CTF-Writeups

Lines of Contact - CTF Challenge

Category: Signal Processing / Misc
Difficulty: Hard

đŸ“„ Download Challenge

Download Lines of Contact.rar - Try to solve it yourself before reading the writeup!

Challenge Summary

You’re aboard a deep-space relay station when an incoming capture hits your buffer: a mono audio recording labeled “0xfun”. The analysts insist it’s alien. But the encoding style feels familiar — like something humanity would send when it wanted to be understood without sharing a language. Dig into the signal, recover what it’s really carrying, and extract the flag.

This is an analog “pictures in audio” challenge (Golden Record–style raster), but you’re not told that up front.

1. First Triage: “Is This Really Just Noise?”

Open ctf_record.wav in Audacity (or any waveform viewer) and zoom in.

There are repeating deep negative spikes at regular-ish intervals. That pattern is your first clue: real noise doesn’t produce clean periodic “events” like that.

Those spikes look like a line sync pulse: something that marks “start of a line.”

2. Hypothesis: It’s a Raster Stream (Audio → Image)

If there’s a repeating sync, the classic structure is:

[sync pulse][signal payload][sync pulse][signal payload]...

If you stack payloads line-by-line, you get an image. This is the “aha”: treat the audio as scanlines.

3. Extract Line Starts (Sync Detection)

Take the audio samples and look at their distribution. Sync pulses are the most negative plateau in the whole file.

A robust approach:

  1. Let m = min(audio)
  2. Choose a threshold like thr = 0.7 * m
  3. Mark runs where audio < thr
  4. Filter by run length: compute run lengths of consecutive samples below threshold. Keep the long ones (e.g., top 10–20% or > median of the longest cluster).

You get a list of sync start indices: starts[].

4. Estimate Line Spacing (Samples Per Line)

Compute:

diffs = starts[i+1] - starts[i]

During picture regions, diffs will cluster tightly around a nominal value (small drift is normal).

Use: nominal_line = median(diffs) (after ignoring huge gaps)

5. Find Image Boundaries

There will be big gaps in diffs between picture segments (preamble / marker / separator).

Find indices where diffs > 3 * nominal_line and split starts[] into chunks (segments). Each chunk with lots of lines is a “picture.”

6. Decode Each Line into Pixels

For each line:

  1. Take samples from start + sync_len to next_start as the “video payload”
  2. Infer sync_len from the tight cluster of long run lengths: sync_len = median(long_run_lengths)
  3. Slice: payload = audio[start + sync_len : next_start]
  4. Resample each payload to a fixed width = median(payload_length) to handle slight time drift

7. Stack Lines into an Image

Convert each resampled payload into a row. Stack rows in order → 2D array. Normalize brightness robustly using percentiles (1%–99%). Save as PNG.

One segment should look like a calibration image (circle or obvious shape). The next contains the flag.

8. Read the Flag

Once the flag image appears, read it and submit.

← Back to Blog