CC BY
27
CROSS-LAYER ADAPTATION PROCEDURE FOR MEAN OPINION SCORE PREDICTION
Vladimir A. Efimushkin,
Deputy Director General, FSUE ZNIIS, Moscow, Russia, [email protected]
Keywords: cross-layer adaptation, radio resource allocation, Long Term Evolution, QoS, LTE, CLA, MOS
Vadim V. Plakhov,
Maintenance Specialist of NPDB, FSUE ZNIIS, Moscow, Russia, [email protected]
The development of LTE and LTE Advanced networks has led to abruptly increasing of their number in the recent years, which naturally contributed to increase the traffic volume and the user density allocation in each LTE cell. These changes brought the problem of effective usage of LTE radio resources as well as ensuring of Quality of Service (QoS) and experience (QoE) to another level. One of the possible solutions, which could be used to solve this problem, is a cross-layer optimization approach for radio resource allocation, i.e. the usage of the Cross-Layer Adaptation (CLA) procedures. There are many different types and modifications of these procedures, but the base of each is the sharing of multiple level resources of the Open System Interconnection (OSI) model with taking into account the dynamic changes of the traffic types, the user's location in the LTE cell as well as the number of the concurrent data sessions.
The article considers the CLA procedure, which allows to produce the radio resource allocation of the LTE network thereby to provide the guaranteed prediction of Mean Opinion Score (MOS). This procedure is the modification of the Adaptive Modulation and Coding (AMC) procedure, which uses the quality indicators of LTE network such as Channel Quality Indicator (CQI) and compares these indicators with pre-com-puted indexes of channel coding and modulation scheme (MCS) to selects the optimally MCS index while fulfilling a certain Block Error Rate (BLER) constraint.
Considered procedure is described in present article by two types of an interconnection algorithm of functional entities, which located on the evolved Node Base Station (eNodeB) side of the LTE network as well as a brief description of algorithm steps and results of their comparing analysis.
Для цитирования:
Ефимушкин В.А., Плахов В.В. Процедура межуровневой адаптации для прогнозирования средней экспертной оценки пользователей // T-Comm: Телекоммуникации и транспорт. 2017. Том 11. №1. С. 57-61.
For citation:
Efimushkin V.A., Plakhov V.V. (2017). Cross-layer adaptation procedure for mean opinion score prediction. T-Comm, vol. 11, no.1, рр. 57-61.
COMMUNICATIONS
Adaptive modulation and coding procedure
CLA procedures are the radio resource allocation procedures, which use resources of multiple levels of the OSI model. These procedures allows to provide effective radio resource allocation between LTE network users in the real-time mode, regardless of users location within the LTE cell, i.e. the distance from user to eNodeB as well as the number of concurrent sessions of data transmission and types of the traffic transmitted over LTE network [1-3 j.
The usage of these procedures allows to set up the operation mode of eNodeB and defines how User Equipment (UE) could compute the MCS indexes that satisfy the BLER requirements and how to ensure the limit scores of QoS which are specified for each type of the traffic by corresponding values of Quality Class Indicator(QCI) [7,9].
Consider one of the most commonly used types of CLA procedures - the AMC procedure {hereafter - AMC). AMC techniques are often used as a basis for extensions and modifications of CLA procedures and expanding of their functionality. The examples of relevant modifications are described below.
AMC is used on the network with dynamically changing channels and is employed to improve channcl throughput. It's achieved by selection of the MCS indexes that arc optimal for current channel conditions and BLER constraint, which should be lower than 10% [5,7].
The CQI feedbacks that are periodically reported by the UE to the eNodeB in collaboration with the Channel State Information (CSI) feedbacks are used for optimal selection of the MCS index by pre-computcd mappings between CQI [6] and MCS [13] indexes which arc collected into look-up table on the eNodeB.
Consider the details of the MCS selection algorithm. In practical LTE network, the same MCS indexes must be assigned to all subcarriers assigned to each UE and for these goals the Sig-nal-to-Noisc Ratio (SNR) values of multiple subscribers are collected and translated into one-dimensional link quality metric (LQM) [7].
After that AMC exploits static mappings between these LQMs and the BLER performance of all MCS indexes to select the best MCS index in term of channel throughput. In other words, for each MCS a range of LQM values is associated via a look-up table, over which that MCS maximizes channcl throughput of LTE network. Functional architecture for realization of the AMC algorithm is shown on Figure 1 [7].
_Q£L
Receives' characteristics
MAC
Statistics
BLER Estimation
LQWBLER Mapping
Target BIER
i_L
MCS Index Search
M C S[5=i-
Transmitter
jaLl
CQI/MCS Mapping
Reallocation
However, the MCS selection could also be performed on the UE' side. For current channel conditions UE selects the MCS index which will prov ide a maximum attainable value of channel throughput by taking into account the BLER constraint described above. After the initial selection, the MCS index is compared with CQI indexes by internal look-up table for determination the CQI feedback that is transmitted to the eNodeB.
As previously mentioned, a large number of extensions and modifications are developed on the basis of the AMC which aim at improving the various aspects of the AMC functioning. For example, [6j presents the AMC extension - prediction procedure, which is built on the basis of machine learning, called reinforcement learning (RL). Relevant procedure called RL-AMC provides the possibility for determination the best MCS index by predicting the CQI indexes, automatically and in real-time mode [7,12].
AMC extensions for optimization the video traffic transmission, generated by video applications and services of LTE network are presented in [5, 8, 10J. There are also a number of modifications aimed at predicting the state of the channel based on the LTE network quality indicators, QoS, QoE and MOS parameters [11].
CLA procedure on the base of MOS prediction
First, consider in detail the AMC modification proposed in [11]. The aim of this modification is a prediction of MOS indexes during the data transmission over LTE network.
MOS prediction function operates in real-time mode. !t uses for comparing the pre-computed mappings between MOS indexes and Peak Signal-to-Noise Ratio (PSNR) calculated on the basis of certain procedure which uses the evaluation results of channel distortion introduced to the transmitting channel during the media traffic transmission from the media server. Procedure of PSNR calculating is also described in [11].
Functional architecture of this modification is shown on Figure 2. It includes the following entities: video application module (hereafter - VA), MOS prediction module (hereafter-MP), cross-layer resource allocation module (hereafter- RA), subscriber buffers, scheduler and transmitter [11],
The initial conditions of the basic procedure algorithm (hereafter-basic algorithm) arc the following:
• UEs are connected to eNodeB;
• The media server stores all possible versions of required data, i.e. in our case - all versions of the video file with different quality.
r* SliKtfUi,
SchEduter " Trasmitter
RA
Fig. 1, AMC functional architecture [7]
Fig. 2. Functional architecture of the basic procedure
7T>
Basic algorithm includes following steps:
1. UE; has sent the request to media server for obtaining the media data (in present example it's a video file), here and then /=1,2.....K\
2. The media server has initiated the transmission of video packets with different versions of the video file after obtaining the request from UE,. These packets will be stored in subscriber buffer (buffer LIE,) until receiving the information about selected version of the video file from the RA module;
3. Simultaneously with obtaining packets of the video file, the VA module evaluates the distortion, which was introduced into the channel between the media server and eNodeB, Obtained results of the evaluation are forwarded to MP module;
4. At the same lime, the distortion of the channel between LIE and eNodeB is evaluated on the base of the Packet Loss Ratio (PLR) which is obtained by eNodeB with CS1 information;
5. Based on the evaluation results of these channel distortions, the MP module initiates the PSNR computing procedure. After that, MP module uses the obtained result for MOS prediction by comparing mappings between PSNR and MOS, Obtained MOS is going to be the maximum achievable MOS for current channel conditions for all versions of the video file and then is used for selection of the optimal MCS index with the help of the internal look-up table which contains pre-computed mapping between MOS and MCS indexes;
6. The selected version of the video file as well as the optimal MCS index are forwarded to RA module for initiating the procedure of radio resource allocation (the same as AMC allocation technique);
7. The RA module forwards the information about the necessary range of resource blocks (RBs) allocated for each subscriber to the scheduler and the information about MCS to the transmitter;
8. Information about allocated RBs is forwarded to the transmitter for system reconfiguration.
Thereby the CLA procedure which was proposed in [1 Ij allows providing the guaranteed superior MOS index to the user by taking into account the requests from other users connected to eNodeB of LTE network in the same time [11],
CLA algorithm on the base of comparing QoS parameters
The modification of CLA procedure which is described above is an effective technique of radio resource allocation focused on the users as a whole. However, it has its own disadvantages such as necessity to forward the different versions of the video file which should be stored on subscriber buffers' side or additional calculations of PSNR and the channel distortions during multimedia transmission.
For elimination of these lacks, we propose an improved variant of this procedure which is based on evaluation of network parameters and their comparing with QoS parameters of LTE network defined by QCI indexes for prediction the best value of MOS.
The initial conditions of the algorithm of the improved procedure (hereafter - improved algorithm) are following
• UEs are connected to eNodeB;
• The media server stores all possible versions of the required video file.
Functional architecture of the improved procedure includes similar modules as the basic procedure, but the logic of their
interconnection is different. Functional architecture of the improved procedure is shown on Figure 3.
edia 5erv 1
Fig. 3. Functional architecture of the improved procedure
Proposed here improved algorithm includes following steps:
1. UE/ sent the request to media server for obtaining the video file containing the information about the file and characteristics of UE; sueh as screen parameters or UE, category which describe the maximum of supporting download and upload speeds and etc.
2. Based on these information media server selects a necessary version of the video file which fulfills the technical constraints of UE,. For example, if the screen resolution of UE, is small the media server will select the video file with the same resolution or with the most suitable resolution of the video. The look-up tables are used for this selection procedure on the media server side. Media server initiates the data transmission to the subscriber buffer after the work of selection procedure is finished;
3. Simultaneously with obtaining this information, the VA module evaluates the transmission delay and PLR into the channel between the media server and eNodeB. If these parameters fulfill the QCI constraints for this type of the traffic they are forwarded to the MP module. At the same time, the MP module receives the CQI from UE,;
4. The MP module initiates the MOS prediction procedure after obtaining the information about the transmission delay, PLR and CQI indexes. This procedure uses the comparing of obtaining parameters within pre-computed MOS and MCS indexes for obtaining the optimal MCS index for current channel conditions;
5. Obtained MCS index is forwarded to the RA module which initiates relevant procedure of radio resource allocation in LTE cell;
6. The RA module forwards the information about the necessary range of resource blocks (RBs) allocated for the subscriber to the scheduler and the information about MCS to the transmitter side;
7. Information about allocated RBs are forwarded to the transmitter for system reconfiguration
The improved procedure allows to provide the best MOS without additional evaluation of the channel distortions and PSNR value as well as additional transmission of different versions of the video file into the subscriber buffer on eNodeB side. The improved procedure is more interactive and responsive to
the changes in the channel state of LTE network because it uses the CQI - constantly updating parameter [1 ].
Comparing analysis
Considering the basic [11J and improved algorithms, which are targeted to MOS prediction we can note the following key differences:
1. The selection of the version of video file of the basic algorithm is carried to the eNodeB side and requires the subscriber buffer with relevant memory size because the frames of all versions of video file should be stored into buffer while the improved algorithm provides the same selection on the media server side on the initiated stage of data transmission. This difference allows to reduce the time of MOS prediction and radio resource allocation in LTE network as well as subscriber buffer size for each UE that are connected to eNodeB;
2. The MOS prediction in the basic algorithms is provided on the base of evaluating of channel distortions and PSNR values. In the improved algorithm this prediction uses only the comparing mappings between CQI and pre-computed MOS and MCS indexes. Also the improved algorithm applies the primary inspection prior MOS prediction procedure which allows to estimate the amount of transmission delay and PLR value to relate the boundary values of QCI.
From the above it can be concluded that the improved algorithm of CLA has greater adaptability to channel state changes, less reaction lime for such changes and produce a minimal impact to the operation mode of eNodeB.
Conclusion
CLA procedures aimed at QoS, QoE or MOS prediction allow to provide the best guaranteed values of these quality parameters and in the same time are the effective tools of radio resource allocation in the wireless network. They have their own advantages and disadvantages, impact to different network parameters and operate the multiple types of media traffic. Some of them provide the optimal selection of CQI index, but are not able to optimize the downlink radio resource or choose an optimal Precoding Matrix Indicator for LTE cell.
The full implementation of the pledged functioning algorithm of the proposed CLA procedure requires the upgrading of the transceiver equipment of media servers or eNodeB of LTE network.
To reduce these disadvantages need to produce further development of CLA procedures to improve the efficiency of radio resource allocation of LTE network, reduce response time with dynamically changing performance data protection and etc. [41,
1. ETSI (2016), TS 123 107 Version 13.0.0: Digital cellular telecommunications system (Phase 2+): Universal Mobile Telecommunications System (UMTS); LTE; Quality of Service (QoS) concept and architecture (3GPP TS 23.107 version 13.0.0 Release 13).
2. ETSI (2016), TS 136 213 Version 13.1.1; Evolved Universal Terrestrial Radio Access (E-UTRA); Physical layer procedures (3GPP TS 36.213 version 13.1.1 Release 13).
3. Long Term Evolution (LTE): an introduction (2007), White Paper, Ericsson.
4. Efimushkin, V. and Plakhov, V. (2016). Cross-layer adaptation procedure on the basis of MOS prediction. MTUC1 10th, Moscow, 16-17 March 2016, p. 372.
5. Efunushkina, T., Gabbouj, M. and Samuylov, K. (2014). Analytical model in discrete time for cross-layer video communication over LTE. Automatic Control and Computer Sciences, vol. 48, no. 6, pp. 345-357.
6. Brüggen, T. (2011). R&S TS8980 test system analyzes LTE quality indicators: CQI, I'M I and Rl, Wireless Technologies, Conformance test systems.
1. Bruno, R., Masarecchia, A, and Passarella, A. (2014). Robust Adaptive Modulation and Coding (AMC) Selection in LTE Systems using Reinforcement Learning. Vehicular Technology Conference, IEEE 80th, pp. 1-6.
8. Kwon, Y„ Suh, D., Kim, C., Hong, K, (2011). Application driven, AMC-based cross-layer optimization for video service over LTE, ELIRASIP. Journal of Wireless Communications and Networking, vol. 31.
9. Quality of Service in LTE (2014). Sandvine Intelligent Broadband Networks.
10. Rakesh, R., Balaaji, T., Amiya, N. (2012). Cross Layer Design for Efficient Video Streaming over LTE Using Scalable Video Coding. Network Protocols and Algorithms. Network Protocols and Algorithms, vol. 4, no. 4, pp. 101-125.
1 I, Yimg, J., Zhaoming, I,., Daabing, L,, Xiangming, W., Wei, Z., Wenmin, M. (2014). QoE-based cross-layer design for video applications over LTE. Multimedia Tools and Applications, vol.72, iss.2, pp. 1093-1113.
12. Wat kins, C., Dayan, P. (1992). Q-Learning. Machine Learning, vol.8, pp. 279-292.
13. MCS Index for 802.1 In and 802.1 lac Chart (2016), available at: http://www.wlanpros.com/mcs-index-802-l I n-802-1 lac-chart-3.html (Accessed 18 February 2016).
References
T-Comm Tom 11. #1-2017
ПРОЦЕДУРА МЕЖУРОВНЕВОЙ АДАПТАЦИИДЛЯ ПРОГНОЗИРОВАНИЯ СРЕДНЕЙ ЭКСПЕРТНОЙ ОЦЕНКИ ПОЛЬЗОВАТЕЛЕЙ
Ефимушкин Владимир Александрович, заместитель генерального директора по научной работе, ФГУП
ЦНИИС, к.ф.-м.н., доцент, Москва, Россия, [email protected] Плахов Вадим Вадимович, специалист по эксплуатации БДПН, ФГУП ЦНИИС, Москва, Россия, [email protected]
Аннотация
Развитие сетей LTE и LTE Advanced за последние годы привело к резкому, скачкообразному увеличению их количества, что естественным образом способствовало увеличению объемов передаваемого трафика и плотности расположения пользователей в каждой соте сети LTE. Данные изменения вывели на новый уровень проблему эффективного использования радио-ресурсов сети LTE для обеспечения гарантированного качества обслуживания и восприятия. Одним из вариантов решения данной проблемы является применение межуровневого подхода к оптимизации радио ресурсов, т.е. использование процедур межуровневой адаптации (Cross Layer Adaptation, CLA). Данные процедуры позволяют учитывать динамическое изменение типов передаваемого трафика, местоположение пользователей в соте сети LTE, а также количество одновременно поддерживаемых сеансов передачи данных. Существует множество различных вариантов и модификаций данных процедур, но в основе каждой из них лежит механизм совместного использования ресурсов нескольких уровней модели взаимодействия открытых систем. Рассматривается процедура CLA, позволяющая осуществлять перераспределение ресурсов сети LTE таким образом, чтобы обеспечить спрогнозированное значение средней экспертной оценки пользователей. Данная процедура является модификацией процедуры адаптивной модуляции и кодирования (Adaptive Modulation and Coding), которая для динамического распределения радио ресурсов использует информацию, содержащуюся в индикаторах качества сети LTE, таких как индикатор качества канала (Channel Quality Indicator), и оперирует заранее определенными значениями схем модуляции и кодирования (Modulation and Coding Scheme) для сопоставления данных, полученных с реальной сети, с уже имеющимися вариантами перераспределения радио ресурсов сети LTE.
Предлагается модифицированный вариант алгоритма взаимодействия функциональных элементов сетевой инфраструктуры, расположенных на стороне базовой станции (evolved Node Base station) сети LTE, а также описания этапов их взаимодействия и результатов сравнительного анализа вариантов алгоритма.
Ключевые слова: межуровневая адаптация, распределение радио ресурсов, показатели качества обслуживания, CLA, LTE, QoS, MOS.
Литература
1. ETSI (2016), TS 123 107 Version 13.0.0: Digital cellular telecommunications system (Phase 2+); Universal Mobile Telecommunications System (UMTS); LTE; Quality of Service (QoS) concept and architecture (3GPP TS 23.107 version 13.0.0 Release 13).
2. ETSI (2016), TS 136 213 Version 13.1.1: Evolved Universal Terrestrial Radio Access (E-UTRA); Physical layer procedures (3GPP TS 36.213 version 13.1.1 Release 13).
3. Long Term Evolution (LTE): an introduction (2007), White Paper, Ericsson.
4. Ефимушкин В.А., Плахов В.В. Процедура межуровневой адаптации на базе прогнозирования средней экспертной оценки пользователя // В сб.: Х Конференция МТУСИ. Москва, 16-17 марта 2016 г. C. 372.
5. Ефимушкина Т.В., Габуж М, Самуйлов К.Е. Аналитическая модель в дискретном времени для передачи видео с учетом межуровневой адаптации в сети LTE // Автоматика и вычислительная техника, 2014. Вып. 48. № 6. С. 33-45.
6. Bruggen T. R&S TS8980 test system analyzes LTE quality indicators: CQI, PMI and RI // Wireless Technologies, Conformance test systems, 2011.
7. Bruno R., Masarecchia A., Passarella A. Robust Adaptive Modulation and Coding (AMC) Selection in LTE Systems using Reinforcement Learning / Vehicular Technology Conference, IEEE 80th, 2014, pp. 1-6.
8. Kwon Y., Suh D., Kim C., Hong K. Application driven, AMC-based cross-layer optimization for video service over LTE // EURASIP Journal of Wireless Communications and Networking, 2011, vol. 31.
9. Quality of Service in LTE (2014), Sandvine Intelligent Broadband Networks.
10. Rakesh R., Balaaji T., Amiya N. Cross Layer Design for Efficient Video Streaming over LTE Using Scalable Video Coding. Network Protocols and
Algorithms // Network Protocols and Algorithms, 2012, vol. 4, no. 4, pp. 101-125.
11. YimgJ., Zhaoming L., Daabing L., Xiangming W., Wei, Z., Wenmin M. QoE-based cross-layer design for video applications over LTE // Multimedia Tools and Applications, 2014, vol. 72, iss. 2, pp. 1093-1113.
12. Watkins C, Dayan P. Q-Learning // Machine Learning, 1992, vol. 8, pp. 279-292.
13. MCS Index for 802.1 In and 802.1 lac Chart (2016), available at: http://www.wlanpros.com/mcs-index-802-l 1 n-802-1 lac-chart-3.html (Accessed 18 February 2016).
7T>
