On Tue, 22 May 2007 16:12:32 +0100, Jim Lesurf
wrote:

In article , Don Pearce
wrote:
On Tue, 22 May 2007 09:19:09 +0100, "Serge Auckland"
wrote:

Indeed, and in my previous post of the criteria, it was stated that THD
should be measured at all frequencies 20-20k and refers to all powers
and all loads for which the amplifier was designed. In practice, the
measurements are actually THD+N as this is what distortion meters
actually measure. Of course the use of a harmonic analyser for
distortion measurement won't pick up the +N component, but as a
practicing engineer, I found the use of such an instrument to be
tedious in the extreme, and unnecessary when an overall THD+N figure
was so easily achieved.

S.


The problem becomes more complex when you use an FFT analyser, as I
suspect most are these days. You then need to consider the number of
points in the FFT, and the way they display noise. Discrete signals are
easy - whatever you do with the FFT, they look the same size, but the
"+noise" bit will change with the number of points.

Erm. It should be the total noise in the audio range. This means that
however many bins it was divided into becomes irrelevant as they are then
summed. Although I'd agree that a small fraction of the noise will be in
the input signal bin and would be 'lost'.

In recent years I've tended to use a Stanford Instruments unit that
combines a test waveform generator and an FFT specan, and 'automates' the
process as you wish. The trick, of course, is to know what process to
specifiy and to understand how to interpret the results - especially when
the spectrum on the screen isn't simple. :-)

The noise floor problem is more significant when reviews simply display the
floor value in terms of the per-bin level without having any clue what
resolution bandwidth they are using. In those cases your comment does
indeed apply, and makes the floors shown in some magazines worthless.
Having tried discuss this with one or two people I fear that this issue
whooshes over the head of some of them. Although there are others who
clearly understand it, but don't use such meaningless plots.


Exactly - although the maths is very easy - just add 10 log (audio
bandwidth / (bin bandwidth * windowing ratio)) to the noise level in
dB. But as you say, this appears to be beyond most people.

The problem is that you must do this to the noise, but not to the
discrete signals, and it can get tricky sometimes separating the one
from the other.

Are there many distortion analysers any more that simply null the
fundamental and display the sum of the rest?

Dunno. The last one I used a lot was the Sound Technology 1000A about two
decades ago. This was very nice, but took a few seconds to settle into a
null, etc, whenever you altered anything. Worked down to about 0.002%
though, IIRC. I think that part of the delay was for the light bulb in the
oscillator to settle when you changed frequency. ;-

Slainte,

Jim

I still have a couple of those tiny bead thermistors in vacuum tubes
that are really good at stabilizing Wien Bridge oscillators. Better
than light bulbs, I think.

d

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