|
|
studio experiment - follow up to yesterday's post
|
I bandwidth limited, ms encoded and decoded Riders on the Storm by The
Doors today in order to replicate what I hear on DAB. Here is what I
|
No, as I said, DAB uses 'intensity stereo' joint stereo coding, not mid/side
joint stereo, so you won't be replicating DAB if you've used mid/side
coding. Mid/side doesn't destroy the phase information, whereas intensity
stereo does destroy it.
|
did and what I heard. As a pre-script...I just listened to "How Soft"
by Led Zep on Planet Rock and literally all you could hear was the
vocal. ONLY the vocal. Nothing else. It's rubbish!!!!
Stage one.
Bandwidth Limiting.
Bandwidth limited 13 versions of Riders On The Storm by The Doors into
sections of 750 Hz each: starting from 20Hz - 770, with 12 more
versions up to 9990. Mixed them all together. Result? Unpleasant
harmonic emphasis above 3k, bass stable but reverb broke down into
discreet time delays somewhere between 1 and 1.5 seconds. Imagine
playing the whole song through a delay of 1-1.5 secs, eq-ued with roll
off below 750 and high shelf boost above 4k of say 6 db. Most
unpleasant.
|
The upper limit of human hearing is 20 kHz, not 10 kHz.
See which of these you can hear:
|
Most PC speakers are not able to reproduce high frequencies, so you need
to test it on proper equipment.
|
Fair point.
|
I'll have to try it through my Hifi.
I couldn't hear the 15Khz on my new laptop despite quite easily hearing
the 14Khz. So perhaps it is just a limitation of my lap top speakers.
Richard E.
|
|
gr, hwh
|
The file names are self-explanatory.
|
I can't find these files at this location, but I've had a look at your
3 seconds of 4kHz files, and it is not dithered. It shows quantization
distortion very clearly. When you are generating your sine waves, you
need to add in a bit worth of noise to the pre-quantised signal.
|
This has got Gaussian noise with 1/2 LSB variance added to it, and the
spectrum shows it's done the trick:
|
If you're listening on a standard PC, the presence of dither or not in
these files is irrelevant. The files are all sampled at 44.1kHz
|
Looking at the spectrum without dither I'd agree with you because the
highest levels of distortion were still 116 dB (IIRC) down from the peak.
|
(suitable for burning onto a CD). Most PCs resample everything to 48kHz
internally, and make quite a poor job. Hence there will be bucket loads
of distortion added. That's even before it comes out of a dodgy
on-board sound chip and through terrible speakers etc.
|
I think the nature of the test demands that a decent PC and
non-resampling sound card are used. I use an Echo Mia, listening
|
I only generated them last week as a bit of fun because Bill Wright wanted
to see how high frequency he could hear these days, and it was just a
coincidence that the subject of the human hearing bandwidth came up so I
posted links to the files. It wasn't meant to be a serious test originally.
|
through a pair of Stax Lambda headphones. Also, when you are listening
to a frequency at which your hearing is severely attenuated, the
presence of quantization artifacts at much lower frequencies is indeed
very important. I can hear them with no problem at all by making a
file which has the first half dithered, and the second half
un-dithered.
|
btw, though MATLAB is entirely suitable if you use it correctly, Cool
Edit Pro works very well and is a little more use friendly.
|
I've got Cool Edit 2000, can I use that?
|
Here are some other pages examining high frequency hearing cut offs:
|
Right, thanks.
BTW David, have you received an email from that bloke yet?
|
Cheers,
David.
|
|
|
|
I then bounced stereo file and mono files of this arrangement.
Stage Two.
Ms Encoding
Within a new mix I imported the mono file. This was sent to bus one
with -6db of gain. Tracks two and three took the stereo versions and,
being panned hard left and right were then sent to busses two and three
respectively. Busses two and three were kept mono. Bus three was then
phase inverted. Busses two and three routed to bus four also with -6db
of gain. Therefore bus one became my mid channel and bus four my side.
Here's where it gets interesting. DAB radio seems to play the right
channel of the Side mix only. Weird huh? Hence keyboard solo on ROTS
goes missing, since it is largely panned to the left.
Stage 3.
MS Decoding
I subsequently set up a decoding matrix by creating a new arrangement
whereby the bounced versions of my mid and side were individually
imported to four tracks. Track one, mid: track two, side: both routed
to bus one with 6db extra gain. Track three, mid: track four, side -
phase inverted - and both routed to bus 2. Busses one and two panned
hard left and right gave me back my stereo image, with ROTS exhibiting
all the nasty aretfacts outlinined in paragraph one.
Stage 4.
Conclusion
So why does DAB seem to broadcast only the right channel of the Side
mix? And why does this only happen to tracks recorded before about
1980, especially to tracks laid down in the late sixties and seventies?
I know what's happening now, I can hear it. But why it should only
happen to certain tracks remains a mystery....
|
|