The proposed ANHA

ANHA is developed based on one of the most successful ACO algorithms, which is called AntNet [75]. AntNet was developed for route optimization in communication networks. The features of AntNet have been discussed in chapter 2. ANHA is an adaptation of AntNet into the handover scenario. In AntNet, the source node and the destination nodes are usually static nodes;

however, in this study one of the nodes, which is the MH, is moving and the number of path between the source node (CH) and the destination node (MH) will change as the MH moves from one network to another. Hence, in adapting the algorithm of AntNet in this work some modifications are needed.

In AntNet ant agents are deployed periodically and these ants search for the path from the node sending the ant to some specific destination. However, in this case, the available path is limited to the number of NIC available on the MH and also the number of network currently available to the MH. Thus in the proposed approach on implementing ACO concept, the ants are modified to be directed ants, where in the previous section, the RA are used to request for duplication. Then, the DA will affect the change of pheromone in each path available to the MH.

The level of pheromone will then determine the path selected by the MH.

There are several crucial AntNet components that are adapted to ANHA. These components are:

1. The Local Traffic Model

2. The Path Goodness Measurement

3. The Handover Decision table updating rule 4. The Local Decision Policy

5. The evaluation of the DA’s path

84 The Local traffic Model: this component is the local estimate of the network status from the MH’s point of view. The information processed here plays an important role in the evaluation of the ant traversed paths and in the statistical filtering of the ant collected information. This component implements a moving exponential estimate of mean and dispersion (basically the Jacobson’s Algorithm [116]) for the observed ant trip time. The ant trip time concept from AntNet here is adapted to the MOS information collected from the DA. In ANHA, this component actually estimates the local MOS estimate from the MH’s point of view. This component is calculated using equations (5.9) and (5.10). μMOS and σMOS2 are the mean and the variance of MOS respectively. η is the sampling rate used to calculate the mean and variance of MOS. For example, if η = 0.1, means that only the last 50 samples of MOS will affect the value of the mean and variance of MOS. These equations are adapted from the AntNet local traffic model as shown in table 5.2. In the original AntNet model the round trip time (rtt) is used to measure the traffic condition of the path between the source node and the end node. In this research, rtt alone is considered insufficient since it does not portray the QoE of the measured path. Thus, MOS is opted to replace the rtt.

𝜇_𝑁 ← 𝜇_𝑁+ 𝜂(𝑀𝑂𝑆_{𝑐𝑢𝑟𝑟𝑒𝑛𝑡}− 𝜇_𝑁) (5.9)

𝜎_𝑁² ← 𝜎_𝑁² + 𝜂((𝑀𝑂𝑆_{𝑐𝑢𝑟𝑟𝑒𝑛𝑡}− 𝜇_𝑁)²− 𝜎_𝑁²) (5.10) Table 5.2 ANHA adaption from AntNet local traffic model.

AntNet local traffic model:

𝜇_𝑁 ← 𝜇_𝑁+ 𝜂(𝑟𝑡𝑡_{𝑐𝑢𝑟𝑟𝑒𝑛𝑡}− 𝜇_𝑁)

𝜎_𝑁² ← 𝜎_𝑁²+ 𝜂((𝑟𝑡𝑡_{𝑐𝑢𝑟𝑟𝑒𝑛𝑡}− 𝜇_𝑁)²− 𝜎_𝑁²)

ANHA adapted local traffic model:

𝜇_𝑁 ← 𝜇_𝑁+ 𝜂(𝑀𝑂𝑆_{𝑐𝑢𝑟𝑟𝑒𝑛𝑡}− 𝜇_𝑁)

𝜎_𝑁² ← 𝜎_𝑁²+ 𝜂((𝑀𝑂𝑆_{𝑐𝑢𝑟𝑟𝑒𝑛𝑡}− 𝜇_𝑁)²− 𝜎_𝑁²)

The Path Goodness measurement: The goodness of the path between MH and CH cannot be obtained directly, but should be dependent on the status and the changes in each path. In ANHA, the value of MOS is associated to a goodness measure of r where the value of r is between 0 and 1. r can be calculated using equation 5.11.

𝑟 = 𝑐₁(𝑀𝑂𝑆_{𝑐𝑢𝑟𝑟𝑒𝑛𝑡}

𝑀𝑂𝑆_{𝑏𝑒𝑠𝑡} ) + 𝑐₂( 𝑀𝑂𝑆_{𝑏𝑒𝑠𝑡}− 𝐼_𝑠𝑢𝑝

(𝑀𝑂𝑆_{𝑏𝑒𝑠𝑡}− 𝐼_𝑠𝑢𝑝) + (𝑀𝑂𝑆_{𝑏𝑒𝑠𝑡}− 𝑀𝑂𝑆_{𝑐𝑢𝑟𝑟𝑒𝑛𝑡})) (5.11) Isup is calculated using equation 5.12:

𝐼_𝑠𝑢𝑝 = 𝜇_𝑀𝑂𝑆+ 𝑧 (𝜎_𝑀𝑂𝑆

√𝑊) (5.12)

85 Where z determines the confidence interval of the Isup, while W is the sliding window size, meaning how many samples of MOS measurement taken for the calculation. A sigmoid filter processes the value of r, so that the values at the highs and lows of r are retained. The sigmoid equation used here is shown by equation 5.13.

𝑟 ← 1

1 + 𝑒^{−16×𝑟+8} (5.13)

The value of z depends on the z-table as shown in table 5.3. For example, when considering 95%

confidence interval, the value of z will be 1.65.

Table 5.3 The z-table

ANHA Path Goodness Measurement equations are adapted from the AntNet model as shown in table 5.4. Notice that the r equation in AntNet, since lower rtt is better while higher MOS is better, the equation for the rtt is reversed, compared to MOS in the adapted equation. Meanwhile, the calculation for Isup and the sigmoid function are the same.

86 Table 5.4 ANHA adaptation from AntNet Path Goodness Measurement Model.

AntNet Path Goodness Measurement:

𝑟 = 𝑐₁(𝑟𝑡𝑡_{𝑐𝑢𝑟𝑟𝑒𝑛𝑡}

𝑟𝑡𝑡_{𝑏𝑒𝑠𝑡} ) + 𝑐₂( 𝐼_𝑠𝑢𝑝− 𝑟𝑡𝑡_{𝑏𝑒𝑠𝑡}

(𝐼_𝑠𝑢𝑝 − 𝑟𝑡𝑡_{𝑏𝑒𝑠𝑡}) + (𝑟𝑡𝑡_{𝑐𝑢𝑟𝑟𝑒𝑛𝑡}− 𝑟𝑡𝑡_{𝑏𝑒𝑠𝑡}))

𝐼_𝑠𝑢𝑝 = 𝜇_𝑟𝑡𝑡+ 𝑧 (𝜎_𝑟𝑡𝑡

√𝑊)

𝑟 ← 1

1 + 𝑒^{−16×𝑟+8} ANHA Path Goodness Measurement:

𝑟 = 𝑐₁(𝑀𝑂𝑆_{𝑐𝑢𝑟𝑟𝑒𝑛𝑡}

𝑀𝑂𝑆_{𝑏𝑒𝑠𝑡} ) + 𝑐₂( 𝑀𝑂𝑆_{𝑏𝑒𝑠𝑡}− 𝐼_𝑠𝑢𝑝

(𝑀𝑂𝑆_{𝑏𝑒𝑠𝑡}− 𝐼_𝑠𝑢𝑝) + (𝑀𝑂𝑆_{𝑏𝑒𝑠𝑡} − 𝑀𝑂𝑆_{𝑐𝑢𝑟𝑟𝑒𝑛𝑡}))

𝐼_𝑠𝑢𝑝 = 𝜇_𝑀𝑂𝑆+ 𝑧 (𝜎_𝑀𝑂𝑆

√𝑊)

𝑟 ← 1

1 + 𝑒^{−16×𝑟+8}

The Handover Decision Table updating rule: The values of pheromone, PN (refer table 5.1) of each path determines the attractiveness of a path. In ANHA, the pheromone of each path between MH and CH is given a fair amount of pheromone. For example, if there are two paths, each path is given an initial value of ½. Then the level of pheromone is updated via the reinforcement process. The positive reinforcement is shown in equation 5.14, whilst the negative reinforcement (evaporation) is shown by equation 5.15.

𝑃_𝑁 ← 𝑃_𝑁+ 𝑟(1 − 𝑃_𝑁) (5.14)

𝑃_𝑁′ ← 𝑃_𝑁′− 𝑟𝑃_𝑁′ (5.15)

Let a MH have two NICs, NIC1 and NIC2. When a DA arrive at NIC1, the pheromone at NIC1 will receive a positive reinforcement, whilst at the same time the pheromone at NIC2 will receive a negative reinforcement. Thus, as time goes by, and DA keeps arriving at both NICs, the NIC with better quality will eventually become bigger rapidly compared to NIC with low quality.

This is the ant stigmergy characteristic that has been incorporated in ANHA. This updating rule is directly adapted from AntNet without any changes.

87 The local decision policy: Then the best path is chosen using the local ant decision policy where the probability of choosing a path is calculated using equation 5.16. This information is kept in the network selection array in figure 5.1.

𝑃_𝑁^′ = 𝑃_𝑁+ 𝛼𝑙_𝑛

1 + 𝛼(|𝑁_𝑘| − 1) (5.16)

P’N is the probability of choosing a path. PN is the pheromone level of each path. Nk is the number of available paths. α is the weight for the heuristic component ln, where ln is the normalized value of RSS of each AP connected to MH as shown by equation 5.17. The path (AP) with the highest P’N is selected as the best path, and in the handover process, this path will be chosen. The calculation for P’N is directly taken from AntNet, whereas the equation for ln is adapted by changing the qLengthN, which represents the queue length of each path in AntNet to RSS, since the parameter the target of proposing ANHA is to integrate both parameters, which are the RSS and MOS in the handover decision and triggering process. Notice that α is the coefficient that controls how much influence RSS has on the handover decision. The value of α can determine whether the MH triggers the handover process earlier or later; for example, when α is bigger, RSS gives more influence and the handover will be triggered earlier while when the α is smaller, RSS have less influence causing the MH to trigger the handover later.

𝑙_𝑁 = 1 − 𝑅𝑆𝑆_𝑁

∑^|𝑁_𝑛′^𝑘=1^| 𝑅𝑆𝑆_𝑛′

(5.17)

Table 5.5 ANHA adaptation from AntNet local decision policy.

AntNet local decision policy:

𝑃_𝑁^′ = 𝑃_𝑁+ 𝛼𝑙_𝑛 1 + 𝛼(|𝑁_𝑘| − 1)

𝑙_𝑁 = 1 − 𝑞𝐿𝑒𝑛𝑔𝑡ℎ_𝑁

∑^|𝑁_𝑛′^𝑘=1^| 𝑞𝐿𝑒𝑛𝑔𝑡ℎ_𝑛′

ANHA local decision policy:

𝑃_𝑁^′ = 𝑃_𝑁+ 𝛼𝑙_𝑛 1 + 𝛼(|𝑁_𝑘| − 1)

𝑙_𝑁= 1 − 𝑅𝑆𝑆_𝑁

∑^|𝑁_𝑛′=1^𝑘| 𝑅𝑆𝑆_𝑛′

The evaluation of the DA’s path: the path available to MH is evaluated in two kinds of way:

 Implicit way: through the decision table update component.

 Explicit way: through the use of RSS value as the heuristic component.

Up to this point, the features of AntNet adapted to ANHA have been discussed. These five components is the core component in implementing a bio-inspired ACO based handover decision algorithm. The algorithm not only take into account the current state of the network (RSS) it also

88 considers the long-term memory, PN (the pheromone that are updated based on the MOS level of a path according to the Handover Decision Table update rule), in deciding the best path.