This was not published as such, it was an internal AES-NL project I did with some friends. One test was with a modified FM-coder, which is a unit that brick-wall filters at 15kHz to make room for the FM mux tone at 19kHz, stacking the second channel on top of that.
Th short story is that all but one of 6 participants could reliably detect the presence, or not, of the filter. However, all preferred the filtered version.
The filter is flat till 15kHz, -60dB @ 19kHz, but has huge phase shifts over the band.
Another test was on SACD audibility. We used a wideband mike and wideband signals (up to 50kHz energy content) and listened to it unmodified or 20kHz limited. Resuls were inconclusive, but in the course of the tests we found that only very few SACDs contained any signal above 20kHz - many seemed to be just copies of a CD version....
Jan
Yes, I can explain it very easily; I goofed. I had lost track of the fact that you were following the Oliver criteria.
Sorry,
Bob
It's definitely not a complete list.
Several important papers from AES Hamburg, Mar 81 are missing.
This was one of the first Conventions that discussed CD & digits seriously.
Toshi Doi (Sony), Roger Lagadec (Studer) & a Polygram big wig who's name I can't remember presented important papers that aren't there.
Maybe its just the papers that have been digitised.
We'd need a program for the Hamburg Mar 1981 convention to identify the paper on slew rates.
Hamburg was my first AES paper presentation and the list reminds me that the Dynamic Duo (Lipsh*tz & Vanderkooy) picked up on a point I made and expanded it into 2 papers later that year.
__________________
Thanks for the data on the bandwidth limitation DBLTs.
Your results are exactly the same as min from the early 1980's with vinyl and tape. I used a multiplex brickwall filter too but screwed up to be flat at 20kHz.
It's definitely not a complete list.
Several important papers from AES Hamburg, Mar 81 are missing.
This was one of the first Conventions that discussed CD & digits seriously.
Toshi Doi (Sony), Roger Lagadec (Studer) & a Polygram big wig who's name I can't remember presented important papers that aren't there.
Maybe its just the papers that have been digitised.
We'd need a program for the Hamburg Mar 1981 convention to identify the paper on slew rates.
Hamburg was my first AES paper presentation and the list reminds me that the Dynamic Duo (Lipsh*tz & Vanderkooy) picked up on a point I made and expanded it into 2 papers later that year.
__________________
Thanks for the data on the bandwidth limitation DBLTs.
Your results are exactly the same as min from the early 1980's with vinyl and tape. I used a multiplex brickwall filter too but screwed up to be flat at 20kHz.
Was this a preprint or was it published in the journal, do you know that?
Jan
When you drive a speaker at resonance, it stores energy in the form of its cone velocity and momentum. It will generate emf, as we all know, of a polarity that acts to reduce the current flow. This counter emf will be on the order of the peak voltage driving it when it is at resonance and the driving current is small.
If you reverse the driving voltage at the instant that the reverse emf is at its peak, it will act to increase the driving current rather than decrease it. This simple description is a big simplification, but from this you can see that in principle you could get a peak current equal to twice the peak driving voltage divided by the resistance of the voice coil.
This is discussed in my AES paper "Open Loop Output Impedance and Interface Intermodulation Distortion in Audio Power Amplifiers", available on my website.
Otala's conclusions were wrong on most things, but his claim that unexpectedly high current can flow into a loudspeaker under some conditions was correct. To the extent that such conditions occur with real program material is a bit speculative.
Cheers,
Bob
Bob, this is exactly The reason why I target resonances within the loudspeaker design. If you use airflow as a mean of damping, rather than relying on cabinet enclosure air-spring you gain several advantages, one is that you reduce strain on the driving amplifier, another and more important you reduce the problems with room/speaker resonance interaction.
This puzzles even some of the most brilliant engineers.Originally Posted by keantoken
I don't see a way that the surge current can exceed the total rail voltage divided by the coil resistance.So a +-25V amp into 4R would have max 12.5A max. Anything wrong with this?
See Self's book on power amplifiers, 6th edition, p380-383.
It quotes both M. Otala
AES E-Library Peak Current Requirement of Commercial Loudspeaker Systems
AES E-Library Peak Current Requirement of Commercial Loudspeaker Systems
AES E-Library Input Current Requirements of High-Quality Loudspeaker Systems
and B. Cordell
"Interface Intermodulation in Amplifiers", Wireless World, Feb 1983, p. 32.
(I do not know this article).
Consider this, the gain of a push pull output stage vs voltage vs bias current:
The 'kink' in the Class AB distortion curve caused by the large step changes (from 'gm-doubling' in Class A) in gain. In order to get decent distortion performance your signal therefore need to remain in the Class A region at all times. To make the Class A region span any decent power output your efficiency will be terrible. Hence you pretty much have created a push pull Class A amplifier with the added advantage that at high power levels you can cross into Class B at a significant distortion penalty.
Using optimal bias class B avoids the step changes in gain but now at low power there is a small wiggle in the gain of the output stage which causes it to be less linear than Class A for small signals.
Class XD displaces the wiggle to a positive or negative voltage so that it takes a larger voltage swing to reach it. For small signals it operates on the linear shoulder portion of the Class B gain curve therefore delivering Class A levels of linearity. For large signals (in XD Const) where the load current exceeds the displacement current, the wiggle region is operated in so distortion becomes comparible to optimally biased Class B.
The distortion penality from operating in the 'wiggle' region is much less than the step changes in gain that occur for an overbiased stage hence why in the previous plot I posted 'Class AB' exceeds the distortion level of Class B when it runs out of bias current, while Class XD does not. Therefore you have the low distortion of Class A for small signals, no worse than Class B for large signals, and not much worse efficiency than optimal bias Class B.
There is no reason you couldn't increase the constant displacement current to a point where the load current never exceeded it, therefore always operating in Class A. This however would come at a significant efficiency penality, which is half of the point of using Class XD vs Class A/AB. The voltage controlled variation of Class XD (XD-PP) aims to maintain just enough displacement current to keep the amplifier operating in the linear region without ruining it's efficiency.
I don't know how to explain it any clearer than that. You seem to be in denial that Class XD is a better solution than Class AB. You can't operate in Class A without having those gain steps which causes the kink in the distortion curve and subsequent worse linearity than Class B, there doesn't exist a bias point which satifies both operating in Class A for small signals and avoiding the gm doubling problem.
Last edited:
This is how I can say I know of only one real life case where the (instantaneous
current demanded exceeds that calculated from the impedance curve. I've looked for and observed this in real life.
Using actual recordings, percussion waveforms with a high degree of asymmetry should be the best candidates.
Higher peak currents will happen even in purely LTI circuits, with the driver replaced by a resistor.
Are these vented enclosures you are talking about? I use AcoustaStuff in the enclosure, lightly stuffing it even for vented designs.
The example I cited was a very rough approximation for a single closed-box woofer with no other drivers or crossover. The low-frequency behavior of a vented design is more complex, and I don't know if it makes the worst-case current scenario worse or not.
I also imagine that energy stored in crossovers can play a role in creating instances of increased current.
Cheers,
Bob
My article was actually first presented at the AES Convention in 1982 and is available on my web site. It shows a made-up waveform that stimulates such high currents. That was done using SPICE way back when, long before it was available on PCs. More than a decade later, a nearly identical waveform was shown in Self's 4th edition (Figure 7.39).
Cheers,
Bob
A closed box is nothing short of terrible, a vented one may be even worse. The optimization and design-rules on this serves only a function of extending the frequency response of the driver enclosure combination, not an enhancement of quality and performance ..
Sometimes forces on the membrane work against the motion other times they work with the motion...
I try to target as close to homogenous damping regardless of position as possible, this is clearly visible on impedance measurements. The price you have to pay is less low frequency output, thus you have to recreate those by other means.
Sometimes forces on the membrane work against the motion other times they work with the motion...
I try to target as close to homogenous damping regardless of position as possible, this is clearly visible on impedance measurements. The price you have to pay is less low frequency output, thus you have to recreate those by other means.
This is not a loudspeaker thread, but I have designed many loudspeakers and I have no idea what you are talking about.
Cheers,
Bob
Sorry, normally I don't discuss speakers at all here. But as this relates to the interface to the amplifier I brought it up.
You can make the impedance of your base system fairly linear by adding parallel LCR circuits, but it rarely is very successful as speaker system resonances are mechanical realated and based on elements that are nothing close to beeing linear and thus they are all over the map.
The only way I have found and the only one that really works is to target and dampen the air motion right behind the driver. This can be done by placing a flow resiative went. This approach takes most of the cabinet related resonances out of the system an leaves you with a much flatter impedance curve. The result is something resembling a gentle roll off approaching 2. Order and a Q off 0,7 (assuming the right amount of flow damping)
Prize is that you no longer have the system resonances to aid the low frequency extension.
I normally cure this by adding an extra driver so I pull more current from the amplifier in the lowest frequencies.
You can make the impedance of your base system fairly linear by adding parallel LCR circuits, but it rarely is very successful as speaker system resonances are mechanical realated and based on elements that are nothing close to beeing linear and thus they are all over the map.
The only way I have found and the only one that really works is to target and dampen the air motion right behind the driver. This can be done by placing a flow resiative went. This approach takes most of the cabinet related resonances out of the system an leaves you with a much flatter impedance curve. The result is something resembling a gentle roll off approaching 2. Order and a Q off 0,7 (assuming the right amount of flow damping)
Prize is that you no longer have the system resonances to aid the low frequency extension.
I normally cure this by adding an extra driver so I pull more current from the amplifier in the lowest frequencies.
Thanks for the clarification. You really got me worry, I read through all those pages last night over and over, and I think over it, did some calculation, and it cannot be wrong.
I still yet to read what is "Self class AB" vs "Self Class B". The way the graph shown kind of make me question whether his "class AB" satisfies Oliver's condition when he print the graph.
Yes, the Oliver's condition is one of the most useful thing I learned from your book, I follow that to the "T". I know now I cannot use 0.12ohm for Re and I have to change to 0.22. That's the reason I am playing with the layout of 8 pairs of output transistors. People might think that's crazy, but is it? With 8 pairs, I can satisfy Oliver's condition AND I get 8W of pure Class A. With 30V rail, I can easily get over 100W Class AB power. I don't think any reasonable Class A amp can touch that. with 8W of Class A, I can cover 99.9% of the programs for listening at home of any size. People don't realize how loud is 8W. the occasional spike in signal is covered by the Class AB that satisfies Oliver's condition and still at optimal.
I think people that concentrate on XD really don't get what I am trying to say even I said it over and over. I did not read Self's book as detail as yours, but I don't recall I've seen any explanation of Oliver's condition in his book.
My philosophy is to satisfy Oliver's condition AND make a sizable Class A region to get the best of both worlds. I am dancing on Class A power vs Class AB power vs heat dissipation I can handle. Those so far, seems to be fighting against each other and I really don't know what to think.
Thanks
Last edited:
I would think an SMPS designer (which I am not) wouldn't have any trouble at all understanding how much surge current a speaker can draw under the right conditions.
Unless the output impedance becomes high (possible due to SOA protection), then the back-EMF can't rise above the rail voltage. Output-to-rail diodes reduce this. Knowing this, we can say that, if when the back-EMF is at the rail voltage, the amplifier output moves to the opposite rail, then you will have the full rail-to-rail voltage across the voicecoil.
So this again supports my suggestion that the max surge current cannot exceed the rail-to-rail input voltage of the amplifier across the voicecoil resistance.
If the current limit kicks in you get a flyback effect, with the back-EMF potentially rising over the rails and then being clamped by saturating transistors or output-to-rail diodes. BUT, since the amp must be a current source rather than voltage source at the time this happens, you still don't get violation of the rule, unless perhaps the negative output transistor has independent and delayed SOA protection.
Unless the output impedance becomes high (possible due to SOA protection), then the back-EMF can't rise above the rail voltage. Output-to-rail diodes reduce this. Knowing this, we can say that, if when the back-EMF is at the rail voltage, the amplifier output moves to the opposite rail, then you will have the full rail-to-rail voltage across the voicecoil.
So this again supports my suggestion that the max surge current cannot exceed the rail-to-rail input voltage of the amplifier across the voicecoil resistance.
If the current limit kicks in you get a flyback effect, with the back-EMF potentially rising over the rails and then being clamped by saturating transistors or output-to-rail diodes. BUT, since the amp must be a current source rather than voltage source at the time this happens, you still don't get violation of the rule, unless perhaps the negative output transistor has independent and delayed SOA protection.
Did you know that, within the class A region, the Oliver condition is not the most linear? You can improve the class A region linearity at a tradeoff of AB linearity. I found 10mV (total voltage across the emitter and base stopper resistors) gives the best A region linearity. It's not easy to maintain this bias voltage though in a high power stage.