Network entity and user equipment for transmission rate control

Abstract

A network entity for a mobile telecommunication system, the network entity comprising circuitry configured to perform a transmission rate control of a data transmission according to a transmission control protocol, wherein the transmission rate control is performed based on an output of a machine learning algorithm comprising a congestion prediction of the data transmission.

Description

Technical Field

[0001] This disclosure generally relates to a network entity and user equipment for a mobile telecommunications system. Background Technology

[0002] Several generations of mobile telecommunications systems are known, such as the third generation (“3G”) based on the International Mobile Telecommunications-2000 (IMT-2000) specification, the fourth generation (“4G”) providing the capabilities defined in the International Mobile Telecommunications-Advanced Standard (IMT-Advanced Standard), and the fifth generation (“5G”) currently under development and likely to be put into practice in 2020.

[0003] One candidate to meet the 5G requirements is the so-called Long Term Evolution (“LTE”), a wireless communication technology that allows mobile phones and data terminals to conduct high-speed data communication and is already used in 4G mobile telecommunications systems. Other candidates to meet the 5G requirements are called New Radio (NR) access technology systems. NR can be based on LTE technology, just as some aspects of LTE are based on previous generations of mobile communication technology.

[0004] LTE is based on GSM / EDGE (“Global System for Mobile Communications” / “Enhanced Data Rate Evolution of GSM”, also known as EGPRS) second-generation (“2G”) network technology and UMTS / HSPA (“Universal Mobile Telecommunications System” / “High-Speed ​​Packet Access”) third-generation (“3G”) network technology.

[0005] LTE is standardized under the control of 3GPP (“3rd Generation Partnership Project”), and there exists a subsequent LTE-A (Advanced LTE) that allows higher data rates than basic LTE and is also standardized under the control of 3GPP.

[0006] In the future, 3GPP plans to further develop LTE-A to make it able to meet the technical requirements of 5G.

[0007] Since 5G systems may be based on either LTE-A or NR, it is assumed that the specific requirements of 5G technology will be largely handled by the features and methods already defined in the LTE-A and NR standard documents.

[0008] Furthermore, Transmission Control Protocol (“TCP”) is a very common protocol on the Internet, and many applications supported in 5G networks will continue to use Transmission Control Protocol.

[0009] As is well known, the rate at which data enters a network according to a transmission control protocol is controlled by several mechanisms, such as the slow start mechanism. In the slow start mechanism, the transmission rate may decrease significantly after network congestion is detected, and may only begin to increase slowly thereafter. This is perceptible to users, for example, in highly user-centric wireless services (e.g., virtual reality), where the gap between the end user and network functionality is minimal.

[0010] Although there are technologies that control the transmission rate of data transmission according to transmission control protocols, there is often a desire to improve existing technologies. Summary of the Invention

[0011] According to a first aspect, this disclosure provides a network entity for a mobile telecommunications system, including circuitry configured to perform transmission rate control of data transmission according to a transmission control protocol, wherein the transmission rate control is performed based on the output of a machine learning algorithm that includes congestion prediction of data transmission.

[0012] According to a second aspect, this disclosure provides a user equipment for a mobile telecommunications system, including circuitry configured to use a Transmission Control Protocol-based service and receive from a network entity a first MAC control element including a recommended bit rate based on the output of a machine learning algorithm, the output of which includes congestion prediction of data transmission according to the Transmission Control Protocol, and the circuitry adjusting the transmission rate of data transmission according to the Transmission Control Protocol in response to and based on the received first MAC control element.

[0013] According to a third aspect, this disclosure provides a user equipment for a mobile telecommunications system, including circuitry configured to coordinate activities at different layers, whereby a modem included in the user equipment obtains information about a transmission control protocol header.

[0014] According to a fourth aspect, this disclosure provides a user equipment for a mobile telecommunications system, including circuitry configured to transmit a buffer status report to a network entity, the report indicating a buffer size smaller than its actual buffer size.

[0015] Further aspects are set forth in the dependent claims, the following description, and the accompanying drawings. Attached Figure Description

[0016] The embodiments are explained by way of example with reference to the accompanying drawings, wherein:

[0017] Figure 1 Two embodiments of a radio access network are illustrated schematically;

[0018] Figure 2The slow start mechanism for data transmission rate according to the transmission control protocol is shown;

[0019] Figure 3 A block diagram illustrating an embodiment of a recurrent neural network in the training phase is shown.

[0020] Figure 4 A block diagram illustrating an embodiment of a recurrent neural network in the inference phase is shown;

[0021] Figure 5 A graph showing the time evolution of the transmission rate of data transmission controlled by a network entity according to the transmission control protocol is presented.

[0022] Figure 6 A state diagram is shown for a first embodiment of transmission rate control of data transmission controlled by a network entity according to a transmission control protocol.

[0023] Figure 7 A state diagram is shown for a second embodiment of data transmission rate control controlled by a network entity according to a transmission control protocol;

[0024] Figure 8 A state diagram is shown for a third embodiment of data transmission rate control controlled by a network entity according to a transmission control protocol.

[0025] Figure 9 A state diagram is shown for a fourth embodiment of data transmission rate control controlled by a network entity according to a transmission control protocol.

[0026] Figure 10 A block diagram of user equipment and network entities is shown;

[0027] Figure 11 A block diagram is shown that can be used to implement a multipurpose computer for user equipment or network entities;

[0028] Figure 12 A block diagram illustrating an embodiment of the user equipment is shown; and

[0029] Figure 13 The diagram illustrates the state of the buffer status report being transmitted from the user equipment to the network entity. Detailed Implementation

[0030] Provide a reference Figure 2 Before a detailed description of the embodiments, a general explanation will be given.

[0031] As mentioned at the beginning, several generations of mobile telecommunications systems are generally known, such as the third generation (“3G”) based on the International Mobile Telecommunications-2000 (IMT-2000) specification, the fourth generation (“4G”) providing the capabilities defined in the International Mobile Telecommunications-Advanced Standard (IMT-Advanced Standard), and the fifth generation (“5G”) currently under development and likely to be put into practice this year.

[0032] One candidate system that meets 5G requirements is called a New Radio (“NR”) access technology system. In some embodiments, some aspects of NR may be based on LTE technology, just as some aspects of LTE are based on previous generations of mobile communication technologies.

[0033] As an example of mobile telecommunications systems, in Figure 1 A typical embodiment of the NR radio access network RAN1a is shown in Figure A. RAN1a has macro cells 2 established by LTE eNodeB 3 and NR cells 4 established by NR eNodeB 5 (also known as gNB (next-generation eNodeB)).

[0034] UE 6 can communicate with LTE eNodeB 3, and can also communicate with NR eNodeB 5 as long as it is within NR cell 4. This embodiment illustrates an NR EN-DC (“E-UTRA-NR Dual Connectivity”) deployment for Telecommunication System 1a.

[0035] In some embodiments, the mobile telecommunications system is an NR-independent system. Typically, data collection at different nodes is known, and the NWDAF (“Network Data Analysis Function”) entity is known. In some embodiments, the NWDAF entity may collect data from, for example, the 5G core network, the 5G RAN, and UE measurements collected in the RAN. Furthermore, in some embodiments, it is specified that data may be collected from entities outside the 3GPP system, i.e., application servers may share data with the NWDAF entity. In some embodiments, these provisions exist within the network.

[0036] therefore, Figure 1 Figure B illustrates another typical embodiment of the NR radio access network RAN ​​in mobile telecommunications system 1b. The RAN has NR cells 4 established by NR eNodeB 5 (gNB).

[0037] UE 6 can communicate with gNB 5, which is connected to the 5G core network (“5GC”) 8. NWDAF entity 9 collects data from 5GC 8, gNB 5, and application server 42 outside of mobile telecommunications system 1b.

[0038] As mentioned at the beginning, Transmission Control Protocol (“TCP”) is a very common protocol on the Internet, and many applications supported in 5G networks will continue to use TCP. The rate at which data enters the network according to TCP is controlled by several mechanisms, such as the slow-start mechanism.

[0039] In some embodiments, a TCP window is configured in both the transmitter and receiver, and a sliding window is used. The TCP window should be small if the network is congested, and typically large if the network is error-free and large bandwidth can be allocated. Other protocols, such as Packet Data Convergence Protocol (“PDCP”) or Radio Link Control (“RLC”) window operations, can be aligned with the TCP window configuration. Both TCP and RLC support sliding window mechanisms. However, radio conditions can change dynamically and may lead to buffer overflows, resulting in a TCP slow start mechanism. This can be identifiable by the user, as mentioned at the beginning, and can be inconvenient for highly user-centric wireless services (e.g., virtual reality), where the gap between the end user and network functions is minimal.

[0040] One approach to addressing this problem, described in the article (“TCP-Aware Scheduling in LTE Networks”, Shojaedin et al.), is to frequently allocate resources to user equipment (“UEs”) with small buffer sizes. However, it has been recognized that this approach can lead to resource waste due to inappropriate resource allocation sizes, or for UEs with larger buffer sizes, the allocated resources being underutilized at the expense of resource scarcity.

[0041] Furthermore, it has been recognized that artificial intelligence (“AI”) and / or machine learning (“ML”) are powerful tools for learning, analyzing, and predicting complex network scenarios; therefore, in some embodiments, machine learning is integrated with wireless communication. In some embodiments, the application of ML and / or AI in wireless communication (i.e., mobile telecommunications systems) is categorized as follows:

[0042] First, the application of ML in wireless systems is to enhance situational awareness and overall network operation by using intelligent and predictive data analytics, such as fault monitoring and user tracking on wireless networks.

[0043] Secondly, in addition to its powerful intelligent and predictive data analytics capabilities, ML is also used as a major driver for intelligent and data-driven wireless network optimization to address a variety of issues, from cell association and radio access technology selection to frequency allocation, spectrum management, power control, and intelligent beamforming.

[0044] Third, in addition to its system-level functions, ML plays a key role in the physical layer of wireless networks, for example, in coding and modulation design, and in the transmitter and receiver stages of general communication systems.

[0045] Fourth, rapidly deploy highly user-centric wireless services, such as VR, where the gap between end users and network functions is minimal, and ML helps wireless networks track and adapt to human user behavior.

[0046] It has also been recognized that, in some embodiments, ML and / or AI methods can be used in scheduler implementations, for example, in base stations such as gNBs. In some embodiments, ML and / or AI methods are used in the scheduler within the network entity to predict network congestion and / or the radio conditions of the UE.

[0047] Furthermore, it has been recognized that, in some embodiments, ML and / or AI techniques can be extended to address the TCP window stalling problem, and that this can be further extended to Fast User Datagram Protocol Internet Connection (“QUIC”) if the window size is known for, for example, the gNB or if the scheduler can predict the window size.

[0048] Therefore, some embodiments relate to network entities for mobile telecommunications systems, including circuitry configured to perform transmission rate control for data transmission according to a transmission control protocol, wherein the transmission rate control is performed based on the output of a machine learning algorithm that includes congestion prediction of data transmission.

[0049] A network entity can be a base station that is part of a mobile telecommunications system, such as an eNodeB, NR gNB, etc., which can be based on UMTS, LTE, LTE-A, or NR, 5G systems, etc. The entity can also be any other entity within the mobile telecommunications system and can be located anywhere within that system.

[0050] The circuit may include at least one of a processor, microprocessor, dedicated circuit, memory, storage device, radio interface, wireless interface, network interface, etc., for example, typical electronic components included in a base station (e.g., eNodeB, NR gNB, user equipment, etc.). It may include interfaces such as mobile telecommunications system interfaces adapted to provide communication to and / or to a mobile telecommunications system. It may also include wireless interfaces, such as wireless LAN interfaces, Bluetooth interfaces, etc.

[0051] As is well known, data transmission according to the Transmission Control Protocol (TCP) is packet-based, and the transmission rate typically depends on the window size and round-trip time (RTT). In some embodiments, the window size is the maximum number of bytes that the source can send in a set of packets for which it has not yet received an acknowledgment. The destination computer sends an acknowledgment back for each correctly received packet. In some embodiments, the RTT is the time from transmitting a packet until the source receives its acknowledgment.

[0052] Therefore, the transmission rate can be controlled by adjusting the window size, the bit rate in wireless transmission, or by affecting the round-trip time, which may depend on connection quality, network congestion, network scheduling, busy servers or base stations, etc.

[0053] Machine learning algorithms can be, or can include, neural networks, decision trees, support vector machines, etc., that generate outputs. The scheduler in the network entity uses this output to perform transmission rate control; that is, it determines whether the network entity needs to perform transmission rate control and takes the necessary actions accordingly. ML algorithms can be trained using supervised, unsupervised, reinforcement, and deep learning strategies. ML algorithms can use historical network data in both supervised and deep learning strategies.

[0054] The ML algorithm outputs a congestion prediction based on data transmission according to the Transmission Control Protocol (TCP) (e.g., the time when congestion might occur and / or the probability of congestion occurring), and provides this output to a scheduler, for example, in a network entity, which determines the control transmission rate. This prediction can be based on learning from TCP congestion mechanisms, low-level protocol configurations, UE radio conditions, etc.

[0055] For example:

[0056] Suppose an 8kB TCP window size is configured in the transmitter and receiver. Due to congestion and latency, an instance's buffer occupancy is 7kB, and data exceeding 1kB may trigger the TCP slow start mechanism. A scheduler or similar entity in the network entity (e.g., a gNB) can perform packet inspection and is aware of the configured TCP window size. Therefore, in some embodiments, the circuitry (of the network entity) is also configured to perform Transmission Control Protocol (TCP) packet inspection.

[0057] This accumulation of data may be due to lost packets or slow scheduling.

[0058] Packet loss at the TCP layer can also lead to a missing ACK at the RLC AM sublayer. The RLC sublayer may notice this at a slightly different time than TCP because TCP ACK and RLC ACK may have different timings (RLC ACK is based on Poll-PDU and Poll-Byte, i.e., the number of PDUs and bytes before sending the ACK), and the network / application may configure the same or different values ​​for TCP ACK. Furthermore, the mechanisms for configuring these parameters are different: TCP windows are dynamically configured using TCP packets, while RLC parameters are configured by RRC and most likely at bearer establishment. Packets may also be lost in the PDCP sublayer due to dropped timers expiring.

[0059] Therefore, the scheduler can organize these events, and ML and / or AI models can learn and predict future possibilities.

[0060] If buffer accumulation is due to congestion and packets arrive late, this can be detected, for example, by using a PDCP drop timer or UL delay parameter.

[0061] Therefore, the scheduler can collect these statistics and predict future activities with a degree of determinism. Schedulers equipped with this information can avoid buffer overflows or saturation and slow down or speed up packet scheduling.

[0062] Therefore, in some embodiments, transmission rate control is performed by controlling the data scheduling rate. The data scheduling rate can be controlled by the network entity, and thus indirectly controls the transmission rate of data transmission according to the Transmission Control Protocol. Since the TCP window is configured within the TCP packet itself, and the RLC window is configured using Radio Resource Control (“RRC”), cross-layer alignment may not work correctly. However, by utilizing ML implemented near the network entity scheduler, ML can understand the characteristics and processing of TCP traffic in the network and adjust the scheduling of data transmission for this service accordingly.

[0063] Typically, the TCP endpoint on the network side is unknown (somewhere on the internet), and another option could be, for example, if the gNB (network entity) and UE communicate and take appropriate action, i.e., take action to control the transmission rate. A more direct option could be for the gNB to communicate any possible TCP endpoint congestion or blockage to the UE, or to avoid triggering a slow start mechanism.

[0064] For example, there are existing technologies regarding new MAC (“Media Access Control”) control elements for forming UEs with recommended bit rates. However, consensus has been reached on specific use cases for MTSI (“Multimedia Telephony Service for IMS”) and ANBR (“Access Network Bit Rate Recommendation”) as specified in 3GPP TS 26.114 and TS 38.321. These are used by RTP / RTCP and sent to the UE by the gNB for recommended data rates for uplink (“UL”) and downlink (“DL”). Here, the average window time is fixed at 2000 milliseconds. This is equivalent to the rate control mechanism for audio.

[0065] In some embodiments, the MAC control element described above (in this document: the first MAC control element) can be modified to extend the rate control mechanism for TCP-based applications so that rate control is implemented when the TCP window is about to close. In such embodiments, the average window time is configurable, allowing it to be adjusted according to different applications and / or radio conditions.

[0066] Therefore, in some embodiments, the circuitry of the network entity is further configured to generate a first MAC control element including a recommended bit rate, and wherein, by transmitting the first MAC control element to the user equipment using a transmission control protocol-based service to perform transmission rate control, the circuitry adjusts the transmission rate of data transmission according to the transmission control protocol in response to and based on the transmitted first MAC control element.

[0067] In some embodiments, the first MAC control element includes the average window time.

[0068] The user equipment (UE) is configured accordingly, meaning it can receive the first MAC control element and adjust the transmission rate accordingly. In response to the received first MAC control element, the UE can adjust the TCP window size and / or transmission bit rate and / or average window time based on the received first MAC control element.

[0069] In some embodiments, the recommended bit rate and / or average window time are based on the output of a machine learning algorithm.

[0070] In some embodiments, network entities perform necessary signaling adaptations regarding congestion and transmission rate through user equipment to implement transmission rate control when the TCP window is about to close, through signaling procedures other than the first MAC control element. In such embodiments, the signaling is based on RRC signaling, physical layer control signaling (DCI), allocated licenses, etc.

[0071] In some embodiments, the circuitry (of the network entity) is further configured to receive a query for a recommended bit rate from the user equipment and, in response to the received query, transmit a first MAC control element to the user equipment.

[0072] Further explanations of the recommended bit rates are described in 3GPP TS 38.321 (e.g., in section 5.8.10), and these explanations can be used / extended / modified for transmission rate control in TCP-based applications.

[0073] Furthermore, a new MAC control element (in this paper: Second MAC Control Element) can be introduced from the UE to the network entity, indicating any data rate preference from the UE's TCP layer perspective. The MDT (“Minimized Driver Test”) framework has already introduced UL PDCP queuing delay (“Channel Quality Indicator”) or 5QI (“5G Quality of Service Indicator”) for QCI as a parameter. This is crucial for AS delay (i.e., from packet arrival at the PDCP layer to receiving UL grant + HARQ), RLC delay, F1 delay, and PDCP reordering delay. This parameter can provide delay at the AS layer, but due to the involvement of different layers, it may not address TCP window stopping issues. The new MAC control element can also modify RLC or PDCP parameters, such as Poll-PDU, Poll-Byte, or PDCP drop timers.

[0074] Therefore, in some embodiments, the circuit is also configured to receive a second MAC control element, including data rate preferences, from the user equipment using a transmission control protocol-based service, and wherein the circuit further performs transmission rate control based on the second MAC control element. The second MAC control element provides, for example, further information about the UE's radio status and transmission channel.

[0075] In some embodiments, the second MAC control element includes uplink packet data aggregation protocol queuing delay for at least one of the channel quality indicator and the 5G service quality indicator.

[0076] In some embodiments, the second MAC control element changes at least one of the radio link control and packet data convergence protocol parameters.

[0077] In some embodiments, the parameters include at least one of Poll-PDU, Poll-Byte, and Packet Data Convergence Protocol drop timer.

[0078] The MAC control element is an example, and in some embodiments, actual signaling can occur via PDCP control PDU, L1 signaling, or RLC control PDU. Although RLC data or PDCP data PDUs can also be used for this purpose, these formats may reduce flexibility because new bits indicating a new format may have backward compatibility issues and may require the ability to understand the new format.

[0079] As described above, in some embodiments, the output of the machine learning algorithm includes predictions about data transmission congestion based on TCP. The network entity can become aware (detect) from the ML output that TCP is about to initiate congestion control (slow start mechanism), and the network (entity) scheduler will, for example, prevent this by changing the transmission rate.

[0080] In some embodiments, the machine algorithm includes a recurrent neural network.

[0081] Typically, neural networks are organized into multiple layers, where each layer includes one or more nodes, and each node in a layer is connected to nodes in the immediately preceding and following layers. The layer that receives external data (input) is the input layer, and the layer that produces results and / or predictions (output) is the output layer. Intermediate layers are intermediate layers containing one or more hidden layers. Each connection between nodes is assigned a weight. A trained neural network can be characterized by trained weights. In a recurrent neural network (“RNN”), the input can be a sequence of data, such as a time series of data, and the recurrent network has an internal state. Nodes in an RNN iteratively receive input data, for example, using the output of the first iteration as the input for the second iteration, and so on. Therefore, future events can be predicted based on the temporal evolution of various input parameters, i.e., it is suitable for predicting time series data (e.g., timing of data transmission congestion).

[0082] In some embodiments, the question is what the key factors of congestion are, since these factors should be input into the ML algorithm. Typically, ML algorithms can automatically find relevant input parameters from a large pool of input parameters. In this sense, any type of input is acceptable. However, too many parameters can lead to additional costs (e.g., a large number of neurons or layers in a neural network, resulting in enormous processing power). Therefore, in some embodiments, it is best to select relevant inputs based on the best knowledge of the communication system designer.

[0083] The following parameters can be used as input and output parameters for predicting TCP congestion:

[0084] Input layer:

[0085] • Radio status:

[0086] Synchronization signal-reference signal received power (SS-RSRP);

[0087] Channel State Information - Reference Signal Received Power (CSI-RSRP);

[0088] Synchronization signal-reference signal reception quality (SS-RSRQ);

[0089] Channel State Information - Reference Signal Received Quality (CSI-RSRQ);

[0090] Channel Quality Indicator (CQI);

[0091] ο Detection Reference Signal (SRS) Measurement; and

[0092] ο block error rate

[0093] • RLC layer:

[0094] Error or missing ACK in ORLC layer

[0095] ·PDCP layer:

[0096] Expiration of the discard timer in the οPDCP layer

[0097] TCP layer:

[0098] Base stations (i.e., network entities) can use deep packet inspection to read application layer data and then interpret the contents of the TCP header; and

[0099] The TCP port number can be a clue to the application (e.g., FTP downloads, web, messaging, video streaming, video conferencing tools, etc.).

[0100] • Business load:

[0101] Historical data on service load with timestamps. Network congestion is likely to occur during peak hours.

[0102] Layer 2 Measurement - UL PDCP Queue Delay

[0103] Output layer:

[0104] • Limit the start time / date;

[0105] • Limit the starting position;

[0106] • The types of restricted services; and

[0107] • Rate control of UL or DL ​​or both.

[0108] Therefore, in some embodiments, the output of the recurrent neural network includes at least one of the following: the timing of the start of connection restriction, the location of the start of connection restriction, the type of restricted service, and predictions of uplink transmission rate control, downlink rate control, or both.

[0109] In some embodiments, the input to the recurrent neural network includes time series data.

[0110] In some embodiments, the time-series data includes radio conditions.

[0111] In some embodiments, radio condition includes at least one of synchronization signal-reference signal received power, channel state information-reference signal received power, synchronization signal-reference signal received quality, channel state information-reference signal received quality, channel quality indicator, probe reference signal measurement, and block error rate.

[0112] In some embodiments, the time-series data includes at least one of radio link control layer errors and lost ACKs.

[0113] In some embodiments, the time-series data includes the expiration of a discard timer in the packet data aggregation protocol layer.

[0114] In some embodiments, time-series data includes information from the Transmission Control Protocol header.

[0115] In some embodiments, the time-series data includes at least one of timestamped traffic load and uplink packet data aggregation queuing delay.

[0116] In some embodiments, the recurrent neural network is trained based on historical training data, which can be obtained from historical network data, including data on the input parameters and data on when TCP congestion occurred.

[0117] In some embodiments, a recurrent neural network is trained offline or during operation.

[0118] In some embodiments, the training process is deployed within a network entity (e.g., a base station, etc.) as described herein, which includes electronic components (circuitets) typically used for training the ML algorithm (i.e., neural network) of the process, such as memory, microprocessors, graphics processing units, etc. In other embodiments, the training process is deployed within an external server / tool ​​for network operation and maintenance (O&M). In some embodiments, the training process is processed offline. In other embodiments, the training process is processed during real-time network operation, wherein the server includes sufficient memory to store historical (training) data. In some embodiments, the network's raw data (historical data) is too large to be stored in the memory within the network entity or server. In such embodiments, the data is processed prior to the training process, for example, by averaging, to reduce its size.

[0119] In some embodiments, a trained ML algorithm (e.g., a recurrent neural network with trained weights) is deployed for inference in network entities (the actual operation of admission control). In such embodiments, the input to the ML algorithm (e.g., the recurrent neural network) is actual (real-time) data from real-time network monitoring, some static configuration, and historical data (which may include real-time network monitoring data from previous iterations).

[0120] Some embodiments relate to user equipment for a mobile telecommunications system, including circuitry configured to use a Transmission Control Protocol-based service and receive from a network entity a first MAC control element including a recommended bit rate, the first MAC control element being based on the output of a machine learning algorithm that includes congestion predictions for data transmission according to the Transmission Control Protocol, and the circuitry responding to and based on the received first MAC control element to adjust the transmission rate of data transmission according to the Transmission Control Protocol.

[0121] User equipment may be or may include smartphones, VR devices, laptops, etc. The circuit may include at least one of the following: a processor, microprocessor, dedicated circuit, memory, storage device, radio interface, wireless interface, network interface, etc., for example, typical electronic components included in the user equipment to implement the functions described herein.

[0122] In some embodiments, the circuitry (of the user equipment) is also configured to transmit a query for the recommended bit rate to the network entity.

[0123] In some embodiments, the circuitry (of the user equipment) is also configured to transmit a second MAC control element, including a data rate preference, to the network entity.

[0124] In some embodiments, the second MAC control element includes uplink packet data aggregation protocol queuing delay for at least one of the channel quality indicator and the 5G service quality indicator.

[0125] A typical mobile phone (user device) may have different processors used for modems and applications.

[0126] Therefore, it has been recognized that the implementation of user equipment can coordinate activities at different layers, that is, the AS (“access layer”) layer in the user equipment’s modem can know the TCP headers generated / received in the application layer.

[0127] Therefore, some embodiments relate to user equipment for mobile telecommunications systems, including circuitry configured to coordinate activities at different layers, such that a modem included in the user equipment obtains information about the transmission control protocol header.

[0128] In some embodiments, from the user equipment's perspective, if the user equipment detects congestion in data transmission, it can send a buffer status report (“BSR”) indicating a smaller buffer size than it actually has, in order to request a smaller license from the network entity. Buffer status reports are typically obtained from LTE.

[0129] Therefore, some embodiments relate to user equipment for mobile telecommunications systems, including circuitry configured to transmit buffer status reports to network entities, the reports indicating a buffer size smaller than its actual buffer size.

[0130] As described above, the user equipment may be or may include smartphones, VR devices, laptops, etc. The circuitry may include at least one of the following: a processor, microprocessor, dedicated circuitry, memory, storage device, radio interface, wireless interface, network interface, etc., for example, typical electronic components included in the user equipment to implement the functions described herein.

[0131] In some embodiments, the network entities and user equipment described herein constitute a transmission rate control system for data transmission according to a transmission control protocol, and / or are part of a mobile telecommunications system (network).

[0132] return Figure 2 The diagram illustrates the slow-start mechanism of data transmission rate according to the transmission control protocol.

[0133] according to Figure 2 The temporal evolution of the transmission rate of data transfer (i.e., TCP segments) in the Transmission Control Protocol (TCP) illustrates a typical problematic situation (which will be avoided by the techniques described in this article).

[0134] At point 10a, the transmission rate increases slowly (initial cwnd in TCP), and begins to increase exponentially at point 10b until it reaches a certain level. At point 10c, the transmission rate increases moderately when the congestion window reaches the TCP slow start threshold (ssthresh in TCP).

[0135] However, TCP is unaware of when congestion might occur, so it continues to increase the transmission rate until the maximum allowed rate is reached. Congestion occurs at some point, and a lost TCP ACK is detected at l0d. This allows TCP to react by detecting congestion at l0e and quickly reducing the transmission rate. When the low transmission rate is reached, the slow start mechanism is restarted at l0f.

[0136] This typical time evolution is useful as training data for machine learning algorithms, for example, Figure 3 The current neural network is shown during the training phase.

[0137] Figure 3 A block diagram of an embodiment of a recurrent neural network 20 in the training phase is shown.

[0138] In this embodiment, the recurrent neural network 20 (indicated by arrows returning to the same node) during the training phase is deployed in network entity 7 and receives input from a data storage device, which includes historical data 21 at input layer 22. In this embodiment, as described above, the input includes historical time-series data, including radio conditions, data / information from RLC, PDCP, TCP layers, and traffic load.

[0139] The nodes of input layer 22 are connected to the first node of intermediate layer 23. Intermediate layer 23 performs computations, and the last node is connected to the output layer, which outputs the prediction of congestion occurrence (the time of congestion and / or the probability of congestion occurrence) and other output values, as described herein.

[0140] The loss function 25 compares the predicted results with the actual results obtained from the stored historical data 21 and uses the backpropagation algorithm to update the weights of the neural network 20 in order to improve the prediction accuracy of the recurrent neural network 20.

[0141] Figure 4 A block diagram of an embodiment of the recurrent neural network 30 in the inference phase is shown.

[0142] Neural network 30 corresponds to Figure 3 A trained recurrent neural network 20 is deployed in network entity 7 for inference, wherein the input layer 32, intermediate layer 33, and output layer 34 have the same characteristics as... Figure 3The same structure applies. The recurrent neural network 30 obtains actual (real-time) data 31 from the real-time network and measurement results (input data as described herein) and outputs a prediction of congestion regarding data transmission according to the transmission control protocol to the scheduler 35 (as part of the network entity).

[0143] Scheduler 35 uses TCP-based service scheduling to schedule data in scheduler queue 36, including data transmission from user equipment. Based on congestion prediction of data transmission from recurrent neural network 30, scheduler 35 determines the transmission rate control for data transmission in scheduler queue 36 and identifies activities 37 for performing transmission rate control, which are described herein.

[0144] Scheduler 35 can, for example, adjust the data scheduling rate of data transmission from TCP-based services to avoid slow start mechanisms and maintain a high transmission rate.

[0145] In other embodiments, the deployment of the machine learning function varies, for example, the input layer 32, intermediate layer 33, and output layer 34. As discussed, one direct variation is the deployment within the network entity (e.g., within a base station). In some embodiments, this is high-performance processing (e.g., cloud / edge computing) outside the network entity and connected to the network entity via an interface (e.g., an O&M network). Network virtualization / cloud-based RANs can provide alternative, flexible deployment options.

[0146] Figure 5 A graph showing the time evolution of the transmission rate of data transmission controlled by network entity 7 according to the transmission control protocol is shown.

[0147] At 11a, the transmission rate increases slowly (initial cwnd in TCP), and begins to increase exponentially at 11b until it reaches a certain level. At 11c, the transmission rate increases moderately when the congestion window reaches the TCP slow start threshold (ssthresh in TCP).

[0148] Figure 4 Network entity 7 continuously monitors the temporal evolution of the data transmission rate based on the transmission control protocol and other input values ​​(as described herein). Based on congestion predictions of data transmission from recurrent neural network 30, scheduler 35 adjusts the data scheduling rate of data transmission from TCP-based services to avoid slow start mechanisms and maintain a high transmission rate at 11d.

[0149] Figure 6 A state diagram is shown for a first embodiment of data transmission rate control controlled by network entity 7 according to the transmission control protocol.

[0150] UE 6 has established a TCP connection with (server) PC 41, and UE 6 will download data from PC 41. At point 50, PC 41 connects via router R 40 and network entity NE 7 (whose configuration is as follows). Figure 4 and Figure 5 (As shown) via network at a low transmission rate (e.g.) Figure 5 (As shown) Sends the first data packet, which confirms its receipt.

[0151] At point 51, PC 41 increases the transmission rate (congestion window), as indicated by the arrow pointing towards UE 6. NE 7 continuously monitors network conditions (radio conditions, TCP header information, etc., as described herein) in parallel to determine the transmission rate control for data transmission based on the output of a recurrent neural network (machine learning algorithm), including predicting congestion for data transmission according to the Transmission Control Protocol.

[0152] At point 52, PC 41 further increases the transmission rate (congestion window), and the recurrent neural network 30 in NE 7 predicts that congestion may occur as the transmission rate increases further (e.g., due to buffer overflow at router R 40 or a decline in radio link conditions, as described herein). Based on the machine learning algorithm and the output of the recurrent neural network 30, NE 7 performs transmission rate control for data transmission according to the transmission control protocol by controlling the data scheduling rate.

[0153] As a result, the transmission rate remains constant at point 53, as indicated by the same number of arrows at points 52 and 53, which avoids triggering the TCP slow start mechanism (such as...). Figure 5 (As shown).

[0154] Figure 7 A state diagram is shown for a second embodiment of data transmission rate control controlled by network entity 7 according to the transmission control protocol.

[0155] UE 6 has established a TCP connection with server PC 41, and UE 6 is about to upload data to PC 41. At point 60, PC 41 connects via router R 40 and network entity NE 7 (whose configuration is as follows). Figure 4 and Figure 5 (As shown) via network at a low transmission rate (e.g.) Figure 5 (As shown) Sends the first data packet, which confirms its receipt.

[0156] At point 61, NE 7 generates a first MAC control element including a recommended bit rate, and forms a transmission rate control by transmitting the first MAC control element to UE 6 using a service based on the transmission control protocol. In response to and based on the transmitted first MAC control element, NE 7 adjusts the transmission rate of data transmission according to the transmission control protocol.

[0157] At position 62, data transmission proceeds at the adjusted transmission rate, which remains unchanged at position 63, as indicated by the same number of arrows at positions 62 and 63. This avoids triggering TCP slow start mechanisms (such as...). Figure 5 (As shown).

[0158] Figure 8 A state diagram is shown for a third embodiment of data transmission rate control controlled by network entity 7 according to the transmission control protocol.

[0159] UE 6 has established a TCP connection with server PC 41, and UE 6 is about to upload data to PC 41. At point 70, PC 41 connects via router R 40 and network entity NE 7 (whose configuration is as follows). Figure 4 and Figure 5 (As shown) via network at a low transmission rate (e.g.) Figure 5 (As shown) The first data packet is transmitted, and the packet confirms its receipt.

[0160] At point 71, UE 6 transmits a query for the recommended bit rate to NE 7, and NE 7, in response to the received query, transmits the first MAC control element (from...) to the user equipment. Figure 7 ), used to perform transmission rate control of data transmission according to the transmission control protocol.

[0161] In another embodiment, in the absence of a response from the network, a query for the first MAC control element can be transmitted from UE 6 to NE 7.

[0162] In other embodiments, congestion may be detected at the receiver, for example in entity PC 41, such that if NWDAF is deployed, the NWDAF entity receives input directly from PC 41.

[0163] At point 72, data transmission proceeds at the adjusted transmission rate, which remains unchanged at point 73, as indicated by the same number of arrows at points 72 and 73. This avoids triggering TCP slow start mechanisms (such as...). Figure 5 (As shown).

[0164] Figure 9 A state diagram is shown for a fourth embodiment of data transmission rate control controlled by network entity 7 according to the transmission control protocol.

[0165] This embodiment is basically the same as Figure 6 The implementation is the same, except that UE 6, which uses a service based on the transmission control protocol, transmits a second MAC control element including data rate preference to NE 7 at 80, and NE 7 further performs transmission rate control based on the second MAC control element.

[0166] Steps 81 to 84 correspond to Figure 6 In steps 50 to 53, in addition to controlling the data scheduling rate in NE 7, the transmission rate control is based on a machine learning algorithm (recurrent neural network 30) and the output of the second MAC control element.

[0167] refer to Figure 10 Embodiments of the UE 6 and network entity (NE) 7 (e.g., NR eNB / gNB) used to implement embodiments of the present disclosure, as well as the communication path 104 between the UE 6 and NE 7, are discussed.

[0168] UE 6 has a transmitter 101, a receiver 102 and a controller 103. The technical functions of the transmitter 101, receiver 102 and controller 103 are generally known to those skilled in the art, and therefore, a more detailed description thereof is omitted.

[0169] NE 7 has a transmitter 105, a receiver 106 and a controller 107, wherein, also here, the functions of the transmitter 105, receiver 106 and controller 107 are generally known to those skilled in the art, and therefore a more detailed description thereof is omitted.

[0170] Communication path 104 has an uplink path 104a from UE 6 to NE 7 and a downlink path 104b from NE 7 to UE 6.

[0171] During operation, the controller 103 of UE 6 controls the reception of downlink signals at receiver 102 via downlink path 104b, and the controller 103 controls the transmission of uplink signals via transmitter 101 through uplink path 104a.

[0172] Similarly, during operation, the controller 107 of NE 7 controls the transmission of downlink signals at transmitter 105 via downlink path 104b, and the controller 107 controls the reception of uplink signals at receiver 106 via uplink path 104a.

[0173] In the following text, see references Figure 11 An embodiment of the general-purpose computer 130 is described.

[0174] Computer 130 can be implemented such that it can be used substantially as any type of network entity, base station or new radio base station, transmission and reception point, or user equipment as described herein. The computer has components 131 to 141 that can form circuits, such as circuits for either a base station or a user equipment, as described herein.

[0175] Embodiments of using software, firmware, programs, etc. to perform the methods described herein may be installed on computer 130, and the computer may then be configured to suit the specific embodiment.

[0176] Computer 130 has a CPU 131 (Central Processing Unit), which can execute various types of processes and methods described herein, for example, based on programs stored in read-only memory (ROM) 132, stored in storage device 137 and loaded into random access memory (RAM) 133, or stored on medium 140 that can be inserted into a corresponding drive 139.

[0177] CPU 131, ROM 132, and RAM 133 are connected to bus 141, which in turn is connected to input / output interface 134. The number of CPUs, memory, and storage devices is merely exemplary, and those skilled in the art will understand that when computer 130 is used as a base station or user equipment, it can be adjusted and configured accordingly to meet specific needs.

[0178] At input / output interface 134, several components are connected: input 135, output 136, storage device 137, communication interface 138 and driver 139. Media 140 (compressed disk, digital video optical disc, compressed flash memory, etc.) can be inserted into these components.

[0179] Entering 135 can be a pointing device (mouse, chart, etc.), keyboard, microphone, camera, touch screen, etc.

[0180] Output 136 can include a display (LCD, CRT, LED, etc.), a speaker, etc.

[0181] Storage device 137 may include hard disks, solid-state drives, etc.

[0182] The communication interface 138 can be adapted to communicate via, for example, a local area network (LAN), a wireless local area network (WLAN), a mobile telecommunications system (GSM, UMTS, LTE, NR, etc.), Bluetooth, infrared, etc.

[0183] It should be noted that the above description pertains only to an example configuration of computer 130. Alternative configurations can be achieved using additional or other sensors, storage devices, interfaces, etc. For example, communication interface 138 can support other radio access technologies besides UMTS, LTE, and NR mentioned.

[0184] When computer 130 is used as a base station, communication interface 138 may also have a corresponding air interface (providing, for example, E-UTRA protocol OFDMA (downlink) and SC-FDMA (uplink)) and network interface (implementing, for example, protocols such as S1-AP, GTP-U, S1-MME, X2-AP, etc.). Computer 130 is also implemented to transmit data according to TCP. Furthermore, computer 130 may have one or more antennas and / or antenna arrays. This disclosure is not limited to any characteristics of these protocols.

[0185] Figure 12 A block diagram of an embodiment of user equipment 6 is shown.

[0186] UE 6 includes an application processor 150, circuitry 151, and a modem 152. The application processor, i.e., at the application layer, generates TCP data transmissions, for example. The circuitry coordinates activities at different layers, allowing the modem 152 to obtain information about the TCP header.

[0187] For illustrative purposes only, the circuit is shown as a separate entity, but it may also be integrated into the application processor 150 or the modem 152.

[0188] Figure 13 The diagram shows the state of the buffer status report being transmitted from user equipment 6 to network entity 7.

[0189] UE 6 connects to the mobile telecommunications system and communicates via NE 7.

[0190] At 160, UE 6 transmits a buffer status report to the network entity, indicating a smaller buffer size than it actually has, in order to request a smaller license from NE 7.

[0191] At point 161, based on the buffer status report received from UE 6, NE 7 transmits a new license to UE 6, including the transmission parameters that UE 6 should use.

[0192] Unless otherwise stated, all units and entities described in this specification and claimed in the appended claims may be implemented as integrated circuit logic, for example, on a chip, and unless otherwise stated, the functionality provided by such units and entities may be implemented by software.

[0193] With regard to the implementation of the above-disclosed embodiments using at least a software-controlled data processing device, it will be understood that providing such a software-controlled computer program and the transmission, storage or other medium providing such a computer program are contemplated as aspects of this disclosure.

[0194] Note that this technology can also be configured as described below.

[0195] (1) A network entity for a mobile telecommunications system, comprising circuitry configured to perform transmission rate control for data transmission according to a transmission control protocol, wherein the transmission rate control is performed based on the output of a machine learning algorithm that includes congestion prediction of data transmission.

[0196] (2) The network entity according to (1), wherein the transmission rate control is performed by controlling the data scheduling rate.

[0197] (3) The network entity according to (1) or (2), wherein the circuit is further configured to generate a first MAC control element including a recommended bit rate, and wherein the transmission rate control is performed by transmitting the first MAC control element to the user equipment using a service based on the transmission control protocol, and the circuit adjusts the transmission rate of the data transmission according to the transmission control protocol in response to and based on the transmitted first MAC control element.

[0198] (4) The network entity according to (3), wherein the recommended bit rate is based on the output of the machine learning algorithm.

[0199] (5) The network entity according to (3) or (4), wherein the first MAC control element includes an average window time.

[0200] (6) The network entity according to (5), wherein the average window time is based on the output of the machine learning algorithm.

[0201] (7) The network entity according to any one of (3) to (6), wherein the circuit is further configured to receive a query for the recommended bit rate from the user equipment and, in response to the received query, transmit the first MAC control element to the user equipment.

[0202] (8) The network entity according to any one of (1) to (7), wherein the circuit is further configured to receive a second MAC control element including a data rate preference from the user equipment using a service based on the transmission control protocol, and wherein the transmission rate control is also performed based on the second MAC control element.

[0203] (9) The network entity according to (8), wherein the second MAC control element includes uplink packet data aggregation protocol queuing delay for at least one of the channel quality indicator and the 5G service quality indicator.

[0204] (10) The network entity according to (8) or (9), wherein the second MAC control element changes at least one of the radio link control and packet data convergence protocol parameters.

[0205] (11) The network entity according to (10), wherein the parameters include at least one of Poll-PDU, Poll-Byte and Packet Data Convergence Protocol drop timer.

[0206] (12) The network entity according to any one of (1) to (11), wherein the circuit is further configured to perform Transmission Control Protocol data packet inspection.

[0207] (13) The network entity according to (12), wherein the machine algorithm includes a recurrent neural network.

[0208] (14) The network entity according to (13), wherein the output of the recurrent neural network includes the timing of the start of connection restriction, the location of the start of connection restriction, the type of restricted service, and at least one of the predictions of uplink transmission rate control, downlink rate control, or both.

[0209] (15) The network entity according to (13) or (14), wherein the input of the recurrent neural network includes time series data.

[0210] (16) The network entity according to (15), wherein the time-series data includes radio conditions.

[0211] (17) The network entity according to (16), wherein the radio condition includes at least one of synchronization signal reference signal received power, channel state information reference signal received power, synchronization signal reference signal received quality, channel state information reference signal received quality, channel quality indicator, probe reference signal measurement and block error rate.

[0212] (18) A network entity according to any one of (15) to (17), wherein the time-series data includes at least one of radio link control layer errors and lost ACKs.

[0213] (19) A network entity according to any one of (15) to (18), wherein the time-series data includes the expiration of a discard timer in the packet data aggregation protocol layer.

[0214] (20) A network entity according to any one of (15) to (19), wherein the time-series data includes information from the Transmission Control Protocol header.

[0215] (21) The network entity according to any one of (15) to (20), wherein the time-series data includes at least one of timestamped service load and uplink packet data aggregation queuing delay.

[0216] (22) The network entity according to any one of (13) to (21), wherein the recurrent neural network is trained based on historical training data.

[0217] (23) The network entity according to any one of (13) to (22), wherein the recurrent neural network is trained offline or during operation.

[0218] (24) A user equipment for a mobile telecommunications system, comprising circuitry configured to use a service based on a transmission control protocol and to receive from a network entity a first MAC control element including a recommended bit rate based on the output of a machine learning algorithm, the output of which includes a congestion prediction of data transmission according to the transmission control protocol, and the circuitry adjusting the transmission rate of data transmission according to the transmission control protocol in response to and based on the received first MAC control element.

[0219] (25) The user equipment according to (24), wherein the circuit is further configured to transmit a query for the recommended bit rate to the network entity.

[0220] (26) The user equipment according to (24), wherein the circuit is further configured to transmit a second MAC control element including data rate preference to the network entity.

[0221] (27) The user equipment according to (26), wherein the second MAC control element includes uplink packet data aggregation protocol queuing delay for at least one of the channel quality indicator and the 5G service quality indicator.

[0222] (28) A user equipment for a mobile telecommunications system, including circuitry configured to coordinate activities at different layers, whereby a modem included in the user equipment obtains information about a transmission control protocol header.

[0223] (29) A user equipment for a mobile telecommunications system includes circuitry configured to transmit a buffer status report to a network entity, the report indicating a buffer size smaller than the actual buffer size it has.

Claims

1. A network entity for a mobile telecommunications system, the network entity including circuitry configured to perform transmission rate control of data transmission according to a transmission control protocol, wherein, The transmission rate control is performed based on the output of a machine learning algorithm that includes congestion prediction of the data transmission; wherein the circuit is further configured to generate a first media access control element including a recommended bit rate, and wherein the transmission rate control is performed by transmitting the first media access control element to the user equipment using a service based on the transmission control protocol, and the circuit adjusts the transmission rate of the data transmission according to the transmission control protocol in response to and based on the transmitted first media access control element.

2. The network entity according to claim 1, wherein, The transmission rate control is performed by controlling the data scheduling rate.

3. The network entity according to claim 1, wherein, The recommended bit rate is based on the output of the machine learning algorithm.

4. The network entity according to claim 1, wherein, The first media access control element includes the average window time.

5. The network entity according to claim 4, wherein, The average window time is based on the output of the machine learning algorithm.

6. The network entity according to claim 1, wherein, The circuit is also configured to receive a query for the recommended bit rate from the user equipment, and in response to the received query, transmit the first media access control element to the user equipment.

7. The network entity according to claim 1, wherein, The circuit is also configured to receive a second media access control element, including a data rate preference, from a user equipment using a service based on the transmission control protocol, and wherein the transmission rate control is also performed based on the second media access control element.

8. The network entity according to claim 7, wherein, The second media access control element includes uplink packet data aggregation protocol queuing delay for at least one of the channel quality indicator and the 5G service quality indicator.

9. The network entity according to claim 7, wherein, The second media access control element changes at least one of the parameters of the radio link control and packet data convergence protocol.

10. The network entity according to claim 9, wherein, The parameters include at least one of Poll-PDU, Poll-Byte, and Packet Data Convergence Protocol drop timer.

11. The network entity according to claim 1, wherein, The circuit is also configured to perform Transmission Control Protocol (TCP) data packet inspection.

12. The network entity according to claim 11, wherein, The machine learning algorithm includes a recurrent neural network.

13. The network entity according to claim 12, wherein, The output of the recurrent neural network includes the timing of the start of connection restriction, the location of the start of connection restriction, the type of restricted service, and at least one of the predictions of uplink transmission rate control, downlink rate control, or both.

14. The network entity according to claim 12, wherein, The input to the recurrent neural network includes time series data.

15. The network entity according to claim 14, wherein, The time-series data includes radio conditions.

16. The network entity according to claim 15, wherein, The radio condition includes at least one of synchronization signal-reference signal received power, channel state information-reference signal received power, synchronization signal-reference signal received quality, channel state information-reference signal received quality, channel quality indicator, probe reference signal measurement, and block error rate.

17. The network entity according to claim 14, wherein, The time-series data includes at least one of errors and lost ACKs from the radio link control layer.

18. The network entity according to claim 14, wherein, The time-series data includes the expiration of discard timers in the packet data aggregation protocol layer.

19. The network entity according to claim 14, wherein, The time-series data includes information from the Transmission Control Protocol header.

20. The network entity according to claim 14, wherein, The time-series data includes at least one of timestamped service load and uplink packet data aggregation queuing delay.

21. The network entity according to claim 12, wherein, The recurrent neural network is trained based on historical training data.

22. The network entity according to claim 12, wherein, The recurrent neural network can be trained offline or during operation.

23. A user equipment for a mobile telecommunications system, the user equipment including circuitry configured to use a service based on a transmission control protocol and to receive from a network entity a first media access control element including a recommended bit rate based on the output of a machine learning algorithm, the output of which includes a congestion prediction of data transmission according to the transmission control protocol, and the circuitry adjusting the transmission rate of the data transmission according to the transmission control protocol in response to and based on the received first media access control element.

24. The user equipment according to claim 23, wherein, The circuit is also configured to transmit a query for the recommended bit rate to the network entity.

25. The user equipment according to claim 23, wherein, The circuit is also configured to transmit a second media access control element, including data rate preferences, to the network entity.

26. The user equipment according to claim 25, wherein, The second media access control element includes uplink packet data aggregation protocol queuing delay for at least one of the channel quality indicator and the 5G service quality indicator.

Images