Source-linked AI summary
LoRa beyond ALOHA: An Investigation of Alternative Random Access Protocols
Luca Beltramelli, Aamir Mahmood, Patrik Österberg, Mikael Gidlund
TL;DR
LoRaWAN’s default pure ALOHA can scale poorly under many same-channel, same-SF transmissions, motivating study of alternative access. The paper develops a LoRa-aware model to compare three random-access mechanisms. Slotted ALOHA has higher reliability than pure ALOHA at an energy-efficiency cost for low device densities; CSMA compares favorably with slotted ALOHA at smaller spreading factors.
Problem
Pure ALOHA’s scalability is limited when many devices transmit on the same channel and spreading factor.
Source: §I · p. 1Method
A unified analytical model compares pure ALOHA, slotted ALOHA, and non-persistent CSMA while incorporating LoRa capture effects and access-specific design parameters.
Source: §I · pp. 1–2, §V · p. 9Results
Slotted ALOHA offers higher reliability than pure ALOHA at the expense of energy efficiency when device numbers are small; CSMA is reported to outperform slotted ALOHA at smaller spreading factors.
Source: §V · p. 9Takeaways & Limitations
Abstract
from arXiv · showhide
We present a stochastic geometry-based model to investigate alternative medium access choices for LoRaWAN---a widely adopted low-power wide-area networking (LPWAN) technology for the Internet-of-things (IoT). LoRaWAN adoption is driven by its simplified network architecture, air interface, and medium access. The physical layer, known as LoRa, provides quasi-orthogonal virtual channels through spreading factors (SFs) and time-power capture gains. However, the adopted pure ALOHA access mechanism suffers, in terms of scalability, under the same-channel same-SF transmissions from a large number of devices. In this paper, our objective is to explore access mechanisms beyond-ALOHA for LoRaWAN. Using recent results on time- and power-capture effects of LoRa, we develop a unified model for the comparative study of other choices, i.e., slotted ALOHA and carrier-sense multiple access (CSMA). The model includes the necessary design parameters of these access mechanisms, such as guard time and synchronization accuracy for slotted ALOHA, carrier sensing threshold for CSMA. It also accounts for the spatial interaction of devices in annular-shaped regions, characteristic of LoRa, for CSMA. The performance derived from the model in terms of coverage probability, channel throughput, and energy efficiency are validated using Monte-Carlo simulations. Our analysis shows that slotted ALOHA indeed has higher reliability than pure ALOHA but at the cost of lower energy efficiency for low device densities. Whereas, CSMA outperforms slotted ALOHA at smaller SFs in terms of reliability and energy efficiency, with its performance degrading to pure ALOHA at higher SFs.
I. INTRODUCTION
This paper examines alternatives to LoRaWAN’s default ALOHA access to address interference, reliability, and scalability while considering energy efficiency. It develops a unified comparison of pure ALOHA, slotted ALOHA, and non-persistent CSMA using LoRa-specific interference effects and access parameters.
I. INTRODUCTION
- LoRaWAN’s default pure ALOHA is simple and suits sporadic monitoring traffic, but has limited scalability and no QoS guarantees.The paper focuses on random-access alternatives intended to improve reliability and scalability without compromising energy efficiency.Source: §I · p. 1
- The paper compares pure ALOHA, slotted ALOHA, and non-persistent CSMA using a unified model that incorporates LoRa’s time- and power-capture effects.The model is intended to support comparisons of channel throughput and energy efficiency while accounting for LoRa interference behavior.Source: §I · pp. 1–2
- The model characterizes interference intensity using access-specific design parameters, including slotted ALOHA guard time and the CSMA sensing threshold.The authors frame these parameters as useful for designing LoRa media access control.Source: §I · p. 2
A. LoRa and LoRaWAN
- LoRaWAN defines medium access and higher layers, while LoRa is its proprietary physical-layer modulation; the standard describes device classes with different downlink and energy behaviors.Class A devices use ALOHA for uplinks, Class B devices synchronize periodically, and Class C devices continuously listen for gateway transmissions.Source: §I · p. 1, A. LoRa and LoRaWAN · p. 2
1) S-ALOHA:
- Prior LoRa access proposals target either improved scalability and reliability through random access, or bounded QoS through resource reservation; slotted ALOHA is one random-access alternative.Slotted access can theoretically double channel capacity relative to pure ALOHA, but energy-efficient synchronization across LoRa’s coverage remains challenging.Source: B. Alternative channel access… · p. 2, 1) S-ALOHA · p. 2
2) CSMA:
This section describes alternative access mechanisms and the assumptions behind the LoRa interference model. It treats dominant co-SF interference as the main analytical case while noting the conditions that qualify that simplification.
2) CSMA:
- CSMA senses the channel before transmission and avoids synchronization, but hidden nodes can reduce its performance; with all nodes hidden, throughput degrades to pure ALOHA.The passage notes that LoRa’s multi-kilometer coverage can leave a significant share of devices hidden from one another.Source: 2) CSMA · p. 2
3) Scheduled MAC:
- Scheduled MAC divides channel resources among devices for contention-free transmissions, but requires conveying schedules and predicting resource needs.The cited LoRa proposals include hybrid ALOHA/TDMA, on-demand TDMA, traffic-periodicity-based slots, self-organizing time slots, and time-slotted channel hopping.Source: 3) Scheduled MAC · p. 2
III. MATHEMATICAL MODELING
- The modeled network has uniformly distributed devices around one gateway, six contiguous SF annuli, unacknowledged fixed-size uplinks, and a shared duty-cycle limit across access methods.The model assumes no gateway limit on simultaneous demodulation, though the authors say it could be modified to account for that hardware constraint.Source: §III · p. 3
A. Interference model
- Inter-SF interference requires many strong interferers before the receiver fails to detect the desired symbol’s peak.The interference analysis also reports that full-overlap co-SF decoding depends on relative received signal strength when interferers transmit different symbols.Source: A. Interference model · p. 3
- The model assumes perfect SF orthogonality and considers only dominant co-SF interference, an assumption the authors say weakens with very many interferers.They note that reducing potential interferers under slotted ALOHA and CSMA makes dominant-interferer modeling more plausible.Source: A. Interference model · p. 4
B. MAC-dependent Intensity of Point Process of Interferers
The model characterizes dominant gateway interference by forming a MAC-dependent subset of the device point process. For P-ALOHA, independent transmission thinning and its vulnerability time determine the interferer intensity.
B. MAC-dependent Intensity of Point Process of Interferers
- The interfering-device set is a MAC-dependent subset of the device PPP, used to characterize dominant interference at the gateway.Its intensity changes the interference distribution and, in turn, system performance.Source: B. MAC-dependent Intensity of Point… · p. 4
- Extreme order statistics give the dominant-interference distribution from the per-interferer fading and path-loss distribution and the random number of interferers.The count is described by the arrival-process PMF; for Poisson arrivals, its parameter for an annulus depends on the annulus radii.Source: B. MAC-dependent Intensity of Point… · p. 4
1) P-ALOHA:
- P-ALOHA independently thins the device PPP to active transmitters of intensity αλ; its vulnerability time is twice the message time-on-air.Devices with a message can transmit immediately.Source: 1) P-ALOHA · p. 4
2) LoRA:
The model adjusts interferer intensity for LoRa’s preamble immunity, slotted-ALOHA timing errors, and CSMA contention. For CSMA, it accounts for contention geometry and approximates the active interferers as an equi-dense PPP.
2) LoRA:
- LoRa’s immunity to interference on the first preamble symbols reduces vulnerability time, with a visible effect for payloads of a few bytes.The reduction depends on preamble duration, symbol time, and message time-on-air.Source: 2) LoRA · p. 4
3) S-ALOHA:
- S-ALOHA’s interferer intensity depends on guard-time ratio and adjacent-slot collision probabilities, which depend on timing-error distribution and guard interval.Previous-slot collisions are less likely because they affect the transmitted message’s preamble.Source: 3) S-ALOHA · p. 4
4) NP-CSMA:
- Asynchronous non-persistent CSMA senses before transmission and reschedules after a busy-channel result; backoff permits energy-saving mode without gateway synchronization.The model represents active interferers as a Matérn hard-core Type II process, assuming ties in backoff values are negligible.Source: 4) NP-CSMA · pp. 4–5
- Because modeling the CSMA active-interferer process is complex, the analysis approximates it as an equi-dense PPP with uniformly and randomly distributed interferers.The contention domain depends on detection threshold and annulus geometry; distance distributions for annuli are more complex than for a disk.Source: 4) NP-CSMA · p. 5
C. Derivation of CCDFs of SNR and SIR
The channel model combines fading, path loss, and noise to evaluate received SNR and SIR. SIR analysis uses the distribution of dominant interference to derive the probability of exceeding the threshold.
C. Derivation of CCDFs of SNR and SIR
- The channel is modeled with Rayleigh block-fading, AWGN, and free-space power-law path loss.The path-loss function is l(r) = γr^-β, with γ determined by carrier wavelength and β the path-loss exponent.Source: C. Derivation of CCDFs of SNR and SIR · p. 6
- SNR success is evaluated as the probability that received SNR exceeds threshold θ, using the received power ptxhl(r) and an exponential channel gain.The channel gain h is exponentially distributed with unit mean.Source: C. Derivation of CCDFs of SNR and SIR · p. 6
- SIR success under co-SF interference is the probability that received signal power divided by dominant interference exceeds threshold δ.The derivation uses the dominant-interference distribution and the distribution of the product of path loss and fading.Source: C. Derivation of CCDFs of SNR and SIR · p. 6
D. Performance Metrics
This section introduces the performance metrics used to compare the random access protocols.
D. Performance Metrics
- The section derives the performance metrics used to compare the random access protocols.Source: D. Performance Metrics · p. 6
1) Success and Coverage Probability:
An end device succeeds only when its transmission meets both the SNR and SIR conditions. Coverage probability is obtained by averaging success probability over devices in the relevant spreading-factor annulus.
1) Success and Coverage Probability:
- Success requires both the SNR and SIR conditions to hold at the gateway.The probability of satisfying both is not treated as the product of independent probabilities.Source: 1) Success and Coverage Probability · p. 6
- The joint success probability accounts for the dependence between satisfying the SNR and SIR conditions.The passages note that messages meeting the SIR condition are more likely to meet the SNR condition.Source: 1) Success and Coverage Probability · p. 6
- Coverage probability averages success probability over the device distribution in the annulus for spreading factor q.The annulus extends from the inner radius for q−1 to the outer radius for q.Source: 1) Success and Coverage Probability · p. 6
2) Channel Throughput:
The model calculates channel throughput from offered traffic and transmission success probability. Throughput is calculated per spreading factor under the orthogonal-SF assumption.
2) Channel Throughput:
- Channel throughput is the product of offered traffic and success probability.Source: 2) Channel Throughput · p. 6
- The model treats spreading factors as orthogonal when calculating throughput.This follows the assumption stated in Sec. III-A.Source: 2) Channel Throughput · p. 6
- Throughput is calculated separately for each spreading factor q.The per-SF expression weights success probability by the end-device distribution.Source: 2) Channel Throughput · p. 6
3) Energy Efficiency:
Energy efficiency is based on payload delivered per message energy, which varies with the access method. The model accounts for synchronization energy in slotted ALOHA and channel-detection energy in NP-CSMA.
3) Energy Efficiency:
- Energy efficiency relates payload size to the energy used to deliver a message to the gateway.The message energy depends on the channel access method; sleep-mode consumption is assumed negligible.Source: 3) Energy Efficiency · p. 6
- For P-ALOHA, the model counts device energy used to transmit messages.The passage states that all energy in this case is used for transmission.Source: 3) Energy Efficiency · p. 6
- For S-ALOHA, periodic gateway beacons maintain synchronization and add energy use.The analysis assumes beacons of duration TB sent every TSYNC, reducing overhead relative to the cited two-way synchronization method.Source: 3) Energy Efficiency · p. 6
- For NP-CSMA, energy efficiency includes average channel-detection energy; retransmissions and backoff energy are excluded.The model assumes no gateway acknowledgements and negligible energy use during energy-saving backoff.Source: 3) Energy Efficiency · pp. 6–7
IV. RESULTS AND DISCUSSION
The results compare LoRa access schemes using a Monte-Carlo-validated analytical model. S-ALOHA offers the highest coverage in the reported comparison, while CSMA’s reliability advantage depends on sensing threshold and spreading factor.
IV. RESULTS AND DISCUSSION
- The analytical model was validated with Monte-Carlo simulations averaging 2000 runs per distance, with 95% confidence-interval error bars.Source: §IV · p. 7
1) LoRa and S-LoRa:
- The S-LoRa comparison assumes beacon synchronization, class B or C devices, and normally distributed synchronization error tied to clock skew and interval.Source: 1) LoRa and S-LoRa · p. 7
- At N = 3000, S-ALOHA increases average success probability over P-ALOHA by 0.16 across spreading factors.Source: 1) LoRa and S-LoRa · p. 7, Figure 2 · p. 7
- Synchronization error lowers S-LoRa’s SIR success probability relative to ideal S-ALOHA, while the LoRa preamble provides some guard time without an explicit guard interval.Source: 1) LoRa and S-LoRa · p. 7
- LoRa and S-LoRa offer similar coverage at typical distances of 3 and 6 km, but S-LoRa scales better as device counts increase.Source: 1) LoRa and S-LoRa · p. 7
2) NP-CSMA:
- At low sensing thresholds CSMA’s success probability exceeds S-ALOHA’s, while high thresholds reduce it to P-ALOHA’s performance.Source: 2) NP-CSMA · p. 8, Figure 4 · p. 8
- S-ALOHA provides the highest coverage probability in the comparison because it outperforms NP-CSMA and P-ALOHA at large spreading factors.Source: Figure 3 · p. 7, 2) NP-CSMA · p. 8
- For outer annuli, NP-CSMA offers little gain over P-ALOHA; equal-interval allocation distributes SIR success probability more fairly across annuli.Source: 2) NP-CSMA · p. 8, Figure 4 · p. 8
B. Energy Efficiency
The comparison finds that access-method performance depends on device density, spreading factor, and service area, while LoRa capture effects improve throughput at high densities. Slotted ALOHA trades energy efficiency for reliability, and CSMA can outperform it in some parameter regions.
B. Energy Efficiency
- At R = 3 km, S-ALOHA is more energy efficient than P-ALOHA for devices near their annulus’s outer radius, while P-ALOHA leads nearest the gateway; low-threshold CSMA can exceed both.The figure reports energy efficiency by distance from the gateway; CSMA efficiency depends on sensing threshold and is nearly constant within each annulus for small P0.Source: Figure 5 · p. 8
- At low device density, P-ALOHA and NP-CSMA have the best mean energy efficiency, but at higher density S-ALOHA becomes most efficient as its higher success probability offsets synchronization energy.The energy accounting includes transmission, S-ALOHA synchronization, and NP-CSMA channel-sensing consumption.Source: B. Energy Efficiency · pp. 8–9
- The energy-efficient choice varies across SF, service radius, and device count: NP-CSMA leads at small radius or low density, while larger SFs can favor CSMA over S-ALOHA.Periodic-beacon synchronization energy increases exponentially with SF, making S-ALOHA less energy efficient at larger SFs.Source: B. Energy Efficiency · p. 9, Figure 6 · p. 9
V. CONCLUSIONS
- The unified model compares P-ALOHA, S-ALOHA, and NP-CSMA using LoRa capture and interference behavior, while accounting for slotted-ALOHA guard intervals and synchronization error and CSMA sensing thresholds.It characterizes interference intensity in relation to the access-mechanism design parameters.Source: §V · p. 9
C. Channel Throughput
- With time- and power-capture effects, P-ALOHA and S-ALOHA retain high channel-access efficiency as device counts grow, unlike theoretical expressions without capture.When messages overlap in time, the strongest can survive collision; the model assumes no two interferers transmit the same symbol.Source: Figure 7 · p. 9, C. Channel Throughput · p. 9
V. CONCLUSIONS
- Slotted ALOHA offers higher reliability than pure ALOHA but can cost energy efficiency at low device counts; CSMA can improve both metrics when devices are closely located, typically at small SFs.The conclusions also report that LoRa power capture significantly improves channel throughput for all three mechanisms at large device densities.Source: §V · p. 9
APPENDIX
The appendix approximates the expected number of active neighbors of an interfering device using the fraction of an annulus inside the transmitter’s contention domain. It scales an upper bound according to that fraction and compares the resulting model with simulations.
A. Neighbors in an interfering ED contention domain
- The approximation uses H, the fraction of devices in an annulus within a transmitter’s contention domain, because conditioning the interfering node’s distance distribution is too complex.H is obtained from equation (10).Source: A. Neighbors in an interfering ED… · p. 9
- The approximation varies with H: it uses the unconditioned neighbor expectation near H = 0, a single-interferer assumption near H = 1, and an average of two interfering devices near H = 0.5.An exponential function g(H), equal to 1 at H = 0 or 1 and 2 at H = 0.5, scales the upper bound for the expected number of interfering devices.Source: A. Neighbors in an interfering ED… · pp. 9–10
- The appendix compares the simulated number of interfering end devices with the proposed mathematical model.The accompanying visuals depict the transmitter contention domain and the scaling function for the interfering-device count.Source: A. Neighbors in an interfering ED… · p. 10, Figure 8 · p. 10, Figure 9 · p. 10