WO2025015507A1

WO2025015507A1 - DEVICE AND METHOD FOR REAL-TIME PREDICTION OF QoE IN VOICE SERVICES

Info

Publication number: WO2025015507A1
Application number: PCT/CN2023/107780
Authority: WO
Inventors: Antonio De Domenico; Fadhel AYED; David LOPEZ-PEREZ; Nicola PIOVESAN; Wenjie Li; Xing Wei; Jun Chen
Original assignee: Huawei Technologies Co Ltd
Current assignee: Huawei Technologies Co Ltd
Priority date: 2023-07-17
Filing date: 2023-07-17
Publication date: 2025-01-23
Anticipated expiration: 2026-01-17

Abstract

The present disclosure relates to voice service quality prediction in a network. The disclosure proposes a network device for quality estimation and/or prediction of a voice service for a UE, wherein the network device is associated with a serving cell of the UE, and the network device being configured to: receive one or more measurement reports from the UE, wherein each measurement report comprises one or more radio measurements collected by the UE during the voice service in the serving cell; and obtain a first probability for the UE based on the one or more measurement reports, wherein the first probability indicates a probability of occurrence of a low-quality service event in the serving cell. This disclosure further proposes a UE for assisting the quality estimation and/or prediction of a voice service.

Description

DEVICE AND METHOD FOR REAL-TIME PREDICTION OF QoE IN VOICE SERVICES

TECHNICAL FIELD

The present disclosure relates to communication networks, particularly to voice services in communication networks. In order to improve the prediction of the user experience in voice services, the disclosure proposes a network device and a method for quality estimation and/or prediction of voice service for user equipment (UE) .

BACKGROUND

In today’s communication networks, while 5G has been deployed throughout the world, Voice over LTE (VoLTE) is continuing to grow. VoLTE subscriptions are predicted to reach 4.8 billion by the end of 2022, and Voice over New Radio (VoNR) is launched. The importance of voice services can be also understood by analyzing the methodology in the Umlaut network test where voice benchmarking accounts for 30%of the overall score. In this test, the audio quality of the transmitted speech samples is evaluated using the HD-voice capable and International Telecommunication Union (ITU) standardized so-called Perceptual Objective Listening Quality Analysis (POLQA) wideband algorithm. This model evaluates the distortion between the ground truth (reference) voice signal and the received (distorted) signal to estimate the Mean Opinion Score (MOS) of the network.

MOS is defined by ITU as the mean of the values on a predefined scale that subjects assign to their opinion of the performance of the telephone transmission system used either for conversation or for listening to spoken material. POLQA wideband algorithm is a very accurate model but not suitable for real-time user performance evaluation and or network optimization.

In addition, with the deployment of 5G worldwide, the number of active carriers in mobile networks has been constantly growing, and to obtain Uplink (UL) and Downlink (DL) rates and voice quality for each cell on each frequency, a full-packet test must be performed on each frequency, which leads to very large operational costs.

Radio resource management (RRM) optimization usually targets other Key Performance Indicators (KPIs) , such as coverage and load, as the relation between the MOS and easily measurable network parameters is complex, and the available MOS estimation process is slow.

Currently, no solution enables the real-time estimation (respectively, prediction) of the MOS experienced during a voice call before (respectively, after) handover (HO) . Therefore, there is a lack of mechanisms optimizing the MOS through RRM and HO mechanisms.

SUMMARY

In view of the above-mentioned limitations, the present disclosure aims to provide a solution to estimate and predict in real-time the MOS in voice services. In particular, an objective is to define a new metric that describes the occurrence of low MOS experienced by the user during voice services. Another objective is to provide a method to evaluate the occurrence of low MOS events through user measurement reports. Another objective is to design a new solution to control HO events such that the low MOS events are minimized.

These and other objectives are achieved by the solutions of this disclosure as provided in the independent claims. Advantageous implementations are further defined in the dependent claims.

A first aspect of the disclosure provides a network device for quality estimation and/or prediction of a voice service for a UE, wherein the network device is associated with a serving cell of the UE, and the network device being configured to: receive one or more measurement reports from the UE, wherein each measurement report comprises one or more radio measurements collected by the UE during the voice service in the serving cell; and obtain a first probability for the UE based on the one or more measurement reports, wherein the first probability indicates a probability of occurrence of a low-quality service event in the serving cell.

This disclosure enables real-time quality estimation and/or prediction of voice services for the UE. In particular, the occurrence of the low-quality service experienced by the UE during voice services can be predicted by the network device 300 based on collected real-time measurement reports.

In an implementation form of the first aspect, the one or more radio measurements comprise one or more of the following types of network performance information: reference signal received power (RSRP) , reference signal received quality (RSRQ) , and signal to interference plus noise ratio (SINR) .

In an implementation form of the first aspect, the network device is further configured to obtain one or more radio parameters, wherein the one or more radio parameters indicate information about a handover event, and/or information about a handover type; and obtain the first probability for the UE further based on the one or more radio parameters. Notably, such radio parameters may be included in Call detail records (CDR) .

In an implementation form of the first aspect, the low-quality service event occurs when a MOS of the voice service drops below a first threshold.

A particular embodiment of this disclosure proposes a parametrical solution to predict in real-time the MOS in voice services. Notably, the MOS is expressed as a single rational number, typically in the range 1 –5, where 1 is the lowest perceived quality, and 5 is the highest perceived quality. For instance, when the MOS of voice service is below 4, it is determined to be a low-quality service. This threshold may be specified by the network operator.

In an implementation form of the first aspect, the network device is further configured to estimate and/or predict the MOS of the voice service for the UE based on the one or more radio measurements and/or the one or more radio parameters.

Possible KPIs include RSRP, SINR, handover event (HO event) , and handover type (HO type) , which can be collected during voice services, will be used for estimating and/or predicting the MOS for the UE.

In an implementation form of the first aspect, the network device is further configured to obtain, from an external network device, a model for estimating and/or predicting the probability of occurrence of the low-quality service event; and/or generate the model for estimating and/or predicting the probability of occurrence of the low-quality service event based on a machine learning algorithm.

Notably, before estimating the first probability based on the one or more measurement reports, the network device may first acquire a model for estimating the first probability given the UE radio condition. Possibly, the network device may receive a trained model from other network devices, or train the model by itself.

In an implementation form of the first aspect, the network device is further configured to generate the model by training the machine learning algorithm using the one or more radio measurements and/or the one or more radio parameters.

Optionally, this model can be based on machine learning algorithms and it may enable mapping measured KPIs to the first probability.

In an implementation form of the first aspect, the network device is further configured to generate the model further based on the estimated/predicted MOS of the voice service.

In an implementation form of the first aspect, the machine learning algorithm is represented by:

wherein γ represents an SINR, HO represents a handover occurrence, IF_HO represents a handover type, MOS_th represents a MOS threshold, and α, β₀, β₁, β₂ are model intercept and coefficients depending on MOS_th.

In an implementation form of the first aspect, the network device is further configured to derive the first probability based on the model using the one or more radio measurements and/or the one or more radio parameters.

In an implementation form of the first aspect, the network device is further configured to provide the model to the UE for the UE to derive the first probability based on the model; and receive the first probability from the UE.

Optionally, the model may be shared by the serving cell, i.e., the network device, to the connected UEs through, e.g., the broadcast channel. Then, the first probability can be estimated directly at the UE side.

In an implementation form of the first aspect, the network device is further configured to determine whether to trigger a measurement report collection in one or more non-serving cells of the UE, based on the first probability.

Optionally, the network device may analyze the estimated first probability to decide whether to trigger the collection of measurement reports in different cells.

In an implementation form of the first aspect, the network device is further configured to trigger the measurement report collection in the one or more non-serving cells of the UE, when the first probability is above a second threshold; and obtain one or more further measurement reports from the UE, wherein each further measurement report comprises one or more radio measurements collected by the UE in the one or more non-serving cells.

For instance, the measurement report collection can be triggered for all the UEs for which P_MOS＞T_mrt, where 0＜T_mrt＜1. P_MOS is the first probability, and T_mrt is the threshold for triggering measurement reports.

In an implementation form of the first aspect, the network device is further configured to predict one or more second probabilities for the UE based on the one or more radio measurements comprised in the further measurement reports and/or the one or more radio parameters, wherein each second probability indicates a probability of occurrence of a low-quality service event in one of the one or more non-serving cells; or receive the one or more second probabilities from the UE.

Notably, each second probability indicates a probability of the occurrence of a low-quality service event in other non-serving cells. The second probability may be referred to as P’_MOS in this disclosure.

In an implementation form of the first aspect, the network device is further configured to determine whether to trigger a handover event for the UE based on the one or more second probabilities.

Optionally, the network device may further analyze the related P’_MOS to check whether the UE should be handover to a new cell.

In an implementation form of the first aspect, the network device is further configured to trigger the handover event for the UE, when at least one second probability is below a third threshold; or delay the handover event for the UE, when the one or more radio measurements from the serving cell are above a fourth threshold and the one or more second probabilities are not below the third threshold.

In particular, low values of P’_MOS, i.e., P’_MOS＜T_HO, can initiate an HO event.

In an implementation form of the first aspect, the network device is further configured to optimize one or more network parameters related to a handover policy based on the first probability and the second probability.

It can be understood that such a proposed procedure enables the network to improve the MOS during voice services, without additional information exchange between UEs and BSs. For instance, frequency-specific offsets, cell-specific offsets, or the HO thresholds and hysteresis, may be optimized, such that the probability that the UEs served in the network perceives a MOS lower than a threshold TMOS during voice services can be minimized.

In an implementation form of the first aspect, the network device is further configured to send a measurement request to the UE, wherein the measurement request indicates the UE to send the one or more measurement reports to the network device.

Optionally, before sending the measurement reports to the network device, the UE may first receive the measurement request from the network device.

A second aspect of the disclosure provides a UE for assisting quality estimation and/or prediction of a voice service of the UE, the UE is configured to send one or more measurement reports to a network device, wherein the network device is associated with a serving cell of the UE, and each measurement report comprises radio measurements collected by the UE during the voice service in the serving cell.

In an implementation form of the second aspect, the UE is further configured to receive a measurement report collection indication from the network device, wherein the measurement report collection indication indicates the UE to provide one or more further measurement reports to the network device, wherein each further measurement report comprises radio measurements collected by the UE in one or more non-serving cells; and send the one or more further measurement reports to the network device.

This disclosure further proposes a UE for assisting the real-time quality estimation and/or prediction of voice services. In particular, the occurrence of the low-quality service experienced by the UE during voice services can be predicted by the network device based on the real-time measurement reports collected by the UE.

In an implementation form of the second aspect, the UE is further configured to obtain a model for estimating a probability of occurrence of a low-quality service event from the network device.

As previously disclosed, the first probability P_MOS may be captured using a machine learning model. This model may be provided to the UE.

In an implementation form of the second aspect, the UE is further configured to derive a first probability based on the model, wherein the first probability indicates a probability of occurrence of the low-quality service event in the serving cell; and send the first probability to the network device.

Optionally, the first probability can be estimated directly at the UE side. For instance, the measurement report collection can be triggered for all the UEs when P_MOS＞T_mrt, where 0＜T_mrt＜1.

In an implementation form of the second aspect, the UE is further configured to derive one or more second probabilities based on the model, wherein the one or more second probabilities indicate a probability of occurrence of the low-quality service event in the one or more non-serving cells; and send the one or more second probabilities to the network device.

The second probability P’_MOS, which indicates a probability of occurrence of a low-quality service event in non-serving cells, can also be estimated directly at the UE side.

In an implementation form of the second aspect, the UE is further configured to receive a measurement request from the network device, wherein the measurement request indicates the UE to send the one or more measurement reports to the network device.

A third aspect of the disclosure provides a method performed by a network device for quality estimation and/or prediction of a voice service for a UE, wherein the network device is associated with a serving cell of the UE, and the method comprises: receiving one or more measurement reports from the UE, wherein each measurement report comprises radio measurements collected by the UE during the voice service in the serving cell; and obtaining a first probability for the UE based on the one or more measurement reports, wherein the first probability indicates a probability of occurrence of a low-quality service event in the serving cell.

Implementation forms of the method of the third aspect may correspond to the implementation forms of the network device of the first aspect described above. The method of the third aspect and its implementation forms achieve the same advantages and effects as described above for the network device of the first aspect and its implementation forms.

A fourth aspect of the disclosure provides a method performed by a UE for assisting quality estimation and/or prediction of a voice service of the UE, wherein the method comprises sending one or more measurement reports to a network device, wherein the network device is associated with a serving cell of the UE, and each measurement report comprises radio measurements collected by the UE during a voice service in the serving cell.

Implementation forms of the method of the fourth aspect may correspond to the implementation forms of the UE of the second aspect described above. The method of the fourth aspect and its implementation forms achieve the same advantages and effects as described above for the UE of the second aspect and its implementation forms.

A fifth aspect of the disclosure provides a computer program product comprising a program code for carrying out when implemented on a processor, the method according to the third aspect, any implementation forms of the third aspect, or the method according to the fourth aspect, or any implementation forms of the fourth aspect.

It has to be noted that all devices, elements, units, and means described in the present application could be implemented in software or hardware elements or any kind of combination thereof. All steps which are performed by the various entities described in the present application as well as the functionalities described to be performed by the various entities are intended to mean that the respective entity is adapted to or configured to perform the respective steps and functionalities. Even if, in the following description of specific embodiments, a specific functionality or step to be performed by external entities is not reflected in the description of a specific detailed element of that entity that performs that specific step or functionality, it should be clear for a skilled person that these methods and functionalities can be implemented in respective software or hardware elements or any kind of combination thereof.

BRIEF DESCRIPTION OF DRAWINGS

The above-described aspects and implementation forms of the present disclosure will be explained in the following description of specific embodiments in relation to the enclosed drawings, in which:

FIG. 1 shows an example of the MOS trends in a live network;

FIG. 2 shows exemplary methodologies for voice quality prediction;

FIG. 3 shows a network device according to an embodiment of the disclosure;

FIG. 4 shows a block diagram of the proposed method according to an embodiment of the disclosure;

FIG. 5 shows a UE according to an embodiment of the disclosure;

FIG. 6 shows a model construction phase according to an embodiment of the disclosure;

FIG. 7 shows a model construction phase according to an embodiment of the disclosure;

FIG. 8 shows a model construction phase according to an embodiment of the disclosure; and

FIG. 9 shows a model inference phase according to an embodiment of the disclosure;

FIG. 10 shows a model inference phase according to an embodiment of the disclosure;

FIG. 11 shows a method according to an embodiment of the disclosure; and

FIG. 12 shows a method according to an embodiment of the disclosure.

DETAILED DESCRIPTION OF EMBODIMENTS

Illustrative embodiments of a computer-implemented method, a wireless digital twin, and a corresponding computer program product for performance prediction in a wireless communication network are described with reference to the figures. Although this description provides a detailed example of possible implementations, it should be noted that the details are intended to be exemplary and in no way limit the scope of the application.

Moreover, an embodiment/example may refer to other embodiments/examples. For example, any description including but not limited to terminology, element, process, explanation, and/or technical advantage mentioned in one embodiment/example is applicative to the other embodiments/examples.

Future network optimization solutions focus on cost reductions and model precision enhancements leveraging available data to replace expensive and iterative processes based on human expertise with machine learning methods and new lightweight approaches to data gathering are needed.

In this context, using an accurate estimate of the MOS, it will be possible to optimize the network configuration and parameters, such that the user MOS during voice service can be enhanced. One important example is the tuning of mobility load balancing parameters, such as the frequency-specific offsets in the network, to distribute the users across the different cells using distinct carrier frequencies, to optimize the user experience. To achieve this goal, it is necessary not only to accurately estimate the MOS but more importantly to understand when the user-experienced MOS can drop below an acceptable level of quality, which can be limited/avoided by accurately controlling network parameters. An example of the MOS behavior, showing the deep spikes related to reduced voice quality experienced on the user side, is shown in FIG. 1.

Currently, multiple solutions have been proposed in the literature to estimate the MOS using radio parameters in the context of the different mobile technologies. FIG. 2 shows two exemplary methodologies for estimating voice service quality. For instance, Huawei introduced the voice quality index (VQI) model to estimate the MOS based on network-related KPIs. The general framework used to compute the VQI is described in FIG. 2 (a) . Similarly, Ericsson designed the Speech Quality Indicator (SQI) , which is also a parametrical solution using radio KPIs to estimate the MOS during voice services. The voice service quality estimator used in SQI is described in FIG. 2 (b) .

In these examples, parametrical solutions are proposed to estimate the MOS using network KPIs. However, they are intricate solutions, as they use speech codec, Block Error Rate (BLER) , Frame Error Rate (FER) , and other hard-to-model info. Therefore, these schemes cannot be used to estimate/predict the MOS in real time.

FIG. 3 shows a network device 300 for quality estimation and/or prediction of a voice service for a UE 310. In particular, the network device 300 is associated with a serving cell of the UE 310.

The network device 300 may comprise processing circuitry (not shown) configured to perform, conduct or initiate the various operations of the network device 300 described herein. The processing circuitry may comprise hardware and software. The hardware may comprise analog circuitry or digital circuitry, or both analog and digital circuitry. The digital circuitry may comprise components such as application-specific integrated circuits (ASICs) , field-programmable arrays (FPGAs) , digital signal processors (DSPs) , or multi-purpose processors. The network device 300 may further comprise memory circuitry, which stores one or more instruction (s) that can be executed by the processor or by the processing circuitry, in particular under the control of the software. For instance, the memory circuitry may comprise a non-transitory storage medium storing executable software code which, when executed by the processor or the processing circuitry, causes the various operations of the network device 300 to be performed. In one embodiment, the processing circuitry comprises one or more processors and a non-transitory memory connected to the one or more processors. The non-transitory memory may carry executable program code which, when executed by the one or more processors, causes the network device 300 to perform, conduct or initiate the operations or methods described herein.

The network device 300 is configured to receive one or more measurement reports 301 from the UE 310. In particular, each measurement report comprises one or more radio measurements collected by the UE 310 during the voice service in the serving cell. The network device 300 is further configured to obtain a first probability 302 for the UE 310 based on the one or more measurement reports 301, wherein the first probability 302 indicates a probability of occurrence of a low-quality service event in the serving cell.

This disclosure enables real-time quality estimation and/or prediction of voice services for the UE. In particular, the occurrence of the low-quality service experienced by the UE during voice services can be predicted by the network device 300 based on collected real-time measurement reports. Optionally, the one or more radio measurements comprised in the measurement reports may include one or more of the following types of network performance information: RSRP, RSRQ, and SINR.

A particular embodiment of this disclosure proposes a parametrical solution to predict in real-time the MOS in voice services. Notably, the MOS is expressed as a single rational number, typically in the range 1 –5, where 1 is the lowest perceived quality, and 5 is the highest perceived quality. Possibly, the low-quality service event occurs when a MOS of the voice service drops below a first threshold. Such the first threshold for determining the low-quality service event may be referred to as T_MOS in this disclosure. Possibly, this threshold may be specified by the network operator (e.g., T_MOS=4) .

An exemplary overall framework of the proposed solution is sketched in FIG. 4. Notably, the network device 300 may also be referred to as the base station (BS) in this disclosure. The first probability 302 may also be referred to as P_MOS in this disclosure.

According to an embodiment of this disclosure, before estimating the P_MOS based on the one or more measurement reports 301, the network device 300 may first acquire a model to capture P_MOS given the UE radio condition, for instance, This model can be based on machine learning algorithms and enable mapping measured KPIs to the first probability 302, i.e., P_MOS. Possible KPIs include RSRP, SINR, which can be collected during voice services. Possibly, the model may be constructed offline, locally, or in a centralized server. Details regarding the model construction will be discussed in a latter part of this application.

Optionally, the network device 300 may be further configured to obtain one or more radio parameters, wherein the one or more radio parameters indicate information about a handover event, and/or information about a handover type. Notably, such radio parameters may be included in CDR. The network device 300 may be further configured to obtain the first probability 302 for the UE 310 further based on the one or more radio parameters.

Notably, the information of the CDR may only be used for constructing the model. When using the model to estimate/predict the MOS, the handover event (HO event) , and handover type (HO type) information is not reported but known. It may be understood that when targeting to estimate the event of low-quality service from the serving cell, the handover event (HO event) , and the handover type (HO type) are set to false (i.e., 0) , for instance. When predicting the low-quality service in a different cell, the handover event (HO event) is set to true, and the handover type (HO type) is set according to the different cell (e.g., if it is in the same frequency or a different frequency of the serving cell) .

According to an embodiment of the disclosure, the network device 300 may be further configured to estimate and/or predict the MOS of the voice service for the UE 310 based on the one or more radio measurements and/or the one or more radio parameters.

During operations, based on the acquired model and real-time measurement reports, the network device 300 can estimate the P_MOS for each of its connected UEs.

The one or more measurement reports 301 usually include radio measurements such as RSRP or RSRQ. Then, the network device 300 may analyze the estimated P_MOS to trigger the collection of measurement reports in different cells.

Optionally, the network device 300 may be configured to determine whether to trigger a measurement report collection in one or more non-serving cells of the UE 310, based on the first probability 302. In particular, the network device 300 may be configured to trigger the measurement report collection in the one or more non-serving cells of the UE 310, when the first probability 302 is above a second threshold. Notably, the second threshold may be referred to as T_mrt (i.e., the threshold for measurement reports triggering) . For instance, the measurement report collection can be triggered for all the UEs for which P_MOS＞T_mrt, where 0＜T_mrt＜1.

Accordingly, the network device 300 may be configured to obtain one or more further measurement reports from the UE 310, wherein each further measurement report comprises one or more radio measurements collected by the UE 310 in the one or more non-serving cells.

According to an embodiment of the disclosure, the network device 300 may be further configured to predict one or more second probabilities for the UE 310 based on the one or more radio measurements comprised in the further measurement reports and/or the one or more radio parameters. Notably, each second probability indicates a probability of the occurrence of a low-quality service event in one of the one or more non-serving cells. The second probability may be referred to as P’_MOS in this disclosure.

That is, the network device 300 may further analyze the related P’_MOS to check if the UE should be handover to a new cell.

Alternatively, the P’_MOS may also be received by the network device 300 from the UE 310. In this example, P’_MOS can be estimated directly at the UE side.

According to an embodiment of the disclosure, the network device 300 may be further configured to determine whether to trigger a handover event for the UE 310 based on the one or more second probabilities.

In particular, low values of P’_MOS, i.e., P’_MOS＜T_HO, can initiate an HO event. Optionally, the network device 300 may be configured to trigger the handover event for the UE 310, when at least one second probability is below a third threshold. Notably, the third threshold may be referred to as T_HO in this disclosure.

Alternatively, when the one or more radio measurements from the serving cell are above a fourth threshold and the one or more second probabilities are not below the third threshold, the network device 300 may be configured to delay the handover event for the UE 310.

Optionally, the network device 300 may be configured to optimize one or more network parameters related to a handover policy based on the first probability 302 and the second probability.

It can be understood that such proposed procedure enables the network to improve the MOS during voice services, without additional information exchange between UEs and BSs. Details regarding network optimization will be discussed in the latter part of the application.

FIG. 5 shows a UE 310 for assisting in quality estimation and/or prediction of a voice service for the UE 310. In particular, the quality estimation and/or prediction is performed at a network device 300, which is associated with a serving cell of the UE 310.

The UE 310 may comprise processing circuitry (not shown) configured to perform, conduct or initiate the various operations of the UE 310 described herein. The processing circuitry may comprise hardware and software. The hardware may comprise analog circuitry or digital circuitry, or both analog and digital circuitry. The digital circuitry may comprise components such as application-specific integrated circuits (ASICs) , field-programmable arrays (FPGAs) , digital signal processors (DSPs) , or multi-purpose processors. The UE 310 may further comprise memory circuitry, which stores one or more instruction (s) that can be executed by the processor or by the processing circuitry, in particular under the control of the software. For instance, the memory circuitry may comprise a non-transitory storage medium storing executable software code which, when executed by the processor or the processing circuitry, causes the various operations of the UE 310 to be performed. In one embodiment, the processing circuitry comprises one or more processors and a non-transitory memory connected to the one or more processors. The non-transitory memory may carry executable program code which, when executed by the one or more processors, causes the UE 310 to perform, conduct or initiate the operations or methods described herein.

The UE 310 is configured to send one or more measurement reports 301 to the network device 300. In particular, each measurement report comprises one or more radio measurements collected by the UE 310 during the voice service in the serving cell.

This disclosure further proposes a UE for assisting the real-time quality estimation and/or prediction of voice services. In particular, the occurrence of the low-quality service experienced by the UE during voice services can be predicted by the network device 300 based on the real-time measurement reports collected by the UE 310.

According to an embodiment of the disclosure, the UE 310 may be further configured to receive a measurement report collection indication from the network device 300. In particular, the measurement report collection indication indicates the UE 310 to provide one or more further measurement reports to the network device 300. Each further measurement report comprises radio measurements collected by the UE 310 in one or more non-serving cells. The UE 310 may be futher configured to send the one or more further measurement reports to the network device 300.

According to an embodiment of the disclosure, the UE 310 may be further configured to obtain a model for estimating a probability of occurrence of a low-quality service event from the network device 300.

According to an embodiment of the disclosure, the UE 310 may be further configured to derive a first probability 302 based on the model, wherein the first probability 302 indicates a probability of occurrence of the low-quality service event in the serving cell. The UE 310 may accordingly be configured to send the first probability 302 to the network device 300.

As previously disclosed, the first probability 302 P_MOS may be captured using a machine learning model. For instance, the measurement report collection can be triggered for the UEs for which P_MOS＞T_mrt, where 0＜T_mrt＜1.

According to an embodiment of the disclosure, the UE 310 may be further configured to derive one or more second probabilities based on the model, wherein the one or more second probabilities indicate a probability of occurrence of the low-quality service event in the one or more non-serving cells; and send the one or more second probabilities to the network device 300.

Notably, the second probability P’_MOS indicates a probability of occurrence of a low-quality service event in one of the one or more non-serving cells. P’_MOS can be estimated directly at the UE side.

Optionally, before sending the one or more measurement reports 301 to the network device 300, the UE 310 may first receive a measurement request from the network device 300, wherein the measurement request indicates the UE 310 to send the one or more measurement reports 301 to the network device 300.

The proposed solution in this disclosure may include three phases: model construction phase, model inference phase, and MOS-based network optimization.

FIG. 6 shows an example of the model construction phase according to an embodiment of this disclosure, which relates to the process of building an ML model to estimate P_MOS.

To construct the model, each RAN node (i.e., the network device 300) first collects radio measurements such as RSRP, RSRQ, SINR, HO, HO type, etc, during voice services (steps 1 and 2 of FIG. 6) . The collected data may be imputed to existing solutions to estimate the MOS such as VQI or SQI (step 3 of FIG. 6) . Alternatively, available CDR data, which includes voice services parameters together with the estimated MOS (based on ITU POLQA) can be also used. Using the estimated MOS, and the collected radio parameters, each RAN node can construct a model to estimate and/or predict (after the HO event) P_MOS (step 4 of FIG. 6) .

That is, optionally, the network device 300 may be configured to generate the model for estimating and/or predicting the probability of occurrence of the low-quality service event based on a machine learning algorithm.

One solution to this goal is to use logistic regression algorithms such as:

where γ represents a SINR, HO represents a handover occurrence, IF_HO represents a handover type, MOS_th represents a MOS threshold, namely T_MOS, these are the selected inputs of the model. α, β₀, β₁, β₂ are model intercept and coefficients depending on MOS_th.

In another embodiment, the model construction phase is centralized across multiple RAN nodes. FIG. 7 shows the centralized model construction according to an embodiment of this disclosure. In this case, a centralized RAN node (the node on the right in FIG. 7) can collect model data from multiple nodes, train a global model, and share it with the nodes participating in the model construction phase.

That is, optionally, the network device 300 may be configured to obtain, from an external network device (e.g., a centralized RAN node) , a model for estimating and/or predicting the probability of occurrence of the low-quality service event.

Alternatively, each RAN node can construct a local model, and the central RAN node (the node on the right in FIG. 8) can construct a global model, e.g., through federated learning or distributed learning schemes. This example is shown in FIG. 8.

The developed model is then used to estimate and/or predict online P_MOS during network operations. Compared with FIG. 6, FIG. 9 further shows an example of online P_MOS inference. This online process is described in steps 5 and 6 of FIG. 9. Specifically, using a fresh measurement report (step 5) , the RAN node infers the P_MOS of its serving UEs leveraging the model produced offline in step 4, where HO = 0 (no handover) .

FIG. 9 shows a model inference phase which is done at the RAN node for each connected UE.In another embodiment, the model may be shared by the serving RAN node to the connected UEs through, e.g., the broadcast channel. Then, P_MOS can be estimated directly at the UE side.

That is, optionally, the network device 300 may be configured to provide the model to the UE 310 for the UE 310 to derive the first probability 302 based on the model; and receive the first probability 302 from the UE 310.

FIG. 10 shows an example where the UE infers P_MOS using the model received by the serving cell. Using the received model, the UE computes P_MOS and periodically reports it in the Physical Uplink Shared Channel (PUSCH) , together with a standard measurement report.

Similarly, each UE reports to the serving cell the P_MOS together with the measured link quality (RSRP/RSRQ) for each target cell.

In another embodiment, the UE computes P_MOS and reports in the PUSCH measurement only for those cells whose predicted P_MOS is smaller than T_mrt.

P_MOS can be used to drive HO functionalities and optimize user performance as described in FIG. 4. According to this disclosure, measurement report triggering is on P_MOS. Specifically, the network device 300 decides to command HO measurements through Radio Resource Control (RRC) configuration if P_MOS is smaller than a given threshold, T_mrt (see FIG. 4) .

If the user's P_MOS is below the above-mentioned threshold, the network device 300 demands its UE 310 to perform intra-frequency and inter-frequency measurements. In one embodiment, to trigger the measurements P_MOS is used in conjunction with other RAN parameters, e.g., RSRP, SINR, throughput, and cell load.

After collecting intra-frequency and inter-frequency measurements, the network device 300 can predict, using the previously computed model and the collected measurements, P’_MOS, the probability that user MOS is lower than the threshold MOS_th after inter-frequency HO (e.g., IF_HO = 1) or intrafrequency HO (e.g., IF_HO = 0) , setting e.g., HO = 1.

In one embodiment, the network device 300 selects the lowest between the two predictions (i.e., P’_MOS for inter-frequency HO and P’_MOS for intra-frequency HO) and initiates the related HO procedure if P’_MOS is lower than a given threshold T_HO, as shown in FIG. 4.

In one embodiment, the network device 300 selects first the prediction related to the prioritized HO (i.e., inter-frequency HO or intra-frequency HO) and initiates the related HO procedure if P’_MOS is lower than a given threshold T_HO. If the condition is not met, the condition with respect to the HO with low priority is checked. The priority between inter-frequency HO or intra-frequency HO can target rate or coverage optimization and it is usually decided as a network level policy.

In one embodiment, to trigger the HO, P’_MOS is used in conjunction with other RAN parameters, e.g., RSRP, SINR, throughput, and cell load.

In addition to the solution of how to estimate P_MOS at the serving cell and the solution of how to predict P’_MOS at the HO target cells, this disclosure further proposes to optimize offline the parameters of the HO policy, e. g,

● the frequency-specific offsets in the network, o_f,

● the cell-specific offsets o_k,j between the ith cell and jth cell in the network,

● the HO thresholds and hysteresis,such that the probability that the UEs served in the network perceives a MOS lower than a threshold T_MOS during voice services is minimized.

To achieve this goal, radio parameters are collected at the location where the policy is built, e.g., RSRPs, where j denotes the index of the jth cell in the network and i the index of the ith sample in the cell.

In one case, samples relate to different UEs; then i is the ith UE in the cell j. In another case, multiple UEs are clustered in a ‘grid’ , and the ith sample indicates the statistical representation (e.g., the mean) of measurements collected in the ith grid.

Using the collected radio parameters, HO models are constructed and used to statistically characterize, for each grid/UE i, the HO probabilities, P_i,j, to be served by the cell j. Similarly, For each grid/UE, based on the MOS model, the HO probabilities, and radio parameters, the P_MOS can be predicted, following different HO events.

Let us denote with γ_i the SINR at grid/UE i, with c_i the serving cell for grid/UE I, and with f_j the frequency operated at the cell j, the optimization problem to determine the parameters of the HO policy can be expressed as follows:

where HO and IF_HO are indicator functions that can be computed as

In other embodiments,

● the minmax function is used instead of the min operator.

● the optimization function includes a cell-level KPI such as the rate while the network average probability that the MOS is lower than MOS_th is constrained to be lower than a given requirement.

● the optimization function is a linear combination of cell-level KPIs and MOS

● this problem is solved using a stochastic optimization algorithm, such as reinforcement learning.

It may be worth further mentioning that the present disclosure may be used to solve the same objective in other types of networks such as WiFi, fixed network, and satellite networks. In different embodiments, the present idea is adopted in the aforementioned types of networks.

In another example, a UE application could run a similar model. However, this would require a dedicated app and would prevent the access network node to react online to changes in UE QoE.

FIG. 11 shows a method 1100 for quality estimation and/or prediction of a voice service for a UE 310, according to an embodiment of the disclosure. The method 1100 is performed by a network device, particularly the network device 300 shown in FIG. 3 or FIG. 5. The method 1100 comprises a step 1101 of receiving one or more measurement reports 301 from the UE 310. Each measurement report comprises one or more radio measurements collected by the UE 310 during the voice service in the serving cell. The method 1100 further comprises a step 1102 of obtaining a first probability 302 for the UE 310 based on the one or more measurement reports 301, wherein the first probability 302 indicates a probability of occurrence of a low-quality service event in the serving cell. Possibly, the UE 310 is the UE shown in FIG. 3 or FIG. 5.

FIG. 12 shows a method 1200 for assisting quality estimation and/or prediction of a voice service for a UE 310, according to an embodiment of the disclosure. The method 1200 is performed by the UE, particularly the UE 310 shown in FIG. 3 or FIG. 5. The method 1200 comprises a step 1201 of sending one or more measurement reports 301 to a network device 300. The network device 300 is associated with a serving cell of the UE 310. Each measurement report comprises one or more radio measurements collected by the UE 310 during the voice service in the serving cell. Possibly, the network device 300 is the network device 300 shown in FIG. 3 or FIG. 5.

To summarize, this disclosure enables real-time estimation of UE QoE in voice services based on radio parameters. This enables the network to estimate in real-time the user QoE. This can be used to drive network operations such as intra/inter-frequency measurement to support HO. Further, this disclosure enables real-time prediction of UE QoE in voice services after HO implementation based on radio parameters. In this case, the network can decide to implement a given HO not only according to the expected rate or coverage but also based on the predicted QoE. This disclosure provides a solution where UE QoE can be predicted directly at the UE using the model provided by the serving BS. This approach reduces the amount of signaling reported by the UEs thus avoiding rate reduction.

The present disclosure has been described in conjunction with various embodiments as examples as well as implementations. However, other variations can be understood and effected by those persons skilled in the art and practicing the claimed embodiments of the disclosure, from the studies of the drawings, this disclosure, and the independent claims. In the claims as well as in the description the word “comprising” does not exclude other elements or steps and the indefinite article “a” or “an” does not exclude a plurality. A single element or other unit may fulfill the functions of several entities or items recited in the claims. The mere fact that certain measures are recited in the mutually different dependent claims does not indicate that a combination of these measures cannot be used in an advantageous implementation.

Furthermore, any method according to embodiments of the disclosure may be implemented in a computer program, having code means, which when run by processing means causes the processing means to execute the steps of the method. The computer program is included in a computer-readable medium of a computer program product. The computer-readable medium may comprise essentially any memory, such as a ROM (Read-Only Memory) , a PROM (Programmable Read-Only Memory) , an EPROM (Erasable PROM) , a Flash memory, an EEPROM (Electrically Erasable PROM) , or a hard disk drive.

Moreover, it is realized by the skilled person that embodiments of the proposed network device 300, or the UE 310 and the corresponding computer program product, comprise the necessary communication capabilities in the form of e.g., functions, means, units, elements, etc., for performing the solution. Examples of other such means, units, elements, and functions are: processors, memory, buffers, control logic, encoders, decoders, rate matchers, de-rate matchers, mapping units, multipliers, decision units, selecting units, switches, interleavers, de-interleavers, modulators, demodulators, inputs, outputs, antennas, amplifiers, receiver units, transmitter units, DSPs, trellis-coded modulation (TCM) encoder, TCM decoder, power supply units, power feeders, communication interfaces, communication protocols, etc. which are suitably arranged together for performing the solution.

Especially, the processor (s) of the network device 300, or the UE 310 may comprise, e.g., one or more instances of a Central Processing Unit (CPU) , a processing unit, a processing circuit, a processor, an Application Specific Integrated Circuit (ASIC) , a microprocessor, or other processing logic that may interpret and execute instructions. The expression “processor” may thus represent a processing circuitry comprising a plurality of processing circuits, such as, e.g., any, some or all of the ones mentioned above. The processing circuitry may further perform data processing functions for inputting, outputting, and processing of data comprising data buffering and device control functions, such as call processing control, user interface control, or the like.

Claims

A network device (300) for quality estimation and/or prediction of a voice service for a user equipment, UE (310) , wherein the network device (300) is associated with a serving cell of the UE (310) , and the network device (300) being configured to:

receive one or more measurement reports (301) from the UE (310) , wherein each measurement report comprises one or more radio measurements collected by the UE (310) during the voice service in the serving cell; and

obtain a first probability (302) for the UE (310) based on the one or more measurement reports (301) , wherein the first probability (302) indicates a probability of occurrence of a low-quality service event in the serving cell.
The network device (300) according to claim 1, wherein the one or more radio measurements comprise one or more of the following types of network performance information: reference signal received power, reference signal received quality, and signal to interference plus noise ratio.
The network device (300) according to claim 1 or 2, configured to:

obtain one or more radio parameters, wherein the one or more radio parameters indicate information about a handover event, and/or information about a handover type; and

obtain the first probability (302) for the UE (310) further based on the one or more radio parameters.
The network device (300) according to claim 1 or 2, wherein the low-quality service event occurs when a mean opinion score, MOS, of the voice service drops below a first threshold.
The network device (300) according to claim 4 and claims 2 and 3, configured to:

estimate and/or predict the MOS of the voice service for the UE (310) based on the one or more radio measurements and/or the one or more radio parameters.
The network device (300) according to one of the claims 1 to 5, configured to:

obtain, from an external network device, a model for estimating and/or predicting the probability of occurrence of the low-quality service event; and/or

generate the model for estimating and/or predicting the probability of occurrence of the low-quality service event based on a machine learning algorithm.
The network device (300) according to claim 6 and claims 2 and 3, configured to:

generate the model by training the machine learning algorithm using the one or more radio measurements and/or the one or more radio parameters.
The network device (300) according to claim 7, configured to:

generate the model further based on the estimated/predicted MOS of the voice service.
The network device (300) according to claim 7 or 8, wherein the machine learning algorithm is represented by:

wherein γ represents a signal to interference plus noise ratio, HO represents a handover occurrence, IF_HO represents a handover type, MOS_th represents a MOS threshold, and α, β₀, β₁, β₂ are model intercept and coefficients depending on MOS_th.
The network device (300) according to one of the claims 6 to 9, configured to:

derive the first probability (302) based on the model using the one or more radio measurements and/or the one or more radio parameters.
The network device (300) according to one of the claims 6 to 10, configured to:

provide the model to the UE (310) for the UE (310) to derive the first probability (302) based on the model; and receive the first probability (302) from the UE (310) .
The network device (300) according to one of the claims 1 to 11, configured to:

determine whether to trigger a measurement report collection in one or more non-serving cells of the UE (310) , based on the first probability (302) .
The network device (300) according to claim 12, configured to:

trigger the measurement report collection in the one or more non-serving cells of the UE (310) , when the first probability (302) is above a second threshold; and

obtain one or more further measurement reports from the UE (310) , wherein each further measurement report comprises one or more radio measurements collected by the UE (310) in the one or more non-serving cells.
The network device (300) according to claim 13 and claim 3, configured to:

predict one or more second probabilities for the UE (310) based on the one or more radio measurements comprised in the further measurement reports and/or the one or more radio parameters, wherein each second probability indicates a probability of occurrence of a low-quality service event in one of the one or more non-serving cells; or

receive the one or more second probabilities from the UE (310) .
The network device (300) according to claim 14, configured to:

determine whether to trigger a handover event for the UE (310) based on the one or more second probabilities.
The network device (300) according to claim 15, configured to:

trigger the handover event for the UE (310) , when at least one second probability is below a third threshold; or

delay the handover event for the UE (310) , when the one or more radio measurements from the serving cell are above a fourth threshold and the one or more second probabilities are not below the third threshold.
The network device (300) according to one of the claims 1 to 16, configured to:

optimize one or more network parameters related to a handover policy based on the first probability (302) and the second probability.
The network device (300) according to one of the claims 1 to 17, configured to:

send a measurement request to the UE (310) , wherein the measurement request indicates the UE (310) to send the one or more measurement reports (301) to the network device (300) .
A user equipment, UE (310) , for assisting quality estimation and/or prediction of a voice service of the UE (310) , the UE (310) being configured to:

send one or more measurement reports (301) to a network device (300) , wherein the network device (300) is associated with a serving cell of the UE (310) , and each measurement report comprises radio measurements collected by the UE (310) during the voice service in the serving cell.
The UE (310) according to claim 19, configured to:

receive a measurement report collection indication from the network device (300) , wherein the measurement report collection indication indicates the UE (310) to provide one or more further measurement reports to the network device (300) , wherein each further measurement report comprises radio measurements collected by the UE (310) in one or more non-serving cells; and

send the one or more further measurement reports to the network device (300) .
The UE (310) according to claim 19 or 20, configured to:

obtain a model for estimating a probability of occurrence of a low-quality service event from the network device (300) .
The UE (310) according to claim 21, configured to:

derive a first probability (302) based on the model, wherein the first probability (302) indicates a probability of occurrence of the low-quality service event in the serving cell; and

send the first probability (302) to the network device (300) .
The UE (310) according to claims 21 or 22, configured to:

derive one or more second probabilities based on the model, wherein the one or more second probabilites indicate a probability of occurrence of the low-quality service event in the one or more non-serving cells; and

send the one or more second probabilities to the network device (300) .
The UE (310) according to one of the claims 19 to 23, configured to:

receive a measurement request from the network device (300) , wherein the measurement request indicates the UE (310) to send the one or more measurement reports (301) to the network device (300) .
A method (1100) performed by a network device (300) for quality estimation and/or prediction of a voice service for a user equipment, UE (310) , wherein the network device (300) is associated with a serving cell of the UE (310) , and the method comprises:

receiving (1101) one or more measurement reports (301) from the UE (310) , wherein each measurement report comprises radio measurements collected by the UE (310) during the voice service in the serving cell; and

obtaining (1102) a first probability (302) for the UE (310) based on the one or more measurement reports (301) , wherein the first probability (302) indicates a probability of occurrence of a low-quality service event in the serving cell.
A method (1200) performed by a user equipment, UE (310) , for assisting quality estimation and/or prediction of a voice service of the UE (310) , and wherein the method comprises:

sending (1201) one or more measurement reports (301) to a network device (300) , wherein the network device (300) is associated with a serving cell of the UE (310) , and each measurement report comprises radio measurements collected by the UE (310) during a voice service in the serving cell.
A computer program product comprising a program code for carrying out, when implemented on a processor, the method (1100, 1200) according to claim 25 or 26.