Energy Efficiency

---
name: energy_efficiency
description: Evaluates the energy efficiency of a wireless power transmission system where an energy source learns optimal transmit power levels for energy-harvesting nodes using a stochastic multi-armed bandit algorithm without channel state information. Use when the user has predictions and gold and needs to compute EE.
metadata:
  skill_kind: metric
  source_arxiv: 2012.15203
  bibtex_key: ghosh2020learning
  confidence: high
---

# energy_efficiency

> Learning to Optimize Energy Efficiency in Energy Harvesting Wireless Sensor Networks — Ghosh et al. (2020) (arXiv:2012.15203, 2020)

## What this evaluates

Evaluates the energy efficiency of a wireless power transmission system where an energy source learns optimal transmit power levels for energy-harvesting nodes using a stochastic multi-armed bandit algorithm without channel state information.

## Datasets

- **Simulated Rayleigh Fading EH Network** — total ?; splits: test (10000)

## Metrics

- `EE` **(primary)** — range: continuous
  - Ratio of successful data transmission rate to total energy consumption. Computed as the average EE over T time slots, comparing the proposed UCB-EH algorithm against a max-power benchmark and an oracle scheme.

## Input / output format

**Input**: System state includes number of nodes k, available transmit power levels P={0..30} dBm, target rate r0, and binary feedback indicating successful decoding.

**Output**: Transmit power level p ∈ P selected for each time slot t.

## Scoring recipe

```python
def compute_ee(predictions, gold):
    successful = [1 if decode_success(p) else 0 for p in predictions]
    total_energy = sum(energy_cost(p) for p in predictions)
    ee = sum(successful) / total_energy
    return ee
```

## Common pitfalls

- EE is highly sensitive to the target rate r0; the optimal r0 differs between schemes (0.75 bpcu for UCB-EH vs 2 bpcu for benchmark).
- CSI-based benchmarks must account for CSI acquisition energy cost; ignoring it overestimates their performance.
- Simulations assume specific Rayleigh fading variances and path loss exponent γ=2.5, limiting direct generalization to other channel models.

## Evidence (verbatim from paper)

> We empirically evaluate the performance of our proposed algorithm [UCB-EH] in terms of average EE and compare it with a benchmark scheme without CSI. In the benchmark scheme without CSI the source always transmits with the maximum power to charge the EH nodes, i.e., it transmits with power 30 dBm. We define the oracle scheme as the scheme based on the optimal policy which maximizes the energy efficiency of the system. The aim is our algorithm to learn the oracle scheme with time. We measure the performance of our algorithm for T=10^4 time slots and the results are averaged over 1000 repetitions.

## Citation

```bibtex
@misc{ghosh2020learning,
  title={Learning to Optimize Energy Efficiency in Energy Harvesting Wireless Sensor Networks},
  author={Ghosh et al. (2020)},
  year={2020},
  note={arXiv:2012.15203}
}
```

- arXiv: 2012.15203