Maestro QoE-Aware Dynamic Resource Allocation in Wi-Fi

Dynamic resource allocation that lifts QoE without touching clients

Maestro: QoE-aware dynamic resource allocation in Wi-Fi networks

Telemetry-only QoE estimators drive a DDQN multi-agent policy that jointly selects access category and traffic priority per flow, yielding higher QoE and fairness across Wi-Fi deployments.

Telemetry-driven QoE DDQN multi-agent RL Works on existing AP + controller No client or app changes Read the paper

25× QoE over WMM in the testbed

+78% fairness QoE fairness vs. WMM

+69% QoE vs. QFlow without client hooks

+12% OTT QoE lift on Netflix/Zoom (up to 29%)

Problem: Poor User Experience in Wi-Fi Networks

How to Allocate Resources to Improve User QoE?

What is the impact of Wi-Fi configuration parameters on user QoE of immersive applications?.
How can we automatically set the access category and priority queues to improve QoE and QoE fairness?.

Limitations of existing solutions

Static policies (e.g., WMM) cannot handle dynamic workloads and conditions.

QoS fails to capture user satisfaction (or QoE)!.

Existing works require modifying UE/client limiting deployability.

Approach

How Maestro works

1 · Sense

Telemetry → QoE

257 controller metrics sampled every 4s feed an LSTM (256/128/64) trained for 50 epochs with MSE + Adam to estimate QoE per app class.

2 · Decide

Policy agent

Multi-agent DDQN observes estimated QoE + telemetry and chooses one of six action pairs (LQ/HQ × BE/VI/VO) per flow every 16s.

3 · Learn fast

Simulation harness

Knowledge base (402 hours, 362K datapoints across 15 configs) fuels off-policy training with Boltzmann exploration and replay buffers to speed convergence.

4 · Deploy

Controller integration

Flow classifier tags app class; Maestro pushes access category + priority queue to the controller/AP without any firmware or client modifications.

Evaluation highlights

Testbed

Aruba controller/AP with mixed clients and servers used to validate Maestro.

QoE and QoE Fairness

QoE lift per application class and Jain fairness across different application mixes.

OTT services

Real Internet traffic QoE gains.

Why it matters

Better QoE without invasive hooks

Everything runs from controller telemetry, keeping clients/apps untouched while boosting experience and fairness.

Generalizes to real OTT apps

Validated with Netflix and Zoom traffic: +12% QoE on average, up to 29%, even with unseen flows.

Practical for operators

Runs atop existing Aruba controller/AP stack, configuring only flow-level knobs (access category + priority queue).

← Back to Research