Simulation Environment

We consider a WCDMA system containing 7 × 7 hexagonal and wrap-around WCDMA cells for simulations. The longest distance between the BS and the cell boundary is one kilometer. The channel of the WCDMA system suffers inter-cell MAI, intra-cell MAI, AWGN noise, log-normal shadowing [42], and multipath fad-ing [43]. The path-loss exponent is 4.35 [40]. The spreadfad-ing factor is from 4 to 128.

Perfect power control is used in the system.

The arrival of new calls in each cell is modeled as a Poisson process with a mean arrival rate λ. Four types of traffic are considered: real-time voice, real-time video, non-real-time data, and non-real-time best effort. The traffic intensity is defined as the product of mean arrival rate and mean session time of a service type. Users are also assumed to be uniformly distributed in cells, and a random-walk model is used to simulate the mobility of every user. Similarly, all mobile users in WLAN are located randomly and activated in a saturation mode of that their access transmissions are always on. The new user arrival process is assumed to be in Poisson distribution.

There are four types of traffic: real-time voice, real-time video stream, non-real-time

Figure 4.5: The topology of WCDMA/WLAN subnetworks for simulations

data, and best-effort. Their QoS requirements are listed in Table 4.1. Since the requirements can be supported in both WCDMA and WLAN systems, there would be no data rate problem to handoff vertically from WLAN (higher bandwidth) to WCDMA (lower bandwidth).

As shown in Fig. 4.5, the WLAN subnetworks are fully overlapped over the WCDMA networks. A WLAN subnetwork group consists of 3×3 round QoS basic service sets (QBSSs). The centers of WLAN subnetwork groups are located at the same place of WCDMA’s BSs and the cross-point of 3 WCDMA cells’ boundaries. The radius of each QBSS is 100 meters, and any two adjacent QBSSs are assumed to use different channel frequencies. Both Rayleigh and log-normal fading channel models are also considered. The WLAN system parameters are based on those in [50, 52], where the SIFS, PIFS, and DIFS are assumed to be 10µs, 20µs, and 40µs, respectively;

a beacon interval is 20ms; the maximum duration of CFP is 15ms; and a slot time (aSlotTime) of PHY is 9µs. The value of ξ and constant c in (4.11) are 500ms and 3, respectively. The bias constant µ in (4.12) and (4.22) are 0. In order to eliminate the handoff latency of the handoff request contention in the WLAN system, the Fast

Table 4.1: QoS Requirements

Traffic type Min. data rate (kbps) Delay bound (ms) BER

Voice 32 100 10⁻³

Video stream 64 100 10⁻³

Data 128 1,000 10⁻⁶

Best-effort 1 10,000 10⁻⁵

Handoff Protocol [92] is adopted to provide efficient inter-AP transitions.

The FQAC will be compared to conventional SIR-based call admission control (CAC) [44] and joint session admission control (JSAC) [79]. The SIR-based CAC is implemented to make admission decision for a new user according to the currently estimated SIR of the system. If the system’s SIR is higher than the threshold SIR^∗, then the call will be admitted; otherwise, the call will be rejected. In the implemen-tation parameters of SIR-based CAC, such as the margin of residual capacity [44] and the margin for handoff [45] to tolerate the misjudged admissions, are finely tuned ac-cording to the greedy approach in [93] to maximize the system utilization and handoff failure constrains. As to JSAC, a brief description is given in Section 4.1.

4.4.2 Simulation Results

Fig. 4.6 depicts the QoS guarantee ratio for all services versus the average traffic intensity. It can be found that FQAC can maintain almost all users’ QoS. When the traffic intensity is as high as 1.5, the QoS guarantee ratio is still over 98%. The JSAC, however, has a few QoS violations over high traffic intensity, and has apparent degradation performance when the traffic intensity comes to very high. Also, the SIR-based CAC get the worst guarantee region. It is because the SIR-SIR-based CAC has only instant channel soundings for admission decisions. It could be affected by temporary channel fluctuations, cause misjudgment of system situation, and eventually lead

Figure 4.6: QoS guarantee ratio abnormal low blocking rate and overload the system capacity.

Figure 4.7 depicts the average blocking rates of new mobile users for four types of service versus the average traffic intensity. It can be found that the FQAC can keep the blocking rate until the traffic intensity becomes extremely high. Because admitting a real-time service has higher impact than accepting a non-real-time service, the blocking rate of a real-time request is higher than that of a non-real-time request.

This will ensure to achieve QoS guarantee of all existing users. The reason of such performance is that the FQAC consider more realistic and essential measures of both WCDMA and WLAN systems. The system states can be really reflected in the admission results. The JSAC method also has the same property, but the blocking rate is a little bit higher. The reason is that JSAC has to mitigate the influence of user mobility, so it sacrifices the system utilization. The blocking rate of SIR-based CAC is the lowest when the traffic intensity is high. It is because the CAC is decided according to a short period of SIR measurement, which is possible to obtain a temporary low value of the signal fluctuation, and would accepted too many users.

Fig. 4.8 shows the average handoff user blocking rates for four types of service

Figure 4.7: Average new user blocking rate for the service of (a) voice, (b) video, (c) data, and (d) best-effort

Figure 4.8: Average handoff user blocking rate for the service of (a) voice, (b) video, (c) data, and (d) best-effort

Figure 4.9: Average number of handoff per minute

versus the average traffic intensity. FQAC has sufficient low blocking rate to decrease the forced termination rate for the real-time services. This is because the dwelling estimation can avoid high-velocity users to enter the small-coverage subnetworks, and select the suitable subnetwork in which the mobile user dwell time is longer, as shown in Fig. 4.9. This is a great advantage to improve the user experience when using voice or video streaming services. Meanwhile, the systems’ overhead of dealing with the handoff processes can be decreased. JSAC also has the same trend, but the forced termination rate would be somewhat higher because it does not consider the realistic channel variation and user mobility conditions. SIR-based CAC has the same problem mentioned above, and the property of admission decision with instant-measure also induces improper forced terminations.

4.5 Concluding Remarks

In this chapter, we propose fuzzy Q-learning admission control (FQAC) for WCDMA/

WLAN heterogeneous networks. The FQAC system adopts neural-fuzzy inference

sys-tem (NFIS) with Q-learning (FQL) method for dwelling estimation and admissibility estimation. It considers multiple realistic system measures, such as the number of mobile users, interference in WCDMA systems, and the busy period in WLAN sys-tems. Meanwhile, the QoS requirements and mobility of mobile users are also taken into account. According to these considerations, FQAC would generate the dwelling cost and admissibility cost in each reachable subnetwork to reflect the possible impact of the requesting mobile user. With FQL, the relations between system states and FQAC actions can be adaptively adjusted by a reinforcement learning signal mea-sured from the states of every reachable network. In order to minimize the expected maximal impact (cost) of the mobile user’s admission request, the decision maker in the FQAC system adopts minimax theorem to jointly estimate the mixed-cost and decide the most suitable subnetwork or reject the mobile user request. Simulation results show that FQAC performs more aggressive admission decisions than JSAC and SIR-based CAC when the traffic intensity is low. Hence FQAC has lower new and handoff user blocking rates and higher system utilization while maintaining the QoS. As long as the traffic intensity grows high, FQAC turns to be more conservative, and the blocking rates increase rapidly to avoid QoS violations. Since FQAC has the feature to estimate users mobility intelligently, the admitted subnetwork is also the one in which the user may dwell longer. This will lower the handoff rate, decrease the handoff blocking rate, and improve the real-time services user experience. And the mentioned performance of the FQAC system is the best evidence of that the FQAC system is capable to adapt to the fluctuation of traffic dynamics in the heterogeneous networks.

Chapter 5