are there instances in commonly available recordings where a brief transient can make good use of a large peak without serious heating and destruction?- guesstimates/examples - ?
Serious heating/destruction of what ?
Speakers are pretty forgiving on transients. Many can take 2-3 times normal power on transients without damage.
The same goes for output transistors. Its the average power that is important.
Speakers are pretty forgiving on transients. Many can take 2-3 times normal power on transients without damage.
The same goes for output transistors. Its the average power that is important.
Hi Freddyi
It depends. What we hear as “loudness” is something like a short term average when one is measuring actual SPL.
Music spans a large range so far as it’s “peak to average ratio”.
Pink noise used for audio power testing has a peak to average ratio of 6dB so the amplifier has to be 4 times larger than the SPL one would read with a sound level meter on “slow” but Smaart and some other spectrum measuring pink noise systems have a P/A ratio of 10 dB so the amp has to be 10X larger than what one might think if one reads an SPL meter.
A decent hifi recording may have a peak to average ratio of 20 or even 30dB so that the instantaneous peaks need to be 100 to 1000 times larger than what one measures with a meter.
A recording like the Fireworks recording I had posted some time back has a P/A ratio of more than 40dB which requires a peak more than 10,000 times larger than the average Wattage.
So, if one wanted 90dB SPL at 1 meter and had a 90dB 1W sensitivity loudspeaker playing the sounds mentioned above, one would need anywhere from 4Watts to 10,000Watts to reproduce the signal unclipped.
The up side and the reason most hifi;s are limited dynamically is that we don’t hear clipping if it’s very short, it’s inaudible as a “flaw” but if one compares A/B to a much larger unclipped amplifier, then what you can hear is the unclipped version is more dynamic.
In the real world, an oscilloscope looking at the amplifier output is the ONLY way to tell if your system clipping instantaneously.
Best,
Tom
I have noticed problems with amplifiers with low voltage supply rails.
They just don't seem to have the dynamic range and just clip transients.
To faithfully reproduce transients you need a decent power supply voltage.
For my own design disco and guitar amplifiers I use high supply rails and often get comments back on how good my amplifiers sound.
I was DJ-ing at a club one night and the announcer was using an in house amplifier for announcements and it sounded terrible. I gave him my mic to use through my gear and he was gobsmacked how clear it sounded.
They just don't seem to have the dynamic range and just clip transients.
To faithfully reproduce transients you need a decent power supply voltage.
For my own design disco and guitar amplifiers I use high supply rails and often get comments back on how good my amplifiers sound.
I was DJ-ing at a club one night and the announcer was using an in house amplifier for announcements and it sounded terrible. I gave him my mic to use through my gear and he was gobsmacked how clear it sounded.
Hi,
Not a chance with currently commonly available recordings.
Unfettered dynamics are nowadays as rare as rockinghorse droppings.
You'd need to be an expert on relatively obscure audiophile recordings
to know ones that would let you pump 1KW peaks in the midrange
into a speaker that can barely handle 100W under normal conditions.
rgds, sreten.
Not a chance with currently commonly available recordings.
Unfettered dynamics are nowadays as rare as rockinghorse droppings.
You'd need to be an expert on relatively obscure audiophile recordings
to know ones that would let you pump 1KW peaks in the midrange
into a speaker that can barely handle 100W under normal conditions.
rgds, sreten.
Last edited:
The OP didn't say anything about faithfully reproducing a transient, but I would assume too that that should be included in the mix. A crescendo is a joy to hear only if the full sound level increase is actually heard and is undistorted.
Another factor to consider is whether or not a relatively high level transient causes any driver in the speaker system to exceed Xmax. That of course will result in a distorted output (not good) but probably is not going to cause damage to the driver. IMO Xmax, especially with a sealed system as opposed to bass reflex, most of the time is going to be more of a limiting factor than amplifier clipping.
Another factor to consider is whether or not a relatively high level transient causes any driver in the speaker system to exceed Xmax. That of course will result in a distorted output (not good) but probably is not going to cause damage to the driver. IMO Xmax, especially with a sealed system as opposed to bass reflex, most of the time is going to be more of a limiting factor than amplifier clipping.
Well, 10dB peaks aren't so rare, but if that's all you're concerned about, it's implied that you're carrying about 100W average in the music already, which is either damned loud for a home system or your speakers are extraordinarily inefficient. Just another 10dB and you're down to 10W average, which could be considered "normal loud", and 20dB peaks are also not entirely rare. Sure, you can build a system that can handle this with some sort of minimal compression, but the paralleling of drivers it takes kind of messes up the idea of maintaining a great image. So, yeah, you're probably a bit past reality here for what might be considered "high end" listening. It'd be more practical to take extreme steps to drop the noise floor in your listening environment if super dynamic range is the goal, not to mention the preservation of human hearing.
Hi
To be clear, what we measure and what we hear as “loud” are two different things which are only somewhat tied together. Most people are familiar with the “equal loudness curves” that show the actual loudness vs “how loud it sounds” subjectively and as is clear, ones ears are nothing like “flat” in their response.
As a result, low bass and high frequency sounds need to be MUCH louder to “sound like” a given SPL in the mid band. To be detectable at 20hz for example, a sound needs to be about 70dB (about 10 million times) louder than in the upper mid range to be equally loud or detectable.
While the frequency in question accounts for one aspect of the giant difference between how loud something sounds vs how loud it measures, there is a similar discrepancy in the time domain as well.
If one has a specialized “peak hold” sound level meter or if one looks at a microphones voltage on an oscilloscope and calculates based on the microphones sensitivity, then this part can also be seen.
For example, I have measured a peak over 130dB by tossing a table spoon onto a tile floor, a peak greater than 140dB by sitting in a car and slamming the door with the windows closed. While those numbers sound like it would be deafening, they are not in part because of the frequency but also the duration of the sound.
We don’t hear very short loud sounds as loudly as when they are longer in duration. Yet if one wanted to accurately reproduce these sounds at a realistic level, those are the peak levels one requires.
What we hear as subjective loudness is weighted by our ears response curve AND is something like a short term average SPL level and NOT a peak level.
Thus, if one looks at SPL with a sound level meter OR looks at amplifier voltage with an analogue meter, what you see is an average, an integrated level averaged over some period of time and NOT the instantaneous peak value.
Most people in audio are familiar with the term “loudness wars” which is the record producers effort to compress the dynamic range so that the average level is as high as possible. For compressed FM pop music, one finds the peak level is only about 8 or 10dB greater than the average level (the average level being more like the heating value in the signal as well as subjective loudness).
On the other hand, older recordings and good modern recordings can have a peak to average ratio more like 20dB or more and specialty recordings much greater than that.
The up side for hifi systems is that we can’t hear instantaneous clipping, sure everyone knows what clipping sounds like if it’s long enough but short term clipping IS INAUDIBLE as a flaw until one can switch back and forth with an unclipped signal.
Then, the unclipped version simply sounds more dynamic, not necessarily “louder”.
If one has a program like “Cool Edit”, one can read the peak to average ratio and other information about the recording and if one has an oscilloscope, one can examine the amplifier voltage to see if it or any other part of the chain is clipping instantaneously and if it is, that means you have dynamic compression as a result. The oscilloscope doesn’t lie.
To be clear, what we measure and what we hear as “loud” are two different things which are only somewhat tied together. Most people are familiar with the “equal loudness curves” that show the actual loudness vs “how loud it sounds” subjectively and as is clear, ones ears are nothing like “flat” in their response.
As a result, low bass and high frequency sounds need to be MUCH louder to “sound like” a given SPL in the mid band. To be detectable at 20hz for example, a sound needs to be about 70dB (about 10 million times) louder than in the upper mid range to be equally loud or detectable.
While the frequency in question accounts for one aspect of the giant difference between how loud something sounds vs how loud it measures, there is a similar discrepancy in the time domain as well.
If one has a specialized “peak hold” sound level meter or if one looks at a microphones voltage on an oscilloscope and calculates based on the microphones sensitivity, then this part can also be seen.
For example, I have measured a peak over 130dB by tossing a table spoon onto a tile floor, a peak greater than 140dB by sitting in a car and slamming the door with the windows closed. While those numbers sound like it would be deafening, they are not in part because of the frequency but also the duration of the sound.
We don’t hear very short loud sounds as loudly as when they are longer in duration. Yet if one wanted to accurately reproduce these sounds at a realistic level, those are the peak levels one requires.
What we hear as subjective loudness is weighted by our ears response curve AND is something like a short term average SPL level and NOT a peak level.
Thus, if one looks at SPL with a sound level meter OR looks at amplifier voltage with an analogue meter, what you see is an average, an integrated level averaged over some period of time and NOT the instantaneous peak value.
Most people in audio are familiar with the term “loudness wars” which is the record producers effort to compress the dynamic range so that the average level is as high as possible. For compressed FM pop music, one finds the peak level is only about 8 or 10dB greater than the average level (the average level being more like the heating value in the signal as well as subjective loudness).
On the other hand, older recordings and good modern recordings can have a peak to average ratio more like 20dB or more and specialty recordings much greater than that.
The up side for hifi systems is that we can’t hear instantaneous clipping, sure everyone knows what clipping sounds like if it’s long enough but short term clipping IS INAUDIBLE as a flaw until one can switch back and forth with an unclipped signal.
Then, the unclipped version simply sounds more dynamic, not necessarily “louder”.
If one has a program like “Cool Edit”, one can read the peak to average ratio and other information about the recording and if one has an oscilloscope, one can examine the amplifier voltage to see if it or any other part of the chain is clipping instantaneously and if it is, that means you have dynamic compression as a result. The oscilloscope doesn’t lie.
Many wise, interesting, and diverse thoughts here. Good thread.
1. Once a few buddies at Bell Labs had a line on fried-egg tweeters (that we measured as the best we ever measured, 1969) to buy by the carton-load. i set up 9 per channel (obviously leading to an 8-Ohm crossover) in a curved array. I truly believe it is beneficial to have tweeter power (or headroom) and a shortcoming of many otherwise good systems.
2. Interesting how easy it is to hear (see?) a triangle in an otherwise very loud orchestra. Is it their very high peak, as Tom recounts? Or seeing the percussionist? Recording producers crank up the triangle mic a whole lot to produce the same effect on recordings, I bet.
3. Loudness peaks is a statistical coincident thing. Women's choirs for example have giant peaks by coincidence. The stats of using crossovers dramatically reduces the peak-stat in a band and hence amp and driver power capability. All statistics.
4. For sure, few of us have sources that aren't compressed. So this discussion is "academic."
Ben
1. Once a few buddies at Bell Labs had a line on fried-egg tweeters (that we measured as the best we ever measured, 1969) to buy by the carton-load. i set up 9 per channel (obviously leading to an 8-Ohm crossover) in a curved array. I truly believe it is beneficial to have tweeter power (or headroom) and a shortcoming of many otherwise good systems.
2. Interesting how easy it is to hear (see?) a triangle in an otherwise very loud orchestra. Is it their very high peak, as Tom recounts? Or seeing the percussionist? Recording producers crank up the triangle mic a whole lot to produce the same effect on recordings, I bet.
3. Loudness peaks is a statistical coincident thing. Women's choirs for example have giant peaks by coincidence. The stats of using crossovers dramatically reduces the peak-stat in a band and hence amp and driver power capability. All statistics.
4. For sure, few of us have sources that aren't compressed. So this discussion is "academic."
Ben
Don Keele used to use a ~1.5kW amp for tests on speakers in Audio magazine. Almost all decent quality tweeters could take the full power output for very short tonebursts with no problems.
In music, the peak/average ratio is a lot higher in the treble, as far as I understand.
In music, the peak/average ratio is a lot higher in the treble, as far as I understand.
Hi,
No its not. I'm talking about using a 1KW amplifier with
a speaker that is easily blown up with a 100W amplifier.
i.e. recordings with a 10dB higher crest factor than normal.
They exist, but are rare and obscure, and nobody buys a
1KW amplifier to play them for a pair of speakers for which
100W is plenty enough usually. A recipe for dead speakers.
Such peaks are routinely removed on nearly all recordings.
THe crest factor is set by average to peak levels
and they are controlled on nearly all recordings.
Classical has the highest CF, especially if gain riding is
not used, and simple recording and minimal limiting
techniques are used.
rgds, sreten.
FWIW a driver can take any sort of short term peak based
on it doesn't cause excursion limiting or related damage and
the simple thermal time constant of the voice coil and the
length of time the peak level is applied.
That might imply a mid driver can take say 2KW peaks on
an obscure recording, but they don't exist in recordings.
Back in the days of comphrehensive speaker protection I think
The KEF 105 allowed 1KW AC for 10mS in the midrange.
One such peak is 10W over a second, easy for a B110.
The time constant for a B110 is about 4 seconds.
Last edited:
Modern, dynamic recordings are not so rare as you say. There's tons of it out there. Obscure? Mostly.
Here is a histogram of the "DR Database" minimum, average, and maximum crest factors, arranged by album, of all 38000+ CDs and vinyl records in the DR Database as of April 2013. The DR scale is approximately representative to average-to-peak (crest factor) values in dB:
An externally hosted image should be here but it was not working when we last tested it.
As you can see almost all commercially available recordings have significantly compressed dynamic range, especially within the last 23 years from 1991 when dynamic compression of CD tracks began in earnest with the commercial availability of multi-band digital compressors, now used by almost all popular music mixing and mastering engineers (but typically not used in classical music). Also note the bi-modal nature of the CD data - implying two types of CDs, i.e., non-compressed and "loud" characteristic of popular music discs.
More info: "Loudness War" Dynamic Range Compression & The DR Database - Observations - diyAudio
Last edited:
I've thought about setting up a fast comparator and a long latch period to indicate clips in hi fi amps, but thought it would be offensive to 'philes who believe they would/ should hear it.
“i.e. recordings with a 10dB higher crest factor than normal.
They exist, but are rare and obscure”
What do you base that assertion on? Perhaps we are talking about two different things, I very rarely see music with less than 10dB peak to average ratio
If you don’t have a program like Cool edit to examine a wave file for it’s peak and average levels and don’t have an oscilloscope or Duroughs VU meter, you can down load this free Orban software meter and still get an idea what the music is actually doing.
ORBAN Loudness Meter
A free program called Audacity also has an ability to show peak and average levels with it’s VU meter function.
Meter Toolbar - Audacity Manual
Hi,
I'm referring to off the scale stuff, that can only be played
back with no amplifier limiting in a typical studio, carefully.
Yes there are good highly dynamic recordings, but very
few that don't impose a limiter of any sort, and allow
the random highest peak level to set overall level.
It just doesn't work like that, it can't, in practice.
rgds, sreten.
Perhaps you should consider the last post that I made, above, in order to substitute data for disagreement. The entire DR Database is represented in that one histogram.
In my experience most audiophiles play mostly commercially available music, represented fairly completely by the DR Database records shown. YMMV.
In my experience most audiophiles play mostly commercially available music, represented fairly completely by the DR Database records shown. YMMV.
There are some truly dismal recordings there especially when you get to low bit mp3's but the wave files from steeley dan approach 20dB p/a and the best ones are better than 20dB p/a. Examine the peak to RMS values for the green recordings.
20dB p/a means the peak value is 100X the average level which is close to what we hear as loudness.
Hi,
Your not following what I'm one about all. You can use
a pair of LS3/5A's with a 1KW amplifier needed for the
peaks not clipping in a very dynamic recording, and
totally trash them with an unclipped 100W amplifier
on mundane recordings.
If a speaker can handle 100W programme with a crest
factor of 20dB, i.e. average power is just 1W, then
it can handle 1KW programme with a crest factor of
of 30dB, as the average power is still just 1W.
Especially if the programme is exactly the same, but
one is limited to 20dB, and the other limited to 30dB.
(The +10dB extra peaks being necessarily rare and sporadic,
for the 20dB peak limiting not to totally ruin the recording.
In reality progressive limiters are used, not hard levels.)
The replay level would be the same for unclipped program.
However the 20dB CF recording and the 1KW amplifier
used unclipped, i.e. +10dB level would certainly destroy
the LS3/5A speakers, even though the average is 10W.
rgds, sreten.
Last edited:
- Status
- Not open for further replies.