Modulation Recognition Using Machine Learning Approaches in CubeSat Signals - Poster

Aug 3, 2026

Upload by Aria

CubeSats are widely used in the small satellite sector, yet many failures occur along the communication chain, including difficulty establishing contact. This poster proposes using machine learning to decode spacecraft signals by targeting modulation detection within the OSI stack. A GNU Radio–based synthetic RF dataset is generated from packetized telemetry, adding randomized code blocks and realistic noise. Neural networks, including a ConvLSTM approach, are trained and evaluated under Gaussian and phase noise, with preliminary results showing BPSK as most accurately detected.

LanguageEnglish

CategoryTechnology

FormatPDF

Pages1

Reading Time3 min

AccessFree to download

Abstract

Gaussian noise impact over different modulations

Methodology

Problem

Discussion

Future Work

Contact

References

Modulation Recognition Using Machine Learning Approaches in CubeSat Signals Vidushi Jain, PhD candidate 1 ; Regina Lee, PhD 1 ; Franz Newland, PhD 2 1York University, 2 University of Ottawa | Abstract | | --- | | • CubeSats make up a large subset of the small satellite industry.
• Many CubeSat failure occur along the communication chain and failure to establish contact with the satellite
• GNU radio is a platform commonly used for processing signal, with the help of GR-satellite
• Establishing such decoding chains are a time-consuming task
• Machine learning can be used as a tool to decode signal coming from spacecrafts
• The focus is to target different layers of the OSI stack, and this poster demonstrates modulation detection layer | Gaussian noise impact over different modulations Percentage detetcted Power (dB) BPSK GMSK GFSK QPSK Figure 2. Illustration of CubeSat OSI stack and noise insertion for synthetic Dataset generation Methodology | Problem | | --- | | | Figure 1. Subsystem contribution to CubeSat failure 0, 30 and 90 days after ejection . • New technology can aid in overcoming or reducing CubeSat failure rate • Development of software tools to automate the signal processing from CubeSat signals will help in the success rate of CubeSats • Using already existing GNU radio software an RF dataset can be synthesized to replicate CubeSat signal. • Machine learning algorithm can be used to detect modulation sequences in CubeSat signals • Different methods and algorithms can be used to compare the efficiency of modulation detection within space-based signals Figure 3. Modulation schemes and noise types used in simulated dataset • Simulated CubeSat telemetry packetized and inserted into frames • Randomized code blocks modulated according to several standard CubeSat transponder modulation schemes; noise added to generate a simulated noisy space dataset relevant to most space missions • Simulated dataset used to train neural networks to detect the modulation scheme used, with a bootstrap method • To test the usability of the algorithm a ConvLSTM algorithm was used on a gaussian and phase noise synthesized signal Phase noise impact over different modulations Power (dB) BPSK GMSK GFSK QPSK Chart 1. Modulation detection performance for preliminary simulated data Discussion • From both noise it is evident that BPSK was detected more accurately than other modulation schemes • GMSK had the worst performance, and detection did get more difficult as the signal to noise ratio decreased. • Analysis chain has been proven, ready for full simulated dataset and subsequent testing on actual flight data . | Future Work | | --- | | • Further generation of signals using different modulation schemes and noise sources
• Testing the algorithm with a random distribution of the synthesised signal
• Checking the feasibility of the algorithm using SATNOGs received signal on an operating CubeSat
• Comparing the efficiency of different algorithms | Contact Vidushi Jain York University [Email: Vidushi.jain.jv@gmail.com](Email: Vidushi.jain.jv@gmail.com) [Phone: 6476331956](Phone: 6476331956)[ ](Phone: 6476331956)[SSC24-WP1-27](SSC24-WP1-27) References 1. Modenini, A., & Ripani, B. (2023) . A tutorial on the tracking, telemetry, and Command (TT&C) for Space Missions. IEEE Communications Surveys & Tutorials, 25(3), 1510–1542. [https://doi.org/10.1109/comst.2023.3287431](https://doi.org/10.1109/comst.2023.3287431) 2. Scholl, S. (2022, June 26). RF signal classification with synthetic training data and its real-world [performance. arXiv.org](performance. arXiv.org). [https://arxiv.org/abs/2206.12967](https://arxiv.org/abs/2206.12967) 3. Astrocast telemetry downlink specification. HB9HSLU. (n.d.). [https://hb9hslu.ch/wordpress/?page_id=691](https://hb9hslu.ch/wordpress/?page_id=691) 4. Pontes, R. P., Martins, E., Ambrósio, A. M., & Villani, E.

FAQs

Why is modulation detection important for CubeSat communications?

Many CubeSat failures occur along the communication chain, making it difficult to establish contact. Automated modulation detection improves the ability to decode downlink signals reliably.

How is the training dataset for the machine learning model created?

An RF dataset is synthesized using existing GNU Radio tools. Simulated telemetry packets are inserted into frames, code blocks are modulated using standard CubeSat schemes, and noise is added to form a noisy space-relevant dataset.

Which modulation scheme performed best in the preliminary results?

BPSK was detected more accurately than the other tested modulation schemes. Detection performance for GMSK was the worst and degraded as signal-to-noise ratio decreased.

Aug 3, 2026

Upload by Aria