⚡ electrical engineering

QoS-Aware Disjoint Multipath Video Streamingfor Healthcare IoT

This paper addresses the challenge of QoS-aware medical video streaming in battery-powered Healthcare IoT networks by proposing HQ-DMVS, a polynomial-time heuristic algorithm that jointly optimizes source admission, node-disjoint multipath routing, and rate control to achieve near-optimal utility, significantly reduced delay, and extended network lifetime compared to existing methods.

Original authors: Nedal Ababneh, Khaled Almi’ani

Published 2026-07-01

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Nedal Ababneh, Khaled Almi’ani

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: The "Medical Video Traffic Jam"

Imagine a hospital network as a busy city where doctors and patients are connected by wireless roads. These roads are made of small, battery-powered relay stations (like streetlights that also act as traffic routers).

The goal is to send live medical video (like ultrasound or endoscopy feeds) from the patient to the specialist. This video is very sensitive:

It can't be blurry (needs high quality).
It can't be late (needs to be real-time).
It can't get lost (needs to arrive intact).

The Problem:
In the past, these networks treated video like regular email or text messages. They would pick the shortest route to the destination. But if that one road gets clogged, has a bad signal, or runs out of battery power, the video freezes or pixelates. Since the relays run on batteries, if one relay works too hard, it dies, and the whole network loses a piece of its map.

The Solution: HQ-DMVS (The "Smart Traffic Controller")

The authors created a new system called HQ-DMVS. Think of this as a super-smart traffic control center that manages three things at once, rather than separately:

Who gets in the city? (Admission): Deciding which video streams are important enough to let in.
Which roads do they take? (Routing): Finding paths that don't share any intersections.
How fast do they drive? (Rate Control): Adjusting the speed of the video based on how much "fuel" (battery) and "road space" (bandwidth) is left.

The Core Innovation: "No Sharing the Middle"

The most important rule of this new system is Node-Disjoint Routing.

Old Way: Imagine two ambulances (Video Stream A and Video Stream B) trying to get to the hospital. The old system might send them both through the same small bridge (Relay Station X). If that bridge collapses or the battery dies, both ambulances are stuck.
HQ-DMVS Way: The system forces the two ambulances to take completely different routes. They might start at different places and end at the same hospital, but they never share a single intermediate bridge.
- Analogy: It's like sending one delivery truck via the highway and another via the back roads. If the highway has an accident, the back-road truck still gets through. This ensures that if one relay fails, it only kills one video stream, not all of them.

How It Works: The Two-Phase Algorithm

Since calculating the perfect route for every camera is mathematically impossible to do instantly (it's too complex, like solving a Sudoku puzzle with a million squares), the authors built a two-step "smart guess" system:

Phase 1: The "Widest Path" Hunt
The system looks at all the available roads and picks the path that has the most "room" (bandwidth) and the most "fuel" (battery) left. It does this for the most important cameras first. Once a path is picked, it "blocks off" the middle bridges so no other camera can use them. This guarantees the "no sharing" rule.

Phase 2: The "Speed Boost"
Once the paths are set, the system checks how fast each camera can actually go. It starts them at a slow, safe speed. Then, it looks for any extra room on the roads and gives a speed boost to the most important cameras first. It keeps doing this until the roads are full or the batteries are too low to go faster.

The Results: Why It Matters

The authors tested this system in a computer simulation with up to 200 relay stations. Here is what they found:

It's Almost Perfect: The system achieves 93.6% of the performance of the "perfect" mathematical solution (which takes hours to calculate), but it does it in less than a tenth of a second.
Faster Delivery: It cuts the delay (lag) by about 32–41%. This means the doctor sees the video almost instantly.
Fewer Lost Packets: The video arrives intact 96% of the time, compared to much lower rates for older methods.
Better Battery Life: Because the system spreads the traffic out and avoids overworking single relays, the network lasts 54% longer before the first battery dies.
Clearer Video: The video quality (measured in PSNR) is 2.4 dB better than standard routing, meaning the image is sharper and easier to diagnose.

The Bottom Line

This paper presents a "traffic cop" for medical video networks that is fast, smart, and fair. By forcing different video streams to take completely separate paths and managing their speed based on battery life, it ensures that critical medical videos get through clearly and quickly, even when the network is crowded or batteries are running low.

The authors have made their code and data public so other researchers can verify these results and build upon them.

Technical Summary: QoS-Aware Disjoint Multipath Video Streaming for Healthcare IoT

1. Problem Statement

The paper addresses the challenge of streaming medical video over Healthcare Internet of Things (IoT) networks. Unlike scalar telemetry, medical video (e.g., endoscopy, ultrasound) requires strict Quality of Service (QoS) targets: high perceptual quality (PSNR ≥36 dB), low interactive delay (≤150 ms), high packet delivery ratios (≥0.95), and uninterrupted continuity.

Existing protocols often fail because they decouple three critical decisions:

Source Admission: Deciding which video streams to accept.
Routing: Determining the path data takes through battery-powered relays.
Rate Control: Assigning transmission rates to admitted sources.

Treating these separately leads to suboptimal outcomes, such as admitting a low-priority stream that starves a high-priority one, or operating at peak rates that prematurely deplete relay batteries. Furthermore, single-path forwarding is brittle; congestion or fading on a single link can degrade an entire stream. The problem is compounded by the need to maximize the "priority-weighted utility" of admitted streams while respecting link/node capacities and per-second residual energy constraints. The authors note that the joint formulation of these decisions is NP-hard.

2. Methodology

The authors propose HQ-DMVS (Healthcare-QoS Disjoint Multipath Video Streaming), a framework that unifies admission, routing, and rate assignment. The methodology consists of two components:

A. Exact Formulation (MILP)

The authors formulate the Joint Admission, Routing, and Rate (JARR) problem as a Mixed-Integer Linear Program (MILP).

Objective: Maximize the sum of priority-weighted utilities for admitted sources.
Constraints:
- Node-Disjointness: No two admitted sources share an intermediate relay (enforced by capping out-degree and in-degree at 1 for relays).
- Flow Conservation: Ensures data flows from sources to sinks without loss.
- Capacity: Respects link bandwidth and node forwarding limits.
- Energy Feasibility: Ensures traffic does not exceed the per-second residual energy of any relay, modeled using a first-order radio energy model (Heinzelman model).
Linearization: Although the utility function involves a bilinear term (admission variable $\times$ rate), the authors show it can be exactly linearized because the admission variable is binary and the rate is bounded, allowing standard branch-and-cut solvers (like CPLEX) to find the global optimum for small-to-medium networks.

B. Algorithmic Solution (HQ-DMVS)

Since the MILP is NP-hard and becomes intractable for large networks (e.g., >60 nodes), the authors propose HQ-DMVS, a two-phase polynomial-time heuristic:

Phase 1: Disjoint Multipath Routing (HQ-DMVS-R):
- Sources are processed in decreasing order of priority.
- For each source, the algorithm computes a "widest-bottleneck" path on a residual graph.
- The path weight combines bandwidth slack and residual energy slack, controlled by a mixing parameter $\alpha$ .
- Once a path is selected, its intermediate nodes are removed from the graph to guarantee node-disjointness for subsequent sources.
Phase 2: QoS-Aware Rate Assignment (HQ-DMVS-A):
- Admitted sources start at a minimum rate ( $R_{min}$ ).
- The algorithm iteratively increases rates by a quantum $\delta$ for the source with the highest priority-weighted marginal utility gain, provided capacity and energy constraints are met.

3. Key Contributions

Unified Optimization: The paper is the first to jointly model binary source admission, node-disjoint multipath routing, continuous rate assignment, and explicit per-second residual energy feasibility in a single optimization framework for healthcare IoT.
Structural Guarantees: The proposed algorithm constructs node- and edge-disjoint paths by design, eliminating failure correlation between independent medical streams.
Theoretical Bounds: The authors derive a conditional additive optimality envelope. If the heuristic and the MILP admit the same set of sources, the heuristic's solution is guaranteed to be within $|V_S| P_{max} \delta U_{step}$ of the MILP optimum.
Reproducibility: The authors release the simulator, AMPL/CPLEX models, and raw logs as open artifacts.

4. Experimental Results

The framework was evaluated on random geometric topologies (up to 200 nodes) using an IEEE 802.11n PHY/MAC model and a first-order radio energy model. Results are based on 30 seeded runs and compared against the MILP optimum, Shortest-Path Routing (MRRA-SP), and Path-Capacity Routing (MRRA-PC).

Key Performance Metrics (at $|V|=50$ , 20 sources):

Utility: HQ-DMVS achieves 93.6% of the MILP optimum utility (statistically significant, $p < 10^{-4}$ ).
Delay: Reduces end-to-end delay by 32–41% compared to MRRA-SP (47 ms vs. 110 ms).
Packet Delivery Ratio (PDR): Achieves ≥0.96, significantly outperforming baselines (MRRA-SP: 0.84).
Bandwidth Utilization: Improves by 18–24% over baselines.
Video Quality: Gains 2.4 dB PSNR over shortest-path routing.
Network Lifetime: Extends the "first-node-dies" time by up to 54%.
Scalability: The algorithm runs in ≤112 ms for 200-node networks, making it suitable for periodic re-planning.

Sensitivity Analysis:

The algorithm is robust to energy heterogeneity and traffic burstiness.
Performance degrades gracefully when controller snapshots become stale (up to 8 seconds).
The mixing parameter $\alpha$ (balancing bandwidth vs. energy) shows a flat optimum in the range [0.5, 0.7].

5. Significance and Claims

The paper claims that HQ-DMVS bridges the gap between theoretical optimality and practical deployability in healthcare IoT.

Clinical Relevance: By meeting strict QoS targets (PSNR, delay, PDR) and extending network lifetime, the framework supports diagnostic-grade video streaming in resource-constrained environments like ambulances and community clinics.
Deterministic Safety: Unlike learning-based approaches (e.g., Reinforcement Learning) which rely on latent policies that are hard to audit, HQ-DMVS is deterministic. Every admission and rate decision is traceable to specific algorithmic steps, which is crucial for clinical regulatory compliance.
Operational Feasibility: The authors assert that the algorithm can run on low-cost hardware (e.g., Raspberry Pi 4) and fits within the control loops of standard network controllers, supporting periodic re-optimization as link conditions drift.
Generalizability: While focused on healthcare, the authors note the framework applies to any application involving priority-differentiated multimedia, battery-powered relays, and capacity-variable wireless links (e.g., industrial IoT, smart-city surveillance).

The paper concludes that decoupling routing from rate control leaves a significant performance gap (13–36%), whereas the joint approach of HQ-DMVS closes this gap to within 10% of the theoretical optimum while maintaining polynomial-time complexity.