Longwave radio signals during a leap second

At the end of 31 December 2005 (UTC), a leap second was inserted into the UTC time scale: between 23:59:59 and 00:00:00, there was a 23:59:60. In the longwave radio spectrum, several radio stations broadcast precision time signals. I used my software defined radio system to record these stations during the leap second; the results are presented on this page. As it turns out, some radio stations apparently didn't get it right...

MSF, HBG and DCF77

In the below waterfall diagram (i.e., a diagram with frequency on the horizontal axis and time on the vertical axis, and colour indicating signal strength at each time and frequency), three time signal stations can be seen:

Each of them interrupts its carrier in a 1-second rithm to provide a time reference; variations in the duration and/or number of interruptions are used to transfer data and indicate the beginning of a new minute. The annotations in the picture should be self-explanatory.

[waterfall showing MSF, HBG and DCF]

Perhaps the most notable thing is that HBG apparently somehow missed the leap second: at the leap second, it transmitted a triple interruption, which is its code for beginning of the first minute of an hour (see here). Unfortunately, their signal faded away into the noise during the next few minutes; by the time it was readable again, about 10 minutes later, their start-of-the-minute mark was aligned with DCF's, so apparently they had inserted the leapsecond by then.
This apparently incorrect transmission by HBG has now (January 7th) been confirmed independently, see http://phk.freebsd.dk/Leap/20051231_HBG/. Also, HBG staff have confirmed the error.

MSF also seems to have done something "wrong". During the first 16 seconds of each minute, it broadcasts the offset between UTC and UT1 (which is related to the earth rotation). As indicated in the above diagram, the offset was -600 ms before the leap second. Looking at the data just after the leap second, we see this:

[MSF just after the leapsecond]

Decoding this gives an offset of +400 ms, while it should have been +300 ms according to the IERS bulletin.
Checking now (2nd January, around 20:00 UTC), the offset is still sent as +400 ms.
(Update May 2006: a reader of this page has told me that the offset could not be changed in time due to the IERS bulletin being issued during the Christmas break; changing the DUT offset broadcast by MSF is apparently done manually at the transmitter site.)

Some further observations:

Audio time signals on Deutschlandfunk and France Inter

Some audio broadcast radio station still transmit "time pips" at the top of the hour: a few short 1 kHz tones, at 1 second intervals, the last of which coincides with the start of the hour. In the recording two such stations are present: Deutschlandfunk on 153 kHz, and France Inter on 162 kHz. These stations use amplitude modulation, which shows up in the waterfall display as a strong continuous carrier, and symmetrical audio sidebands around that carrier.

Here's an annotated waterfall display of these two broadcasters, with DCF77 at the left, around the leap second:

[longwave AM stations]

As can be seen, the start of the last pip of France Inter nicely coincides with the start of the new hour, taking proper care of the leap second.

However, Deutschlandfunk's timing is weird: their last pip is way to early, it already starts before the leap second has started!
Listening today (2nd January, at 20:00 UTC), Deutschlandfunk's time pips are nicely synchronized to DCF77, with the start of the long pip at the start of the hour.

Test setup

As noted above, I made these recordings with my home-built software defined radio system. This system was configured to sample in-phase and quadrature signals at 195.3125 ksamples/seconds (at about 20 bits of effective resolution), around a centerfrequency of 107 kHz, thus recording everything between 10 and 204 kHz. I recorded from about 4 minutes before until about 11 minutes after the leap second. The antenna was a simple loop of wire indoors in my living room, which no doubt is far from optimal. It was connected to the radio hardware through a fifth-order low-pass filter.
(I still have the raw recorded data available, in case more analysis is to be done.)

Comments are welcome, at