⚡ electrical engineering

On the Detection Optimality and Exact BER Analysis of NOMA

This paper establishes the theoretical optimality of per-user maximum a posteriori (MAP) detectors for NOMA, proves the equivalence of conventional successive interference cancellation and joint maximum likelihood detection, and provides an exact bit error rate analysis that accounts for SIC-induced statistical changes, revealing modest performance gains at low-to-moderate SNR.

Original authors: Tasneem Assaf, Hamad Yahya, Arafat Al-Dweik

Published 2026-07-21

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Tasneem Assaf, Hamad Yahya, Arafat Al-Dweik

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 at a crowded party where two people, a loud friend and a quiet friend, are trying to tell you a secret at the exact same time. They are shouting over the music, and their voices are blending into a single, messy noise. This is the daily reality of modern wireless networks, where many devices try to send data to a cell tower simultaneously. To make sense of this chaos, engineers use a clever trick called "Non-Orthogonal Multiple Access" (NOMA). Instead of giving each person a separate room (which wastes space), they let them share the same room but give the loud friend a megaphone and the quiet friend a whisper. The receiver then has to figure out who said what.

The standard way to do this is like peeling an onion: you listen for the loud voice first, guess what it said, and then subtract it from the noise to hear the quiet voice. This method is called "Successive Interference Cancellation" (SIC). It's been the go-to recipe for years because it works well for the system as a whole. However, there's a nagging doubt: does this "peeling" method actually give the best possible chance for the quiet friend to be understood correctly? And when we try to calculate exactly how often mistakes happen, do we accidentally ignore how the act of guessing the loud voice changes the statistical nature of the noise for the quiet voice? These are the questions that Tasneem Assaf, Hamad Yahya, and Arafat Al-Dweik set out to answer in their 2026 study.

The researchers decided to stop assuming the standard "peeling" method was perfect and instead asked, "What is the mathematically perfect way to guess the quiet friend's message?" They built a new detector based on a statistical rule called Maximum A Posteriori (MAP), which essentially means finding the most likely answer given all the messy information available. They discovered that for the "far user" (the one with the weak signal), the standard method is actually already the best possible choice. But for the "near user" (the one with the strong signal), the standard method is slightly off. The new, optimal detector they designed acts like a smarter listener who doesn't just subtract the loud voice but adjusts their hearing threshold based on exactly how loud the noise is at that specific moment.

One of the paper's most significant findings is a correction to how we calculate error rates. Previous studies often assumed that once the loud voice is removed, the remaining noise behaves exactly as it did before. The authors proved that this is wrong. Because the decision to remove the loud voice is a guess, it changes the statistical "personality" of the remaining noise and signal. They derived a brand-new, exact formula to calculate the Bit Error Rate (BER)—which is just a fancy way of saying "how often the message gets garbled"—that accounts for this change. They showed that if you ignore this change, your calculations are inaccurate.

When they tested their new ideas, the results were clear. In simulations, their new optimal detector consistently made fewer mistakes than the standard "peeling" method, especially when the signal was weak or the power between the two users was balanced. The improvement wasn't a massive explosion of performance, but a "modest but consistent" gain, which in the world of wireless engineering is a very valuable thing. They also proved that two other popular methods, called Joint Multiuser Maximum Likelihood (JML) and standard SIC, are actually identical twins when it comes to the near user; they make the exact same decisions.

Ultimately, this paper doesn't just offer a slightly better way to listen to a party; it fixes the math we use to predict how well our phones will work. By proving that the old way of calculating errors was missing a crucial piece of the puzzle, and by providing a detector that listens more intelligently, the authors have tightened the screws on how we design future networks. Their work shows that even in a system that has been studied for decades, there is still room to refine the rules of the game, ensuring that every bit of data gets through as clearly as possible.

Technical Summary: On the Detection Optimality and Exact BER Analysis of NOMA

Problem Statement
While Non-Orthogonal Multiple Access (NOMA) with Successive Interference Cancellation (SIC) and Joint Multiuser Maximum Likelihood (JML) detection is widely studied, two critical gaps exist in the current literature. First, the presumed optimality of these detectors for minimizing individual-user Bit Error Rate (BER) has not been rigorously established from a statistical signal-processing perspective; they are often adopted based on information-theoretic capacity arguments rather than per-user detection optimality. Second, existing BER analyses frequently overlook the statistical changes induced by the SIC decision process. Specifically, standard analyses often assume that post-SIC fading and noise remain independent with their original marginal distributions, an assumption that fails to account for the conditional dependence introduced by the SIC decision outcome.

Methodology
The paper addresses these gaps through a statistical signal-processing approach centered on the Maximum A Posteriori (MAP) rule. The authors focus on a two-user power-domain downlink NOMA system using Binary Phase Shift Keying (BPSK) modulation over Rayleigh fading channels.

Optimal Detector Derivation: The authors derive the optimal MAP detectors for both the far-user ( $U_1$ ) and the near-user ( $U_2$ ) to minimize individual BER.
- For the far-user ( $U_1$ ), the analysis demonstrates that the optimal detector reduces to the conventional single-user Maximum Likelihood (ML) detector and is equivalent to the JML detector.
- For the near-user ( $U_2$ ), the authors derive a novel decision rule that accounts for the statistical mixture of hypotheses. This rule introduces a dynamic decision boundary, $\varepsilon(y_2)$ , which depends on the received signal, channel gain, and noise variance. The paper proves that this optimal detector reduces to the conventional SIC detector only in the limit of high Signal-to-Noise Ratio (SNR).
Equivalence Proofs: The paper formally proves the equivalence between the conventional SIC detector and the JML detector for the near-user, clarifying that they induce identical decision regions. However, it also establishes that the derived optimal MAP detector for the near-user is distinct from the JML detector, as the former minimizes the near-user's BER while the latter minimizes the NOMA symbol error rate.
Exact BER Analysis: A rigorous exact average BER analysis is presented for the near-user. Crucially, this analysis conditions the statistical behavior of the fading amplitude and noise on the SIC outcome (successful or unsuccessful). The authors utilize joint post-SIC probability density functions (PDFs) to model the fact that the SIC decision renders the fading and noise variables jointly dependent, correcting the inaccurate assumption of independence found in prior works.

Key Contributions

Optimal Detector Design: Derivation of the exact optimal MAP detectors for individual users in a two-user NOMA system, identifying the specific conditions under which these detectors simplify to conventional SIC or JML structures.
Theoretical Equivalence: Proof that conventional SIC and JML detectors are equivalent for the near-user, while simultaneously demonstrating that the optimal MAP detector is superior to both in minimizing individual BER.
Exact BER Formulation: Presentation of an exact average BER analysis for the near-user over Rayleigh fading channels that explicitly incorporates the statistical impact of SIC decisions, including the conditional dependence of fading and noise.
Validation: Theoretical results are validated through numerical simulations, showing perfect alignment between the derived analytical expressions and simulation data.

Results
Numerical results indicate that the proposed optimal detector achieves modest but consistent BER gains over conventional SIC/JML detectors, particularly in low-to-moderate SNR regimes.

Low-to-Moderate SNR: In scenarios with balanced power allocation (e.g., $\alpha_1 = 0.51$ ) and lower SNRs, the Gaussian likelihoods overlap significantly. Here, the optimal decision boundary deviates notably from the conventional fixed boundary ( $\beta\sqrt{\alpha_1}$ ), resulting in measurable performance improvements.
High SNR / Dominant Power: When the far-user's signal is dominant (e.g., $\alpha_1 = 0.9$ ) or at high SNR, the SIC decision becomes highly reliable. Consequently, the optimal detector converges to the conventional SIC detector, yielding nearly identical BER performance.
Power Allocation Sensitivity: The analysis reveals that the BER is sensitive to power allocation coefficients ( $\alpha_1$ ). The optimal detector consistently outperforms the conventional approach across the tested range of $\alpha_1$ and SNRs, with the largest gains observed in regions where the optimal boundary deviates most from the conventional one.

Significance
The paper claims significance by bridging the gap between information-theoretic capacity arguments and practical per-user detection optimality. It establishes that while conventional SIC is optimal in high-SNR or specific power-allocation regimes, it is not universally optimal for minimizing individual user BER. Furthermore, the work provides the first exact BER analysis that correctly accounts for the statistical dependencies created by the SIC process, offering a more accurate theoretical framework for evaluating NOMA system performance. The results suggest that while the performance gains of the optimal detector are modest, they are consistent and theoretically grounded, providing a necessary correction to existing BER analysis methodologies.

More like this