Another realization of Bob Cordell's THD Analyzer

[Cncjerry]

I'll go looking, but who is Victor if I don't find him?

Zapbuzz, thanks for the PM, look for my reply.

I mentioned in a note to Zapbuzz that I have a full CNC shop with mills and lathes. I also have tons of high quality poly caps (and others) many high voltage. I have many, many sprague tantalums, 150D NOS, orange drops, blue drops, bumblebees, etc, etc. Some guy sold me his Tek and HP repair stock. I think I paid $200 for about $40k of parts. I added up just the sprague tantalums at low, low ebay prices and it came to some "pay off the kids school loans" kind of number. I've been sending bags to friends. As I mentioned to zapbuzz, if there is an opportunity to do a joint build on something and we use my component footprints, we might save some money. The only thing about these parts is the many are axial, not radial. Like all the 150Ds you see all of HP gear. By the way, I don't think I have ever had one of them fail. I have had some orange drops fail. They go 'bang' like a .22.

I was active here years ago and then lost my email address to pac bell/att merger and rejoined in 2014. I can't remember the project from 2014 but around then is when my son blew my amp. Just a rambling bit more detail on the KSA50S, the high voltage is soldered to the rails (+/- like 50V or something) and I needed a really big soldering iron to re-solderit. But what happened was all that current across a not so great soldering joint with about 1ohm of resistance heated the joint until the solder melted and the lead fell out. I wish it had just reflowed... I wonder how many amplifiers can do that? reflow themselves?

 

[1audio]

He is Vicnic here on Diyaudio. I'm on a phone so limited in links. The end of the great "Low distortion oscillator" thread has more on his boards.

 

[Cncjerry]

quick question while I'm here. Do analyzers generally measure THD continuously or measure, stimulate, measure? Is there a standard protocol for how they work or should work? I know the weighting filters and test frequencies have standards but what about how the various measurements are made? I know it shouldn't make a difference, but how do you know the harmonic you are looking at has harmonic energy other than comparing it to a baseline?

I have to go do some more reading I guess. this will be one of those dumb questions I'll kick myself for asking when I read the replies. ha!

Thanks

 

[1audio]

Most traditional analyzers are real time indicators. The computer types measure analyze and display. They can do this on a continuous basis but it's a delayed process and sometime a measure multiple times process and then display process.

 

[Bob Cordell]

quick question while I'm here. Do analyzers generally measure THD continuously or measure, stimulate, measure? Is there a standard protocol for how they work or should work? I know the weighting filters and test frequencies have standards but what about how the various measurements are made? I know it shouldn't make a difference, but how do you know the harmonic you are looking at has harmonic energy other than comparing it to a baseline?

I have to go do some more reading I guess. this will be one of those dumb questions I'll kick myself for asking when I read the replies. ha!

Thanks

One thing to point out is that most traditional analyzers provide an output of the distortion residual waveform and amplitude; i.e., all that is left over after the notch takes out the fundamental. This permits one to see the distortion waveform on a scope, along with the noise. It permits a judgement call as to how much of the THD+N is noise and how much is actual distortion. It further permits one to make a judgement as to the relative amount of second and third-order distortion there is in many cases. It is especially valuable in seeing crossover distortion. Finally, it permits the distortion to be further analyzed by a spectrum analyzer and can permit distortion measurement largely in the absence of noise. This is especially valuable when making distortion measurements at low power levels, like 1 watt and below. Many FFT-based distortion analysis tools and instruments do not provide a distortion residual output

Some of this is discussed in my book. Also, on my website can be found the original 3-part construction article from Audio magazine for my THD analyzer. It provides a fairly good explanation of how traditional distortion analyzers work. I use that distortion analyzer heavily to this day.

Cheers,
Bob

 

[phofman]

An FFT-based analyzer shows every harmonics + THD + THD+N, i.e. the harmonics are individually separated from the noise. Is it less convenient? I am learning the practical aspects, thanks.

 

[Bob Cordell]

An FFT-based analyzer shows every harmonics + THD + THD+N, i.e. the harmonics are individually separated from the noise. Is it less convenient? I am learning the practical aspects, thanks.

Yes, and this is a good thing. It just often does not show the real-time distortion residual.

Cheers,
Bob

 

[1audio]

The distortion residual display is the special value of DiAna. I don't know of other fit analyzers that can do that.

 

[phofman]

I do not understand this - a good analog notch filter will have let's say 60dB attenuation. Harmonics at -100dB will be still 40dB below the fundamental in such waveform view - that is just a few pixels at 800x600 resolution.

In digital processing the "notch filter" can be significantly steeper.

The distortion residual display is the special value of DiAna. I don't know of other fit analyzers that can do that.

I could add the fundamental subtraction into my compensation software. Fundamental is being determined precisely by sine curve fitting (amplitude, frequency and phase against the head of the buffer) every 250ms cycle, it can be "compensated" out just like any other harmonics. The analyzing software downstream will receive samples with the fundamental subtracted. If it can display a waveform, it will display the residuals.

Actually both fundamental and all harmonics determined in the cycle can be subtracted, leaving only the noise and spurials for the downstream analyzer, should such an output make sense.

 

[Mark Tillotson]

I do not understand this - a good analog notch filter will have let's say 60dB attenuation. Harmonics at -100dB will be still 40dB below the fundamental in such waveform view - that is just a few pixels at 800x600 resolution.

I think there's a second filter to extract the residual - a non-critical active notch plus a multi-pole Butterworth high-pass wouldn't be hard to implement. Active notch would have enough Q to not depress H2 too far but not so much that frequency drift is an issue. Maybe Bessel filter is needed for lower phase error in the residual.

 

[BrianL]

Bob's comments lead to another useful possibility. Take the 'residual output' of something like an HP339A and feed that into an FFT analyzer ( hp3562A recently mentioned or computer 'sound card'). That will help you further sort things out. With some work you might be able to calibrate it and with a low-distortion source be able to get good measurements. (I think AP may do something like this with some of their products)

 

[Cncjerry]

Thanks, Bob, I'll go back and read the build and book. This is all very interesting. I guess I misunderstood how these analyzers worked. They notch out the fundamental and sum the rest. I was thinking they did successive harmonic energy readings from the FFT like I do with a spectrum analyzer. I know see that as being much more complicated than necessary and would have been impractical with any analyzer.

Playing around today with SpectrumLab, is there is an easy way to calculate THD from the db readings? Spectrumlab also has a sinad function with variable notch I was comparing against. For instance, if I stimulate the system with 1v peak and I see a fundamental of -10db, with 2nd, 3rd, 4th, 5th of -96.4, -99.1, -110.7, -104.9, do I have to convert them to voltages or can I calculate the THD from them more easily?

Getting back to summation of the system after the notch vs FFT, which is preferred? Telling me to read the book is fine as well.

thanks again.

Jerry

 

[Bob Cordell]

Bob's comments lead to another useful possibility. Take the 'residual output' of something like an HP339A and feed that into an FFT analyzer ( hp3562A recently mentioned or computer 'sound card'). That will help you further sort things out. With some work you might be able to calibrate it and with a low-distortion source be able to get good measurements. (I think AP may do something like this with some of their products)

Yes, I very often run the residual into a spectrum analyzer, often my analog HP3580A. It is a great analyzer, but only goes up to 50 kHz. You can see way down below the noise with this approach. I've often thought of running the residual into my DSO, which has a very wideband spectrum analysis funtion, but only 8-bit A2D. But even 8-bit A2D can be useful when you are just looking at the residual. This approach would allow viewing THD-20 harmonics out as far as the THD analyzer's bandwidth can go (200 kHz for a 20 kHz fundamental for my analyzer).

Cheers,
Bob

 

[blmn]

Still I mean even the Asus loopback. I have never seen such noise figure for a soundcard, let alone an internal one.

Phofman,

Last week arrived here a NU Audio card by EVGA (180$ card). This card is newer than ASUS Essence STX and has an advantage: sampling rate higher (384kHz).

I replaced the "audiophile" amp pop ( at line output for an LME49990 x2 and the THD and SNR were improved a lot.

The ASIO driver is not working properly so, initially, I´m using ASIO4ALL.

Below are the loopback tests, one is the best THD spot and the other is the best SNR spot (different line levels).

The THD numbers are worse than the ASUS ones, but SNR is better and the card is unshielded in a PCI-e extender over the bench. When I shield it, the numbers would be better.

Regards,