Areas to Investigate

1. Requirements for Data
2. How to serve data to public (sigmf-torrent)
3. How do we process in real-time
4. How do we visualize for bipeds
5. How to we design our ML solution (see below)

Goal for this Week

Utilize a multi-stage pipeline on raw IQ data in order to better filter, store less data, and hopefully find ET in the fire-hose of raw data. We envision this pipeline as modular filters that can be swapped depending on their accuracy and throughput. Higher throughput filters should happen earlier in the pipeline when possible.

Proposed ML Pipeline

Stage 0 Data Reduction

FFT, threshold based on power => classify as HAS_POWER or NOISE. if HAS_POWER send to Stage 1

Stage 1 Anomaly Classifier

Small neural net => classify as ABNORMAL or TYPICAL. if ABNORMAL send to Stage 2

Stage 2 Isolate signals

Network to draw boxes around signals => frequency shifts => Filter => Downsample => send to Stage 3

Stage 3 "ET" Classifiers

Utilize best Signal Differentiators for ET, cribbed from Steve Croft's HowToFindET

Localization (Best)

Classify signals as being from WITHIN THE SOLAR SYSTEM or EXTRASOLAR. We are only interested in extra-terrestrial signals. Terrestrial interferers should be filtered out. This will only leave signals from satellites or martians or deep space origins. Multiple options to do this:

• Aim antennas at different locations and filter out signals that appear on multiple
• Use a calibrated array to figure out AOA and remove "earth angles"
• Any other ideas

Doppler (Better)

We expect technosignatures from earth have very low doppler (how low) that should differentiate them from technosignatures from a galaxy far far away.

"Narrow" Bandwidth (Good)

Filter out any large bandwidth signals

Experiments

We performed 6 experiments using 4 different antennas: 1g, 2a, 2b, 3L

• All with Y polarization
• Recorded on a X310 w/dual twinrx daughterboards.
• Run using IF frequency of 629.1456MHz.
• Radios tuned to 620MHz when collecting data.
• Syncing to 10MHz clock.

Data will be stored in sigmf format and can be found: xxx

Target #	Frequency (MHz)	Description
0a,0b	1420,1575	Cassiopia-A
1a,1b	1420,1575	3c84 - Radio Galaxy
2a,2b	1420,1575	Tracked GPS satellite
3a,3b	1420,1575	Celestial North

Experiment #	Beams (0,1,2,3)
0	(T0a, T0a, T0a, T0a)
1	(T1a, T1a, T1a, T1a)
2	(T2b, T2b, T2b, T2b)
3	(T3a, T3a, T3a, T3a)
4	(T0a, T1a, T2a, T3a)
5	(T0b, T1b, T2b, T3b)