TRANSMISSION SYSTEM, TRANSMISSION METHOD, AND TRANSMISSION PROGRAM
The transmission system dynamically adjusts reception buffer sizes based on network conditions using a learning model to minimize delays, enhancing performance in real-time applications.
Patent Information
- Application Number
- JP2022579293
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2021-02-08
- Publication Date
- 2024-11-20
- Estimated Expiration
- 2041-02-08
AI Technical Summary
Existing transmission systems using SMPTE ST2110 experience unnecessary delays due to fixed buffer settings, which do not account for network fluctuations, impacting real-time performance in applications like web conferences and autonomous driving.
A transmission system that calculates a jitter pattern sequence using a learning model to determine a network state index, allowing for dynamic adjustment of the reception buffer size based on network conditions, optimizing buffer size to minimize delays.
Improves transmission performance by adapting buffer sizes to network states, reducing delays and ensuring low-latency transmission in varying network conditions.
Smart Images

Figure 0007587170000001 
Figure 0007587170000002 
Figure 0007587170000003
Abstract
Description
[Technical field]
[0001] The present invention relates to a transmission system, a transmission method, and a transmission program. [Background technology]
[0002] SMPTE ST2110 is a method for transmitting uncompressed video and audio over a network. SMPTE ST2110-20 describes a method for storing video essence in the payload of RTP (Real-time Transport Protocol) packets for transmission (Non-Patent Document 1). In uncompressed video transmission, the number of packets per frame is constant as long as the video parameters and the pixel mapping method to the RTP packets do not change.
[0003] SMPTE ST2110-21 describes a desirable RTP packet transmission timing model for uncompressed video transmission (Non-Patent Document 2). In the Linear model, the RTP packet transmission interval is always constant. In the Gapped model, a timing model based on SDI (Serial Digital Interface), RTP packets are transmitted at constant intervals only during a period corresponding to the effective pixels of SDI in each frame. The receiving device is designed based on the transmission timing model. [Prior art documents] [Non-patent literature]
[0004] [Non-Patent Document 1] SMPTE ST 2110-20:2017, Professional Media Over Managed IP Networks: Uncompressed Active Video [Non-Patent Document 2] SMPTE ST 2110-21:2017, Professional Media Over Managed IP Networks: Traffic Shaping and Delivery Timing for Video Summary of the Invention [Problem to be solved by the invention]
[0005] It is known that various problems occur in transmission via a network depending on the network conditions. For example, video can be played when a certain amount of packets are available, but delays occur depending on the network conditions, and the time it takes for a certain amount of packets to be available varies. To deal with this, a buffer is provided to store the waiting time until a certain amount of packets are available for playback. In conventional transmission devices, a certain value is set in the buffer to absorb the time fluctuations until data is available in a congested state.
[0006] However, even when the network is stable, the same value is applied to the buffer as when the network is congested, which causes unnecessary delays. Such delays are an obstacle in cases where real-time performance is required, such as online communication in web conferences, remote medical care, and video control for autonomous driving.
[0007] Furthermore, since the transmission standard SMPTE ST2110 is intended to send video and audio signals over a network instead of via coaxial cable, it is desirable to minimize network delays as the transition from coaxial cable to a network takes place.
[0008] The present invention has been made in consideration of the above circumstances, and an object of the present invention is to improve transmission performance in accordance with the state of a network. [Means for solving the problem]
[0009] In order to achieve the above-mentioned object, one aspect of the present invention is a transmission system comprising: a calculation unit that calculates a jitter pattern sequence of a plurality of packets transmitted by a transmitting device over a network using a transmission interval of the packets and a difference in arrival time between the packets and a receiving device; a generation unit that inputs the jitter pattern sequence to a first learning model to generate a state indicator indicating a state of the network; and a buffer estimation unit that determines a buffer size of a receiving buffer of the receiving device based on the state indicator.
[0010] One aspect of the present invention is a transmission method performed by a transmission system, which includes the steps of: calculating a jitter pattern sequence of a plurality of packets transmitted by a transmitting device over a network using the transmission interval of the packets and the difference in arrival time of each of the packets at a receiving device; inputting the jitter pattern sequence into a first learning model to generate a status index indicating the status of the network; and determining a buffer size of a receiving buffer of the receiving device based on the status index.
[0011] One aspect of the present invention is a transmission program that causes a computer to function as the above-mentioned transmission system. Effect of the Invention
[0012] According to the present invention, it is possible to improve transmission performance according to the network state. [Brief description of the drawings]
[0013] [Figure 1] FIG. 1 illustrates an example of a configuration of a transmission system according to a first embodiment. [Diagram 2] FIG. 2 is a block diagram showing a configuration of a network estimation unit. [Diagram 3] 13 is a flowchart showing a transmission control process. [Figure 4] FIG. 2 is a sequence diagram showing the operation of the transmission system. [Diagram 5] FIG. 11 is a diagram illustrating an example of the configuration of a transmission system according to a modified example of the first embodiment. [Figure 6] FIG. 11 illustrates an example of the configuration of a transmission system according to a second embodiment. [Figure 7] FIG. 13 is a diagram illustrating an example of the configuration of a transmission system according to a modified example of the second embodiment. [Figure 8] 2 is an example of a hardware configuration. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0014] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
[0015] <First embodiment> 1 is a diagram showing an example of the configuration of a transmission system according to the first embodiment. In this embodiment, the transmission interval of RTP packets transmitted by the transmitting device 1 is fixed (known) according to a known rule determined by the transmission method. That is, this embodiment is applied to a transmission method with a fixed transmission interval, such as video transmission according to SMPTE ST 2110. In this case, only the receiving device 2 performs packet capture.
[0016] In this embodiment, RTP packets are used, but the present invention is not limited to RTP packets, and packets other than RTP packets may be used.
[0017] The illustrated transmission system includes a transmitting device 1 and a receiving device 2, and the transmitting device 1 and the receiving device 2 are connected via a network (not shown).
[0018] The transmitting device 1 has a packet generating unit 11, a transmission buffer 12, and a network interface 13. The packet generating unit 11 generates RTP packets and outputs them to the network interface 13 via the transmission buffer 12. The network interface 13 stores the RTP packets in network protocol packets (hereinafter referred to as "network packets") and transmits them to the receiving device 2 via the network. The network interface 13 of this embodiment transmits the transmission intervals of the RTP packets to the network estimation unit 25 of the receiving device 2.
[0019] The receiving device 2 includes a packet capture unit 21 , a network interface 22 , a receiving buffer 23 , a reproduction unit 24 , a network estimation unit 25 , a buffer estimation unit 26 , and a control unit 27 .
[0020] The packet capture unit 21 captures a network packet transmitted via a network, and outputs to the network estimation unit 25 the arrival time at which the network packet arrives at the receiving device 2 .
[0021] The network interface 22 extracts the RTP packets from the received network packets, and outputs the RTP packets to the playback unit 24 via the receiving buffer 23. The playback unit 24 extracts and arranges the data portion of the RTP packets, thereby playing back the RTP packets as video.
[0022] The network estimation unit 25 generates a state indicator indicating the state of the network between the transmitting device 1 and the receiving device 2, using the transmission interval and arrival time of the packets, and outputs the state indicator to the buffer estimation unit 26. The network estimation unit 25 will be described later.
[0023] The buffer estimation unit 26 determines the buffer size of the receiving buffer 23 of the receiving device 2 based on the state index. The buffer estimation unit 26 may input the state index to a learning model for buffer estimation (second learning model) to determine the buffer size of the receiving buffer 23. The buffer estimation unit 26 may also acquire an observed value output from the receiving buffer 23, and determine the buffer size of the receiving buffer 23 using a deviation between the observed value or an estimated value estimated from the observed value and the state index.
[0024] The control unit 27 updates the receiving buffer 23 to the buffer size determined by the buffer estimation unit 26 .
[0025] In the following description, RTP packets and network packets may be referred to as packets.
[0026] 2 is a configuration diagram showing the configuration of the network estimation unit 25. The illustrated network estimation unit 25 includes a calculation unit 251, a generation unit 252, a learning unit 253, and a learning model 254.
[0027] The calculation unit 251 generates a jitter pattern sequence of a plurality of packets, which the transmitting device 1 transmits via a network, by using the transmission intervals of the packets and the difference between the arrival times (arrival times) of the packets at the receiving device 2. Specifically, the calculation unit 251 acquires the arrival times of a plurality of packets that are consecutive in time series from the packet capture device 21, and calculates the difference between the arrival times of these consecutive packets.
[0028] The generator 252 inputs the jitter pattern sequence to a learning model 254 (first learning model) to generate a state index indicating the state of the network.
[0029] The learning unit 253 generates the learning model 254 by machine learning. The learning model 254 may be a neural network generated by deep learning. The deep learning may be unsupervised learning or supervised learning. Specifically, the learning unit 253 may generate a neural network by performing unsupervised learning using off-peak time periods of the network (route) as learning data, or may generate a neural network by performing supervised learning in which labels are assigned to the transmission intervals collected from the transmitting device 2 and the multiple arrival times collected from the receiving device 2.
[0030] In this embodiment, the network estimation unit 25 is provided in the receiving device 2. However, the network estimation unit 25 may be provided in an external device instead of the receiving device 2.
[0031] Next, a basic flow of video transmission and playback in this embodiment will be described.
[0032] In the transmitting device 1, a packet generating unit 11 divides a video stream to be transmitted into RTP packets in accordance with SMPTE ST 2110, and outputs the RTP packets to a network interface 13 via a transmitting buffer 12. The network interface 13 stores the RTP packets in network packets of an arbitrary network protocol, and transmits them onto a network toward the receiving device 2.
[0033] As a network protocol, User Datagram Protocol (UDP), Internet Protocol (IP), Ethernet, etc. may be used. The type of network is not important. A wavelength path with a dedicated wavelength assigned to the network may be used.
[0034] In the receiving device 2, the network interface 22 extracts the RTP packets from the network packets received via the network, and outputs them to the playback unit 24 via the receiving buffer 23. The playback unit 24 extracts and arranges the data portion of the RTP packets to play back the RTP packets as video.
[0035] In addition to transmitting and playing back video, the transmission system of this embodiment controls transmission in response to the state of the network, thereby improving transmission performance.
[0036] FIG. 3 is a flowchart showing the transmission control process.
[0037] The transmitting device 1 starts transmitting RTP packets (S11). At that time, the network interface 13 of the transmitting device 1 transmits transmission interval information indicating a prescribed transmission interval of the RTP packets to the network estimation unit 25. In the receiving device 2, the packet capture 21 sequentially collects the network packets that arrive at the network interface 22, and outputs the arrival time of each network packet to the network estimation unit 25.
[0038] The network estimation unit 25 estimates the state of the network using the transmission interval information acquired from the transmitting device 1 and the arrival time acquired from the packet capture 21, and generates a state indicator indicating the state (S12).
[0039] First, the calculation unit 251 calculates the arrival time difference of packets using the arrival times notified from the receiving device 2. Specifically, the calculation unit 251 calculates the arrival time difference using the difference between the arrival times of two consecutive packets. Then, the generation unit 252 calculates a jitter pattern sequence using the transmission interval and the arrival time difference. Specifically, the calculation of the jitter pattern sequence is as follows.
[0040] The ideal sequence of RTP packet transmission times for one frame based on the transmission timing model is denoted as T ideal Let (k) be the index of the RTP packet in one frame, and 0≦k≦N-1, where N is the number of RTP packets per frame.
[0041] If the first RTP packet is sent at time 0, T ideal(k) = C * k. In this case, the packet transmission interval between the k-th packet and the (k-1)-th packet is as follows:
[0042] ΔT ideal (k) = T ideal (k)-T ideal (k-1) = C The calculation unit 251 determines the head packet of the frame based on the packet whose timestamp has changed, the addition of marker information, etc. The transmission device 1 notifies the network estimation unit 25 of the transmission interval C of this RTP packet.
[0043] As for the timing of notification, the transmitting device 1 may notify the transmission interval C by adding the transmission interval C to the RTP packet at the start of transmission. Alternatively, a default value for the transmission interval C may be set in advance in the network estimation unit 25, and the transmitting device 1 may notify the network estimation unit 25 of the transmission interval C only when the transmission interval C is different from the default value.
[0044] The packet capture 21 of the receiving device 2 captures a packet arrival time sequence T RX (k), 0≦k≦N-1 is obtained. The calculation unit 251 calculates the arrival time difference using the difference between the arrival times of two consecutive packets. The arrival time difference is as follows.
[0045] ΔT RX (k) = T RX (k)-T RX (k-1) The calculation unit 251 defines a jitter pattern using the difference in transmission delay. The jitter pattern in this embodiment is, for example, as follows.
[0046] J 1 (k)=(T RX (k)-T ideal (k))-(T RX (k-1)-T ideal (k-1) =(T RX (k)-TRX (k-1))-(T ideal (k)-T ideal (k-1) = ΔT RX (k)-C The calculation unit 251 performs the above calculation for multiple packets to generate a time-series jitter pattern sequence. In this way, since the jitter pattern can be calculated from the relative time in the receiving device 2, it is clear that time synchronization between the transmitting device 1 and the receiving device 2 is not necessary.
[0047] Although the calculation unit 251 of this embodiment calculates the jitter pattern based on the transmission interval and arrival time difference between two consecutive packets, the interval between sampled packets may be lengthened, or the average of delay differences between a plurality of packets may be calculated.
[0048] Next, the generating unit 252 generates a state index based on the jitter pattern sequence. First, the generating unit 252 performs preprocessing such as descending sorting, absolute value processing, feature extraction, and frequency on the jitter pattern sequence. The generating unit 252 inputs the preprocessed jitter pattern sequence to, for example, a neural network (learning model) generated by deep learning, and obtains a state index output by the neural network. The neural network is trained so as to output an optimal size of the receiving buffer 23 for the input jitter pattern sequence.
[0049] The neural network structure used for learning can be a simple fully connected one, a CNN, or an RNN for time series data. The weights used for learning can be learned for each network path or can use pre-trained weights.
[0050] When performing deep learning for each route, transmissions are performed for each route for a certain period of time without transmission control, and the data obtained is used to train the neural network from scratch, pre-trained weights are used as is in the neural network, or pre-trained weights are applied to the neural network and adjusted and re-trained using a small amount of data while it is in operation.
[0051] When learning or relearning weights from scratch, the learning unit 253 may obtain the maximum jitter (delay time) and network state (congested state, quiet state) from the observed values (actual operational data) in the receiving buffer 23, and use the obtained data to train the neural network.
[0052] Possible state indices output from the neural network include, for example, indices showing the degree of congestion, such as the time period, traffic volume, and number of simultaneously connected terminals. In addition, when the state indices are applied to the optimization of the buffer size, since the receiving buffer 23 affects the minimum waiting time according to the network state, the state indices may also include the predicted maximum jitter and the sustainability of the current state (congested state, uncongested state, etc.). The sustainability of the current state means the possibility or probability that a state transition will occur.
[0053] When transmission control is performed by optimizing the buffer size, the buffer estimation unit 26 receives a state indicator from the network estimation unit 25 and determines an optimal buffer size using the state indicator. The control unit 27 updates the receiving buffer 23 with the determined buffer size (S13).
[0054] The buffer estimation unit 26 may determine the buffer size by feeding back an observation value from the receiving buffer 23. In the case of feedback control, the buffer estimation unit 26 may obtain an observation value output from the receiving buffer 23, and determine the buffer size using a deviation (difference) between the observed value or an estimated value of the receiving buffer 23 estimated from the observed value, and the state index. The observed value is, for example, the actual maximum jitter, the degree of filling of the buffer, the occurrence of packet loss, etc.
[0055] Specifically, the buffer estimation unit 26 may calculate the deviation between the actual maximum jitter, which is an observed value, and the maximum jitter of the state index, and determine the buffer size based on the calculated deviation. The buffer estimation unit 26 may also estimate whether a state transition has occurred based on the observed buffer filling level, the occurrence of packet loss, etc., and adjust the buffer size to match the actual network path, thereby improving the accuracy of controlling the buffer size.
[0056] Furthermore, the buffer estimation unit 26 may determine the buffer size by using a neural network (learning model) generated by reinforcement learning. Specifically, the buffer estimation unit 26 may input the state index output from the network estimation unit 25 to the neural network, and determine the buffer size output by the neural network as the size of the receiving buffer 23.
[0057] The neural network learns to output an optimal receiving buffer size using observed values (actual operation data) of the receiving buffer 23 actually operated for a specific route in the network. In this case, the weights may be learned for a certain period of time within the receiving device 2, such as the buffer estimation unit 26. Alternatively, only data collection may be performed within the receiving device 2, the collected data may be transmitted to a learning unit 253 installed in a device outside the receiving device 2, and the updated weights may be returned to the receiving device 2 to update the neural network provided in the buffer estimation unit 26.
[0058] In addition to the above, the buffer estimation unit 26 may determine a stepwise buffer size according to a state indicator output from the network estimation unit 25. Specifically, a threshold value may be set for the state indicator, and a stepwise buffer size may be set according to the state.
[0059] FIG. 4 is a sequence diagram showing the processing of the transmission system of the present embodiment.
[0060] The transmitting device 1 transmits a packet to the receiving device 2 (S21). The transmitting device 1 transmits the packet transmission interval to the network estimation unit 25 (S22).
[0061] The receiving device 2 transmits the arrival time of the packet to the network estimation unit 25 (S23). The network estimation unit 25 calculates a jitter pattern sequence using the transmission interval and the arrival time, inputs the jitter pattern sequence to a learning model to obtain a network status index, and outputs the status index to the buffer estimation unit 26 (S24).
[0062] The buffer estimation unit 26 determines the buffer size of the receiving buffer 23 using the state index, and outputs the buffer size to the control unit 27 (S25). The control unit 27 updates the receiving buffer 23 with the buffer size output from the buffer estimation unit 26 (S26). The reproduction unit 24 reproduces the packets stored in the updated receiving buffer 23 as video (S27).
[0063] 5 shows an example of the configuration of a transmission system according to a modification of the first embodiment. In this modification, a network estimation unit 25 is provided in an external device 3, not in a receiving device 2. The rest is the same as in the first embodiment.
[0064] <Second embodiment> 6 is a system configuration diagram showing a configuration example of a transmission system according to the second embodiment. In this embodiment, the transmission system is a system in which the transmission interval of packets transmitted by the transmitting device 1A does not follow a specific rule, but dynamically changes and is unknown. In this case, the transmitting device 1A and the receiving device 2A perform packet capture. In this embodiment, there is no need to perform time synchronization between the transmitting device 1A and the receiving device 2A.
[0065] The illustrated transmission system of the present embodiment includes a transmitting device 1A and a receiving device 2A, which are connected to each other via a network. The transmitting device 1A of the present embodiment differs from the first embodiment in that it includes a packet capturer 14, but is otherwise similar to the transmitting device 1 of the first embodiment. The packet capturer 14 captures network packets sent from the network interface 13, and transmits the sending time (sending time) of the network packets to the network estimation unit 25.
[0066] The receiving device 2A differs from the first embodiment in that a network estimation unit 25A (calculation unit) collects the transmission times of a plurality of network packets from the transmitting device 1 and calculates the difference in the transmission times of successive network packets as a transmission interval, but is otherwise similar to the receiving device 2 of the first embodiment. The network estimation unit 25A of this embodiment also includes a calculation unit 251, a generation unit 252, a learning unit 253, and a learning model 254, as shown in FIG.
[0067] In this embodiment, the packet transmission time sequence T observed (k) does not have a fixed transmission interval. The information on the transmitting side transmitted to the network estimation unit 25 is the transmission time of the packet obtained by the packet capture 14.
[0068] Also, series T observed(k) may be a frame unit by adding a time stamp and marker information of the first packet as in the first embodiment. Alternatively, a specific number of packets may be set, an arbitrary packet may be set as the first packet, and the set number of packets from that point may be processed as one sequence. k is the index of the RTP packet in one sequence, and 0≦k≦N-1, where N is the number of RTP packets included in one sequence.
[0069] In this case, the packet transmission interval sequence (transmission interval) for the k-th packet and the (k-1)-th packet is as follows:
[0070] ΔT observed (k) = T observed (k)-T observed (k-1) The network estimation unit 25A estimates the packet transmission interval sequence ΔT observed Calculate (k). ΔT observed (k) is a timing for estimating the state of the network so that the necessary amount of data is present in the network estimation unit 25A. The timing of transmission of the transmission time by the transmitting device 1A can be, for example, transmitting as additional information during packet transmission, or transmitting 1 to several sequences collectively after transmission is completed.
[0071] If the packet arrival time sequence on the receiving device 2 side is ΔTRX(k) as in the first embodiment, the packet arrival interval is as follows:
[0072] ΔT RX (k) = T RX (k)-T RX (k-1) Therefore, the jitter pattern in this embodiment is as follows:
[0073] J 2 (k)=(T RX (k)-T observed (k))-(T RX (k-1)-T observed (k-1) =(T RX (k)-TRX (k-1))-(T observed (k)-T observed (k-1) = ΔT RX (k)- ΔT observed (k) ΔT RX (k) and ΔT observed Since (k) is the relative time difference between the transmitting device 1A and the receiving device 2A, it can be seen that in this embodiment as well, time synchronization between the transmitting device 1A and the receiving device 2A is not necessary.
[0074] The network estimation unit 25A performs the above calculations for multiple packets to generate a time-series jitter pattern sequence. Then, similar to the first embodiment, the network estimation unit 25A inputs the jitter pattern sequence to a learning model and outputs a state index.
[0075] The rest is the same as in the first embodiment, so a description thereof will be omitted here.
[0076] 7 shows an example of the configuration of a transmission system according to a modification of the second embodiment. In this modification, a network estimation unit 25 is provided in an external device 3, not in a receiving device 2. The rest is the same as in the second embodiment.
[0077] The transmission systems of the first and second embodiments described above include a calculation unit 251 that calculates a jitter pattern sequence of a plurality of packets transmitted by the transmitting devices 1, 1A via a network, using the transmission intervals of the packets and the difference in arrival times of the packets at the receiving devices 2, 2A, respectively, a generation unit 252 that inputs the jitter pattern sequence to a learning model 254 and generates a state index indicative of the state of the network, and a buffer estimation unit 26 that determines the buffer size of the receiving buffer 23 of the receiving devices 2, 2B based on the state index.
[0078] As a result, in this embodiment, it is possible to improve the transmission performance according to the network state. Specifically, by optimizing the buffer size and changing the waiting time according to the network state, it is possible to prevent overflows in a congested state and enable low-latency transmission in a stable state.
[0079] In this embodiment, the network estimation units 25, 25A acquire the transmission intervals and arrival intervals of packets in the transmitting devices 1, 1A and the receiving devices 2, 2A, input the jitter pattern, which is the variation in delay time difference, to the learning model, and estimate and index the network state from the transmission information that can be acquired during transmission. As a result, in this embodiment, even if the user cannot obtain information on the entire network, the transmitting device 1 transmits the transmission interval together with the transmission data, making it possible to use the transmission system of this embodiment.
[0080] Furthermore, since the jitter pattern can be calculated based on the relative time in each of the devices 1, 1A, 2, 2B, rather than on absolute time, it can be used even if the transmitting devices 1, 1A and the receiving devices 2, 2A are not time-synchronized.
[0081] Furthermore, the network estimation units 25, 25A may be incorporated in the receiving devices 2, 2A, or may be incorporated in the external device 3. In the former case, the process can be completed within the receiving device 2, except for obtaining the transmission interval transmitted from the transmitting devices 1, 1A, and improved stability can be expected. On the other hand, in the latter case, the generation of the network status indicator is performed externally, so that the dependency on the performance of the computing unit of the receiving devices 2, 2A is reduced, and the range of use of the transmission system of this embodiment can be expected to be expanded. In addition, in the latter case, scalability such as integrating and processing route information of multiple terminals, such as multicast, can be expected.
[0082] In this embodiment, since the buffer size is optimized using a network status index, it is not necessary to set a constant large buffer size, and low latency can be achieved when the network is in a stable state. On the other hand, even if the network becomes congested while transmission is being performed according to the stable state, it is possible to control the buffer size without disturbing the video.
[0083] In this embodiment, the buffer size is optimized by feedback control or reinforcement learning. In the case of a device with high computing performance, highly flexible optimization is performed in real time by feedback control, and in the case of a device with low computing performance, reinforcement learning is performed from data sampled through actual use to obtain weights appropriate for the route, and the process can be selected taking into account the device performance, and it is expected that the system can be used in a variety of devices.
[0084] As another method of controlling the buffer size, a threshold value may be set for a network status indicator, and the buffer size may be set in stages according to the status.
[0085] For example, a general-purpose computer system as shown in FIG. 8 can be used for the above-described transmitting devices 1 and 1A, receiving devices 2 and 2A, and external device 3. The illustrated computer system includes a CPU (Central Processing Unit, processor) 901, a memory 902, a storage 903 (HDD: Hard Disk Drive, SSD: Solid State Drive), a communication device 904, an input device 905, and an output device 906. The memory 902 and the storage 903 are storage devices. In this computer system, the CPU 901 executes a predetermined program loaded on the memory 902, thereby realizing each function of each device. For example, the functions of the transmitting devices 1 and 1A and the receiving devices 2 and 2A are realized by the CPUs of the transmitting devices 1 and 1A executing a program for the transmitting devices, and the CPUs of the receiving devices 2 and 2A executing a program for the receiving devices, respectively.
[0086] Moreover, these devices 1, 1A, 2, 2B, and 3 may be implemented in one computer, or may be implemented in multiple computers. Moreover, these devices 1, 1A, 2, 2B, and 3 may be virtual machines implemented in a computer. The programs for the transmitting device, the receiving device, and the external device may be stored in a computer-readable recording medium such as an HDD, an SSD, a Universal Serial Bus (USB) memory, a Compact Disc (CD), or a Digital Versatile Disc (DVD), or may be distributed via a network.
[0087] The present invention is not limited to the above-described embodiment, and various modifications are possible within the scope of the present invention.
[0088] For example, in the above embodiment, the transmission is controlled by optimizing the buffer size of the receiving buffer 23, but other transmission control may be performed. For example, redundant transmission, introduction of error correction code, etc. are conceivable. In this case, the state indicator output from the network estimation unit 25 is transmitted to the transmitting device 1, not to the receiving device 2. The transmitting device 1 includes a control unit that determines the presence or absence of redundant transmission based on the state indicator, or determines an error correction coding method, and generates control information for controlling the transmitting device 1 with the determined content. Furthermore, the control unit may select a path for redundant transmission, or may determine the adoption of a coding method that can handle a high error rate. The redundant transmission can improve the stability of transmission. In this case, the network state estimation unit 25 may be provided on the transmitting device 1 side. In addition, as a transmission control, a control unit that outputs an alert when the state of the network estimated from the state indicator is bad may be provided. [Explanation of symbols]
[0089] 1: Transmitting device 11: Packet generation unit 12: Send buffer 13: Network interface 14: Packet capture 2: Receiving device 21: Packet capture 22: Network interface 23: Receive buffer 24: Playback section 25: Network estimation unit 251: Calculation section 252: Generation part 253: Learning Department 254: Learning model (neural network) 26: Buffer estimation unit 27: Control unit
Claims
1. A calculation unit that calculates a jitter pattern sequence of a plurality of packets transmitted by a transmitting device over a network using a transmission interval of the packets and a difference in arrival time of the packets at a receiving device; a generation unit that inputs the jitter pattern sequence into a first learning model to generate a state index that indicates a state of the network; a buffer estimation unit that determines a buffer size of a receiving buffer of the receiving device based on the status indicator; The buffer estimation unit inputs the state indicator into a second learning model to determine a buffer size of the receiving buffer. Transmission system.
2. A calculation unit that calculates a jitter pattern sequence of a plurality of packets transmitted by a transmitting device via a network using a transmission interval of the packets and a difference in arrival time of the packets at a receiving device; a generation unit that inputs the jitter pattern sequence into a first learning model to generate a state index that indicates a state of the network; a buffer estimation unit that determines a buffer size of a receiving buffer of the receiving device based on the status indicator; The buffer estimation unit acquires an observed value output from the receiving buffer, and determines a buffer size of the receiving buffer using a deviation between the observed value or an estimated value estimated from the observed value and the state indicator. Transmission system.
3. A calculation unit that calculates a jitter pattern sequence of a plurality of packets, the packets being transmitted by a transmitting device via a network, using a transmission interval of the packets and a difference in arrival time of the packets at a receiving device; a generation unit that inputs the jitter pattern sequence into a first learning model to generate a state index that indicates a state of the network; a buffer estimation unit that determines a buffer size of a receiving buffer of the receiving device based on the status indicator; The first learning model is a neural network generated by deep learning, A learning unit is provided that generates the neural network by performing unsupervised learning using off-peak time periods of the network as learning data, or generates the neural network by performing supervised learning in which labels are added to the transmission intervals collected from the transmitting device and the arrival times collected from the receiving device. Transmission system.
4. A transmission method performed by a transmission system, comprising: calculating a jitter pattern sequence of a plurality of packets, the jitter pattern sequence being calculated based on the transmission intervals of the plurality of packets transmitted by a transmitting device via a network and the difference between the arrival times of the packets at a receiving device; inputting the jitter pattern sequence into a first learning model to generate a state indicator indicative of a state of the network; determining a buffer size of a receiving buffer of the receiving device based on the status indicator; The step of determining the buffer size includes inputting the state indicator into a second learning model to determine the buffer size of the receiving buffer. Transmission method.
5. A transmission method performed by a transmission system, comprising: calculating a jitter pattern sequence of a plurality of packets, the jitter pattern sequence being calculated based on the transmission intervals of the plurality of packets transmitted by a transmitting device via a network and the difference between the arrival times of the packets at a receiving device; inputting the jitter pattern sequence into a first learning model to generate a state indicator indicative of a state of the network; determining a buffer size of a receiving buffer of the receiving device based on the status indicator; The step of determining the buffer size includes acquiring an observed value output from the receiving buffer, and determining the buffer size of the receiving buffer using a deviation between the observed value or an estimated value estimated from the observed value and the state indicator. Transmission method.
6. A transmission method performed by a transmission system, comprising: calculating a jitter pattern sequence of a plurality of packets, the jitter pattern sequence being calculated based on the transmission intervals of the plurality of packets transmitted by a transmitting device via a network and the difference between the arrival times of the packets at a receiving device; inputting the jitter pattern sequence into a first learning model to generate a state indicator indicative of a state of the network; determining a buffer size of a receiving buffer of the receiving device based on the status indicator; The first learning model is a neural network generated by deep learning, A learning step is performed to generate the neural network by performing unsupervised learning using the off-peak time period of the network as learning data, or to generate the neural network by performing supervised learning in which the transmission interval collected from the transmitting device and the arrival time collected from the receiving device are labeled. Transmission method.
7. A transmission program that causes a computer to function as the transmission system according to any one of claims 1 to 3.
Citation Information
Patent Citations
Packet receiving apparatus, method and program for determining necessary and sufficient reception buffer size
JP2008244894A
Apparatus and method for predicting delay fluctuation
JP2014135685A
Information processing device, information processing system, information processing program, and information processing method
JP2019008554A
Fraud communication detection device, and fraud communication detection program
JP2020135816A
