Yeah. Progress isn't always progress. It's only in the eyes of the beholder. When the beholder is just the vendor, that doesn't always help the customer.
I may be whistling past the graveyard, but one of the things I like about USB interfaced products is that USB has, so far anyway, been backward compatible. Mostly. That doesn't always help with drivers, but I suspect it does some.
My previous workbench computer was a 2005 iMac with a Core2Duo processor running Windows 7 32 bit natively through Boot Camp. It actually worked great. But, when Apple got out of the Intel using business, I stumbled across a NOS (aka refurb) iMac with an i7 processor. The vendor (MacSales) is partially in the business of selling SSD's and memory and they installed a nice SSD and 32 Gb of memory. The deal was too good to pass up. That allowed me to run Windows 10 64 bit.
That all was appealing at the time because I wanted to upgrade from a QuantAsylum QA401 to a QA403 and the QA software was targeted toward Windows 10 and later. As it turned out, the QA403 was delayed due to parts shortages so I tried the E1DA Cosmos route. So far, it's worked pretty well. But, I am an amateur hobbyist so that also sways my options and choices. I found that I prefer the REW and Multitone software packages for making audio measurements over the QA system. Personal choice...
Turns out that it really helps if one calibrates both the output from and input to REW. It seems REW can get the signal input calibration from its own generator, but that leaves external noise uncalibrated. At least I think that was my issue. After 'calibration' of the input and output potentiometers on the Focusrite Solo, REW-generated noise displays and scales correctly.
Regarding noise calibration, has anybody used or built an external noise calibrator, e.g. from a resistor and an amplifier for gain and buffering? Analog Devices has a note describing such a thing, but the op-amp they recommend is quite limited in gain bandwidth product, so the device is only good for noise up to a few hundred Hz.
Would maybe a Microchip MCP6V81 work in a similar circuit? It also has much lower input noise (13 nV/Sqrt(Hz) instead of 230 nV/Sqrt(Hz)) than the LTC2063, and requires only a single supply voltage. I have zero experience with these 'zero-drift' op-amps.
Schematic from the Analog Devices note linked above:
I would think that some of the modern low-noise, low-drift opamps could be of interest. The challenge will be to find one with low 1/f noise.
That's where the zero-drift opamps are nice. Zero-drift can be implemented by swapping polarity on the input and output at some rate that's generally above the audio band. Take the LMP2021 for example. There's some switching going on at around 30 kHz as evidenced by the spurs in the noise plot below.
This switching has the advantage that it nulls out the input DC offset, but it also cancels out the 1/f noise so if you build a noise source with an auto-zero amp, you'll get white. The drawback, as you've discovered, is that these amps tend to have limited useful bandwidth.
But as I mentioned, some of the newer low-offset, low-drift opamps also have very low 1/f noise, so they could be useful if a white noise source is what you're after.
I'm not sure what you'd get from such a noise calibration, though.
Tom
That's where the zero-drift opamps are nice. Zero-drift can be implemented by swapping polarity on the input and output at some rate that's generally above the audio band. Take the LMP2021 for example. There's some switching going on at around 30 kHz as evidenced by the spurs in the noise plot below.
This switching has the advantage that it nulls out the input DC offset, but it also cancels out the 1/f noise so if you build a noise source with an auto-zero amp, you'll get white. The drawback, as you've discovered, is that these amps tend to have limited useful bandwidth.
But as I mentioned, some of the newer low-offset, low-drift opamps also have very low 1/f noise, so they could be useful if a white noise source is what you're after.
I'm not sure what you'd get from such a noise calibration, though.
Tom
Making a calibrated flat noise source is not easy, and then what are you trying to do with it? Assuming the software has the correct math, calibrating the chain with a sine wave is the best starting point. Assuming your DUT has some gain (20 dB) a reasonably low noise input on the analysis chain should not limit your efforts. A noise source can be useful for validating your chain.
A noise diode is the usual starting point for a noise source. You can use a resistor. It should have a significantly higher noise than the input amp of your noise source. And I would recommend a FET input or input noise current will mess with your efforts. Another way is a pseudi-random source using a shift register or software etc. Or get a General Radio 139X noise generator. Around $80 to $130. However they are not calibrated in mV/rtHz. I can help with that if you find you need.
A noise diode is the usual starting point for a noise source. You can use a resistor. It should have a significantly higher noise than the input amp of your noise source. And I would recommend a FET input or input noise current will mess with your efforts. Another way is a pseudi-random source using a shift register or software etc. Or get a General Radio 139X noise generator. Around $80 to $130. However they are not calibrated in mV/rtHz. I can help with that if you find you need.
Something like this? That one is in Irvine, right next door. That seller has a veritable cornucopia of stuff. I could even pick up a spare copy of my espresso machine! ;-)
But then such a device probably needs recapping and retubing, plus it seems like a bit of a space hog, and what seemed like good performance in the 50s might not quite be state-of-the-art anymore...
We'll see where this goes. Maybe I'll try to build something.
Thanks for the quick replies!
But then such a device probably needs recapping and retubing, plus it seems like a bit of a space hog, and what seemed like good performance in the 50s might not quite be state-of-the-art anymore...
We'll see where this goes. Maybe I'll try to build something.
Thanks for the quick replies!
I have one of those. Untouched since its left their factory. Still meets spec. Noise spec is complex with both frequency and randomness requirements. They are built to last a very long time unlike consumer stuff.
Still what are you intending to do with noise? There may be a better way.
Still what are you intending to do with noise? There may be a better way.
So you have such a device, but you find it not all that useful? Interesting that it's based on a thyratron, which is quite close to Lee DeForest's earliest (weak) vacuum tubes.
I'm not sure what I'm doing with more than one amplifier, to be honest. A simple stand-alone noise source signal standard dependent on nothing more than a resistor at a known temperature seems interesting. To be useful it would have to be quite low noise, which (as you guys know better than I do) is not all that easily achievable. So, we'll see. I ordered some parts from Digikey (thanks again for the TI alternative suggestion, Tom). If nothing else, it will keep me out of trouble. ;-)
I'm not sure what I'm doing with more than one amplifier, to be honest. A simple stand-alone noise source signal standard dependent on nothing more than a resistor at a known temperature seems interesting. To be useful it would have to be quite low noise, which (as you guys know better than I do) is not all that easily achievable. So, we'll see. I ordered some parts from Digikey (thanks again for the TI alternative suggestion, Tom). If nothing else, it will keep me out of trouble. ;-)
I just tested a possible workaround for people who can't run Multitone with their test gear setup.
It's possible to save a generated test signal from Multitone to a wav file. So, if you have access to a Windows 10 or later computer with 64 bits, you could just set up the test signal and save it for use on whatever computer you wish.
I did just that and played it back through Audacity into the same DAC as before and recorded the spectrum from the same ADC as before in REW. You lose the display of IMD numbers and so on that you get with using a full Multitone setup, but the spectrum display is there.
I figure that this might be useful for making full system measurements and comparisons some day, as discussed a little bit in another thread hereabouts.
Note: You can do the same thing with LTspice, although it's a little more effort and you lose dithering capability unless you build a dithering circuit in LTspice. There's almost certainly other software programs that will do this.
Good idea to generate on a 64 bit system and measure on whatever. More broadly speaking the measurements suggest that there is not likely to be audible improvements from here. The numbers are far beyond anything electromechanical can create.
Multitone
OT, a little. I've been trying Kagi instead of Google as a search engine. Much less irrelevance.
Ahh, but this is the test system. No preamp or power amp. A nice baseline, but nothing you can actually listen to in your living room.
Plus, as mentioned in that other thread, no unmentionable complications like common mode currents conducted through the AC mains and the interconnects, and so forth. Lots of care is placed on isolation in the test system, from USB isolation to batteries for some components.
Since it's fresh in my mind, I tried a couple more things.
This is what could be called Phi14. I added a low tone at 29 Hz and a high one at 23173 Hz.
This is the NID Multitone test built into REW. The sequence is based upon a paper by Evan and Rees, which is behind an IEEE paywall. It's pretty punishing, at least with flat (white shaped) spectrum. Pink shaped is probably much more realistic when compared to music.
This is what could be called Phi14. I added a low tone at 29 Hz and a high one at 23173 Hz.
This is the NID Multitone test built into REW. The sequence is based upon a paper by Evan and Rees, which is behind an IEEE paywall. It's pretty punishing, at least with flat (white shaped) spectrum. Pink shaped is probably much more realistic when compared to music.
Different tests. But, exactly the same hardware in every way. I didn't even move the cables between tests - just the mouse to run the software.
Sometime I'll try adding the exact same test to Multitone and run it from there.
OK, here's the most accurate comparison I can make.
I ran the NID Multitone test in REW. I asked for a lower frequency limit of 20 Hz and an upper limit of 23 KHz. REW automatically sets the frequencies for all the tones. John (the REW author) adjusts the tones to fit into in FFT bins, so there's some alterations from exactly what is asked for.
Based on the tone sequence progression that John had described for his NID tone placement on a thread somewhere, I then calculated the tone frequencies using that 20.5 Hz as a base. First warning: My top frequency was a couple Hz off of what REW told me was being used. That's undoubtedly due to the adjustment John does mentioned above.
I then added that tone sequence as a test in Multitone. Because it's easier to adjust test levels in REW, I ran the Multitone test first and measured the RMS level of the test spectrum with my Fluke 87. Yeah, the 87 probably isn't table top flat out to 40 KHz, but the results should be directly comparable regardless of which test software was being used. It was .710 Vrms.
Ran the distortion test in Multitone:
Then, I switched to REW and adjusted the generator output level to the same 0.710 Vrms.
Ran the distortion test in REW:
The FFT length was the same for both tests (1 Meg) and 16 averages were used. Same filter, too.
But, it's not 100% apples to apples comparison because of the tweaked tone frequencies in REW and I don't know how similar the crest reduction algorithms are between the two software applications. That probably explains the dB or so difference in TD+N between the two sets of results. But, that's actually not so bad. To me.
Of course, the same hardware from computer to cables to power to DAC to Scaler to ADC was used for both tests.
BTW, the NID test I showed earlier from REW was at a many dB higher signal level.
Edit: Next time I have the gear set up, I will look at the signal in REW's scope mode to get an idea of the crest factor.
I ran the NID Multitone test in REW. I asked for a lower frequency limit of 20 Hz and an upper limit of 23 KHz. REW automatically sets the frequencies for all the tones. John (the REW author) adjusts the tones to fit into in FFT bins, so there's some alterations from exactly what is asked for.
Based on the tone sequence progression that John had described for his NID tone placement on a thread somewhere, I then calculated the tone frequencies using that 20.5 Hz as a base. First warning: My top frequency was a couple Hz off of what REW told me was being used. That's undoubtedly due to the adjustment John does mentioned above.
I then added that tone sequence as a test in Multitone. Because it's easier to adjust test levels in REW, I ran the Multitone test first and measured the RMS level of the test spectrum with my Fluke 87. Yeah, the 87 probably isn't table top flat out to 40 KHz, but the results should be directly comparable regardless of which test software was being used. It was .710 Vrms.
Ran the distortion test in Multitone:
Then, I switched to REW and adjusted the generator output level to the same 0.710 Vrms.
Ran the distortion test in REW:
The FFT length was the same for both tests (1 Meg) and 16 averages were used. Same filter, too.
But, it's not 100% apples to apples comparison because of the tweaked tone frequencies in REW and I don't know how similar the crest reduction algorithms are between the two software applications. That probably explains the dB or so difference in TD+N between the two sets of results. But, that's actually not so bad. To me.
Of course, the same hardware from computer to cables to power to DAC to Scaler to ADC was used for both tests.
BTW, the NID test I showed earlier from REW was at a many dB higher signal level.
Edit: Next time I have the gear set up, I will look at the signal in REW's scope mode to get an idea of the crest factor.
Last edited:
After a little more reading in the details of the REW Help files, I learned that the crest factor for these NID sequences is just about 12 dB, according to John Mulcahy. That's a good enough reference for me. He says that the sorta random frequency alignments like that don't respond well to tweaking the phase alignments to minimize crest factor.
Since I was running the generator at -3 dB in Multitone to avoid the pointy end of the DAC performance curve, the numbers pretty much match up.
Since I was running the generator at -3 dB in Multitone to avoid the pointy end of the DAC performance curve, the numbers pretty much match up.