KR100412040B1 - System for scheduling dynamic slot assignment using neural network in a wireless communication network - Google Patents

Abstract

According to the present invention, a cell scheduling algorithm using a neural network is applied to a slot allocation scheme in a wireless ATM network, so that slot allocation for each traffic can be efficiently performed adaptively to an irregular data flow of dynamic parameters involved in dynamic slot allocation. To provide a dynamic slot allocation scheduling system using a neural network in a wireless communication network.

To this end, the present invention provides a dynamic slot allocation scheduling system for a wireless ATM network, wherein a plurality of dynamic parameters each having a value for a traffic generation rate change and a queue length change of a buffer measured in each cell's frame period are transmitted. Determining the priority of CBR (Constant Bit Rate), VBR (Variable Bit Rate), and ABR (Avariable Bit rate) traffic based on the priority bits for each traffic according to the dynamic parameters received from the wireless terminal and the plurality of wireless terminals, respectively. In order to determine the priority according to queue length for VBR and ABR traffic, and predict the slot allocation number for traffic generation using neural network according to the traffic generation rate change and buffer queue length change. It is characterized by consisting of a base station for allocating slots according to priority.

Description

System for scheduling dynamic slot assignment using neural network in a wireless communication network

The present invention relates to a dynamic slot allocation scheduling system using a neural network in a wireless communication network, and more particularly, by allowing the slot to be adaptively allocated according to the occurrence of traffic in a wireless ATM (Asynchronous Transfer Mode) network. The present invention relates to a dynamic slot allocation scheduling system using neural networks in a wireless communication network for maximizing the use of network resources while improving the quality of service of a user.

In general, a signaling scheme for assigning dynamic slots in a wireless ATM network includes a method of transmitting dynamic parameters by an in-band signaling scheme and an out-of-band scheme. ) It is divided into a method of transmitting dynamic parameters by a signal method.

In the in-band signaling method, there is an advantage in that the necessary information can be transmitted in a timely manner by piggybacking the data cell transmitted on the uplink, but this has the disadvantage that the amount of information that can be transmitted is extremely limited.

In addition, the out-band signaling method has the advantage of transmitting a lot of information in accordance with the transmission of dynamic parameters. However, in general, since it is necessary to approach a designated signal slot competitively, it is not possible to transmit necessary information in a timely manner. have.

That is, in the case of such a signaling method, each optimization should be performed in association with the limitation of the bandwidth required for signaling and the efficiency of the signaling protocol. In such a traffic-specific dynamic slot allocation method, a dynamic slot assignment (DSA) is performed. / DSA ++), Bandwidth on Demand Fair Sharing Round Robin (BOD-FSRR), Estimated Prorated Slot Assignment (EP-SA), Equivalent Capacity based Dynamic Release Slot Assignment (EC-DRSA), Non-Preemptive Priority (NP) polling have.

However, in the case of DSA / DSA ++ among the dynamic slot allocation schemes, the characteristics and quality requirements of each service traffic are reflected by calculating the priority using dynamic parameters without applying a separate bandwidth allocation scheme for each traffic. As a result, cell loss and delay for VBR traffic having instantaneous traffic generation characteristics occur. In particular, the transmission of dynamic parameters using a specific field has a problem that can result in overhead bit problems of the channel.

In addition, in the case of the EC-DRSA scheme among the dynamic slot allocation schemes, in order to guarantee basic quality of service (QoS) requirements of VBR traffic, such as a peak rate, a mean rate, and a burst, Although the concept of equivalent band is introduced to guarantee the target cell loss rate and the maximum transmission delay time for a given traffic and buffer size, the channel wasted or lost and delayed for VBR traffic having instantaneous traffic generation characteristics. Since it is generated, there is a problem that it is difficult to apply to a traffic characteristic having a variable traffic generation rate.

Accordingly, the present invention has been made to solve the above-mentioned conventional problems, and its object is to apply a cell scheduling algorithm using a neural network to a slot allocation method in a wireless ATM network, and to determine dynamic parameters involved in dynamic slot allocation. The present invention provides a dynamic slot allocation scheduling system using a neural network in a wireless communication network that can efficiently perform slot allocation for each traffic adaptively to an irregular data flow.

1 is a diagram illustrating a configuration of a wireless communication network system to which a dynamic slot allocation scheduling system using a neural network according to the present invention is applied;

2 is a diagram illustrating a configuration of a dynamic slot allocation scheduling system using a neural network in a wireless communication network according to the present invention;

3 is a diagram illustrating a configuration of a neural network of the scheduler shown in FIG. 2;

4A is a view showing a traffic change amount and a buffer queue length change amount measurement state at a wireless terminal according to a preferred embodiment of the present invention;

4B is a diagram showing values of a cell generation rate change and a queue length change of a frame received at a base station according to a preferred embodiment of the present invention;

5 is a flowchart for explaining an operation for slot allocation increase / decrease processing of a base station according to an embodiment of the present invention;

6a and 6b are graphs showing simulation results for constant bit rate (CBR) traffic according to a preferred embodiment of the present invention;

7A and 7B are graphs showing simulation results for Variable Bit Rate (VBR) traffic according to a preferred embodiment of the present invention.

8A and 8B are graphs showing simulation results for ABR (Available Bandwidth Rate) traffic according to a preferred embodiment of the present invention.

<Description of Symbols for Main Parts of Drawings>

10: traffic incidence change measuring instrument, 12: buffer waiting matrix length change measuring instrument,

14: temporary memory, 16: traffic generator,

18: queue monitoring unit, 20: transmission unit,

22: receiver, 24: cell header processor,

26: priority calculator, 28: channel information processor,

30: slot calculator,

32: Scheduler.

In order to achieve the above object, according to the present invention, in a dynamic slot allocation scheduling system of a wireless ATM network, a dynamic value having a value for a traffic generation rate change and a queue length change of a buffer measured in units of frame periods of each cell A plurality of wireless terminals each transmitting a parameter, and a constant bit rate (CBR), variable bit rate (VBR), variable bit rate (ABR) according to priority bits for each traffic according to dynamic parameters received from the plurality of wireless terminals, respectively. After deciding the priority of the traffic, priority is determined by queue length for VBR and ABR traffic, and slot allocation for traffic generation using a neural network is performed according to the change in traffic generation rate and the queue length of the buffer. And a base station for predicting the number and allocating slots according to the determined priority. The.

Hereinafter, the present invention configured as described above will be described in detail with reference to the accompanying drawings.

That is, FIG. 1 is a diagram illustrating a configuration of a wireless communication network system to which a dynamic slot allocation scheduling system using a neural network according to the present invention is applied.

As shown in FIG. 1, according to the dynamic slot allocation scheduling system according to the present invention, a neural DSA algorithm is generated by improving a dynamic slot allocation scheme of VBR traffic in a wireless ATM network and integrating a cell scheduling algorithm. Dynamic parameters involved in dynamic slot allocation allow the slot allocation to be adaptively applied to irregular data flows using the neural network's learning and recognition capabilities.

According to the present invention, by using a neural network, by learning a dynamic parameter representing a change in cell generation rate and a change in queue length, a next change in traffic generation rate and a queue length are generated based on past traffic generation rate and queue length change. The amount of change can be predicted. Here, the QoS of the VBR traffic user is guaranteed by determining the number of slots to be allocated to the next frame using the predicted value, and the channel wasted and prevented due to the equivalent bandwidth allocation, thereby monitoring and adjusting the VBR traffic. Bandwidth allocation scheme can be used.

In the figure, a plurality of wireless terminals MT1 to MTn transmit a dynamic parameter indicating a traffic generation change state and a queue length change to a neural network scheduler of a base station (BTS).

The base station (BTS) uses a neural network scheduler to perform priority control for each traffic and to detect idle slots of channels, thereby rapidly generating traffic at the wireless terminals MT1 to MTn, thereby increasing the length of the queue. Increasing, it is possible to proceed the adaptive and intelligent traffic control to reduce the loss rate while suppressing the increase in cell delay of each wireless terminal.

That is, in the present invention, the traffic generation and the queue length change are observed through the dynamic parameters representing the traffic characteristics of each of the wireless terminals MT1 to MTn, and the number of slots to be allocated to the next frame using the neural network is determined according to the result. By the determination, real-time control of the wireless terminal is made possible.

2 is a diagram illustrating a configuration of a dynamic slot allocation scheduling system using a neural network in a wireless communication network according to the present invention.

As shown in FIG. 2, the wireless terminal MT includes a traffic generation rate change meter 10, a buffer queue length change meter 12, a temporary memory 14, a traffic generator 16, and a queue monitor. 18 and a transmission unit 20.

The base station (BTS) also includes a receiver 22, a cell header processor 24, a priority calculator 26, a channel information processor 28, a slot operator 30, and a scheduler 32.

In the figure, the dynamic parameter of the wireless terminal MT uses an ATM cell header portion and a GFC field (4 bits) with a change in cell generation rate measured in frame periods and a change in queue length of a buffer. To transmit in in-band signaling.

In the wireless terminal MT, the traffic occurrence rate change measuring unit 10 performs a process of measuring a traffic change rate for each frame period of data and encoding the data into two bits, as shown in FIG. 4A. Measuring the number of cell occurrences (g _i-1 ) during the i-1th frame period in the period, and if the number of cells generated in the i-1th frame becomes larger than the number of cells generated in the i-2th frame, the first constant (a ₀ ). Is set to "1" and the second constant a ₁ is set to "1" when the value of the number of cells becomes equal to or greater than a predetermined threshold value Δ (ie, twice the average number of requested slots).

The relationship between the traffic change rate and the buffer queue matrix length change rate using the traffic generation rate change meter 10 is expressed by Equation 1 below.

a0 = a1 = 0

In the figure, the buffer queue length change measuring unit 12 performs the same measurement as the traffic generation rate changing measuring unit 10, in which the number of cells generated during the i-1th frame period is the previous frame (i-2). If the length of the buffer increases as the number of cells generated by increases, the first constant (b ₀ ) bit is set to "1", and the increase width of the buffer length is set to a predetermined threshold value Δ (ie, a general buffer). When the size exceeds 0.7 times, the second constant b ₁ is set to "1".

Here, the relationship between the traffic change rate and the buffer queue length change rate measurement using the buffer queue length change measurer 12 is shown in Equation 2 below.

b0 = b1 = 0

In the figure, the temporary memory 14 is for storing the number of cell occurrences in the previous frame and the queue length value of the buffer.

The traffic generator 16 generates a uniform pulse according to the traffic change rate and the buffer queue matrix change of the traffic generation rate change meter 10 and the buffer queue matrix length change meter 12. It comprises a uniform traffic generator for generating a Poison pulse (Poission Pulse).

The uniform traffic generator includes a CBR traffic generator having a mean burst size of "1" and a mean inter-pulse time maintained at a constant interval, and the Poison traffic generator includes: The average burst size is " 1 " and the average inter-pulse duration consists of a VBR traffic generator according to the Poisson distribution.

In addition, the transmitter 20 generates a slot request cell when a cell is generated in the traffic generator 16 under queue monitoring by in-band information from the queue monitor 18. In addition to transmitting the cell through the access slot, the data cell is transmitted through the data slot while the reservation is made by the requesting cell.

In this case, when the slot 20 transmits the slot request cell, the transmitter 20 transmits the average request slot information and the parameters of the traffic type by the random access method, and in order to prevent the loss due to the collision of the random access. The cell is stored in a temporary buffer so that retransmission can occur if a collision occurs.

In addition, in order to prevent time delay during transmission of the slot cell, the transmitter 20 sets a threshold value (that is, three times) of the number of collisions and discards the cells of the temporary buffer when the number of collisions exceeds the threshold value. When a data cell is transmitted, a dynamic parameter representing a cell generation rate change and a queue change is piggybacked to the last cell of the allocated slots per frame.

As shown in FIG. 2, the receiving unit 22 transmits an acknowledgment signal for a slot request cell transmitted from the wireless terminal MT in the base station BTS, and transmits an acknowledgment signal to the wireless terminal MT from the base station BTS. As data is transmitted, it checks whether there is an available slot by checking a collision state of a slot request cell transmitted through a random access slot, and accepts or rejects a reservation request according to the available slot. Will be performed.

In this case, when the reception unit 22 initially accepts the reservation request for the wireless terminal, it is determined whether the reception unit 22 accepts the request according to the average band request transmitted by the wireless terminal, and data transmission is performed according to the determined average band. The slot allocation number for every frame can be adjusted according to the dynamic parameters from the wireless terminal.

In the same figure, the cell header processor 24 is for processing the information of the cell header for the slot request cell received through the receiver 22.

In addition, the priority calculator 26 classifies priorities based on priority bits for each traffic and performs two steps of control for classifying priorities by the length of the queue. In the classification step of priority based on bits, priority is determined for each traffic of CBR, VBR, and ABR, and then, slots are allocated by using a fixed allocation method for CBR traffic, and standby for VBR and ABR traffic. Slots are allocated by determining the priority according to the matrix length. In the ABR traffic, slots are allocated using an extra slot after slot allocation for CBR and VBR traffic is made.

The slot operator 30 monitors the number of slot distributions per frame based on the channel state information of the channel information processor 28, calculates the remaining spare slots, and transfers the spare slots to VBR traffic in the next frame. For use in allocation.

In addition, the scheduler 32 determines the number of slots to be allocated to the next frame by determining the increase or decrease of the slot allocation through the neural network using the dynamic parameters of the wireless terminal MT. As shown, a neural network 40 that receives in-band information, channel state information, and first bit information to determine an increase or decrease of slot allocation, and a pattern in which a weight value of the neural network 40 is stored. It consists of a table 42.

Here, the neural network 40 plays a role of determining the increase / decrease of slot allocation for input patterns using a multilayer perceptron using an error backpropagation learning algorithm. That is, when the neural network 40 is used, as the learning is performed on the traffic in which the cell generation rate change and the queue length change occur for all the wireless terminals, the slots are flexibly changed according to the generation change of the VBR traffic. Being able to adjust the number of allocations reduces bandwidth waste and prevents cell delays and losses in VBR traffic.

In the neural network 40, as shown in FIG. 4B, a total of nine pieces of information data are input. As the input values, traffic generation rate changes and queues during the past three frame periods transmitted as dynamic parameters of the wireless terminal. A value indicating a change in length, an increase in the number of slots (that is, (number of slots allocated in frame i)-(number of slots allocated in frame i-1)) will be given. The slot allocation is adaptively predicted by predicting the number of slot allocations generated by traffic of the wireless terminal.

Next, the operation of the present invention made as described above will be described in detail with reference to the flowchart of FIG. 5.

First, when a new data packet is generated as a new traffic is generated through the traffic generator 16 in a predetermined wireless terminal among the plurality of wireless terminals MT1 to MTn and is transmitted through the transmitter 20 (step S10), The base station (BTS) receives the traffic data through the receiving unit 22 and determines whether the reservation request information is included (step S11).

As a result of the determination, when the base station (BTS) determines that the reservation request information is included in the traffic data received from the wireless terminal through the receiving unit 22, the base station (BTS) of the traffic cell by the cell header processor 24 In the state where processing is performed and priority calculation and slot allocation for each traffic are performed by the priority calculator 26, it is determined whether there is an idle slot according to the idle slot monitoring process by the slot calculator 30. (Step S12).

As a result of the determination, when the slot operator 30 determines that there is no idle slot for the received traffic cell, the reservation request is rejected (step S13).

On the other hand, when the slot operator 30 determines that there is an idle slot for the traffic cell, the base station (BTS) schedules the in-band information, channel state information, and priority bit information. As provided in FIG. 5, the average bandwidth is allocated through prediction of the slot allocation number by the neural network 40 of the scheduler 32 and adaptive allocation of the slots (step S14).

On the other hand, if it is determined in the determination result of step S11 that the base station (BTS) does not include the reservation request information in the traffic data received from the receiver 22, the in-band signaling information is received from the wireless terminal. In step S15, the neural network 40 of the scheduler 32 receives in-band signaling information, channel state information, and priority bit information, according to a change in traffic generation rate and a change in queue length for a predetermined frame. It is determined whether or not the slot is increased or decreased (step S16), and the number of slots is increased / decreased according to the determination result to perform adaptive slot allocation (step S17).

Next, the scheduler 32 of the base station (BTS) is increased / decreased as a result of predicting the number of slots of the neural network 40, and the allocated slot value is added to the average bandwidth and output through the output link ( Step S18).

Next, the simulation results of the model according to the characteristics for each traffic will be described in detail with reference to FIGS. 6A and 6B and FIGS. 7A and 7B, 8A and 8B.

That is, in the present invention, the model according to the traffic characteristics is assumed to be three types of uniform traffic, Poison traffic, and burst traffic. The uniform traffic generator has an average burst size of "1", and the average inter- Assume that the CBR traffic terminals allow the pulse periods to be maintained at regular intervals, while the Poisson traffic generator assumes that the average burst size is "1" and the average inter-pulse periods are VBR traffic terminals that conform to the Poison distribution. The burst traffic generator is composed of a burst section in which pulses are generated per unit time and a silent section in which pulses are not generated. It is assumed that the burst traffic generator is an ABR traffic terminal as a source model such as file transmission.

Each wireless terminal has a single buffer, and the access method uses the TDMA / TDD method, and the conventional DSA ++ / TDD method for comparison with the dynamic slot allocation algorithm (Neural DSA) to which the neural network of the present invention is applied, By configuring the BoNeS tool of MASCARA and Non-Neural Dynamic Slot Allocation Algorithm, we can compare the performance of cell loss rate, delay and throughput of VBR traffic.

However, the number of wireless terminals uses 5 CBR traffic terminals, 5 VBR traffic terminals, and 6 ABR traffic terminals. The average number of slot requests (Load = 0.5) is the same for all terminals. Use a fixed length of 2ms.

On the other hand, Table 1 below shows the simulation parameters used in the experiment.

Parameter value Setting value Average Slot Requests (Load) 0.3-0.7 Channel Capacity 21,200,000 Frame Length 2 ms Number of Terminals CBR (5), VBR (5), ABR (6) Learning Rate 0.35 Buffer Size CBR (20), VBR (50), ABR (100) Slot Size 1 ATM Cell Simulation Time Frame * 10000

In the present invention, the buffer size is set according to the cell loss rate according to the size change of the buffer (i.e., 10 ^-4 to 10 ^{-6 for} CBR traffic terminals, 10 ^-6 to 10 ^{-9 for} VBR traffic terminals, and for ABR traffic terminals. 10 ^-9 to 10 ^-12 ) should be taken into consideration.

Meanwhile, when the simulation is performed by applying the DSA ++ / TDD method, the MASCARA method, the non-neural dynamic slot allocation algorithm (Non-Neural DSA) method, and applying the dynamic slot allocation algorithm method applying the neural network according to the present invention. For CBR traffic, as shown in FIGS. 6A and 6B, since the priority of each traffic is the highest for CBR traffic, QoS for the average number of requested slots (Load = 0.5) can be satisfied without any difference. have.

However, in the case of DSA ++ / TDD, priority is given only by queue length, not by traffic priority, resulting in greater cell loss and delay in CBR traffic than other algorithms.

That is, in the dynamic slot allocation algorithm to which the neural network of the present invention is applied, QoS can be guaranteed for CBR traffic in the same manner as other algorithms except for the DSA ++ / TDD algorithm.

In addition, according to the cell delay and loss rate for the VBR traffic, as shown in FIGS. 7A and 7B, in the system of the dynamic slot allocation algorithm using the neural network according to the present invention, slots are adaptively allocated according to generation of VBR traffic. As a result, it can be seen that the cell delay or loss rate in the average number of requested slots (Load = 0.5) is improved compared to other algorithms. That is, not only can QoS be improved for VBR traffic with irregular traffic generation, but also QoS of CBR traffic can be guaranteed.

Accordingly, in the dynamic slot allocation algorithm using the neural network of the present invention, the neural network adaptively adjusts and allocates slot allocation numbers according to generation of VBR traffic, so that cell delay and loss rate for VBR traffic are higher than those of other algorithms. Results in improvement.

On the other hand, in the case of cell delay and loss rate for ABR traffic, after allocating slots of CBR traffic and VBR traffic without applying a separate band allocation algorithm, the remaining bandwidth is used, as shown in FIG. 8A, a neural network. In the system using the dynamic slot allocation algorithm, the delay is high, whereas in the system using the DSA ++ / TDD algorithm, the queue length of the buffer is given priority, so the delay or loss rate is lower than that of other algorithms. . However, if it is assumed that ABR traffic is loss-sensitive traffic as shown in FIG.

It is a matter of course that the present invention having the above-described embodiments can be variously modified and implemented without departing from the gist of the present invention without departing from the gist thereof.

As described above, according to the present invention, even though the occurrence of traffic is irregular and instantaneous by the VBR traffic dynamic slot allocation algorithm using the neural network, the slot can be adaptively allocated to the change, thereby averaging the VBR traffic. Compared to the MASCARA algorithm that allocates bandwidth, the efficiency of channel use can be improved, and the cell loss rate and delay characteristics according to QoS of VBR traffic can be improved. The real-time Internet service or multimedia service can have a smoother and better effect.

Claims (5)

In the dynamic slot allocation scheduling system of a wireless ATM network, A plurality of wireless terminals each transmitting a dynamic parameter having a value for a change in traffic generation rate and a change in queue length of a buffer measured in units of frame periods of each cell; After determining the priority of the constant bit rate (CBR), variable bit rate (VBR), and variable bit rate (ABR) traffic based on the priority bits for each traffic according to the dynamic parameters received from the plurality of wireless terminals, the VBR, Determining priority by queue length for ABR traffic, and predicting slot allocation number for traffic generation using neural network according to traffic generation rate change and buffer queue length change, and according to the determined priority Dynamic slot allocation scheduling system using a neural network in a wireless communication network, characterized in that the base station for allocating slots. 2. The apparatus of claim 1, wherein each wireless terminal measures a change rate of traffic at every frame period of a traffic cell and encodes it into a specific bit, and measures a change in length of a buffer waiting matrix of the traffic cell generated at every frame period. A buffer queue length change measuring device for encoding a specific bit to generate a specific bit, a temporary memory for storing the number of cell occurrences and queue length values of a previous frame, and generating traffic according to the measurement results of the traffic generation rate change measuring device and the buffer waiting matrix length measuring device. Dynamic slot allocation scheduling system using a neural network in a wireless communication network, characterized in that it comprises a traffic generator and a transmission unit for transmitting the traffic cells generated by the traffic generator with a dynamic parameter. The mobile station of claim 2, wherein the base station comprises: a receiver for receiving a traffic cell and a dynamic parameter transmitted from each of the wireless terminals, a cell header processor for processing a cell header of the received traffic cell, and traffic for each of the received traffic cells A priority operator for classifying priorities, a slot operator for allocating idle slots by monitoring the number of slot distributions per frame of traffic cells according to channel state information, and a cell generation rate and queue length change of the dynamic parameters. A dynamic slot allocation scheduling system using a neural network in a wireless communication network, characterized in that it comprises a scheduler for dynamically adjusting the number of slot allocation by learning using a learning algorithm of the neural network for the value. According to claim 2 or 3, wherein the transmission unit of each wireless terminal to form a slot request cell to transmit a traffic cell through a random access slot, the data cell is a data slot in the state that is reserved by the request cell Via The receiving unit of the base station checks the collision state of the slot request cell transmitted through the random access slot, and determines the acceptance of the reservation request according to the existence of the available slot, the dynamic slot using a neural network in a wireless communication network Allocation Scheduling System. delete