Spectral-Efficient LoRa Modulation

• SE-LoRa is a family of modulation enhancements that increases bit-rate-per-Hz for LPWAN through multi-chirp, layered chirps, and index modulation.
• It employs overlapping chirps, I/Q multiplexing, and combinatorial mapping to boost spectral efficiency by 2–10× while preserving low-power operation.
• Advanced receiver designs using SIC and FFT-detection mitigate complexity, ensuring robust performance under fading and interference.

Spectral-efficient LoRa (SE-LoRa) modulation denotes a family of waveform and protocol enhancements to classical LoRa chirp spread spectrum (CSS) systems, engineered to dramatically increase the bit-rate-per-Hz (spectral efficiency, SE) for Low-Power Wide-Area Network (LPWAN) applications. In contrast to legacy LoRa, which is severely limited in SE due to its single-chirp, single-index noncoherent design, SE-LoRa approaches leverage multi-chirp, index modulation, I/Q multiplexing, layered chirps, superposition/collision exploitation, and combinatorial or algebraic symbol mapping to boost SE by 2–10× or more at a modest penalty in receiver complexity, energy efficiency, or SNR budget. The field synthesizes core advances including multi-factor index-aided schemes, overlapping-symbol chirp stacking, dual-mode time-domain multiplexing, chirp-rate index modulation, and layered CSS frameworks.

1. Fundamentals of Spectral Efficiency in LoRa and the Need for SE-LoRa

The classical LoRa physical layer employs a single chirp per symbol, with each $M=2^{\mathrm{SF}}$-point symbol carrying $\mathrm{SF}$ bits over a duration $T_s = M/B$ seconds on bandwidth $B$. Its spectral efficiency is $\eta = \mathrm{SF}/2^{\mathrm{SF}}$ bits/s/Hz, which drops sharply with increased spreading factor. For instance, with $\mathrm{SF}=12$, $\eta \approx 0.003$ bits/s/Hz. This low SE is incompatible with emerging LPWAN applications that demand higher data rates for advanced sensor data, firmware delivery, and industrial telemetry (Almeida et al., 2021, Maleki et al., 6 Jan 2026).

Conventional LoRa achieves robustness through large redundancy in codewords; however, throughput is constrained by physical symbol rate and the 1-chirp-per-symbol structure. Increasing SE thus requires designing modulation and detection schemes that encode more bits into each symbol period, often exploiting greater diversity in chirp structure, index sets, time/frequency layering, and coherent/noncoherent detection (Azim et al., 2022, Azim et al., 2024).

2. Overlapping-Chirp and Superposition-Based SE-LoRa

A direct approach to SE improvement is to transmit $K > 1$ chirps in the duration of a conventional symbol, overlapping them such that at each $T_s$, the receiver observes the desired symbol along with partial, shifted versions of previous and subsequent chirps. In "Spectral-Efficient LoRa with Low Complexity Detection," the composite baseband signal is modeled as:

$$X_{\mathrm{SE}}[n;\{s_i\}_{i=-K+1}^{K-1}] = x_L[n;s_0] + \sum_{i=-K+1}^{-1} x_L[n-i\lambda;s_i]g_M[n] + \sum_{i=1}^{K-1} x_L[n-i\lambda;s_i]g_M[n]$$

where $\lambda = \left\lfloor M / K \right\rfloor$, and $s_i$ are the overlapping symbol indices (Maleki et al., 6 Jan 2026).

By packing $K$ symbols per $T_s$, the spectral efficiency scales as:

$$\eta_{\mathrm{SE-LoRa}} \approx K \cdot \eta_0 = K \cdot \frac{\mathrm{SF}}{2^{\mathrm{SF}}}$$

yielding, for example, $+445.45\%$ (SF=7), $+1011.11\%$ (SF=9), and $+1071.88\%$ (SF=11) SE gain with carefully chosen $K$ values (Maleki et al., 6 Jan 2026).

Joint maximum-likelihood detection is theoretically optimal but computationally intractable for realistic payload lengths and SF. To address this, a successive interference cancellation (SIC) detector architecture is proposed, operating at approximately the same complexity as legacy LoRa FFT demodulation. SIC uses sequential chirp estimation and cancellation across the overlapping windows to iteratively recover chirp indices with minimal SNR penalty (≤3 dB at SER $=10^{-3}$ in Rician fading) (Maleki et al., 6 Jan 2026).

3. Index Modulation, Layered, and Combinatorial Approaches

Several SE-LoRa variants utilize index modulation (IM), where symbols encode information by activating a unique pattern of frequency shifts, chirp rates, or spreading factor combinations within each symbol period, often leveraging combinatorial encoding.

3.1. Spreading-Factor-Index-Aided LoRa

SFI-LoRa partitions input bits into “index bits” (choosing $M$ out of $N_{\rm av}$ candidate SFs from $S_{\mathrm{av}}=\{7,8,\dots,12\}$) and “modulation bits” (selecting cyclic starting-frequency bins of constituent chirps). Each symbol’s superposition waveform is constructed by concatenating all selected chirp blocks, each with its own length and cyclic shift, and mapping data to both the combination of SFs and internal frequency indices (Zeng et al., 7 Jun 2025).

The total throughput and spectral efficiency scale with the number of superposed SFs ($M$), but the BER and receiver complexity increase with $M$ due to inter-block interference and increased DFT operations per symbol. For example, at $\mathrm{SF}\approx9$, SFI-LoRa achieves $$R\approx 6.003\,\mathrm{kbps} \to 0.0480\,\mathrm{b/s/Hz}$$, a 2.7× improvement over standard LoRa (Zeng et al., 7 Jun 2025).

3.2. Multiple Chirp-Rate Index Modulation

MCR-IM encodes symbols via the simultaneous activation and superposition of Zadoff-Chu chirps of distinct roots (chirp rates) and frequency offsets. The number of bits per symbol equals $$\eta_b = \left\lfloor \log_2 \binom{2^\nu P}{L} \right\rfloor$$, where $P$ is the number of candidate chirp rates and $L$ the number of active indices (tones) (Zhu et al., 17 Jul 2025). The quasi-orthogonality of ZC sequences allows for large-scale channelization and robust multi-user access, especially when combined with PD-SIC multiuser detection.

Spectral efficiencies exceeding 50% over prior art have been demonstrated, with gains scaling as $P$ and $L$ increase, and collision scenarios managed by SIC. For instance, with $\nu=7, P=4, L=4$, $\eta_b=31$, compared to classic LoRa's $7$ (Zhu et al., 17 Jul 2025).

3.3. Layered Chirp Spread Spectrum (LCSS) and Layered Dual-Mode CSS

The LCSS/LDMCSS framework sends $L$ layers of chirps, each with a distinct slope $\alpha_{\ell}$, over a common symbol period. Each layer independently encodes $\lambda = \log_2 M$ bits via cyclic time/frequency shift; total per-symbol bit load is $L\lambda$. Spectral efficiency is $\mathrm{SE}=L\lambda/M$ (Azim et al., 2024). LDMCSS, using even/odd frequency-shift pairs per layer, achieves nearly the same SE as LCSS but with half as many layers, reducing DFT complexity.

For large $L$, LCSS and LDMCSS surpass TDM-CSS and DM-TDM-CSS in SE-flexibility and robustness to phase/frequency offsets and fading. LCSS achieves up to 8× SE increase over LoRa for $<1$ dB $E_b/N_0$ penalty at BER $10^{-3}$ (Azim et al., 2024).

4. I/Q Multiplexing and Time-Domain Multiplexed SE-LoRa

4.1 In-Phase and Quadrature CSS

IQ-CSS encodes two independent SF-bit symbols per period on orthogonal I and Q branches of the chirp, requiring coherent channel estimation at the receiver (Almeida et al., 2020, Almeida et al., 2021). Each channel uses an independent cyclic shift, and detection is performed via real/imag FFT peak estimation after equalization:

$$\hat{k}_i = \arg\max_f \Re\{\tilde R(f)\}; \quad \hat{k}_q = \arg\max_f \Im\{\tilde R(f)\}$$

This configuration doubles the SE over standard LoRa (e.g., for $\mathrm{SF}=7$, $\eta_{\mathrm{IQCSS}} = 0.109$ bps/Hz) and yields $\sim1$ dB $E_b/N_0$ improvement under optimal channel estimation. The main challenge is increased receiver processing and channel estimation overhead, which is manageable in fixed or low-mobility scenarios (Almeida et al., 2020).

4.2 Dual-Mode and IQ TDM-CSS

DM-TDM-CSS and IQ-TDM-CSS expand on layered ideas using both up- and down-chirps, with even and odd cyclic frequency indices packed in time. DM-TDM-CSS enables nearly fourfold SE increase over LoRa, supporting both coherent and non-coherent detection, and exhibits robust performance under phase/frequency offsets, with only $0.2$–$0.5$ dB penalty versus IQ-TDM-CSS (Azim et al., 2022).

5. Receiver Design, Detection Algorithms, and Complexity

5.1 Noncoherent vs. Coherent Detection

SE-LoRa variants support both noncoherent DFT-peak detection (robust, low-complexity) and coherent maximum likelihood (ML) or minimum distance detection (higher complexity, lower BER, sensitive to channel errors). Layered and index-modulated schemes rely on multi-FFT branches; IQ-CSS and some TDM variants require phase tracking and equalization (Azim et al., 2022, Azim et al., 2024, Almeida et al., 2020).

For superposed/overlapping chirps with significant inter-chirp interference (e.g., overlapping-chirp, MCR-IM), SIC is requisite for practical real-time decoding. For MCR-IM, the PD-SIC algorithm assigns user chirp-rates and iteratively cancels the dominant reconstructed signals in the DFT domain (Zhu et al., 17 Jul 2025). For SFI-LoRa, combinatorial mapping and block-wise dechirp enable independent extraction of index and modulation bits (Zeng et al., 7 Jun 2025).

5.2 Complexity Analysis

• Overlapping-chirp SE-LoRa with SIC operates at per-symbol complexity $O(M\log M)$, similar to LoRa (Maleki et al., 6 Jan 2026).
• Layered CSS requires $L$ (or $L/2$ for LDMCSS) dechirp/FFT units per symbol, with per-unit cost $O(M\log M)$ (Azim et al., 2024).
• SFI-LoRa and MCR-IM rely on multi-rate or multi-root DFT operations and combinatorial index mapping/unmapping, with an increase in FFT count proportional to $M$ or the number of roots/layers.
• IQ-CSS and its layered/TDM generalizations mostly double FFT operations compared to classical LoRa (Almeida et al., 2020, Azim et al., 2022).

A plausible implication is that hardware supporting parallel high-throughput FFTs at the gateway is critical for the viability of high-SE LoRa variants at scale.

6. SE-LoRa Performance, Limitations, and Application Scenarios

6.1 Performance Metrics

Extensive simulation and analytical results show:

• SE-LoRa can yield 2–10× improvements in bits/s/Hz relative to classical LoRa, with a typical SNR penalty of $0.3$–$3$ dB at target SER or BER ($10^{-3}$–$10^{-4}$) (Maleki et al., 6 Jan 2026, Zeng et al., 7 Jun 2025, Zhu et al., 17 Jul 2025, Azim et al., 2024).
• Under fading and moderate phase/frequency offsets, noncoherent detection preserves most of the SE gains with robust BER, whereas coherent methods provide further SNR savings but at the cost of sensitivity to channel nonstationarity (Almeida et al., 2020, Azim et al., 2022).
• For MCR-IM and LCSS/LDMCSS schemes, interference between multi-chirp or multi-layer signals is controlled either by sequence design (e.g., Zadoff–Chu orthogonality) or parameter selection to minimize cross-correlation (Zhu et al., 17 Jul 2025, Azim et al., 2024).

6.2 Limitations and Trade-offs

• Increased peak-to-average power ratio (PAPR), especially for index-modulated or layered superpositions, requires more linear PA designs or envelope management (Azim et al., 2022).
• Receiver complexity, particularly multi-FFT and SIC/PD-SIC processing, scales with SE, presenting a trade-off with network density and gateway hardware (Maleki et al., 6 Jan 2026, Zhu et al., 17 Jul 2025).
• Large index sets (e.g., high $L$, $P$ in MCR-IM, or $M$ in SFI-LoRa) amplify LUT or sorting operations, but practical top-$K$ search with parallel FFTs has been shown feasible at the gateway.
• Under high mobility, the advantages of coherent variants (e.g., IQ-CSS) diminish due to channel variation within the estimation window (Almeida et al., 2021).

6.3 Application Domains

SE-LoRa modulation schemes are best suited for:

• Edge/video frame uplinks and sensor burst uploads in industrial IoT where higher bitrates matter more than absolute range (Zeng et al., 7 Jun 2025, Maleki et al., 6 Jan 2026).
• LPWAN deployments with moderate cell size (≤1 km) and high node density, especially under spectrum scarcity (Azim et al., 2022, Azim et al., 2024).
• Collision-prone scenarios (e.g., large-scale networks) where multichirp/multirate features paired with SIC/PD-SIC facilitate robust multi-user access (Zhu et al., 17 Jul 2025).

7. Comparative Overview of Major SE-LoRa Schemes

Scheme	SE Gain (vs LoRa)	Detection Method	Notable Features / Trade-offs
Overlapping-chirp + SIC (Maleki et al., 6 Jan 2026)	4–10×	Frequency-domain SIC	Minimal complexity overhead, <3 dB SNR loss
SFI-LoRa (Zeng et al., 7 Jun 2025)	2–3×	SF index + DFT	Index modulation on SF choices, combinatorial mapping
MCR-IM (Zhu et al., 17 Jul 2025)	34–55% > prior art	PD-SIC	Multi-chirp (ZC), robust multi-user access
IQ-CSS (Almeida et al., 2020)	2×	Coherent FFT	Requires channel estimation, doubled SE
LCSS/LDMCSS (Azim et al., 2024)	up to 8×	Multi-FFT	Fully tunable SE, layered chirps, flexible
DM-TDM-CSS (Azim et al., 2022)	≈4×	Noncoh./coh. FFT	Dual-mode, robust to FO/PO, low-moderate complexity

This comparison table groups major SE-LoRa schemes by spectral efficiency gain, principal detection architecture, and distinguishing traits.

---

SE-LoRa encompasses a spectrum of modulation and protocol designs that preserve core LoRa characteristics—constant envelope, frequency spreading, low-power operation—while addressing the throughput bottleneck inherent to classic CSS. By judiciously exploiting time/frequency layering, orthogonality, index modulation, and advanced receiver processing, SE-LoRa variants enable order-of-magnitude improvements in spectral efficiency, with implications for both near-term LPWAN upgrades and long-term IoT scaling requirements across industries (Maleki et al., 6 Jan 2026, Zeng et al., 7 Jun 2025, Zhu et al., 17 Jul 2025, Azim et al., 2024, Azim et al., 2022, Almeida et al., 2021, Almeida et al., 2020, Azim et al., 2022).