Collaborative Research: NeTS: Medium: Towards High-Performing LoRa with Embedded Intelligence on the Edge

Project Goals

Recent years have witnessed the emergence of Low-Power Wide-Area Networks (LPWANs) as a promising mechanism to connect billions of low-cost Internet of Things (IoT) devices for wide-area data collection. Among all the existing LPWAN technologies, LoRa has attracted significant interest from both academia and industry due to its open-source and unlicensed spectrum nature. Ideally, LoRa allows IoT devices to communicate with gateways several or even tens of miles away. Unfortunately, recent studies show that the communication range of LoRa is far from the expectation in non-line-of-sight (NLoS) real-world environments. This is because the blockage attenuation in real-world settings could severely degrade the Signal-to-Noise Ratio (SNR) of LoRa packets, causing decoding failures even at a sub-kilometer distance. To address this critical issue, the key is to obtain extra SNR gain. State-of-the-art works obtain extra SNR gain by leveraging information collected from multiple LoRa devices or gateways. Although these approaches have achieved impressive SNR gains, they all require extra overhead to calibrate and synchronize the LoRa devices and gateways during deployment. Such a requirement, unfortunately, makes the real-world deployment of LoRa systems extremely hard to scale. The project goals are to (1) uses the feature extraction capability of modern deep neural networks (DNN) and the computational resources now available on edge devices to create better performing LoRa networks. (2) reduce the cost of deploying and maintaining real-world LoRa networks, and thus accelerate adoption of wide-area IoT applications which will enhance efficiency of smart cities and other verticals.

Intellectual Merit

This project targets the following four technical contributions.

  • DNN Decoder:For a single gateway, we will design a multi-dimension multi-resolution neural-enhanced LoRa decoder with various DNN models to effectively capture LoRa signals’ multi-dimension features even when the signal strength is far below the noise floor. By doing so, we can improve the energy efficiency of LoRa.
  • Encoder-decoder Co-design: For a single gateway, we will develop a neural-efficient SF-configuration-based encoder-decoder co-design and an energy-efficient mixed-chip encoder-decoder co-design, in which the newly designed encoder provides a much richer feature space for the neural-enhanced decoding. By doing so, we extend the communication range and data rate of LoRa, respectively.
  • Spatial Deiversity Decoder: For multiple gateways, we will develop a co-design of a neural-enhanced multi-gateway symbol decoder and a frequency-aware encoder that utilizes the spatial diversity of multiple gateways to enhance the SNR of received signals. By doing so, we increase the energy efficiency of LoRa
  • End-to-End DNN-Empowered LoRa Platform Design and Prototype Development: Existing LoRa platforms are not designed to support DNN-based encoding/decoding. We will design and develop the first end-to-end DNN-empowered LoRa platform, which integrates the innovations proposed before.

Broader Impact

  • Impact on Society:The outcomes of the proposed project are expected to push the frontier of the state-of-the-art LoRa technology, and to advance the deployment and maintenance feasibility of real-world LoRa networks. Our project will also benefit a wide range of real-world LoRa-based IoT applications related to smart cities, precision agriculture, smart industry, and smart logistics. Moreover, we believe our proposed research is an important step towards using ML techniques to solving low-power wireless networking problems. We will make our codes and collected datasets open-source. As such, our project can benefit other researchers in wireless networking, AI, and edge computing communities.
  • Impact on Undergraduate and Graduate Research: Undergraduate and graduate research is a very important component of the PIs’ career as educators and researchers. For the proposed project, the PIs will actively recruit undergraduate and graduate students and work closely with them as collaborators and mentors, and will co-author with them on potential publications.
  • Impact on K-12 Education: All three PIs have accumulated valuable experiences in K-12 education. PI Cao had the experience of training the teachers from primary schools for an IoT programming competition and giving an IoT application talk to the primary school students. Since joining MSU in Fall 2014, Co-PI Zhang has been developing a comprehensive K-12 education program in intelligent computing systems. As an example, in Summer 2016, Co-PI Zhang participated in the High School Engineering Institute (HSEI) Program at MSU where he led a hands-on project to guide a group of 43 high school students to build a prototype of Fitbit-like wearable activity tracker using Arduino microcontroller board and accelerometer. PI Chen taught in the COMPS club (Computer Olympia and Mathematical Problem Solving) at the local Buchholz High School and has been continuously involved in the activities of the club as an advisor. The proposed project will help the PIs to develop new educational modules for future K-12 education programs.
  • Dissemination of Research Results: The PIs plan to use three venues to broadly disseminate the research results: scientific publications, open source and dataset distribution, and industrial collaboration. (1) Scientific Publications: the PIs will continue their tradition of publishing research results in top-tier conferences and journals. (2) Open Source and Dataset Distribution: the PIs plan to make the hardware and software that they develop for this project open source, and publish the suitably anonymized and ethically approved dataset. All three PIs have strong records of doing this in the past. (3) Agriculture Community Collaboration: the PIs will leverage their established collaborative relationships with the agriculture community partners from MSU to disseminate their research results to the agricultural industry with the goal to make an impact on the development of smart agriculture.

Activities

  • Year 1:
    • For research thrust II, we design new features for encoding and optimize the corresponding decoding methods in various application cases.
    • For research thrust V, we design a new hardware prototype at the LoRa node.
  • Year 2:
    • For research thrust I, based on our preliminary work NELoRa, we optimize our model training method to evaluate the performance gain and computation efficiency on large SFs (e.g., 11 and 12).
    • For research thrust I, we design a geolocation-aware FL framework, which support to train our ML decoder distributedly across different countries.
    • For research thrust II, based on the encoder-decoder co-design framework ChirpTransformer (Year 1), we carefully study the reliability and computation efficiency of neural-enhance decoder in decoding symbol-hopping encoded symbols.
    • For research thrust V, we design a satellite+LoRa framework to establish an end-to-end MLLoRa system in rural areas.
  • Year 3:
    • For research trhust I, we design a new DNN decoder with hierachical structures, which obtains fine-grained temporal-spatial features to enhance the ability of signal denoising and reducing the computation complexity simultaneously.
    • For research thrust I, we continue to complete the design of the geolocation-aware FL framework, which support to train our ML decoder distributedly across different countries.

Outputs

  • Year 1:
    • Yidong Ren, Wei Sun, Jialuo Du, Huaili Zeng, Younsuk Dong, Mi Zhang, Shigang Chen, Yunhao Liu, Tianxing Li, Zhichao Cao, "Demeter: Reliable Cross-soil LPWAN with Low-cost Signal Polarization Alignment", accepted by Proceedings of ACM the 24th Annual International Conference on Mobile Computing and Networking (MobiCom 2024)
    • Chenning Li*, Yidong Ren*, Shuai Tong, Shakhrul Iman Siam, Mi Zhang, Jiliang Wang, Yunhao Liu, Zhichao Cao, "ChirpTransformer: Versatile LoRa Encoding for Low-power Wide-area IoT", accepted by Proceedings of ACM the 28th Annual International Conference On Mobile Computing And Networking (MobiSys 2024)
  • Year 2:
    • Yidong Ren, Amalinda Gamage, Li Liu, Mo Li, Shigang Chen, Younsuk Dong, Zhichao Cao, "SateRIoT: High-performance Ground-Space Networking for Rural IoT", accepted by Proceedings of ACM the 24th Annual International Conference on Mobile Computing and Networking (MobiCom 2024)
    • Maolin Gan, Lanpeng Li, Samiul Alam, Li Liu, Luyang Liu, Mi Zhang, Zhichao Cao, "GeoFL: A Framework for Efficient Geo-Distributed Cross-Device Federated Learning", accepted by Proceedings of IEEE International Conference on Computer Communications (INFOCOM 2025)
    • Yidong Ren, Chenning Li, Shakhrul Iman Siam, Mi Zhang, Shigang Chen, Zhichao Cao, "Morph: ChirpTransformer-based Encoder-decoder Co-design for Reliable LoRa Communication", arXiv Pre-print arXiv:2507.22851
    • Maolin Gan, Khang Nguyen, Jialuo Du, Yidong Ren, Zhichao Cao, "NELoRa++: Towards General Neural-enhanced LoRa Demodulation", MSU EIN Group Technical Report, 2025
  • Year 3:
    • Khang Nguyen, Yidong Ren, Jialuo Du, Jingkai Lin, Maolin Gan, Shigang Chen, Mi Zhang, Chunyi Peng, Zhichao Cao, "LoRaSeek: Boosting Denoising Ability in Neural-enhanced LoRa Decoder via Hierachical Feature Extraction", accepted by Proceedings of ACM the 25th Annual International Conference on Mobile Computing and Networking (MobiCom 2025)
    • Maolin Gan, Lanpeng Li, Samiul Alam, Li Liu, Luyang Liu, Mi Zhang, Hucheng Zeng, Zhichao Cao, "GeoFL: A Framework for Efficient Geo-Distributed Cross-Device Federated Learning", accepted by ACM/IEEE Transactions on Networking (ToN), 2026.

Personnel

  • Dr. Zhichao Cao (Leading PI at MSU)
  • Dr. Mi Zhang (PI at OSU)
  • Dr. Shigang Chen (PI at UF)
  • Yidong Ren (Ph.D. student at MSU, Graduated in 2025)
  • Jinkai Lin (Ph.D. student at MSU)
  • Maolin Gan (Ph.D. student at MSU)
  • Shakhrul Iman Siam (Ph.D. student at OSU)
  • Benjamin Nagoshi (Ph.D. student at UF)