Post-Quantum Cryptography Nearly Doubles Transmission Time Across LoRa Wireless Networks

A controlled experimental investigation into secure wireless data transmission has revealed that integrating lattice-based post-quantum cryptography onto the physical layer of Long Range, or LoRa, networks nearly doubles transmission airtime and energy consumption, whereas lightweight symmetric encryption introduces virtually undetectable operational penalties. The peer-reviewed study, conducted by researchers Viacheslav Shkuratskyy, Aminu Bello Usman, Hamidreza Bagheri, Sam Hill, and Shuaib Karim Memon, evaluated the delicate operational balance between data compression, cryptographic protection, and physical-layer transmission efficiency. Deploying connected devices across environmental monitoring, precision agriculture, and critical infrastructure has historically relied on LoRa technology because of its robust chirp spread spectrum modulation, extensive geographic reach, and exceptionally low operating power. However, LoRa devices face strict duty-cycle regulations and an unyielding physical-layer payload ceiling of two hundred fifty-five bytes per frame. As edge artificial intelligence shifts computation and real-time inference directly onto remote sensor nodes, connected hardware increasingly must compress and transmit complex data structures such as visual images without violating regulatory bandwidth constraints or rapidly depleting compact battery reserves.

The engineering team addressed these fundamental wireless limits by designing an end-to-end processing pipeline that couples neural image compression with contrasting cryptographic protocols under an application-layer threat model. Because the native LoRa physical layer lacks built-in cryptographic mechanisms, raw data frames are inherently susceptible to passive eavesdropping, injection, modification, replaying, or unauthorized dropping by malicious actors. In high-stakes settings such as industrial telemetry or healthcare systems, leaving transmitted payloads unencrypted exposes sensitive infrastructure to compromise, whereas in ordinary environmental monitoring, arbitrary cryptographic overhead unnecessarily inflates transmission airtime and depletes battery reserves. To reconcile these competing demands, the researchers deployed a convolutional variational autoencoder, or VAE, directly on the transmitting hardware. The compression model was tasked with reducing high-resolution environmental imagery into compact binary bitstreams before LoRa modulation. By creating a substantial reduction in raw data volume, the neural autoencoder establishes an operational resource cushion, allowing constrained edge nodes to accommodate subsequent cryptographic metadata and frame structures without exceeding physical hardware limits or standard channel occupancy rules.

To rigorously quantify how cryptographic workloads interact with this compressed transmission baseline, the authors designed a standardized experimental architecture operating within an isolated radio-frequency environment. All transmission sequences were executed inside an enclosed Faraday cage to eliminate ambient radio interference, maintain repeatable signal paths, and comply with strict ISM-band duty-cycle regulations. The experimental hardware paired a Raspberry Pi single-board computer with a Semtech SX1262 transceiver module operating at eight hundred sixty-eight megahertz, conforming to European telecommunications standards. Radio parameters were fixed across every test, utilizing a spreading factor of seven, an operating bandwidth of one hundred twenty-five kilohertz, an error-correcting coding rate of four-fifths, and a constant transmission output power of twenty-two decibels referenced to one milliwatt. A statutory duty cycle of one percent was strictly enforced by adjusting temporal pauses between successive wireless bursts. During testing, electrical current and supply voltage were continuously recorded using an INA226 digital monitoring module, enabling high-precision empirical measurement of electrical energy consumption across both computational preparation and physical radio transmission phases.

The test workload centered on an uncompressed, high-resolution environmental capture measuring approximately 5.5 megabytes, chosen specifically to reflect the detailed, feature-rich scenes encountered by intelligent field monitoring systems. When processed through the convolutional variational autoencoder on the Raspberry Pi transmitter, this large graphic file was compressed into a dense, one-thousand-sixteen-byte bitstream, achieving an extraordinary payload reduction of 99.98 percent. The authors noted that executing this computationally intensive compression locally at the network edge is critical for preserving energy, as transmitting uncompressed visual data over wide-area low-power links would be physically impossible. Because the compressed bitstream still exceeded LoRa’s single-frame maximum capacity of two hundred fifty-five bytes, the source payload had to be segmented into consecutive discrete chunks.

Across one hundred repeated experimental runs per configuration, the researchers benchmarked three distinct operational modes: a compression-only baseline, compression paired with lightweight ChaCha20 encryption, and compression paired with a hybrid post-quantum architecture combining CRYSTALS-Kyber key encapsulation with ChaCha20 symmetric encryption. Symmetric cryptographic parameters—consisting of a two-hundred-fifty-six-bit secret key, a ninety-six-bit nonce, and associated integrity data—were generated offline in advance to isolate core encryption behavior from auxiliary key-distribution delays. In the lightweight configuration, ChaCha20 performed stream encryption across the VAE bitstream, generating authentication tags via the Poly1305 algorithm to guarantee message authenticity and tampering detection. Under the hybrid post-quantum scheme, the transmitter executed Kyber-512 key encapsulation using a pre-shared public key, generating a temporary shared session key alongside an eight-hundred-byte Kyber ciphertext.

The empirical findings revealed that lightweight encryption using ChaCha20 introduces a negligible performance penalty, remaining virtually indistinguishable from the unencrypted VAE baseline. In both the compression-only and ChaCha20 configurations, the mean physical transmission time remained locked at approximately 5.03 seconds, exhibiting exceptionally tight consistency with standard deviations below 0.006 seconds across trials. Computational processing consumed an average of 3.80 joules for compression alone and remained effectively identical when ChaCha20 encryption was engaged, with most processing durations clustering tightly between 8.8 and 9.8 seconds. Transmission energy consumption for ChaCha20 fluctuated within a narrow margin between 1.49 and 1.55 joules, matching the unencrypted baseline and yielding calculated transmission energy overheads that hovered harmlessly between negative two percent and positive three percent. The authors observed minor negative overhead percentages in scattered runs, attributing these small anomalies to physical measurement variability, slight thermal fluctuations, and operating-system execution jitter rather than intrinsic algorithmic gains.

In sharp contrast to the lightweight cipher, integrating post-quantum security imposed severe physical and temporal costs that directly challenged LoRa’s foundational operating premises. The addition of the eight-hundred-byte CRYSTALS-Kyber ciphertext significantly expanded the aggregate payload volume that needed to traverse the physical radio link. Because an eight-hundred-byte payload far exceeds LoRa’s two-hundred-fifty-five-byte frame boundary, the key encapsulation data alone forced substantial frame fragmentation. Consequently, the mean transmission duration for the post-quantum configuration jumped to approximately nine seconds, representing an average airtime overhead of 80.12 percent compared to the unencrypted baseline. Radio transmission energy consumption rose proportionally, climbing from approximately 1.5 joules in the baseline trials to an average of 2.70 joules under CRYSTALS-Kyber, reflecting an empirical transmission energy overhead ranging between seventy-five and eighty-five percent across the repeated trials.

Crucially, the experimental measurements demonstrated that while post-quantum cryptography slightly increased local processing workloads, its dominant performance bottleneck stems from channel transmission rather than on-device mathematical computation. When examining on-device preparation times and energy figures, CRYSTALS-Kyber operations clustered near the upper edge of the common 8.8-to-9.8-second and 3.7-to-4.0-joule intervals, reflecting a modest computational overhead for lattice polynomial calculations. However, the subsequent requirement to broadcast the expanded ciphertext across the airwaves generated a profound resource penalty. In low-power wide-area networks governed by strict regional duty cycles, nearly doubling the radio transmission duration severely restricts how frequently a sensor node can report data. Furthermore, elevated transmission energy rapidly exhausts the finite battery capacities typical of remote environmental installations, demonstrating that the physical size of post-quantum cryptographic primitives poses a far greater barrier to edge deployment than microcontroller execution speed.

The experimental trials also highlighted a critical reliability limitation directly linked to post-quantum payload expansion and excessive packet fragmentation. Across the one hundred repeated post-quantum transmission sequences conducted inside the shielding of the Faraday cage, one trial suffered lost packets that prevented the receiver from assembling the complete bitstream, resulting in failed image reconstruction. Despite an entirely controlled laboratory environment devoid of external radio-frequency interference or dynamic physical obstacles, splitting the combined ciphertext and image bitstream across multiple LoRa frames increased packet loss vulnerability. The authors noted that in real-world agricultural or industrial deployments, ambient interference, multi-path signal fading, and environmental channel degradation would inevitably amplify frame dropping. Because corrupted or missing key-encapsulation frames render subsequent encrypted payloads completely unrecoverable, excessive fragmentation introduces severe systemic risks that threaten data integrity and network dependability.

Ultimately, the researchers concluded that combining advanced neural compression with lightweight symmetric ciphers offers an immediate, deployable path for secure edge-enabled IoT architectures, whereas post-quantum cryptography requires substantial architectural refinement before achieving field readiness on low-power wide-area links. Neural compression successfully carves out an operational margin by shrinking multi-megabyte sensory data by 99.98 percent, easily offsetting the tiny footprint of algorithms like ChaCha20 without compromising battery longevity or transmission reliability. For post-quantum mechanisms to become practical across constrained radio channels like LoRa, future engineering efforts must focus on optimizing packet segmentation, integrating robust forward error correction, implementing adaptive retransmission protocols, and developing specialized post-quantum algorithms featuring dramatically smaller ciphertext sizes. Until such compact primitives mature, low-power monitoring networks must carefully balance theoretical quantum resistance against immediate transmission efficiency and delivery success.

Subject of Research: Evaluation of lightweight and post-quantum encryption overhead for compressed image transmission across the LoRa physical layer

Article Title: Evaluating lightweight and post-quantum encryption for compressed image transmission over LoRa in the era of edge AI

Article References: Shkuratskyy, V., Usman, A. B., Bagheri, H., Hill, S., & Memon, S. K. (2026). Evaluating lightweight and post-quantum encryption for compressed image transmission over LoRa in the era of edge AI. Discover Electronics, 3(1), Article 114. https://doi.org/10.1007/s44291-026-00261-4

Image Credits: AI Generated

DOI: 10.1007/s44291-026-00261-4

Keywords: LoRa physical layer, edge AI, variational autoencoder, image compression, lightweight cryptography, ChaCha20, CRYSTALS-Kyber, post-quantum cryptography, key encapsulation mechanism, Semtech SX1262, Faraday cage testing, INA226 sensor, payload fragmentation, airtime overhead, transmission energy, Internet of Things, duty-cycle regulations, packet delivery reliability