JP6033069B2 - Communication quality estimation device - Google Patents

Description

  The present invention relates to a communication quality estimation device, and in particular, passively captures communication packets, analyzes communication parameters, and estimates throughput of a session with a small data size based on an analysis result of a session with a large data size. The present invention relates to a communication quality estimation apparatus.

  Wireless communication system operators are required to obtain communication quality information for each local area in a wide service area, create an area map, and thoroughly evaluate the communication quality status of the entire service area. .

  In Patent Document 1, in order to reduce the load on the network and estimate the TCP quality with a small amount of calculation, one quality measuring device is connected to the other between two quality measuring devices connected via a packet communication network. The measurement packet is sent to the quality measurement device, and the other sends the corresponding measurement packet back to one quality measurement device. The other quality measurement device returns the quality based on the reception status of the returned measurement packet. A technique for actively estimating the TCP throughput between the measurement apparatuses, the retransmission occurrence rate, the average packet round-trip delay, and the probability of occurrence of timeout retransmission is disclosed.

  Patent Document 2 discloses a technique for calculating an average throughput of a TCP connection based on a transmission band of a communication line that is shared by a plurality of terminals and connections.

  In Patent Document 3, between two quality measurement devices connected via a packet communication network, one quality measurement device intermittently sends measurement packets to the other quality measurement device at regular intervals, and each quality measurement device A technique for actively measuring communication quality related to a measurement packet based on a transmission / reception state related to the measurement packet transmitted / received by an apparatus is disclosed.

  In Patent Literature 4, a communication quality management device obtains, from each radio base station, radio channel quality information at the current position of the portable information terminal, radio base station radio channel resource information, and radio base station network channel usage information. A technique for collecting and determining communication quality at each position and creating communication quality map information in which the determined communication quality is associated with position information is disclosed.

JP 2004-140596 A JP 2003-209574 A JP 2004-7339 A JP 2010-62783 A

  Since Patent Documents 1 and 3 are active systems that send and receive dedicated packets for measurement, extra traffic and load are generated during quality measurement, and a large number of quality measurement devices must be provided to expand the measurement range. Scalability issues arise, such as having to. In Patent Document 2, only the quality of the communication line is measured, and the quality of experience of the client user cannot be measured.

  In Patent Document 4, since the communication quality information management device operates actively and must request the measurement and notification of communication quality from the radio base station and the packet switching device, the number of requested devices becomes large. There is a scalability issue. In addition, since communication quality monitoring is performed as an additional process in response to a request from the management apparatus, there is a problem that the processing load and traffic increase with measurement.

  The object of the present invention is to solve all the problems of the prior art, passively capture communication packets and analyze communication parameters, and based on the analysis results of sessions and the like that have a large data size and are not easily affected by flow control. An object of the present invention is to provide a communication quality estimation device that can estimate the throughput of a session or the like that is small in size and easily affected by flow control with high accuracy.

  In order to achieve the above object, the present invention provides the following means in a communication quality estimation apparatus that passively monitors packets transmitted and received using a session or connection as a communication unit to estimate communication quality. There is a feature.

  (1) Based on the monitoring result management means for managing the attribute information, packet size and arrival time of the packets captured on the route where each communication unit is aggregated as the monitoring result for each communication unit, and the monitoring result of each packet Network quality analysis means that analyzes communication parameters including communication rate for each communication unit, and throughput from any communication unit communication rate, with the communication rate and throughput of the communication unit that is less susceptible to flow control as the teaching data Learning means for constructing a communication quality dictionary for estimating the throughput, and throughput estimation means for estimating the throughput of a communication unit whose throughput is susceptible to flow control by applying the communication rate of the communication unit to the communication quality dictionary. Equipped.

  (2) TCP initial window size, advertisement window size, window size time series transition, RTT or bandwidth delay product is calculated as one of the communication parameters for each communication unit, and throughput estimation results are used as these parameters. Based on this, a communication quality dictionary is constructed for each parameter category.

  According to the present invention, the following effects are achieved.

  (1) Because the data size is large, the relationship between the communication rate and throughput is learned and modeled in advance for connections where the influence of slow start immediately after the start of communication does not affect the throughput, and the data size is small In addition, if the throughput is calculated based on the communication rate, the connection throughput that is calculated lower due to the slow start can be estimated by applying the communication rate to the learning model. And the like can be accurately estimated.

  (2) Throughput estimation for TCP connections or sessions is corrected based on TCP initial window size, advertising window size, window size time series transition, RTT or bandwidth delay product. The influence of TCP flow control on the result can be eliminated, and for example, the accuracy degradation estimated to be low can be improved as compared with a connection having a large throughput of a connection having a small initial window size.

It is a block diagram of the network to which the communication quality measurement method of the present invention is applied. It is the functional block diagram which showed the structure of the capture apparatus. It is the flowchart which showed operation | movement of one Embodiment of this invention. It is the figure which expressed typically the contents of the TCP connection table. It is the sequence flow which showed the method of measuring the delay characteristic of a TCP connection. It is the sequence flow which showed the method of measuring the response characteristic of the end server in HTTP session. It is the sequence flow which showed the method of analyzing the characteristic for every HTTP session. It is the flowchart which showed the construction procedure of a communication quality dictionary. It is the flowchart which showed the estimation procedure of the throughput using a communication quality dictionary. It is the figure which showed the necessity of correct | amending the estimation result of a throughput according to an initial window size. It is the figure which showed the relationship between a data size and a throughput. It is a diagram expressing a throughput estimation mechanism according to the present invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Here, the outline of the present invention will be described first, and then the details will be specifically described.

  Like TCP flow control, the bandwidth is temporarily limited due to protocol specifications and restrictions in addition to network quality, which may cause the apparent communication rate and throughput to decrease. For example, in the TCP slow start algorithm, the communication rate is kept low by reducing the window size immediately after the connection is established, and then the communication rate is gradually increased by gradually increasing the window size according to the actual quality of the network. . Therefore, as shown in FIG. 11 (a), after the time t1 when the data size transferred by the TCP connection is sufficiently large and the slow start is finished and the window size is expanded to a size according to the actual quality of the network. If the communication period is sufficiently long, the throughput can be obtained with sufficient accuracy based on the average value AVE of the communication rates during the period.

  On the other hand, because the data size is small, communication is completed at time t2 before the window size is expanded to the size corresponding to the actual quality of the network, as shown in FIG. Alternatively, as shown in FIG. 4C, if the communication period after time t3 when the window size is expanded to the size corresponding to the actual quality of the network is short, the throughput can be obtained as an average value of the communication rate. Can not.

  On the other hand, the communication rate time series after the connection or session is established is slightly affected by the initial window size and RTT, but the communication rate after the window size is expanded to the size corresponding to the actual quality of the network. It has been confirmed by the inventors' observation that there is a strong correlation.

  Therefore, in the present invention, a communication quality dictionary is constructed by learning in advance the relationship between the communication rate time series and throughput in a connection or session having a large data size, and thereafter, for connections and sessions having a small data size, FIG. As shown in (a) and (b), the communication rate and throughput are estimated by reversely searching the communication quality dictionary using the change in communication rate immediately after the start of the session or the like as a parameter.

  FIG. 1 is a block diagram showing the configuration of a network to which the communication quality estimation method of the present invention is applied. A radio base station BS is installed in each area of the service providing range, and a client in this area (this embodiment) In the form, a radio mobile terminal MN) is accommodated in each radio base station BS. Each radio base station BS is connected to a radio access network RAN, and the radio access network RAN is connected to a gateway (GW) of the core network. The core network is connected to the Internet at the Internet Exchange (IX).

  Various servers that provide services in response to requests from clients are connected to the Internet. In the present embodiment, as a line capable of aggregating traffic between each terminal MN (access) and each server (end), the line L connecting the radio access network RAN and the core network is captured as a communication quality measuring device. Device 1 is connected.

  FIG. 2 is a functional block diagram showing the configuration of the capture device 1. Here, the configuration unnecessary for the description of the present invention is omitted.

  The packet capture unit 101 captures an IP packet from a TCP connection (including an HTTP session) on the line L. The monitoring result management unit 102 includes a table (here, a TCP connection table) 102a that manages captured packets for each predetermined communication unit such as a TCP connection or an HTTP session.

  The network (NW) quality analysis unit 103 includes various analysis means including a communication rate measurement unit 103a, a window size detection unit 103b, a data integration amount calculation unit 103c, and an RTT measurement unit 103d. The analysis result database 104 holds a monitoring result for each communication unit (here, TCP connection).

  The learning unit 107 refers to the analysis result database 104 and constructs a learning model using the analysis result as teacher data for communication whose throughput is not easily affected by TCP flow control because the data size is equal to or larger than a predetermined size threshold. To do. In the present embodiment, the communication rate, window size, total amount of transfer data, RTT, and the like are recorded for each elapsed time with respect to the establishment time of the TCP connection (or HTTP session), and each session is recorded in time zone and A communication quality dictionary 107a that estimates the throughput based on the communication rate of an arbitrary connection and the like is constructed by classifying the day of the week and linking with the average throughput observed from the establishment to the end of the connection.

  The user quality estimating unit 105 includes a classifying unit 105a that reverses the communication quality dictionary 107a using a captured packet analysis result (communication rate time series) as a key, and has a data size of a predetermined size. For communication over the size threshold, the throughput is calculated based on the average value of the communication rate, while for communication with a data size less than the predetermined size threshold, the connection is established for a certain period or until the end of the session. The throughput is estimated by applying the time series of the communication rate observed in the above to the communication quality dictionary. The result output unit 106 outputs the calculation or estimation result of the throughput.

  Note that whether throughput is easily affected by TCP flow control does not depend only on the data size of communication, and the fluctuation range of the time series transition of the ratio between communication rate and throughput (communication rate / throughput) It also depends on the time to settle down. In other words, the fluctuation range of the time series transition of the communication rate / throughput is the time slope of the communication rate / throughput (the delta Δ component that is the differential or the difference), so that the effect of the slow start has become smaller Means.

  Therefore, in place of the selection based on the data size as described above, a communication quality dictionary using as a teacher data communication whose time until the fluctuation range of the time series transition of the communication rate / throughput settles within the predetermined threshold is greater than the predetermined time threshold. To estimate the throughput by applying the communication rate to the communication quality dictionary for the communication whose time until the fluctuation range of the time series transition of the communication rate / throughput settles within the predetermined time is less than the predetermined time threshold Anyway.

  Next, the operation of the capture device 1 will be described in detail with reference to the flowchart of FIG. Note that the flowchart of FIG. 3 mainly shows the operation of the monitoring result management unit 102, and is repeatedly executed at a predetermined cycle.

  In step S <b> 1, a packet that has arrived at the aggregated link L to which the capture device 1 is connected is captured by the packet capture unit 101. In step S2, the source IP address srcIP and its TCP port number srcPort of the captured packet, and if necessary, the destination IP address dstIP and its TCP port number dstPort and the protocol number identify each packet. Extracted as a key. The information adopted as the identification key is not limited to the above. For example, if the information can uniquely identify the client (user) and the communication (connection or session), the user ID in HTTP is adopted as the identification key. You may do it.

  In step S3, it is determined whether or not a record having the same key as the key of the captured packet is already registered in the TCP connection table 102a. In this embodiment, as shown in FIG. 4, for example, the type of packet to be monitored (SYN, SYN + ACK, ACK, Data, etc.), arrival time t, position information P, direction (upstream / downstream) ), Attribute information such as packet size and sequence number is recorded in a record format for each connection using the key as an index.

  If it is immediately after the start of the capture, it is determined that it has not been registered, so the process proceeds to step S4, where a predetermined random number R (0 <R <1.0) is generated. In step S5, the sampling ratio Rsampling preset as the frequency of capturing, recording and analyzing the packet is compared with the random number R. If R ≦ Rsampling, the current packet is determined to be monitored, and step S6. Proceed to In step S6, the attribute information of the captured packet is newly recorded in the TCP connection table 102a by the record method using the key as an index. Note that packets determined as R> Rsampling in step S5 are discarded in step S12.

  On the other hand, if it is determined in step S3 that a record having the same key as the identification key of the captured packet is already registered in the TCP connection table 102a, the process jumps to step S6, and a record related to the packet is stored in the TCP connection table 102a. It is additionally registered. In step S7, it is determined whether or not the packet is a TCP connection disconnection request (FIN) or a forced disconnection (RST). At first, since it is determined that neither FIN nor RST is determined, the process proceeds to step S9. Note that a TCP timeout (TO) may be detected instead of the FIN or RST. Thereafter, when the FIN or RST packet is captured, the process proceeds from step S7 to S8, and the end flag Fend is set in the record recorded in the connection.

  In step S9, it is determined whether or not there is a record with Fend = 1, that is, whether or not there is a newly disconnected connection. If such a record exists, it will progress to step S10 and the said record will be provided to the NW quality analysis part 103 as a communication log. In step S11, all records provided to the NW quality analysis unit 103 are discarded from the TCP connection table 102a. The NW quality analysis unit 103 analyzes the communication log provided from the TCP connection management unit 102 and calculates end-to-end communication quality. This calculation result is notified to the analysis result database 104 and held.

  FIG. 5 is a diagram for explaining an analysis method by the NW quality analysis unit 103. Here, a method for measuring the delay characteristics of a connection that can be captured from a SYN packet of a TCP_3way handshake executed between a client and a server when a TCP connection is established will be described.

  In this case, the server side based on the difference (t2-t1) between the arrival time t1 of the SYN packet first transmitted from the terminal MN to the server and the arrival time t2 of the SYN + ACK packet returned from the server to the terminal MN. RTT (round trip) delay is calculated. The client-side RTT delay is calculated based on the difference (t3-t2) between the arrival time t2 of the SYN + ACK packet and the arrival time t3 of the ACK packet last transmitted from the terminal MN to the server.

  Further, based on the difference (t4-t1) between the arrival time t1 of the first SYN packet and the arrival time t4 of the data packet first transmitted from the terminal MN to the server after the 3-way handshake, the TCP connection required time is calculated. The Furthermore, based on the difference (t5-t1) from the arrival time t1 of the first data transmitted from the terminal MN after the 3-way handshake to the arrival time t5 of the FIN or RST packet, and the amount of transmitted / received data captured within the difference time Thus, the throughput characteristic of the TCP connection is calculated.

  When packet capture is started from the middle of the connection, only possible analysis is selectively performed from the obtained arrival time. That is, if the capture is started from the SYN + ACK packet, only the client-side RTT delay is calculated based on the difference (t3-t2) from the arrival time t2 to the arrival time t3 of the ACK packet.

  In addition, since the throughput characteristics and TCP connection required time of the TCP connection depend not only on the delay on the client side but also on the server side, these characteristics calculated when the server side delay is large are The communication quality based on the location base cannot be accurately represented. Therefore, when the server-side RTT delay exceeds a predetermined threshold, or when the packet arrival interval such as data or ACK that can represent the server-side delay exceeds a predetermined threshold, TCP characteristics and TCP connection It is desirable to exclude the time required from quality analysis.

  FIG. 6 is a diagram showing a method for calculating the response characteristics of the end server in the HTTP session. The proxy server that has received the HTTP request transmitted from the terminal MN executes a connection process with the end server by a three-way handshake. Next, transmission of an HTTP request and reception of an HTTP response are repeated, and after disconnecting from the server, an HTTP response is returned to the terminal HM. The configuration in FIG. 6 can also be realized by terminating the TCP connection in the GW in FIG. 1 and causing the GW to function as a proxy.

  In this case, in addition to the client-side RTT delay (radio section and RAN delay) and server-side RTT delay calculated in FIG. 5, the arrival time t4 of the HTTP request transmitted from the terminal MN to the server and the proxy server Based on the difference (t5-t4) from the arrival time t5 of the HTTP response returned to the terminal MN, the response time of the end server experienced on the client side can be calculated.

  FIG. 7 is a diagram illustrating a method of analyzing characteristics for each HTTP session. In a connection in which multiple HTTP sessions are established sequentially, the analysis is performed for each HTTP session, and for the first HTTP session, the server ends last from the arrival time t31 of HTTP Req1-1 sent first by the terminal MN. The time until the arrival time t32 of HTTP Rep1-3 to be transmitted to is the required time of the first HTTP session, and the amount of data transmitted and received during that time is the amount of communication in the HTTP session.

  Similarly, for the second HTTP session, the time from the arrival time t33 of the first HTTP Req2-1 sent by the terminal MN to the arrival time t34 of the HTTP Rep2-1 sent last by the server is the second HTTP session. The required amount of time is the amount of data transmitted and received during that time, which is the amount of communication in the HTTP session.

  Thus, if the required time and the data amount are obtained for each session, the throughput can be calculated for each HTTP session by dividing the data amount by the required time.

  FIG. 8 is a flowchart showing a procedure for constructing the communication quality dictionary 107a by the learning unit 107. In step S31, a connection whose data size exceeds a predetermined size threshold is selected from the analysis result database 104, and the analysis is performed. Results are captured.

  It is desirable that the size threshold is not a common value for all connections, but is different for each connection. For example, the size threshold value can be a predetermined multiple (for example, 10 times) of the integrated value of the amount of data transmitted and received until the slow start is completed in each connection.

  In this case, the integrated value of the amount of data sent and received until the transmission window size of each connection exceeds the advertisement window size notified from the data receiver for the first time is integrated with the amount of data sent and received before the slow start is completed. It can be estimated as a value. Alternatively, the integrated value of the amount of data transmitted and received before the transmission window size of the connection decreases for the first time compared to the previous value is estimated as the integrated value of the amount of data transmitted and received before the slow start is completed. May be. Further, in addition to the data size threshold, a connection whose holding time TM falls below a predetermined time threshold may be selected.

  In step S32, every elapse time based on the establishment time of an arbitrary connection, using the communication rate and accumulated data amount for each elapsed time based on the throughput of each selected connection and the establishment time of each connection as teacher data. A learning model for estimating the throughput of the connection is constructed for the input of the communication rate and the accumulated data amount. In step S33, the learning model is registered as the communication quality dictionary 107a.

  FIG. 9 is a flowchart showing the throughput estimation procedure by the user quality estimation unit 105. In step S41, the analysis result of the current connection of interest is read from the analysis result database 104. In step S42, it is determined whether or not the throughput of the connection of interest is an estimation target. In the present embodiment, connections whose data size is below a predetermined size threshold are targeted for throughput estimation.

  If it is an estimation target, the process proceeds to step S43, and the communication rate and accumulated data amount for each elapsed time with reference to the connection establishment time are applied to the communication quality dictionary 107a to estimate the throughput. On the other hand, if it is not an estimation target, the process proceeds to step S44, and the throughput is calculated based on the analysis result of the connection. In this embodiment, the average value of the communication rates after connection establishment is calculated as the throughput.

  In step S45, the throughput estimation result or calculation result is output. In step S46, it is determined whether throughput estimation / calculation has been completed for all analysis results. If not completed, the process returns to step S41, and the same processing as described above is executed for the next analysis result.

  According to the present embodiment, since the data size is large, the relationship between the communication rate and the throughput is learned and modeled in advance for a connection in which the influence of the slow start immediately after the start of communication does not easily affect the throughput. Connections with small data size can be estimated by applying the communication rate to the learning model because the connection throughput, which is calculated to be low due to the slow start if the throughput is calculated from the communication rate, is small. And the like can be accurately estimated.

  In the above embodiment, one communication quality dictionary 107a has been described as being shared for throughput estimation of all connections, but the present invention is not limited to this. That is, when the packet of each connection is captured, the time zone and day of the week, and when the location information of the wireless mobile terminal MN is described in each packet, each connection is further classified for each location information. Alternatively, the communication quality dictionary 107a may be constructed. In this way, the throughput can be estimated using the optimum communication quality dictionary according to the time zone, day of the week, and position information at which the connection to be estimated is captured, so that the accuracy can be improved.

  In HTTP, the UserAgent notified by the browser to the server includes terminal information such as hardware information, mobile carrier name, host OS name, and application name, and HTTP request path and URI information that identifies the wireless communication method. It may be listed. Therefore, the client terminal may be identified based on these pieces of information, and for example, only the analysis result of a terminal that uses only EV-DO or LTE as a communication method may be used as teacher data.

  For example, LTE generally has a higher communication rate than other communication methods such as EV-DO. Therefore, by using only the analysis result during LTE communication determined from UserAgent, URI, etc. as teacher data, the communication rate / Throughput estimation can be further improved, and the potential capability of communication can be estimated.

  Therefore, the communication quality dictionary 107a may be constructed for each communication method, and the communication quality dictionary 107a may be switched according to the determination result of the communication method, or only the communication quality dictionary 107a unique to LTE is constructed, Only the analysis result at the time of LTE communication may be applied to the communication quality dictionary 107a to estimate the throughput.

  Furthermore, the communication rate and accumulated data amount for each elapsed time based on the connection establishment time depend not only on the network environment but also on the initial TCP window size and RTT. FIG. 10 is a diagram comparing the communication rates of [solid line A] when the initial window size is small and [solid line B] when the initial window size is large, compared to the case where the initial window size is large and small. The soup rate will increase. Therefore, if the initial window size is not taken into account, even in the same network environment, a connection with a large initial window size tends to have a higher throughput estimation result than a small connection.

  Accordingly, the relationship between the TCP initial window size and the throughput may be statistically analyzed in advance, and the throughput estimation result may be multiplied by the correction coefficient corresponding to the observed initial window size. Alternatively, teacher data may be created for each initial window size. Further, instead of the initial window size, the correction coefficient may be multiplied or teacher data may be created according to the time series transition of the window size or the value of the advertisement window size, or these may be combined. .

  Similarly, RTT depends not only on the network environment but also on the communication method, and a communication method with a small RTT has a higher cost than a communication method with a large RTT. Therefore, if RTT is not taken into account, even in the same network environment, a connection with a small RTT tends to have a higher throughput estimation result than a large connection. Therefore, the relationship between the RTT and the throughput may be statistically analyzed in advance, and the throughput estimation result may be multiplied by the correction coefficient corresponding to the observed RTT. In addition, teacher data may be created for each RTT classification.

  Furthermore, the bandwidth delay product obtained as the product of the bandwidth and the delay time also affects the throughput estimation result as in the case of the RTT, and flows on the line in a network or a line with a large bandwidth delay product (in TCP ACK waiting state) Packet data increases. Therefore, instead of the RTT or together with the RTT, a bandwidth delay product may be obtained, and a correction coefficient may be multiplied according to the bandwidth delay product, or teacher data may be created.

  In the above embodiment, the relationship between the time series of the communication rate and the throughput has been described as being learned in advance, but the present invention is not limited to this, and instead of the time series of the communication rate, You may make it learn beforehand the relationship between the inclination of a communication rate, the time-series transition of a differential value, distribution, a statistic, and a throughput.

  DESCRIPTION OF SYMBOLS 101 ... Packet capture part, 102 ... TCP connection management part, 102a ... TCP connection table, 103 ... Network (NW) quality analysis part, 104 ... Analysis result database, 105 ... User quality estimation part, 106 ... Result output part, 107 ... Learning unit, 107a ... Communication quality dictionary

Claims (13)

In a communication quality estimation device that passively monitors packets sent and received using a session or connection as a communication unit to estimate communication quality,
Capture means for capturing packets arriving on a route where each communication unit is aggregated;
Monitoring result management means for managing the captured packet attribute information, packet size and arrival time as a monitoring result for each communication unit;
Network quality analysis means for analyzing communication parameters including a communication rate for each communication unit based on the monitoring result of each packet;
Learning means for constructing a communication quality dictionary that estimates throughput from a time series of communication rates of an arbitrary communication unit, using as a teacher data the communication rate time series and throughput of a communication unit in which the throughput is not easily affected by flow control,
A communication quality estimation apparatus comprising: quality estimation means for estimating a throughput of a communication unit by applying a time series of communication rates of a communication unit in which the throughput is susceptible to flow control to the communication quality dictionary . The learning means constructs a communication quality dictionary that estimates the throughput from the time series of the communication rate of any communication unit, using the time series and throughput of the communication rate of the communication unit having a data size equal to or greater than a predetermined threshold as teacher data. ,
The said quality estimation means estimates the throughput of the said communication unit by applying the time series of the communication rate of the communication unit whose data size is less than the said threshold value to the said communication quality dictionary. Communication quality estimation device. The learning unit determines an integrated value of the amount of data transmitted and received until it is estimated that the slow start is completed, and a data size is a predetermined multiple of the integrated value as a communication unit equal to or greater than the predetermined threshold. 3. The communication quality estimation apparatus according to claim 2, wherein a communication quality dictionary for estimating throughput is constructed from the time series of communication rates of communication units as described above. It said learning means, in each communication unit, the transmission window size, transmitting and receiving the integrated value of the transmission and reception data amount before the advertised window size notified from the data receiving side exceeds the first time, before the slow start is completed The communication quality estimation apparatus according to claim 3, wherein the communication quality estimation apparatus estimates the integrated value of the data amount.   The learning means calculates, in each communication unit, an integrated value of the amount of data transmitted and received until the transmission window size decreases for the first time as compared to the previous value, and the amount of data transmitted and received until the slow start is completed. The communication quality estimation apparatus according to claim 3, wherein the communication quality estimation apparatus estimates the integrated value of. The learning means uses the time series and throughput of the communication unit in which the time until the fluctuation range of the time series transition of the communication rate / throughput settles within the predetermined threshold is equal to or greater than the predetermined time threshold as the teacher data. Build a communication quality dictionary that estimates throughput from the time series of communication rates of communication units.
The quality estimating means applies a time series of communication rates of communication units in which a time until a fluctuation range of a time series transition of a communication rate / throughput settles within the threshold is less than a predetermined time threshold to the communication quality dictionary. The communication quality estimation apparatus according to claim 1, wherein throughput of the communication unit is estimated. The network quality analysis means analyzes at least one of TCP initial window size, advertisement window size, window size time series transition, RTT and bandwidth delay product as one of communication parameters for each communication unit,
It further comprises correction means for correcting the throughput estimation result by the quality estimation means based on at least one of the initial window size, advertisement window size, window size time series transition, RTT, and bandwidth delay product of the TCP. The communication quality estimation apparatus according to any one of claims 1 to 6, The network quality analysis means analyzes TCP as an initial window size, advertisement window size, time series transition of window size, RTT and one of the communication parameters of bandwidth delay product for each communication unit,
The learning means constructs a communication quality dictionary for each of at least one of the initial window size, advertisement window size, window size time series transition, RTT, and bandwidth delay product of the TCP. The communication quality estimation apparatus according to any one of claims 6 to 6.   9. The communication quality estimation apparatus according to claim 1, wherein the network quality analysis unit analyzes a differential value of the communication rate instead of the communication rate.   The communication quality estimation apparatus according to claim 1, wherein the communication rate is analyzed for each elapsed time based on a communication unit establishment time.   The communication quality estimation apparatus according to any one of claims 1 to 10, wherein the throughput is an average value of a total amount of data transmitted and received between establishment and termination of a communication unit. Means for identifying a communication method of each communication unit;
The learning means constructs a unique communication quality dictionary for each communication method,
12. The communication quality estimation apparatus according to claim 1, wherein the quality estimation unit estimates a throughput by applying a time series of communication rates of each communication method to a corresponding communication quality dictionary. . The learning means constructs a communication quality dictionary unique to LTE,
13. The communication quality estimation apparatus according to claim 1, wherein the quality estimation unit estimates a throughput by applying a time series of LTE communication rates to a communication quality dictionary unique to the LTE. .