Confirmation timeout (Round Trip Time, RTT) is an important value in data networks. Correct prediction of this characteristic allows us to estimate the network load in order to adequately select packet sending and retransmission parameters. Approximate heuristic models are used in the Transmission Control Protocol (TCP) for Round Trip Time evaluation. The values of the coeficients in these models were obtained experimentally for fixed topology networks. Therefore, the use of these models in a dynamic topology network (mobile ad-hoc network) is ineficient. This article represents an RTT prediction model based on application of fuzzy neural network theory. This model relies upon zeroorder Sugeno-Type Fuzzy Inference algorithm. The input values of fuzzy neural network are RTT values measured in the current and two previous cycles. The output value is RTT value expected in the next cycle. The proposed model is set up and examined by means of simulation experiments. In these experiments the functioning of mobile ad-hoc network which is used for communication software while counteracting emergencies was simulated.
Introduction
Mobile ad hoc networks (MANET) are a promising direction in the development
of telecommunication technologies [
The process of information exchange in a mobile ad hoc network is based
on the implementation of the packet data transfer. One of the important
characteristics in this case is RTT. Correct prediction of this value allows us to
estimate the network load to adequately select the parameters of packet sending
and retransmissions. Approximate heuristic models are used in the TCP for its
evaluation [
Development of Neuro-Fuzzy Model The neuro-fuzzy model is suggested to predict RTT in an ad hoc network. The following values are used in this model: is RTT value measured in the current cycle; 1 is RTT value measured in the previous cycle; 2 is RTT value measured in the cycle preceding the previous one. The model allows us to calcycles. culate the estimated value ̃︁
of the confirmation timeout for each of the next
Construction of the model is carried out on the criterion of minimal
complexity. The following parameters correspond to this criterion: fuzzy inference
algorithm is the zero-order Sugeno [
( 2 = 1), ℎ ( 1 = 1)
( 2 = 2), ℎ ( = 2) ( 1 = 2)
( 2 = 2), ℎ
2; 1 ... 8 are individual conclusions of the fuzzy rules.
Type and parameters of the membership functions for each input value are shown in Fig. 1, Fig. 2 and Fig. 3.
Mathematical and Information Technologies, MIT-2016 | Information technologies 1 µ1(M) µ2(M) Aggregation procedure is performed by the second layer of neurons: 1 = 1( ) ∧ 1( 1
) ∧ 1( 1); 2 = 1( ) ∧ 1( 1) ∧ 2( 1);
...
8 = 2( ) ∧ 2( 1) ∧ 2( 1).
The model of forecasting Round Trip Time includes four structural neural layers. Fuzzification procedure is performed by means of the first layer of neurons: (4) (5) (6) (7) (8) (9) (10) (11) (12) (13)
Activation is a part of the defuzzification procedure. Calculation of the amount of aggregated results ∑︀ 8 =1 and the weighted sum of the aggregate results ∑︀ 8=1 are performed by means of the third layer of neurons.
The final part of defuzzification procedure is performed by means of the fourth layer: ︁̃ = ︀∑ 8=1 ︀∑ 8=1 .
In order to obtain the coefficient values needed to calculate membership
function, it is required to set the weights of neurons of the first layer. Training
of the neurons of the third layer is needed for evaluating the values of individual
fuzzy rules conclusions [
Modeling Information Streams Transmission Let us consider an example in which mobile ad hoc network is used for communication in the construction of underground facilities. Fig. 4 and Fig. 5 show the area of the construction works (limited by bold dotted line).
2 4 7 1 8 5 3 + 6 9
This building belongs to the dangerous construction projects, because works on its construction are carried out in the conditions of a possible collapse of rocks. Works are carried out by a personnel shift which consists of: 1) head of the shift who uses ad hoc node 1; 2) eight workers equipped with ad hoc nodes with numbers 2-9.
Ad hoc units are denoted by small numbered circles, and coverage areas of these units are limited by the corresponding circles of larger radius. The following functions are performed by means of ad hoc nodes:
1) video streams to monitor the status of the facility, conditions and the course of the work;
2) exchange of voice messages to control the construction process and the coordination of countering emergencies;
3) transfer of data on the functional status and current location coordinates of the builders, as well as data of monitoring external conditions on the construction site.
In the given example the transmission of information streams is carried out in an ad hoc network for a period of time of observation which lasts 50 seconds. The characteristics of the streams are represented in Table 1 and Table 2.
Fig. 4 shows the situation where the network topology remains unchanged during the considered time interval. The routes of information streams trans4 2 7 mission correspond to the broken lines which connect the nodes-senders and nodes-recipients.
Fig. 5 shows a scenario where an ad hoc network topology changes due to the collapse of rock, which began at time =4 s. Collapse zone is highlighted in gray. As a result of emergency workers who used ad hoc nodes 6 and 9 were in the collapse zone and node 6 malfunctioned (corresponding circle in Fig. 5 is crossed).
In response to the collapse, workers with ad hoc nodes 3-5, 8 and 9 have moved. The locations of these nodes at the initial time in Fig. 5 are marked by the dashed circles. In the modified network structure in Fig. 5 routes that transmit information streams (numbered 2, 3, 5–8) differ from the corresponding streams marked in Fig. 4.
Dynamism of network topology had an impact on the radio channels workload and throughput available for transmission of data flows. For example, a radio channel connecting the node 2 to node 1, except for the main streams of 1, 4 and 6, additional streams 2 and 8 started to transmit.
The responsiveness of the node 1 receiving the data file transmitted from the node 9 is of great importance in an emergency. This file contains information about the current parameters of health status and location of the worker who has been exposed to the collapse. On the basis of the data head of the shift can quickly and effectively coordinate the actions of other workers to rescue the injured builder.
For the file to be delivered, streams 4 and 5 need to be transferred. The
combination of these interrelated streams is called a controlled flow (CF) [
Node 9 (Sender) hannel 6
C C hannel 1
Node 5
Node 2 Channel 5
Channel 2 Node 5
Node 2
Duration of data file delivery from node 9 to node 1 is directly dependent on
the value ( ), the current CF-circuit throughput available for CF transmission.
To calculate this value, one should use the expression:
( ) =
{ ( )},
where ( ) is the current value of the channel throughput of the CF-circuit
[
The value ( ) can be figured out from the formula: ( ) = ︃{ 0, − ( () ) , ( ) < ; ( ) ≥ ; where is the throughput of the radio channel; ( ) is the current value of the channel throughput required to transmit real-time streams; ( ) is the number of data streams, having to be transmitted over the channel by the time , ( ) ≥ 1.
The value ( ) can be determined using the following expression: ︁∫
= ( ) , where is the size of the message transmitted by data flow. (14) (15) (16) (17) (18) (19) ( ) =
( ), ︁∑ =1 where ( ) is the current value of the channel throughput required for realtime stream transmission ; is the number of real-time streams, which need to be transmitted on the CF-circuit channels.
The value of ( ) can be found from the formula: ( ) = ︂{ ,
0,
<
≤ <
;
or ≥
,
is required on real-time stream channel ;
beginning and the end of transmission of real-time stream .
and
where is the value of the bandwidth of the channel required to transmit
real-time stream [
The minimum possible duration of the CF transmission can be determined using the following formula: =
− , where stream transmission without packet loss and an ideal correspondence between the intensity of sending data of the stream and the bandwidth of CF-circuit is starting time of CF transmission; is closure time of CFThe value
this, use the formula: available for the transmission.
is calculated on the basis of the obtained values ( ). To do E(t), Kb/s 400 300 200 100 1 2 3 4 1 2 3 4 5 6 , bit/s 256 128 128 256 =1 0 0 1 0 0 0 E(t), Kb/s 800 600 400 200 12 14 16 18 20 22 24 t, s
To calculate the function ( ) in the case shown in Fig. 4 and Fig. 5, we used data contained in Table 3 and Table 4.
Using these inputs, the function ( ) is calculated and its form is shown in Fig. 7. The fixed network topology function ( ) has the form shown in Fig. 8.
Analysis of Fig. 7 and Fig. 8 shows that the change in the network topology during information exchange leads to a significant deceleration of data file transmission duration. In a network with a dynamic topology the value is set to 39.7 s, and in case of a fixed network structure it is =24.2 s. 4
Setting Parameters of the Model and the Evaluation of the Efectiveness of Its Application In real operating conditions an ad hoc network overload and packet loss frequently occur, so the actual value of the data file transfer duration can significantly exceed the calculated value . To evaluate these characteristics a number of simulation experiments were made, in which various scenarios of applying an ad hoc network for providing connectivity on dangerous construction sites were simulated. For this purpose, a simulation model of information streams transmissions in a network with dynamic topology was used. It was developed in MatLab Simulink software environment. The simulation results provided evidence for setting developed neuro-fuzzy model forecasting round trip time. On the basis of these data the training matrix of the following form is made: ⎛ ⎜⎜⎜⎜⎜⎜ ⎝ 1 2
. .
2 3 .
3 4 .
5 ⎟ ( +1) ( +2) (.+3) ⎟⎟⎟⎟ . . . ⎟⎠
4 ⎞ ( −3) ( −2) ( −1) (20) where is a round trip time of confirmation in the loop ; is the number of cycles in each simulation experiment, =750.
Setting neuro-fuzzy model was carried out using software tools Fuzzy Logic Toolbox. Table 5 shows the results of training the neurons of the first layer, while Table 6 contains the results of training the neurons of the third layer.
To assess the efficiency of the developed and customized models, a number of simulations for the transfer of information streams in an ad hoc network were conducted. The selection of retransmission was simulated on the basis of the suggested neuro-fuzzy forecasting RTT and the classical model of evaluation of this quantity used in TCP. The results showed that the use of neuro-fuzzy forecasting RTT in a large network load reduces deviations of timeout retransmission on 5,7-19,2 percents. This contributes to minimizing retransmissions count and average data stream transmission time by 4.2-9.6 percents. Thus, the model of neuro-fuzzy prediction of confirmation timeout in the mobile ad hoc network is synthesized. The model includes four neuron layers, performing fuzzy inference procedure (fuzzification, aggregation, revitalization and defuzzification). To adjust the weights neurons we used training data, reflecting the dynamics of the RTT in the ad hoc network used for communication on dangerous construction sites. Simulations have shown that the use of the proposed model for selecting timeout retransmission will significantly reduce the duration of the transmission data flows in the mobile ad hoc network.
Acknowledgements. The research have been carried out with the financial support of the Ministry of Education and Science of the Russian Federation (the unique identifier of the project is RFMEFI57815X0138).