Folks, this is the situation:
Yes, you can synthesize test tones and even interesting effects, BUT you must pass through 'live music', or at least the next best thing, in order to evaluate what a fairly good quality op amp does to the music. Let us say that you have a 'perfect' CD source. Well, how do you play it back? It must be badwidth limited, simply because it is a CD and what about the op amps in the D-A stage of the CD player? They would add their own phase modulation, and potentially obscure any listening differences.
Yes, you can synthesize test tones and even interesting effects, BUT you must pass through 'live music', or at least the next best thing, in order to evaluate what a fairly good quality op amp does to the music. Let us say that you have a 'perfect' CD source. Well, how do you play it back? It must be badwidth limited, simply because it is a CD and what about the op amps in the D-A stage of the CD player? They would add their own phase modulation, and potentially obscure any listening differences.
MikeB
Yes, this is correct although the open-loop vs closed-loop have been discussed somewhat at the earlier pages of this thread...
As the frequency rises the "available bandwidth" for feedback decreases -> more distortion. On a similar note you can see that PSRR (of opamps for example) is dropping with frequency too (with PSSR- usually being the most sensitive due to the most common design before the output stage).
I can provide you with the mathematical background behind basic nonlinear distortion if you want to. If you know sine and cosine functions it really is very simple and very instructive as you clearly can see why THD & IMD are created.
While a 1 kHz THD measurement would probably not be very useful here (and IMO pretty useless for audio altogether unless applied to a speaker driver) – try one at 10 or 15 kHz (if your system is not bandwidth-limited to 20 kHz!).
Also, the standard CCIF (IMD test at 14 & 15 kHz) would probably yield pretty ugly numbers here!
As a next step you could cascade two nonlinear stages (say one with even order distortion and one with odd order, or just two similar ones). Now you will probably get really funky results! The math behind this pretty much just grows exponentially with the number of stages so playing around graphically (or playing the audio stream) is much more instructive!
Remember that real-world speakers have terrible distortion specs – especially the ”audiophile” crap where 5-10% isn’t uncommon. There is a reason most manufacturers don’t list the distortion specs but rather like to talk about their magic speaker wires or special green-painted resistors in their fancy pamflets.
The only speaker manufacturers I know that DO publish distortion plots are JBL, Beyma and Fane for their series of pro sound drivers. JBL:s generally have very low distortion, usually < 1% even at high SPL:s as a result of excellent and sound engineering. These are the drivers you will find if you come home to my place!
Yes, this is correct although the open-loop vs closed-loop have been discussed somewhat at the earlier pages of this thread...
As the frequency rises the "available bandwidth" for feedback decreases -> more distortion. On a similar note you can see that PSRR (of opamps for example) is dropping with frequency too (with PSSR- usually being the most sensitive due to the most common design before the output stage).I can provide you with the mathematical background behind basic nonlinear distortion if you want to. If you know sine and cosine functions it really is very simple and very instructive as you clearly can see why THD & IMD are created.
While a 1 kHz THD measurement would probably not be very useful here (and IMO pretty useless for audio altogether unless applied to a speaker driver) – try one at 10 or 15 kHz (if your system is not bandwidth-limited to 20 kHz!).
Also, the standard CCIF (IMD test at 14 & 15 kHz) would probably yield pretty ugly numbers here!
As a next step you could cascade two nonlinear stages (say one with even order distortion and one with odd order, or just two similar ones). Now you will probably get really funky results! The math behind this pretty much just grows exponentially with the number of stages so playing around graphically (or playing the audio stream) is much more instructive!
Remember that real-world speakers have terrible distortion specs – especially the ”audiophile” crap where 5-10% isn’t uncommon. There is a reason most manufacturers don’t list the distortion specs but rather like to talk about their magic speaker wires or special green-painted resistors in their fancy pamflets.
The only speaker manufacturers I know that DO publish distortion plots are JBL, Beyma and Fane for their series of pro sound drivers. JBL:s generally have very low distortion, usually < 1% even at high SPL:s as a result of excellent and sound engineering. These are the drivers you will find if you come home to my place!
Folks, I can do the live singing:
Take all your overgrown infants away somewhere
and build them a home
a little place on their own
the Chef's Memorial Home for incurable gurus
and kings
they can appear to themselves every day
on an open loop amp
to make sure they’re still inept
it's the only topology they'll accept
"ladies and gentlemen, please welcome curl and otala
mr sleewrate and nonlinear friend leinonen
mr evil phase shift and crap
the ghost of a black gate cap
and the memories of 741:s
and now adding colour a group of anonymous digital
-hating meat packing punters"
did they expect us to treat them with any respect?
they can polish their resistors and sharpen their
ears, and abuse themselves playing games through the years
pim pim, tim tim, opamps are bad!
safe in the permanent non double-blind bubble of high-end
their favourite toy
they'll be good girls and boys
In the Chef's Memorial Home for obscurable
wasters of bandwidth and amps
is everyone in?
are you having a nice time?
now the global feedback loop can be applied.
To be released on vinyl shortly.
Take all your overgrown infants away somewhere
and build them a home
a little place on their own
the Chef's Memorial Home for incurable gurus
and kings
they can appear to themselves every day
on an open loop amp
to make sure they’re still inept
it's the only topology they'll accept
"ladies and gentlemen, please welcome curl and otala
mr sleewrate and nonlinear friend leinonen
mr evil phase shift and crap
the ghost of a black gate cap
and the memories of 741:s
and now adding colour a group of anonymous digital
-hating meat packing punters"
did they expect us to treat them with any respect?
they can polish their resistors and sharpen their
ears, and abuse themselves playing games through the years
pim pim, tim tim, opamps are bad!
safe in the permanent non double-blind bubble of high-end
their favourite toy
they'll be good girls and boys
In the Chef's Memorial Home for obscurable
wasters of bandwidth and amps
is everyone in?
are you having a nice time?
now the global feedback loop can be applied.
To be released on vinyl shortly.
Hi swedish chef !
I think you did not fully understood me... Of course feedback decreases
if openloopgain drops with higher freqs. I wanted to show that this
enables the "full" openloopdistortion at output even with frequencies
within the openloopbandwidth !
Okay, i did my program applying distortions to a wave. (i was already
familiar with the basical mathematic for simple unlinear distortions)
The results are very interesting and explain why 3rd harmonic is EVIL !
I took a wavefile ripped from cd, and applied 4 different distortions,
all with about 0.6%thd for a single sinewave.
Here the results:
2nd harmonic only: THD: 0.62% -> 0.03%
3rd harmonic only: THD: 0.62% -> 2%
2nd+3rd harmonic: THD: 0.65% -> 0.3%
tanh distortion: THD: 0.6% -> 0.6%
for 2nd harmonic i use a simple y = x*(1-n) + x^2*n
for 2nd+3rd i use y = x*(1-n) + x^3*n
for 3rd only, too complex, but based on 2nd+3rd, 2nd cancelled out,
very identical to diffamp-behaviour.
And tanh speaks for itself...
All thd values were calculated with FFT. Of course these values vary with different music.
I think the first 2 numbers speak for themself, not surprising, but
explains some things !
About thd from speakers...
I made measurements with my diy-speakers, see attached pic !
The most spikes are from roomacoustics...
Observe 3rd harmonic from tweeter ! (Crossover at 2.5khz)
MikeB
I think you did not fully understood me... Of course feedback decreases
if openloopgain drops with higher freqs. I wanted to show that this
enables the "full" openloopdistortion at output even with frequencies
within the openloopbandwidth !
Okay, i did my program applying distortions to a wave. (i was already
familiar with the basical mathematic for simple unlinear distortions)
The results are very interesting and explain why 3rd harmonic is EVIL !
I took a wavefile ripped from cd, and applied 4 different distortions,
all with about 0.6%thd for a single sinewave.
Here the results:
2nd harmonic only: THD: 0.62% -> 0.03%
3rd harmonic only: THD: 0.62% -> 2%
2nd+3rd harmonic: THD: 0.65% -> 0.3%
tanh distortion: THD: 0.6% -> 0.6%
for 2nd harmonic i use a simple y = x*(1-n) + x^2*n
for 2nd+3rd i use y = x*(1-n) + x^3*n
for 3rd only, too complex, but based on 2nd+3rd, 2nd cancelled out,
very identical to diffamp-behaviour.
And tanh speaks for itself...
All thd values were calculated with FFT. Of course these values vary with different music.
I think the first 2 numbers speak for themself, not surprising, but
explains some things !
About thd from speakers...
I made measurements with my diy-speakers, see attached pic !
The most spikes are from roomacoustics...
Observe 3rd harmonic from tweeter ! (Crossover at 2.5khz)
MikeB
Attachments
Folks I just ran a test measuring harmonic distortion in a sine wave generated by a function generator. I also was able to take a simultaneous fft measurement as well as a thd measurement. Then I added FM modulation to the sine wave. It could be easily seen on the fft, and the oscilloscope tracing instability, BUT it took a lot of FM to change the thd measurement. This means that the thd test is relatively insensitive to FM modulation. So there!
John,
Was the modulation freq = the carrier freq? If the modulation freq is low, the modulation sidebands might be getting notched out if there's a notch filter for the fundamental in the distortion analyzer. And the case we're looking at is really the carrier modulating its own phase, right?
Was the modulation freq = the carrier freq? If the modulation freq is low, the modulation sidebands might be getting notched out if there's a notch filter for the fundamental in the distortion analyzer. And the case we're looking at is really the carrier modulating its own phase, right?
Andy C, I could use some help here, as I have never studied FM modulation formally. However, what I did find was that at a large range of modulation frequencies, I could get results on the FFT, but not with THD. This makes sense, since the THD notch is what can't follow the change in frequency.
MikeB, interesting plots. I would not have first thought the distortion quite so low, but I do believe your plots.
MikeB, interesting plots. I would not have first thought the distortion quite so low, but I do believe your plots.
The sideband amplitudes are purely a function of the modulation index x, where x = delta_f / f_mod. They involve a bunch of Bessel functions that I won't get into here unless you need for more info.
delta_f = peak frequency deviation
f_mod = frequency of the modulating signal (modulation input of generator)
Normally the modulation input of these signal generators produces a peak frequency deviation proportional to the peak voltage deviation at the mdulation input. So as your modulation frequency goes up, assuming a constant amplitude of the modulating signal, the modulation index will go down proportional to modulation frequency.
The modulation sideband frequencies are the carrier plus and minus integer multiples of the modulation frequency. As the modulation index gets large, the higher-order sidebands become higher in amplitude. As the modulation index becomes very small, the f_carrier +/- f_mod will be the only sidebands seen.
So if you've got a very low modulation frequency, you'll potentially have a large modulation index, giving lots of sidebands. Yet these sidebands could be very close in to the carrier, and many could be notched out by the notch filter in the THD analyzer if it has such a filter. So I could see in this situation things could get pretty wobbly, and yet the THD analyzer might not register much in the way of distortion.
delta_f = peak frequency deviation
f_mod = frequency of the modulating signal (modulation input of generator)
Normally the modulation input of these signal generators produces a peak frequency deviation proportional to the peak voltage deviation at the mdulation input. So as your modulation frequency goes up, assuming a constant amplitude of the modulating signal, the modulation index will go down proportional to modulation frequency.
The modulation sideband frequencies are the carrier plus and minus integer multiples of the modulation frequency. As the modulation index gets large, the higher-order sidebands become higher in amplitude. As the modulation index becomes very small, the f_carrier +/- f_mod will be the only sidebands seen.
So if you've got a very low modulation frequency, you'll potentially have a large modulation index, giving lots of sidebands. Yet these sidebands could be very close in to the carrier, and many could be notched out by the notch filter in the THD analyzer if it has such a filter. So I could see in this situation things could get pretty wobbly, and yet the THD analyzer might not register much in the way of distortion.
Opamps are mady by big manufacturer. Opamps are practical, and have big market. Why the manufacturers do not recognize the "bad side" of audio opamp and make such a simple and good audio opamp inside those plastic case? The ones that is said "For Audio" like OPA2134 seems not sounding good too.
Lumanauw, you have it right! It is NOT that we dislike the concept of op amps. I have worked with op amps since they were first designed, 40 years ago. I have ua709's that are 38 years old, in my lab stock. I have used many hundreds of IC op amps in designs, and I test almost every new IC op amp design as they become available. I can still build discrete circuits that are essentially op amp based, that work and sound better than any op amp that I have tested. Why? The only factor that is obvious is the relatively low open loop bandwidth of most of the IC op amps. This is why we are addressing this issue.
If manufacturers can offer a truly superior IC op amp that is not compromised compared to discrete designs, we will use them, almost exclusively, and only use discrete designs for special applications. We want this, as it would save money, time, and space for us, and cost reduction for our customers. So far, even the OPA134 and AD797 are just OK, not quite what we can make ourselves in specialized situations such as buffers and line stages.
If manufacturers can offer a truly superior IC op amp that is not compromised compared to discrete designs, we will use them, almost exclusively, and only use discrete designs for special applications. We want this, as it would save money, time, and space for us, and cost reduction for our customers. So far, even the OPA134 and AD797 are just OK, not quite what we can make ourselves in specialized situations such as buffers and line stages.
Hi all,
I am not sure whether this is on or off topic, but it HAS to do with belief. It is something I came across today which I think is interesting.
Source: International Herald Tribune, Science page (10), 21 Oct 04.
This articles quotes research by is a team on the shaping of preferences by cultural messages.
First they do a "blind taste" with 67 subjects. They are asked for their preference between two brands of Coke, Coke Classic and Pepsi. The tasting, as said, is blind. There is no indication which is which. Preferences are about equally split between the two products.
Next they do a sighted test, where the subjects can actually see which brand they are tasting. Result: 3 out of 4 prefer Coke Classic....
Now, the results as such are not mentioned as particularly interesting. It is a given that the blind vs sighted can and often will produce very different results. What was interesting is that during the tests, brain activity was monitored. The main difference was that in the sighted test, brain areas believed to be involved in memories and brand recognition lighted up in the scanners. Follow-on research suggested that in shaping preferences, brain areas involved in memory, decision making and self-image play important roles.
Now, I will be the first to agree that listening to audio is not the same as tasting Coke. But these types of tests results fit very well with other research in perception and preference shaping, including audio.
Jan Didden
I am not sure whether this is on or off topic, but it HAS to do with belief. It is something I came across today which I think is interesting.
Source: International Herald Tribune, Science page (10), 21 Oct 04.
This articles quotes research by is a team on the shaping of preferences by cultural messages.
First they do a "blind taste" with 67 subjects. They are asked for their preference between two brands of Coke, Coke Classic and Pepsi. The tasting, as said, is blind. There is no indication which is which. Preferences are about equally split between the two products.
Next they do a sighted test, where the subjects can actually see which brand they are tasting. Result: 3 out of 4 prefer Coke Classic....
Now, the results as such are not mentioned as particularly interesting. It is a given that the blind vs sighted can and often will produce very different results. What was interesting is that during the tests, brain activity was monitored. The main difference was that in the sighted test, brain areas believed to be involved in memories and brand recognition lighted up in the scanners. Follow-on research suggested that in shaping preferences, brain areas involved in memory, decision making and self-image play important roles.
Now, I will be the first to agree that listening to audio is not the same as tasting Coke. But these types of tests results fit very well with other research in perception and preference shaping, including audio.
Jan Didden
MikeB,
I am still not sure I understand exactly what you have simulated but nevertheless... For the polynomial describing the simple nonlinear transfer function you should just set:
y(t) = a1*x(t) + a2*x^2(t) + a3*x^3(t) +...
and then alter the coefficients to model the particular behaviour you want.
And regarding the meausurements on your loudspeakers - do you have a ball park figure as to at what SPL they were recorded? The distortion seemed very low so either you have very capable speakers or the SPL were not that loud. Personally I think in the vicinity of 100 dB @ 2-4 m distance would be a starting point for a home system. Remember that classical music easily reaches 120 dB at loud passages. IIRC a full orchestra puts out something like 10 W of clean acoustic power and it would simply take a massive system to recreate that. But good work anyway!
And if you have understood the theory behind THD you can now figure out why THD is essentially useless for specifying a tweeter much above 6 kHz!
/Magnus
I am still not sure I understand exactly what you have simulated but nevertheless... For the polynomial describing the simple nonlinear transfer function you should just set:
y(t) = a1*x(t) + a2*x^2(t) + a3*x^3(t) +...
and then alter the coefficients to model the particular behaviour you want.
And regarding the meausurements on your loudspeakers - do you have a ball park figure as to at what SPL they were recorded? The distortion seemed very low so either you have very capable speakers or the SPL were not that loud. Personally I think in the vicinity of 100 dB @ 2-4 m distance would be a starting point for a home system. Remember that classical music easily reaches 120 dB at loud passages. IIRC a full orchestra puts out something like 10 W of clean acoustic power and it would simply take a massive system to recreate that. But good work anyway!
And if you have understood the theory behind THD you can now figure out why THD is essentially useless for specifying a tweeter much above 6 kHz!
/Magnus
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.