encoding the unpunctured subset of the predetermined number of bits using a FEC block code comprising an error correcting ability of bits.

[0219] In various example embodiments, the ERCC encoder and ERCC decoder maybe trained by a neural network. In certain example embodiments, the ERCC encodermay be generated according to at least one block code wherein each segment of theERCC encoder output is transformed with a separate error conversion code. In someexample embodiments, the ERCC encoder may include a neural network type encoderconfigured to convert an input importance pattern to an error rate guaranteed by a network.

[0220] FIG. 38 illustrates an example of a system according to certain exampleembodiments. In one example embodiment, a system may include multiple devices, such as, for example, NE 3810 and / or UE 3820.

[0221] NE 3810 may be one or more of a base station (e.g., 3G UMTS NodeB, 4G LTEEvolved NodeB, or 5G NR Next Generation NodeB), a serving gateway, a server,and / or any other access node or combination thereof.

[0222] NE 3810 may further include at least one gNB-centralized unit (CU), which may be associated with at least one gNB-distributed unit (DU). The at least one gNB-CU and the at least one gNB-DU may be in communication via at least one F1 interface, at least one Xn-C interface, and / or at least one NG interface via a 5thgeneration core (5GC).

[0223] UE 3820 may include one or more of a mobile device, such as a mobile phone, smart phone, personal digital assistant (PDA), tablet, or portable media player, digital camera, pocket video camera, video game console, navigation unit, such as a global positioning system (GPS) device, desktop or laptop computer, single-location device, such as a sensor or smart meter, or any combination thereof. Furthermore, NE 3810 and / or UE 3820 may be one or more of a citizens broadband radio service device(CBSD). For example, UE 3820 may be in a cloud (e.g., a virtual function / applicationrunning on a cloud).

[0224] NE 3810 and / or UE 3820 may include at least one processor, respectively indicated as 3811 and 3821. Processors 3811 and 3821 may be embodied by any computational or data processing device, such as a central processing unit (CPU), application specific integrated circuit (ASIC), or comparable device. The processors may be implemented as a single controller, or a plurality of controllers or processors.

[0225] At least one memory may be provided in one or more of the devices, as indicated at 3812 and 3822. The memory may be fixed or removable. The memory may include computer program instructions or computer code contained therein. Memories 3812 and 3822 may independently be any suitable storage device, such as a non-transitorycomputer-readable medium. The term “non-transitory,” as used herein, maycorrespond to a limitation of the medium itself (i.e., tangible, not a signal) as opposedto a limitation on data storage persistency (e.g., random access memory (RAM) vs.read-only memory (ROM)). A hard disk drive (HDD), random access memory (RAM), flash memory, or other suitable memory may be used. The memories may becombined on a single integrated circuit as the processor, or may be separate from the one or more processors. Furthermore, the computer program instructions stored in the memory, and which may be processed by the processors, may be any suitable form of computer program code, for example, a compiled or interpreted computer program written in any suitable programming language.

[0226] Processors 3811 and 3821, memories 3812 and 3822, and any subset thereof,may be configured to provide means corresponding to the various blocks of FIGs. 1-37.Although not shown, the devices may also include positioning hardware, such as GPS or micro electrical mechanical system (MEMS) hardware, which may be used to determine a location of the device. Other sensors are also permitted, and may be configured to determine location, elevation, velocity, orientation, and so forth, such as barometers, compasses, and the like.

[0227] As shown in FIG.38, transceivers 3813 and 3823 may be provided, and one or more devices may also include at least one antenna, respectively illustrated as 3814 and 3824. The device may have many antennas, such as an array of antennas configured for multiple input multiple output (MIMO) communications, or multiple antennas for multiple RATs. Other configurations of these devices, for example, may be provided. Transceivers 3813 and 3823 may be a transmitter, a receiver, both a transmitter and a receiver, or a unit or device that may be configured both for transmission and reception.

[0228] The memory and the computer program instructions may be configured, with the processor for the particular device, to cause a hardware apparatus, such as UE, toperform any of the processes described above (i.e., FIGs. 1-37). Therefore, in certainexample embodiments, a non-transitory computer-readable medium may be encoded with computer instructions that, when executed in hardware, perform a process such as one of the processes described herein. Alternatively, certain example embodiments may be performed entirely in hardware.

[0229] In certain example embodiments, an apparatus may include circuitryconfigured to perform any of the processes or functions illustrated in FIGs. 1-37. Asused in this application, the term “circuitry” may refer to one or more or all of thefollowing: (a) hardware-only circuit implementations (such as implementations in only analog and / or digital circuitry), (b) combinations of hardware circuits and software, such as (as applicable): (i) a combination of analog and / or digital hardware circuit(s) with software / firmware and (ii) any portions of hardware processor(s) with software (including digital signal processor(s)), software, and memory(ies) that work together to cause an apparatus, such as a mobile phone or server, to perform various functions), and (c) hardware circuit(s) and or processor(s), such as a microprocessor(s) or a portion of a microprocessor(s), that requires software (e.g., firmware) for operation, but the software may not be present when it is not needed for operation. This definition of circuitry applies to all uses of this term in this application, including in any claims. As a further example, as used in this application, the term circuitry also covers an implementation of merely a hardware circuit or processor (or multiple processors) or portion of a hardware circuit or processor and its (or their) accompanying software and / or firmware. The term circuitry also covers, for example and if applicable to the particular claim element, a baseband integrated circuit or processor integrated circuit for a mobile device or a similar integrated circuit in server, a cellular network device, or other computing or network device.

[0230] FIG. 39 illustrates an example of a 5G network and system architectureaccording to certain example embodiments. Shown are multiple network functions that may be implemented as software operating as part of a network device or dedicated hardware, as a network device itself or dedicated hardware, or as a virtual function operating as a network device or dedicated hardware. The NE and UE illustrated in FIG.39 may be similar to NE 3810 and UE 3820, respectively. The user plane function (UPF)may provide services such as intra-RAT and inter-RAT mobility, routing and forwarding of data packets, inspection of packets, user plane QoS processing, buffering of downlink packets, and / or triggering of downlink data notifications. The application function (AF) may primarily interface with the core network to facilitate application usage of traffic routing and interact with the policy framework.

[0231] According to certain example embodiments, processors 3811 and 3821, andmemories 3812 and 3822, may be included in or may form a part of processingcircuitry or control circuitry. In addition, in some example embodiments, transceivers 3813 and 3823 may be included in or may form a part of transceiving circuitry.

[0232] In some example embodiments, an apparatus (e.g., NE 3810 and / or UE 3820)may include means for performing a method, a process, or any of the variants discussed herein. Examples of the means may include one or more processors, memory, controllers, transmitters, receivers, and / or computer program code forcausing the performance of the operations.

[0233] In various example embodiments, apparatus 3810 may be controlled by memory3812 and processor 3811 to receive, from a network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for a communication session; generate an application source encoder and decoder configured to encode and decode, respectively, a plurality of levels corresponding tothe at least one bit error rate; and train the application source encoder and decoderaccording to the bit error rate.

[0234] Certain example embodiments may be directed to an apparatus that includes means for performing any of the methods described herein including, for example, means for receiving, from a network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for a communication session; means for generating an application source encoder and decoder configured to encode and decode, respectively, a plurality of levelscorresponding to the at least one bit error rate; and means for training the applicationsource encoder and decoder according to the bit error rate.

[0235] In various example embodiments, apparatus 3810 may be controlled by memory3812 and processor 3811 to transmit, to a network entity, a request for a semantics communication session; receive, from the network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for the semantics communication session; calculate a semantic loss of at least one datapacket according to the received multi-level bit error rate; and transmit, to the network entity, the calculated semantic loss.

[0236] Certain example embodiments may be directed to an apparatus that includes means for performing any of the methods described herein including, for example, means for transmitting, to a network entity, a request for a semantics communication session; means for receiving, from the network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for the semantics communication session; means for calculating a semantic loss of at least one data packet according to the received multi-level bit error rate; and means for transmitting, to the network entity, the calculated semantic loss.

[0237] In various example embodiments, apparatus 3820 may be controlled by memory3822 and processor 3821 to establish at least one link configured for a plurality of QoS levels of the link; and multiplex a plurality of segments of a data packet into a multi- level packet with a corresponding header.

[0238] Certain example embodiments may be directed to an apparatus that includes means for performing any of the methods described herein including, for example,means for establishing at least one link configured for a plurality of QoS levels of thelink; and means for multiplexing a plurality of segments of a data packet into a multi- level packet with a corresponding header.

[0239] In various example embodiments, apparatus 3810 may be controlled by memory3812 and processor 3811 to establish at least one link with an application configured fora plurality of QoS levels of the link; process each of the plurality of segments of the datapacket based upon a corresponding QoS in a separate subchannel; and transmit themulti-level packet to an application.

[0240] Certain example embodiments may be directed to an apparatus that includes means for performing any of the methods described herein including, for example, means for establishing at least one link with an application configured for a plurality ofQoS levels of the link; means for processing each of the plurality of segments of thedata packet based upon a corresponding QoS in a separate subchannel; and means fortransmitting the multi-level packet to an application.

[0241] In various example embodiments, apparatus 3810 may be controlled by memory3812 and processor 3811 to transform a first bit error rate pattern of a first end-to-end channel associated with at least one of a first set of bit error rates to a second bit error rate pattern of a second end-to-end channel associated with at least one of a second set of bit error rates. The transforming may include changing a packet length of the firstend-to-end channel. The transforming may be performed using an ERCC encoder andan ERCC decoder.

[0242] Certain example embodiments may be directed to an apparatus that includes means for performing any of the methods described herein including, for example, means for transforming a first bit error rate pattern of a first end-to-end channel associated with at least one of a first set of bit error rates to a second bit error rate pattern of a second end-to-end channel associated with at least one of a second set of bit error rates. The transforming may include changing a packet length of the firstend-to-end channel. The transforming may be performed using an ERCC encoder andan ERCC decoder.

[0243] In various example embodiments, apparatus 3810 may be controlled by memory3812 and processor 3811 to receive, from a computing device, a first bit error rate pattern of a first end-to-end channel associated with at least one of a first set of bit error rates, the first bit error rate pattern having been transformed from a second bit error rate pattern of a second end-to-end channel associated with at least one of a second set of biterror rates by changing a packet length of the second end-to-end channel using an ERCCencoder and ERCC decoder.

[0244] Certain example embodiments may be directed to an apparatus that includes means for performing any of the methods described herein including, for example, means for receiving a first bit error rate pattern of a first end-to-end channel associated with at least one of a first set of bit error rates, the first bit error rate pattern having been transformed from a second bit error rate pattern of a second end-to-endchannel associated with at least one of a second set of bit error rates by changing apacket length of the second end-to-end channel using an ERCC encoder and ERCCdecoder.

[0245] The features, structures, or characteristics of example embodiments described throughout this specification may be combined in any suitable manner in one or more example embodiments. For example, the usage of the phrases “various embodiments,” “certain embodiments,” “some embodiments,” or other similar language throughout this specification refers to the fact that a particular feature, structure, or characteristic described in connection with an example embodiment may be included in at least one example embodiment. Thus, appearances of the phrases “in various embodiments,” “in certain embodiments,” “in some embodiments,” or other similar language throughout this specification does not necessarily all refer to the same group of example embodiments, and the described features, structures, or characteristics may be combined in any suitable manner in one or more example embodiments.

[0246] As used herein, “at least one of the following: ” and “at least one of ” and similar wording, where the list of two or more elements are joined by “and” or “or,” mean at least any one of the elements, or at least any two or more of the elements, or at least all the elements.

[0247] Additionally, if desired, the different functions or procedures discussed above may be performed in a different order and / or concurrently with each other. Furthermore, if desired, one or more of the described functions or procedures may be optional or maybe combined. As such, the description above should be considered as illustrative of theprinciples and teachings of certain example embodiments, and not in limitation thereof.

[0248] One having ordinary skill in the art will readily understand that the example embodiments discussed above may be practiced with procedures in a different order, and / or with hardware elements in configurations which are different than those which are disclosed. Therefore, although some embodiments have been described based upon these example embodiments, it would be apparent to those of skill in the art that certainmodifications, variations, and alternative constructions would be apparent, while remaining within the spirit and scope of the example embodiments.

[0249] Partial Glossary

[0250] 3GPP 3rd Generation Partnership Project

[0251] 5G 5th Generation

[0252] 5GC 5th Generation Core

[0253] 6G 6th Generation

[0254] AF Application Function

[0255] API Application Programming Interface

[0256] AR Augmented Reality

[0257] ARQ Automatic Repeat Request

[0258] ASIC Application Specific Integrated Circuit

[0259] AWGN Additive White Gaussian Noise

[0260] BER Block Error Rate

[0261] BLER Block Error Rate

[0262] BSC Binary Symmetric Channel

[0263] CBSD Citizens Broadband Radio Service Device

[0264] CCTV Closed Circuit Television

[0265] CPU Central Processing Unit

[0266] CRC Cyclic Redundancy Check

[0267] CU Centralized Unit

[0268] DNN Deep Neural Networks

[0269] DU Distributed Unit

[0270] EIP Effective Information Packet

[0271] eMBB Enhanced Mobile Broadband

[0272] eNB Evolved Node B

[0273] ERCC Error Rate Conversion Code

[0274] ERCAC Error Rate Conversion Autoencoder Code

[0275] ERL Effective Radio Link

[0276] FEC Forward Error Correction

[0277] gNB Next Generation Node B

[0278] GPS Global Positioning System

[0279] HARQ Hybrid Automatic Repeat Request

[0280] HDD Hard Disk Drive

[0281] IoT Internet of Things

[0282] JSCC Joint Source and Channel Coding

[0283] LLR Log-Likelihood Ratio

[0284] LTE Long-Term Evolution

[0285] LTE-A Long-Term Evolution Advanced

[0286] MAC Medium Access Control

[0287] MEMS Micro Electrical Mechanical System

[0288] MIMO Multiple Input Multiple Output

[0289] mMTC Massive Machine Type Communication

[0290] NE Network Entity

[0291] NG Next Generation

[0292] NG-eNB Next Generation Evolved Node B

[0293] NG-RAN Next Generation Radio Access Network

[0294] NR New Radio

[0295] PDA Personal Digital Assistance

[0296] PDCP Packet Data Convergence Protocol

[0297] PHY Physical

[0298] PRB Physical Resource Block

[0299] QoE Quality of Experience

[0300] QoS Quality of Service

[0301] RAM Random Access Memory

[0302] RAN Radio Access Network

[0303] RAT Radio Access Technology

[0304] RF Radio Frequency

[0305] RLC Radio Link Control

[0306] ROM Read-Only Memory

[0307] UE User Equipment

[0308] UMTS Universal Mobile Telecommunications System

[0309] UPF User Plane Function

[0310] URLLC Ultra-Reliable and Low-Latency Communication

[0311] UTRAN Universal Mobile Telecommunications System TerrestrialRadio Access Network

[0312] VR Virtual Reality

[0313] XR Extended Reality

Claims

WE CLAIM:

1. A method comprising:receiving, by a first network entity, from a second network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon therequest for a communication session;generating, by the first network entity, an application source encoder and decoderconfigured to encode and decode, respectively, a plurality of levels corresponding to theat least one bit error rate; and training, by the first network entity, the application source encoder and decoderaccording to the bit error rate.

2. The method of claim 1, further comprising, prior to receiving the multi-level bit error rate interface configuration: transmitting, by the first network entity, to the second network entity, a requestfor the communication session.

3. The method of claim 1, further comprising, prior to receiving the multi-level bit error rate interface configuration: transmitting, by the first network entity, to the second network entity, a requestfor the multi-level bit error rate interface configuration.

4. The method of claim 3, wherein the request for the multi-level bit errorrate interface configuration is transmitted via a pre-trained source.

5. The method of claim 1, further comprising:transmitting, by the first network entity, to the second network entity, a trainingdataset associated with the application source encoder and decoder.

6. The method of claim 5, wherein the training dataset is transmitted via anapplication programming interface configured to engage loss reporting.

7. The method of claim 1, further comprising:freezing, by the first network entity, the application source encoder and decoder;and generating, by the first network entity, the training dataset.

8. The method of claim 1, further comprising:receiving, by the first network entity, from the second network entity, anapproval for the requested semantics communication session.

9. A method comprising:transmitting, by a first network entity, to a second network entity, a request for asemantics communication session; receiving, by the first network entity, from the second network entity, a multi- level bit error rate interface configuration comprising at least one bit error rate based upon the request for the semantics communication session; calculating, by the first network entity, a semantic loss of at least one data packetaccording to the received multi-level bit error rate; and transmitting, by the first network entity, to the second network entity, thecalculated semantic loss.

10. The method of claim 9, wherein the multi-level bit error rate is receivedvia an application programming interface configured for loss reporting.

11. The method of claim 9, wherein the calculated semantic loss istransmitted via an application programming interface.

12. A method comprising:establishing, by user equipment, at least one link configured for a plurality of quality of service levels of the link; and multiplexing, by the user equipment, a plurality of segments of a data packet into a multi-level packet with a corresponding header.

13. The method of claim 12, wherein the establishing further comprisesconfiguration at least one of: a range of acceptable sizes for each of the plurality of segments; a range of packet sizes for each of the plurality of segments; and a header configured for the network entity to determine at least one of the plurality of quality of service levels is active within each of the plurality of packets and a corresponding segment size.

14. A method comprising:establishing, by a network entity, at least one link with an application configured for a plurality of quality of service levels of the link; processing, by the network entity, each of the plurality of segments of the data packet based upon a corresponding quality of service in a separate subchannel; and transmitting, by the network entity, the multi-level packet to an application.

15. The method of claim 14, wherein the establishing further comprisesconfiguration at least one of: a range of acceptable sizes for each of the plurality of segments; a range of packet sizes for each of the plurality of segments; and a header configured for the network entity to determine at least one of the plurality of quality of service levels is active within each of the plurality of packets and a corresponding segment size.

16. The method of claim 14, wherein the establishing is performed accordingto at least one of the following: an application type; a configuration established through a handshake process between the network entity and the application.

17. A method comprising:transforming, by a first computing device, a first bit error rate pattern of a firstend-to-end channel associated with at least one of a first set of bit error rates to a secondbit error rate pattern of a second end-to-end channel associated with at least one of a second set of bit error rates, wherein the transforming comprises changing a packet length of the first end-to-endchannel, and the transforming is performed using an error rate conversion code encoder and an error rate conversion code decoder.

18. The method of claim 17, further comprising:receiving, by the first computing device, from a second computing device, the first bit error rate pattern, wherein the first bit error rate pattern is configured to provide a multi-level bit error rate comprising at least one bit error rate, wherein each bit error rate is associated with a respective packet length; generating, by the first computing device, an application source encoder and application source decoder configured to encode and decode, respectively, a plurality of levels corresponding to the at least one bit error rate; generating, by the first computing device, an error rate conversion codetransformer based upon the application source encoder and the received first bit errorrate pattern; and transforming the first bit error rate pattern to the second bit error rate pattern according to the generated error rate conversion code transformer.

19. The method of claim 17, wherein the first computing device comprisesuser equipment or cloud software application.

20. The method of claim 17, wherein the second computing device comprisesa network entity.

21. A method comprising:receiving, by a first computing device, from a second computing device, a firstbit error rate pattern of a first end-to-end channel associated with at least one of a firstset of bit error rates, the first bit error rate pattern having been transformed from a second bit error rate pattern of a second end-to-end channel associated with at least one of a second set of bit error rates by changing a packet length of the second end-to-end channel using an error rate conversion code encoder and error rate conversion code decoder.

22. The method of claim 21, wherein the first computing device comprises anetwork entity.

23. The method of claim 21, wherein the second computing device comprisesuser equipment or cloud software application.

24. The method of claim 21, wherein the first bit error rate pattern isassociated with at least one forward error correction channel code.

25. The method of claim 21, further comprising:puncturing a subset of a predetermined number of bits; and encoding the unpunctured subset of the predetermined number of bits using aforward error correction block code comprising an error correcting ability of bits.

26. The method of claim 21, wherein the error rate conversion code encoderand error rate conversion code decoder are trained by a neural network.

27. The method of claim 21, wherein the error rate conversion code encoderis generated according to at least one block code wherein each segment of the error rateconversion code encoder output is transformed with a separate error conversion code.

28. The method of claim 21, wherein the error rate conversion code encodercomprises a neural network type encoder configured to convert an input importancepattern to an error rate guaranteed by a network.

29. An apparatus comprising:at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to: receive, from a network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for a communication session; generate an application source encoder and decoder configured to encode anddecode, respectively, a plurality of levels corresponding to the at least one bit error rate;and train the application source encoder and decoder according to the bit error rate.

30. An apparatus comprising:at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to: transmit, to a network entity, a request for a semantics communication session;receive, from the network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for the semantics communication session; calculate a semantic loss of at least one data packet according to the received multi-level bit error rate; and transmit, to the network entity, the calculated semantic loss.

31. An apparatus comprising:at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to: establish at least one link configured for a plurality of quality of service levels ofthe link; and multiplex a plurality of segments of a data packet into a multi-level packet with a corresponding header.

32. An apparatus comprising:at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to: establish at least one link with an application configured for a plurality of qualityof service levels of the link; process each of the plurality of segments of the data packet based upon a corresponding quality of service in a separate subchannel; and transmit the multi-level packet to an application.

33. An apparatus comprising:at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to:transform a first bit error rate pattern of a first end-to-end channel associated withat least one of a first set of bit error rates to a second bit error rate pattern of a secondend-to-end channel associated with at least one of a second set of bit error rates, wherein the transforming comprises changing a packet length of the first end-to-end channel, and the transforming is performed using an error rate conversion code encoder and an error rate conversion code decoder.

34. An apparatus comprising:at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to: receive, from a computing device, a first bit error rate pattern of a first end-to-end channel associated with at least one of a first set of bit error rates, the first bit errorrate pattern having been transformed from a second bit error rate pattern of a second end-to-end channel associated with at least one of a second set of bit error rates by changing a packet length of the second end-to-end channel using an error rate conversion code encoder and error rate conversion code decoder.

35. An apparatus comprising:means for receiving, from a network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for acommunication session;means for generating an application source encoder and decoder configured toencode and decode, respectively, a plurality of levels corresponding to the at least onebit error rate; and means for training the application source encoder and decoder according to thebit error rate.

36. An apparatus comprising:means for transmitting, to a network entity, a request for a semantics communication session; means for receiving, from the network entity, a multi-level bit error rate interface configuration comprising at least one bit error rate based upon the request for the semantics communication session; means for calculating a semantic loss of at least one data packet according to the received multi-level bit error rate; and means for transmitting, to the network entity, the calculated semantic loss.

37. An apparatus comprising:means for establishing at least one link configured for a plurality of quality ofservice levels of the link; and means for multiplexing a plurality of segments of a data packet into a multi-level packet with a corresponding header.

38. An apparatus comprising:means for establishing at least one link with an application configured for aplurality of quality of service levels of the link; means for processing each of the plurality of segments of the data packet based upon a corresponding quality of service in a separate subchannel; and means for transmitting the multi-level packet to an application.

39. An apparatus comprising:means for transforming a first bit error rate pattern of a first end-to-end channelassociated with at least one of a first set of bit error rates to a second bit error rate patternof a second end-to-end channel associated with at least one of a second set of bit error rates, wherein the transforming comprises changing a packet length of the first end-to-end channel, andthe transforming is performed using an error rate conversion code encoder and an error rate conversion code decoder.

40. An apparatus comprising:means for receiving, from a computing device, a first bit error rate pattern of afirst end-to-end channel associated with at least one of a first set of bit error rates, thefirst bit error rate pattern having been transformed from a second bit error rate pattern of a second end-to-end channel associated with at least one of a second set of bit error rates by changing a packet length of the second end-to-end channel using an error rate conversion code encoder and error rate conversion code decoder.

41. A non-transitory computer readable medium comprising programinstructions that, when executed by an apparatus, cause the apparatus to perform atleast a method according to any of claims 1-28.

42. An apparatus comprising circuitry configured to perform a methodaccording to any of claims 1-28.

43. A computer program comprising instructions, which, when executed byan apparatus, cause the apparatus to perform the method of any of claims 1-28.

Images