Tandem Queueing Analysis and Airtime Validation for UAV-Assisted Heterogeneous Wi-Fi–LoRa Disaster Networks

Natural disasters such as earthquakes, floods, and hurricanes frequently incapacitate ground-based telecommunication infrastructure, severing critical connectivity between affected populations and emergency rescue coordination teams. The inability to transmit geographical coordinates, medical status, and situational updates severely delays triage operations. Consequently, establishing ad hoc, resilient, infrastructure independent aerial communication bridges is paramount for disaster recovery [1].

Unmanned aerial vehicles (UAVs) provide a viable aerial relay solution because of their rapid deployment, flexible trajectory control, and elevated line-of-sight (LoS) propagation profiles. However, designing heterogeneous aerial networks requires addressing the disparate operational envelopes of constituent wireless technologies. Combining IEEE 802.11n Wi-Fi for localized, high-throughput ground-to-UAV access with Long Range (LoRa) modulation for extended UAV-to-ground backhaul forwarding establishes an effective heterogeneous compromise. Nevertheless, this architectural coupling introduces a significant capacity mismatch since the multi-megabit access link can rapidly ingest traffic that overwhelms the sub-kilobit LoRa backhaul, causing buffer accumulation, excessive latency, and queue overflow during simultaneous multi-node reporting.

In order to analyze these dynamics, this paper presents a UAV-assisted two-hop Wi-Fi–LoRa emergency network which can be quite common for the scenarios considered [1]. Ground nodes broadcast a compact 11-byte telemetry frame containing vital positioning and status indicators to an overhead UAV, which transparently relays the packets to a remote base station over LoRa. We model this pipeline using a tandem M/D/1 queueing framework where service durations are analytically bounded by deterministic transmission time-on-air (ToA). This framework characterizes queue delay, packet loss, and operational stability across varied network loads [2].

The primary contributions of this paper are:

• A hybrid Wi-Fi–LoRa relay architecture featuring an 11-byte payload designed to minimize aerial transmission overhead.
• An analytical airtime-derived tandem queueing formulation characterizing throughput and buffer bottlenecks across asymmetric wireless links.
• Cross-validation of network performance through packet-level simulation in OMNeT++ (utilizing INET and FLoRa frameworks) and discrete-event queueing simulations.

Aerial relays have been widely investigated for restoring post-disaster connectivity. Panda et al. [3] demonstrated the feasibility of deploying UAV-mounted Wi-Fi access points for field coordination. Casmin and Ever [1] formulated open tandem queueing models to analyze flow limits and service rates for UAV-assisted base station offloading, establishing the viability of queueing frameworks in aerial network performance assessment.

Experimental trials corroborate the choice of IEEE 802.11n for ground-to-UAV links. Ullah et al. [4] benchmarked 802.11 protocols across various aerial altitudes, confirming that 2.4-GHz 802.11n on 20-MHz channels offers a reliable balance of link budget and throughput. However, localized Wi-Fi coverage remains insufficient for long-range forwarding to remote control centers.

To extend backhaul coverage, hybrid and multi-hop architectures have received significant focus. Yao et al. [5] analyzed multi-UAV relay networks integrated with satellite backhauls for remote IoT telemetry, while Cakmakci [6] evaluated LoRa-satellite gateway topologies. LoRa provides low-power, long-range backhaul connectivity, yet its low transmission bit rate imposes strict channel capacity boundaries. Semtech’s physical specifications [7] and empirical studies by Adelantado et al. [8] characterize how spreading factors (SF), channel band-widths, and regional duty-cycle limitations constrain network throughput.

Queueing approximations frequently serve to evaluate such capacity constraints. Huh et al. [2] applied ideal M/D/1 benchmarks to evaluate multi-channel LoRa scheduling, while Sorensen et al. [9] examined deterministic service delays and gateway blocking in machine-type IoT traffic. Ever et al. [10] highlighted the necessity of modeling finite buffer thresholds to capture congestion in resource-constrained IoT relays. Extending these foundations, this study investigates the performance boundaries of a tandem Wi-Fi–LoRa UAV relay under asymmetric airtimes and finite buffering.

Although existing studies have demonstrated the feasibility of UAV-assisted communication, investigated heterogeneous wireless architectures, and developed queueing models to evaluate network performance, several limitations remain. Existing studies mainly address communication technologies, relay architectures, or queueing behaviour independently, offering limited insight into how their interactions affect end-to-end performance under demanding traffic conditions. In particular, the substantial service-rate asymmetry between high-throughput access technologies such as Wi-Fi and low-rate backhaul technologies such as LoRa has not been sufficiently examined in the context of finite relay buffers and time-critical disaster communications. Furthermore, the relationship between physical-layer transmission airtime, queue congestion, packet delivery, and end-to-end latency requires more systematic investigation using complementary analytical and simulation-based approaches. To address these limitations, this study integrates an airtime-derived tandem queueing framework with packet-level network simulations to characterise the performance boundaries of a UAV-assisted Wi-Fi–LoRa architecture, with particular emphasis on identifying communication bottlenecks and evaluating trade-offs among buffer capacity, packet delivery reliability, and latency.

The proposed network topology is depicted in Fig. 1. The infrastructure comprises dispersed ground survivor nodes, a dual-interface UAV relay, a terrestrial LoRa gateway, and an emergency network server.

Ground nodes transmit compact telemetry frames over IEEE 802.11n Wi-Fi at 2.4 GHz. The UAV functions as an intermediate store-and-forward relay. The packets received at the Wi-Fi interface are ingested, placed in an internal buffer, and retransmitted over the 868-MHz LoRa physical interface toward the remote ground gateway.

To minimize medium access contention and transmission airtime, we structure an 11-byte fixed-size payload (Table I). Latitude and longitude coordinates are quantized into 24-bit integer fields, providing sub-meter geographic accuracy while minimizing packet overhead.

The access stage utilizes IEEE 802.11n HT mixed-mode at 2.4 GHz over a 20-MHz channel with a single spatial stream (MCS0, nominal PHY rate 6.5 Mb/s). The 11-byte payload is encapsulated within standard UDP, IPv4, and 802.11 MAC headers, resulting in an 81-byte physical frame in the air. Under baseline analytical conditions, this corresponds to an idealized transmission service time T_W = 0.142 ms, yielding an aggregate nominal service rate μ_W = 1/T_W ≈ 7,042 packets/s.

The relay backhaul operates at 868 MHz with a bandwidth BW = 125 kHz and coding rate CR = 4/5. Transmission time-on-air (ToA) is governed by spreading factor (SF), preamble length (N_pre), and total physical payload size (PL) [7]. The symbol duration T_sym and preamble duration T_pre are:

T_sym = 2^SF / BW    (1)

T_pre = (N_pre + 4.25) T_sym    (2)

The payload symbol count N_pay is determined by:

Γ = (8PL − 4SF + 28 + 16C − 20H) / [4(SF − 2D)]    (3)

N_pay = 8 + max{0, ⌈Γ⌉ (CR + 4)}    (4)

where H = 0 denotes an explicit header, C = 1 enables the cyclic redundancy check (CRC), and D = 1 enables low data rate optimization when T_sym > 16 ms. Total LoRa airtime T_L is:

T_L = (N_pre + 4.25 + N_pay) T_sym    (5)

For an encapsulated LoRa frame with PL = 28 bytes (N_pre = 8, BW = 125 kHz), the resulting symbol duration at SF = 7 is 1.024 ms, yielding N_pay = 53 symbols and a total airtime T_L = 66.816 ms, which matches our FLoRa simulation trace. Table 2 compiles calculated airtimes and deterministic service rates μ_L = 1/T_L across SF ∈ {7, 8, 9}.

The end-to-end network is abstracted as a two-stage tandem queueing model (Fig. 2). The first stage models Wi-Fi ingestion at the UAV, while the second stage models LoRa backhaul transmission. Both queues operate under a First-In-First-Out (FIFO) discipline with deterministic airtime service durations.

Assuming N independent ground nodes each generating reports at mean rate r via a Poisson process, the aggregate arrival rate to the Wi-Fi stage is:

λ₁ = Nr    (1)

Denoting p₁ and p₂ as the packet delivery probabilities across the Wi-Fi and LoRa queues, respectively, the departure rate from Stage 1 and subsequent arrival rate to Stage 2 are:

d₁ = λ₁p₁,   λ₂ = d₁,   d₂ = λ₂p₂    (2)

With deterministic service times T_W and T_L, the traffic intensities are:

ρ₁ = λ₁ / μ₁ = λ₁T_W,   ρ₂ = λ₂ / μ₂ = λ₂T_L    (3)

Under stable conditions (ρ_i < 1), the mean queue waiting time for an infinite-capacity M/D/1 queue is given by the Pollaczek–Khinchine formula:

W_q,i = λ_iT_i² / [2(1 − ρ_i)]    (4)

The total stage delay is W_i = W_q,i + T_i. Hence, end-to-end transfer latency for successfully forwarded reports is approximated by:

D_E2E ≈ W₁ + W₂ + T_proc    (5)

where T_proc accounts for onboard MCU processing and interface switching.

To capture packet drops under congestion, we evaluate finite waiting-buffer capacities (L₁, L₂). The primary evaluated topology allocates a finite LoRa buffer of L₂ = 20 packets (total capacity K = L₂ + 1 = 21 packets including the server). Under an offered load of N = 20 devices each generating r = 1 packet/s (λ₁ = 20 packets/s) and assuming lossless Wi-Fi reception (p₁ = 1):

ρ₂ = 20 × 0.066816 ≈ 1.336    (6)

Because ρ₂ > 1, the arrival rate exceeds the physical LoRa service rate (μ_L ≈ 14.97 packets/s). In a finite-capacity system, this saturation directly manifests as queue overflow losses, identifying the LoRa link as the critical performance bottleneck.

Three approaches were used to evaluate the proposed communication system architecture, namely, OMNet++ network simulation, python based discrete event queueing simulation, and lastly an analytical M/D/1/K queueing analysis.

First, OMNeT++ 6 along with INET 4.4 and FLoRa 1.1.0 modules were used to check the packet level operation of our proposed architecture. In this implementation, we model IEEE 802.11n communication network between the ground devices and the UAV, followed by LoRa transmission to the ground station. A single device scenario was used to examine the end to end communication path, and then a 20 device scenario was used to examine Wi-Fi packet collection. This provides us with insight into the network level observations of packet reception, tramission airtime, and relay behaviour. Ground nodes generate 11-byte payloads via UdpBasicApp. The UAV runs UdpSink linked via an event-driven bridging module to SimpleLoRaApp, which transmits packets to the terrestrial gateway at 868 MHz (14 dBm transmission power).

Two operational scenarios were benchmarked:

1. Single-Node Baseline: A single node transmitting to a UAV at 50 m altitude to validate physical-layer timings and end-to-end stack traversal.
2. Multi-Node Scenario: 20 nodes randomly deployed across a 60 × 60 m field, each transmitting an emergency packet between t = 6 s and t = 8 s, evaluated over a 100-s simulation horizon.

Secondly, a python based discrete event simulation was developed to evaluate the two stage M/D/1 queue model under different traffic conditions. The arrival rate of packets were generated using a Poisson process with exponential distributed inter-arrival times, while the packet service times were deterministic and calculated using the Wi-Fi and LoRa tramission airtime equations. This simulation processes packets through FIFO queue. This approach was used to investigate packet delievery ratio (PDR), end to end delay, and congestion as the arrival rate, and LoRa spreading factor change.

And lastly, an analytical M/D/1/K model was also added in python to provide us with mathematical reference for the simulated LoRa queue. This method calculates the packet blocking probability using the arrival rate, deterministic service time, and finite system capacticy. The theoretical PDR is found using:

PDR_analytical = 1 − P_blocking.

This approach calculates queueing probablity directly.

Together, the three approaches provide complementary evidence: OMNeT++ examines network-level communication, the Python simulation evaluates queueing performance under varying traffic conditions, and the analytical model provides a theoretical reference for validating the simulated packet delivery results.¹

¹ https://github.com/arishaahmad/Tandem-Queueing-Analysis-and-Airtime-Validation-for-UAV-Assisted-Wi-Fi-LoRa-Disaster-Networks

In the single-node OMNeT++ run, the emergency report traversed both hops successfully, reaching the network server with an end-to-end delay of 67.461 ms. As shown in Fig. 3, the empirical physical transmission airtimes were 0.132 ms for Wi-Fi and 66.816 ms for LoRa at SF = 7. This matches the theoretical ToA formulation (66.816 ms in Table 2), highlighting the severe service-time disparity between the two interfaces (T_L / T_W ≈ 506).

In the 20-node deployment, the UAV successfully received all 20 incoming Wi-Fi reports without packet loss, confirming effective medium access performance over the local access link.

Fig. 4 illustrates the Packet Delivery Ratio (PDR) as a function of aggregate arrival rate λ₁. At low traffic rates (λ₁ < 10 packets/s for SF7), the system achieves near-lossless throughput (PDR ≈ 100%). However, as arrival rates exceed the LoRa physical service capacity μ_L, queue overflow occurs in the finite buffer (L₂ = 20). At an offered load of 20 packets/s, the delivered PDR drops to 75.24% for SF7, 40.74% for SF8, and 22.21% for SF9.

Fig. 5 illustrates the corresponding mean end-to-end delay for successfully delivered packets. As the LoRa buffer saturates, queue waiting times dominate end-to-end latency. At 20 packets/s, the mean latencies reach 1.293 s for SF7, 2.535 s for SF8, and 4.702 s for SF9. The elevated delay at higher spreading factors directly reflects the extended symbol duration (T_sym) and corresponding reduction in service rate.

Figs. 6 to 8 evaluate network performance across varying LoRa buffer capacities L₂. As shown in Fig. 6, the simulation closely matches the analytical M/D/1/K results for SF7, demonstrating agreement between the two approaches. Increasing buffer capacity improves PDR under marginal traffic loads. However, under severe overload (λ₁ >> μ_L), Fig. 7 demonstrates that expanding buffer capacity yields diminishing throughput improvements, while Fig. 8 shows a substantial increase in queueing delay. For time-critical emergency reporting, oversized buffers can therefore increase latency without overcoming the LoRa service-rate limitations.¹

¹ https://github.com/arishaahmad/Tandem-Queueing-Analysis-and-Airtime-Validation-for-UAV-Assisted-Wi-Fi-LoRa-Disaster-Networks

This study presented a UAV-assisted two-hop hybrid Wi-Fi–LoRa emergency network designed for rapid post-disaster communication. By deploying a compact 11-byte telemetry format, the architecture minimizes transmission overhead while preserving critical location and status telemetry. OMNeT++ simulations coupled with tandem M/D/1/K queueing models validate the physical airtime formulations (0.132 ms for Wi-Fi and 66.816 ms for LoRa at SF7). The results demonstrate that while IEEE 802.11n reliably ingests bursty multi-node reports, the LoRa backhaul constitutes the primary network bottleneck. Under heavy traffic (λ ≥ 20 packets/s), spreading factors and buffer allocations dictate performance: larger buffers slightly reduce packet loss at the cost of substantial queueing latency.

Future work will expand this framework by incorporating adaptive data-rate (ADR) algorithms on the LoRa link, evaluating multi-channel gateway reception, and implementing packet aggregation mechanisms at the UAV to balance throughput and latency in dynamic post-disaster deployments.

This work was supported by the International Science Partnerships Fund (ISPF) Research Collaboration Grants, British Council, under the Institutional Links collaboration “Enhancing Post-Earthquake Communication Resilience: Integrating UAVs, LoRa Technology, and GDPR-Compliant Systems (EPECR)” (Project Reference ID: 935Turkey, 2025-2027).