UniversitÓ degli Studi di Bologna


Mobility Prediction Project


Handover and Mobility Prediction mechanisms for
Reactive and Hard/Soft-Proactive Handover Strategies

Several communication-level handover strategies are possible, which mainly differ in the event used to trigger the handover. It is possible to distinguish between two main strategy categories, reactive and proactive. Reactive handover strategies tend to delay handover as much as possible: handover starts only when wireless clients completely lose their current AP signal. Reactive strategies are effective in minimizing the number of handovers, e.g., by avoiding to trigger a handover process when a client approaches a new wireless cell, without losing the origin signal, and immediately returns back to the origin AP. However, reactive handovers tend to be long because they include looking for new APs, choosing one of them, and asking for re-association. Moreover before e reactive handover available bandwidth is limited, since signal quality is low (in first analysis signal quality depends on AP-wireless client distance). Proactive strategies, instead, tend to trigger handover before the complete loss of the origin cell signal, e.g., when the new cell RSSI overpasses the origin one. These strategies are less effective in reducing the number of useless handovers but are usually prompter, by performing search operations for new APs before the handover procedure starts.

By concentrating on proactive strategies, a further classification is possible. On the one hand, Hard Proactive (HP) strategies trigger a handover any time the RSSI of a visible AP is greater than the RSSI of the currently associated AP plus an Hysteresis Handover Threshold (HHT); HHT is introduced mainly to prevent heavy bouncing effects. On the other hand, Soft Proactive (SP) strategies are "less proactive" in the sense that they trigger handover only if i) the HP condition applies (there is an AP with RSSI greater than current AP RSSI plus HHT), and ii) the current AP RSSI is lower than a Fixed Handover Threshold (FHT).

For instance, the handover strategies implemented by Cisco Aironet 350 and Orinoco Gold Wi-Fi cards follow, respectively, the HP and SP models. More in detail, Cisco Aironet 350 permits to configure its handover strategy with the "Scan for a Better AP" option: if the current AP RSSI is lower than a settable threshold, the Wi-Fi card monitors RSSI data for all visible APs; for sufficiently high threshold values, the Cisco cards behave according to the HP model. Orinoco Gold cards exactly implements the SP strategy, without giving any possibility to configure the used thresholds.

We have designed and implemented two prediction mechanisms: Handover Prediction and Mobility Prediction.
Handover Prediction supplies the probability an handover procedure starts, Mobility Prediction if an handover procedure is probable and which is the most probable next AP.

Our handover/mobility prediction solution is based on a two pipelined module architecture. The first module (Filter) is in charge of filtering RSSI sequences to mitigate RSSI fluctuations due to signal noise. The second module (Prob) tries to estimate the probability a handover happens in the near future and which is the most probable next AP based on RSSI values provided at its input from Filter.

Figure 1. Filter and Prob Modules.

The modular architecture of our predictor permits a completely separated implementation and deployment of Filter and Prob, thus simplifying the exploitation and experimentation of different filtering and handover/mobility prediction mechanisms, even dynamically composed at provision time by downloading the needed module code. In particular we compare prediction mechanism performance when based on RSSI or wireless client position and RSSI filtered with several low-pass filters.

By delving into finer details, we have implemented two variants of the Prob module, one suitable for communication-level HP handovers and the other for SP ones. We have decided not to work on Prob prototypes for reactive strategies because of two reasons: first, handover prediction is less challenging in the case of reactive handovers than of proactive ones since the triggering of a reactive handover only depends on the RSSI data from one single AP; secondly, reactive communication-level handovers are of minor interest for services with session continuity requirements, given their longer time needed to complete handover.
The HP-variant of our Prob module is in the state:
  • LowProb, if the filtered value for the current AP RSSI is greater than the filtered RSSI values for any visible AP plus a Hysteresis Prediction Threshold (HPT);
  • HighProb, otherwise.
The SP-variant of the Prob module can assume the following states:
  • Low-Prob, if the filtered RSSI value for the current AP is greater than either a Fixed Prediction Threshold (FPT) or the filtered RSSI value for any visible AP plus HPT;
  • HighProb, otherwise.
Let us stress a Prob Module with only two probability states (High/Low) is sufficient for many application scenarios; however it is possible to develop more sophisticated Prob Modules with several intermediate states, if needed.
Handover Probability supplies to the application level the Prob Module output, without any further refinement. When handover probability is High Mobility Prediction supplies the MAC address of the most probable AP, i.e. with the strongest RSSI.

Figure 2 depicts predicted RSSI values for the current AP and the next one, in proximity of an HP handover. A wireless client, moving from the origin AP to the destination AP, is first associated with the origin (white background), then with the destination (grey background). When the predicted RSSI of the destination AP overcomes the predicted RSSI of the origin AP plus HHT, the handover is triggered.

Figure 2. HP-variant prediction and handover triggers.

The implemented SP-variant of the handover predictor triggers a prediction when the predicted RSSI value for the current AP is lower than i) a Fixed Prediction Threshold (FPT) and ii) a predicted RSSI value for one visible AP plus HPT. Similarly to HP, the SP-variant predictor only considers the most probable future locality in the case of several predictions simultaneously enabled. Figures 3 and 4 show predicted RSSI values for the origin and the destination APs in proximity of an SP handover. Figure 3 depicts a case where predicted RSSI values change quite slowly: it is the overcoming of hysteresis thresholds that triggers handover prediction. In Figure 4, instead, predicted RSSI values rapidly evolve, and the passing of fixed thresholds produces handover prediction.

Figure 3. SP-variant prediction and handover triggers for relatively slow RSSI evolution.

Figure 4. SP-variant prediction and handover triggers for relatively fast RSSI evolution.

Last update: 4-may-10

Privacy Policy