More Like this

1. Cloud-LoRa: Enabling Cloud Radio Access LoRa Networks Using Reinforcement Learning Based Bandwidth-Adaptive Compression

   Shahid, Mohammad_Osama; Koch, Daniel; Rahghuram, Jayaram; Krishnaswamy, Bhuvana; Chintalapudi, Krishna; Banerjee, Suman (April 2024, 21st USENIX Symposium on Networked Systems Design and Implementation (NSDI 24))

   The Cloud Radio Access Network (CRAN) architecture has been proposed as a way of addressing the network throughput and scalability challenges of large-scale LoRa networks. CRANs can improve network throughput by coherently combining signals, and scale to multiple channels by implementing the receivers in the cloud. However, in remote LoRa deployments, a CRAN's demand for high-backhaul bandwidths can be challenging to meet. Therefore, bandwidth-aware compression of LoRa samples is needed to reap the benefits of CRANs. We introduce Cloud-LoRa, the first practical CRAN for LoRa, that can detect sub-noise LoRa signals and perform bandwidth-adaptive compression. To the best of our knowledge, this is the first demonstration of CRAN for LoRa operating in real-time. We deploy Cloud-LoRa in an agricultural field over multiple days with USRP as the gateway. A cellular backhaul hotspot is then used to stream the compressed samples to a Microsoft Azure server. We demonstrate SNR gains of over 6 dB using joint multi-gateway decoding and over 2x throughput improvement using state-of-the-art receivers, enabled by CRAN in real-world deployments. 

   more » « less
2. TnB: resolving collisions in LoRa based on the peak matching cost and block error correction

   https://doi.org/10.1145/3555050.3569132  

   Rathi, Raghav; Zhang, Zhenghao (November 2022, International Conference On Emerging Networking Experiments And Technologies)

   ABSTRACT LoRa has emerged as one of the main candidates for connecting low-power wireless IoT devices. Packet collisions occur in LoRa networks when multiple nodes transmit wireless signals simultaneously. In this paper, a novel solution, referred to as TnB, is proposed to decode collided LoRa signals. Two major components of TnB are Thrive and Block Error Correction (BEC). Thrive is a simple algorithm to resolve collisions by assigning an observed signal to a node according to a matching cost that reflects the likelihood for the node to have transmitted the signal. BEC is a novel algorithm for decoding the Hamming code used in LoRa, and is capable of correcting more errors than the default decoder by jointly decoding multiple codewords. TnB does not need any modification of the LoRa nodes and can be adopted by simply replacing the gateway. TnB has been tested with real-world experimental traces collected with commodity LoRa devices, and the results show that TnB can increase the median throughput by 1.36× and 2.46× over the state-of-the-art for Spreading Factors (SF) 8 and 10, respectively. Simulations further show that the improvement is even higher under more challenging channel conditions. 

   more » « less
3. Concurrent interference cancellation: decoding multi-packet collisions in LoRa

   https://doi.org/10.1145/3452296.3472931  

   Shahid, Muhammad Osama; Philipose, Millan; Chintalapudi, Krishna; Banerjee, Suman; Krishnaswamy, Bhuvana (August 2021, Proceedings of the 2021 ACM SIGCOMM Conference)

   LoRa has seen widespread adoption as a long range IoT technology. As the number of LoRa deployments grow, packet collisions undermine its overall network throughput. In this paper, we propose a novel interference cancellation technique -- Concurrent Interference Cancellation (CIC), that enables concurrent decoding of multiple collided LoRa packets. CIC fundamentally differs from existing approaches as it demodulates symbols by canceling out all other interfering symbols. It achieves this cancellation by carefully selecting a set of sub-symbols -- pieces of the original symbol such that no interfering symbol is common across all sub-symbols in this set. Thus, after demodulating each sub-symbol, an intersection across their spectra cancels out all the interfering symbols. Through LoRa deployments using COTS devices, we demonstrate that CIC can increase the network capacity of standard LoRa by up to 10x and up to 4x over the state-of-the-art research. While beneficial across all scenarios, CIC has even more significant benefits under low SNR conditions that are common to LoRa deployments, in which prior approaches appear to perform quite poorly. 

   more » « less
4. BYOG : Multi-Channel, Real-time LoRaWAN Gateway Testbed using General-purpose Software Defined Radio

   https://doi.org/10.1145/3656299  

   Shahid, Muhammad Osama; Krishnaswamy, Bhuvana (June 2024, Proceedings of the ACM on Networking)

   Adaptive Data Rate (ADR) is used by multi-channel LoRaWANs to meet the demanding capacity needs of LoRa networks. The network server running ADR in each channel determines the optimum data rate and assigns the appropriate spreading factor for each LoRa device to maximize the network throughput. This in turn requires the gateway to be capable of receiving LoRa packets of all possible spreading factors. Existing gateways achieve this by using multiple RF front ends, increasing the overall cost and complexity. In this work, we propose BYOG (Bring Your Own Gateway), a LoRaWAN receiver that can receive and decode 10 channels simultaneously in real-time. Towards this pipeline, we develop self-dechirping, an SF-agnostic packet detection algorithm that also detects the spreading factor of the packet. This computationally lightweight algorithm can be implemented on any general-purpose software-defined radio, bringing down the cost and ease of LoRaWAN gateway implementations. BYOG will enable research and development in LoRaWAN ADR. Using experimental, real-world datasets, we show that the proposed algorithm can detect the spreading factor accurately and operate over a wide range of SNRs using three different SDRs (RTL-SDR, HackRF One, USRP B210). BYOG performs as well as a high-end LoRaWAN gateway in terms of network throughput. 

   more » « less
5. SLoRA: Scalable Serving of Thousands of LoRA Adapters

   Sheng, Ying; Cao, Shiyi; Li, Dacheng; Hooper, Coleman; Lee, Nicholas; Yang, Shuo; Chou, Christopher; Zhu, Banghua; Zheng, Lianmin; Keutzer, Kurt; et al (May 2024, Proceedings of Machine Learning and Systems 6 (MLSys 2024))

   The "pretrain-then-finetune" paradigm is commonly adopted in the deployment of large language models. Low-Rank Adaptation (LoRA), a parameter-efficient fine-tuning method, is often employed to adapt a base model to a multitude of tasks, resulting in a substantial collection of LoRA adapters derived from one base model. We observe that this paradigm presents significant opportunities for batched inference during serving. To capitalize on these opportunities, we present S-LoRA, a system designed for the scalable serving of many LoRA adapters. S-LoRA stores all adapters in the main memory and fetches the adapters used by the currently running queries to the GPU memory. To efficiently use the GPU memory and reduce fragmentation, S-LoRA proposes Unified Paging. Unified Paging uses a unified memory pool to manage dynamic adapter weights with different ranks and KV cache tensors with varying sequence lengths. Additionally, S-LoRA employs a novel tensor parallelism strategy and highly optimized custom CUDA kernels for heterogeneous batching of LoRA computation. Collectively, these features enable S-LoRA to serve thousands of LoRA adapters on a single GPU or across multiple GPUs with a small overhead. Compared to state-of-the-art libraries such as HuggingFace PEFT and vLLM (with naive support of LoRA serving), S-LoRA can improve the throughput by up to 4 times and increase the number of served adapters by several orders of magnitude. As a result, S-LoRA enables scalable serving of many task-specific fine-tuned models and offers the potential for large-scale customized fine-tuning services. The code is available at this https URL 

   more » « less