Zond 6 (1968): the beeps, decoded

n4ze3m · 17 October 2026

In November 1968 Jodrell Bank tracked Zond 6, the uncrewed Soviet test of a crewed flight around the Moon. Jodrell released the tape in 2018. The Russian voices on it were translated, but I couldn't find a published bit-level decoding of the beeps. This page decodes them.

In one line: every second the probe sends a short marker tone. On "data" seconds, eight bit cells follow: a 978 Hz pulse is a 1 and the 700 Hz base tone is a 0. Each word is sent four times. For more than a minute after the probe came out from behind the Moon, it sent the same word: 11010000.

Contents:

1. The tape
2. What the beeps are
3. The structure
4. The words
5. The voices: a timetable?
6. What is uncertain
7. How it was checked
8. Data and code
9. Credits and sources

1. The tape

Jodrell Bank tape #42 was digitised and published in 2018 with a transcript by Tim O'Brien and Kostya Novoselov: jb.man.ac.uk/multimedia/audio/zond6.html. The audio is licensed CC BY-NC-ND 4.0 and is not copied here. Please listen to it there.

Two stretches of the tape carry the probe's signal without voices:

• 0:00 – 1:03: about 01:51 UT on 14 November 1968, one to two hours before the Moon.
• 2:15 – 3:32: about 03:25 UT, just after the probe reappeared from behind the Moon (Lovell: "signals disappeared at 02:50 and reappeared again at 03:20").

These times come from Lovell's spoken notes; the tape is edited. The PDF transcript marks these stretches only as "Telemetry".

2. What the beeps are

Sven Grahn identified the tones (base ~710 Hz, bit ~1000 Hz, one-second sync ~1600 Hz, an extra ~2500 Hz pulse on seconds with data). He matched them to the command-verification link on 922.76 MHz, the same link later heard from Soyuz and Progress. Commands go up at 64 bit/s on 768.96 MHz, and the spacecraft sends back confirmation that it received them. The same 2.5 MHz subcarrier carries the relayed voices.

Measured on this tape:

All four tones are one ~4.9 kHz oscillator divided by 7, 5, 3 and 2, within 0.3%. Grahn's Soyuz/Progress tones (750, 1050 and 1750 Hz) are exactly 5250 Hz divided by 7, 5 and 3.

3. The structure

• Seconds. Marker onsets fall on a regular 1.042 s rhythm on this tape. If the onboard second was exactly 1.000 s, the tape plays about 4% slow. That is an estimate, not a measurement.
• Markers. A 1632 Hz marker opens an idle second and a 2451 Hz marker opens a data second. The marker lasts about half a cell (measured 72 ms; the envelope filter adds roughly 10 ms).
• Bit cells. The second divides into 8.5 cells of 122.5 ms. The marker takes the first half cell, then come 8 bit cells. In data seconds the pulses start on this grid, the first cell about 64 ms after the marker onset. Pulses are about 70–100 ms long. A run of 1s can sound as one long pulse. A few pulses start on a half-cell boundary instead (see section 6).
• Repetition. Each word is sent in four consecutive seconds, followed by idle seconds. A new word comes every 10 seconds before the Moon (10.42 s on the tape) and every 9 after it (9.37 s).
• Voice mode. Just before the relayed Russian voice starts at 3:32, the idle tone is held continuously for about 2.5 s.

Each row is one second. Blue = data marker, green = idle marker, red = 978 Hz pulses, grey = base tone.

4. The words

Bit b0 is the first cell after the marker. Words in red are uncertain (see section 6).

Read with b0 as the least significant bit, 11010000 is 11, 11111000 is 31 and 00011000 is 24. Read the other way they are 208, 248 and 24. The bit order is not known.

5. The voices: a timetable?

On 17 November the probe relayed numbered messages ("615 / 704 / 01", then 705, then 706). Each repeats the same items with new clock times. The item names are Russian spelling-alphabet letters: Борис 25 = Б25, Владимир 26 = В26, Женя 24 = Ж24, Дмитрий 26 = Д26. Times below are as transcribed by Jodrell Bank:

Pattern: the three named windows seem to run end to start between consecutive items: knock from В25 to В26, pine from В26 to Ж24, willow from Ж24 to Д26. As transcribed, 9 of the 12 window boundaries match an item time exactly. A tenth matches if "24:07", which is not a clock time, is read as 14:07. Two do not match: willow should end at 12:52, and knock should start at 13:39. The gaps between items are the same in both messages (+1, +12, +16, +10 minutes) while the whole block moves 85 minutes later. Coverage in 2018 called these "simulated instrument readings". They look more like a timetable being re-issued. What was being timed is unknown.

6. What is uncertain

• Meaning. If Grahn is right, the words are command receipts, but which commands they confirm is unknown. With b0 as the least significant bit, the post-Moon word is 11. On 13 November the relayed voice gives a "pre-sanction for number eleven". That may be a coincidence.
• Two early words. In 11000000 and 00001000 (0:08–0:19) one pulse starts at 491–493 ms. That is exactly a half-cell boundary, between b3 and b4, not the start of a cell. Reading each cell as one bit may be the wrong model for those seconds, so their words may be wrong.
• Bits 5–7. They are set only once in a clean word (b6 at 2:18.8), possibly noise.
• Tape speed and absolute timing are estimates.

7. How it was checked

• Tone tracks. Band-pass envelopes at the four tones, 2 ms resolution, each compared with its own running noise floor.
• Data seconds. Only 2451 Hz markers that last at least 60 ms and fall on the one-second rhythm count; 49 qualify. Pulse edges use hysteresis and a 40 ms minimum length; 123 pulses were found.
• Grid fit. A free fit gives 123 ms cells starting 63 ms after the marker. With the cell fixed at 1/8.5 of the second (122.5 ms), the median distance of a pulse start from the grid is 1.2 ms, and 87% are within 10 ms. On the half-cell grid (61.3 ms), 95% are within 10 ms.
• Chance test. Random pulse times, same count and window, fit as well in 0 of 200 trials; their typical median distance was 20 ms.
• Decoder agreement. Reading a cell as 1 when the 978 Hz tone fills more than 35%, 50% or 65% of its first 80 ms gives the same word in 48 of 49 seconds. A second method compares the 978 Hz and 700 Hz power with a margin of 0.5 (log10 units) and agrees in 43 of 49. Without the margin it fails, because the 700 Hz tone is often weak in 0 cells.
• Tones. One FFT over the voice-free stretches with peak interpolation.

8. Data and code

Every number on this page comes from one script run on the MP3 from Jodrell Bank. The code, data and figures are also on GitHub: github.com/n4ze3m/zond6.

• zond6.py: the decoder and all checks (Python 3, numpy, scipy, ffmpeg). Run python zond6.py zond6.mp3 out/.
• words.csv: every data second with its word and how full each cell is.
• pulses.csv: every pulse, with its start and end after the marker, its cell, and its distance from the grid.
• summary.json: tone frequencies, the second, the grid fit, the chance test and decoder agreement.
• README.txt: what each column means. make_figures.py draws the figures.

9. Credits and sources

• Audio and transcript: Jodrell Bank Centre for Astrophysics, University of Manchester (CC BY-NC-ND 4.0).
• Sven Grahn, Radio systems of the Zond 4–8 series of spacecraft, citing E. P. Molotov, Terrestrial radio systems of space vehicles (2004).
• University of Manchester release, November 2018.

17 October 2026: the timing values were corrected by 0.2% after a sampling-step error in my first analysis. The decoded words did not change.

Corrections are welcome. GitHub · n4ze3m.com · @n4ze3m