On the Practical Performance of Noise Modulation for Ultra-Low-Power IoT: Limitations, Capacity, and Energy Trade-offs

1. Problem Statement

The proliferation of Ultra-Low-Power (ULP) Internet of Things (IoT) devices demands communication architectures that minimize energy consumption to support battery life or energy harvesting. Traditional RF architectures rely on power-hungry components like local oscillators (LO), phase-locked loops (PLL), and high-resolution ADCs, creating a bottleneck for scalability.

Noise Modulation (NoiseMod) has emerged as a minimalist alternative that encodes data via the statistical variance (power level) of a noise-like signal rather than the amplitude, phase, or frequency of a coherent carrier. While theoretically promising for eliminating oscillators, there is a lack of grounded, practical comparative evaluations. Key gaps include:

• How does NoiseMod compare to legacy standards (BPSK, NC-FSK) in terms of true energy consumption and capacity?
• How does Rayleigh fading affect non-coherent NoiseMod, and what mitigation strategies are required?
• What are the trade-offs between oversampling (for variance estimation) and system throughput/energy?

2. Methodology

The authors conducted a rigorous comparative study combining analytical derivations, energy modeling, and numerical simulations.

• System Model:
  • Hypothesis Testing: $H_0$ (Silence/Background Noise) vs. $H_1$ (Transmitted Noise with power $P$).
  • Detection: An optimal energy detector using a Likelihood Ratio Test (LRT) calculates the sum of squared samples ($T = \sum y_n^2$) to determine the bit.
  • Channels: Performance was analyzed under Additive White Gaussian Noise (AWGN) and Frequency-Flat Rayleigh Fading.
• Analytical Derivations:
  • Derived the optimal detection threshold ($\gamma_{th}$) and Bit Error Rate (BER) expressions for both AWGN and Rayleigh channels.
  • Modeled the impact of Selection Diversity (2-antenna) to mitigate fading.
• Energy Modeling:
  • Developed a granular energy consumption model ($E_{bit}$) that separates static circuit power ($P_{ckt}$) from dynamic ADC power.
  • Crucially, the model accounts for the fact that NoiseMod requires oversampling ($N$ samples per bit), making ADC energy linearly dependent on $N$.
• Simulation Scenarios:
  • Monte Carlo simulations were run across ISM bands (2.4 GHz, 5.725 GHz, 24 GHz).
  • Benchmarked against Coherent BPSK (gold standard for efficiency) and Non-Coherent FSK (NC-FSK) (common in BLE).
  • Parameters included varying Signal-to-Noise Ratios (SNR), sample sizes ($N \in \{10, 50, 100\}$), and transmission distances.

3. Key Contributions

1. Theoretical Bounds: Analytically proved that non-coherent NoiseMod suffers a catastrophic error floor in fading environments without Channel State Information (CSI). Demonstrated that 2-antenna selection diversity is mandatory to restore performance.
2. ADC-Aware Energy Model: Established a three-way trade-off between BER performance, SNR requirements, and hardware energy costs. Showed that increasing sample size ($N$) to improve variance estimation linearly increases ADC energy consumption.
3. Capacity Analysis: Quantified that NoiseMod's reliance on oversampling fundamentally limits spectral efficiency to $1/N$ bits/symbol, restricting it to low-data-rate regimes compared to coherent schemes.
4. Energy Crossover Distance: Identified a critical distance threshold where the energy efficiency of NoiseMod flips. Below this distance, its low circuit power wins; beyond it, the high radiated power needed to overcome its SNR penalty makes it less efficient than BPSK.

4. Key Results

A. Performance in AWGN

• SNR Penalty: To achieve a target BER of $10^{-3}$, NoiseMod suffers an 8 dB SNR penalty compared to NC-FSK and an even larger gap compared to BPSK.
• Sample Size Impact: Increasing the number of integrated samples ($N$) improves BER but drastically reduces throughput (spectral efficiency). For example, $N=100$ yields better reliability but limits capacity to 0.01 bits/symbol.

B. Vulnerability to Fading

• Error Floor: In Rayleigh fading, the non-coherent detector cannot distinguish between a deep fade and a logical '0', causing the BER curve to flatten (error floor).
• Mitigation: Implementing 2-antenna selection diversity successfully removes the error floor, restoring the "waterfall" BER curve.

C. Energy and Distance Trade-offs

• Circuit Power: NoiseMod drastically reduces baseline circuit power (approx. 0.6 mW total vs. 4.5 mW for BPSK) by eliminating oscillators and mixers.
• Crossover Points: Due to the high SNR penalty, NoiseMod requires significantly more radiated power ($P_{PA}$) as distance increases. The "energy crossover distance" (where NoiseMod becomes less efficient than BPSK) is highly frequency-dependent:
  • 2.4 GHz: ~41.7 meters
  • 5.725 GHz: ~17.5 meters
  • 24 GHz: ~4.2 meters
• Conclusion: NoiseMod is superior only for ultra-short-range, near-field applications. As frequency increases or distance extends, the path loss penalty outweighs the circuit savings.

5. Significance and Implications

This paper provides a critical reality check for the deployment of NoiseMod in IoT. While the oscillator-free design offers a compelling path to ultra-low power, the study reveals that:

• It is not a universal solution: NoiseMod is strictly limited to short-range, low-frequency, low-data-rate scenarios.
• Diversity is non-negotiable: Practical deployment in real-world (fading) environments requires spatial diversity (multiple antennas), adding hardware complexity that partially offsets the "minimalist" benefit.
• Design Guidance: System designers must calculate the specific "energy crossover distance" for their application frequency. If the link exceeds this distance, traditional coherent schemes (like BPSK) are more energy-efficient despite their complex circuitry.

The work bridges the gap between theoretical noise modulation concepts and practical engineering constraints, offering a clear framework for when to adopt NoiseMod and when to stick with conventional modulation.