Sample RF Data Analysis Report- Two Examples

ITAR technical-data screening (required)

Uploaded files screened (2):

Burst QPSK Control Sensor-Interference-Clipping Event-30s-SigMF.sigmf-data
FHSS-like GFSK Telemetry-Dropouts-Periodic Noise Rises-30s-Sigmf.sigmf-data

ITAR Screen Result = CLEAR (no USML/ITAR indicators observed in filenames, content type, or extracted signal characteristics; appears consistent with commercial/industrial ISM-band telemetry/control).

Report 1 — Burst QPSK Control Sensor-Interference-Clipping Event-30s-SigMF.sigmf-data

1. Executive Summary

Duration analyzed: ~30 s (inferred from filename; see Data Characteristics)
Signals/events: bursty emitter with many short bursts plus one long burst; intermittent elevated wideband noise consistent with local interference/overload during the long burst.
Overall data quality: usable; metadata missing (SigMF .sigmf-meta not provided), so absolute frequency/power calibration is not available.

2. Data Characteristics

Data type: A) Raw complex IQ (interleaved int16)
Samples: 6,000,000 complex samples
Sample rate (inferred): ~200,000 S/s (6,000,000 / 30 s)
Timebase: derived from inferred sample rate (no timestamps in file)
Center frequency / units / gain: unknown (no .sigmf-meta)
DC offset: small (I ≈ +0.92 counts, Q ≈ +1.77 counts; negligible vs signal RMS)
Overload/clipping check (ADC-rail): no hard rail hits detected at ±32767; however, interference/overload-like behavior appears as a ~0.35 s interval of elevated median spectral power (details below). This can happen without “flat-top” ADC saturation (e.g., analog compression, AGC action, near-field impulsive EMI).

3. Detected Signals/Events

Burst structure (energy-threshold, envelope-based)

Burst count (threshold-based): ~749 detected transmissions/events
Typical burst duration: ~1.0–1.2 ms (median ≈ 1.17 ms)
Long burst: ~0.3506 s (dominant event)
Interpretation: consistent with a control/telemetry node sending brief packets plus an occasional longer transaction (e.g., status dump, retries, firmware/parameter push, or a prolonged collision/backoff scenario).

Spectral placement (relative to capture center)

Dominant energy peak (during long burst): ~+18 kHz offset from capture center (baseband estimate)
Occupied bandwidth (long burst, relative):
- ~100.8 kHz (95% OBW)
- ~177.2 kHz (99% OBW)
  (Note: wide 99% OBW is consistent with interference/noise elevation and/or multi-component content during the long event.)

Interference / noise-floor anomaly

Elevated “noise-median” interval: starts ~17.20 s, lasts ~0.35 s
Behavior: median spectral power rises by >6 dB relative to baseline during this interval, aligned with the long burst duration.
Interpretation: near-field industrial EMI (VFD/relay/supply switching), front-end compression, or a co-channel emitter becoming active during the long burst.

Modulation family evidence (conservative)

After coarse frequency correction using the long-burst peak, a 4-cluster constellation structure is present (rotated quadrants), consistent with QPSK-like signaling or another 4-state PSK/QAM family under filtering/noise.
Classification: QPSK-like burst data (family-level)

Confidence: 70% (Moderate)

Supports: burst timing + clear 4-cluster structure post-correction.
Reduces confidence: missing center frequency/sample-rate metadata (inferred), interference elevation during key burst, unknown filtering/symbol rate, unknown AGC behavior.

4. Classification Summary Table

Signal/group Type/service (most likely)Time occupancy (approx)ConfidenceNotesBursty packetized emitter Industrial ISM telemetry/control, PSK-family Low-to-moderate (many short packets)70%One ~0.35 s event coincident with strong interference/noise rise Interference episode Conducted/radiated EMI or front-end compression~0.35 s65%Wideband median-power lift during long burst

5. Limitations & Assumptions

No SigMF metadata: absolute center frequency, gain, units, and true sample rate not confirmed.
Occupied bandwidth values are relative-to-capture and can be inflated by interference.
Industrial multipath/near-field coupling can distort constellation and bias classification.

6. Recommendations

Capture again with SigMF .sigmf-meta (center freq, sample rate, gain/AGC state, antenna, timestamp).
If near-field EMI is suspected: repeat capture with attenuator, band-pass filter (902–928 MHz), and/or notch known local interferers to prevent compression.
Collect a longer capture including multiple long-burst events; log plant equipment state (VFD speed changes, relay cycles).
If you need higher confidence: capture wider bandwidth (if allowed) to observe adjacent-channel structure and confirm if the wide OBW is real modulation or interference.

Under optimal conditions, classification can exceed 95%; real confidence depends on sample rate/bandwidth, front-end dynamic range, gain/AGC, clipping, noise figure, antenna & placement/polarization, channel conditions, clock error, frequency accuracy, filtering, compression, capture completeness, and ground truth.

Report 2 — FHSS-like GFSK Telemetry-Dropouts-Periodic Noise Rises-30s-Sigmf.sigmf-data

1. Executive Summary

Duration analyzed: ~30 s (inferred from filename; see Data Characteristics)
Signals/events: strong frequency-hopping-like activity across a set of discrete channels, with:
- Periodic wideband noise rises every ~4.0 s lasting ~0.205 s
- Two major dropouts (~0.84 s and ~0.64 s) where overall received power dips notably
Overall data quality: usable; metadata missing limits absolute frequency/power reporting.

2. Data Characteristics

Data type: A) Raw complex IQ (interleaved int16)
Samples: 7,200,000 complex samples
Sample rate (inferred): ~240,000 S/s (7,200,000 / 30 s)
Center frequency / units / gain: unknown (no .sigmf-meta)
DC offset: small (I ≈ +0.89 counts, Q ≈ −0.15 counts)
Hard ADC clipping: no rail hits detected; however, noise-floor surges suggest environmental EMI or front-end behavior changes.

3. Detected Signals/Events

Hopping behavior (relative to capture center)

Using short-time FFT peak-tracking (frame step ≈ 4.27 ms), the dominant energy peaks cluster into ~11 discrete frequency offsets (quantized to ~1 kHz resolution), consistent with FHSS-like channelization within the captured bandwidth.

Approx. channel offsets (kHz, relative): −90, −59, −39, −19, +1, +9, +21, +41, +61, +95, +103
Dwell time (frame-based):
- Median: ~4.27 ms (one analysis frame)
- Mean: ~18.8 ms
- Max observed: ~0.84 s (often associated with “dropout”/hold behavior)

Periodic noise rises

Pattern: every ~4.00 s
Duration: ~0.205 s per occurrence
Start times (approx): 4.25 s, 8.25 s, 12.25 s, 16.25 s, 20.25 s, 24.25 s, 28.25 s
Interpretation: periodic industrial EMI source (e.g., switching supply cycle, motor controller event, charger cycle) or periodic receiver/AGC behavior change.

Dropouts (overall received power dips)

Using a smoothed power envelope:

Major dropout 1: ~0.84 s around 9.78–10.62 s
Major dropout 2: ~0.64 s around 19.37–20.01 s
Additional smaller dips clustered near ~20.06–20.16 s (short, ~11 ms each)

Modulation family evidence (conservative)

The FHSS-like multi-channel pattern is strong. Within any single dwell, the capture bandwidth is narrow, making robust modulation-ID difficult without confirmed deviation/symbol rate and a wider per-hop view.
Classification: FHSS-like telemetry, FSK/GFSK-family plausible (family-level)

Confidence:

FHSS-like behavior: 85% (High) (discrete channel set + rapid hops + repeatability)
GFSK/FSK family: 60% (Moderate) (plausible for ISM telemetry, but not uniquely proven from this narrowband-per-hop view + missing metadata)

4. Classification Summary Table

Signal/group Type/service (most likely)Time occupancy (approx)Confidence Notes Hopper (11-channel set)Industrial ISM FHSS telemetry/control High (dominant throughout)85%Channel set visible as discrete baseband offsets Noise-rise events Periodic EMI / receiver state change~7 × 0.205 s ≈ 1.43 s total75%Highly periodic at ~4.0 s Dropouts Link fade/obstruction, interferer, or receiver gain event~1.48 s total (2 major)70%Two main dips; may correlate with plant activity

5. Limitations & Assumptions

No SigMF metadata: absolute channel frequencies in MHz cannot be stated (only offsets).
Peak-tracking will always return “a peak,” even in noise; however, the discrete clustering strongly suggests true hopping rather than random noise.
Modulation family (GFSK vs FSK vs filtered CPFSK variants) is not uniquely identifiable without per-hop demod/feature extraction and confirmed deviation/symbol timing.

6. Recommendations

Re-capture with SigMF .sigmf-meta (especially center freq, sample rate, gain/AGC on/off, antenna).
If possible, capture with wider instantaneous bandwidth so each hop channel is fully contained with margins; this enables deviation/symbol-rate estimation and stronger GFSK vs FSK discrimination.
Correlate the 4-second periodic noise rise with plant equipment logs (VFD ramps, relay test cycles, battery chargers, welders, compressors).
Add front-end filtering (902–928 MHz BPF) and consider attenuation to avoid compression near machinery.
If troubleshooting reliability: record simultaneous RSSI/SNR vs time and attempt directional sweeps (handheld Yagi / near-field probe) to separate emitter vs EMI source.