SIC-Based Detector for Multiuser Communications

Updated 13 January 2026

Successive interference cancellation (SIC) is a method that sequentially detects and subtracts interfering signals to reveal weaker components in multi-signal environments.
It employs dynamic ordering strategies and multi-branch extensions to reach near-maximum likelihood performance with reduced computational complexity.
Advanced SIC variants integrate candidate feedback, lattice reduction, and neural network techniques to mitigate error propagation and optimize detection efficiency.

Successive interference cancellation (SIC)-based detection is a class of multiuser or multi-signal processing techniques that sequentially detect and subtract the strongest or most reliably decoded interfering signals from a received composite signal, thereby exposing weaker or less dominant signal components for subsequent detection. SIC is widely deployed in overloaded multiple-input multiple-output (MIMO) systems, code-division multiple access (CDMA), non-orthogonal multiple access (NOMA), satellite receivers, wireless sensor networks, fiber-optic systems, and integrated sensing-communication architectures, where traditional linear detectors (e.g., MMSE, ZF) are suboptimal due to strong, structured interference. Optimally ordered SIC with appropriate cancellation and candidate selection approaches near-maximum likelihood (ML) performance at substantially reduced computational cost, making it indispensable for high-dimensional communication scenarios.

1. Fundamental SIC Principles and System Architecture

SIC operates by decomposing the multi-signal detection problem into a sequence of stages. At each stage, the receiver detects one user's symbol (or a signal layer), subtracts the reconstructed interference contribution from the composite received signal, and passes the resulting residual to the next stage. The detection sequence (i.e., ordering) is often based on received power, channel norm, or instantaneous reliability metrics, aiming to minimize the cumulative impact of error propagation. Linear or non-linear detection filters, quantization, or soft-decision modules are deployed per stage, followed by hard or soft reconstruction of the interfering signal:

For a generic vector multiple-access model $\mathbf{r} = \mathbf{H}\mathbf{s} + \mathbf{n}$ , with $\mathbf{H}$ stacked by user/channel columns, SIC proceeds by (1) selecting the most reliable stream, (2) detecting its symbol via MMSE or ML filter, (3) subtracting its contribution $\mathbf{h}_i\hat s_i$ from the receive vector, (4) repeating the process for remaining streams (Arevalo et al., 2014).
Candidate selection and error checking (e.g., via shadow area constraints or multiple feedback) can be incorporated to mitigate error propagation at each stage (Li et al., 2013, Mandloi et al., 2015).

This reduction from the exponential joint detection complexity $O(M^K)$ to a multi-stage polynomial cost is especially impactful in high-load systems.

2. Ordering Strategies and Multi-Branch Extensions

The SIC ordering—the sequence in which signals are detected and canceled—significantly impacts performance. Standard approaches use fixed power ordering, but channel-dependent orderings (e.g., column-norm-based, SINR-based, dynamic LLR metrics) exploit instantaneous channel diversity (Arevalo et al., 2014, Mandloi et al., 2015). Multi-branch (MB) SIC further enhances reliability by generating detection candidates along several parallel orderings, typically using pre-stored permutations and cyclic shifts. For each branch, an independent SIC run yields an estimated symbol vector, and the best candidate is selected using the ML criterion:

$b^* = \arg\min_{1 \le b \le B} \|\mathbf{r} - \mathbf{H}\hat{\mathbf{s}}_b\|^2$

This MB-SIC approach yields near-ML reliability with polynomial computational complexity. In multiuser MIMO, for $N_t = 8$ transmit and $N_r = 8$ receive antennas (QPSK/16-QAM), multi-branch LR-SIC attains within 1 dB of the true ML detector at $10^{-3}$ BER, outperforming conventional SIC by up to 5 dB SNR (Arevalo et al., 2014).

3. Error Propagation Mitigation: Candidate and Feedback Techniques

A critical limitation of conventional SIC is error propagation: early-stage decoding errors contaminate subsequent stages due to incorrect subtraction. Modern SIC variants introduce multiple feedback (MF) and candidate symbol lists at unreliable detection steps. The shadow area constraint (SAC) defines unreliability regions in the constellation; if a soft estimate lies in this region, MF-SIC selects $M$ nearest constellation points as candidates and recursively invokes SIC on each (Li et al., 2013, Mandloi et al., 2015). Improved MF-SIC (IMF-SIC) checks SAC recursively in deeper layers before committing to a candidate symbol, further suppressing error propagation. Ordered IMF-SIC (OIMF-SIC) employs log-likelihood ratio (LLR)-based dynamic ordering, updating the detection sequence after every stage.

Simulation evidence confirms that MF-SIC, IMF-SIC, and OIMF-SIC yield substantial BER improvements—up to 2 dB gains over MF-SIC and essentially ML performance at $L=2$ –$3$ recursion depth (Mandloi et al., 2015).

4. Lattice-Reduction-Aided, Widely-Linear, Signal-Adaptive and Neural Approaches

Lattice-reduction (LR)-aided SIC preconditions the channel matrix using a unimodular transformation $\mathbf{T}$ such that $\tilde{\mathbf{H}} = \mathbf{H}\mathbf{T}$ is nearly orthogonal, reducing noise enhancement and improving diversity (Arevalo et al., 2014). Widely-linear processing extends SIC to improper signals (e.g., BPSK), using augmented vectors and MMSE filtering, often combined with vector space projection (VSP) for jamming suppression (Yang et al., 2014). Blind adaptive SIC designs utilize constant modulus (CM) criterion-based despreading, weight adaptation, and dynamic amplitude scaling across MC-DS-CDMA subcarriers (Shakya et al., 2011).

Recent advances embed deep neural network blocks into successive cancellation architectures. SICNet is a DNN-aided detector that learns soft interference cancellation directly from data, is robust to CSI mismatch, power order changes, and fading, and achieves BER close to classical SIC with perfect CSI (2207.14468). Similarly, model-based NN equalizers for non-linear, bandlimited channels deploy stage-wise RNNs to approximate forward-backward sorting, with SIC stages reducing complexity versus JDD, yielding near-capacity performance (Plabst et al., 2024).

5. SIC in Specialized Systems and Joint Estimation Frameworks

SIC has been systematically adapted to application domains where interference is inherent or purposely superimposed:

In LoRa networks, gateway SIC enables recovery of colliding packets: two-user capture–cancel schemes improve edge reliability by up to 34%, and capacity by 159% (Sant'Ana et al., 2020).
Cooperative DS-CDMA uplinks use greedy list-based SIC with RAKE receivers and multi-relay selection, achieving near-ML detection performance for moderate complexity (Gu et al., 2014).
Hybrid beamforming SIC in overloaded satellite receivers leverages spatial structure via MRC/CAR beamformers in each cancellation step, yielding up to 12 dB gains over non-adaptive beam approaches at $10^{-4}$ BER (Abu-Shaban et al., 2014).
SIC-aided generative diffusion models implement joint channel estimation and data detection for low-rank multiuser MIMO: they split channel matrices into full-rank sub-blocks, sequentially denoise per block, and outperform classical joint diffusion or MMSE by up to 10 dB NMSE and an order-of-magnitude SER reduction in rank-deficient settings (Bhattacharya et al., 20 Jan 2025).

6. Complexity Analysis, Robustness and Performance Characterization

The computational complexity of SIC-based detectors is typically polynomial, growing as $O(KN_R^2)$ for conventional linear SIC and $O(N_t^4)$ for MB-LR-SIC (Arevalo et al., 2014). Introducing MF, MB, or recursive candidate strategies increases cost proportional to the number of candidates/branches or recursion layers but remains tractable (e.g., $O(SN_t^2)$ for MF-SIC at unreliable layers) (Li et al., 2013, Mandloi et al., 2015). NN-SIC and SICNet scale polynomially in network/constellation sizes, while achieving efficiency $1$–$2$ orders of magnitude lower than trellis-based JDD or Gibbs sampling (Plabst et al., 2024, 2207.14468).

SIC is susceptible to error floors due to hardware impairments and imperfect CSI. Analytical outage probability and ASEP expressions specify explicit degradation due to distortion noise and estimation error for ZF/MMSE-SIC (Miridakis et al., 2016). MMSE-SIC preserves receive diversity with ideal hardware, while ZF-SIC is more vulnerable to CSI uncertainty. Modern error propagation mitigation and dynamic ordering schemes recapture most of the lost diversity at modest complexity increases.

7. SIC in Integrated Sensing and Communication and Fiber-Optics

In integrated sensing and communication (ISAC) architectures, SIC enables successive decoding/detection of coexisting radar and uplink signals at full-duplex base stations. The optimal ordering (decode-first vs. detect-first) shifts with target distance and UE transmit power—a threshold regime exists for which each order is preferable. Residual self-interference and intercell interference critically affect performance, and careful selection of SIC order is mandatory for maximizing joint signal retrieval (Ali et al., 2024).

In contemporary optical fiber studies, SIC paired with multistage coding and Gaussian message passing approaches the achievable information rates of JDD at linear complexity. Appropriately constructed SIC detectors over memory-and-nonlinearity-dominated channels close within 0.1–0.2 bpcu of JDD rates using only 2–16 SIC stages. This delivers a practical solution for long-haul, high-capacity links with correlated phase noise, where separate detection–decoding is fundamentally suboptimal (Jäger et al., 2024, Jäger et al., 2024, Prinz et al., 2022).

References:

Multi-branch lattice-reduction SIC: (Arevalo et al., 2014)
MF-SIC and candidate feedback: (Li et al., 2013, Mandloi et al., 2015)
Adaptive despreading/CDMA: (Shakya et al., 2011)
Widely-linear MC-SIC: (Yang et al., 2014)
Cooperative relay selection: (Gu et al., 2014)
Hybrid beamforming satellite SIC: (Abu-Shaban et al., 2014)
LoRa SIC: (Sant'Ana et al., 2020)
Diffusion-model SIC: (Bhattacharya et al., 20 Jan 2025)
Neural SIC and SICNet: (Plabst et al., 2024, 2207.14468)
Turbo SIC-MMSE for OTFS: (Li et al., 2024)
Fiber-optic and bandlimited SIC: (Jäger et al., 2024, Jäger et al., 2024, Prinz et al., 2022)
Hardware and CSI impairment: (Miridakis et al., 2016)
ISAC SIC ordering: (Ali et al., 2024)

In summary, SIC-based detectors offer strong trade-offs between near-ML reliability and tractable complexity in dense, overloaded, high-order multi-signal environments. Innovations in ordering, multi-branching, candidate feedback, signal adaptation, and neural augmentation have entrenched SIC as a foundational strategy in modern multiuser detection, non-orthogonal access, and interference-limited communication systems.