A WLAN handoff composes of three phases, as shown in Fig. 1. The scanning phase discovers the APs that an STA can hear and measures the signal strengths from the APs. The re-authentication phase verifies the access rights of an STA to a specific AP. Finally, the re-association phase negotiates with the target access point and re-establishes the connection [11]. Table 1 shows the latency of each phase. It can be seen that the latter two phases spend less than 10 ms while the scanning phase contributes the major part about 90% of the total handoff latency. The same observations can be made from different implementations of wireless network interface cards, as shown in Fig. 2.
Table 1. Latencies of handoff procedures
Fig. 1. Standard IEEE802.11 handoff procedures [9]
Fig. 2. Handoff latencies of NICs from different vendors [9]
The IEEE 802.11 specifies two scanning mechanisms, i.e. active and passive scan. While performing an active scan, an STA sends probe requests and waits for the responses from APs on each channel. For a passive scan, an STA has to wait for certain amount of time to receive beacons from all APs on a channel, and then the STA continues with another channel. Table 2 compares these two kinds of mechanisms. The passive scanning mechanism suffers from a major defect that it has to tie in with the exact time when the APs send the beacons. This characteristic results in a long waiting time on a single channel. The active scanning mechanism does not require staying in a channel for that long. An STA sends a request whenever it needs to update the environment information. However, since the active scanning mechanism stays on each channel with APs for the maximum channel time, it may waste too much time on channels with only few APs. To overcome this drawback, previous works learn from the environment, intelligently ignore the empty channels [13] and complete the scan right after all messages from the expected APs’ responses are successfully received [13].
These enhancements based on the active scan reduces the channel waiting time, however, they do not consider the requirements such as packet delay for the connection. The QoS requirements of the connection may not be satisfied if the WLAN scan takes too long time to complete.
Based on the above observation, some enhanced solutions proposed to take advantage of
the beacon regularity in order to avoid overheads and reduce the channel waiting time [10].
The idea is to receiving AP beacons which are predictable and to eliminate the channel occupied time. Then, an STA can get the background information without the overheads caused by probe requests and responses in the active scan. The difference can be seen in an ideal scenario as follows. To investigate the information of a single AP, the standard active scan takes (R+M) milliseconds, in which R stands for the request transmission time, 1 ms for example, and M stands for the predefined maximum channel time, typically 11 ms. Therefore, the latency of the active scan for a single AP is about 12 ms. However, listening to a beacon for one AP requires only a packet receiving time which is less than 1ms.
Table 2. The comparison of the standard scanning mechanisms
Although the existing work improves the scanning latencies, they have to obey the beacons from APs, which leads to the bottleneck and limits the enhancement. The more APs which need to be scan are, the higher possibility the beacons collide. If beacons collide with each other, the performance is downgraded. As shown in Fig. 3, because of a beacon collision
for AP1 and AP2, an STA cannot process AP2 (the dot-line block) after finishing AP1 scan. The STA has to wait for another beacon interval to get the next beacon (the solid block). The performance downgrades significantly due to a long beacon interval. The total scan time may become double as the latency in the non-collision case. SyncScan [10] solves this problem by asking the APs on the same channel to send the beacons at the same time t (or very close to it), and APs on the next channel to send the beacons at time (t+d), and so on. The beacon arrival time is staggered to avoid collisions between channels. However, the APs on the same channel broadcast their beacons at a very close time, and STAs are not guaranteed to receive beacons from all APs successfully in the first try. Moreover, this approach requires the coordination between APs and may have deployment problem.
Fig. 3. The bottleneck of a passive scan
All the previous works mentioned above have good performance only in some specific cases. The active scan mechanisms may spend long channel waiting time for hotspots with low AP density. On the other hand, passive scan mechanism may suffer from serious beacon conflicts in hotspots with high AP density. Moreover, the existing methods do not consider the
various delay constraints of real-time services. User applications send the data packets (ex.
voice or video) regularly and tolerate various ranges of delays. In this thesis, a novel mechanism which minimizes the channel waiting time by the passive scan approach and resolves the beacon conflicts by using the active scan strategy is proposed. Moreover, the proposed mechanisms consider both handoff delay and various QoS requirements such as delay and jitter constraints for different applications.