s/n measures

[Paul Schreiber]

1) First thing, everything you have said is correct.
2) The problem is with making low level measurements of say "almost-to-CD quality". A 14 bit (cheesy $59 CD) system has a 78dB SNR, or almost 1 part per million. Thats 1uV (microvolt) of noise on a 1 volt signal (RMS). So now we are back to the "if I can't hear it, it doesn't exist" arguments of rec.audio.high-end. I don't know of many meters that can resolve 1 ppm that cost <$1500. (usualy their voltage references are 5-10 ppm).
3) There was a HP Distortion meter, basically a notching vector voltmeter, that was in use for about 14 years. It was  absolutely sufficient for measuring any audio equipment pre CD. It is quite accurate to THDs of say down to 0.01%. I had like 3 of them. But post-CD, the AP is about it. The Model I can do 0.0005% all day.
4) The problem inherent in listening is the noise in the power amp.
5) Of course, one can argue "Who the heck needs to know if my OBXa is 0.001% or not!" That's for sure! But the reader did ask about 100dB stuff, in which only the AP can tread.

lastly, a little (true) story:

I am redesigning the DKI Synergy synth (Wendy Carlos' favorite). They use 24 bit math, a (at that time) $250 16bit DAC, and 5534 output amps. But guess what drives the 1/4" outputs? A LM348 quad 741! Why in the world?

Well, it turns out that there is so much digital noise in the system that DKI "dummyed down" the audio so the average user can't hear all the noise! If you AP BEFORE the LM348, the SNR is about 40dB. The low GB and slew rate of the LM348 is filtering out alot of the high frequency stuff. GAG!!

Hope this helps.

Paul Schreiber
Synthesis Technology

----------
From: 	brad sanders[SMTP:brad.sanders at circellar.com]
Sent: 	Saturday, March 22, 1997 1:17 AM
To: 	synth-diy
Subject: 	s/n measures


Paul, that's fine if you NEED an FFT - but c'mon... S/N measures have
been made accurately for decades now - without FFTs!

To check noise output of a circuit, what you need is a good mV meter
(or amp) and a rectifier/averager circuit. If you wanna get really
fancy, an A/C-weighting filter would be nice.

If you can build a synth, you can build all these. Use a very low
noise opamp (even, say, a 5534) in a 40db gain config, and connect to
your VOM. Measure the voltage with a shorted input, then connect your
DUT and note the voltage again. Remembering to correct for
A^2+B^2=C^2, you can determine the noise voltage output of your
circuit. For best results be careful with grounds, and stick the whole
circuit in a shielded metal box with a battery.

You can't determine the *spectral* density of the noise like this, but
you can determine the overall noise output. A pair of headphones
attached to the output of your noise amp might allow you to make a
good estimation of the (audible) noise spectrum.

Hey, Paul - what about this? Build a good noise amp, tweak a PC
soundcard (I've done it; you CAN improve these things) and use
commonly available PC software to build a low-end Model1? You can even
use the same soundcard to generate test signals....

>If I'm trying to match the performance (or at least get close to) of a
>CEM3330 VCA chip (100 db S/N), how can I measure just what the noise floor
>is???

>There really are only about 3 instruments made capable of accurately =
>making these measurements. The most widely used (by far) is an Audio =
>Precision Model 1. It is a true 18bit (100dB) device. This generates all =
>of the SNR plots you see in magazines like Stereophile. There is a new =
>Model 2 out that is 22bit (128db). The downside is that these are $10K =
>boxes. I'm trying to find a used Model 1 for $4500.