LoRA Crosstalk: Interference & Mitigation

• LoRA crosstalk is the interference from overlapping chirp signals among transmitters, degrading performance at both symbol and frame levels.
• The research details advanced detection methods like ML-based multi-user detection and SIC, yielding up to 20× capacity and 2.5–10 dB SNR improvements.
• Practical mitigation strategies leverage tailored chirp designs, multi-antenna architectures, and optimized power allocation to boost network throughput by up to 60%.

LoRA crosstalk characterizes the mutual interference arising when multiple LoRa transmitters—whether using the same or different spreading factors—simultaneously emit chirp-based signals on overlapping time-frequency resources. While LoRa’s chirp spread spectrum (CSS) and ALOHA MAC aim for network simplicity and resilience, crosstalk fundamentally limits capacity, reliability, and scalability by producing destructive or non-orthogonal superpositions at the receiver, with effects observed at both symbol and frame level. Modern research comprehensively quantifies crosstalk mechanisms, characterizes system throughput under both ideal and non-ideal conditions, and proposes dense algorithmic and architectural mitigation across PHY and MAC.

1. Signal and Interference Models Underlying Crosstalk

The canonical LoRa baseband symbol for spreading factor $SF$ ($N=2^{SF}$ chips) and symbol index $s$ is expressed as

$$x_s[n] = \exp\left(j 2\pi\left[\frac{n^2}{2N} + \left(\frac{s}{N}-\frac{1}{2}\right) n \right]\right), \quad n=0,\dots,N-1.$$

A received gateway signal with $K$ colliding transmitters is

$y[n] = \sum_{i=1}^K h_i\,x_{m_i}[n-\tau_i] + w[n],$

where $h_i$ are complex channel gains, $m_i$ symbol indices, $\tau_i$ chip-level timing offsets, and $w[n]\sim\mathcal{CN}(0,\sigma^2)$ is AWGN.

Crosstalk arises when either co-SF (same spreading factor) or inter-SF transmissions result in energy from multiple users occupying overlapping spectral bins post-dechirp and FFT. Unlike ideal orthogonality, practical channel conditions, power dispersion, and hardware nonidealities result in mutual interference even between nominally orthogonal SFs (Waret et al., 2018).

2. Error Mechanisms and Theoretical Analysis

Under interference, demodulated bins transform from single prominent peaks (collision-free) to multiple local maxima, elevating symbol and frame error rates (SER, FER). Precise SER expressions for a single interferer, accounting for arbitrary relative chip and phase alignment between SoI and interferer, are given as

$$P(\hat{s}\neq s) = 1 - E[\int_0^\infty f_{Ri}(y;v_s,1)\prod_{k\neq s} F_{Ri}(y;v_k,1) dy],$$

where $f_{Ri}$, $F_{Ri}$ are Ricean PDF and CDF, $v_s$, $v_k$ encode signal and interference amplitude, and the expectation is over time/phase offset and symbol values (Afisiadis et al., 2019). Crucially, misalignment between signals mitigates crosstalk severity by spreading interfering energy over multiple FFT bins, offering up to 1 dB SNR benefit versus fully chip-aligned models.

Coherent demodulation, which aligns signal phase using the estimated channel or preamble phase, further reduces error rates under crosstalk, enabling $2.5$–$10$ dB SNR improvement at moderate to high interference, compared to standard non-coherent methods (Afisiadis et al., 2020).

3. Practical Crosstalk Mitigation: Multistage and SIC Receivers

Crosstalk mitigation centers on exploiting signal structure, power differences, and timing offsets:

• Maximum Likelihood (ML) Multi-User Detection: The joint likelihood over symbol sequences of attackers and targets,

$$L = \mathrm{const} + \sum_k \Big\{ \text{SoI matched DFT term} + \text{Interference terms} \Big\},$$

can be solved at per-symbol complexity $O(N^2)$ (vs. $O(N^3)$ for full trellis), enabling practical implementation for two-user collisions with strong synchronization and time offset exploitation (Xhonneux et al., 2020).

• Serial Interference Cancellation (SIC): A power-domain NOMA approach. The receiver decodes the strongest user, reconstructs and subtracts its signal, proceeding iteratively down the power hierarchy. This yields up to $20\times$ same-SF channel capacity, contingent on sufficient received power disparities among users. Trade-offs include error propagation to weaker signals and increased computational and memory load (Tesfay et al., 2021).
• Asynchronous/Desynchronous Decoders: Algorithms that track frequency/time edges from preambles or symbol boundaries can recover all frames in 2-user settings when timing offsets are sufficiently diverse, yielding $+60\%$ throughput improvements. For $N\ge3$ collided users, recovery rate depends on symbol diversity and surface area in the frequency-time space (Rachkidy et al., 2018).

4. Algorithmic Innovations in Collision-Resistant Symbol Detection

Symbol-level crosstalk mitigation has advanced via low-complexity Bayesian classification and functional feature extraction:

• CoRa Detector: Replaces brittle peak-finding with two per-bin features—the half-period discriminator (HPD, distinguishing clipped from complete tones) and peak magnitude deviation (PMD, normalizing observed magnitudes)—fed to a Bayesian classifier pre-trained on empirical collision data. This yields up to $29\%$ better decoding than TnB and $178\%$ over conventional interference cancellation at constant-complexity, real-time cost (Álamos et al., 18 Dec 2024).
• Complexity Comparison:

Algorithm	Per-symbol Complexity	Typical Throughput Gain
Baseline FFT	$O(N\log N)$	1x (reference)
CIC	$O((1+2K)N\log N)$	Up to 2x (SF=8)
TnB/Thrive	$O(N^2)$	Up to 3–4x
CoRa	$O(2N\log N)$	Up to 11.5x

(Álamos et al., 18 Dec 2024)

5. Crosstalk Across Spreading Factors and Networks

Crosstalk can occur:

• Within a single SF (“same-SF”): Co-SF crosstalk is the primary limitation at high densities (Waret et al., 2018).
• Across different SFs (“inter-SF”): Empirical SF orthogonality is imperfect, causing inter-SF SINR thresholds $q_{iSF,k} < \infty$. Analytical throughput models show up to 50% aggregate throughput penalty due to inter-SF crosstalk for fully loaded networks, and the choice between SF-distance and SF-random allocation governs individual device and total throughput (Waret et al., 2018).
• Between networks (“inter-network”): Multiple co-located LoRa deployments further lower packet extraction probability. Mitigation is most efficiently provided by spatial diversity (multiple gateways), improving DER from 0.24 (single gateway) to 0.56 (3 gateways) under severe interference; node-side directionality gives only modest incremental gain (Voigt et al., 2016).

6. Advanced PHY Layer and Multiuser Architectures

Next-generation crosstalk mitigation leverages both advanced signal design and multiuser detection:

• Nonlinear Chirps (CurvingLoRa): Using symbol-orthogonal nonlinear frequency trajectories, e.g., quadratic or quartic polynomials, enforces strong crosscorrelation suppression among users. This enables separation of fully aligned symbols, resolving near-far blocking and sustaining throughput gains up to $7.6\times$ in large networks (Li et al., 2022).
• Multi-Antenna and Joint Detection: With $L$ gateways (or $N_t$ receive antennas per gateway), the joint ML rule over all antennas preserves interference structure, enabling scalable uplink (capacity doubled/tripled for two/three concurrent users, at the price of a $3$–$5$ dB transmit power increase) (Nguyen et al., 2021).
• Optimized Power Allocation: Using Jaccard coefficient-based similarity minimization across expected power vectors enhances bin separation post-detection, an effect realized through convex approximations in the uplink scheduler (Nguyen et al., 2021).

7. Crosstalk from Narrowband Interference and Coexistence

LoRa’s susceptibility to crosstalk from external non-LoRa sources—narrowband interferers—shows distinct physics:

• BPSK and GMSK Interference: Symbol-level MC simulations reveal that AWGN modeling is conservative; true structured narrowband modulation incurs 2–5 dB less impairment for equal INR, and constant-envelope interferers (e.g., GMSK) are markedly less harmful than BPSK (Huang et al., 30 Nov 2025).
• An analytical summary formula,

$$\mathrm{INR}_{\max}(\mathrm{SNR}) = \alpha\,\mathrm{SNR} + \beta + \frac{\gamma}{\mathrm{SNR} - (R_T - N_0 - 1)},$$

with parameters fit per SF and interferer type, enables correct coexistence threshold planning.

• Practical mitigation: Deployments should avoid AWGN-only coexistence models, randomize symbol timing, use appropriate FEC depth, and calibrate LBT algorithms to signal type (Huang et al., 30 Nov 2025).

---

In aggregate, LoRa crosstalk research demonstrates that network performance degrades continuously with load—rather than at hard capture thresholds—across interactions between signal structure, timing, power, and system design. High-fidelity models and algorithmic advances now enable near-ML collision recovery, multi-user decoding up to practical network densities, and robust operation even with spectrally-adjacent protocol coexistence. Recent directions stress tailored chirp designs (CurvingLoRa), hybrid PHY-MAC schemes, full-featured multi-antenna architecture, and calibrated coexistence analysis as central engineering principles for LPWAN scalability and reliability in dense, interference-rich environments (Xhonneux et al., 2020, Tesfay et al., 2021, Rachkidy et al., 2018, Álamos et al., 18 Dec 2024, Li et al., 2022, Waret et al., 2018, Voigt et al., 2016, Nguyen et al., 2021, Afisiadis et al., 2019, Afisiadis et al., 2020, Huang et al., 30 Nov 2025).