In modern wireless communication, particularly Massive MIMO (Multiple-Input Multiple-Output), the energy consumption of high-resolution analog-to-digital converters (ADCs) and digital-to-analog converters (DACs) has become a critical bottleneck. As Sohrabi et al. and Sankar and Chepuri note, scaling these systems to hundreds of antennas requires a radical shift toward low-resolution hardware—often as extreme as 1-bit quantization—to maintain power efficiency.
However, reducing resolution to a single bit introduces severe non-linear distortion. A 1-bit converter effectively acts as a sign function, discarding all magnitude information and leaving only the phase (or polarity) of the signal. This transformation invalidates the linear assumptions that underpin traditional signal processing.
The Geometry of 1-Bit Distortion
The fundamental challenge of 1-bit quantization is that it collapses a continuous signal space into a discrete set of corners. As Huang 0003 et al. explain, the sign function is a harsh nonlinearity that makes reconstructing the original signal's scale and detail difficult using standard linear recovery methods.
In a communication context, this distortion manifests as "quantization noise" that is highly dependent on the signal itself. Yapici et al. argue that conventional precoding strategies fail in these environments because they do not account for this signal-dependent nature. When a transmitter uses 1-bit DACs, the resulting signal is essentially a square wave; the subtle amplitude variations required for high-order modulation schemes like 64-QAM are lost.
Linearizing the Non-Linear
To bridge the gap between non-linear hardware and linear processing, researchers often employ the Bussgang decomposition. This mathematical tool allows a non-linear function to be modeled as a linear gain plus an uncorrelated distortion term. Ye demonstrates that this approach can enable joint activity detection and channel estimation even with low-resolution ADCs.
However, this linearization has limits. Ma et al. point out that Bussgang-based tools become inaccurate when the input signal does not follow a Gaussian distribution—a common occurrence in "box-constrained" precoding where signals are restricted to specific ranges. Furthermore, Wu et al. show that while linear approximations are useful, optimizing performance requires rigorous asymptotic analysis to derive the exact Symbol Error Probability (SEP) in the presence of 1-bit noise.
Strategies for Mitigation and Recovery
If the hardware is inherently non-linear, the system must be designed to be "quantization-aware." Several strategies have emerged to recover signal fidelity:
- Constellation Scaling: Sohrabi et al. found that by scaling the QAM constellation range down by a factor of $\sqrt{2/\pi}$ (approximately 0.8), 1-bit systems can achieve performance remarkably close to ideal infinite-resolution systems.
- Spatial Oversampling: Instead of high temporal resolution, systems can use spatial redundancy. Sankar and Chepuri propose using spatial sigma-delta ($\Sigma\Delta$) ADCs, which shape the quantization noise into higher spatial frequencies, effectively "pushing" the distortion away from the signal of interest.
- Dithering: Aabel demonstrates that adding controlled noise (dithering) before quantization can help preserve signal information that would otherwise be lost below the quantization threshold, allowing 1-bit receivers to support high-order modulations.
- Learned Estimators: Because the distortion is deterministic but complex, neural networks are particularly effective. Sattari et al. show that shallow convolutional networks can extract channel state information from 1-bit signals more accurately than traditional linear algorithms, as they can learn to "invert" the non-linear smearing caused by the hardware.
Beyond Communication
The problem of quantization-induced non-linearity extends beyond MIMO. In wireless sensor networks, Ahmadi et al. show that 1-bit quantization breaks the assumptions required for Granger causality, necessitating new frameworks to infer relationships between signals. Similarly, in imaging, Shah et al. explore how periodic nonlinearities (like sensor wrapping) combined with low bit-depths create a measurement space that requires "harmonic multipliers" for successful reconstruction.
Ultimately, the research suggests that while 1-bit quantization is a "harsh" impairment, it is not an absolute barrier. By treating the quantization not as random noise, but as a specific non-linear transformation, it is possible to recover high-fidelity information from the most impoverished digital representations.
Key Foundational Work
One-Bit Precoding and Constellation Range Design for Massive MIMO With QAM Signaling (2018) — contributed 1 concept elements to this direction.
Robust Decoding from 1-Bit Compressive Sampling with Ordinary and Regularized Least Squares (2018) — contributed 1 concept elements to this direction.
Throughput Analysis of Massive MIMO Uplink With Low-Resolution ADCs (2017) — contributed 1 concept elements to this direction.