Compatibility check data
The generation and transmission of compatibility check data within two-sided machine learning models in wireless communication networks address the challenge of maintaining compatibility between vendor-specific devices, ensuring efficient and independent management of model parts, thereby enhancing ML functionalities and preventing performance degradation.
Patent Information
- Application Number
- PCT/EP2024/060264
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-11
- Filing Date
- 2024-04-16
- Publication Date
- 2024-11-14
AI Technical Summary
In wireless communication networks, especially in 5G and LTE, the challenge lies in maintaining compatibility between different vendor-specific wireless devices using two-sided machine learning models, where parts of the model are independently managed, leading to potential performance degradation and incompatibility issues.
The generation and transmission of compatibility check data using a part of the two-sided model between wireless devices allow for proactive compatibility checks, ensuring that the model parts remain aligned and functional without direct model transfer, enabling lifecycle management and avoiding performance degradation.
This approach ensures that the two-sided machine learning model parts remain compatible, preventing performance degradation and allowing for independent management by respective vendors, thus enhancing the exploitation of ML functionalities in communication networks.
Smart Images

Figure EP2024060264_14112024_PF_FP_ABST
Abstract
Description
COMPATIBILITY CHECK DATAFIELD
[0001] Various example embodiments relate in general to wireless communication networks and more specifically, to machine learning in such networks.BACKGROUND
[0002] Machine Learning, ML, functionalities may be exploited to enhance performance of wireless communication networks, such as cellular communication networks operating according to Long Term Evolution, LTE, and / or 5G radio access technology. 5G radio access technology may also be referred to as New Radio, NR, access technology. Since its inception, LTE has been widely deployed and 3rd Generation Partnership Project, 3GPP, still develops LTE. Similarly, 3GPP also develops standards for 5G / NR. Some topics in the 3GPP discussions are related to exploitation of ML functionalities and there is a need to provide improved methods, apparatuses and computer programs for exploitation of ML functionalities, in particularly for air interface.SUMMARY
[0003] According to some aspects, there is provided the subject-matter of the independent claims. Some example embodiments are defined in the dependent claims.
[0004] The scope of protection sought for various example embodiments of the invention is set out by the independent claims. The example embodiments and features, if any, described in this specification that do not fall under the scope of the independent claims are to be interpreted as examples useful for understanding various example embodiments of the invention.
[0005] According to a first aspect of the present invention, there is provided a wireless device comprising at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the wireless device at least to perform at least one operation concerning a machine learning functionality, wherein the machine learning functionality is associated with a two-sided model between the wireless device and another wireless device, generate, using a part of the two-sided model, compatibility checkdata concerning the machine learning functionality and transmit said compatibility check data to said another wireless device.
[0006] Example embodiments of the first aspect may comprise at least one feature from the following bulleted list or any combination of the following features:• wherein the wireless device and said another wireless device are devices of different vendors;• wherein the two-sided model is a paired model over which a joint inference is performed;• wherein said part of the two-sided model is independently managed from another part of the two-sided model;• wherein said part of the two-sided model is an encoder;• wherein the stored instructions cause, when executed by the at least one processor, the wireless device at least to transmit a request to generate input data for at least said part of the two-sided model, receive said input data and generate at least a part of said compatibility check data by inputting said input data to said part of the two-sided model;• wherein the stored instructions cause, when executed by the at least one processor, the wireless device at least to determine to trigger the compatibility check concerning the two-sided model and generate, after said determination, said compatibility check data;• wherein the stored instructions cause, when executed by the at least one processor, the wireless device at least to perform, after said determination, a handshake with said another wireless device to determine whether the wireless device or said another wireless device is to perform the compatibility check;• wherein the stored instructions cause, when executed by the at least one processor, the wireless device at least to determine to trigger said compatibility check based on at least one of a model update rate, total number of model updates, local update, performance degradation or a request received from said another wireless device;• wherein the at least one operation concerning the machine learning functionality comprises at least one of running the two-sided model, pre-processing of the two- sided model, post-processing of the two-sided model or configuration of the two- sided model;• wherein the stored instructions cause, when executed by the at least one processor, the wireless device at least to receive feedback from the second node based on the transmitted compatibility check data;• wherein said part of the two-sided model is run at the wireless device and said another part is run at said another wireless device.
[0007] According to a second aspect of the present invention, there is provided a wireless device comprising at least one processing core, at least one memory including computer program code, the at least one memory and the computer program code being configured to, with the at least one processing core, cause the wireless device at least to receive, from another wireless device, compatibility check data concerning a machine learning functionality, wherein the machine learning functionality is associated with a two- sided model between the wireless device and said another wireless device, and the two-sided model comprises a part and another part and perform a compatibility check by inputting at least said compatibility check data to said another part of the two-sided model.
[0008] Example embodiments of the second aspect may comprise at least one feature from the following bulleted list or any combination of the following features:• wherein the stored instructions further cause, when executed by the at least one processor, the wireless device at least to receive input data for the two-sided model and perform the compatibility check by comparing an output of said another part of the two-sided model to said input data;• wherein the stored instructions further cause, when executed by the at least one processor, the wireless device at least to transmit a result of the compatibility check to a vendor of the wireless device;• wherein said part of the two-sided model is run at the wireless device and said another part is run at said another wireless device.
[0009] According to a third aspect of the present invention, there is provided a method, comprising performing, by a wireless device, at least one operation concerning a machine learning functionality, wherein the machine learning functionality is associated with a two- sided model between the wireless device and another wireless device, generating by the wireless device, using a part of the two-sided model, compatibility check data concerningthe machine learning functionality and transmitting, by the wireless device, said compatibility check data to said another wireless device.
[0010] According to a fourth aspect of the present invention, there is provided a method, comprising receiving by a wireless device, from another wireless device, compatibility check data concerning a machine learning functionality, wherein the machine learning functionality is associated with a two-sided model between the wireless device and said another wireless device, and the two-sided model comprises a part and another part and performing, by the wireless device, a compatibility check by inputting at least said compatibility check data to said another part of the two-sided model.
[0011] According to a fifth aspect of the present invention, there is provided a wireless device, comprising means for performing at least one operation concerning a machine learning functionality, wherein the machine learning functionality is associated with a two- sided model between the wireless device and another wireless device, means for generating, using a part of the two-sided model, compatibility check data concerning the machine learning functionality and means for transmitting said compatibility check data to said another wireless device.
[0012] According to a sixth aspect of the present invention, there is provided a wireless device, comprising means for receiving, from another wireless device, compatibility check data concerning a machine learning functionality, wherein the machine learning functionality is associated with a two-sided model between the wireless device and said another wireless device, and the two-sided model comprises a part and another part and means for performing a compatibility check by inputting at least said compatibility check data to said another part of the two-sided model.
[0013] According to a seventh aspect of the present invention, there is provided a non- transitory computer readable medium having stored thereon a set of computer readable instructions that, when executed by at least one processor, cause an apparatus to at least perform the first or the second method. According to an eight aspect of the present invention, there is provided a computer program comprising instructions which, when the program is executed by an apparatus, cause the apparatus to carry out the first or the second method.BRIEF DESCRIPTION OF THE DRAWINGS
[0014] FIG. 1 illustrates an example of a network scenario in accordance with at least some example embodiments;
[0015] FIG. 2 illustrates a first signalling graph in accordance with at least some example embodiments;
[0016] FIG. 3 illustrates a second signalling graph accordance with at least some example embodiments;
[0017] FIG. 4 illustrates a third signalling graph in accordance with at least some example embodiments;
[0018] FIG. 5 illustrates an example apparatus capable of supporting at least some example embodiments;
[0019] FIG. 6 illustrates a flow graph of a first method in accordance with at least some example embodiments;
[0020] FIG. 7 illustrates a flow graph of a second method in accordance with at least some example embodiments.EXAMPLE EMBODIMENTS
[0021] Embodiments of the present disclosure provide enhancements related to Machine Learning, ML, functionalities in communication networks. More specifically, embodiments of the present disclosure make it possible to keep a two-sided model compatible, wherein the two-sided model is a model used by a wireless device and another device. The wireless device may use a part of the two-sided model and said another wireless device may use another part of the two-sided model. Said parts of the two-sided may be kept compatible by generating Compatibility Check Data, CCD, using said part of the two-sided model, but not the entire two-sided model, i.e., without using said another part. The wireless device may then transmit the CCD to said another wireless device and said another wireless device may perform a compatibility check by inputting at least the CCD to said another part of the two-sided model. Hence, exploitation of ML functionalities may be enhanced in communication networks by keeping said parts compatible.
[0022] FIG. 1 illustrates an example of a network scenario in accordance with at least some example embodiments. According to the example scenario of FIG. 1, there may be a communication system, which comprises wireless terminal 110, wireless network node 120 and core network element 130. Wireless terminal 110 and wireless network node 120 may be referred to as wireless devices.
[0023] Wireless terminal 110 may be a wireless device, like a User Equipment, UE. Wireless terminal 110 may comprise, or be incorporated into, for example a smartphone, a cellular phone, a Machine-to-Machine, M2M, node, Machine-Type Communications, MTC, node, an Internet of Things, loT, node, a car telemetry unit, a laptop computer, a tablet computer or, indeed, any kind of suitable wireless terminal, station or a relay.
[0024] In the example network scenario of FIG. 1, wireless terminal 110 may communicate wirelessly with wireless network node 120 via air interface 115. Air interface 115 may be a beam-based interface. That is, wireless terminal 110 and wireless network node 120 may communicate using one or more beams.
[0025] Air interface 115 between wireless terminal 110 and wireless network node 120 may be configured in accordance with a Radio Access Technology, RAT, which both terminal device 110 and wireless network node 120 are configured to support. Examples of cellular RATs include Long Term Evolution, LTE, New Radio, NR, which may also be known as fifth generation, 5G, radio access technology and MulteFire. On the other hand, examples of non-cellular RATs comprise Wireless Local Area Network, WLAN, and Worldwide Interoperability for Microwave Access, WiMAX.
[0026] For instance, example embodiments may be exploited in 5G / NR networks standardized by 3rd Generation Partnership Project, 3GPP, and / or in WLANs, like networks operating at least partly according to Institute of Electrical and Electronics Engineers, IEEE, 802.11 -standard. That is, wireless terminal 110 and wireless network node 120 may be configured to operate according to at least one standard specification specified by the 3 GPP and / or IEEE.
[0027] For example in the context of LTE, wireless network node 120 may be referred to as eNB while wireless network node 120 may be referred to as gNB in the context of NR. In some example embodiments, wireless network node 120 may be referred to as a Transmission and Reception Point, TRP, or control multiple TRPs that may be co-located ornon-co-located. In case of non-cellular RATs, wireless network node 120 may be referred to as an access point. In some example embodiments, wireless network node 120 may refer to a relay. In any case, example embodiments of the present disclosure are not restricted to any particular wireless technology. Instead, example embodiments may be exploited in any wireless communication system, wherein two-sided ML models are used.
[0028] In some example embodiments, the network scenario may comprise a relay node instead of, or in addition to, wireless terminal 110 and / or wireless network node 120. Relaying may be used for example when operating on millimeter-wave frequencies. One example of the relay node may be an Integrated Access and Backhaul, IAB, node. The IAB node may be referred to as a self-backhauling relay as well. Another example of a relay may be an out-band relay. In general, the relay node may comprise two parts:1) Distributed Unit, DU, part which may facilitate functionalities of wireless network node 120, such as a gNB. Thus, in some example embodiments, the DU part of a relay may be referred to as wireless network node 120 and the DU may perform tasks of wireless network node 120;2) Mobile Termination, MT, part which may facilitate functionalities of wireless terminal 110, i.e., a backhaul link which may be the communication link between a parent node (DU), such as a DU part of wireless network node 120, and the relay, such as an IAB node. In some example embodiments, the MT part may be referred to as wireless terminal 110 and perform tasks of wireless terminal 110.
[0029] Wireless network node 120 may be connected, directly or via at least one intermediate node, with core network 130 via interface 125. Core network 130 may be, in turn, coupled via interface 135 with another network (not shown in FIG. 1), via which connectivity to further networks may be obtained, for example via a worldwide interconnection network. Wireless network node 120 may be connected, directly or via at least one intermediate node, with core network 130 or with another core network.
[0030] The network may further comprise first vendor 141 and second vendor 142. First vendor 141 may be a vendor of wireless terminal 110 and second vendor 142 may be a vendor of wireless network node 120. First vendor 141 and / or second vendor 142 may be located outside of core network 130, as illustrated in FIG. l, and connected to core network 130. In some example embodiments, first vendor 141 and / or second vendor 142 may be connected to core network 130 via the worldwide interconnection network, possibly viainterface 135. In some example embodiments, first vendor 141 and / or second vendor 142 may be located in core network 130.
[0031] An ML model may be used for air interface 115. The ML model may be, or comprise, an Artificial Intelligence, Al, model as well. The ML model may be a two-sided model between a wireless device, such as wireless terminal 110, and another wireless device, such as wireless network node 120. The two-sided model may be a paired ML model over which joint inference is performed. Joint inference may be performed across the wireless device and said another wireless device. For example, a part of inference may performed by wireless terminal 110 first and then the remaining, another part may be performed by wireless network node 120, or vice versa. In some example embodiments, joint inference may refer to a procedure, wherein wireless terminal 110 may perform said part of the two- sided model (e.g., group of layers), i.e., the first partial of the inference process. Wireless network node 120 may then perform said another part of the two-sided model (e.g., group of layers), i.e., the final part of the inference process. In some example embodiments, said part of the two-sided model may be run at wireless terminal 110 and said another part may be run at wireless network node 120, or vice versa.
[0032] An example of the two-sided model may be Channel State Information, CSI, feedback compression with an autoencoder. At least in case of CSI, a part of the autoencoder may be an encoder and another part of the autoencoder may be a decoder. In case of autoencoders, the encoder part may receive, as an input, channel state information H. The total number of feedback parameters in H may be NtxNrx Nc. Where may be a number of transmit antennas at wireless network node 120, Nrmay be a number of receiver antennas at wireless terminal 110, and Ncmay be a number of subcarriers an Orthogonal Frequency Division Multiplexing, OFDM, system operates over. A neural network of the encoder part may compress H into a codeword S of smaller size. In some example embodiments, S may be 64 bits. A neural network of the decoder part may take S as input and recover the H matrix.
[0033] The two-sided ML model may be trained as a whole, or said part and said another part may be trained separately. For example, the autoencoder may be trained as a whole, or the encoder and decoder parts may be trained separately. In case of separate training, said part may be trained first, for example by wireless terminal 110, possibly with an unsupervised algorithm. After performing the training of said part, wireless terminal 110 may use the trained part to generate labelled data for supervised training of said another part,for example at wireless network node 120. In some example embodiments, said part may be referred to as a first part and said another part may be referred to as a second part of the two- sided ML model.
[0034] In some example embodiments, the two-sided model may be a model with an architecture comprised of blocks that have been distributed across several hardware platforms, which might not be mounted / located on the same wireless device. The two-sided model may require alignment between said part and said another part, running at different wireless devices. The wireless devices might not be co-located (e.g., wireless terminal 110 and wireless network node 120). One option to ensure the alignment of said part and said another part may be to train and update an autoencoder at one wireless device jointly and then transfer said part of the model (“encoder” or “decoder”) to another device. For example wireless terminal 110 may transmit the required “decoder” to wireless network node 120, or vice versa. If wireless terminal 110 and wireless network node 120 belong to different vendors, inferencing a model generated by another vendor may be a challenging task, e.g., due to differences in hardware designs of the vendors, such as ML accelerators. Extensive details may need to be specified for correct inference of the model at a wireless device of another vendor.
[0035] Furthermore, this implies that one vendor may need to understand the details of implementation of the other vendor, which may not be only infeasible but also impractical. Sharing of hardware and software platform details may be infeasible due to the wide variations in wireless devices. An alternative may be to train and / or update another part of the two-sided model using compatible datasets, which inherently requires that the updates are synchronized. Nevertheless, different vendors may have different requirements and capabilities in terms of tuning and transfer of the two-sided model and hence, it may be preferable that each vendor manages its models in an independent manner. In such a case, the risk is that said part and said another part of the two-sided model, i.e., the parts at the two sides, become incompatible to each other.
[0036] At least some example embodiments of the present disclosure therefore make it possible to keep said part and said another part of the two-sided model compatible. For example, the compatibility of said part and said another part of the two-sided model for 5G- Adv NR. air interface may be ensured in 5G networks, or for any other air interface in other communication networks. At least some example embodiments of the present disclosureallow wireless terminal 110 and wireless network node 120 to avoid direct model transfer and enable keeping the ML model lifecycle management implementation-specific.
[0037] In some example embodiments, a compatibility check of the two-sided model may be performed to confirm that said part and said another part of the two-sided model are well aligned. That is, performance degradation may be avoided such that the performance does not degrade below a target threshold, even if said part and said another part of the two- sided model would be independently managed by their respective vendors. Moreover, if said part and said another part of the two-sided model would become incompatible, it may be discovered and corrective actions taken to re-align the models.
[0038] In some example embodiments, evolution of the two-sided model may be tracked and said compatibility check triggered proactively after concluding that compatibility may be compromised. The CCD may be generated and exchanged after triggering said compatibility check by either of the wireless devices, or by both sides simultaneously, possibly after agreeing on a CCD sharing strategy.
[0039] FIGs. 2 - 4 illustrate signalling graphs in accordance with at least some example embodiments of the present disclosure. Even though wireless terminal 110 and wireless network node 120 are used as examples, operations of wireless terminal 110 and wireless network node 120 may be performed by any wireless devices. For example, wireless terminal 110 may perform the operations of wireless network node 120 and wireless network node 120 may perform the operations of wireless terminal 110. In some example embodiments, another wireless terminal may perform the operations of wireless network node 120. In some example embodiments, another wireless network node may perform the operations of wireless terminal 110.
[0040] In some example embodiments, wireless terminal 110 and wireless network node 120 may run the two-sided model for CSI compression. However, the procedures may be applied similarly for any other two-sided model between two wireless devices.
[0041] That is, example embodiments of the present disclosure may also be applied in use cases other than CSI compression two-sided model scenarios. For example, when performing predictions for cell and beam level mobility by wireless terminal 110 and wireless network node 120 simultaneously, the ML models in wireless terminal 110 and wireless network node 120 may be complementary to each other. Similarly, in a positioninguse case, when training a two-sided ML-based location estimator of wireless terminal 110, the ML model of wireless terminal 110 providing positioning measurements to wireless network node 120 may be trained independently from the ML model of wireless network node 120 computing the final location estimate. The ML model of wireless terminal 110 may use the same dataset identity as the ML model of wireless network node 120. A compatibility check of the two-sided model may be intended to confirm that parts of the two-sided model are well aligned, i.e., the performance does not degrade below a target threshold, even when the two models may be independently managed.
[0042] FIG. 2 illustrates a first signalling graph in accordance with at least some example embodiments. On the vertical axes are disposed, from the left to the right, first vendor 141, wireless terminal 110, wireless network node 120 and second vendor 142 of FIG. 1. Time advances from the top towards the bottom. More specifically, FIG. 2 illustrates a cross-validation procedure, without direct communication between first vendor 141 and second vendor 142.
[0043] The procedure may be used to verify that said part and another part of the two- sided model, residing at different communication ends, remain compatible with each other. Said part and another part of the two-sided model may have been initially synchronized and aligned to be compatible e.g., via model identity matching and / or joint training. Each of said part and said another part of the two-sided model may have been updated and / or fine-tuned autonomously by the respective vendor during their proprietary Lifecycle Management, LCM, procedure, without explicitly informing the peer vendor(s).
[0044] At step 202, wireless terminal 110 may receive a model update from first vendor 141, wherein first vendor 141 may be a vendor of wireless terminal 110. Wireless terminal 110 may for example receive the model update from a server of first vendor 141. The server of first vendor 141 might not be controlled or linked with the server of second vendor 142. Reception of the model update may be an operation concerning an ML functionality. The ML model functionality may be associated with the two-sided model between wireless terminal 110 and wireless network node 120. That is, wireless terminal 110 may perform, at step 204, at least one operation concerning the ML functionality. In addition, or alternatively, performing the at least one operation concerning the ML functionality may comprise at least one of running the two-sided model, pre-processing ofthe two-sided model, post-processing of the two-sided model or configuration of the two- sided model.
[0045] At step 204, wireless terminal 110 may assess whether compatibility check should be triggered. For example, wireless terminal 110 may determine to trigger said compatibility check based on at least one of a model update rate, total number of model updates, local update, performance degradation or a request received from wireless network node 120. That is, wireless terminal 110 may track how often the two-sided model is updated or if the total number of updates since the last compatibility check exceeds a certain threshold. Alternatively, or in addition, wireless terminal 110 may update model locally, trigger said compatibility check due to performance degradation or trigger said compatibility check due to a request from wireless network node 120. In some example embodiments, wireless terminal 110 may track the performance of the two-sided model, number of updates and trigger conditions of said compatibility check. For example, if rate is larger than a threshold and the number of total updates is larger than another threshold, wireless terminal 110 may trigger said compatibility check. Alternatively, or in addition, wireless terminal 110 may trigger said compatibility check when there is a change in the model input, e.g., some features cannot be extracted or do not match statistically with the past ones. In some example embodiments, wireless terminal 110 may trigger said compatibility check when there are changes in the radio configuration of wireless terminal 110 executing the model part.
[0046] Wireless terminal 110 may hence determine to trigger a compatibility check procedure concerning the two-sided model. If wireless terminal 110 determines that said compatibility check is to be performed, wireless terminal 110 may transmit after said determination, at step 206, a request to generate input data for a part of the machine learning two-sided model. The request may be transmitted to first vendor 141. Said input data may be synthetic input data. So if wireless terminal 110 determines that a compatibility check is due, it may request the generation of synthetic input data set(s) from its own vendor server. In some example embodiments, synthetic data may refer to a simulated or ray traced data, or a combination of these. Alternatively, or in addition, synthetic data may refer to not explicitly / directly measured or collected data collected by a device from the real environment it operates in.
[0047] At step 208, first vendor 141 may transmit said input data to wireless terminal 110. For example, a server of first vendor 141 may provide the generated synthetic input data to wireless terminal 110.
[0048] At step 210, wireless terminal 110 may generate the CCD using a part of the two-sided model. For example, wireless terminal 110 may use said input to generate a model output, and the pair (synthetic input, output) may form the CCD, to be transmitted to wireless network node 120. The exact number of input data samples to be used for compatibility check may vary, but at least one sample may be needed to perform a compatibility check. In some example embodiments, wireless terminal 110 may generate the CCD responsive to said determination at step 204, to trigger said compatibility check.
[0049] In some example embodiments, wireless terminal 110 may run inference of the encoder to generate an output. Wireless terminal 110 may then generate the CCD by incorporating said input and output to the CCD.
[0050] At step 212, wireless terminal 110 may transmit the CCD to wireless network node 120. In some example embodiments, real data may be used for said compatibility check. In such a case, wireless terminal 110 may provide a real encoder input, such as CSI, coupled with encoded output to wireless network node 120, instead of said synthetic data received from first vendor 141. The used real input data may be comprised in the generated CCD.
[0051] At step 214, wireless network node 120 may perform said compatibility check by inputting at least the CCD to said another part of the two-sided model. Wireless network node 120 may apply said output to said another part of the two-sided model and evaluate the result with said input received from wireless terminal 110. That is, wireless network node 120 may perform said compatibility check by comparing an output of the two-sided model to said input data. For example, wireless terminal 110 may run the remaining, decoder part of the two-sided model using the CCD. Then, wireless terminal 110 may check the compatibility and accuracy of the full encoder-decoder processing.
[0052] At step 216, wireless network node 120 may report the result of said compatibility check to second vendor 142, e.g., to a server of second vendor 142. If incompatibility is detected at step 214, wireless network node 120 may further indicate viasignalling to wireless terminal 110 (not shown in FIG. 2) that incompatibility has been detected and optionally include the recommended corrective actions.
[0053] In some example embodiments, the procedure may be initiated by the network, for example by wireless network node 120, Location Management Function, LMF, etc. In some example embodiments, one part of the two-sided model e.g., CSI encoder, may be used to generate the CCD.
[0054] FIG. 3 illustrates a second signalling graph accordance with at least some example embodiments. On the vertical axes are disposed, from the left to the right, first vendor 141, wireless terminal 110, wireless network node 120 and second vendor 142 of FIG. 1. Time advances from the top towards the bottom. More specifically, FIG. 3 illustrates a compatibility check, procedure with direct communication between first vendor 141 and second vendor 142, wherein input data may be generated by first vendor 141. Said input data may be synthetic input data.
[0055] Steps 302 - 306 in FIG. 3 may correspond to steps 202 - 206 in FIG. 2, respectively. At step 308, first vendor 141 may generate said input data. At step 310, first vendor 141 may transmit said input data to second vendor 142. Said input data may be hence transmitted directly from first vendor 141 to second vendor 142.
[0056] Step 312 may correspond to step 208 of FIG. 2. At step 314, wireless terminal 110 may generate the CCD. In the example of FIG. 3, the CCD may comprise output of the two-sided model, but not said input data. That is, wireless terminal 110 may generate the CCD without embedding said input data to it.
[0057] At step 316, wireless terminal 110 may transmit the CCD to wireless network node 120. The CCD may be provided to wireless network node 120 as normal feedback from wireless terminal, e.g. as a CSI report. In some example embodiments, wireless terminal 110 may indicate that a given feedback is used for the CCD via a Radio Resource Control, RRC, or Medium Access Control, MAC, indication. The indication may be a flag.
[0058] At step 318, wireless network node 120 may receive said input data from second vendor 142. Steps 320 and 322 in FIG. 3 may correspond to steps 214 and 216 in FIG. 2, respectively.
[0059] Alternatively, wireless network node 120 may trigger a specific report from wireless terminal 110 (not shown in FIG. 3) after receiving said input data from second vendor 142. In such a case, the order of steps 316 and 318 may be reversed.
[0060] FIG. 4 illustrates a third signalling graph in accordance with at least some example embodiments. On the vertical axes are disposed, from the left to the right, first vendor 141, wireless terminal 110, wireless network node 120 and second vendor 142 of FIG. 1. Time advances from the top towards the bottom. More specifically, FIG. 4 illustrates an example for a situation wherein said compatibility check may be simultaneously triggered by both, wireless terminal 110 and wireless network node 120. If such a situation arises, a handshake procedure on deciding who generates the CCD may be activated.
[0061] For example, wireless terminal 110 and wireless network node 120 may inform each other about triggering of said compatibility check. Wireless terminal 110 and wireless network node 120 may also inform each other about whether they prefer to generate the CCD themselves, or if they prefer the CCD to be generated by the peer. Such indications may be realized by sending two flags: CC-trigger = yes / no, CCD-generation = yes / no. For example, the wireless terminal 110 may transmit, to the wireless network node 120, information comprising the CC-trigger flag and / or CCD-generation flag. Similarly, the wireless network node 120 may transmit, to the wireless terminal 110, information comprising the CC-trigger flag and / or CCD-generation flag.
[0062] Steps 402 and 404 in FIG. 4 may correspond to steps 302 and 304 in FIG. 3, respectively. Wireless terminal 110 may then perform a handshake with wireless network node 120 to determine whether wireless terminal 110 or wireless network node 120 is to perform said compatibility check. At step 406, wireless terminal 110 may transmit to wireless network node 120 an indication or a request for said compatibility check. Wireless terminal 110 may for example transmit CC-trigger flag and CCD-generation flag (e.g. indicating: CC-trigger = yes, CCD-generation = no). At step 408, wireless network node 120 may respond by acknowledging the request. At step 410, wireless network node 120 may trigger generation of said input data and second vendor 142 may, at step 412, generate said input data. At step 414, second vendor 142 may then transmit said input data to wireless network node 120. At step 416, wireless network node 120 may further transmit said input data to wireless terminal 110. Steps 418 and 420 in FIG. 4 may correspond to steps 314 and 316 in FIG. 3, respectively. Steps 422 and 424 in FIG. 4 may correspond to steps 320 and322 in FIG. 3, respectively. In some example embodiments, whomever starts said compatibility check may need to obtain the CCD and share the CCD with the other device. The outcome of said compatibility check may trigger an action such as a model update at each part, model deactivation, etc.
[0063] In some example embodiments, wireless terminal 110 and wireless network node 120 may trigger said compatibility check or agree on the need for a more complete compatibility check, and may be available to generate the CCD as well. In such a case, first vendor 141 may generate first input data and second vendor 142 may generate second input data simultaneously. Thus, two CCDs may be obtained and both used at wireless network node 120 to check the compatibility of the two-sided model.
[0064] FIG. 5 illustrates an example apparatus capable of supporting at least some example embodiments. Illustrated is device 500, which may comprise, for example, wireless terminal 110 or wireless network node 120, or a control device configured to control the functioning thereof, possibly when installed therein. Comprised in device 500 is processor 510, which may comprise, for example, a single- or multi-core processor wherein a singlecore processor comprises one processing core and a multi-core processor comprises more than one processing core. Processor 510 may comprise, in general, a control device. Processor 510 may comprise more than one processor. Processor 510 may be a control device. A processing core may comprise, for example, a Cortex-A8 processing core manufactured by ARM Holdings or a Steamroller processing core produced by Advanced Micro Devices Corporation. Processor 510 may comprise at least one Qualcomm Snapdragon and / or Intel Atom processor. Processor 510 may comprise at least one application-specific integrated circuit, ASIC. Processor 510 may comprise at least one field- programmable gate array, FPGA. Processor 510 may be means for performing method steps in device 500. Processor 510 may be configured, at least in part by computer instructions, to perform actions.
[0065] A processor may comprise circuitry, or be constituted as circuitry or circuitries, the circuitry or circuitries being configured to perform phases of methods in accordance with example embodiments described herein. As used in this application, the term “circuitry” may refer to one or more or all of the following: (a) hardware-only circuit implementations, such as implementations in only analog and / or digital circuitry, and (b) combinations of hardware circuits and software, such as, as applicable: (i) a combination of analog and / or digitalhardware 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.
[0066] 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.
[0067] Device 500 may comprise memory 520. Memory 520 may comprise randomaccess memory and / or permanent memory. Memory 520 may comprise at least one RAM chip. Memory 520 may comprise solid-state, magnetic, optical and / or holographic memory, for example. Memory 520 may be at least in part accessible to processor 510. Memory 520 may be at least in part comprised in processor 510. Memory 520 may be means for storing information. Memory 520 may comprise computer instructions that processor 510 is configured to execute. When computer instructions configured to cause processor 510 to perform certain actions are stored in memory 520, and device 500 overall is configured to run under the direction of processor 510 using computer instructions from memory 520, processor 510 and / or its at least one processing core may be considered to be configured to perform said certain actions. Memory 520 may be at least in part comprised in processor 510. Memory 520 may be at least in part external to device 500 but accessible to device 500.
[0068] Device 500 may comprise a transmitter 530. Device 500 may comprise a receiver 540. Transmitter 530 and receiver 540 may be configured to transmit and receive, respectively, information in accordance with at least one cellular or non-cellular standard. Transmitter 530 may comprise more than one transmitter. Receiver 540 may comprise more than one receiver. Transmitter 530 and / or receiver 540 may be configured to operate in accordance with Global System for Mobile communication, GSM, Wideband Code DivisionMultiple Access, WCDMA, Long Term Evolution, LTE, and / or 5G / NR standards, for example.
[0069] Device 500 may comprise a Near-Field Communication, NFC, transceiver 550. NFC transceiver 550 may support at least one NFC technology, such as Bluetooth, Wibree or similar technologies.
[0070] Device 500 may comprise User Interface, UI, 560. UI 560 may comprise at least one of a display, a keyboard, a touchscreen, a vibrator arranged to signal to a user by causing device 500 to vibrate, a speaker and a microphone. A user may be able to operate device 500 via UI 560, for example to accept incoming telephone calls, to originate telephone calls or video calls, to browse the Internet, to manage digital files stored in memory 520 or on a cloud accessible via transmitter 530 and receiver 540, or via NFC transceiver 550, and / or to play games.
[0071] Device 500 may comprise or be arranged to accept a user identity module 570. User identity module 570 may comprise, for example, a Subscriber Identity Module, SIM, card installable in device 500. A user identity module 570 may comprise information identifying a subscription of a user of device 500. A user identity module 570 may comprise cryptographic information usable to verify the identity of a user of device 500 and / or to facilitate encryption of communicated information and billing of the user of device 500 for communication effected via device 500.
[0072] Processor 510 may be furnished with a transmitter arranged to output information from processor 510, via electrical leads internal to device 500, to other devices comprised in device 500. Such a transmitter may comprise a serial bus transmitter arranged to, for example, output information via at least one electrical lead to memory 520 for storage therein. Alternatively to a serial bus, the transmitter may comprise a parallel bus transmitter. Likewise processor 510 may comprise a receiver arranged to receive information in processor 510, via electrical leads internal to device 500, from other devices comprised in device 500. Such a receiver may comprise a serial bus receiver arranged to, for example, receive information via at least one electrical lead from receiver 540 for processing in processor 510. Alternatively to a serial bus, the receiver may comprise a parallel bus receiver.
[0073] Device 500 may comprise further devices not illustrated in FIG. 5. For example, where device 500 comprises a smartphone, it may comprise at least one digital camera. Some devices 500 may comprise a back-facing camera and a front-facing camera, wherein the back-facing camera may be intended for digital photography and the frontfacing camera for video telephony. Device 500 may comprise a fingerprint sensor arranged to authenticate, at least in part, a user of device 500. In some example embodiments, device 500 lacks at least one device described above. For example, some devices 500 may lack a NFC transceiver 550 and / or user identity module 570.
[0074] Processor 510, memory 520, transmitter 530, receiver 540, NFC transceiver 550, UI 560 and / or user identity module 570 may be interconnected by electrical leads internal to device 500 in a multitude of different ways. For example, each of the aforementioned devices may be separately connected to a master bus internal to device 500, to allow for the devices to exchange information. However, as the skilled person will appreciate, this is only one example and depending on the example embodiment, various ways of interconnecting at least two of the aforementioned devices may be selected without departing from the scope of the example embodiments.
[0075] FIG. 6 is a flow graph of a first method in accordance with at least some example embodiments. The steps of the first method may be performed by a wireless device, such as wireless terminal 110 or wireless network node 120, or by a control device configured to control the functioning thereof, possibly when installed therein.
[0076] The first method may comprise, at step 610, performing at least one operation concerning a machine learning functionality, wherein the machine learning functionality is associated with a two-sided model between the wireless device and another wireless device. The first method may further comprise, at step 620, generate, using a part of the two-sided model, compatibility check data concerning the machine learning functionality. Finally, the first method may comprise, at step 630, transmit said compatibility check data to said another wireless device.
[0077] FIG. 7 is a flow graph of a second method in accordance with at least some example embodiments. The steps of the second method may be performed by another wireless device, such as wireless terminal 110 or wireless network node 120, or by a control device configured to control the functioning thereof, possibly when installed therein.
[0078] The second method may comprise, at step 710, receiving, from another wireless device, compatibility check data concerning a machine learning functionality, wherein the machine learning functionality is associated with a two-sided model between the wireless device and said another wireless device, and the two-sided model comprises a part and another part. The second method may further comprise performing a compatibility check by inputting at least said compatibility check data to said another part of the two-sided model.
[0079] It is to be understood that the example embodiments disclosed are not limited to the particular structures, process steps, or materials disclosed herein, but are extended to equivalents thereof as would be recognized by those ordinarily skilled in the relevant arts. It should also be understood that terminology employed herein is used for the purpose of describing particular example embodiments only and is not intended to be limiting.
[0080] Reference throughout this specification to one example embodiment or an example embodiment means that a particular feature, structure, or characteristic described in connection with the example embodiment is included in at least one example embodiment. Thus, appearances of the phrases “in one example embodiment” or “in an example embodiment” in various places throughout this specification are not necessarily all referring to the same example embodiment. Where reference is made to a numerical value using a term such as, for example, about or substantially, the exact numerical value is also disclosed.
[0081] As used herein, a plurality of items, structural elements, compositional elements, and / or materials may be presented in a common list for convenience. However, these lists should be construed as though each member of the list is individually identified as a separate and unique member. Thus, no individual member of such list should be construed as a de facto equivalent of any other member of the same list solely based on their presentation in a common group without indications to the contrary. In addition, various example embodiments and examples may be referred to herein along with alternatives for the various components thereof. It is understood that such example embodiments, examples, and alternatives are not to be construed as de facto equivalents of one another, but are to be considered as separate and autonomous representations.
[0082] In an example embodiment, a wireless device, such as, for example, wireless terminal 110 or wireless network node 120, may comprise means for carrying out the example embodiments described above and any combination thereof.
[0083] In an example embodiment, a computer program may be configured to cause a method in accordance with the example embodiments described above and any combination thereof. In an example embodiment, a computer program product, embodied on a non-transitory computer readable medium, may be configured to control a processor to perform a process comprising the example embodiments described above and any combination thereof.
[0084] In an example embodiment, a wireless device, such as, for example, wireless terminal 110 or wireless network node 120, may comprise at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the wireless device at least to perform the example embodiments described above and any combination thereof.
[0085] Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more example embodiments. In the preceding description, numerous specific details are provided, such as examples of lengths, widths, shapes, etc., to provide a thorough understanding of example embodiments of the disclosure. One skilled in the relevant art will recognize, however, that the disclosure can be practiced without one or more of the specific details, or with other methods, components, materials, etc. In other instances, well-known structures, materials, or operations are not shown or described in detail to avoid obscuring aspects of the disclosure.
[0086] While the forgoing examples are illustrative of the principles of the example embodiments in one or more particular applications, it will be apparent to those of ordinary skill in the art that numerous modifications in form, usage and details of implementation can be made without the exercise of inventive faculty, and without departing from the principles and concepts of the disclosure. Accordingly, it is not intended that the disclosure be limited, except as by the claims set forth below.
[0087] The verbs “to comprise” and “to include” are used in this document as open limitations that neither exclude nor require the existence of also un-recited features. The features recited in depending claims are mutually freely combinable unless otherwise explicitly stated. Furthermore, it is to be understood that the use of "a" or "an", that is, a singular form, throughout this document does not exclude a plurality.INDUSTRIAL APPLICABILITY
[0088] At least some example embodiments find industrial application in wireless communication networks, for example in 3 GPP networks.ACRONYMS LIST3GPP 3rd Generation Partnership ProjectAl Artificial IntelligenceBS Base StationCCD Compatibility Check DataCSI Channel State InformationDU Distributed UnitGSM Global System for Mobile communicationIAB Integrated Access and BackhaulIEEE Institute of Electrical and Electronics Engineers loT Internet of ThingsLCM Lifecycle ManagementLMF Location Management FunctionLTE Long-Term EvolutionM2M Machine-to-MachineMAC Medium Access ControlML Machine LearningMT Mobile TerminationNFC Near-Field CommunicationNR New RadioOFDM Orthogonal Frequency Division MultiplexingRAT Radio Access TechnologyRRC Radio Resource ControlTRP Transmission and Reception PointUE User EquipmentUI User InterfaceWCDMA Wideband Code Division Multiple AccessWiMAX Worldwide Interoperability for Microwave AccessWLAN Wireless Local Area NetworkREFERENCE SIGNS LIST
Claims
CLAIMS:
1. A wireless device comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the wireless device at least to:- perform at least one operation concerning a machine learning functionality, wherein the machine learning functionality is associated with a two-sided model between the wireless device and another wireless device;- generate, using a part of the two-sided model, compatibility check data concerning the machine learning functionality; and- transmit said compatibility check data to said another wireless device.
2. The wireless device according to claim 1, wherein the wireless device and said another wireless device are devices of different vendors.
3. The wireless device according to claim 1 or claim 2, wherein the two-sided model is a paired model over which a joint inference is performed.
4. The wireless device according to any of the preceding claims, wherein said part of the two-sided model is independently managed from another part of the two-sided model.
5. The wireless device according to any of the preceding claims, wherein said part of the two-sided model is an encoder.
6. The wireless device according to any of the preceding claims, wherein the stored instructions cause, when executed by the at least one processor, the wireless device at least to:- transmit a request to generate input data for at least said part of the two-sided model;- receive said input data; and- generate at least a part of said compatibility check data by inputting said input data to said part of the two-sided model.
7. The wireless device according to any of the preceding claims, wherein the stored instructions cause, when executed by the at least one processor, the wireless device at least to:- determine to trigger the compatibility check concerning the two-sided model; and- generate, after said determination, said compatibility check data.
8. The wireless device according to claim 7, wherein the stored instructions cause, when executed by the at least one processor, the wireless device at least to:- perform, after said determination, a handshake with said another wireless device to determine whether the wireless device or said another wireless device is to perform the compatibility check.
9. The wireless device according to claim 7 or claim 8, wherein the stored instructions cause, when executed by the at least one processor, the wireless device at least to:- determine to trigger said compatibility check based on at least one of a model update rate, total number of model updates, local update, performance degradation or a request received from said another wireless device.
10. The wireless device according to any of the preceding claims, wherein the at least one operation concerning the machine learning functionality comprises at least one of running the two-sided model, pre-processing of the two-sided model, post-processing of the two- sided model or configuration of the two-sided model.
11. The wireless device according to any of the preceding claims, wherein the stored instructions cause, when executed by the at least one processor, the wireless device at least to:- receive feedback from the second node based on the transmitted compatibility check data.
12. A wireless device comprising: at least one processor; andat least one memory storing instructions that, when executed by the at least one processor, cause the wireless device at least to:- receive, from another wireless device, compatibility check data concerning a machine learning functionality, wherein the machine learning functionality is associated with a two-sided model between the wireless device and said another wireless device, and the two-sided model comprises a part and another part; and- perform a compatibility check by inputting at least said compatibility check data to said another part of the two-sided model.
13. The wireless device according to claim 12, wherein the stored instructions further cause, when executed by the at least one processor, the wireless device at least to:- receive input data for the two-sided model; and- perform the compatibility check by comparing an output of said another part of the two-sided model to said input data.
14. The wireless device according to claim 12 or claim 13, wherein the stored instructions further cause, when executed by the at least one processor, the wireless device at least to:- transmit a result of the compatibility check to a vendor of the wireless device.
15. The wireless device according to any of the preceding claims, wherein said part of the two-sided model is run at the wireless device and said another part is run at said another wireless device.
Citation Information
Patent Citations
Communication devices and methods for alignment of generalized ai / ML model
WO2024098181A1
