Introduction
Massive Multiple-Input Multiple-Output (MIMO) is a cornerstone technology for modern wireless communication, enabling base stations with hundreds of antennas to serve multiple users simultaneously. However, the deployment of such large-scale arrays is often hindered by the high power consumption and cost associated with high-resolution digital-to-analog converters (DACs). A promising solution is the use of 1-bit DACs, which significantly reduce hardware complexity and power requirements.
While 1-bit DACs are efficient, they introduce severe quantization noise. Traditional linear precoding techniques, such as Zero-Forcing (ZF), perform poorly under 1-bit constraints, especially for higher-order modulations like 16-QAM. This paper explores nonlinear precoding algorithms that mitigate these effects, making high-order modulation practical for 1-bit hardware.
System Model and the 1-Bit Bottleneck
The system under consideration consists of a base station (BS) with $B$ antennas serving $U$ single-antenna users over a frequency-flat wireless channel. Each antenna at the BS is equipped with a 1-bit DAC, which quantizes the complex-valued precoded signals into a discrete set of four possible values (corresponding to the signs of the real and imaginary parts).
The primary challenge is that 1-bit quantization is a highly nonlinear process. Simple "linear-quantized" precoders—which calculate a standard precoded signal and then round it to 1-bit—suffer from significant error floors at high signal-to-noise ratios (SNR). This makes reliable communication using modulations beyond QPSK extremely difficult.
Nonlinear Precoding via SQUID
To overcome the limitations of linear precoding, the authors propose treating precoding as a discrete optimization problem. The goal is to find the 1-bit transmit vector that minimizes the mean-square error (MSE) at the user terminals. Since this is an NP-hard combinatorial problem, the paper introduces the Squared $\ell_\infty$-Norm Relaxation (SQUID) algorithm.
SQUID relaxes the discrete 1-bit constraint into a continuous convex problem using a squared infinity-norm. Unlike Semidefinite Relaxation (SDR), which is computationally prohibitive for massive MIMO, SQUID can be solved efficiently using matrix-vector operations, making it suitable for real-time implementation in large-scale arrays. Furthermore, because 1-bit quantization removes amplitude information, the authors developed blind estimation techniques for users to locally determine the necessary scaling factor ($\beta$) for decoding non-constant modulus signals like 16-QAM.
Performance Evaluation
The proposed SQUID algorithm was evaluated through numerical simulations of a system with 128 BS antennas and 16 users. The results demonstrate a dramatic improvement in Bit Error Rate (BER) compared to traditional methods.
As shown in the comparison above, linear-quantized ZF precoding (top) exhibits a massive error floor for all modulations, rendering 16-QAM and higher completely unusable. In contrast, the SQUID algorithm (bottom) significantly lowers these error floors. Notably, 16-QAM becomes viable and even outperforms 16-PSK due to its more efficient constellation packing. While 64-QAM still faces challenges, the results prove that 1-bit systems are not strictly limited to low-order phase modulations.
Conclusion
This work demonstrates that the performance bottleneck in 1-bit massive MIMO systems is not the hardware itself, but the reliance on linear precoding. By employing nonlinear algorithms like SQUID, it is possible to support high-order modulations such as 16-QAM with power-efficient 1-bit DACs. This paves the way for more sustainable and cost-effective massive MIMO deployments in future wireless networks.
Summary updated