GhostHunter: Exploring the Noise Floor with Anti-LIF Spiking Neural Networks

SatNOGS
1

Hi everyone,

I’m Edwin Romano, and I’ve been focusing my research on a specific niche in satellite communications: finding and recovering data where traditional decoders see only noise.

Traditional DSP algorithms often reach their operational limits when the carrier is buried deep in the noise floor or when dealing with hardware-induced artifacts. To address this, I have developed the Anti-LIF (Leaky Integrated Fire) v3.9 architecture, a Spiking Neural Network (SNN) specifically designed for “Signal Intelligence” and deep-noise telemetry recovery.

Searching Where Nobody Else Looks

My approach is not just about decoding strong signals, but about rescuing data from observations often classified as “Bad” or “Unknown.” By utilizing neuromorphic dynamics, the system can identify structured patterns (headers, sync words, and frame counters) within the noise floor that standard receivers typically ignore.

Technical Capabilities Validated:

Edge Detection Intelligence: The SNN identifies high-speed transitions (e.g., detecting double-edge events at ~19.3k bps for standard 9.6k signals), allowing for precise clock recovery in unstable conditions.

Resilient Synchronization: Successful extraction of primary sync markers (ASM) from raw SigMF-IQ data where traditional phase-locked loops (PLL) fail to maintain lock.

Protocol Versatility: Adaptive analysis for GMSK, AFSK, ASK, and asynchronous pulsed transmissions.

Efficiency by Design (Sparsity)

Beyond robustness, the Anti-LIF architecture is built for the extreme power constraints of orbital hardware:

Network Sparsity: 95.17% (The neurons only fire when there is actual information to process).

Energy Impact: A 93.5% reduction in computational operations compared to equivalent standard Artificial Neural Networks (ANN).

Real-time Latency: Inference times under 0.25 ms, enabling high-frequency baseband analysis.

Let’s Rescue Some Data

I am looking to collaborate with the community to test the limits of this technology. If you have “Unknown” status observations or recordings with extremely faint signals that you’d like to investigate, I would be happy to run them through the Anti-LIF core to see what we can recover.

I believe SNNs are the future of Edge Computing in space, allowing us to process massive amounts of data with minimal power.

You can find further technical details in the attached REV 2026 White Paper.

Best regards,

Edwin Romano

Founder, Anti-LIF Neuromorphic Architecture

2 Likes
2

Sounds nice! I can’t find the attached whitepaper though.

I have 1.3TB of compressed iq file recordings from my satnogs station if you want data to run on.

3

Hi mfalkvidd, thank you for the feedback and that incredible data offer!

Whitepaper & Theory: My apologies for the oversight. You can find the full technical architecture of the Anti-LIF SNN in this paper on TechRxiv: Anti-LIF: A Universal Spiking Neural Network Architecture. It details how the beta dynamics (\beta=0.88) manage to operate near Shannon’s limits in high-noise environments.

Code & Proof of Concept: To see the SNN in action, I’ve prepared this GitHub repository: Anti-LIF-Universal-SNN. It contains a demo specifically designed to extract signals within a high-interference Wi-Fi noise environment, showcasing the core’s ability to trigger spikes only on structured data.

The 1.3TB Challenge: This is a goldmine. I would love to start by running the SNN on a subset of your “Unknown” status observations where traditional DSP failed. Solving those “lost” signals would be the ultimate benchmark for GhostHunter.

Community Goal: My aim is to contribute cleaned bitstreams back to the community, just as I did with the Technosat (Observation #13141687), where I recovered telemetry from what was considered a lost signal.

Looking forward to collaborating and diving into those recordings!

Best regards,

Edwin Romano

Email: temporaledwin58@gmail

1 Like
4

very interesting. any software can we try? is it opensource?

5

The software is currently proprietary and not open-source, as it is a specialized architecture optimized for space applications and low-SNR environments. We are currently developing a SaaS platform to allow users to process their own datasets in the future.

In the meantime, I am accepting raw IQ data for Proof of Concept (PoC) processing under the following terms:

Free Tier: Limited to files up to 100MB.

Processing Time: Please note that free requests have a lower priority and will take longer to process as they are handled manually.

Deliverables: I will return the cleaned bitstream and a detailed structural report.

This allows you to validate the engine’s performance on your toughest signals without needing the source code.

Check the latest mission results here: :backhand_index_pointing_right: GhostHunter | SNN Signal Recovery Lab "

6

IQ file: here

this is afsk. i’m curious how much data can be restored.

thank you

7

​*“Thanks for the challenge! AFSK is a great test for our Anti-LIF neuromorphic engine. I’ll run it through our ‘Frequency Discriminator’ neuron layers to see how much of the frame structure we can recover from the noise floor. I’ll get back to you with the results and a full report. Stay tuned!”*

8

superb. btw is the result also in clean iq file? if yes i hope can share here to.

9

Glad you liked the results! Currently, the Anti-LIF engine outputs the raw bitstream and spike timestamps directly to maximize data integrity. However, I can look into synthesizing a ‘reconstructed’ clean IQ file from the detected spikes if you’re interested in running it through standard decoders. For now, the most accurate data is in the bitstream CSV.

10

yes. i’m interest

11

I have completed the forensic analysis of the baseband_435887855Hz signal. After processing the data stream through my GhostHunter (Anti-LIF v4.0) neuromorphic engine, I have successfully identified a Pulse Interval Modulation (PIM/Burst) signal operating at a native Baud Rate of 1333.33 Hz.

By utilizing Spiking Neural Networks (SNN), we successfully synchronized the clock and extracted the payload from the primary burst detected at T=59.10s.

Protocol: Structured frame featuring an #R header, comma delimiters, and a q5 integrity footer.

Satellite Status: Nominal telemetry readings showing 94% energy (240u), 161u inertia stability, and a 216u operating temperature (direct solar exposure).

I will be uploading the processed data and the full GhostHunter code to my GitHub repository soon.

12

Hello,

I am writing to inform you that I have completed the forensic analysis of the baseband file you provided. Using the GhostHunter (Anti-LIF v4.0) neuromorphic engine, we successfully penetrated the noise floor and extracted real-time telemetry from the system.

We have identified a PIM (Pulse Interval Modulation) burst signal operating at a native Baud Rate of 1333.33 Hz. Data recovered from T=59.10s reveals a structured frame (#R … q5) with sensor readings showing a 94% energy level and nominal operational status.

I have published all the evidence—including spike files, the bitstream, and interpreted results—so you can personally verify the accuracy of these findings.

You can view the final results here:

:link: Reporte GhostHunter-M3 | SNN Forensic Analysis

13

Hi everyone. I have concluded the stress tests for my neuromorphic engine, GhostHunter (Anti-LIF v4.0), on the baseband_435887855Hz signal. What began as a detection challenge has resulted in the full telemetry extraction of a proprietary protocol.

Operation Summary:

Low-Level Detection: We successfully penetrated the noise floor of the baseband signal, identifying a Pulse Interval Modulation (PIM/Burst) instead of conventional protocols.

Clock Synchronization: The SNN model identified a native Baud Rate of 1333.33 Hz (0.75 ms base pulse).

Burst Localization: The primary data packet was located with surgical precision at T=59.10s.

Data Extraction: We decoded a structured frame featuring an #R header and a q5 footer. Sensors report: 94% Energy (240u), nominal stability (161u), and elevated temperature (216u).

Call for Verification:

This mission served as a proof of concept to demonstrate that Spiking Neural Networks can “see” intelligence where traditional decoders fail. We want the community to see for themselves whether these results hold true; transparency is the core of our research.

You can view the detailed report, download the raw data, and verify the findings at the following link:

:link: Reporte GhostHunter-M3 | SNN Forensic Analysis

14

Given that the tool is currently proprietary and not open-source, I don’t think it is the appropriate place here to discuss it in this forum that supports the open source values.

Thus I’m going to close this topic here and if and when the tool will be open source, we are glad to host discussions about it.

1 Like