In wireless communication, the two-user model is a powerful sandbox. It allows information theorists and network engineers to design elegant strategies for sharing spectrum, such as Non-Orthogonal Multiple Access (NOMA), Rate-Splitting (RS), and Interference Alignment (IA). In a simple two-user world, we can easily map capacity regions, optimize power allocation, and orchestrate clean interference cancellation.
However, a profound challenge emerges when we attempt to scale these architectures to dense, real-world networks. As the number of simultaneous users ($K$) grows, the mathematical and computational complexity does not scale linearly—it explodes exponentially. This phenomenon creates a "scalability bottleneck" that separates theoretical multi-user designs from practical, massive-scale deployments.
The Combinatorial Explosion of Successive Interference Cancellation
To understand why scaling is so difficult, we must look at the mechanical heart of non-orthogonal networks: Successive Interference Cancellation (SIC). In a basic power-domain NOMA system, a base station superimposes multiple user signals into a single waveform. The receiver then decodes these signals one by one, treating the remaining un-decoded signals as noise, and subtracting each decoded signal from the aggregate waveform before moving to the next.
2-User SIC (Linear):
Received Signal ───► Decode User 2 ───► Subtract User 2 ───► Decode User 1 (Clean)
K-User SIC (Permutational Explosion):
Received Signal ───► Sort K Users ───► [K! Possible Decoding Orders] ───► Risk of Compound Error Propagation
In a two-user system, this process is trivial: there is only one interference layer to strip away. But as Zhiguo Ding et al. demonstrate in MIMO-NOMA Design for Small Packet Transmission in the Internet of Things, the complexity of power allocation and SIC grows significantly once a network expands beyond a simple pair. This bottleneck stems from two primary issues:
- Decoding Order Permutations: To perform SIC effectively, users must be sorted based on their effective channel gains. In a dynamic network with $K$ users, the number of possible decoding sequences scales combinatorially ($K!$). As Yue Zhao et al. observe in On the Sum-Capacity with Successive Decoding in Interference Channels, determining the optimal decoding sequence in dense wireless networks quickly becomes computationally intractable.
- Error Propagation: SIC assumes perfect decoding at each stage. If a receiver introduces a minor error while decoding the first user's signal, that error is carried forward into the subtraction phase, corrupting the remaining signal pool. The authors of Energy-Efficient Power Allocation in Uplink mmWave Massive MIMO with NOMA note that this compounding risk forces practical systems to artificially restrict user clustering to pairs, sacrificing the broader spectral efficiency gains that higher-order non-orthogonal grouping could theoretically provide.
The Curse of Dimensionality in Capacity and Secrecy Characterization
The scalability bottleneck is not just a hardware or computational constraint; it is a fundamental theoretical barrier. In information theory, defining the exact boundaries of what a network can achieve requires tracking how signals interact across space and time.
When transmitters only have access to delayed or imperfect Channel State Information at the Transmitter (CSIT), they must rely on retrospective interference management—using stale feedback to clean up past interference. In a two-user network, this is highly effective. However, Pritam Mukherjee et al. show in MISO broadcast channels with confidential messages and alternating CSIT that scaling beyond two users triggers a combinatorial explosion of potential joint channel states and feedback delays.
Similarly, Alireza Vahid et al. demonstrate in Capacity Results for Binary Fading Interference Channels with Delayed CSIT that as nodes are added, the number of potential bit combinations that can simultaneously benefit multiple receivers becomes logically unmanageable. The state-space reduction techniques that work for two users fail because the overlapping regions of shared information grow exponentially.
State Space Complexity:
2 Users: [ State A ] ─── [ State B ] (Manageable Overlap)
K Users: [ Exponentially growing permutations of overlapping channel states ]
This mathematical tractability issue intensifies when physical layer security constraints are introduced. To guarantee that a message remains secret, the system must ensure that an eavesdropper or a non-intended user cannot decode it. Hung D. Ly notes in Multiple-Input Multiple-Output Gaussian Broadcast Channels with Common and Confidential Messages that characterizing secrecy capacity regions is heavily constrained by the matrix interaction complexity of multiple simultaneous receivers.
Proving these boundaries requires eliminating a massive number of variables from a large system of inequalities. As Yingbin Liang et al. highlight in Degraded broadcast channel: Secrecy outside of a bounded range, the number of bounds grows quadratically with the user count, making general closed-form solutions for arbitrary multi-user networks nearly impossible to derive.
Stream Management and Alignment Failures in Advanced Coding
Advanced transmission frameworks designed to bypass the limitations of traditional orthogonal access face their own scaling hurdles. Two prominent examples are Rate-Splitting (RS) and Interference Alignment (IA).
Rate-Splitting Stream Management
Rate-Splitting manages interference by dividing a user's message into a common part (which can be decoded by all users to mitigate interference) and a private part (decoded only by the intended recipient). In a simple two-user MISO broadcast channel, this requires managing just three streams: one shared common stream and two private streams.
However, Yijie Mao explains in Rate-Splitting Unifying SDMA, OMA, NOMA, and Multicasting in MISO Broadcast Channel: A Simple Two-User Rate Analysis that scaling this strategy to larger networks introduces severe combinatorial complexity. In a true $K$-user network, you can split messages into common parts shared among subsets of users (e.g., a stream for users 1 and 2, another for users 2 and 3, and so on). The possible combinations of common and private message parts expand rapidly, making optimal power allocation and stream management computationally demanding.
Interference Alignment Subspace Coordination
Interference Alignment works by coordinating multiple transmitters so that their interfering signals overlap perfectly within a small, discarded subspace at each receiver, leaving the remaining dimensions clean for data transmission.
Interference Alignment (3-User vs K-User):
3-User: [ Clean Subspace ] [ Aligned Interference Subspace (Signals 2 & 3 Overlap) ] ───► Success
K-User: [ Clean Subspace ] [ Multi-source Alignment Coordination Fails to Compress ] ───► Dimensionality Collapse
While this can be elegantly demonstrated for three-user networks, the authors of A Layered Lattice Coding Scheme for a Class of Three User Gaussian Interference Channels show that extending these principles to larger systems remains an open challenge. The precise coordination required to align multiple independent interference sources into a single decodable structure scales in complexity. As Abhik Kumar Das et al. note in Network Coding for Multiple Unicasts: An Interference Alignment Approach, scaling the approach requires entirely different mathematical frameworks and does not automatically guarantee the capacity improvements observed in simpler configurations.
Moving Beyond User-Pairing Heuristics
Because the mathematical and computational complexity of managing $K$-user interference scales exponentially, the wireless industry has historically relied on user-pairing heuristics—grouping users into isolated pairs or triplets to keep the math manageable.
However, as networks grow denser and demand for massive connectivity increases, these heuristics limit the potential of next-generation wireless systems. The core focus of modern multi-user scalability research is to design low-complexity alternatives, structured codebooks, and decentralized optimization frameworks that can handle multi-user interference without succumbing to the curse of dimensionality.