Gibb's effect.
Are you referring to some form of overshoot after the analogue filters?
Have you seen audio or music signals that can create sufficient overshoot in the analogue domain that the maximum specified output voltage peak can be exceeded?
By what percentage does this Gibb's effect (overshoot) exceed that specified peak?
Are you referring to some form of overshoot after the analogue filters?
Have you seen audio or music signals that can create sufficient overshoot in the analogue domain that the maximum specified output voltage peak can be exceeded?
By what percentage does this Gibb's effect (overshoot) exceed that specified peak?
Yes, that would be my question too. It was discussed earlier in the thread.
E.G.: If a DAC has a peak analog signal output of 2 volts on a full value sine wave, will we ever see voltages higher than 2 volts on the analog output? If so, how much? And if so, under what conditions? Is this simply ringing, as AndrewT mentions above, or something inherent in certain signals?
E.G.: If a DAC has a peak analog signal output of 2 volts on a full value sine wave, will we ever see voltages higher than 2 volts on the analog output? If so, how much? And if so, under what conditions? Is this simply ringing, as AndrewT mentions above, or something inherent in certain signals?
What I describe in #717 are all valid waveforms. I can generate a swept sine at -6dBFS and play it back through sound card (E-MU 0404 USB) using no digital attenuation. Output is fed to input and analog output pot and microphone pre amplifier are set so monitor in software gives -6dBFS. Recorded result is exactly as expected, peak samples are -6dBFS. Without changing any settings, MLS wave with -6dBFS sample values is played back; the input signal is clipped. Digitally output is lowered 3dB, and now monitor peaks at -1.2dB. Therefore, output peaks of MLS signal crest 10.2dB above swept sine signal.
Overshoot is in the analog filters, as it must for Gibbs and Nyquist sampling theorem.
Regards,
Andrew
Overshoot is in the analog filters, as it must for Gibbs and Nyquist sampling theorem.
Regards,
Andrew
Andrew Barleywater:
I was cusrious about the overshoot, so just took a look at a DAC that was sitting here in the bench, an M-Audio Fast Treack Pro. I looked at the RCA outputs on the o'scope terminated with 22K.
On square and sawtooth waves there is definite overshoot by about 1.6dB that does not show on sine waves.
Looking at a few music tracks some show no overshoot at all (as far as I can tell) but some have consistent and regular overshoot. Strings recorded very hot, for example. There I can see many overshoots of 1dB, some a little higher. I don't know if they come from the DAC chip or the following circuit, but it doesn't matter in practice, they are there. It will be interesting to see what other DACs do on overshoot. But I won't have time to test that for another week or so.
So maybe I need to revise the test a bit. When measuring the -12dB tone, peaks could be as high as 13dB, and maybe 14dB more. Leaving yourself a few dB headroom is always a good idea. For 35% of those who responded, that would mean a 40WPC (@ 8 ohms) amp. For another 38%, that means they need at least a 6WPC amp. The remaining 25% need amps with high output voltages to play as loud as they like.
I was cusrious about the overshoot, so just took a look at a DAC that was sitting here in the bench, an M-Audio Fast Treack Pro. I looked at the RCA outputs on the o'scope terminated with 22K.
On square and sawtooth waves there is definite overshoot by about 1.6dB that does not show on sine waves.
Looking at a few music tracks some show no overshoot at all (as far as I can tell) but some have consistent and regular overshoot. Strings recorded very hot, for example. There I can see many overshoots of 1dB, some a little higher. I don't know if they come from the DAC chip or the following circuit, but it doesn't matter in practice, they are there. It will be interesting to see what other DACs do on overshoot. But I won't have time to test that for another week or so.
So maybe I need to revise the test a bit. When measuring the -12dB tone, peaks could be as high as 13dB, and maybe 14dB more. Leaving yourself a few dB headroom is always a good idea. For 35% of those who responded, that would mean a 40WPC (@ 8 ohms) amp. For another 38%, that means they need at least a 6WPC amp. The remaining 25% need amps with high output voltages to play as loud as they like.
I also ran MLS signals and it depends on the peak value (tops of the waves) how much the signal goes past the 0dB output. If the flat tops are at 0dBFS, then they show on the scope exactly where a square wave does. Maximum output seems limited. There is some overshoot, same on MLS as Sqaure.
However, if the flat tops are digitally reduced 6dB, the the MLS signal does show as a bit higher than a normal square wave with the top at the same value. About 6dB higher as far as I can tell.
However, if the flat tops are digitally reduced 6dB, the the MLS signal does show as a bit higher than a normal square wave with the top at the same value. About 6dB higher as far as I can tell.
I haven't run Pano's test, but have used the method in various forms; what I described in #723 is the exact same scenario.
There are very detailed papers on MLS, and swept sine techniques for measuring impulse response of a given system. The math has existed since long before digital audio. Current computing power, software, and of course the web make all of this accessible.
In a nutshell: A single sample central to a long series of zero value samples is set to a nonzero value. In frequency domain this signal has perfectly flat spectrum from DC to fs/2. When passed through a system the resultant output is no longer a single sample (unless the system has perfect transmission). The resultant signal is the impulse response of the system. Energy delivered by the single sample impulse is very small, and resultant signal to noise ratio is typically poor.
This is overcome by MLS; it looks and sounds like white noise, but it is not, it is mathematically derived series with lengths going as (2^n)-1, and has further property that autocorrelation of the sequence results in a single value ((2^n)-1)/2^n and all other values -1/(2^n). Bottom line: vast improvement of S/N.
I'll work on getting a thread up on MLS and Exponential Swept Sine measurement techniques, with practical demonstrations and references.
Regards,
Andrew
You're welcome. I always measure when I can. Understanding the results isn't always easy, tho.
MLS is another subject and does belong in another thread. DAC peaking is relevant to this thread. I do see peaking out of my M-Audio DAC, but I also seem to see more output a high frequencies, at least on the o'scope. Will have to test further. next week.
I still see this test as valid. Play the test tone(s) at your loudest setting and measure. You now know what your peak voltage will be on digital playback.
I'm interested to run my old Shure tracking test LP in an attempt to find the maximum levels out of my cart+ phono preamp. I think it has some tracks recorded about as loud as any LP would be.
Pano, thanks for this fantastic thread. I applaud the simplicity of the test.
I have a pair of Pensil 7s, Mark Audio Alpair 7s in a mass loaded transmission line variant. There is a thread on it here in the Mark Audio section. My room is about 4 metres, by 4.5 metres. Difficult acoustically. Anyway.
I set the system level according to the loudest I felt comfortable running it, about 2-3 clicks on the volume control above where I would usually listen. (I would listen at 12 oclock, I had it set at about 2 oclock-ish, its a passive preamp)
On the test tone I measured 2.22V.
As I understand it this is 9dB below full scale digital right?
Hence my potential maximum loudspeaker voltage when driven from my digital source at this volume setting would be 2.22 x 2.818 = 6.256V. This is equivalent (in 8 Ohm terms) to a 4.89W amplifier? (V squared / R load)
Do I have that right?
I have a pair of Pensil 7s, Mark Audio Alpair 7s in a mass loaded transmission line variant. There is a thread on it here in the Mark Audio section. My room is about 4 metres, by 4.5 metres. Difficult acoustically. Anyway.
I set the system level according to the loudest I felt comfortable running it, about 2-3 clicks on the volume control above where I would usually listen. (I would listen at 12 oclock, I had it set at about 2 oclock-ish, its a passive preamp)
On the test tone I measured 2.22V.
As I understand it this is 9dB below full scale digital right?
Hence my potential maximum loudspeaker voltage when driven from my digital source at this volume setting would be 2.22 x 2.818 = 6.256V. This is equivalent (in 8 Ohm terms) to a 4.89W amplifier? (V squared / R load)
Do I have that right?
Last edited:
You are quite welcome. My hope is that this test will allow fellow DIYers to know how much voltage they need to run their speakers very loud. It helps to know how much amplification you need and helps to set your gain structure.
9dB below a full value sine wave, correct.
That looks right (but I've just returned from a night of bar hopping and heavy drinking, so math skills might not be the best).
Seems from your measurements that with the volume knob at 2:00 you would just be under clipping a 5 Watt amp. As to how well that amp might handle near clipping levels, that's another thread.
I need to change my vote to >20v.
Since I got to uni, my stereo has been used for flat parties a few times.
As with all parties, you crank it up, and keep going until the clipping lights are starting to come on.
Back it off slightly, and you're there.
Anyway, that's a 120w/ch amp that's been run clipping. With the 4ohm speakers i'm using, that's just over 20v.
Its an active system, but that's what the lowest sensitivity drivers (TB W6-1139) are soaking up. The Fostex FR drivers (run from 500Hz up) have a mere 10w/ch to play with, but drivers are ~10dB more sensitive than the TB woofers.
Haven't tested the subwoofer yet, but I suspect that's taking some power, too - ~28mm p/p travel on a 12" cone sure is impressive.
Chris
PS - normal listening levels are much lower - I don't think the people here would be too impressed if I ran my stereo flat out all the time.
Since I got to uni, my stereo has been used for flat parties a few times.
As with all parties, you crank it up, and keep going until the clipping lights are starting to come on.
Back it off slightly, and you're there.
Anyway, that's a 120w/ch amp that's been run clipping. With the 4ohm speakers i'm using, that's just over 20v.
Its an active system, but that's what the lowest sensitivity drivers (TB W6-1139) are soaking up. The Fostex FR drivers (run from 500Hz up) have a mere 10w/ch to play with, but drivers are ~10dB more sensitive than the TB woofers.
Haven't tested the subwoofer yet, but I suspect that's taking some power, too - ~28mm p/p travel on a 12" cone sure is impressive.
Chris
PS - normal listening levels are much lower - I don't think the people here would be too impressed if I ran my stereo flat out all the time.
It may be even more, if you are clipping an amp rated as 120W into 8 ohms. But remember, your vote is the not the maximum voltage, it's the voltage you measured on the test tone.
It's 9dB below the loudest sine wave value, or 12dB (RMS) below peak.
My guess is that you would measure around 12 volts on the test tone, playing at that volume setting. But that's just a guess based on your amplifier rating.
Badly..... 
That's why I also supplied a higher tone. I do like to keep the tones low so that most volt meters will read them with good accuracy.
http://www.diyaudio.com/forums/mult...ower-do-your-speakers-need-3.html#post2870958
That's why I also supplied a higher tone. I do like to keep the tones low so that most volt meters will read them with good accuracy.
http://www.diyaudio.com/forums/mult...ower-do-your-speakers-need-3.html#post2870958
Its rated for 120w/ch into 4ohm. It has 50v DC PSU rails, but I suspect there's some sag under continuous high current draw.
If I'm using 20v RMS on the loud bits (and party music is all loud), then that'd be 5v on the test-tone: 1/4 * voltage = -12dB.
I can't do the test right now, due to hungover flatmates. Will report back when I can. Probably on Wednesday.
Chris
PS - I tend to bring up the subwoofer level somewhat to get the nightclub feel. Next time its loud, I'll try measuring the voltage at the subwoofer terminals. The sub's hooked up to a bass guitar amp that also has 50v rails, and I think (with what looks like a 1kVA transformer) that'll do the full 300w into 4ohm.
PPS - the subwoofer is crossed 40Hz LR4, and the mains are 11L ported enclosured tuned to 40Hz, with a handful of stuffing in each port. They roll off gently a little above 60Hz. There's a severe standing wave at 63Hz in this room, so I try to avoid exciting that, hence the low subwoofer XO point.
Last edited:
Thanks Chris for the info and the clarification.
Also, for those who can't use the 120Hz or 220Hz tones for some reason, I put some sweeps here:
http://www.diyaudio.com/forums/mult...wer-do-your-speakers-need-14.html#post2881475
Also, for those who can't use the 120Hz or 220Hz tones for some reason, I put some sweeps here:
http://www.diyaudio.com/forums/mult...wer-do-your-speakers-need-14.html#post2881475