MP3 w/ high freq?

I recieved a 256kbps mp3 from an ad agency last week.  What makes it interesting is the fact that the high frequency information was completely intact up to 20kHz...From there, there was a sharp slope.  When I make mp3s, they tend to cut off around 16-17kHz.  Any idea what they may have been using to encode?

I have to assume that to keep the highs intact, it must have been masking the entire file more, but I couldn't really hear anything that sounded worse than other encoders I use...In fact, I thought this file sounded somewhat better.  

Any ideas?

Scroll down to the 256kbit/s graph here:  (almost to the bottom)

http://www.fliptech.net/bitrate.shtml

Does that explain what you are hearing / seeing?

Re. 256kbit/s:

This one is nearly identical to the original right through the 17 KHz range. And even up to that point, it is very faithful to the original--not dropping off until the 21 KHz range ...

Similar, but not quite.  Obviously, I didn't have the original wav file to compare to, but the mp3 almost had the look of a typical -6dB curve.  I've tried both LAME and Audition, but I'm unable to duplicate it, even at 320kbps.  I keep Auditions spectral analyzer open 100% of the time, and deal primarily with mp3 or mp2 files.  I've never seen and mp3 behave this way...It looked exactly like an uncompressed file up to 20kHz.  I didn't think to check file size.

Is it possible this is a custom bitrate setting for mp3 PRO?  That's something I haven't tested yet...

Encspot will provide a lot of info about both mp3 and mp2 files.

I was thinking perhaps a -q setting of 0 (which I always use for archivals — Lame 3.90.3) might preserve audio to 20kHz? Not sure.

With a steep slope at 20Khz, that's more like Dolby Digital and less like an mp3.  Perhaps this is a result of filtering or a/d conversion external to the mp3 encoder?

Did some checking with Lame: made a wav file with  15 seconds, 20kHz sine only, 16-bit, normalised to -1dB; encoded the wav with Lame in the following configurations:

1) 256k, q=0;
2) 256k, q=0, -k (keep all frequencies, or turn off the polyphase filter entirely);
3) 320k, q=0;
4) 320k, q=0, -k

results:
1) no signal (20kHz gone)
2) signal present (but with amplitide "bumps" at the beginning and end)
3) no signal (as above)
4) signal present (per 2) above)

so its the -k switch that preserves frequencies above the polyphase by simply turning it off, while the -q switch controls the frequency and turnover of the filter, with Lame anyway; others encoders will be different.

Best bet is to use Encspot to see what the encoder actually was, and go from there.