Low-distortion Audio-range Oscillator

[Bob Cordell]

Originally Posted by Terry McGee

Agreed, Victor. I'm looking forward to receiving one of your oscillators to be my primary distortion reference. I have a thermistor-style variable frequency oscillator and a function generator for other tests.

When I built my Noise and Distortion meter, I settled on three frequencies to avoid the complexity of full range tuning:

40Hz, being about the lowest musical notes likely to be encountered
400hz, being in the middle of the vocal and instrumental fundamentals range, and
4kHz, being about the top of the instrumental fundamentals range.

I didn't see much point in measuring harmonic distortion at 10kHz or so, as I was not going to hear even gross 2nd harmonic distortion at 20kHz, let alone 3rd at 30k. I'd be interested in comments on that strategy.

Terry

Hi Terry,

My vote is for 50Hz, 1kHz and 20kHz. Those are the frequencies used by many others, including John Atkinson of Stereophile. This allows you to make meaningful comparisons.

The chosen frequency for the mid-band is non-critical, and in this sense 400Hz and 1kHz are about equivalent.

50Hz and 40Hz should not give much different results. Many solid-state amplifiers don't do as well at 50Hz as one might expect, so the LF THD test is important. If looking at the spectrum of the residual, you can see any 60Hz components as well under these stressed conditions of medium to high power.

Many people dismiss 20kHz THD because they cannot hear any of the harmonics. But it is important to recognize that the THD-20 measurement is measuring a symptom of high-frequency nonlinearity. If THD-20 is not so good, there will surely be IM products reflected down into the audio band (look at 19+20kHz CCIF IM distortion spectra).

Cheers,
Bob

[vicnic]

Originally Posted by Terry McGee

Agreed, Victor. I'm looking forward to receiving one of your oscillators to be my primary distortion reference. I have a thermistor-style variable frequency oscillator and a function generator for other tests.

When I built my Noise and Distortion meter, I settled on three frequencies to avoid the complexity of full range tuning:

40Hz, being about the lowest musical notes likely to be encountered
400hz, being in the middle of the vocal and instrumental fundamentals range, and
4kHz, being about the top of the instrumental fundamentals range.

I didn't see much point in measuring harmonic distortion at 10kHz or so, as I was not going to hear even gross 2nd harmonic distortion at 20kHz, let alone 3rd at 30k. I'd be interested in comments on that strategy.

Terry

Thank you, Terry.
1kHz oscillator already shipped to you.
About measuring. I think, that there are no limits for the audio equipment performance. Our ears are so magic tool. We can hear differences, that practically no one measurment tool can.
In measurment practice, I think, that more interesting and more informative are points on the audio range borders and middle point - app. 20Hz, 20kHz, 1kHz. Some combinations for the IMD testing are also interesting. Low frequency measuring is important for RIAA amplifiers and, in my opinion, for possible specific thermal distortions. I accept, that Bob said about 20kHz distortions.
Regards,

Victor

[Terry McGee]

Bob, I used 40Hz rather than 50 Hz as 50Hz is our local power frequency and I wanted to be able to differentiate between the test note and hum. I agree that 40 or 50 Hz are likely to give the same LF results. Good bass measurement. And I agree 400 Hz and 1KHz are also going to give essentially the same results. Good midrange measurement.

I still have my doubts though about the significance of distortion measurements much above say 5000Hz. I'm clinging to the conservative view that if you can't hear the artefact, it doesn't matter, but I'm happy to have the view exploded. You mentioned:

"Many people dismiss 20kHz THD because they cannot hear any of the harmonics. But it is important to recognize that the THD-20 measurement is measuring a symptom of high-frequency nonlinearity. If THD-20 is not so good, there will surely be IM products reflected down into the audio band (look at 19+20kHz CCIF IM distortion spectra)."

But even if high frequency non-linearity were significant, wouldn't this require pretty high levels of content at these elevated frequencies to trigger audible amounts of the IM products? I'm unaware of any musical instruments with significant energy up there.

Terry

[benb]

Trumpets.

[Terry McGee]

Fletcher and Rossing (the bible on musical instrument acoustics) cites a study by Luce & Clarke (1967) on the topic of Acoustic Spectra of brass instruments (p453). From the graph, the "cut-off frequency" of the trumpet played mezzo-forte (moderately loud) is about 1kHz. They give the roll-off above cut-off as 15 to 25dB/octave. They do say that when played louder, the slope above cutoff decreases, but don't give a figure. If we guessed at 10dB/octave, and counted four octaves up from 1k to 16k, we'd expect the energy to be 40dB down wrt the peak level in the musical range. So the question becomes, how bad would the IM distortion at 16kHz have to be that harmonics 40dB down would create intermodulation products at lower frequencies that could be heard over the infernal racket the trumpet player is making anyway!

I do like baroque trumpet in particular, so I'll be very pleased is someone can convince me the maths above do not stack up!

Incidentally, be aware that when musical scientists use the expression "cut-off frequency", they mean more like what we would call "turnover frequency" or "corner frequency". It's the start of the downhill slide, not some point down the slope. That's why there can still be substantial energy above that frequency.

Terry

[Terry McGee]

Originally Posted by vicnic

Thank you, Terry.
1kHz oscillator already shipped to you.

Thanks Victor! Looking forward to having a "gold standard"!

About measuring. I think, that there are no limits for the audio equipment performance. Our ears are so magic tool. We can hear differences, that practically no one measurment tool can.
In measurment practice, I think, that more interesting and more informative are points on the audio range borders and middle point - app. 20Hz, 20kHz, 1kHz. Some combinations for the IMD testing are also interesting. Low frequency measuring is important for RIAA amplifiers and, in my opinion, for possible specific thermal distortions. I accept, that Bob said about 20kHz distortions.

I notice you offer oscillators at 1kHz and 10kHz, but not a low frequency. Or is that still to come?

I am a bit tempted to change my 40, 400, 4kHz but still not sure what to. It is an inconvenience not to have "plain vanilla" 1kHz available, so that's probably one point. 50Hz is ruled out as being our mains frequency, so maybe 20Hz makes the best sense. But at the high end, 10kHz because it's still clearly audible, 15kHz because it's on my limit of audibility, or 20k because of the intermod theory? Decisions, decisions!

Terry

[1audio]

I chose 3 specific frequencies for oscillators from Victor. My need is for testing audio ADC's so those were 997 Hz (to avoid any interaction with sample rate subharmonics and to test every bit level) 11.025 KHz (44.1 KHz/4) and 12 KHz (48KHz/4) which are commonly used for Jitter testing. I can also use those for IM testing.

I'll be battery powering them (3 12V NiMH batteries in series from eBay @ $8.00 ea + a 36 V charger for $15 I think and some current limiting resistors). I get the best results from the stacked shunt regulators by powering them from an LM317 set as a current source. Also short circuit proof.

For low frequencies the level detection and agc are difficult. I think the only way to get reasonably fast settling is a sample and hold solution. That's much easier to implement with a state variable oscillator (you have a zero crossing to exploit at every peak). I have done some work towards this if anyone wants to pursue it. The simulations are promising but they are not working circuits yet.

[vicnic]

Originally Posted by Terry McGee

I notice you offer oscillators at 1kHz and 10kHz, but not a low frequency. Or is that still to come?

Terry

Is possible to build low frequency oscillators, but there are two reasons why I do not offer them. ( I have one 20Hz oscillator for myself):
1) Practically no one asked me about low frequency similar oscillators.
2) In this configuration low frequency oscillators become very inertial for any output level changes and have very big settling time after switching ON. My 20Hz oscillator have more than 5 minutes settling time after switching ON. This cause of big time constants in AGC.
But, I still thinking about it

Victor.

[1audio]

Here is the unfinished design for the fast settling state variable oscillator. It borrows the additional peak detecting ideas from Cordell's oscillator, adding more and uses a trick I have used making an LM339 work as a peak detector.

The LM339 is a fast comparator but internally its like an unfinished opamp with open collector outputs. It has the virtue of not going nuts when the output is in an odd state. By feeding back the output to the input as shown it works as a "perfect" rectifier. Being open collector we can wire or several outputs together as shown.

The next stage is borrowed from the Fluke 510A oscillator which is designed to have extremely stable output. The DC offset from R2 moves the peak negative voltage to 0V at the output of U5. U6 is a comparator that switches to turn on the sample and hold at the peak voltage. This is modeled in the LM339 rectifier sim attached.

I have not added a sample and hold (did not see one in the library) or the AGC. I think Victors AGC would be well suited to this.

Also not done are the frequency tuning, the startup circuit or some of the other details. I found Victors modules and did not need to finish it.

[Jakob2]

Originally Posted by Terry McGee

Fletcher and Rossing (the bible on musical instrument acoustics) cites a study by Luce & Clarke (1967) on the topic of Acoustic Spectra of brass instruments (p453). From the graph, the "cut-off frequency" of the trumpet played mezzo-forte (moderately loud) is about 1kHz. They give the roll-off above cut-off as 15 to 25dB/octave. They do say that when played louder, the slope above cutoff decreases, but don't give a figure. If we guessed at 10dB/octave, and counted four octaves up from 1k to 16k, we'd expect the energy to be 40dB down wrt the peak level in the musical range. So the question becomes, how bad would the IM distortion at 16kHz have to be that harmonics 40dB down would create intermodulation products at lower frequencies that could be heard over the infernal racket the trumpet player is making anyway!

I do like baroque trumpet in particular, so I'll be very pleased is someone can convince me the maths above do not stack up!

Incidentally, be aware that when musical scientists use the expression "cut-off frequency", they mean more like what we would call "turnover frequency" or "corner frequency". It's the start of the downhill slide, not some point down the slope. That's why there can still be substantial energy above that frequency.

Terry

Still a good source for the spectral content of various instruments is:

There's life above 20 kilohertz! A survey of musical instrument spectra to 102.4 kHz