**4.1 Contention based bandwidth access**

The IEEE 802.16 standard supports a mandatory Point-to-Multipoint (PMP) architecture operating in Time Division Duplex (TDD) mode. In such network conditions, the frames are divided to downlink (DL) and uplink (UL) subframes. The BS transmits uplink map (UL-

MAP) messages at the beginning of the DL subframe, in order to schedule the uplink traffic from the subscriber stations (SSs) to the BS. The beginning of UL subframe contains Information Elements (IEs) dedicated for initial ranging and bandwidth request procedures, followed by slots for the actual data transmission. The MAC layer of IEEE 802.16 specifies the rules for the contention-mode bandwidth request procedure. A contention period, as mentioned, is allocated at the beginning of the uplink subframe. It is divided into an integer number of transmission slots and is called an information element. Each transmission slot can be used for a transmission of only one bandwidth request. The SSs use the contention slots to send bandwidth request messages. If a SS's request message transmission is successful, the BS grants contention-free data transmission slot for that particular SS in one of the following frames by placing the SS's Connection ID (CID) in the UL-MAP message.

If more than one SS tries to transmit its request in the same transmission slot, a collision happens. Since it is not practically possible for SSs to sense the UL channel and to detect a collision, the SSs can only know of the success of their bandwidth request transmission if they receive a response in the form of a bandwidth grant from the BS in the subsequent frames. A subscriber station that does not receive a response to its bandwidth request by a certain deadline assumes that either a collision happened or resources are not available at the BS. In either case, since the SS can not determine the cause, it assumes that a collision happened and uses an exponential binary back-off procedure to resolve the collision. In particular, the SS starts a contention-based procedure by setting a so called initial backoff window which is an integer number. Next step is selection of a random number within the window, which determines the number of contention slot for which the SS will defer its next request message transmission. Only the slots for which the SS is eligible to send are counted. When the SS's counter reaches zero, the SS sends its request message. The SS considers the contention transmission as lost if no data grant has been given within the period of time defined by a timer. Then the SS enters in the next stage of the backoff algorithm by doubling the size of the backoff window and selecting another random number. This repeats with each loss of the request massage, until the backoff window size reaches its maximum size.

#### **4.2 Multicast-grouping based bandwidth access**

The BS controls the access rights of the SSs for each contention slot. If the BS declares one contention slot as a *broadcast* type of slot, then all SSs have the right to transmit their bandwidth request messages in that particular slot. Contrary to this principle, the BS can mark certain slot as a *multicast* type of slot. In this case only the members of the specified multicast group can access the slot. The BS controls the membership of each SS into multicast groups. For this purpose, it uses a special MAC message called MCA-REQ (Multicast Polling Assignment Request). Each MCA-REQ message contains three basic parameters: the basic CID of the SS, the index of the multicast group, and one of the two possible commands, *join* or *leave*. One SS can belong to several multicast groups.

In order to evaluate the influence of multicast-group implementation over the system's performance, we will calculate the ratio of the number of successful bandwidth request transmissions per frame and the number of active SSs (*maxNsuc/n*). The value of this ratio *maxNsuc/n* = 1 means that all active SSs are served in one TDD frame. The following figure (Fig. 4) presents the results regarding *maxNsuc/n* for different number of active SSs (*n*) and

MAP) messages at the beginning of the DL subframe, in order to schedule the uplink traffic from the subscriber stations (SSs) to the BS. The beginning of UL subframe contains Information Elements (IEs) dedicated for initial ranging and bandwidth request procedures, followed by slots for the actual data transmission. The MAC layer of IEEE 802.16 specifies the rules for the contention-mode bandwidth request procedure. A contention period, as mentioned, is allocated at the beginning of the uplink subframe. It is divided into an integer number of transmission slots and is called an information element. Each transmission slot can be used for a transmission of only one bandwidth request. The SSs use the contention slots to send bandwidth request messages. If a SS's request message transmission is successful, the BS grants contention-free data transmission slot for that particular SS in one of the following frames by placing the SS's Connection ID (CID) in the UL-MAP message. If more than one SS tries to transmit its request in the same transmission slot, a collision happens. Since it is not practically possible for SSs to sense the UL channel and to detect a collision, the SSs can only know of the success of their bandwidth request transmission if they receive a response in the form of a bandwidth grant from the BS in the subsequent frames. A subscriber station that does not receive a response to its bandwidth request by a certain deadline assumes that either a collision happened or resources are not available at the BS. In either case, since the SS can not determine the cause, it assumes that a collision happened and uses an exponential binary back-off procedure to resolve the collision. In particular, the SS starts a contention-based procedure by setting a so called initial backoff window which is an integer number. Next step is selection of a random number within the window, which determines the number of contention slot for which the SS will defer its next request message transmission. Only the slots for which the SS is eligible to send are counted. When the SS's counter reaches zero, the SS sends its request message. The SS considers the contention transmission as lost if no data grant has been given within the period of time defined by a timer. Then the SS enters in the next stage of the backoff algorithm by doubling the size of the backoff window and selecting another random number. This repeats with each loss of the request massage, until the backoff window size

The BS controls the access rights of the SSs for each contention slot. If the BS declares one contention slot as a *broadcast* type of slot, then all SSs have the right to transmit their bandwidth request messages in that particular slot. Contrary to this principle, the BS can mark certain slot as a *multicast* type of slot. In this case only the members of the specified multicast group can access the slot. The BS controls the membership of each SS into multicast groups. For this purpose, it uses a special MAC message called MCA-REQ (Multicast Polling Assignment Request). Each MCA-REQ message contains three basic parameters: the basic CID of the SS, the index of the multicast group, and one of the two

In order to evaluate the influence of multicast-group implementation over the system's performance, we will calculate the ratio of the number of successful bandwidth request transmissions per frame and the number of active SSs (*maxNsuc/n*). The value of this ratio *maxNsuc/n* = 1 means that all active SSs are served in one TDD frame. The following figure (Fig. 4) presents the results regarding *maxNsuc/n* for different number of active SSs (*n*) and

possible commands, *join* or *leave*. One SS can belong to several multicast groups.

reaches its maximum size.

**4.2 Multicast-grouping based bandwidth access** 

different number of transmission slots per frame (*Nr*). The results are obtained by using the analytical equations provided by (Latkoski, 2010).

Furthermore, Fig. 4 provides additional insight regarding the maximization of the bandwidth procedure success rate. It is obvious that instead of using all *Nr* slots for the contention of all *n* subscriber stations, it is better to split the SSs into *M* groups, and to give each group a portion of *Nr/M* slots for contention. The colored lines in Fig. 4, give the possibilities for implementation of this idea. For example, instead of using *Nr* = 16 slots for *n* = 16 users, it is more efficient to use *M* = 8 multicast groups, as the *Nsuc*/*n* is higher for (*n, Nr*) = (2, 2) compared to the case where (*n, Nr*) = (16, 16).

Fig. 4. Normalized success rate for conation-based scheme.

The obvious challenge here is to obtain a precise estimation of the number of active SSs (*n*) by the serving BS, which controls the contention and multicasting.

#### **4.3 Round robin polling bandwidth access**

Instead of using contention based bandwidth distribution among the SSs, the BS has an option to use the round-robin polling based procedure for bandwidth access. In this case, the BS asks each of the registered SSs whether they need bandwidth, starting with the first SS and ending with the last *Nall* SS. Then the circle of polling repeats again from the first SS. Considering that not all *Nall* SSs need bandwidth at a time, but only *n* of them have such need, we can calculate the efficiency of this method through the performance parameter defined as utilization of the slots. We can compute the utilization of the transmission slots in the case of round-robin polling (*RRutil*) as:

$$RR\_{util} = \frac{n}{N\_{all}} \,\prime\,\tag{1}$$

while the average bandwidth access delay seen by the SSs (*RRTd*) is:

*all Td frm r <sup>N</sup> RR t <sup>N</sup>* , (2)

where the *Nr* represents the total number of transmitting slots, and *tfrm* is the TDD frame duration. These simple equations reveal that the utilization in the case of round-robin polling scheme does not depend on *Nr*, while the delay does not depend on *n*. Consequently, this bandwidth allocation scheme is expected to have higher performance in scenarios where the number of active SSs (SSs which need bandwidth) is close to the number of registered SSs.

The previous conclusion can be proven by making a comparison of the transmission slot utilization in the case of round-robin polling and conation based schemes for different numbers of active users. For this purpose we have used the equations provided by (Latkoski, 2010) for the maximal utilization of the transmitting slots provided by contention scheme. The results presented in the following figure are obtained for different values of *Nr* and *Nall*.

Fig. 5. Comparison of the schemes performance.

The range of values where *n ≈ Nall* (right part of the figure) is more appropriate for roundrobin polling scheme utilization, compared to the contention-based scheme.

#### **4.4 SDL models**

For the purpose of analytical results validation, as well as for testing and improvement of the communication protocol for bandwidth allocation, we have created according to the methodology presented in the previous section, both behavior and performance evaluating models. Actually, we have built several behavior models for different MAC-layer processes involved by the communication protocol, located in both base station and subscriber station. After the functional testing of each protocol entity, the behavior models are implemented into fully operational performance model. The highest level of this model is presented in the

where the *Nr* represents the total number of transmitting slots, and *tfrm* is the TDD frame duration. These simple equations reveal that the utilization in the case of round-robin polling scheme does not depend on *Nr*, while the delay does not depend on *n*. Consequently, this bandwidth allocation scheme is expected to have higher performance in scenarios where the number of active SSs (SSs which need bandwidth) is close to the

The previous conclusion can be proven by making a comparison of the transmission slot utilization in the case of round-robin polling and conation based schemes for different numbers of active users. For this purpose we have used the equations provided by (Latkoski, 2010) for the maximal utilization of the transmitting slots provided by contention scheme. The results presented in the following figure are obtained for different values of *Nr*

1 3 5 7 9 11 13 15

The range of values where *n ≈ Nall* (right part of the figure) is more appropriate for round-

For the purpose of analytical results validation, as well as for testing and improvement of the communication protocol for bandwidth allocation, we have created according to the methodology presented in the previous section, both behavior and performance evaluating models. Actually, we have built several behavior models for different MAC-layer processes involved by the communication protocol, located in both base station and subscriber station. After the functional testing of each protocol entity, the behavior models are implemented into fully operational performance model. The highest level of this model is presented in the

robin polling scheme utilization, compared to the contention-based scheme.

*<sup>N</sup> RR t*

number of registered SSs.

0.00

**4.4 SDL models** 

Fig. 5. Comparison of the schemes performance.

0.10

0.20

0.30

0.40

maxUtil

0.50

0.60

0.70

and *Nall*.

*all Td frm r*

*<sup>N</sup>* , (2)

n

Nr = 2 Nr = 4 Nr = 6 Nr = 8 Nr = 10 Nr = 12 Nr = 14 RR (Nall=20) RR (Nall=30) RR (Nall=40) RR (Nall=50) following Fig. 6. It contains a behavior model of the BS which contains several processes: Optimizer, Estimator and MsgCreator. The Optimizer determines the optimal values of the following contention parameters: the initial contention window, the number of allowed consequent unsuccessful bandwidth request transmissions, the number of multicast groups, and the number of contention slots per frame. These parameters are sent to the process which constructs the MAC management messages (MsgCreator), as well as to the process Estimator. The purpose of the Estimator is to estimate several network condition related parameters, such as: the number of active users, the probability of collision, the probability of transmission, etc. These parameters are needed for an accurate operation of the Optimizer.

Fig. 6. Case study Performance model.

The performance model also contains blocks for channel emulation (collisionMng) and simulation control block (SSmng) which provides SSs PID management.

Besides the BS, the performance model contains multiple instances of the subscriber station block. All instances of the SS block operate as independent user equipment stations. Through the SSmng block we are able to define and control the number of active stations in the simulation scenario. The BS through the MsgCreator block controls the mode of operation (contention or round-robin, along with the number of multicast groups), according to the Optimizer commands. The Estimator operates dynamically, and feeds the Optimizer with the necessary information regarding the network conditions. The active stations are sending bandwidth request messages and then register the outcome of every attempt (success or failure).

In this communication protocol engineering case study, the Optimizer is the newly proposed entity which operation will be described in details. Actually, the Optimizer performs several steps presented formally in Fig. 7. These steps are:

Fig. 7. Functional steps of the Optimizer.


#### **4.5 Results**

352 Wireless Communications and Networks – Recent Advances

In this communication protocol engineering case study, the Optimizer is the newly proposed entity which operation will be described in details. Actually, the Optimizer

performs several steps presented formally in Fig. 7. These steps are:

Fig. 7. Functional steps of the Optimizer.

The performance model was tested in simulation scenario where the number of active stations (*n*) is changed with the time, as presented in Fig. 8. The scenario simulates 24 hour network operation. The next two figures (Fig. 9 and Fig. 10) provide the measured performance of the bandwidth request procedure for three different modes of operation: round-robin polling, contention without multicast grouping, and contention with multicast grouping. Fig. 9 presents the transmission slots utilization, while Fig. 10 presents the average bandwidth access delay.

Fig. 8. Number of active SSs during the simulation (actual and estimated).

Fig. 9. Measured utilization of the transmission slots.

Fig. 10. Measured delay.

During the simulations we have used the following parameter values: *Nr* = 10, *Nall* = 150, *q* = 1, *tfrm* = 10ms.

From the results we can conclude that the contention based bandwidth request procedure which uses multicast grouping almost always outperforms the case where no multicasting groups are used. This is the case if our comparison criterion is based on the transmission slots utilization. The same conclusion is not entirely valid if the criterion is based on the average bandwidth access delay. The round-robin mode of operation, as expected, outperforms the other modes of operation when the number of active users is close to the number of registered users.
