---
title: Successive Interference Cancellation (SIC)
url: https://www.emergentmind.com/topics/successive-interference-cancellation-sic
type: topic
---

# Successive Interference Cancellation (SIC)

Successive interference cancellation (SIC) is a multiuser detection and decoding technique in which a receiver identifies and decodes the strongest transmitted signal component, subtracts its reconstructed contribution from the received composite waveform, and then recursively applies the process to the remaining signal. SIC is fundamental in modern communication systems with superposed transmissions, including fiber-optic links with memory, wireless multiple-input multiple-output (MIMO) systems, non-orthogonal multiple access (NOMA) networks, grant-free random-access, heterogeneous cellular architectures, and emerging semantic communication paradigms.

## 1. Fundamental Principles and Channel Models

SIC relies on sequentially detecting transmitted components in the presence of mutual interference. For finite-memory and/or nonlinear channels such as optical fibers, the surrogate channel may be modeled as
$$
y_i = x_i e^{j\theta_i} + n_i
$$
where $x_i$ is the symbol (potentially from a shaped discrete or continuous constellation), $\theta_i$ is a phase-noise process (CPAN model), and $n_i$ is complex Gaussian noise. The notion of SIC generalizes to multiuser systems such as MIMO uplink, where for $N_t$ transmitters,
$$
\mathbf{y} = H\mathbf{s} + \mathbf{n}
$$
with $H$ the fading matrix and $\mathbf{s}$ the transmitted symbol vector.

Ordering in SIC is determined by the detection strategy: in fiber-optic channels, SIC "layers" typically correspond to amplitude then phase dimensions, while in multiuser spatial multiplexing, the order is determined by measures such as MMSE SINRs, log likelihood ratios (LLRs), or the shadow area constraints (see below). In heterogeneous cellular and wireless networks, order is based on received powers or channel gains (see [1402.1557], [1309.6788]).

## 2. Receiver Algorithms and Advanced SIC Techniques

SIC can be instantiated in several distinct algorithmic frameworks, sometimes augmented by auxiliary methods to mitigate error propagation, enhance robustness, or reduce complexity.

- **Layered SIC for Discrete Constellations:** In high-dimensional fiber-optic channels employing shaped star-QAM, the SIC procedure operates in "layers": amplitude detection, followed by phase detection sublayers, with message passing on a CPAN surrogate graph and statistical interference subtraction. The principal detection metric per layer involves integrating over phase (using the modified Bessel function $I_0$ for amplitude) or via Gaussian approximations for phases, propagating soft information and LLRs across layers. Approximate message passing (AMP) algorithms and factor graphs are applied for efficient marginalization of strong phase noise [2405.03232].

- **Multiple-Feedback and Shadow Constraint Algorithms in MIMO:** In MU-MIMO, error propagation is mitigated by the MF-SIC and improved multiple feedback (IMF-SIC) algorithms, which, upon encountering unreliable (shadowed) soft symbol estimates (distance from decision boundary above a threshold $d_\text{th}$), enumerate several nearest neighbor candidates recursively. The detection branch minimizing the overall ML metric is selected. Ordered IMF-SIC (OIMF-SIC) uses LLR-based dynamic reordering at every layer [1509.00645], [1302.5955].

- **Adaptive Despreading and Blind Estimation:** In multicarrier CDMA, blind adaptive SIC integrates a constant-modulus algorithm within the despreader for stage-wise interference suppression, with weight update via stochastic-gradient descent and dynamic per-stage scaling. After each stage, the residual is combined by MRC or EGC across subcarriers [1102.5442].

- **Multiple-Candidate and Widely-Linear Processing:** For DS-CDMA with jamming, MC-SIC with widely-linear MMSE filtering and vector space projection (VSP) for jammer suppression processes unreliable estimates by jointly considering multiple symbol pairs and integrating their feedback into the SIC loop, thus mitigating error propagation and enhancing robustness to non-circular modulation [1407.1442].

- **Neural and Deep Learning Aided SIC:** SIC has been implemented with neural receivers (SICNet, model-based NNs) that replace decision blocks with DNNs or RNNs trained to produce soft outputs/APPs in each cancellation stage. These architectures can learn to replace explicit CSI, model non-linearities, operate under block fading with online retraining (via FEC-informed pseudo-labels), and generalize to semantic-level cancellation in MAC networks [2207.14468], [2401.09217], [2501.10926].

## 3. Achievable Rates, Capacity, and Asymptotic Analysis

SIC's fundamental information-theoretic advantage lies in its ability to approach the rates of optimal joint detection and decoding (JDD) but with far lower computational complexity. Performance is quantified by achievable information rates (AIRs), which—on surrogate models—admit decomposition by SIC layers:
$$
R = I(\text{Amplitude}; Y) + \sum_{\text{phases}} I(\text{Phase}; Y,\text{prior layers}) + \cdots
$$
On linear and phase-noise-corrupted optical links, finite-stage SIC with Gaussian message passing or Gibbs sampling achieves AIRs within 0.1–0.3 bpcu of the JDD upper bound for shaped constellations (K=32 rings, M=128, S=16 stages), and with complexity $O(n)$ in the number of symbols [2405.03232], [2403.15240], [2212.07761], [2401.09217].

Asymptotically, in many-user MAC with Rayleigh fading and power control, sum-rate under SIC scales with the fraction of decodable users $\zeta$ determined by a deterministic threshold equation, and the optimal per-user rate parameter must vanish $\gamma_n \sim 1/n$ for the sum-rate limit to be nonzero, with $U_\infty = \zeta^*/(\alpha^* \ln 2)$ [2405.12937]. In grant-free massive access, SIC enables a deterministic, nonzero spectral efficiency in the limit $n \to \infty$.

In stochastic wireless networks, the coverage and aggregate throughput induced by SIC reveal a rapid, super-exponential decay in the marginal gain from canceling the $k$th interferer, with nearly all benefit accruing to the first cancellation stage (see Section 5) [1402.1557], [1309.6788].

## 4. SIC in Practical Multiuser, Cellular, and Random Access Networks

SIC is deployed in a range of practical network settings:

- **Wireless Networks and Random Access:** In ad hoc and massive IoT settings, SIC enables resolution of packet collisions by iterative decoding of the strongest signals, with upper bounds for decodable layers (due to SIR thresholds and hardware limits). SIC increases the number of served users and reliability in LoRa and random-access protocols, attainable throughput in slotted random access $S^*_M$ asymptotes to $\ln 2$ for ideal SIC, with even 2-layer SIC achieving roughly 80% of this bound [2010.09625], [2001.03248], [2310.06285].

- **Heterogeneous Cellular Networks (HCN):** SIC enables a receiver in multi-tier architectures to cancel dominant non-accessible interferers (e.g., in range-expanded cells), significantly boosting rate coverage and fairness under various association policies (e.g., max-average-power, min-load, biasing). Empirically and analytically, most of the SIC gain is realized with $N=1$ (single-stage) cancellation [1309.6788], [1402.1557].

- **MAC, NOMA, and ISAC:** In NOMA downlink/uplink, SIC is central to distinguishing simultaneously served users, with decoding order schemes based on channel state information (CSI), quality-of-service (QoS) constraints, or hybrid metrics, each trading off multiuser diversity and outage floor [2005.10215]. In integrated sensing-and-communication (ISAC) for full-duplex cellular systems, the optimal SIC order depends on path loss, power disparity, and location, with closed-form SINR expressions guiding switching thresholds [2405.00109].

- **Semantic SIC:** In semantic communications, SIC may operate directly in the embedding-feature domain, reconstructing and canceling users' semantic vectors, rather than symbol-wise, and employing integrated side information from earlier decoded users. Pretraining and partial retraining schemes optimize training efficiency for dynamic user populations [2501.10926].

## 5. Complexity, Error Propagation, and Ordering

The practical efficiency of SIC stems from the linear or polynomial scaling of complexity in the number of detection stages and the feasibility of layer-wise or branch pruning.

- **Complexity Estimates:** In fiber/channel memory settings, SIC preserves $O(n)$ scaling in block length for a fixed number of layers and constellation size, even with Gaussian message passing or neural nets. In MIMO/MU-MIMO, MF-SIC and IMF-SIC increase complexity multiplicatively by the number of checked candidates $S$ and recursion depth $L$, but in practical systems, $L=2$ or $3$ suffices for near-ML performance [1509.00645], [2405.03232], [2403.15240], [2401.09217].

- **Error Propagation and Mitigation:** All SIC methods are subject to error propagation, with earlier erroneous decisions polluting residuals. Multiple-candidate recursion, dynamic ordering, or side information integration mitigate these effects. Performance gains over naive SIC can be as large as 1–2 dB in MIMO and up to 40% increase in semantic similarity metrics in semantic MAC settings (e.g., BERT, BLEU-4 scores), while the cost in hardware (for analog cancellation, high-SNR digital subtraction, and neural inference) and the need for accurate CSI remain important constraints.

- **Ordering Criteria:** In multiuser/MIMO and NOMA, optimal ordering may be computed using per-layer LLRs, dynamic detection reliability, CSI, or application-specific decodability. In coupled systems like ISAC and MAC, dynamic order selection guided by instantaneous SINR or threshold distances is necessary [2405.00109], [2005.10215], [1509.00645].

## 6. Limitations, Design Guidelines, and Trade-Offs

Despite SIC’s theoretical optimality, key limitations emerge in practice:

- **Marginal utility versus complexity:** The gain from further SIC stages drops super-exponentially with each added layer, especially past the first or second cancellation [1402.1557], [1309.6788]. Hence, deploying more than one or two stages yields low returns relative to the cost.

- **Hardware and delay constraints:** Accurate channel estimation, fine-grained timing, and high dynamic range are required for analog/digital subtraction, especially in crowded or fast-fading environments.

- **Robustness via learning-enabled strategies:** Deep and neural SIC architectures provide robustness to CSI errors and permit adaptation in environments with non-linear distortion or complex memory. In semantic communications, SIC must operate across high-level feature spaces with retained performance even under non-i.i.d. sources and evolving user sets [2207.14468], [2501.10926].

- **Interference-limited versus noise-limited regimes:** In purely interference-limited scenarios, lowering code rates and clustering user populations maximize SIC gains, but, in noise-limited regimes, there exists an optimal per-user rate that balances cancellation with noise amplification [1402.1557].

- **Association and load balancing:** In HCN, minimum-load or biased association policies require at least one layer of SIC to recover the rate or coverage lost by deliberately breaking the SIR-optimal association, thereby achieving fairness without sacrificing efficiency [1309.6788].

In sum, successive interference cancellation represents a unifying, adaptable class of receiver techniques that enables multiuser, high-capacity, and robust decoding in diverse physical layers. The dominant gains arise from the first few cancellations, and practical deployment is coupled tightly to system- and channel-specific constraints, hardware support, and, increasingly, advanced algorithmic enhancements such as deep learning and semantic feature integration.

Source: https://www.emergentmind.com/topics/successive-interference-cancellation-sic