Preference Function

We consider two preference functions in the selection of networks. One is from user’s side and the other is from network’s side. For the user, the degree of the satisfaction of QoS can represent for the preference of the user. The preference function for the user corresponding to the candidate network i is denoted by UP , _i which consists of several functions of QoS-related factors.

UP is defined as i

( ) ( ), for the voice or the video call request, ( ), for the HTTP or the best effort call request,

D i P i bargains with one of candidate networks over the acceptation of the delay of packets.

For non-real time service, user will bargain with one of candidate networks over the bit rate that candidate network can afford to allocate.

The design of f b_B( )_i is

28 the minimum transmission rate for this non-real service. For FTP services, the reason of the setting of this value is keeping the session continued. b_i[B B^*, _max^* ] is the maximum preference value 1.5. Besides, this function is defined as the sigmoid curve (S shape) so that when b_i is very large or very small, the increase of b_i will not make user feel much better or worse. Hence, the values of  and  in (4.7) are constants and set to 10 and 10, respectively.

D( )i

29

accepted by network i, respectively, and D^* is the maximum tolerant packet delay for this service request. If d_i is close to D_min, it means that the call request can obtain more resource and the value of f_D( )d_i will be higher, meaning that the value of f_D( )d_i will be close to 2. In this case, user will prefer this network. On the contrary, the value of f_D( )d_i will be close to 1. In the middle of D^* and D_min is corresponding to the half of maximum preference value. The range of d_i for user is

* maximum acceptable packet dropping rate for specific service request. This function is considered only for the real time services because of QoS requirement. In order to guarantee the QoS requirement, if measured average packet dropping rate is larger than the requirement this function will be zero, because this network should not be chosen. When p is much smaller than _i P^*, this function will be close to two. When p is half of i P^*, the preference value will decrease to about 1.5. When p is equal _i to P^*, the value of this function will decrease to one which means the probability of this network cannot guarantee the QoS requirement will be very high. Based on these design principles, the values of  and  in (4.9) are set to 1/9 and 8, respectively.

For candidate network i, the preference function is denoted by NP . The _i

30

concept of NP is the degree of the suitability of the candidate network _i i for the call request, which is defined as

( ) ( ) ( ), for the voice or the video call request, delay, loading intensity increment, and the mobility factor measured at network i for this call request, respectively.

We define g b_B( )_i as

31

Figure 4.2 shows the concept of design, which is adverse to that in Fig. 4.1. Fig. 4.2, candidate network i prefers to allocate the data rate which is close to the minimum transmission rate to the call request to accommodate as many users as possible. When bi is close to B^*, the value of g b_B( )_i is close to 2, which is the maximum preference value of network i. When b_i is close to B^*_max,i, the value of g b_B( )_i is close to 1. The values of  and  in (4.11) are set to 100 and 13, respectively.

This function is considered for non-real time services only.

D( )i packet delay of an observed duration for candidate network i. D^* is the maximum tolerant packet delay for this service request. The range of d_i for network i is

*

min, max,

[D _i, min (D _i,D )]. When d_i is close to D_min,i, network i needs to allocate more resource to guarantee the packet delay, which means the value of g_D( )d_i will be lower. This function is also considered only for the real time services because of the QoS requirement. The design concept of this function is the same as equation (4.11), and the values of  and  in (4.12) are also set to 5 and 8, respectively.

32 threshold for network i, the value of the preference will decrease 1.5 approximately.

If the loading intensity reaches predefined loading threshold, which means the loading of network i is very heavy, the value of this function will decrease to one. For these principles of design, the values of  and  in (4.13) are set to 1/9 and 7, respectively. By these design concepts, the call request can be accepted in the network with light loading and the system can achieve load balancing.

m( )i predefined threshold of the specific service for the network i. If the dwell time of the call request is very long in network i, then x can be much larger than _i x_{th i,} , and the

33

value of the preference will close to two. The value of this function will decrease to 1.5 if x is 1.5 times of _i x_{th i,} . If x is equal to _i x_{th i,} , the preference value will decrease to one. For these principles of design, the values of  and  in (4.14) are set to 150 and 2, respectively. By the design of this function, candidate networks can avoid accepting the call request with too short dwell to decrease the number of handoffs.