On Wed, 18 Oct 2006 21:10:55 +0100, "Keith G"
wrote:


"Rob" wrote in message
...
Don Pearce wrote:
On Wed, 18 Oct 2006 20:34:24 +0100, Rob
wrote:

Don Pearce wrote:
Apropos of some stuff we were doing a while ago, here is the pink
noise track from th HFN-RR test disk played on my system, which is:

Systemdek IV
SME 3009ii arm
AT-OC9 microline cartridge
V.nice indeed.

http://81.174.169.10/odds/dspkr/atoc9.gif

It shows an essentially flat response from about 12Hz, with a slight
lift at 10kHz, then returning to the proper level above that. The
recording is made at 96kHz sampling, and it is clear that the test
record cuts off sharply at 20kHz. The cartridge claims to go to 60kHz
(ahem!)

I've offset the levels of the left and right tracks slightly so they
are more visible.

d
I'm quite used to being alone on this, but could you explain a couple of
things please:

How is time represented on the graph?

Time isn't represented on the graph. The whole thing is about 20
seconds of pink noise, converted by FFT to a dB vs frequency graph.

I look at the graph and I can't see the things you describe:

I'd have expected the db (loudness?!) to remain constant for a flat
response (that is, a straight horizontal line)?


Pink noise isn't flat for an FFT. It drops as you go up by 3dB per
octave, or 10dB per decade. If you put a ruler along the average of
the graph, you can see the deviations above and below the straight
line.

If you have another look, I've drawn in a line that represents what a
flat response should be.

I wouldn't have commented on the 10k blip, but consider the 10-20k range
to roll off?

The line rises above the mean diagonal before it drops again - in fact
it drops back to just below the wanted line at 20kHz.

I've often seen these things in magazines and frankly they've always
been a mystery!

And, clearly not ashamed of public disclosure here, I tried to 'tune' my
system adopting (I think) the same twisted mentality associated with my
interpretation of your graph, and it sounded totally bizarre - bass just
swamped the entire thing.

Thanks, Rob


Much easier when you know what is happening. To be technical, with
pink noise you have equal power per relative bandwidth - so as you go
up in frequency the power gets more spread out, and there is less of
it in each finite frequency bin that the FFT produces.


Many thanks - appreciated.



Same here, but not fully *understood* yet!! :-)





OK, let me explain a bit more about pink noise. Suppose we are at a
frequency of 1kHz and the noise power contained within 1Hz of
bandwidth is a milliwatt; if we now look at 2kHz, that milliwatt of
noise is now spread over 2Hz (same percentage bandwidth), so there is
only half a milliwatt per 1Hz, likewise up at 4kHz, the milliwatt is
spread over 4Hz, so there is a quarter of a milliwatt per Hz. So the
result is that for each doubling of frequency the power per Hz drops
by a half, or 3dB. If you go up ten times in frequency, the power
drops to 1/10th, or by 10dB - hence the description of pink noise as
-3dB per octave, or -10dB per decade.

Why use pink noise instead of white noise, which would have looked
flat? If they had tried that on vinyl the high frequencies would have
melted the cutter head. They have to cool them with liquid helium as
it is.

d

--
Pearce Consulting
http://www.pearce.uk.com

--
Posted via a free Usenet account from http://www.teranews.com