Edmond Stuart said:
Hi Jan,
Yes, I did. Not in the traditional way, but implemented in software rather than hardware
Cheers, Edmond.
Details? There are some vintage lock-in amps available that can extract 100nV harmonics from noise that's 1000x higher. 100nV is 140dB below 1V if I'm not mistaken. That's full scale, actual performance will be better by another 10dB.
There are newer, OEM-type single board analog or digital lock-in amps with PC control with slightly less perfoamne (1uV full scale).
Jan Didden
janneman said:
Yes this should work. I forgot that I published some results in an AES preprint (don't remember the number) using a filtered Soundtech generator. HP used to make a low freqency all analog spectrum analyser with a 10Hz IF BW and 7nV/rtHz noise, you could resolve better than -110dB below a 10mV signal. You could see thirds (.1 ppm) from waving a pair of cutters near my filter inductor due to the hysterysis in the steel.
The PC environment, when you really get down to the noise floor, is a mess. I still contend that 30yr. old analog technology easily betters even the best soundcard. Let alone the versatility and general usefulness for other things. This stuff, after all, is available on the used market sometimes for a couple of hundred bucks.
janneman said:
Hi Jan,
Details? Well, some. I'm using a 24bits 192kHz sound card (Waveterminal 192X + mods) and my own software (at the moment not available to others). In addition to THD-N (yes, minus N) I can also measure the frequency response, even when the volume pot of my amp is turned off.
Cheers, Edmond.
janneman said:
I used a Princeton Applied Research 128A lock-in amplifier 20 years ago, for low frequency (<100KHz) spectral analysis. Worked amazingly well at nV levels, however operating this thing was a very difficult manual process. This equipment has no computer data interface, so data was collected on paper and processed offline.
You can get today an 128A on EBay. They are quite expensive, though.
I don't believe anything better than a good soundcard is really necessary to see the minimally audible V in the output of a power amp
Consider a moderate sensitivity home loudspeaker might give ~ 80 dB SPL at the listening position with 1 W input
1 W into 8 Ohms nominal speaker is 2.8 Vrms
with structured tones spread over multiple critical bands it appears the ear can perceive their linear sum despite each component being below the single tone threshold of hearing in each respective critical band
This gives the occasionally cited "hearing below the noise floor" claim but this is still a limited phenomena - frequency masking by nearby tones/noise still occurs, masking by system noise in the same critical band is still effective, the perceptual noise floor still rises at both frequency extremes
the perceptual noise floor lowest level dips a few dB below the 0 dB spl ref for only a few critical band widths in the low KHz
So if we take the 20 dB number I've seen people bandy about as a estimate of the best sub threshold enhancement possible then we are looking for a collection of single tones, each at worst ~ -20 dB spl
80 dB - (-20 dB) gives -100 dB re 2.8 Vrms
-100 dB re 2.8 Vrms = 28 uVrms, is within my Juli@ s/n with < 1 s averaging, in fact harmonic distortion components should appear as single lines which could readily be seen above the spot noise floor, –120 dB re 2 Vrms card reference V should be easy to pick out with –130 dB spot noise floor
So for the low level harmonic distortion measurements Geddes is proposing here there really is little need for any hardware enhancement to measure the expected levels corresponding to the most extreme estimate of human auditory thresholds I know of
Consider a moderate sensitivity home loudspeaker might give ~ 80 dB SPL at the listening position with 1 W input
1 W into 8 Ohms nominal speaker is 2.8 Vrms
with structured tones spread over multiple critical bands it appears the ear can perceive their linear sum despite each component being below the single tone threshold of hearing in each respective critical band
This gives the occasionally cited "hearing below the noise floor" claim but this is still a limited phenomena - frequency masking by nearby tones/noise still occurs, masking by system noise in the same critical band is still effective, the perceptual noise floor still rises at both frequency extremes
the perceptual noise floor lowest level dips a few dB below the 0 dB spl ref for only a few critical band widths in the low KHz
So if we take the 20 dB number I've seen people bandy about as a estimate of the best sub threshold enhancement possible then we are looking for a collection of single tones, each at worst ~ -20 dB spl
80 dB - (-20 dB) gives -100 dB re 2.8 Vrms
-100 dB re 2.8 Vrms = 28 uVrms, is within my Juli@ s/n with < 1 s averaging, in fact harmonic distortion components should appear as single lines which could readily be seen above the spot noise floor, –120 dB re 2 Vrms card reference V should be easy to pick out with –130 dB spot noise floor
So for the low level harmonic distortion measurements Geddes is proposing here there really is little need for any hardware enhancement to measure the expected levels corresponding to the most extreme estimate of human auditory thresholds I know of
PMA said:
Arny Kruger has high res spectra of most of the major highend soundcards on his web site (pcavtech.com) doing 24bits on -60dB inputs. Even his top card has spurs. What I meant is that when you are actually digitize low level signals there are non-harmonically related spurs (sometimes a lot of them) rather than just a thermal noise floor.
I can't imagine a useful amplifier THD measurement that you couldn't do with a good sound card and a little work.
EDIT - I just noticed that Lynx2 soundcard is $1100
scott wurcer said:
I love my HP3580A spectrum analyzer. I think it actually goes down to 3 Hz BW (if you're willing to wait the time!). I bought it for $700 used about 10 years ago. Probably available for a lot less now.
The only trouble is, it can only show spectra out to 50 kHz - not much good for 20 kHz THD. That's one of several reasons that I prefer 19+20 kHz CCIF IM with spectral analysis.
Also, the 3580A only has about 90 dB dynamic range (it was spec'd at 80), so in order to really sniff out CCIF IM at really low levels I have to put my distortion magnifier (DM) in front of it, giving it an additional 40 dB of dynamic range to be able to see CCIF IM products down to about -130 dB. I can do a little better if I go for a full null on the DM (i.e., lose virtually all of the original stimuli at 19 & 20 kHz) as deep as 60 or so dB.
Bob
Hi
PMA, here is what the firewire Mackie Onyx 400F does on line-in and mic-in.
Note that the line-in does have a roughly 10-20dB headroom here translating in +/- the same performance shown from your card. I assume that in the pic of post 126 your 0 dB = 1 Vrms also equals full scale, no ?
The mic-in gain was set to provide full scale at 1 Vrms hence the " better " performance than line-in which in fact is misleading.
Signal was taken from the output with a 60 dB resistive attenuation.
Averaging was 100 for all plots.
Thanks for the schematic.
---------------------
scott wurcer, thanks for the www.pcavtech.com link. Sadly the Mackie isn't tested.
----------------------
Sent you two mails but seems they didn't reach you ?
Greetings
Michael
PMA said:
PMA, here is what the firewire Mackie Onyx 400F does on line-in and mic-in.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
Note that the line-in does have a roughly 10-20dB headroom here translating in +/- the same performance shown from your card. I assume that in the pic of post 126 your 0 dB = 1 Vrms also equals full scale, no ?
The mic-in gain was set to provide full scale at 1 Vrms hence the " better " performance than line-in which in fact is misleading.
Signal was taken from the output with a 60 dB resistive attenuation.
Averaging was 100 for all plots.
Thanks for the schematic.
---------------------
scott wurcer, thanks for the www.pcavtech.com link. Sadly the Mackie isn't tested.
----------------------
gedlee said:
Sent you two mails but seems they didn't reach you ?
Greetings
Michael
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.