AI-Driven Green Cognitive Radio for Sustainable Spectrum Management

Shujaatali Badami

Home / Research / Green Cognitive Radio

Cognitive radio, that finally counts the watts.

Deep reinforcement learning over partially observable spectrum, with energy-aware reward shaping that turns spectrum efficiency and power efficiency into the same optimisation problem.

Shujaatali Badami ·Co-author ·Bentham CYBPRO · 2025 ·1 December 2025 ·~6 min read

At a glance

VenueBentham CYBPRO2025 (accepted)

MethodDeep RLpartial observability

RewardEnergy-awarecomposite throughput-per-watt

TargetsB5G / 6Gdense heterogeneous deployments

Co-authorsMultipleco-author position

Companion paperRL CSMsame POMDP playbook

Conference ↗ Cite All publications

Abstract, verbatim

Cognitive radio promises dynamic spectrum access, but most existing implementations treat spectrum efficiency and energy efficiency as separable objectives. This paper introduces an AI-driven green cognitive radio framework that unifies the two through energy-aware reward shaping in a deep reinforcement learning controller. The agent observes partial spectrum state through cooperative sensing, selects channels and transmit power jointly, and is rewarded by a composite metric weighting throughput against energy expenditure. Simulation results demonstrate substantial gains in both spectrum utilisation and energy efficiency over baseline DSA strategies, with particularly strong performance in dense, heterogeneous deployments characteristic of beyond-5G and 6G networks.

01Why cognitive radio still wastes energy

Cognitive radio was supposed to be the answer to spectrum scarcity: secondary users opportunistically inhabit unused primary-user channels, dynamic spectrum access replaces static licensing, total system throughput rises. The story works.

The story leaves out energy. Most CR implementations optimise for spectrum efficiency, then bolt energy considerations on as a constraint. The result is high-throughput radios that drink batteries. In an era where most CR endpoints are battery-powered IoT, that is the wrong trade.

02Energy-aware reward shaping

The contribution is a reward function that treats throughput and energy as a single composite metric, not as a primary-and-constraint pair. The DRL agent observes partial spectrum state through cooperative sensing, then jointly selects channel and transmit power. The reward weights throughput against power expenditure with an explicit Pareto coefficient.

The architecture inherits the POMDP-handling logic from earlier RL work: heuristic-bounded exploration, softmax sampling, instance-density adaptation. The novelty is in the reward, not the learner.

03What the simulations show

"Spectrum efficiency and energy efficiency are not opposing constraints. The right reward function reveals they are the same problem under different lighting."

Simulation across dense, heterogeneous deployments characteristic of beyond-5G and 6G networks shows substantial gains on both axes. The improvement is largest in the regime where conventional CR struggles hardest: many secondary users, fast primary-user dynamics, tight battery budgets.

04Where this connects

The DRL controller is the cognitive-radio cousin of the CSM algorithm in the IEEE ICEACE 2024 paper, applied to a different POMDP. The energy-awareness story rhymes with the AI-optimised VLSI work: at every layer of the stack, treating energy as a first-class optimisation target rather than an afterthought changes the achievable envelope.

FAQWhat people ask me about this paper

Q1How does this differ from existing energy-efficient CR work?

Existing approaches typically optimise spectrum efficiency under a fixed energy budget, or vice versa. This work treats them as a single composite objective with an explicit Pareto coefficient, so the controller can move along the trade-off curve at runtime.

Q2Why deep RL instead of classical control?

Cognitive radio is a POMDP with high-dimensional observations from cooperative sensing. Deep RL handles the dimensionality classical control flounders at, and the energy-aware reward fits its function-approximation strengths.

Q3What primary-user model does this assume?

The framework is largely model-agnostic on PU dynamics; the simulation uses standard PU activity models from the literature. The DRL controller adapts to whatever PU pattern it sees.

Q4Is this 5G, 6G or something else?

The framework is technology-neutral but tuned in simulation against beyond-5G and 6G deployment patterns: dense, heterogeneous, energy-constrained. The principles transfer down to LoRa and up to mmWave.

Q5How does this connect to the rest of my work?

Same POMDP playbook as CSM. Same energy-as-first-class-citizen instinct as the AI-optimised VLSI paper. Same end-to-end IoT vision as the quantum-resilient metering work.

CITEHow to cite this paper

@inproceedings{badami2025gcr,
  author    = {Shujaatali Badami and others},
  title     = {AI-Driven Green Cognitive Radio for Sustainable Spectrum Management},
  booktitle = {Bentham CYBPRO 2025},
  year      = {2025},
  publisher = {Bentham Science}
}

S. Badami et al., "AI-Driven Green Cognitive Radio for Sustainable Spectrum Management," in Bentham CYBPRO 2025, 2025.

Badami, S., et al. (2025). AI-Driven Green Cognitive Radio for Sustainable Spectrum Management. In Bentham CYBPRO 2025.

TY  - CONF
AU  - Badami, Shujaatali
TI  - AI-Driven Green Cognitive Radio for Sustainable Spectrum Management
T2  - Bentham CYBPRO 2025
PB  - Bentham Science
PY  - 2025
ER  -

At a glance

01Why cognitive radio still wastes energy

02Energy-aware reward shaping

03What the simulations show

04Where this connects

FAQWhat people ask me about this paper

CITEHow to cite this paper

SEE ALSORelated work in this portfolio

Compressed Suffix Memory for POMDPs

AI-optimised VLSI for sustainable IoT

Quantum-resilient IoT energy metering