Writeup · Cyberweek Africa 2026 - National CTF (Finals)

The Lost Signal 2: Voices Through the Air

stegmediumsolvedFlag: FLAG{REDACTED}
  1. Use calibration.wav (known plaintext A6 3C) to calibrate the FSK tones, bit ordering, and time slots.

Points: 250 (MEDIUM)
R2 (retained for Chapter 3): 60AAA70740D0

A fictional Coastal Research Relay (CR-MVT) monitor recording contains several FSK-encoded bursts interleaved with a noisy adjacent-channel programme, a corrupted dropout, and an independent replay. The genuine archive frames must be separated from decoys, their payloads reversed with the Chapter 1 relay token R1, and the message reconstructed.

All traffic, stations, and the protocol are synthetic; the historical citations (1953 African Broadcasting Service, 1954 Kenya Broadcasting Services) are genuine context only.


Overview

The task is to:

  1. Use calibration.wav (known plaintext A6 3C) to calibrate the FSK tones, bit ordering, and time slots.
  2. Demodulate coastal_monitor_stereo.wav, using the LEFT channel only (Monitor A); the RIGHT channel is a noisy adjacent-channel programme and must not be averaged in.
  3. Reconstruct valid CR-MVT frames (origin = 0x53) for sequences 0–4, rejecting the BK-TRN test loop (origin = 0x54), the corrupted dropout, and the replay.
  4. Reverse the documented payload transform with R1 (8EC5C9EBF7C3, recovered in Chapter 1).
  5. Compute the SHA-256 proof and submit FLAG{LS02_<R2>}.

Recon — calibrating the audio alphabet

From operator_bulletin.txt and calibration.wav:

  • Sample rate: 8 kHz, 16-bit PCM.
  • Each burst opens with a 1900 Hz marker tone, then a silence gap.
  • Binary data is carried by two octave-related lower tones.
  • Calibration contains exactly two octets: A6 3C.

Demodulating calibration.wav (32-sample / 4 ms windows, Goertzel tone power):

Tone Meaning Measured
~670 Hz bit 0 matches calibration
~1340 Hz bit 1 octave of the low tone
1900 Hz burst marker —

Bit slot ≈ 448 samples (56 ms), with the tone occupying the first ~360 samples. Bits are sent LSB first within each octet. Verified against calibration: the 16 data slots decode to A6 3C exactly, confirming the tone map, bit order, and slot timing.


Exploitation — demodulation and frame selection

Seven marker bursts were found (matching broadcast_log.csv):

Burst Start (s) Origin Seq Length Checksum Verdict
B01 0.70 0x53 2 8 ok genuine
B02 6.97 0x53 0 8 ok genuine
B03 13.24 0x54 1 8 ok decoy — BK-TRN test loop
B04 19.51 0x53 4 8 ok genuine
B05 25.78 0x53 1 8 ok genuine
B06 32.05 0x53 3 8 BAD corrupted dropout
B07 38.32 0x53 3 8 ok genuine — replay of B06

Check byte (per bulletin §C): S = 0x6D; for each frame octet B excluding the check byte, S = ROTL8(S,1) XOR B; expected check = S XOR 0xA7. B06 fails this, consistent with the logged “short signal dropout”; B07 is the independent replay and passes.

Frame format: Origin ID | Sequence | Payload length | Payload (8) | Check byte. The genuine archive set is origin 0x53 (CR-MVT) with sequences 0–4 each exactly once:

Seq Burst Ciphertext payload
0 B02 77 94 A7 09 03 01 A9 D0
1 B05 87 A6 1C 03 66 92 3B C3
2 B01 97 EC 35 74 81 33 C5 B2
3 B07 FA 34 73 87 34 21 81 9C
4 B04 CD 49 F8 0A 36 91 97 EE

BK-TRN (B03, origin 0x54) passes checksum but is excluded on origin — a valid checksum is not a signature, per the bulletin.


Payload recovery — ledger key

R1 from Chapter 1: 8EC5C9EBF7C3 → six octets K = [0x8E, 0xC5, 0xC9, 0xEB, 0xF7, 0xC3].

For payload octet i in sequence s:

M = ROTL8(K[(s+i) mod 6], 1) XOR ((0x39 + 17*s + 29*i) mod 256)
P[i] = C[i] XOR M

Concatenating sequences 0–4 (no separators) yields readable ASCII:

SIGNALS_FROM_THE_COAST_CARRY_OLD_TRUTHS!

A correct R1 is confirmed by the fact that the recovered bytes are fully printable ASCII rather than garbage.


Flag computation

Final proof (bulletin §E):

LS02|8EC5C9EBF7C3|SIGNALS_FROM_THE_COAST_CARRY_OLD_TRUTHS!|CR-MVT

SHA-256 (uppercase): 60AAA70740D0B703845AA316610897C8197B0ED0119339D50D391255512CFEDE
R2 (first 12 hex chars): 60AAA70740D0

FLAG{REDACTED}

Lessons / Takeaways

  • Demodulate from the correct channel only: Monitor A is LEFT; averaging with the adjacent-channel RIGHT programme corrupts the tones.
  • Calibrate tone/frequency, bit order, and slot timing from a known plaintext sample before trusting any demodulated output.
  • A valid checksum never proves authenticity: corroborate origin (0x53 CR-MVT vs 0x54 BK-TRN), recorder observations, and expected sequence coverage.
  • Corrupted and replayed bursts look structurally identical to good ones; select frames by origin + sequence set, not checksum alone.

Techniques

  • Audio Steganography
  • FSK Demodulation
  • Checksum Validation
  • Protocol Reconstruction