⚡ electrical engineering

Data Detection for Massive MIMO Systems with 1-Bit Quantized Dithered Linear Precoding

This paper proposes and evaluates novel maximum-likelihood-based data detection methods, including low-complexity variants, for massive MIMO systems employing 1-bit DACs with dithered linear precoding, demonstrating significant performance gains over existing baselines for both full-resolution and 1-bit receiver architectures.

Original authors: Amin Radbord, Italo Atzeni, Antti Tölli

Published 2026-06-05

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Amin Radbord, Italo Atzeni, Antti Tölli

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to send a complex, high-definition message (like a 4K video) from one building to another using a massive array of antennas. This is the world of Massive MIMO, a technology crucial for future 6G networks.

The problem is that the equipment needed to send these signals is incredibly power-hungry and expensive. To fix this, engineers want to use "cheap" parts that only speak in very simple terms: 1-bit DACs (Digital-to-Analog Converters). Think of these as a radio station that can only broadcast "Yes" or "No" (or "On" or "Off") instead of a full range of volume levels.

While this saves massive amounts of electricity, it creates a huge mess. It's like trying to describe a painting using only the words "black" and "white." The original message gets distorted, and the receiver struggles to figure out what was actually sent.

This paper proposes a clever solution to clean up that mess, using a concept called "Dithering."

The Core Idea: The "Static" Trick

Imagine you are trying to whisper a secret to a friend, but there is a loud, annoying hum (static) in the room. Usually, static is bad. But in this paper, the engineers add a controlled amount of random noise (dither) to the signal before it gets chopped up into "Yes/No" bits.

Think of it like this: If you try to draw a straight line on a piece of paper using only a stamp that leaves a single dot, you can't draw a smooth line. But if you shake the paper slightly and randomly while stamping, the dots will eventually form a shape that looks like a line. That random shaking is the dither.

The paper assumes the receiver (the friend) knows exactly how the transmitter (you) shook the paper. With that knowledge, the receiver can mathematically "undo" the shaking and reconstruct the original smooth line.

The Two Main Strategies

The authors developed two main ways to decode these messy signals, depending on how advanced the receiver's equipment is:

1. The "Soft" Approach (BLMMSE-DR)

How it works: This method is like taking a blurry photo and applying a standard "sharpen" filter. It tries to estimate the original message by averaging out the noise and removing the known "shaking" (dither) first.
Pros: It's fast and computationally light. It's like using a quick photo-editing app.
Cons: It's an approximation. It gets the general idea right but might miss fine details.

2. The "Hard" Approach (ML-DR and D-ML)

How it works: This is the "Maximum Likelihood" method. Instead of just guessing, it acts like a detective who checks every possible original message against the received signal to see which one fits best.
The Innovation: The paper introduces a new way to mathematically describe the signal after it gets chopped up. This allows the detective to solve the puzzle directly, even with the "Yes/No" distortion.
Pros: It is incredibly accurate. The paper shows it makes far fewer mistakes than the "Soft" approach or older methods.
Cons: It is computationally heavy. Checking every possibility takes a lot of brainpower (processing power).
The Fix: To make this faster, the authors created a "Low-Complexity" version. Instead of checking every possible message, it checks the most likely ones first (like a detective focusing on the prime suspects). This keeps the accuracy high while speeding up the process.

Key Findings from the Experiments

The authors ran simulations to see how well these methods worked:

The "Goldilocks" Zone: The amount of random noise (dither) added is critical. Too little, and the signal stays distorted. Too much, and the random noise drowns out the message. There is a "sweet spot" (moderate-to-high signal strength) where the method works best.
Removing the Noise: When the receiver knows the dither pattern, removing it (Dither Removal) significantly improves performance. It's like realizing the paper was shaking and holding it still to read the dots clearly.
Beating the Competition: The new "Hard" methods (ML-based) consistently outperformed the best existing methods. In some cases, they reduced errors by 10 times compared to other techniques.
Receiver Matters:
- If the receiver has high-quality equipment (Full-Resolution ADCs), the new methods work wonders.
- Even if the receiver also uses cheap "Yes/No" equipment (1-bit ADCs), the new methods still work better than the old ways, though the signal is naturally noisier.

The Bottom Line

This paper doesn't just say "let's use cheap parts." It says, "If we use cheap parts, we can't just ignore the mess they create. We need to add a specific type of random noise (dither) at the start, and then use advanced math at the end to remove that noise and reconstruct the original message."

By doing this, they proved we can build massive, energy-efficient 6G networks without sacrificing the ability to send clear, accurate data. They offer a toolkit that balances speed and accuracy, allowing engineers to choose the right method based on how much computing power they have available.

Technical Summary: Data Detection for Massive MIMO Systems with 1-Bit Quantized Dithered Linear Precoding

Problem Statement
The deployment of fully digital massive multiple-input multiple-output (MIMO) systems is constrained by the high power consumption and hardware complexity of analog-to-digital converters (ADCs) and digital-to-analog converters (DACs). While adopting 1-bit DACs at the transmitter significantly reduces power and complexity, it introduces severe quantization distortion that degrades signal integrity. Although extensive research exists on precoding design for 1-bit DACs, direct data detection methods at the receiver that explicitly account for transmit-side 1-bit quantization in point-to-point massive MIMO systems remain unexplored. Existing detection methods are primarily tailored for multi-user uplink systems with 1-bit ADCs at the receiver, where quantization occurs after channel propagation, a fundamentally different scenario from the transmit-side quantization considered here. Furthermore, standard baselines often involve a two-stage approach (detecting the binary transmit signal first, then recovering data), which risks error propagation.

Methodology
The paper investigates a point-to-point massive MIMO system where the transmitter employs 1-bit DACs with Gaussian dithering applied prior to quantization, and the receiver utilizes either full-resolution or 1-bit ADCs. The study assumes the dither vector is known at the receiver, enabling dither removal. The methodology is structured around two primary linearization approaches for the quantized signals:

Symbol-Independent Linearization (Bussgang Decomposition): This approach treats the quantization distortion as uncorrelated noise. It is used to derive soft-estimation-based data detection methods. The authors propose BLMMSE-DR (Bussgang Linear Minimum Mean Squared Error with Dither Removal) for full-resolution ADCs and D-BLMMSE-DR (Doubly 1-bit quantized BLMMSE-DR) for 1-bit ADCs. These methods subtract the estimated dither-induced bias from the received signal before linear combining.
Symbol-Dependent Linearization: A novel approach is introduced where the transmitted signal is linearized conditioned on the specific data symbol vector. This allows for the derivation of Maximum-Likelihood (ML) based detection methods that directly recover the data symbol vector without an intermediate soft-estimation step.
- For full-resolution ADCs, the authors derive the full likelihood function of the received signal. They propose ML (without dither removal) and ML-DR (with symbol-dependent dither removal). The ML-DR method simplifies the covariance matrix by neglecting off-diagonal elements, significantly reducing complexity while maintaining performance.
- For 1-bit ADCs, the authors develop D-ML (Doubly 1-bit quantized ML). Since symbol-dependent dither removal is not directly applicable to the 1-bit quantized received signal, this method exploits the statistics of the pre-quantized received signal to approximate the likelihood function of the 1-bit output.

To address the exponential computational complexity growth with the number of data streams ( $K$ ) inherent in exhaustive ML search, low-complexity variants are proposed. These variants restrict the search space to a subset of candidates identified by the corresponding soft-estimation methods (e.g., BLMMSE-DR or D-BLMMSE-DR).

Key Contributions

New Detection Frameworks: The paper proposes the first direct data detection methods for point-to-point massive MIMO with 1-bit DACs and dithering, covering both full-resolution and 1-bit receiver architectures.
Dither Removal Techniques: It introduces specific dither removal strategies for both soft-estimation (symbol-independent) and ML-based (symbol-dependent) detection, demonstrating that removing the known dither bias improves detection accuracy, particularly at moderate-to-high signal-to-noise ratios (SNR).
Symbol-Dependent Linearization: A new linearization technique is developed for the 1-bit quantized transmitted signal, enabling the derivation of ML-based detectors that directly map received signals to data symbols.
Complexity Reduction: Low-complexity variants of the ML-based methods are developed, reducing the search space from $L^K$ (where $L$ is the constellation size) to $\nu^K$ (where $\nu \leq L$ ), making the methods feasible for practical stream counts.

Numerical Results
Numerical evaluations using Symbol Error Rate (SER) as the metric highlight the following:

Role of Dither Power: An optimal dither power exists. For low dither power, quantization distortion components remain highly correlated; for high power, the dither dominates. Optimal performance is generally observed at moderate-to-high SNR.
Performance Gains: The proposed ML-based methods (ML-DR and D-ML) consistently outperform soft-estimation-based methods and a baseline two-stage homotopy optimization ML (two-stage hoML) method.
- For full-resolution ADCs, ML-DR achieves up to two orders of magnitude lower SER compared to soft-estimation methods and significantly outperforms the two-stage hoML baseline. Dither removal in ML-DR provides an additional 2× SER reduction compared to the non-removed case at optimal dither powers.
- For 1-bit ADCs, the D-ML method achieves a 10× SER reduction over other methods, including the two-stage hoML baseline, by avoiding error propagation inherent in the two-stage approach.
Scalability: The low-complexity variants (with reduced search space, e.g., $\nu=3$ or $4$) achieve performance nearly indistinguishable from full-search ML methods while drastically reducing computational load.
Robustness: The proposed methods maintain superiority over the baseline even under imperfect Channel State Information (CSI) at the receiver.

Significance and Claims
The paper claims that the proposed ML-based detection methods, particularly when combined with dither removal and low-complexity search strategies, offer a critical advancement in handling the severe quantization distortion of 1-bit DACs in massive MIMO systems. By directly modeling the transmit-side quantization and dithering effects, these methods overcome the limitations of existing baselines that treat the problem as a standard binary detection task followed by symbol recovery. The results demonstrate that accurate data recovery is achievable with low-resolution hardware, provided that the receiver employs detection algorithms specifically designed to exploit the statistical properties of dithered, quantized signals. The work establishes that dither power is a critical system parameter and that the proposed frameworks provide a favorable trade-off between performance and computational complexity for future 6G and wireless backhaul applications.

The Core Idea: The "Static" Trick

The Two Main Strategies

Key Findings from the Experiments

The Bottom Line

More like this