Home▸Signals & Telecommunications▸AI Cognitive Radio: 2D Bandit Spectrum-Access Simulator

AI Cognitive Radio: 2D Bandit Spectrum-Access Simulator

2D companion to the 3D cognitive-radio waterfall, computed independently: a Q-learning agent picks channels against configurable i.i.d. or bursty Markov primary-user traffic, with a scrubbable spectrum-history strip, a live Q-value bar chart and a rolling throughput/collision strip chart — channel count, learning rate, exploration decay and traffic burstiness are all tunable.

Signals & Telecommunications2DModerate60 FPS📱 Mobile-adapted⇄ 3D version

Details

CategorySignals & Telecommunications

RenderingCanvas 2D

DifficultyModerate

Target FPS60

ControlsSliders for channel count, primary-user traffic load, traffic persistence, agent exploration rate (epsilon), Q-learning rate (alpha), simulation speed and visible waterfall history; toggle buttons for i.i.d. vs bursty Markov traffic and epsilon decay; a Reset learning button; drag the waterfall vertically to scrub history (mouse wheel zooms the visible window) and a Jump to live button to snap back.

cognitive radiodynamic spectrum accessQ-learning6Gwireless AIreinforcement learningmulti-armed banditMarkov traffic

Related simulations

AI Attack Path Planner 2D: Q-Learning on a Network GraphA pannable 2D view of a reinforcement-learning agent exploring a network graph from a DMZ…

AI Escalation Policy Learner (2D) — Contextual Bandit for Human Handoff2D pannable/zoomable flow diagram of a contextual bandit learning, ticket type by ticket…

Multi-Armed Bandit — Marketing Budget AllocatorWatch epsilon-greedy, UCB1 and a random baseline compete to allocate a marketing budget… ⊞ All 28 in Signals & Telecommunications →

An AI secondary-user agent shares a bank of radio channels with unpredictable primary-user traffic, learning which channels tend to be free using tabular Q-learning and ε-greedy exploration. This 2D companion to the 3D spectrum waterfall computes the same dynamic-spectrum-access mechanism independently, and goes further: primary-user traffic can be switched from independent per-slot draws to a bursty two-state Markov chain with a tunable persistence, and exploration can be set to decay automatically as the agent gains experience. Three linked panels — a scrubbable spectrum-history strip you can drag through, a live Q-value bar chart highlighting the best-learned channel, and a rolling throughput/collision strip chart — make the bandit learning process visible in ways a single 3D scene can't.

⚙ Under the hood

cognitive radiodynamic spectrum accessQ-learning6Gwireless AIreinforcement learningmulti-armed banditMarkov traffic

2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install