Hi,
Its an old thread but this "engineering rigour" has obviously been bugging
me for a quite awhile as it is simply so wrong and I cannot understand the
engineering "rigour" that blatantly ignores reality, what I know to be true.
I've only returned to this due to finding a free player called Foobar2000.
In this player you can monitor Peak levels, VU levels and spectrum levels.
Strangely the spectrum is split into 20 bands, with peak and average shown.
Use this player to play your favourite music with the above display options.
It will graphically illustrate the above is complete nonsense, as it always
has been as it ignores the frequency range allocated to each amplifier.
FWIW I've always the understood the maths, the above does not, but I'm
not interested in using maths to get people to understand, the above is a
classic example of using mathematical principles to get it very wrong.
What the above does not show is how to correctly allocate amplifier power
to each frequency band, as this depends on the bandwidth and spectrum.
/Sreten.
P.S. I'm not not having a pop at Andrew at all, just the common wrong idea.
Last edited:
Er, OK, but isn't it usual to make some other detailed suggestion as to what might be appropriate in circumstances where you're suggesting that somebody else's mathematical approach is flawed?
Bully for you.
w
Hi,
As i intimated this is not a new argument and Andrew knows me from the past.
The "reality" is not the issue, what is new is an easy way of seeing the reality.
Andrew admits in this case he disagrees with lots of other people.
As for the possible reality read the rest of the thread.
/Sreten.
Last edited:
Hi Andrew,
For this purpose it does not matter whether its a real CD or compressed.
MP3 etc does not change the large scale envelope or spectral content much.
Run the VU meter, Peak Meter and Spectrum Analyser together.
Imagine you've built a 20 way (not a good idea) active speaker and ponder
the consequences of what you can see. Most decent music looks very similar.
/Sreten.
I am not so sure.
There is so much compression in original CDs and much of that is to reduce the fast transients.
If we deliberately reduce the fast transients, will that skew the peak data? I think it might.
I'm thinking that all sources should be analysed, But I only have CD copies on my PC, so my data is already biased. That cannot give a reliable result in my book.
Listen to real snare and other non bass drumming and compare to the anemic version we hear from our stereo recordings.
I know they are not comparable.
Last edited:
One wonders - if you're not listening to media available to you, then what are you listening to?
Semantics aside this thread got me wondering about it so I turned on the listening oscilloscope to observe some peaks in some of the music that I have.
My system is biamplified, with an XO sitting at about 2.1kHz. Other vital statistics include: 6.5" woofer, 90dB (8ohms, claimed) and 1" tweeter, 89.7dB (4ohms, claimed).
Unfortunately this system uses a Linkwitz transform filter to compensate for the sealed box bass response. A correction factor of 2.5x was calculated from the 50Hz point of the theoretical filter transfer function. 50Hz was selected as this is where kick drums (and electronic varieties) produce the greatest amount of their power. There is probably a 1dB error at this point of the slope, should the frequency content be lower, however through spectral analysis I did on these songs before, the variation of the fundamental bass is no more than 8Hz off 50Hz.
An visual observation was made of the transient peaks. Sections of the song were repeated to confirm the transients.
The following raw values were observed:
Sample 1: 4Vpp woofer, 0.45Vpp tweeter.
Sample 2: 2.15Vpp woofer, 0.5Vpp tweeter.
The estimated corrected values are as follows:
Sample 1: 1.6Vpp/0.9Vpp
Sample 2: 0.86Vpp/1Vpp (!)
Sample 1 is odd, in that it actually has a substantial amount of content at 17-19kHz. This is because it actually contains in part a recording of a inkjet printer head moving around, which includes its PWM harmonic frequency (though due to encoding, not any of the harmonics beyond 20kHz). This did not cause the peak, however, the instantaneous taps of the head loading did. If you're wondering, it is Toner by Cornelius.
Sample 2 was just standard other music containing cymbals, electric guitars and drums.
So from these observations it would appear to support the idea that the voltage response needs to be roughly the same if the drivers were of the same impedance and sensitivity. But in my setup this would be a falsehood due to the bass correction and 4 ohm tweeter.
Semantics aside this thread got me wondering about it so I turned on the listening oscilloscope to observe some peaks in some of the music that I have.
My system is biamplified, with an XO sitting at about 2.1kHz. Other vital statistics include: 6.5" woofer, 90dB (8ohms, claimed) and 1" tweeter, 89.7dB (4ohms, claimed).
Unfortunately this system uses a Linkwitz transform filter to compensate for the sealed box bass response. A correction factor of 2.5x was calculated from the 50Hz point of the theoretical filter transfer function. 50Hz was selected as this is where kick drums (and electronic varieties) produce the greatest amount of their power. There is probably a 1dB error at this point of the slope, should the frequency content be lower, however through spectral analysis I did on these songs before, the variation of the fundamental bass is no more than 8Hz off 50Hz.
An visual observation was made of the transient peaks. Sections of the song were repeated to confirm the transients.
The following raw values were observed:
Sample 1: 4Vpp woofer, 0.45Vpp tweeter.
Sample 2: 2.15Vpp woofer, 0.5Vpp tweeter.
The estimated corrected values are as follows:
Sample 1: 1.6Vpp/0.9Vpp
Sample 2: 0.86Vpp/1Vpp (!)
Sample 1 is odd, in that it actually has a substantial amount of content at 17-19kHz. This is because it actually contains in part a recording of a inkjet printer head moving around, which includes its PWM harmonic frequency (though due to encoding, not any of the harmonics beyond 20kHz). This did not cause the peak, however, the instantaneous taps of the head loading did. If you're wondering, it is Toner by Cornelius.
Sample 2 was just standard other music containing cymbals, electric guitars and drums.
So from these observations it would appear to support the idea that the voltage response needs to be roughly the same if the drivers were of the same impedance and sensitivity. But in my setup this would be a falsehood due to the bass correction and 4 ohm tweeter.
Last edited:
Hi,
I quite agree, note I said decent music, e.g. the Trinity Sessions by the
Cowboy Junkies (live, one microphone, no compression). FWIW the CD
red book standard is capable of higher performance than any tape based
mastering machine, especially in the context of high frequency transients
where it literally pees all over any analogue tape system dynamically.
Classic vinyl records are limited by the mastering process, the tape machine.
In no way are vinyl records in any way dynamically or transiently more taxing
than a good digital recording to CD standards, this is simply the way it is.
Whilst I accept your points, a system nearly all the time plays back recorded
music, building it to do something it never has to do is not to me engineering.
/Sreten.
Hi,
For arguments sake lets say you built a twenty way active speaker
with twenty 1W amplifiers, split like to foobars 20 band display.
Further for arguments sake each of the twenty drivers is flat over
its required range and all the drivers have the same 90dB senstivity.
What is the nature of this beast ?
**********************************
Well maximum output at any single frequency is 1W or 90dB.
With a pink noise signal (equal energy per band) you might suppose
maximum output is 20W or 103dB but this ignores the crest factor or
mean to peak ratio for unclipped random pink noise, so somewhat lower
for average levels but essentially equivalent to 20W unclipped peaks.
Pink noise is a wideband uncorrelated signal, what about a wideband
correlated signal such that all amplifiers and drivers contribute
to a leading edge peak ? The rather surprising answer is off the
chart at a theoretical 400W 116dB unclipped wideband transient.
(Remember its Vsquared/R and in this case all voltages add).
Our 20 way beast is obviously impractical as its characteristics
bear no relation to the statistics of real music typically (stereo)
recorded to play over a single channel to a passive multi-way speaker.
What about a 2 way active speaker ? say a 3kHz c/o point.
***********************************
Here it should be obvious the wider the range of each amplifier the
more power is needed. 100W + 10OW might suit a say 300Hz crossover
point but is totally wrong for a 3kHz crossover point. The peak
level at a single frequency is not relevant (because that is not
music either). If you look at foobar peak levels drop drastically
above 3KHz, and for 20 bands over 3Khz is only 1/3 of all the bands.
This simple fact would suggest 100W / 50W might suit a 3KHz point,
but the falling treble levels suggest an even larger ideal ratio.
YMMV on diiferent music but bass and midrange bands typically peak
out at around -10dB each for overall 0dB peaks, at the same time the
lower treble regions peak out at around -30dB higher around -40dB.
I would be quite happy to tell someone a 50W amplifier for bass/mid
and 10W (class A) for treble would work extremely well all the time.
The bass / midrange bands hover at around -20dB each for 0dB peaks.
At no point does a single band ever hit 0dB, the peak around -10dB.
The number of bands an amplifier needs to reproduce and their peak
levels essentially determines the ideal power for an active amplifier.
What about a 3 way active speaker ? say ~ 300Hz and 3KHz c/o points.
***********************************
Assuming you accept the points regarding the treble above, the bass
/ mid crossover point is a thornier subject. Some acoustic music has
mainly loads of midrange, some heavier music has mainly loads of bass.
Even though generally midrange band levels are lower in level than
the bass bands, because midrange clipping is worse than bass I
would play it safe and allocate for equal power amplifiers a
near equal number of bands, i.e. for 300Hz/3kHz 25W/25W/10W.
Note that compared to the 50W/10W 2 way though the total power is the
same the 3 way active can reproduce 100W peaks in the bass/ midrange,
with the completely false "disadvantage" of only being able to do 25W
at a single frequency in the bass / midrange versus 50W for the 2 Way.
With music the 3 way will go louder than the 2 way.
The bass and midrange amplifiers do not need to be equal. If the bass
is more powerful you would tend to move up the bass / mid c/o point,
conversely the other way round move it down. A 50W bass amplifier can
and does work with a 100W midrange amplifier with a lower c/o point.
Conclusion
**********
FWIW I think a 3 or 4 way active speaker hits the sweet spot of ~
matching the average / peak statistics of music with their average
/ peak capabilities. But there is also a lot to be said for a "half
breed", active equal power at around 300Hz and passive at 3kHz,
and whilst your at it doing baffle step actively for the "half breed".
/Sreten.
For arguments sake lets say you built a twenty way active speaker
with twenty 1W amplifiers, split like to foobars 20 band display.
Further for arguments sake each of the twenty drivers is flat over
its required range and all the drivers have the same 90dB senstivity.
What is the nature of this beast ?
**********************************
Well maximum output at any single frequency is 1W or 90dB.
With a pink noise signal (equal energy per band) you might suppose
maximum output is 20W or 103dB but this ignores the crest factor or
mean to peak ratio for unclipped random pink noise, so somewhat lower
for average levels but essentially equivalent to 20W unclipped peaks.
Pink noise is a wideband uncorrelated signal, what about a wideband
correlated signal such that all amplifiers and drivers contribute
to a leading edge peak ? The rather surprising answer is off the
chart at a theoretical 400W 116dB unclipped wideband transient.
(Remember its Vsquared/R and in this case all voltages add).
Our 20 way beast is obviously impractical as its characteristics
bear no relation to the statistics of real music typically (stereo)
recorded to play over a single channel to a passive multi-way speaker.
What about a 2 way active speaker ? say a 3kHz c/o point.
***********************************
Here it should be obvious the wider the range of each amplifier the
more power is needed. 100W + 10OW might suit a say 300Hz crossover
point but is totally wrong for a 3kHz crossover point. The peak
level at a single frequency is not relevant (because that is not
music either). If you look at foobar peak levels drop drastically
above 3KHz, and for 20 bands over 3Khz is only 1/3 of all the bands.
This simple fact would suggest 100W / 50W might suit a 3KHz point,
but the falling treble levels suggest an even larger ideal ratio.
YMMV on diiferent music but bass and midrange bands typically peak
out at around -10dB each for overall 0dB peaks, at the same time the
lower treble regions peak out at around -30dB higher around -40dB.
I would be quite happy to tell someone a 50W amplifier for bass/mid
and 10W (class A) for treble would work extremely well all the time.
The bass / midrange bands hover at around -20dB each for 0dB peaks.
At no point does a single band ever hit 0dB, the peak around -10dB.
The number of bands an amplifier needs to reproduce and their peak
levels essentially determines the ideal power for an active amplifier.
What about a 3 way active speaker ? say ~ 300Hz and 3KHz c/o points.
***********************************
Assuming you accept the points regarding the treble above, the bass
/ mid crossover point is a thornier subject. Some acoustic music has
mainly loads of midrange, some heavier music has mainly loads of bass.
Even though generally midrange band levels are lower in level than
the bass bands, because midrange clipping is worse than bass I
would play it safe and allocate for equal power amplifiers a
near equal number of bands, i.e. for 300Hz/3kHz 25W/25W/10W.
Note that compared to the 50W/10W 2 way though the total power is the
same the 3 way active can reproduce 100W peaks in the bass/ midrange,
with the completely false "disadvantage" of only being able to do 25W
at a single frequency in the bass / midrange versus 50W for the 2 Way.
With music the 3 way will go louder than the 2 way.
The bass and midrange amplifiers do not need to be equal. If the bass
is more powerful you would tend to move up the bass / mid c/o point,
conversely the other way round move it down. A 50W bass amplifier can
and does work with a 100W midrange amplifier with a lower c/o point.
Conclusion
**********
FWIW I think a 3 or 4 way active speaker hits the sweet spot of ~
matching the average / peak statistics of music with their average
/ peak capabilities. But there is also a lot to be said for a "half
breed", active equal power at around 300Hz and passive at 3kHz,
and whilst your at it doing baffle step actively for the "half breed".
/Sreten.
Last edited:
Sreten,
Andrew and I agree on this, HOWEVER, it all depends in practical terms on two big factors:
- the average listening level
- the max power the tweeter can handle.
There's no point in putting a 1,000 watt amp on a 30 watt tweeter. ONLY the levels between drivers need to be matched. In other words the limit is the max SPL of the LOWEST rated driver for the system.
IF you never run your system loud enough where it is possible to clip an amplifier, then the ratio of power amps that is being suggested based upon typical power vs. spectrum distributions of recorded music will work ok.
But in a no-hold-barred system striving for the best possible then it is essential that the system be NOT limited by amplifier clipping limitations - or put another way one needs to be able to handle the peak transients that are possible for the recorded media (or "zero" VU redbook level for CD) without clipping either the amp or the speakers.
In most systems something is getting clipped on peaks a lot of the time, either the drivers are running into non-linearity (at the edge of their xmax) or the amps are clipped on peaks... it's a fact. It is tough to make a system that does not clip - and even if you do, probably you'll have to watch your levels to make certain nothing gets the little teeny top lopped off...
This is one reason that ZFB (zero global loop feeback) tube amps sound so good - especially on tweeters - they have a very "soft clip" with compression.
MANY times the sound of "loudness" that people associate with "loud" is actually the increase in distortion with loudness. A really clean LOUD system simply does not sound loud at all. It sounds BIG that's all.
Upon hearing such a thing, some people will tell you that "it doesn't go loud enough for me"... In the mean time the SPL is telling us otherwise...
Just my opinion...
_-_-bear
PS. max SPL and best fidelity/lowest distortion are related sometimes but not quite identical... per above..
Andrew and I agree on this, HOWEVER, it all depends in practical terms on two big factors:
- the average listening level
- the max power the tweeter can handle.
There's no point in putting a 1,000 watt amp on a 30 watt tweeter. ONLY the levels between drivers need to be matched. In other words the limit is the max SPL of the LOWEST rated driver for the system.
IF you never run your system loud enough where it is possible to clip an amplifier, then the ratio of power amps that is being suggested based upon typical power vs. spectrum distributions of recorded music will work ok.
But in a no-hold-barred system striving for the best possible then it is essential that the system be NOT limited by amplifier clipping limitations - or put another way one needs to be able to handle the peak transients that are possible for the recorded media (or "zero" VU redbook level for CD) without clipping either the amp or the speakers.
In most systems something is getting clipped on peaks a lot of the time, either the drivers are running into non-linearity (at the edge of their xmax) or the amps are clipped on peaks... it's a fact. It is tough to make a system that does not clip - and even if you do, probably you'll have to watch your levels to make certain nothing gets the little teeny top lopped off...
This is one reason that ZFB (zero global loop feeback) tube amps sound so good - especially on tweeters - they have a very "soft clip" with compression.
MANY times the sound of "loudness" that people associate with "loud" is actually the increase in distortion with loudness. A really clean LOUD system simply does not sound loud at all. It sounds BIG that's all.
Upon hearing such a thing, some people will tell you that "it doesn't go loud enough for me"... In the mean time the SPL is telling us otherwise...
Just my opinion...
_-_-bear
PS. max SPL and best fidelity/lowest distortion are related sometimes but not quite identical... per above..
Hi,
I hate terms such as "no holds barred", they are meaningless as everything
is a compromise, and its usually used to "justify" pointless over engineering,
and an excuse to allegedly prevent a possible problem that does not exist,
rather than understanding the engineering problem and then optimising it.
Red book CD's do contain 0dB peak transients, they do not contain 0dB
transients at any particular frequency (band) and in particular they do
not contain transient levels in the treble anywhere near bass levels.
The fact that the red book standard allows such things is not relevant,
unless you spend your time listening to test signals rather than music.
How can a 100W class aB treble amplifier be "no holds barred" when a similar
cost 30W class A treble amplifier would actually do the required job better ?
/Sreten.
I hate terms such as "no holds barred", they are meaningless as everything
is a compromise, and its usually used to "justify" pointless over engineering,
and an excuse to allegedly prevent a possible problem that does not exist,
rather than understanding the engineering problem and then optimising it.
Red book CD's do contain 0dB peak transients, they do not contain 0dB
transients at any particular frequency (band) and in particular they do
not contain transient levels in the treble anywhere near bass levels.
The fact that the red book standard allows such things is not relevant,
unless you spend your time listening to test signals rather than music.
How can a 100W class aB treble amplifier be "no holds barred" when a similar
cost 30W class A treble amplifier would actually do the required job better ?
/Sreten.
Not true at all. False.
Most do not, some do.
Most "pop" CDs (etc.) are scrubbed of all peaks... so what?
I dunno - I never said a word about the type of the amp or the class it runs in.
So, just to check, how about a 100w Class A amp compared to your two choices?? This is your strawman, not mine... go for it!
What I did say is that SPL and fidelity are not exactly the same thing.
_-_-bear
Not true at all. False.
Most do not, some do.
Most "pop" CDs (etc.) are scrubbed of all peaks... so what?
I dunno - I never said a word about the type of the amp or the class it runs in.
So, just to check, how about a 100w Class A amp compared to your two choices?? This is your strawman, not mine... go for it!
What I did say is that SPL and fidelity are not exactly the same thing.
_-_-bear
It's a difficult argument.
It's an attempt to relate the maximum signal excursion to the average power. This is not straightforward even in a full-spectrum audio amplifier, given the essential intractability to analysis of music taken as a whole.
Further to that, it's an attempt to relate the maximum signal excursion in any band to the average power in that band.
The problem is simplified, however, insofar as the very large majority of recorded music is 16-bit 44k1 stereo, or can be accommodated adequately to the satisfaction of most listeners therein. There is little point in dragging the characteristics of UN-recorded music into the discussion.
One thing is clear. All other things being equal (the sensitivities of the drivers that is) then the gains of the individual amplifiers must be equal, to ensure that scaling is maintained.
The envelope of the signal in each band results from the superposition of the amplitudes of the individual frequencies within that band in whatever phase relationship they occur. It seems not unreasonable to suggest that the maximum signal excursion in a band should be roughly related to the average power in that band, but it's better not to trust one's intuition in many cases, after all we have the expression 'counter-intuitive'.
It may well be interesting to look at the display in Foobar and attempt to glean some insight from that, but exactly how this display is derived, whether it is an amplitude or power spectrum, is unclear.
Unfortunately reliable information on this subject is not easy to find.
The good news is that a pragmatic solution to this conundrum is straightforward, in concept, if not necessarily in practise.
Simply (!) write a brick-wall digital filter at the crossover frequency or frequencies, chuck in a bunch of CDs, in each case create a file containing e.g. the lower frequency band, subtract this sample-for-sample from the original, record the maximum excursion in each and calculate the ratio. Determine the average and worst cases. It'd be nice if the software allowed for the entry of crossover frequencies on-the-fly.
It's at least 10 years since I wrote a digital filter, and I don't even know the structure of a .wav file, so it might take me a while to revive these skills, but I'll have a look at how much work is involved. I'm not promising anything. Maybe I can do some of it in Cubase, or perhaps someone who's done some of this stuff recently and who can rattle it off quickly will have a go...
w
It's an attempt to relate the maximum signal excursion to the average power. This is not straightforward even in a full-spectrum audio amplifier, given the essential intractability to analysis of music taken as a whole.
Further to that, it's an attempt to relate the maximum signal excursion in any band to the average power in that band.
The problem is simplified, however, insofar as the very large majority of recorded music is 16-bit 44k1 stereo, or can be accommodated adequately to the satisfaction of most listeners therein. There is little point in dragging the characteristics of UN-recorded music into the discussion.
One thing is clear. All other things being equal (the sensitivities of the drivers that is) then the gains of the individual amplifiers must be equal, to ensure that scaling is maintained.
The envelope of the signal in each band results from the superposition of the amplitudes of the individual frequencies within that band in whatever phase relationship they occur. It seems not unreasonable to suggest that the maximum signal excursion in a band should be roughly related to the average power in that band, but it's better not to trust one's intuition in many cases, after all we have the expression 'counter-intuitive'.
It may well be interesting to look at the display in Foobar and attempt to glean some insight from that, but exactly how this display is derived, whether it is an amplitude or power spectrum, is unclear.
Unfortunately reliable information on this subject is not easy to find.
The good news is that a pragmatic solution to this conundrum is straightforward, in concept, if not necessarily in practise.
Simply (!) write a brick-wall digital filter at the crossover frequency or frequencies, chuck in a bunch of CDs, in each case create a file containing e.g. the lower frequency band, subtract this sample-for-sample from the original, record the maximum excursion in each and calculate the ratio. Determine the average and worst cases. It'd be nice if the software allowed for the entry of crossover frequencies on-the-fly.
It's at least 10 years since I wrote a digital filter, and I don't even know the structure of a .wav file, so it might take me a while to revive these skills, but I'll have a look at how much work is involved. I'm not promising anything. Maybe I can do some of it in Cubase, or perhaps someone who's done some of this stuff recently and who can rattle it off quickly will have a go...
w
oh gee... a double post... sreten, not sure what I am ignoring...
If one adopts the assumption you have made that no HF signals ever get near or to "0" vu on a CD, then ur safe, i guess... I think I said as much... so I am not sure what I am ignoring.
Perhaps if one records one's own material then one's perspective on what might be present in what frequency band could just possibly be different??
In other words, while workable, the system with the reduced headroom in the HF is clearly a compromise. Which to me looks like the forest?
Assuming that some tweeters (for example) have 96dB/1w/1m specs while a typical woofer or midrange might be more like 89dB/1w/1m, then using a tweeter amp that is 6dB less power is clearly not a problem.
I guess if you limit your listening to only commercial recordings, ur probably quite safe, since they run them through the mastering process and make certain that there is nothing too challenging left in the recording...
Go out and record the sound of a freight train going around a curve, the squeal from the metal on metal is likely to have very high level components above 10kHz... ok, it's just an example, and it isn't "music"... but I'd prefer to have a system that could handle such things over one that could not...
_-_-bear
If one adopts the assumption you have made that no HF signals ever get near or to "0" vu on a CD, then ur safe, i guess... I think I said as much... so I am not sure what I am ignoring.
Perhaps if one records one's own material then one's perspective on what might be present in what frequency band could just possibly be different??
In other words, while workable, the system with the reduced headroom in the HF is clearly a compromise. Which to me looks like the forest?
Assuming that some tweeters (for example) have 96dB/1w/1m specs while a typical woofer or midrange might be more like 89dB/1w/1m, then using a tweeter amp that is 6dB less power is clearly not a problem.
I guess if you limit your listening to only commercial recordings, ur probably quite safe, since they run them through the mastering process and make certain that there is nothing too challenging left in the recording...
Go out and record the sound of a freight train going around a curve, the squeal from the metal on metal is likely to have very high level components above 10kHz... ok, it's just an example, and it isn't "music"... but I'd prefer to have a system that could handle such things over one that could not...
_-_-bear
Last edited:
Hi,
I quite agree if you want a system that can handle a live recording of
a madman let loose in a pan shop with a couple of steel bars, and
want that to be "realistic" - but that is missing or ignoring the points
I'm trying to make as I see it.
/Sreten.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.