CASH: Context-Aware Smart Handover

• Context-Aware Smart Handover (CASH) is a protocol that utilizes diverse context inputs—signal strength, mobility, load, and policy—to optimize handover decisions across heterogeneous networks.
• It integrates predictive, fuzzy logic, and learning-based methods to balance seamless connectivity with reduced signaling overhead and minimized ping-pong effects.
• CASH demonstrates practical benefits in scenarios such as UAV corridors and HetNets, achieving up to 78% fewer handovers and substantial gains in throughput and reliability.

Context-Aware Smart Handover (CASH) denotes handover mechanisms in which decisions are driven by context rather than by a single instantaneous signal-strength threshold. In the cited literature, the term appears explicitly as a protocol for reliable UAV connectivity on aerial corridors, but the underlying idea is older and broader: handover is made “smarter” by combining link, mobility, location, load, policy, or service information, and in some cases by adding prediction, learning, or control-plane context generation before execution. A plausible implication is that CASH is best understood as a family of context-driven handover designs spanning IEEE 802.11/Mobile IPv6, heterogeneous vertical handover, HetNet mobility management, lower-layer triggered mobility, and UAV corridor communications (Saboor et al., 5 Aug 2025, Lim et al., 2012, Gonzalez-Horta et al., 2011).

1. Conceptual lineage and scope

The conceptual predecessor of CASH in the supplied literature is “cognitive handoff,” defined as a multipurpose, multi-criteria, context-aware, self-aware, and policy-based process that explicitly trades off conflicting objectives. That formulation arose from a critique of traditional handoffs as “single-purpose” systems that may be seamless but not adaptive, adaptive but not secure, or secure but not autonomous. Within that view, handoff is not merely a radio trigger; it is a dynamic control problem over heterogeneous networks, services, providers, devices, and user goals (Gonzalez-Horta et al., 2011).

A second strand is the context-aware vertical handover literature, which treats service continuity under heterogeneity and QoS constraints as the central problem. That survey does not use the name CASH, but it characterizes context-aware handover as collecting relevant context from user, device, network, application, and environment, and then using that context to decide whether and where to hand over while preserving QoS/QoE and continuity. It also notes that “handoff” and “handover” are effectively synonymous in this literature (Alhazmi et al., 2011).

The 2025 UAV study gives CASH its explicit name: “Context-Aware Smart Handover” for reliable UAV connectivity on aerial corridors. There, CASH is a trajectory-aware, geometry-based protocol that makes proactive handover decisions using corridor structure and forward-looking scoring, rather than relying only on instantaneous signal strength and fixed thresholds (Saboor et al., 5 Aug 2025).

2. Context sources, knowledge models, and architecture

Across the cited work, the defining feature of CASH is the use of multiple context domains. The cognitive handoff model separates external context from internal context. External context comes from five entities—users, terminals, applications, networks, and providers—while internal context captures the handoff process’s own behavior and performance, including latency, success rate, number of handoffs, signaling overhead, authentication latency, and user intervention rate. This split makes the system not only context-aware but also self-aware (Gonzalez-Horta et al., 2011).

The vertical-handover survey provides a more granular taxonomy. It lists user context such as preferences, profile, intention, states, trip information, location, speed, trajectory, and QoS requirements; device context such as capabilities, characteristics, specifications, software types, available device services, and multi-homed terminal state; network context such as available bandwidth, signal strength / RSS, power consumption, cost, delay, jitter, packet error rate, and channel busy time ratio; application/service context such as service type and SLS requirements; environmental / location context such as 2D coverage maps and impending network profile; and IMS / session / protocol context such as SIP signaling and handover prediction state. It also distinguishes static profile from dynamic profile, a distinction used explicitly in surveyed schemes (Alhazmi et al., 2011).

At the mobile-core level, the Context Awareness Framework (CAF) and the Context Generation and Handling Function (CGHF) provide an architectural substrate for CASH-like decisions. In that model, raw information from UEs, RAN, other NFs, and third-party services is transformed into facts, and then into context, defined as “the knowledge describing the situation of an entity.” CGHF includes a Pub/Sub Engine, Fact Generator, Context Generator, and Context Manager. The processing chain supports data fusion, averages, rates of change, forecasting, classification, clustering, and trend analysis, while the Context Manager uses Event-Condition-Action rules. The resulting relevant context can then “Trigger UE traffic redirection or handover towards new access technology point of attachment” (Marquezan et al., 2016).

3. Algorithmic realizations

One concrete realization is FL Trend (RSD) for IEEE 802.11 / Mobile IPv6. It uses RSSI, speed, and distance as input decision criteria in a fuzzy logic handover trigger. Each input has Low, Medium, and High membership functions, and the 27-rule base maps the context vector $(\text{RSSI}, \text{speed}, \text{distance})$ to handover or not handover outputs. The key novelty is that the fuzzy output is not used instantaneously; it is monitored over a sliding window so that short-lived fluctuations do not trigger handover. The paper’s example uses $N_{ws}=10$, and the post-handover delay is adapted so that low speed implies longer delay and high speed implies shorter delay, with the stated goal of reducing false alarms and ping-pong effects (Lim et al., 2012).

A second realization is Context Aware Mobile Initiated Handoff in an IEEE 802.11 ESS. Here the mobile node (MN) detects degradation using either a packet-drop trigger or the exponentially weighted moving average (EWMA) of interface queue length, scans reachable APs through an interleaved DSProbe / DSProbeResponse mechanism, sends a MoveRequest to the serving AP, and receives an AP-filtered HandoffTargetMessage after LoadRequest / LoadResponse exchanges over the backbone. The candidate list is filtered by residual load margin, and the MN finally chooses the target AP with the strongest SINR. If no AP is suitable, the MN performs a vertical handoff to a WiMAX BS (Sarma et al., 2011).

A third realization appears in HetNets, where mobility management is made context-aware through reinforcement learning. Macro and pico BSs learn long-term traffic loads and cell range expansion (CRE) settings, while scheduling uses UE velocity and historical service rate exchanged across tiers via X2-interface coordination. The paper treats the BSs as players in a game, with discrete REB actions and utilities defined by total served rate/load, and it compares multi-armed bandit (MAB) learning with satisfaction-based learning (Simsek et al., 2015).

In next-generation wireless networks, a further development combines L1/L2-triggered mobility (LTM) with LMMSE-predicted received signal power, a Contextual Multi-Armed Bandit (CMAB) for target-gNB selection, and RIS-assisted coverage enhancement. The CMAB observes a context vector including candidate-gNB RSRP, moving-average MCS / throughput-related stability metrics, SINR, UE speed, time since last HO, and serving-gNB identity, and it uses KernelUCB with an arc-cosine kernel rather than a purely linear bandit. RIS then improves robustness, especially in NLOS and cell-edge regions, through channel shaping and phase optimization (Moreno-Locubiche et al., 9 Dec 2025).

The explicitly named UAV CASH protocol applies the same general idea to pre-defined aerial corridors. It first filters candidate GBSs to those ahead of the UAV and above the minimum usable threshold, then ranks them using a forward-looking geometric score that prefers GBSs farther ahead in the corridor and closer to the corridor centerline. Handover is executed only when the selected candidate is stronger than the serving BS by at least $\Delta$ and the serving link is sufficiently close to the threshold region, controlled by the Handover Safety Margin (HSM) $\delta$ (Saboor et al., 5 Aug 2025).

4. Formalization and decision criteria

The most general formalization in the cited work is the cognitive-handoff multi-objective model. Let $F$ be the set of desirable features and $C$ the set of context data. For each feature $f$, a context variable $V_i$ may be positively or negatively correlated, and weights satisfy $W_i \in \mathbb{R}[0,1]$ with $\sum W_i = 1$. The objective is written as

$N_{ws}=10$0

subject to lower and upper bounds on the variables. In that formulation, features such as seamlessness, autonomy, security, correctness, and adaptability are translated into measurable goals and criteria, and trade-offs are treated as intrinsic rather than exceptional (Gonzalez-Horta et al., 2011).

The predictive LTM work formalizes anticipatory triggering through a linear predictor,

$N_{ws}=10$1

where the input vector contains current and past L3-filtered RSRP samples for a given gNB and the covariance terms are estimated with an exponential window. In the same paper, the CMAB objective is to maximize long-term reward while minimizing regret, and the reward combines average throughput over the last 100 ms with indicators for HO, HOF, RLF, and ping-pong (PP), balanced by $N_{ws}=10$2-weights (Moreno-Locubiche et al., 9 Dec 2025).

The IEEE 802.11 mobile-initiated scheme uses an EWMA trigger printed as

$N_{ws}=10$3

with $N_{ws}=10$4 in experiments, and a candidate-AP selection criterion

$N_{ws}=10$5

where $N_{ws}=10$6 is a safety margin and the simulation value is 250 kbps. This makes the trigger reactive to performance degradation but proactive in timing when queue growth is detected early (Sarma et al., 2011).

The UAV CASH protocol preserves the threshold-and-TTT logic of A3T-style handover while changing candidate selection. The A3 baseline is

$N_{ws}=10$7

sustained over the TTT window, whereas CASH adds the serving-link condition $N_{ws}=10$8 and uses a geometric score favoring forward-aligned base stations. The paper states that larger $N_{ws}=10$9 leads to earlier handover with fewer outages and more handovers, while smaller $\Delta$0 leads to later handover with fewer handovers and more outage risk (Saboor et al., 5 Aug 2025).

5. Deployment environments and reported performance

In IEEE 802.11 / Mobile IPv6 simulations using Omnet++ 4.0, FL Trend (RSD) is compared with RSSI Threshold, Change of RSSI, and FR Threshold over ten scenarios with speeds from 5 to 30 kmph. The evaluated metrics are number of handovers, average packet loss, average throughput, minimum packet delay, maximum packet delay, and mean packet delay. The reported result is that FL Trend (RSD) generally provides the best balance, especially in the 16–30 kmph region, with fewer handovers, lower packet loss, higher throughput, and lower packet delay (Lim et al., 2012).

In the IEEE 802.11 ESS load-balancing study using ns-3 version 3.8, the main findings are that EWMA-based queue detection outperforms packet-drop-based detection, that earlier detection improves throughput and reduces delay, and that the best throughput and delay are observed in case VI, where the queue-based trigger uses a low threshold and BS fallback is enabled. The paper states that up to three MNs hand off to the BS, that the best performance occurs when the number of MNs per AP is within about five to six, and that handoff latency is within 50 ms (Sarma et al., 2011).

In the HetNet reinforcement-learning study, the proposed context-aware mobility management achieves, on average, a gain of up to 80\% for UE throughput, while the handover failure probability is reduced up to a factor of three. The paper also reports 43% and 75% average-UE-throughput improvements for MAB and satisfaction-based learning, respectively, and states that for high-velocity users and short TTT the ping-pong probability is about half of the classical approach (Simsek et al., 2015).

In the LTM-based next-generation wireless study, the lower-layer substrate can execute in under 5 ms, reduces latency by up to 70%, and lowers signaling overhead because RRC commands are not needed. Across eight evaluated configurations, LTM + CMAB + LMMSE + RIS is reported as the best overall balance, with high reliability, low HOF and RLF, the lowest ping-pong rate, reduced control-plane overhead, and only a small tradeoff in resource reservations (Moreno-Locubiche et al., 9 Dec 2025).

In UAV aerial corridors, the named CASH protocol is evaluated in a MATLAB-based custom simulator over 500 unique cities. The principal result is that CASH reduces handover frequency by up to 78% compared to A3 while maintaining low outage probability. The study further reports that for dense urban aerial corridors a GBS density of about 4–6 GBS/km$\Delta$1 offers the best balance among outage, handovers, and deployment cost, and that densities above 6 GBS/km$\Delta$2 do not significantly improve reliability for A3/Fuzzy while potentially increasing unnecessary handovers (Saboor et al., 5 Aug 2025).

6. Trade-offs, limitations, and unresolved issues

The literature is explicit that no single handover scheme fully resolves the problem space. The cognitive-handoff framework identifies several fundamental conflicts: seamlessness vs. handoff rate, decision speed vs. success rate, rich context vs. signaling overhead, and user vs. provider preferences. It therefore treats handoff as a multi-objective optimization problem whose solutions are Pareto-like compromises rather than single-metric optima (Gonzalez-Horta et al., 2011).

The vertical-handover survey adds system-level limitations: single point of failure in centralized context repositories, radio link overhead from frequent context exchange, complexity from extra middleware and proxies, tight coupling cost in some IMS settings, make-before-break assumption in some continuity mechanisms, increased power consumption, signaling delay for SIP operations, and cases in which QoS or handover performance improves without a full continuity guarantee (Alhazmi et al., 2011).

Algorithm-specific limitations also recur. In FL Trend (RSD), the fuzzy membership ranges and rule base are tailored to a particular mobility scenario, speed and distance estimation are assumed rather than fully integrated into evaluation, the low-speed regime shows trade-offs, and the evidence is simulation-based rather than a formal analytical proof of optimality (Lim et al., 2012). In the LTM + CMAB + LMMSE + RIS framework, CMAB may slightly increase resource reservation and some RLFs due to exploration, while RIS may slightly increase cell preparation/resource reservation because more candidate links become viable (Moreno-Locubiche et al., 9 Dec 2025).

The UAV CASH protocol likewise depends on strong structural assumptions: a predefined straight aerial corridor, known or predictable trajectory, rooftop-mounted static base stations, and a specific urban grid and mmWave path-loss model at 28 GHz. The paper states that the HSM $\Delta$3 is optimized empirically, not dynamically, and that CASH may be less effective in free-flight / corridor-free scenarios (Saboor et al., 5 Aug 2025).

At the control-plane level, CGHF introduces a further coordination problem: the framework explicitly asks how to benefit from context reusability while handling conflicting changes triggered by different NFs or applications based on the same context. This suggests that future CASH systems may need not only richer inference, but also stronger conflict resolution across mobility, policy, flow management, anchoring, and service relocation functions (Marquezan et al., 2016).