⚡ electrical engineering

Enhanced Uplink Data Detection for Massive MIMO with 1-Bit ADCs: Analysis and Joint Detection

This paper presents a new analytical framework for uplink data detection in massive MIMO systems with 1-bit ADCs, deriving expected soft-estimated symbols to design a superior linear minimum mean dispersion (LMMD) receiver and a joint detection strategy that significantly outperform conventional methods in symbol error rate.

Original authors: Amin Radbord, Italo Atzeni, Antti Tolli

Published 2026-03-26

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Amin Radbord, Italo Atzeni, Antti Tolli

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 listen to a crowded room full of people talking at once. You want to hear what specific people are saying, but there are two major problems:

The Room is Huge (Massive MIMO): You have a massive array of microphones (antennas) to help you hear better.
The Microphones are Broken (1-Bit ADCs): To save money and power, your microphones are extremely simple. They don't record the full volume or tone of the voice. They only say, "Is the sound louder than a whisper? Yes or No?" and "Is the sound positive or negative?" They turn complex voices into a simple stream of "Yes/No" or "Up/Down" signals.

This paper is about how to figure out exactly what people are saying in that noisy room, even though your microphones are so simple they can barely distinguish between a shout and a whisper.

Here is the breakdown of their solution, using simple analogies:

1. The Problem: The "Pixelated" Voice

When you use these super-simple microphones, the sound gets "pixelated." It's like taking a high-definition photo and turning it into a low-resolution 1-bit black-and-white image. You lose all the details (the volume, the nuance).

In the past, engineers tried to fix this by using standard math tricks designed for perfect microphones. But those tricks failed because they didn't account for the fact that the microphones were "dumb."

2. The Discovery: The "Ghost" of the Voice

The authors realized something fascinating. Even though the microphones are dumb, if you look at the average of all the "Yes/No" signals over time, a pattern emerges.

Imagine you are trying to guess a friend's location based on a blurry, pixelated map. You can't see their exact face, but you know that if they are wearing a red hat, the pixels in the "red" area will light up in a specific way.

The authors calculated exactly what this "average pattern" (or Expected Value) looks like for every possible word someone could say. They found that:

The pattern changes depending on how loud the room is (Signal-to-Noise Ratio).
The pattern changes depending on what the other people in the room are saying (Interference).

Key Insight: The "dumb" microphones distort the sound in a predictable way. If you know the distortion, you can reverse-engineer the original voice.

3. The New Receiver: The "Smart Filter" (LMMD)

Standard receivers try to clean up the signal using old rules. The authors built a new filter called LMMD (Linear Minimum Mean Dispersion).

Old Way: "I hear a 'Yes'. I'll assume the person said 'Hello'." (This is often wrong because the 'Yes' could mean many things).
New Way (LMMD): "I know that if Person A says 'Hello' while Person B says 'Goodbye', the microphones will produce a specific 'Yes/No' pattern. If Person A says 'Hello' and Person B says 'Goodbye', the pattern is different. I will compare the actual pattern I hear against a library of all possible patterns to find the closest match."

This new filter is tailored specifically for these "dumb" microphones, making it much better at guessing the original words than the old filters.

4. The Strategy: Solving the Puzzle Together (Joint Detection)

Here is the most creative part. In a crowded room, people talk over each other.

Old Strategy (UE-Specific): You try to listen to Person A while ignoring Person B. You guess what A is saying based on what you think B is saying. This is like trying to solve a puzzle while only looking at one piece at a time.
New Strategy (Joint Detection - JD): You realize that Person A's voice and Person B's voice are linked. The "Yes/No" signal you hear is a mix of both. Instead of guessing them separately, you guess both at the same time.

The Analogy: Imagine trying to guess two people's passwords simultaneously. If you guess Person A's password, it changes the clues for Person B's password. The authors' method looks at all the clues together to solve the whole puzzle at once.

They also created a "Low-Complexity" version (N-JD). Instead of checking every possible combination of passwords (which takes forever), they check the top 3 or 4 most likely guesses for each person and then combine them. It's like narrowing down the search to the most probable suspects before interrogating them.

5. The Results: Why It Matters

The paper ran simulations (computer experiments) and found:

The "Smart Filter" (LMMD) was much better at hearing clearly than the old standard methods.
The "Joint Strategy" (JD) was a huge leap forward. By listening to everyone together, they could understand the conversation much better, even when the microphones were very simple.
The Sweet Spot: Interestingly, they found that having a little bit of background noise actually helps! It's called "Stochastic Resonance." A tiny bit of static helps the "dumb" microphones distinguish between signals better than total silence does.

Summary

This paper teaches us how to build a super-efficient communication system for the future (6G). By using cheap, low-power microphones (1-bit ADCs) and applying smart math that understands how those microphones "break" the signal, we can still have crystal-clear conversations.

They didn't just fix the broken microphone; they learned to speak the microphone's new, simple language and translate it back into perfect English.

1. Problem Statement

The paper addresses the challenge of uplink data detection in Massive MIMO systems equipped with 1-bit Analog-to-Digital Converters (ADCs).

Context: To utilize sub-THz bandwidths, Massive MIMO is required, but high-resolution ADCs consume excessive power. 1-bit ADCs offer a power-efficient alternative but introduce severe non-linear quantization distortion.
Specific Challenges:
- Quantization Distortion: 1-bit quantization distorts the transmit constellation, making amplitude information unrecoverable at high Signal-to-Noise Ratios (SNR).
- Imperfect Channel State Information (CSI): Channel estimation itself is performed using 1-bit quantized pilot signals, leading to estimation errors that conventional receivers (designed for infinite-resolution) fail to handle optimally.
- Inter-user Interference: In multi-user scenarios, the soft-estimated symbols of interfering users are statistically interdependent due to the non-linear quantization process. Conventional detection strategies often treat users independently, ignoring this interdependence.
- Limitations of Existing Work: Previous studies often focused on single-user models, idealized i.i.d. fading, or high-complexity iterative algorithms (like GAMP/VAMP) that degrade under imperfect CSI.

2. Methodology

The authors propose a comprehensive framework involving analytical characterization, receiver design, and joint detection strategies.

A. System Model

Setup: Single-cell Massive MIMO with $M$ base station (BS) antennas and $K$ single-antenna User Equipments (UEs).
Channel: Correlated Rayleigh fading (more realistic for high frequencies than i.i.d.).
Quantization: 1-bit quantization applied to both channel estimation pilots and uplink data signals. The quantization function $Q(\cdot)$ is defined with a scaling factor to match signal variance.
Estimation: The BS uses a Bussgang Linear MMSE (BLMMSE) estimator to estimate channels from 1-bit quantized pilots.

B. Analytical Characterization of Soft-Estimated Symbols

The core theoretical contribution is the derivation of the expected values of soft-estimated symbols ( $\hat{x}$ ) after linear combining but before data detection.

MRC Analysis: The authors derive a closed-form expression for the expected value of the soft-estimated symbol for a specific UE under Maximum Ratio Combining (MRC). Crucially, they show this value depends on:
- The SNR.
- The number of UEs and BS antennas.
- The pilot matrix.
- The data symbols of all interfering UEs.
ZF and MMSE Extension: They extend this analysis to Zero-Forcing (ZF) and Minimum Mean Squared Error (MMSE) receivers. They prove that asymptotically (as $M \to \infty$ ), the expected values for ZF and MMSE are simply scaled versions of the MRC expected values.
Key Insight: The distortion caused by 1-bit quantization creates a specific "cloud" of expected values in the complex plane. At high SNR, symbols with the same phase but different amplitudes become indistinguishable.

C. Receiver Design: LMMD

Based on the derived expected values, the authors design a new receiver: the Linear Minimum Mean Dispersion (LMMD) receiver.

Objective: Minimize the mean dispersion of the soft-estimated symbols around their derived expected values.
Mechanism: Unlike conventional receivers that ignore quantization effects, LMMD explicitly incorporates the Bussgang gain matrix and the covariance of the 1-bit quantized signal to optimally decorrelate the received signal.

D. Data Detection Strategies

The paper proposes strategies that exploit the interdependence among the soft-estimated symbols of interfering UEs:

UE-Specific Strategies:
- Exhaustive (E-USD): Searches all $L^K$ combinations (high complexity).
- Heuristic (H-USD): Uses the average expected value over interfering symbols (low complexity).
- Genie-aided: Assumes perfect knowledge of interference (theoretical lower bound).
Joint Data Detection (JD):
- Performs a joint search over all $K$ users simultaneously using the minimum distance criterion against the vector of expected values.
- Complexity: $O(L^K)$ , exponential in $K$ but independent of $M$ .
N-point Joint Detection (N-JD):
- A low-complexity variant of JD. It restricts the search space to the $N$ most likely candidates for each user, reducing complexity to $O(N^K)$ .

3. Key Contributions

Generalized Analytical Framework: Characterization of expected soft-estimated symbols for multi-UE systems with correlated Rayleigh fading, extending previous single-UE/i.i.d. work.
Closed-Form Expressions: Derivation of exact expected values for MRC and asymptotic scaling factors for ZF and MMSE receivers under 1-bit quantization.
LMMD Receiver: A novel receiver design tailored specifically for 1-bit ADCs that minimizes dispersion around the derived expected values, outperforming conventional MRC/MMSE.
Joint Detection Strategy: Introduction of JD and N-JD, which leverage the statistical coupling between users caused by quantization, significantly improving Symbol Error Rate (SER).
SER Analysis: Derivation of an upper bound on SER for JD and exact SER for H-USD under simplified i.i.d. fading assumptions.

4. Numerical Results

Receiver Performance:
- LMMD vs. Conventional: LMMD significantly outperforms MRC and MMSE, particularly around the optimal SNR operating point (approx. 6 dB in simulations).
- Imperfect CSI: Conventional receivers (MRC/MMSE) surprisingly perform better under imperfect CSI than perfect CSI because the channel estimate implicitly accounts for quantization scaling. However, LMMD is superior at the optimal SNR.
- SNR Behavior: Performance exhibits non-monotonic behavior (stochastic resonance); too little noise makes quantization distortion dominant, while too much noise drowns the signal.
Detection Strategy Performance:
- JD/N-JD vs. UE-Specific: Joint detection (JD) and its variant (N-JD) provide a massive boost in SER compared to treating users individually (E-USD/H-USD).
- RML Comparison: The proposed JD/N-JD strategies outperform the Robust Maximum Likelihood (RML) method at moderate-to-high SNRs. RML fails at high SNR because it cannot distinguish between symbols of the same phase but different amplitudes due to 1-bit quantization limits.
- Complexity: N-JD with a small $N$ (e.g., $N=3$ ) achieves performance nearly identical to full JD but with significantly lower computational cost.
System Parameters:
- Increasing the number of BS antennas ( $M$ ) monotonically improves SER.
- Increasing the number of UEs ( $K$ ) initially improves performance (due to array gain) but eventually degrades it due to increased interference and overlapping expected values.

5. Significance

This paper bridges the gap between theoretical analysis and practical implementation for 1-bit Massive MIMO systems.

Theoretical Impact: It provides the first rigorous characterization of how 1-bit quantization distorts the joint constellation of multiple users, revealing that user interference is not just additive noise but a structured distortion that can be exploited.
Practical Impact:
- It demonstrates that Joint Detection is essential for 1-bit systems; treating users independently is suboptimal.
- The LMMD receiver and N-JD strategy offer a practical path to high-performance, low-power 6G systems, proving that massive arrays with 1-bit ADCs can outperform hybrid architectures in spectral and energy efficiency.
- The results suggest that for high-SNR operation, systems should either use PSK modulation (to avoid amplitude ambiguity) or employ power control, rather than relying solely on higher-order QAM.

In summary, the paper establishes that by modeling the specific distortion of 1-bit ADCs and exploiting the interdependence of user signals, one can design receivers and detection algorithms that significantly outperform state-of-the-art robust methods, even with imperfect channel knowledge.