One-bit massive MIMO: Channel estimation and high-order modulations

Sven Jacobsson, Giuseppe Durisi, Mikael Coldrey, Ulf Gustavsson, Christoph Studer

Abstract

Massive MIMO systems are pivotal for next-generation wireless networks but suffer from high power consumption due to high-resolution analog-to-digital converters (ADCs). This paper investigates a radical hardware simplification: using one-bit ADCs. We demonstrate that despite severe nonlinearity, simple least-squares (LS) channel estimation is remarkably effective. Furthermore, we show that spatial oversampling—leveraging hundreds of antennas—allows the system to support high-order modulations like 16-QAM, challenging the conventional belief that one-bit quantization restricts signaling to QPSK.

Introduction

In Massive MIMO, base stations equipped with hundreds of antennas serve multiple users simultaneously. However, scaling these systems to wider bandwidths leads to an exponential increase in the power consumption of high-resolution ADCs. One-bit ADCs offer a solution by using simple comparators that only detect the sign of the signal, eliminating the need for complex automatic gain control and significantly reducing hardware costs. The primary challenge lies in recovering complex signals from this highly distorted, nonlinear output.

System Model and Architecture

We consider the uplink of a Massive MIMO system where $K$ single-antenna users transmit to a base station with $N$ antennas. Each antenna at the base station is followed by an RF chain and a pair of one-bit ADCs for the in-phase (Re) and quadrature (Im) components.

System model of a Massive MIMO uplink with one-bit ADCs at the base station.

The resulting output is a quantized signal where each component is mapped to the set {1, -1}. This architecture represents the extreme limit of low-resolution quantization.

Channel Estimation: The Power of Simplicity

A common assumption is that nonlinear quantization requires complex, computationally intensive channel estimation algorithms. However, this research shows that standard Least-Squares (LS) estimation, performed on one-bit quantized pilot symbols, provides sufficient accuracy for Massive MIMO operations. In the single-antenna (SISO) case, combining LS with joint pilot and data processing can even reach theoretical capacity limits. The key insight is that the massive number of antennas provides a form of "spatial oversampling" that compensates for the low temporal resolution of each individual bit.

High-Order Modulation and the "Dithering" Effect

Traditionally, one-bit systems were thought to be limited to QPSK because amplitude information is seemingly lost. This paper refutes that notion, demonstrating that 16-QAM is not only viable but superior to QPSK in many regimes.

This is made possible by two factors:

  1. Spatial Averaging: Across hundreds of antennas, the statistical average of the one-bit outputs can represent amplitude.
  2. Beneficial Noise: Counter-intuitively, thermal noise and random channel phases act as a "dither." This noise smears the signal across the zero-threshold of the quantizer, allowing the receiver to infer amplitude information from the probability of a bit being +1 or -1.

Performance Analysis

The viability of high-order modulation is clearly visible when examining achievable rates. As shown in the figure below, 16-QAM begins to outperform QPSK at SNR levels as low as -15 dB in a system with 400 antennas.

Achievable rate per user versus SNR for QPSK and 16-QAM modulations with 1 and 20 users.

At high SNR, the 16-QAM rate reaches its maximum of 4 bits per channel use in single-user scenarios. However, a "constellation collapse" occurs at extremely high SNR where the lack of noise prevents the dithering effect, causing performance to degrade. In multi-user scenarios (e.g., 20 users), the system remains robust but becomes interference-limited more quickly than high-resolution counterparts.

Conclusion

One-bit Massive MIMO is a promising architecture for energy-efficient wideband communication. By leveraging spatial oversampling and simple LS estimation, these systems can support high-order modulations like 16-QAM. Future work will focus on mitigating multi-user interference and exploring the optimal balance between ADC resolution and spectral efficiency.

Summary updated

BibTeX
@inproceedings{jacobsson2015one,
  author = {Sven Jacobsson and Giuseppe Durisi and Mikael Coldrey and Ulf Gustavsson and Christoph Studer},
  title = {{One-bit massive MIMO: Channel estimation and high-order modulations}},
  booktitle = {Workshops @ ICC},
  year = {2015},
  doi = {10.1109/iccw.2015.7247358},
}
Figure