Thanks, Hugh and x-pro for your input, you both know what it takes to make a successful audio product.
Conrad, thanks for giving me a little more background. This is helpful.
Let me give an example of audio measurement development:
In the old days, people used wave analyzers, and fairly high distortion (.035%) audio oscillators to measure audio equipment. THD was then considered almost useless, even before 1940.
In the 1950's the film industry found that harmonic tone testing was not giving the whole picture, and they (SMPTE) developed a 2 tone IM test for distortion measurement. Papers were written about it, and correlations to THD were made as well.
By 1960, everybody, except analog tape recorder designers switched over to SMPTE IM testing. It was fast, it could be VERY sensitive, and you could look at the IM waveform residual and do further evaluation, if you wanted to. In 1965, the hi fi shop that I worked at could measure as low as 0.005% IM. By 1970, Crown introduced the IMA that could measure as low as 0.001% IM. Wow! I got one of these instruments in 1974. Then I would put a wave analyzer on the output residual and separate the harmonics (double Wow!). That should have been good enough for rock and roll, and anything else, BUT it wasn't.
Why? Because SMPTE IM completely ignored high frequency distortion caused by rate of change of the input signal. Did Crown know this? No. Why was SMPTE IM so successful in the 1950 and 1960's? Because we did NOT design with op amps for audio,
as a rule, in those years.
About 1975, when the BAS was still doing useful audio projects, Tom Holman came up with a test that he thought was reasonable. It was a 1KHz square wave, that was bandwidth limited by 1 pole at 30KHz. He reasoned that ANY amp or preamp should be able to pass this test, without much trouble. But he was wrong! Most amps failed the test, but the test was not perfected yet. It ONLY easily showed even order harmonics, and tended to obscure the odd order harmonics, because they mapped back on to the original square wave, itself.
In 1976, I was invited by Dr. Otala to work in his government lab in Finland for 1 month to help develop a better test. He and his associates had already developed an alternative square wave test that used an extra sine wave that was mathematically out of sync with the normal square wave fundamental frequency. He chose 3.18KHz for the square wave and 15KHz for the sine wave.
Now what ratios should be chosen between the square wave and the sine wave? Well, why not closely follow the SMPTE example of 4:1? So we chose it.
Finally, what about the controlling bandwidth that limited the risetime to something finite. Well, Matti thought that a REAL amp might be able to pass a 100KHz bandwidth, but we knew that most real world IC's and commercial power amps could not pass this test. This was 3.5us rise time. Too fast for most stuff at the time.
Matti had initally tried a 20KHz roll-off, or a 17.5us rise time. This worked well, even if 741 op amps still were terrible in this test, but I suggested 30KHz, just like Holman. Why not? Audio bandwidth did NOT stop, just because the ear could not hear it anymore, or at least very well. Remember, this was the age before brick wall filters in every piece of audio equipment. Ticks, pops, noise, distortion, all had extended bandwidth, (and they still do).
We settled on TIM 30 for general testing, TIM 100 for super testing. The paper was given in the fall of 1976 at the AES and in the Journal a few months later.
Interestingly enough, when working with the test, I developed a good insight of the differences between IM and TIM. What I did was switch the TEK square wave generator to triangle wave, and I had a very close SMPTE IM test. Switch back and forth a few times between squared wave and triangle wave of the same peak amplitude, and you would gain insight as well.
This is one example of how it is done, with many months of research, discussion, and testing. It costs real money to do it properly, and nobody want to invest in it, anymore, so things change slowly.
Conrad, thanks for giving me a little more background. This is helpful.
Let me give an example of audio measurement development:
In the old days, people used wave analyzers, and fairly high distortion (.035%) audio oscillators to measure audio equipment. THD was then considered almost useless, even before 1940.
In the 1950's the film industry found that harmonic tone testing was not giving the whole picture, and they (SMPTE) developed a 2 tone IM test for distortion measurement. Papers were written about it, and correlations to THD were made as well.
By 1960, everybody, except analog tape recorder designers switched over to SMPTE IM testing. It was fast, it could be VERY sensitive, and you could look at the IM waveform residual and do further evaluation, if you wanted to. In 1965, the hi fi shop that I worked at could measure as low as 0.005% IM. By 1970, Crown introduced the IMA that could measure as low as 0.001% IM. Wow! I got one of these instruments in 1974. Then I would put a wave analyzer on the output residual and separate the harmonics (double Wow!). That should have been good enough for rock and roll, and anything else, BUT it wasn't.
Why? Because SMPTE IM completely ignored high frequency distortion caused by rate of change of the input signal. Did Crown know this? No. Why was SMPTE IM so successful in the 1950 and 1960's? Because we did NOT design with op amps for audio,
as a rule, in those years.
About 1975, when the BAS was still doing useful audio projects, Tom Holman came up with a test that he thought was reasonable. It was a 1KHz square wave, that was bandwidth limited by 1 pole at 30KHz. He reasoned that ANY amp or preamp should be able to pass this test, without much trouble. But he was wrong! Most amps failed the test, but the test was not perfected yet. It ONLY easily showed even order harmonics, and tended to obscure the odd order harmonics, because they mapped back on to the original square wave, itself.
In 1976, I was invited by Dr. Otala to work in his government lab in Finland for 1 month to help develop a better test. He and his associates had already developed an alternative square wave test that used an extra sine wave that was mathematically out of sync with the normal square wave fundamental frequency. He chose 3.18KHz for the square wave and 15KHz for the sine wave.
Now what ratios should be chosen between the square wave and the sine wave? Well, why not closely follow the SMPTE example of 4:1? So we chose it.
Finally, what about the controlling bandwidth that limited the risetime to something finite. Well, Matti thought that a REAL amp might be able to pass a 100KHz bandwidth, but we knew that most real world IC's and commercial power amps could not pass this test. This was 3.5us rise time. Too fast for most stuff at the time.
Matti had initally tried a 20KHz roll-off, or a 17.5us rise time. This worked well, even if 741 op amps still were terrible in this test, but I suggested 30KHz, just like Holman. Why not? Audio bandwidth did NOT stop, just because the ear could not hear it anymore, or at least very well. Remember, this was the age before brick wall filters in every piece of audio equipment. Ticks, pops, noise, distortion, all had extended bandwidth, (and they still do).
We settled on TIM 30 for general testing, TIM 100 for super testing. The paper was given in the fall of 1976 at the AES and in the Journal a few months later.
Interestingly enough, when working with the test, I developed a good insight of the differences between IM and TIM. What I did was switch the TEK square wave generator to triangle wave, and I had a very close SMPTE IM test. Switch back and forth a few times between squared wave and triangle wave of the same peak amplitude, and you would gain insight as well.
This is one example of how it is done, with many months of research, discussion, and testing. It costs real money to do it properly, and nobody want to invest in it, anymore, so things change slowly.
john curl said:
Thanks for this overview, John. This is very helpful, and I really mean that. The DIM test you were referring to was a good test, it was just difficult to do and tended to require an expensive spectrum analyzer. I also like what you said about the SMPTE IM test, along with its virtues and its limitations.
Here are my only caveats. While it is certainly true that the op amps of the day were terrible at TIM, the good audio op amps of today produce exceedingly low TIM. Production of TIM is not strongly associated with op amps, or op amp topologies, or low open loop bandwidth. It is associated with slew rate and high frequency linearity (hard and soft TIM, respectively).
As far as the DIM test goes, its only fault was that it was difficult and often the dynamic range of spectrum analyzers limited its sensitivity. That is why I showed that THD-20 was just as sensitive in exposing TIM as was DIM - because it was so much easier to do. Not enough people paid attention to THD-20, and too many people paid attention to THD-1. In my paper "Another View of TIM" (on my web site at www.cordellaudio.com) I show numerous op amps and power amps with both DIM and THD-20 results, and they are highly correlated. In many cases the measurement floor of the THD-20 test was just as low as that of DIM, or lower, due to the instrumentation available and the dynamic range limitations of the spectrum analyzer.
On the other hand, it is certainly true that THD-20 can be at a disadvantage when, for whatever reason, there is a bandlimited situation. In pursuit of cheap instrumentation, that is why I developed the Multitone Intermodulation (MIM) test, which allowed one to see the odd and even order products of IM in-band. Three equal-level frequencies of 9.0, 10.05 and 20 kHz were used. That test never really caught on, and with the falling cost of PC-based spectrum analyzers, many of the disadvantages of the two-tone CCIF test (its need for a spectrum analyzer) went away. That's why I like the CCIF IM test with full spectral analysis today. Also, the CCIF test is not just limited to a single number and the relative strengths of the different orders of distortion products are pretty clearly presented.
Interestingly, with the availability of PC-based spectrum analyzers and a little software, the DIM test could be a more practical test these days 🙂.
Cheers,
Bob
John,
Has anyone ever tried something like IM, but using two square waves, instead of sines? If you visualize an orchestral climax with, say, cymbals and brass hitting FFFF at the same time, you're going to have a prodigious amount of high frequency information--well into the ultrasonic--and high signal levels at the same time. Given that one of the more common subjective complaints is that "it falls apart on climaxes," that might give some insight into what's going on.
Grey
Has anyone ever tried something like IM, but using two square waves, instead of sines? If you visualize an orchestral climax with, say, cymbals and brass hitting FFFF at the same time, you're going to have a prodigious amount of high frequency information--well into the ultrasonic--and high signal levels at the same time. Given that one of the more common subjective complaints is that "it falls apart on climaxes," that might give some insight into what's going on.
Grey
Bob, of course, you are correct, as far as it goes. I agree that THD 20 is OK. Walt Jung started publishing this back at about the same time that Matti and I gave our paper. In fact, we all met at the NY, AES at the same time. Walt called it SID and it led to some contention until 1979 or so, when I got Matti and Walt together, to iron out the details. That is why our letter of rebuttal to you in 1980, was signed by all of us.
Unfortunately most mid fi still uses the 4558. This is also the sort of op amp in Dolby like processors, etc. Trust me, TIM is still around.
Unfortunately most mid fi still uses the 4558. This is also the sort of op amp in Dolby like processors, etc. Trust me, TIM is still around.
Grey, it is too complex. It would be possible but interpretation would take a computer program just for figuring out what you were looking at.
Complex, yes. But I fear that it will take something brutal like this to shake out useful information.
I don't have sufficient test equipment on hand to rig this--mainly lacking a spectrum analyzer. I'm sure there's a way to get my 7603 to sweep frequency vs. level, but I've never sat down and figured it out. At any rate, I envision something like 1kHz and 10kHz, relative levels open to negotiation, then null them at the output (try to filter a square? huh, I've got better ways to give myself migraines). Okay, so what do we get? I imagine we'll get a lot of broad spectrum hash because the circuit will not cope gracefully. Yes, the two fundamentals will interact in a normal IM-ish sort of way. Let's disregard that. Perhaps even the second components. But after that I'm curious to know if there are any unexpectedly high spikes. It might be a complete failure as a test, burying the tester in an avalanche of data as you suggest. But...if there were to be a notably high spike that didn't correlate to any of the more obvious harmonic interrelations, that might be easy.
I do not anticipate that it would lead to an immediate Eureka! moment. Even if there was an obvious spike, you still have to test a number of circuit topologies/bias points/device selection to get some ideas as to what caused it.
Grey
I don't have sufficient test equipment on hand to rig this--mainly lacking a spectrum analyzer. I'm sure there's a way to get my 7603 to sweep frequency vs. level, but I've never sat down and figured it out. At any rate, I envision something like 1kHz and 10kHz, relative levels open to negotiation, then null them at the output (try to filter a square? huh, I've got better ways to give myself migraines). Okay, so what do we get? I imagine we'll get a lot of broad spectrum hash because the circuit will not cope gracefully. Yes, the two fundamentals will interact in a normal IM-ish sort of way. Let's disregard that. Perhaps even the second components. But after that I'm curious to know if there are any unexpectedly high spikes. It might be a complete failure as a test, burying the tester in an avalanche of data as you suggest. But...if there were to be a notably high spike that didn't correlate to any of the more obvious harmonic interrelations, that might be easy.
I do not anticipate that it would lead to an immediate Eureka! moment. Even if there was an obvious spike, you still have to test a number of circuit topologies/bias points/device selection to get some ideas as to what caused it.
Grey
I watched them on Ebay for a while, then went with a HP something or other and got badly burned. It wouldn't even start properly. Once I began reading more closely, I noticed that a lot of the HPs weren't working if you read between the lines. They don't seem to hold up well. I managed to get rid of the HP.
The Tektronix unit might be the ticket, especially since I've got two 7603s, one bench and one rack mount and can afford to dedicate one to the plug-in, which if I recall is a two space unit.
Grey
The Tektronix unit might be the ticket, especially since I've got two 7603s, one bench and one rack mount and can afford to dedicate one to the plug-in, which if I recall is a two space unit.
Grey
I'm not exactly sure if it will work, because I originally used it in a 7633, but it IS a good unit. Probably expensive too. There may be another plug-in that is cheaper and will work pretty well.
Hey, now we're getting somewhere! Would y'all say that a CCIF IM test has at least some correlation with sound, and should be part (and only a part) of a meaningful suite of tests? Or, given the computer power available to everyone, should it be a different version like the 3-tone test? Along the same lines, about a mile and a half up the thread, there was an actual highly regarded amplifier under discussion. Are there any published IM results for it, or most other amps? My guess is, few want to reveal their results in a subjective oriented market. If I set up said IM test, where would those in the know draw the line between excellent performance, and everything else? I'm looking for a few courageous souls to say, "Yes, knowing what we know today, I'd do A and try for such-and-such limits." Then we'll go after B.
CCIF IM is problematic also. Three or more tones is better, because it brings out a 'triple beat' that has twice the amplitude as normal harmonics, or IM products.
Last evening I downloaded the 1977 JAES article "A Method of Measuring TIM..." (and comments and corrections) I note that at the end of the article the statement is made that "...some of the harmonics may be outside the closed loop bandwidth of the amplifier, above which an added attenuation of -6 to -18 dB per octave is generated. This may be rephrased by stating that most high-frequency distortion is generated in the driver circuits. The ultrasonic harmonic components may then be attenuated in the intrisically slow power output stages..."
Isn't the first problem splendidly tackled by National's engineers with the LM4702 and LME49810?.
Isn't the first problem splendidly tackled by National's engineers with the LM4702 and LME49810?.
Jack, look at the comments by Matti Otala on the relative importance of CCIF 2 tone distortion in that paper. It is very accurate.
Conrad Hoffman said:
Tried, even more tones. Again, it does not give answers, as it is not a big problem to design an amp that measueres perfectly in 3-tone or multitone.
What we need are transient signals. Every time limited signal has unlimited frequency spectrum, that makes a small problem. Maybe AM modulation might say something, I do not know. I tried FM and again did not get answers. Earl Geddes has a point in exploring spectra from very low power levels, like mW.
PMA said:
Think about exciting the circuit with an impulse and looking at the residual spectrum. Every circuit is a resonant structure similar to a speaker cabinet. You bang on it and you see it ring. Varying energy levels are interesting.
Mike
MikeBettinger said:
I'll throw this out -- Bill Perkins proposed this signal as an input -- the discontinuities at the zero crossings gernerate a lot of HF stuff.... he actually had an analog prototype at one time.
An externally hosted image should be here but it was not working when we last tested it.
Illustration of a stepwise approximation of 5-cycle cosine burst
and it's harmonic spectrum by Fourier analysis
dave
Ok, sounds like CCIF IM isn't challenging enough. Interesting about the above function, as I've been thinking along similar lines. Since I've been transferring a lot of my LPs to CD, I've spent a lot of time looking at waveforms, particularly pops and clicks. One thing I notice is the signal immediately after a unipolar disruption is offset, and takes a while to settle back to the correct average. A surprisingly long while. No idea what the mechanism is for this, but it got me to thinking about non-symmetrical waveforms. I'm thinking what's commonly known as the "Mexican Hat Wavelet" might be a good place to start. Each cycle should have a precisely defined amplitude, and it's the sort of signal I think is the culprit behind the errors I see in my differential I/O comparisons. Google will show the formula, which I've no idea how to type here directly.
planet10 said:
Dave,
discontinuities at zero (step change of derivative) have infinite frequency spectrum.
PMA said:
Another way of saying the same thing (at least in the mathematical sense)... in a practical application we won't get infinite thou...
dave
- Status
- Not open for further replies.