PaperlayerView exact revision on arXiv

Source-linked AI summary

Scalability analysis of large-scale LoRaWAN networks in ns-3

Floris Van den Abeele, Jetmir Haxhibeqiri, Ingrid Moerman, Jeroen Hoebeke

arXiv:1705.05899v1cs.NI

TL;DR

LoRaWAN deployments raise unresolved questions about scalability and capacity as device populations grow, while simplified models omit capture and interference effects. This paper builds a baseband-derived error model into an ns-3 LoRaWAN module and finds that downstream capacity deteriorates confirmed upstream delivery, with gateway density mitigating but not eliminating the effect.

  • Problem

    LoRaWAN scalability and capacity remain uncertain as deployments grow to hundreds or thousands of end devices per gateway, and existing models may omit capture and interference effects.

  • Method

    The paper constructs a LoRa error model from complex baseband bit error rate simulations and combines it with LoRaWAN MAC and PHY models in ns-3.

  • Results

    Downstream traffic substantially reduces confirmed upstream delivery, while increasing gateway density ameliorates but does not eliminate this effect.

  • Takeaways & Limitations

    Gateway densification can delay the impact of limited downstream capacity, but stringent gateway duty-cycle requirements continue to constrain downstream opportunities.

  • Takeaways & Limitations

    Modeling all interference as noise is less accurate for co-spreading-factor interference than a stochastic approach.

Abstract

from arXiv · show

As LoRaWAN networks are actively being deployed in the field, it is important to comprehend the limitations of this Low Power Wide Area Network technology. Previous work has raised questions in terms of the scalability and capacity of LoRaWAN networks as the number of end devices grows to hundreds or thousands per gateway. Some works have modeled LoRaWAN networks as pure ALOHA networks, which fails to capture important characteristics such as the capture effect and the effects of interference. Other works provide a more comprehensive model by relying on empirical and stochastic techniques. This work uses a different approach where a LoRa error model is constructed from extensive complex baseband bit error rate simulations and used as an interference model. The error model is combined with the LoRaWAN MAC protocol in an ns-3 module that enables to study multi channel, multi spreading factor, multi gateway, bi-directional LoRaWAN networks with thousands of end devices. Using the lorawan ns-3 module, a scalability analysis of LoRaWAN shows the detrimental impact of downstream traffic on the delivery ratio of confirmed upstream traffic. The analysis shows that increasing gateway density can ameliorate but not eliminate this effect, as stringent duty cycle requirements for gateways continue to limit downstream opportunities.

I. INTRODUCTION

LoRaWAN is being deployed for low-rate, long-range, energy-efficient IoT connectivity, but its scalability and capacity remain open questions. This work addresses those questions with a comprehensive ns-3 model and scalability analysis covering traffic, interference, and gateway density.

  • LoRaWAN combines sub-GHz propagation with LoRa modulation to provide large-area coverage using limited infrastructure.
  • LoRaWAN scalability remains uncertain as deployments grow, including the effects of device count, network parameters, traffic, and multiple gateways.
  • Existing studies often omit downstream traffic, transmitter interference, or multiple gateways, while pure ALOHA models miss capture and interference effects.
  • The work contributes a LoRa error model, a comprehensive ns-3 LoRaWAN implementation, and a scalability study of confirmed messages and downstream traffic.
  • LoRaWAN uses one-hop star topologies in which gateways forward packets between end devices and a central network server.
  • Class A devices sleep most of the time, transmit upstream, and open one or two receive windows afterward for possible downlinks.The first and second windows open one and two seconds after the upstream transmission ends.

III. PROBLEM STATEMENT AND APPROACH

The paper investigates how LoRaWAN performance changes with network size and configuration. It uses simulation to study networks with up to 10 000 end devices and varying gateway density, traffic, and data-rate assignments.

  • The study targets unresolved questions about how LoRaWAN networks scale as their size increases.
  • It examines the effects of end-device density, message types, data-rate assignment, and gateway density on network performance.
  • Simulation enables the study of large-scale LoRaWAN networks with up to 10 000 end devices.
  • The approach uses ns-3 because of its adoption in network research and its potential for modeling multiple LPWAN technologies and coexistence.

IV. LORAWAN NS-3 MODULE

The ns-3 module models LoRaWAN applications, protocol layers, gateways, class A end devices, and a simple network server. Its architecture supports configurable multi-spreading-factor gateway implementations and publicly available examples and tests.

  • The module includes a LoRa error model, LoRaWAN PHY and MAC layers, class A end-device and gateway applications, and a simple network server.
  • End-device nodes contain one MAC/PHY pair, while gateways use one MAC/PHY pair per supported spreading factor.
  • A six-channel gateway supporting all spreading factors contains 36 MAC/PHY pairs.
  • The lorawan ns-3 module includes unit tests and examples and is publicly available.

1) LoRa PHY baseband implementation:

The LoRa PHY baseband model simulates transmission through an AWGN channel and receiver-side demodulation and error correction. Its processing chain includes coding, interleaving, whitening, reverse Gray mapping, and chirp modulation.

  • 1) LoRa PHY baseband implementation:: The BER model is based on complex baseband simulations of a LoRa transceiver over an AWGN channel.The simulations measure bit errors for different LoRa PHY configurations.
  • 1) LoRa PHY baseband implementation:: Information bits are encoded using the available 5/4, 7/4, and 8/4 LoRa code rates.The 7/4 and 8/4 rates use Hamming codes that correct one bit error and detect up to two.
  • 1) LoRa PHY baseband implementation:: A diagonal interleaver groups bits across consecutive code words, with its interleaver length equal to the LoRa spreading factor.The number of bits mapped per LoRa symbol consequently equals the spreading factor.
  • 1) LoRa PHY baseband implementation:: The encoded and interleaved stream is whitened, reverse Gray mapped, and converted into integers for LoRa modulation.The modulator generates time-shifted complex baseband up-chirp samples from these integers.
  • 1) LoRa PHY baseband implementation:: At reception, correlation-based demodulation selects the symbol with maximum correlation, and error rate is measured after error correction.The receiver reverses the sender chain before measuring information-bit errors.

2) LoRa PHY BER simulations:

The BER simulations evaluate LoRa PHY configurations over AWGN at stepped SNR values and fit exponential curves to selected measured BER points. The fitted model uses a PDR-based SNR cutoff.

  • 2) LoRa PHY BER simulations:: LoRa BER simulations use the PHY parameters in Table I and evaluate SNR values in 1 dB steps.There is no oversampling, so N = 2^SF.
  • 2) LoRa PHY BER simulations:: The simulations cover the LoRa PHY parameter combinations listed in Table I.The table defines the configurations used for the BER measurements.
  • 2) LoRa PHY BER simulations:: An exponential curve is fitted to selected logarithmic BER measurements after zero-valued BER points are discarded.Points are retained until the corresponding PDR for a 13B packet falls below one in a million.

B. LoRaWAN PHY layer

The ns-3 LoRaWAN PHY and MAC layers model reception, interference, transmission states, receive windows, duty cycles, and retransmissions. Separate state-machine behavior captures the differing roles of end devices and gateways.

  • B. LoRaWAN PHY layer: The LoRaWAN PHY is built on ns-3’s SpectrumPhy concept to support reception modeling and future inter-technology simulations.The PHY error model is combined with the ns-3 PHY implementation.
  • B. LoRaWAN PHY layer: Packet reception uses a chunk-based SINR approach that evaluates each constant-SINR interval using the BER model.Interfering transmissions starting or ending create new reception chunks.
  • B. LoRaWAN PHY layer: The LoRaWANPhy finite state machine has six states, including ready, busy-reception, and busy-transmission states.Incoming transmissions below the cutoff are dropped immediately.
  • C. LoRaWAN MAC layer: The LoRaWANMac class queues packets, opens receive windows, handles end-device retransmissions, and tracks radio duty cycles.Gateway MACs continuously listen when not transmitting and do not use receive windows.
  • C. LoRaWAN MAC layer: The MAC state machine gives class A end devices seven states, while gateway MAC objects use three states.A gateway enters UNAVAIL when another gateway MAC is transmitting, preventing simultaneous transmissions on the same gateway.
  • C. LoRaWAN MAC layer: After transmission, class A end devices wait one second before RW1 and two seconds before RW2.Confirmed packets remain queued for retransmission until transmissions expire or an acknowledgment arrives.

D. LoRaWAN class A end device ns-3 application

The ns-3 applications model class A end devices, gateways, and a network server, while the simulation example varies deployment, traffic, and timing parameters. Delivery is measured differently for confirmed and unconfirmed upstream packets.

  • D. LoRaWAN class A end device ns-3 application: LoRaWANEndDeviceApplication represents class A end devices and configures upstream data rate, packet length, message type, channel selection, and generation times.It generates the MAC payload and adds the LoRaWAN frame header.
  • D. LoRaWAN class A end device ns-3 application: LoRaWANGatewayApplication passes packets between gateways and the network server and tags downstream packets with PHY parameters.The gateway selects the MAC/PHY pair matching the tagged spreading factor and channel.
  • D. LoRaWAN class A end device ns-3 application: The singleton network server accepts upstream packets, sends downstream traffic, detects duplicate packets, and queues per-device downstream data.RW1 and RW2 timers schedule downstream transmissions through recently known gateways.
  • D. LoRaWAN class A end device ns-3 application: Simulations use one, two, or four gateways and configurable end-device populations deployed in a disc with a 6 100m radius.Devices share a 125kHz upstream channel, while RW2 uses a high-power channel with a 10% RDC restriction.
  • D. LoRaWAN class A end device ns-3 application: Upstream generation periods are 600, 6 000, or 60 000 seconds, while downstream arrivals average one packet every ten or hundred upstream packets.Downstream traffic uses an end-device-specific exponential random variable.
  • D. LoRaWAN class A end device ns-3 application: Unconfirmed packets count as delivered after successful gateway reception, whereas confirmed packets also require an acknowledgment from the network server.The measured PDR includes all generated packets, including queued packets.

A. Assigning LoRa spreading factors to end devices

The study compares three strategies for assigning LoRa spreading factors to end devices. The PER strategy achieves the best packet delivery ratio and is selected for the remainder of the paper with a threshold of 0.01.

  • Underestimating a spreading factor can cause low-SNR reception errors, while overestimating it wastes air time.
  • The evaluated strategies are Random, Fixed, and PER, which assigns the lowest spreading factor whose packet error ratio is below a threshold.
  • The simulations vary end-device counts over a 600-second upstream period, with PER thresholds of 0.001, 0.01, 0.1, and 0.25.
  • The PER strategy performs best in PDR, although no tested threshold is best for every network size.
  • A PER threshold of 0.01 is selected for subsequent experiments.

B. Unconfirmed vs confirmed upstream data

Confirmed upstream messages do not consistently improve packet delivery ratio because acknowledgments and retransmissions add gateway and channel load. Confirmed-message PDR exceeds unconfirmed-message PDR only under sufficiently low traffic loads.

  • PDR decreases as transmissions become more frequent and as the number of end devices increases.
  • For a 600-second data period, collisions account for close to 90% of undelivered unconfirmed packets, while interference accounts for another 9%.
  • Undelivered packets are disproportionately sent at SF11 or SF12, comprising 80.9%, 93.6%, and 95.8% for 600, 6 000, and 60 000 seconds, respectively.
  • Confirmed-message PDR is higher only at low traffic loads; in all other simulated cases, it is lower than unconfirmed-message PDR.
  • As traffic load increases, missed receive windows and the average transmissions per confirmed message increase, while acknowledgments abort ongoing gateway receptions.

2) Multi gateway LoRaWAN networks:

Increasing gateway density improves data rates and substantially raises unconfirmed upstream PDR, while confirmed upstream PDR improves less because gateway duty-cycle limits still constrain acknowledgments. Downstream traffic saturates gateways and reduces upstream delivery, especially in dense networks with high downstream rates.

  • Multi gateway LoRaWAN networks:: Higher gateway density leads to faster overall data rates in a 10 000-end-device network.
  • Multi gateway LoRaWAN networks:: Unconfirmed-message PDR increases greatly with more gateways, whereas confirmed-message PDR increases noticeably but less sharply.
  • Multi gateway LoRaWAN networks:: More gateways increase successfully received upstream messages and sent acknowledgments, but duty-cycle restrictions can increase missed receive windows in saturated scenarios.
  • C. Downstream data traffic: With one gateway and at least 5 000 nodes, downstream airtime saturation is clearly visible; distributing traffic across gateways increases downstream PDR.
  • C. Downstream data traffic: Downstream traffic causes negligible upstream-PDR loss at low rates and a small loss at high rates for 5 000 and 10 000 nodes.
  • C. Downstream data traffic: Downstream transmissions block upstream reception, but more gateways reduce this effect by occupying less time per gateway.

VI. RELATED WORK

Prior scalability studies use analytical, stochastic, empirical, and simulator-based approaches, but several omit interference details, downstream traffic, or multiple gateways. This work positions its ns-3 module as a comprehensive alternative for bidirectional multi-gateway LoRaWAN analysis.

  • Analytical capacity studies account for receive windows and duty-cycle restrictions but may model LoRaWAN access as pure ALOHA.
  • Pure ALOHA models do not capture interference dependence on received power, timing, and forward error correction.
  • Stochastic-geometry work models SNR-related outage and co-spreading-factor interference, predicting declining coverage probability as end-device counts increase.
  • Empirical work uses measured communication range and capture effects in LoRaSim, concluding that dynamic transmission parameters or multiple sinks can improve scalability.
  • LoRaWANSim and the ns-3 module both find that gateway duty-cycle limits constrain downlink acknowledgments and data messages as end-device density grows.

VII. DISCUSSION

The scalability analysis finds that confirmed upstream delivery is strongly constrained by limited downstream capacity. More gateways help, but the effect remains, while the modeled interference and path-loss assumptions leave clear avenues for future refinement.

  • Confirmed messages severely reduce upstream packet delivery ratios because gateway downstream capacity is limited.The analysis identifies downstream capacity as the limiting factor for confirmed upstream traffic.
  • The tested downstream data load had negligible impact on upstream messages, with little DS PDR difference between confirmed and unconfirmed downstream data.A significant difference appeared only for the single-gateway, μ = 10 scenario.
  • Modeling all interference as noise is a limitation because co-spreading-factor interference may require a stochastic model.The authors also identify path-loss tuning and downstream data-rate studies as future work.
  • The ns-3 module combines a LoRa error model with class A MAC behavior and supports upstream/downstream messages and multiple gateways.It models range and interference across LoRa spreading factors and coding rates.
  • Increasing gateway density delays but does not eliminate the deterioration of confirmed upstream delivery.Even a four-gateway network can retain low PDR during the 600-second data period.
Loading 1705.05899v1…