OT: S/N-Ratio measurment - help

[Tim Ressel]

I am not at all surprised to hear it. Think about the setup: here you have an
A/D converter with the bottom bit at around 25 uV. Then you put into a PC case,
one that is metal and sealed up to keep those nasty emissions from exiting the
case. So you have this sound card sticking up in the noise breeze like a
high-tech sail. And oh by the way, no shielding what so ever on the card, except
for whatever carefull layout they may have done. Noisy? Of course it is.

Fortunately, steps can be taken. Extensive shielding if the card should help
quite a bit. For example, a cheap shield can be made by sandwiching aluminum
foil between two layers of shelf paper. Carefully wrap the shield around the
card, and then ground the shield real well. Make sure nothing gets shorted by
the shield. 

I can't guarantee this will help, but its definitely worth a try.


Tim Ressel--EMC Compliance Engineer
Hewlett-Packard
Verifone Division
916-630-2541  
tim_r1 at verifone.com                     



-----Original Message-----
From: Martin Czech [mailto:martin.czech at intermetall.de]
Sent: Monday, February 07, 2000 3:47 AM
To: x_ray at hanfi.com
Cc: synth-diy at mailhost.bpa.nl
Subject: Re: OT: S/N-Ratio measurment - help



:::I've got a somewhat OT-question: Does someone know, how I could measure
:::the Signal/Noise ratio of an amplifier, without the need of expensive
:::audio-analyzing meters?
:::It would be really great, if somebody could help me.
:::It could also be a schematic for a circuit or something like this. I'm
:::an advanced electronic hobbyist, so this shouldn't be a problem.

I think this is absolutely not OT!
Because without any figures one can not decide which design offers
advantage to others if there is no direct A/B comparison possibility.
In the past we had pointless discussions like: My circuit is
better, no mine!

I assume it is all about hissing white semiconductor noise.
(1/f noise is also there, perhaps an analog dc voltmeter can be used
to observe this, or a slow scope trace, like in many data sheets.
1/f is interesting for control voltages/currents, e.g. for pitch).
No hum or other artefacts like RF, put everything in a metall box, we
are talking about low level signals...  I assume the circuit is linear
and time independent (no compander etc.).

I'd say the task needs two measurements:
a) the maximum allowable input level for a sine (different frequencys)
b) determination of the circuits own noise.

I would propose:

1. Rather cheap estimation. First you will need to determine the
maximum allowable level for a sine input. A Wien-bridge sine with light
bulb stabilisation is cheap and gives reasonable distortion. National
semiconductor have the schematics in their appl. notes (they say Wine
bridge, I think that is wrong, must be Wien). I think that a listening
test can give an idea about the maximum level, because distortion will
increase very rapidly (at least in most semiconductor circuits). Here
is error #1, since we got no real idea of distortion figures.


You could also compute a sine wave (cooledit) and record that digitally
on CD. (i.e. several tracks with several frequencys etc.)
This is a sine wave you can hardly beat for that money.
Comes nice with aportable CD player, because of less hum problems.

Next you need a 100x amplifier or so, say 2 NE5534, to raise the noise
floor of the circuit when input is loaded with the average generator
resistance.  You could compare the noise level with some known source
(adjustable, see the random shift register thread) with a A/B switch or
look at it on the scope. There is a rough formula that relates the scope
view to noise power, I have forgotten the figures yet.

A CD with white noise with predetermined level per track would
be also very good to have here.

Thus you can get a signal to noise ratio estimation.


It is nonsense just to listen to the circuits noise and tell:
"Huu this is noisy". Because you allways need the ratio between
signal and noise, as your question implies.



2. Not so cheap, but better:
If you have a reasonable sound card 

[this is the next problem, 90% of soundcards are trash, just got a mail
from a friend with a "very good rated" soundcard, Terratek, in the 400
Euro range, 18BIT RESOLUTION.  I asked him to record silence (i.e. a plug
with 1k in the input), and record a silent connection to his mixing desk.
You allready can guess what the result was: the silence measurement shows
larke peaks for different frequencys, about 12kHz, and a predominant
590Hz, I have no clue what that could be, but it must come from the
power supply. The connection to the mixing desk gives additional (2N+1)
* 50Hz hum (N=0,1,2,...). The resulting resolution is 12bit, in words:
twelve bits!) at most, I assumed 0dB full scale level, which might be
not true, because of distortion....This is the kind of ADC hardware that
ruined the name of digital audio up to the present days...]


or digital recording possibility, you're much better of.
You can determine the maximum sine level by looking at the outputs
DFT, you can e.g. raise the amplitude every second by 1dB,
record that and hop your DFT forward until your criteria 
are met.

I've just written a command line interface tool that
does this DFT and power estimation stuff with any .wav file.

Switching in your 100x amplifier you can also DFT the noise,
this will enable you to control, wether it is white, or 
hf, hum or whatever.

I think this way you can get a good figure.
A reasonable ADC hardware will give results that need not be inferior
to 40000 Euro measurement systems, they do it automatically
waht we have to to manually.



m.c.