Zero-Delay Source-Channel Coding With a Low-Resolution ADC Front End

Morteza Varasteh, Borzoo Rassouli, Osvaldo Simeone, Deniz GündüzIEEE Trans. Information Theory 2018 · DOI · OpenReview

TL;DR

This paper tackles the challenge of transmitting analog data efficiently over wireless channels when the receiver is equipped with low-resolution Analog-to-Digital Converters (ADCs), common in power-constrained IoT devices. The authors jointly optimize the encoding and decoding strategies for zero-delay transmission, revealing that optimal strategies deviate significantly from traditional linear methods, especially with one-bit ADCs.

What Problem Does This Paper Address?

Modern wireless systems, particularly in the Internet of Things (IoT) and sensor networks, face a critical constraint: power consumption. Analog-to-Digital Converters (ADCs), which convert real-world analog signals into digital ones, are significant power consumers, with their power usage growing exponentially with resolution. This has led to a strong interest in low-resolution ADCs, especially one-bit ADCs, which are simple and power-efficient but drastically limit the information that can be captured.

Transmitting analog data over a noisy channel with a low-resolution ADC at the receiver presents a complex joint source-channel coding problem. Traditional approaches often assume high-resolution ADCs and linear encoding/decoding. However, with low-resolution ADCs, the optimal strategies are unknown and likely non-linear. This paper addresses this gap by investigating the optimal design of encoders and decoders for transmitting a Gaussian source over a SIMO AWGN channel with a low-resolution ADC front end, focusing on zero-delay transmission (i.e., transmitting a single sample at a time). The authors consider two key performance metrics: Mean Squared Error (MSE) distortion and Distortion Outage Probability (DOP).

Methods

The core of this work lies in the joint optimization of the encoder (mapping the source signal to the channel input) and the decoder (estimating the original source signal from the quantized channel output). The authors formulate this as an optimization problem, minimizing a weighted sum of distortion and average transmission power, controlled by a Lagrange multiplier.

Key methodological contributions include:

  • Derivation of Optimal Mappings for One-Bit ADCs: The paper provides explicit (though implicit in some cases) optimal encoder and decoder mappings for the simplest case of a one-bit ADC under both MSE and DOP criteria.
  • Analysis of Encoder Behavior: For MSE distortion, they show that the optimal encoder is generally non-linear. It approaches a linear encoder in the low Signal-to-Noise Ratio (SNR) regime and an antipodal digital encoder in the high SNR regime. This is a significant departure from the always-linear optimal encoder for full-precision ADCs. For DOP, the optimal encoder is found to be piecewise constant, taking only two opposite values.
  • Development of Necessary Optimality Conditions: For more general $K$-level ADCs and multiple one-bit ADCs (N observations), the authors derive necessary optimality conditions. These conditions are crucial for developing algorithms to find near-optimal solutions.
  • Numerical Optimization: Leveraging these conditions, the authors employ a gradient-descent algorithm to numerically obtain optimized encoder and decoder mappings for various ADC configurations and SNRs.

Datasets

Not discussed in the paper. The paper focuses on theoretical analysis and numerical simulations based on synthetic Gaussian sources and AWGN channels.

Key Insights and Ideas

  • Low-Resolution ADCs Fundamentally Change Optimal Strategies: The assumption of linear encoding, optimal for full-precision ADCs, breaks down with low-resolution ADCs. The optimal encoder becomes non-linear and highly dependent on the SNR.
  • One-Bit ADCs Lead to Distinct Behaviors: For MSE, the optimal encoder saturates at high SNRs, resembling digital transmission. For DOP, it becomes a simple two-value piecewise constant function.
  • SNR Dictates Optimal Architecture: At low SNRs, increasing the number of observations (multiple ADCs) is more beneficial for improving performance than increasing ADC resolution. Conversely, at high SNRs, higher ADC resolution is more advantageous.
  • Trade-off Between Resolution and Observations: The paper highlights a fundamental trade-off: investing bits in higher ADC resolution versus using those bits to increase the number of independent measurements from the channel.

Key Results

  • MSE Distortion:
    • For a one-bit ADC, the optimal encoder is generally non-linear. It tends towards linear at low SNR and antipodal digital at high SNR.
  • Numerical results show that for K-level ADCs, the optimal mappings approach linear behavior at low SNR and digital-like behavior at high SNR.
  • For multiple one-bit ADCs, the linear behavior around the origin persists even at higher SNRs, unlike the single ADC case.
  • DOP Criterion:
    • For a one-bit ADC, the optimal encoder is piecewise constant and takes only two opposite values.
  • The optimal reconstruction points for the decoder tend towards zero at low SNR and $\pm\sqrt{D}$ at high SNR.
  • Numerical results for DOP demonstrate that at low SNR, multiple one-bit ADCs outperform a single higher-resolution ADC for the same total number of output bits. At high SNR, the opposite is true.

Limitations and Future Work

  • Joint Optimization of Quantization Intervals: The paper assumes fixed quantization intervals for the ADCs. A significant future direction is the joint optimization of these intervals along with the encoder and decoder functions.
  • Fading Channels: The current work focuses on AWGN channels. Extending the analysis to fading channels would be a valuable next step.
  • Correlated Sources/Noise: The analysis assumes independent Gaussian sources and AWGN. Investigating scenarios with correlated sources or noise could lead to further insights.
  • Algorithm Convergence: The gradient-descent algorithm used for numerical optimization is not guaranteed to converge to a global optimum.
  • Complexity: While low-resolution ADCs are power-efficient, the optimal encoder and decoder mappings can be complex, potentially requiring significant computational resources at the transmitter or receiver.

Summary updated

Abstract

Motivated by the practical constraints arising in emerging sensor network and Internet-of-Things (IoT) applications, the zero-delay transmission of a Gaussian measurement over a real single-input multiple-output (SIMO) additive white Gaussian noise (AWGN) channel is studied with a low-resolution analog-to-digital converter (ADC) front end. Joint optimization of the encoder and the decoder mapping is tackled under both the mean squared error (MSE) distortion and the distortion outage probability (DOP) criteria, with an average power constraint on the channel input. Optimal encoder and decoder mappings are identified for a one-bit ADC front end under both criteria. For the MSE distortion, the optimal encoder mapping is shown to be non-linear in general, while it tends to a linear encoder in the low signal-to-noise ratio (SNR) regime, and to an antipodal digital encoder in the high SNR regime. This is in contrast to the optimality of linear encoding at all SNR values in the presence of a full-precision front end. For the DOP criterion, it is shown that the optimal encoder mapping is piecewise constant and can take only two opposite values when it is non-zero. For both the MSE distortion and the DOP criteria, necessary optimality conditions are then derived for $K$ -level ADC front ends as well as front ends with multiple one-bit ADCs. These conditions are used to obtain numerically optimized solutions. Extensive numerical results are also provided in order to gain insights into the structure of the optimal encoding and decoding mappings.
BibTeX
@article{varasteh2018zero,
  author = {Morteza Varasteh and Borzoo Rassouli and Osvaldo Simeone and Deniz Gündüz},
  title = {{Zero-Delay Source-Channel Coding With a Low-Resolution ADC Front End}},
  journal = {IEEE Trans. Information Theory},
  year = {2018},
  doi = {10.1109/tit.2017.2719708},
}
Figure