Passives: HF first to clip?

[popej]

I wonder what it is "extracting"? Evidently extracted 1kHz still contain 50Hz component. Probably clipped

 

[wwenze]

but I do know amps can clip either due to a lack of current capacity, or from a rail voltage limitation

They do... when their capacitor is damaged

Speaker-out from an actual amp.

You could say that the envelop you got has something similar

But of course, this shouldn't be the case for a strong amp or preamp driving no load, there should be way too much spare capacity for this to happen.

Rather, the shape of the envelope makes me think there is probably something else going on. Perhaps some limiter or soft-clipping circuit. Which counts as an active circuit, or some would call it non-linear. Whatever the case, most of the (simplified) theories rely on the system being linear time-invariant. And this circuit is not linear time-invariant, because the amount of clipping does not depend on just the signal value, but rather also on the previous signal value, apparently. Which means the value at t=0 depends on what happened at t= -1, so it's no longer time-invariant. When this happens, all the laws of superposition and what not, you can throw them out of the window.

How was the signal summed and re-measured anyway?
EDIT: Ok, I see post #18.
And the simplest answer is, that sum-er is not doing simple clipping. That's all. You can probably even just throw a square wave / step response at it and see what happens.
EDIT2: Also why would the "clipping sum" be clipping anyway when both your sines are -6dB (usually referring to peak)
EDIT3: Ok, so "clipping sum" is outputting 0dB, but there's a +3dB somewhere down the line resulting in the final digital ch1 being +3dB. Altho I can't actually see where that +3dB is from

Also, I should point out the clipping light is already on at "digital", before the signal left the soundcard. Or before it even entered the DAC.

Clipping itself is just a defined transfer function, or at least the one that everybody thinks of when nothing else is specified.
If you get an output that is not expected of a transfer function, it just means the sum-er or clipper is using a different transfer function.

 

[wwenze]

I wonder what it is "extracting"? Evidently extracted 1kHz still contain 50Hz component. Probably clipped

Yea, the extraction process probably went like "This is wave A, extract 50Hz from it". Then of course the extracted wave would be a clean 50Hz.

"But we don't have enough 50Hz because some of it was clipped"

"Take it from the residual 1000Hz wave, nobody will notice"

Hold on, don't tell me all the mathematical question is just a result of confusing the meaning of a function and a measurement, further complicated by using (clipping) functions that we don't know the formula of

 

[gnarly]

Notice your clipping is not at a fixed value which is what is expected or rather defined as clipping.

Meaning the mathematical function of the clipping is not "if value above >1, set to 1", but rather something else.

How are you generating the combined wave? For all we know the equipment could be doing "If ( channel 1 + channel 2 ) > 1, clip channel 2"

The combined wave is per #18....two sine generators being summed ala mixer, and sent to processors DAC.
My goal has been to get the DAC to have analog overload, simply running out of output voltage, but I'm not sure if digital clipping is occurring.
That's really the only uncertainty about the experiment bugging me right now.

I wonder what it is "extracting"? Evidently extracted 1kHz still contain 50Hz component. Probably clipped

The 'extraction' worked like this: the 50Hz &1000Hz summation that is clipped, was sent via XLR to mixer. So was the 50Hz unclipped signal.
In the external mixer, the 50Hz was inverted and summed with the clipped summation.....netting the clipped 1000Hz.

But of course, this shouldn't be the case for a strong amp or preamp driving no load, there should be way too much spare capacity for this to happen.

Even preamps have rail limits, I think the Core110f that is generating the clipped summation, is metering at the output+27dbu.
Had to put inline input pads on the measuring RME soundcard in #18, and then move them to the inputs of processor/mixer used in #19, to feed back to measuring soundcard.

EDIT2: Also why would the "clipping sum" be clipping anyway when both your sines are -6dB (usually referring to peak) so when added together they should be 0dB peak which is not clipping. This is giving more suspicion that the sum-er is starting to clip stuff near but not yet reached 0dB.

I've been trying to catch the onset of clipping, hence -6dB. I'll trying moving deeper into clipping.

 

[Chr1]

Slightly OT, but I am curious why most home hifi amps don't have clipping indicators. Presumably they wouldn't cost much to implement?
My theory, which admittedly is groundless (and cynical,) is that they would actually prefer you not to know your/their amp is struggling. Thoughts?

 

[wwenze]

You're calling this the "clipped 1000Hz" signal, right?

But notice the shape of the clipping is clearly following a 50Hz shape. This is where the 50Hz is missing from.

Below I've generated a similar waveform. It's not exact, because my clipping is horizontal, not diagonal.

Run a FFT on it, and, boom

Notice it contains 50Hz and harmonics.

Interestingly tho, the 3rd harmonic of both 50Hz and 1kHz are both at -34dB in Audacity. Which suggests that the harmonic content generated by clipping is the same for both the 50Hz and 1kHz. Granted, this wasn't a very clean cut and Audacity FFT isn't precise enough to match mathematical calculations. (I could use Arta, but, meh.)

I'm going to withhold commenting on the amplitude of the fundamental, because the value of the fundamental has been incorrectly extracted. So we'll have to settle for the value of the 3rd harmonic.

 

[gnarly]

You're calling this the "clipped 1000Hz" signal, right?

But notice the shape of the clipping is clearly following a 50Hz shape. This is where the 50Hz is missing from.

Yes, that is the clipped 1000Hz. And yes again, the clipping has the period or the 50Hz sine.
Here is again with time cursors.

What i think it is, is clipping due to riding on TOP of the 50Hz ...superposition again.
Riding on top of the 50Hz sine pushes the 1000Hz into clipping zone, thus showing the depressions in actual 1000Hz output.

Here's the 1000Hz clipped material shown positively.

Anyway I've looked at it so far, the 50Hz component still continues to appear to stay unchanged in level, while the 1000Hz clips

 

[wwenze]

50Hz appears unchanged because

The 'extraction' worked like this: the 50Hz &1000Hz summation that is clipped, was sent via XLR to mixer. So was the 50Hz unclipped signal.
In the external mixer, the 50Hz was inverted and summed with the clipped summation.....netting the clipped 1000Hz.

How about performing the opposite experiment? Send the 50+1000 summation. And a 1000Hz unclipped signal.
In the external mixer, the 1000Hz is inverted and summed with the clipped summation. You will get what you would call the clipped 50Hz. Which would be a 50Hz wave with 1000Hz components carved out of it. Just like the current "1000Hz clipped signal" has 50Hz carved out of it.

 

[gnarly]

50Hz appears unchanged because



How about performing the opposite experiment? Send the 50+1000 summation. And a 1000Hz unclipped signal.
In the external mixer, the 1000Hz is inverted and summed with the clipped summation. You will get what you would call the clipped 50Hz. Which would be a 50Hz wave with 1000Hz components carved out of it. Just like the current "1000Hz clipped signal" has 50Hz carved out of it.

Good idea, ......and yes, you are right about the carve out.

 

[danadam]

The combined wave is per #18....two sine generators being summed ala mixer, and sent to processors DAC.

So you sum them still in the digital domain? Then of course you'll get digital clipping. Even if the mixer works in floats, the DAC will only accept integers, so you'll get clipping during float->integer conversion.

 

[antcollinet]

Yes, it is the sum of the sine waves that clips. That is what i meant to be saying. Superposition of frequencies has the higher frequency sine riding on the lower.
Seems to me that always puts higher frequency content at the peak voltage .of any sine summations.

Here's a 50Hz and 1000Hz scope shot of what I'm trying to convey.
Blue trace is 50Hz alone
Yellow is 1000Hz alone
Green is their summation.

Summation is of course a higher voltage than either single sine on it's own, but note how the 1000Hz sine rides on top the 50Hz sine. And that superposition is what increases the summed voltage....and the increase in voltage is all 1000Hz sine content.

This example starts with equal voltage sines. Clearly if gain is raised to both, the 1000Hz sine will be the first to hit an amps rail voltage limit.
So let me move on to unequal voltage sines..

View attachment 509281




Here I've 4X the voltage of the 50Hz sine, leaving the 1000Hz unchanged. Equivalent to +12dB SPL @ 50Hz...hey my sub needed some boost, lol !
1000Hz still rides superimposed, and will be the first to hit the rails, and clip.

To the extent music waveforms are a collection of sine waves ala Fourier, seems to me higher frequency content clips first, as amp gain is raised.

View attachment 509282

The mistake you are making is with your waveform visualisations with a low frequency and the high frequency "on top" while showing the LF as a "separate" entity.

If the peaks clip you don't only hit the high frequency - since the lower frequency doesn't exist as a separate sine wave as you have shown it, such that the high frequency clips and the low frequency remains unaffected. Both the low and high frequency are encapsulated in the single time domain wave form. As soon as it clips, you are clipping the low frequency content as well, and creating a spray of distortion tones throughout the bandwidth.

And your third waveform (yellow) that you describe as the extracted HF - is what the combined clipped waveform should look like. Not the second - which I've no idea how you've interpreted as clipping.


And to really see what is going on you need a complete FFT of the clipped waveform.

 

[gnarly]

Ok, thx everyone, particularly @wwenze

2 mistakes I see I was making...
first, making experiments that upon reflection, were designed to give the results I had in mind..
second, not seeing that what is clipping is the waveform summation, and not separable once summed.

thx again all, for bearing with me...mark

 

[valerianf]

As we now know that both 50Hz and 1000Hz signals are damaged (clipped), the interesting question is which one has the greatest distortion?
Usually, in music, the larger amplitude is in the low frequency area (50 Hz in this example).
But the human ear is more sensitive for the 1 kHz frequency.
Could I guess that at the end we are hearing the 1 kHz with more distortion when the signal is clipping because of the large 50 Hz amplitude?
If it is the case I fully understand why many system are using a separated subwoofer with a separated amplification from the medium and high frequencies channel.
The distortion from the bass channel will not pollute the medium and treble one.
It is one advantage of a bi-amplification.

 

[antcollinet]

It is one advantage of a bi-amplification.

Only works if the crossover (frequency separation) is done upstream of the amplifier. IE Active crossovers.

Leave the crossover in the speaker - or otherwise downstream of amplification, then there is no benefit.

 

[gnarly]

Hi guys, sorry to do another about face. The last scope grap I posted showing "1000Hz teeth" in the 50Hz sine #29, had me scratching my head enough to make me think I was making mistakes.
In another of my proaudio measurement training classes, I had the opportunity to again ask about superposition, and if/how it effects amp clipping. And the answer I got was i was thinking straight .....in line with the prior series of posts and scope grabs.

It was explained that in the case of two sine waves, like the 50Hz and 1000Hz example I've been using, if they have equal amplitude, the 1000Hz will ride on top of the 50Hz, and show a +6dB voltage gain relative to either sine voltage singly.
When amp output is raised into the onset of clipping, the +6dB superposition is first to clip where the 1000Hz is riding on the crests of the 50Hz sine.
And the summed signal will be the only signal to clip until amp gain is raised further begin to clip both the 50Hz sine and the 1000Hz sine regardless of summation.
Or more simply said, at the onset of clipping, and until the 50Hz would clip on its own, the 50Hz sine is untouched.

It was also explained the reason this is not more widely known, is that in music it's not likely to encounter hf content with equal amplitude to low. So an equal amplitude test like the two sines i've been using is a bit of an extreme case.
However, no matter the ratio of amplitudes, higher frequencies are superimposed on lower frequencies, and will be the first to clip within the range of the onset of clipping into full clipping saturation of all waves. Music, multitone, two sines, whatever....

Ok, the guys convinced me I had been understanding correctly, but for further proof I came up with a better test bed.
All analog this time, real scope, and no soundcard. Same 50Hz and 1000Hz equal amplitude.

Arb function generator into XONE 464 analog mixer.
Each sine a little short of clipping mixer's output on its own.
Yellow: Ch1 Summed output.
Blue: Ch2 50Hz output alone.

I could go thru subtracting the sines from summed response again, as a further test. But it seems like a why bother.
I'm told the "1000Hz teeth" in the 50Hz sine (#29 again) is exactly what to expect and is more demonstration of superposition.

Bottom line imo, it does seem to be one of the cases where the time vs frequency resolution tradeoff in out FFT/IFT measurements,
doesn't allow FFT to paint the full picture, apparently because it's not a constant state condition.

 

[NTK]

...
However, no matter the ratio of amplitudes, higher frequencies are superimposed on lower frequencies, and will be the first to clip within the range of the onset of clipping into full clipping saturation of all waves. Music, multitone, two sines, whatever....
...

Clipping is a nonlinear phenomenon, e.g. 2clip(x) ≠ clip(2x). Superposition does not apply.

 

[wwenze]

if they have equal amplitude, the 1000Hz will ride on top of the 50Hz,

Don't think of it this way. The two waves are not independent when they are clipping. Neither of the two waves will come out unscathed.

BTW here is probably a good place to point out why the "subtraction method" is flawed fundamentally and made you start this thread to begin with. Because you subtract assuming wave A is unchanged. Then of course the subtracted result will show you what remains if wave A is unchanged, "thus proving wave A is unchanged" <-- incorrect conclusion because this is not a measured evidence, but an arbitrary calculated result; you did not measure wave A. And the evidence of the mistake is that the remaining wave B contains wave A content of negative polarity.

doesn't allow FFT to paint the full picture, apparently because it's not a constant state condition.

The word you're looking for is "periodic" - FT requires the signal to be periodic.

If the measurement window is large enough that the signal is repeating itself multiple times within the window, FFT paints the full picture.

BTW this is how Amirm's 32-tone IMD waveform looks like

And this is the FFT that comes from it

Review and Measurements of Gustard A20H DAC & AMP

This is a review and detailed measurements of Gustard A20H DAC and headphone amplifier. It is on kind loan from a forum member. It retails for USD $850 on Amazon with free shipping (from China?). I am accelerating this review because it is on sale on massdrop for USD $700. Still not cheap...

www.audiosciencereview.com

If FFT can get 32 frequency peaks out from that scrambled mess, rest assured it can "constant state condition" a waveform with 2 frequencies.

Now I'll show you how to do this properly... start with your clipped waveform of 2 components (1:1 in my case)

Next, top waveform, apply low pass filter to filter away the 1kHz content, leaving only the 50Hz and their harmonics. Example, corner frequency of 250Hz so we can see till the 5th harmonic.
Bottom waveform, apply high pass filter, same settings, so it is equivalent to the subtracted waveform

Notice something? The top wave is now an imperfect 50Hz wave with an amplitude lower than 0.5, while the bottom wave is an imperfect 1kHz wave with an amplitude below 0.5, if you average it. Interesting thing to note, it kind of looks like it has a lower frequency wobble, however the frequency of that is not 50Hz from both time domain and frequency domain observations. The answer lies in the FFT of the second wave, showing the next highest peaks are not the 2nd harmonic, but rather have a frequency value of 900Hz and 1.1kHz - These are IMD, as a result of the 50Hz wave having effect on the transfer function of the 1kHz wave, or in other words the effect of 50Hz causing 1kHz to clip that generated these frequencies as a result of modulation. In theory the 50Hz would see it too, but the IMD frequency is an integer ± combination of the two original frequencies, so it got LPFed out of existence.

 

[gnarly]

Another round of thanks to you, wwenze!

Anyway, I had already caught the logic problem with subtraction, which led to the idea of metering the two sides of the summation, post a hpf and lpf..
I ran the mixers summed output to another processor, where I used brick wall high-pass and low-pass (coincidentally choosing a 250Hz xover frequency ) to feed RMS and Pk meters on both sides.
50Hz had a reduction in level from straight feed vs summed, just like the 1000Hz. Just as you've been saying...

Here's the two sides sent to scope post hp and lp. The yellow cursors were the 50Hz voltage prior to summation & clipping.
Looks similar/familiar, huh?

Between you and my measurement class, I'm think I'm finally going to get this fully sorted. The measurement class is international, and language issues can sometimes form misleading conclusions. thx again

 

[antcollinet]

Between you and my measurement class, I'm think I'm finally going to get this fully sorted. The measurement class is international, and language issues can sometimes form misleading conclusions. thx again

You may actually also find the teachers don't fully understand what is going on either. As an alternative way of getting your head around it. If you think of the low pass filter to extract the 50Hz, what it will do is essentially draw a line through the mid points of the high freqency modulated waveform. When you cut off the top and bottom peaks of that waveform, the midpoint shifts also.

@wwenze I was just about to set up my ultralight and demo with real time signal generator and analysis in REW, but you've already nailed it and saved me the effort. Thanks.

 

[gnarly]

You may actually also find the teachers don't fully understand what is going on either. As an alternative way of getting your head around it. If you think of the low pass filter to extract the 50Hz, what it will do is essentially draw a line through the mid points of the high freqency modulated waveform. When you cut off the top and bottom peaks of that waveform, the midpoint shifts also.

Yeah, that's what I'd be thinking too, that my teacher(s) don't fully understand either....based on the posts I've been making....and then what you guys have helped me to see.
But the lack of understanding is all on me. The classes are primarily for the purpose of providing training on a specific measurement and simulation software package designed for large scale PA. Not beginner classes at all....maybe I should get kicked out Lol
So topics/question like this thread are a bit of a sidetrack, and I'm junior enough to not want to take up too much time...which can leave me with false starts in understanding.

Thx again for helping fill in my blanks