Insanity Can Be Yours

1,500W / 4 Ohms Power Amplifier

Rod Elliott (ESP)

Introduction
This project is mainly in answer to those for whom no amount of power is enough. I have lost count of the number of times people have asked if it's
alright to increase the supply voltage on every circuit published, and in general the answer is no - it's not alright. Every design on this site is
optimised for the stated power. There is always some flexibility, but you must be very careful to make sure that transistor safe operating area (SOA)
is not exceeded. There is also a maximum voltage for any semiconductor, and devices must be selected to ensure they are used within their ratings.
While (lateral) MOSFETs offer some real advantages, they are relatively expensive, and difficult to obtain with voltage ratings above 200V. Vertical
MOSFETs (e.g. HEXFETs and the like) are a possibility, but suffer gross non-linearity at very low currents. Therefore, a relatively high quiescent
current is needed, and this makes heat removal that much more difficult. There are other issues as well, but a discussion of them is outside the scope
of this article.

Capable of 2kW peak and a minimum of 1.5kW continuous, it has to be said that this amplifier will blow up any speaker connected to it. Regardless
of the claimed power that various drivers can handle, they can't. To put this whole issue into perspective, take the most powerful and robust driver
you can (8 ohms), and connect it directly to the 110V mains (I recommend this as a 'thought experiment', rather than actually doing it!). 110V RMS
into 8 ohms is 1,500W. How long would you expect the speaker to last? Most will be toast within perhaps 30 seconds or less! A very few will last
slightly longer, but none will take that level of abuse for more than a few minutes.
I strongly suggest that you read Power Vs. Efficiency before continuing.
Have a look at the voicecoil of any speaker. Imagine how hot it will get with even 100W of continuous power - feel a 100W light bulb - 100W is
enough to make any small mass get very hot indeed, very quickly. 1,500W is an awesome (scary even) amount of power! Look at the size cable
needed to carry 20A, then look at the wire size used for the voicecoil. If you don't see a real problem, then I suggest you abandon electronics take up
flower arranging as a hobby.
It must be understood that this is a 'brute force' approach, and that much is deliberate. Although it would be possible to use more finesse in the final
design (such as using a tracking power supply, or a Class-G multi supply rail approach), these are harder to design, and would require building a
prototype to verify performance. Since I have no need at all for this much power, I am not prepared to spend the time and money to build and test
something I'll never need.
I know that no speaker I have (or am likely to ever have) can take that much power, and the amp would be a waste of money. Should someone be
silly enough to pay me the AU$12,000 I would charge to build a mono version of the amp, then I would happily do so. So, I am confident that it will

work as described, but it will almost certainly never be built by me. I hope that my readers share my pragmatism.
Should you (wisely) decide that this amp is as silly as I think it is, then go back to Project 68. The dual board version with ±70V supplies is still
capable of destroying many drivers, but there are loudspeakers than can take its output short term. This makes it ideal for subwoofer duty, easily
giving over 500W into 4 ohms for transient signals, or 450W continuous (which it can do all day with a fan forced heatsink). This is a proven design,
and although not inexpensive, it still represents fairly good value for such a high power amplifier. IMHO there is no point trying for more power, as
few drivers can handle more than a couple of hundred Watts without suffering severe power compression.

Description
First, let's look at the requirements to get 1.5kW into 4 ohms. We need 77.5V RMS across the load, but we need to have a bit more, because the
supply voltage will collapse under load, and there will always be some voltage lost across the transistors, emitter resistors, etc. The supply voltage
needs to be ...
VDC = VRMS * 1.414
VDC = 77.5 * 1.414 = ±109.6V DC
Since we have not allowed for losses yet, we need to allow around 3-5V for the amplifier, and an additional 10V or so to allow for the supply voltage
falling when the amp is loaded. The higher the current, the greater the I²R (resistive) losses, so 5V was used in this design. With a transformer rated at
2 x 90V, this gives an unloaded supply of ±130V DC (260V DC total), so the supply has to be treated with extreme care - it is very dangerous indeed.
There is an old term used by those who work with high voltages ...
One flash and you're ash !
... and you would do well to remember this. Add on the auxiliary supplies (taking the total to 270V DC), and the supply is capable of killing you
several times over, even after it is disconnected from the mains. Even the output signal to the speakers must be treated with care, as 77V is enough to
give you a nasty shock.
The final supply voltage will be around ±120V, because even with the biggest transformer and filter capacitors, you will lose voltage. The current
demand is also prodigious. With a peak voltage of 110V, the peak supply current is 27.5A into a 4 ohm load. RMS speaker current is just under 20A
at full power. Everything you thought you knew about amp building needs to be re-thought. PCB tracks cannot be used for these current levels,
because the extra resistance will cause current balancing problems with the power transistors. All wiring needs to be extremely robust, and must
absolutely not allow any possibility of contact (it will kill you) or short circuits (which will kill the amp). The supply is capable of vaporising thin
wires and PCB tracks.
Because of the issues discussed above, bipolar transistors were selected as most appropriate for the output stage. This was primarily dictated by the
supply voltage, which exceeds that allowed for any affordable lateral MOSFET. It is even a challenge for affordable BJTs, but the MJ15004/5 or
MJ21193/4 pairs are within ratings, so these are suggested. While I would normally specify a compound pair (also called a Sziklai pair) for the output
stage, in this case it is a triple stage, and the Sziklai pair (much as I like it) can be prone to oscillation, primarily on the negative side. This is highly
undesirable for an amp with the power of that described here. Despite reservations, the triple Darlington is more appropriate for this application.
Next, we need to look closely at the power dissipation of the devices. Worst case resistive load dissipation occurs when ½ the supply voltage is across
both load and transistors. This occurs at a voltage of 65V across the load (worst case), and gives a peak (instantaneous) power in both load and output
stage of ...
P = V² / R = 65² / 4 = 1056W
This is only a starting point, because we must have a safety margin. Remember that the peak dissipation into a reactive load with a 45° phase angle is
almost double that calculated above, about 1900W. If the transistors can be maintained at 25°C (not likely), that's fine, but we need to add more to
allow for elevated temperature. I have elected to use 9 output devices, with a tenth device used as a driver. This maintains worst case peak dissipation
at 212W - not much of a safety margin, but it should be ok in practice - in part because the supply voltage will collapse under load. Cooling is vitally
important - this amplifier will need a very substantial heatsink, and fan cooling is essential. Fans should cut in at no higher than 35°C.
The MJ15024/5 (or MJ21193/4) devices are TO-3 packages, and are specified for 250W dissipation at 25°C. It is worth noting that the driver in this
arrangement will contribute some of the output, but it only reduces the main transistor's peak dissipation by about 5W. TO-3 devices are specified
because they have the highest power rating of any general purpose package, because thermal resistance is lower than any flat-pack plastic device.
The MJE340/350 pre-drivers reduce the loading on the VAS (voltage amplification stage) and ensures good linearity with acceptably low dissipation
of the VAS transistor (Q4) and its current source (Q6). Even so, with around 12mA through the VAS, dissipation is 0.72W, so Q4, Q6, Q9 and Q10
must have heatsinks (or a common heatsink that is suitable for the power dissipated). The bias servo transistor (Q5) should be mounted in thermal
contact with the main heatsink.
The protection circuit will limit the peak transistor power to around 180W, with a short circuit current of about 12A. This is slightly outside the SOA
of the output transistors. Although it is possible to get the protection circuit to force the output stage to follow the SOA curve, this almost inevitably
means that maximum power cannot be achieved unless the protection circuit is made considerably more complex. For an amp that (hopefully) will
never be built, this was not warranted. The alternative is to add more output transistors.
 Figure 1 - 1.5kW Power Amplifier
The circuit is completely conventional, using a long tailed pair input stage, direct coupled to the VAS. No current mirror was used for the LTP, as this
increases open loop gain and may give rise to stability issues. In a very high power amp, stability is paramount. The amp must never oscillate under
any normal load condition, because the heat created can cause almost instant transistor failure.
* Note: It is imperative that Q5 (the bias servo transistor) is mounted on the heatsink, in excellent thermal contact. This is because, unlike most of my
other designs, this amp uses conventional Darlington output configuration. It is necessary to use a Darlington arrangement (or a low power
Darlington transistor as shown) for Q5 to ensure that the bias remains at a safe value with temperature. This is left to the constructor, because as
noted I will not provide technical assistance for this design. There is probably good cause to model and test this aspect of the design very carefully,
because it is so important. The arrangement as shown will reduce quiescent current at elevated temperatures. For example, if total Iq at 24°C is
165mA, this will fall to ~40mA at 70°C. This is probably fine, because there is some delay between the a power 'surge' and the output transistors
transferring their heat to the bias servo via the heatsink.
The additional feedback components (R6a and C3a, shown dotted) are optional. They may be needed to ensure stability. The output 'flyback' diodes
(D9 and D10) will normally only ever conduct if the protection circuit operates while the amp is driving a reactive load. The 1N5404 diodes can
withstand a peak non-repetitive current of 200A. Higher rated components may be used if desired. The voltage rating needs to be at least 400V.
The 100 ohm trimpot used in the LTP is used to adjust the DC offset to minimum. With the component values as shown, offset should be within
±25mV, before adjustment. The second trimpot is used to set quiescent current. This should be set to a value just sufficient to minimise crossover
distortion. High quiescent current is not desirable, simply because of the power dissipation. Quiescent current is set so that 150mV is measured across
R19 or R20.
You will also note that SOA protection has been incorporated. I don't like protection circuits very much, because they can cause very audible sound
quality degradation if they operate, but in an amplifier with so many transistors (not to mention the massive power supply needed), the available
energy will cause instantaneous destruction of the output stage if protection is not used. Would I actually risk applying a short circuit to the output
terminals? No, I would not. I haven't built this design, and I have no intention of doing so. A full simulation tells me that the protection circuits
should ensure that nothing blows up, but I do not intend to find out.
Input sensitivity is about 1.77V for 900W into 8 ohms, or 1800W into 4 ohms, or 1.6V RMS for rated power (1.5kW into 4 ohms).
More Power?
Believe it or not, the design can be pushed even further. You will need to add more power transistors and upgrade the power supply though. Supply
voltages of up to ±150V can be used without changing anything other than increasing the number of output devices. Although the MJ15024/5 are
rated at 250V, they will take more because the base is tied to the emitter with a very low resistance. 1200W into 8 ohms is quite possible, with around
2kW into 4 ohms (or 4kW / 8 ohms in BTL configuration). I suggest that the number of output devices be increased by at least 25% - a total of 13
devices for each supply rail (26 in total, including drivers).
There would seem to be some valid reasons for using MOSFETs, not least because the voltage rating is easily achieved with low cost (switching)
devices. The use of high voltage devices with a relatively high RDSon is necessary to minimise distortion at low levels. Given that these usually use
the TO220 package, I recommend that peak dissipation is limited to 35W or less (~15W average). The TO220 package is convenient, but is hopeless
for transferring significant heat from junction to heatsink. This would require around 50 x TO220 devices for each rail (100 in all). Remember that
vertical MOSFETs must be matched, so expect to purchase at least 300-500 transistors to allow for matching. All parallel devices must have as close
as possible to the same VTH (threshold voltage, gate to source). Note that a vertical MOSFET version has been simulated, built and tested at lower
power (a couple of hundred Watts), but is not shown here.
Given that the peak output power of the amp as described will be around 950W into 8 ohms and close to 1800W into 4 ohms, I doubt that any

upgrades will be needed .

2 Ohms?
No. Not a chance. Adding more transistors will certainly allow it - in fact, the transistors there already will come very close. The problem is simply a
matter of current. Over 40A RMS at full power just means that large quantities of the output power will be dissipated as heat in connectors and
speaker leads, and it's just not worth the effort. Who wants to drag around industrial welding cables as speaker leads? Even at 4 Ohms, you have to
deal with over 20A RMS (and 30A peaks), so leads and connectors need to be very robust.

Power Supply
The power supply needed for an amp of this size is massive. Grown welding machines will look at it and cry. For intermittent operation, you need a
minimum of a 1kVA transformer (or 1.5kVA for the 2kW version), and it will have to be custom made because of the voltages used. If you expect to
run the amp at continuous high power, then transformers should be 2kVA and 3kVA respectively. Filter capacitors will pose a problem - because you
need caps rated for 150V, these will be hard to find. Because high voltage high value caps can be difficult to find, it may be necessary to use two
electros in series for each capacitor location. This is the arrangement shown. You must include the resistors in parallel - these equalise the voltage
across each capacitor so that they have the same voltage. Remember to verify the ripple current rating! This can be expected to be over 10A, and
under-rated capacitors will blow up.
Another difficulty is the bridge rectifier. Although 35A bridges would seem to be adequate, the peak repetitive current is so high that they may not be
up to the task. I suggest that you use two (or even three) in parallel as shown. The bridge rectifier voltage rating should be a minimum of 400V, and
they must be mounted on a substantial heatsink.

Note that the supply shown is suitable for one amplifier only! For a stereo version, you will need a transformer with double the rating for a single
amp, and double the capacitance. Although a standard P39 soft start circuit controller board will work fine, the resistors will need to be upgraded. The
series soft start resistor should be around 33 ohms, and rated at 50W or more. As you can see, the power switch simply applies low voltage AC to the
auxiliary supply bridge rectifier and to the soft start circuit via relay contacts. The relay is located on the control board which also has DC and
thermal detection.
The additional 5V supplies shown will give a small increase in peak output power, but may be omitted. With the extra voltage, peak power is about
2,048W, vs 1,920W without it. While this may appear to be worthwhile, in real terms it is only 0.5dB more. You will gain far more by using heavier
gauge mains and speaker leads or a different power outlet.
DC Protection - You cannot use output relays with this amplifier! Should a DC fault be detected at the output, the only option is to switch off the
power. A relay that will withstand breaking 115V or 150V DC at 25A or more is going to be hard to get, and extremely expensive. Although the
speakers will be subjected to the full supply voltage until the filter caps discharge, as the builder of the amp, you are confident that they will
withstand the power.
Me, I'm not so sure.
You will notice that the AC mains is specified for 230/240V only. Use at 115V is not recommended because of the current. At full power, the amp
will draw a minimum of 10A (slightly above the transformer rating), but with 115V, that will increase to over 20A. The losses are too high with that
much current, so in 115/120V countries, I suggest that the amp be connected to a two phase power source as a matter of course. Even at the higher
supply voltage, the limit for a standard power outlet in Australia is 10A, so the continuous power input is limited to 2400W, and continuous power
output will be substantially less than this.

Figure 3 - Protection and Control Circuits

The control circuit need not be complex, but is very important. Although P33 could be used for DC detection, it would be better to use a dedicated
circuit (which has not been designed yet). The thermal sensors can be transistors or dedicated ICs, and a simple comparator circuit detects that the
temperature is above the trip value of 35°C to turn on the fans. The fans need to be high output types, as they will be called upon to dispose of a
prodigious amount of heat when the amp is being pushed. Thermal switches act as a backup - if the fan controller fails to operate for any reason,
normally open thermal switches will start the fans.
Water cooling is a viable option for an amp like this, especially for long term high power usage in a fixed location.

Construction
If you decide to build this amp, you will be prepared to spend a lot of money and time. You will also have sufficient experience to be able to work out
the construction processes yourself. For a single channel, the parts alone will cost upwards of AU$1,000, probably closer to AU$1,500 or even more.
Just the power transformer is likely to be around $250-300, and there's probably another $300 or more for filter capacitors. You will need a heatsink
rated at about 0.03°C/W with forced cooling. I cannot suggest a suitable heatsink, but you can be sure that it will be large and expensive.
Please note that this project is provided as information only, and I will not provide any assistance to prospective builders. The entire project is your
responsibility if you want to take it on.
Testing
If you are crazy enough to build this amp, then you will have sufficient skills to be able to work out what is needed to test it. Remember that the
smallest mistake could easily despatch many expensive transistors, blow the tracks off a PCB, melt wiring, and all manner of other distasteful
possibilities.
As with any high powered amplifier, initial testing should be done with a current limited power supply and no load. The amp will be functional with
as little as ±10V or less, and the power supply and complete amplifier must be tested using a Variac (bypassing the soft start for initial tests). The
Variac needs to be rated the same as the power transformer (i.e. at least 2kVA).
I will leave the remainder of the test procedure to the constructor, since the only people who should even attempt building an amp of this power
should be very experienced with high power systems. If this does not describe you, then don't even think about it.
1 - Introduction
If I never see another loudspeaker rated at 1,000W (or more) again, it will still be too soon. Well apart from that fact that no voicecoil can withstand
that kind of power for more than a few seconds or so without self destruction, why would anyone think that a 1kW loudspeaker was a good idea?
In the first instance, just consider a typical loudspeaker voicecoil. It is typically wound on some type of cardboard, thin aluminium, Kapton,
fibreglass or some other similar material. I have never seen a ceramic or quartz voicecoil former, but that's what would be needed to take the
temperatures involved at such an insanely high power - not to mention the wire and insulation used. I doubt that asbestos insulation would be
considered a good idea these days. Think in terms of a typical old-style bar radiator or an electric toaster. These were/are typically around 1,000W

and the resistance wire glows red hot (not surprisingly, this is the whole idea ).
Outrageous power ratings for both amplifiers and loudspeaker drivers are like maximum top speed for cars - many people would love to have a car
that can do 300km/h even though it is illegal in most countries to get even close to the maximum (for example, the absolute maximum in Australia is
110km/h).
These power ratings for amplifiers and speakers are designed to appeal to those who have no understanding of efficiency, and think that power is the
only thing that matters. For such people, a 1000W speaker must be better than a 200W speaker. What they don't understand is that a 200W speaker at
100dB/W/m is louder than a 1000W speaker at 90dB/W/m - the higher efficiency driver will achieve 123dB with 200W, vs.120dB for the 1000W
driver. This is ignoring all losses, which are dramatically higher in the high power speaker - see below to find out why.
Demonstration videos of low frequency drivers accepting vast amounts of power can be found on the Net, but prove nothing. They
are simply marketing ploys, designed to convince the buyer that the claims are real. Those that I have seen use the driver completely
open - no box, and not even a basic baffle. This ensures maximum cone movement and maximum cooling because fresh air can
circulate.

In addition, all that is needed to 'prove' the point is to operate the driver at resonance. The resonance impedance may be 10 times
that at other frequencies, so the amplifier output voltage is meaningless as a measure of power. If the impedance is 40 Ohms at
resonance (for a nominal 4 Ohm driver), the nominal voltage is ~63V RMS (1kW / 4Ω), but at resonance the actual power is only
100W for that voltage. This same test proced
ure can be used with the driver in an enclosure, but the drive frequency is simply increased to match the driver's resonance in the
box.
If you want to burn out a competitor's product for the demonstration, simply drive it at a frequency far enough from resonance to
give a spectacular looking failure. I leave it to the reader to figure out if anyone could be so dishonest as to do this.
So, what alternatives are there? Read on - this article explains the issues with dynamic drivers, and shows the deficiencies with
many high powered loudspeakers. There are drivers that claim to take 1.5kW continuous power, yet the parameters of one such
driver examined simply will not allow this much power to be used at low frequencies without exceeding the maximum excursion (to
the point of damage). Further to this, the driver parameters are such that the actual performance cannot be optimised for any known
enclosure - in short, the driver is a pig, and it is extremely difficult to make it perform well regardless of the enclosure type. The
driver in question will not be named, but was 'commended' to me to prove that I am wrong. It has done nothing of the sort
(predictably), because the driver is designed solely to appeal to those who continue to think that power is important. Worth noting is
the fact that as a subwoofer in a 300 litre sealed box, power is limited to less than 300W below 40Hz - a true sub needs to be able to
get to 20Hz to warrant the name "subwoofer". Drivers such as this are the equivalent of putting a Formula 1 engine with full race

tuning into a small sedan, and wondering why it kills you on the first corner . The combination simply doesn't work properly, and
there is no point pursuing such silliness.
While it may appear that many of the calculations in this article are based on the type of SPL (Sound Pressure Level) usually needed
only for large scale public address, the same things apply for audio and home theatre. The effects are reduced because of the lower
sound level normally used in a home environment, but are no less real since domestic loudspeaker drivers are normally rated for
significantly lower power and efficiency than professional drivers.
There is absolutely no good reason that anyone should imagine that a loudspeaker driver capable of 1kW is a good idea - it isn't
now, and never was. There are so many other areas in audio where outrageous claims are made - the proliferation of PMPO
advertising power (having no connection whatsoever with reality), stunning lies about the importance of "specialty" cables in
systems, "magic" components - the list is endless. Very high power, depressingly low efficiency loudspeakers are just another thing
to create FUD (Fear, Uncertainty & Doubt) for buyers and DIY people alike. I hope this article helps a little.

This article has been in preparation for a considerable time. What originally looked like a relatively simple task turned out to require a lot of time, effort and
perseverance. From the initial idea to publication has taken well over a year, and even now, there are bound to be some (hopefully) minor errors. The tests
were conducted exactly as described, but lack of the very sophisticated equipment required to guarantee complete accuracy (especially with magnetic
compression) means that some errors are inevitable.

The above notwithstanding, the results do show the effects as described, and the article is intended to inform, not to criticise or endorse any manufacturer or
specific product. All effects described are real, and although some may seem "off-the-wall", all results are measured - not simulated or obtained theoretically.

2 Efficiency
Consider that the average high efficiency loudspeaker is typically no more than about 5% efficient. This means that only 5% of the
applied electrical energy is converted into sound, the rest is dissipated as heat from the voice coil. This 5% efficiency speaker will
be rated at 99dB/W/m - this is much higher than normally achieved.
If we could get one, a 100% efficient (direct radiating) speaker would convert 1W of electrical energy into 1W of acoustical energy.
This will give us 112dB SPL (at 1W, 1 metre, when radiating into half space). Since no such loudspeaker exists, we must use what is
available. Typical hi-fi loudspeakers are typically around 90dB/W/m - only 0.62% efficiency! 99.38% of all applied power is wasted
as heat.
At one stage, professional sound reinforcement speakers were commonly around the 100dB/W/m efficiency level, but this is now
rare. Only a few of the traditional professional manufacturers (and a small number of specialist speaker makers) have drivers this
efficient, and herein lies the problem (or part of the problem).
There is a realisation that 'power is cheap', and this is quite true. High power amplifiers are now very cheap compared to even a few
years ago. Since power is so cheap, loudspeakers with efficiencies even below 90dB/W/m are common - all you need to do is use a
more powerful amp and everything is back where it should be, right? Wrong!
Since the majority of all electrical power is converted to heat, the higher the power applied to a speaker, the more heat you have to
get rid of. The typical loudspeaker is not a good design for heat disposal, and many of the more dedicated manufacturers have gone
to extreme lengths to get the best possible cooling for their driver's voicecoils.
Even so, there are limits. These limits are physical, metallurgical and chemical, and no amount of marketing hype will change any
of these. The adhesive that bonds the voicecoil to the former has a difficult job, high temperatures, often extreme forces acting upon
it, and high vibration levels all stress the adhesives. It is not uncommon for a voicecoil to reach (or exceed) 200°C, and the more
power that is wasted as heat (because the speaker is inefficient) the more power you need to put into it to get the sound pressure
level (SPL) you had when the voicecoil was cold.
 Figure 1 - Basic Loudspeaker Motor Construction
Figure 1 shows the typical basic construction of the loudspeaker motor. Various proprietary variations exist, but the essential
elements remain much the same. The voicecoil has two ways to get rid of heat - radiation and convection. We can forget convection,
as there is nowhere for the hot air to 'convect' to, other than within the motor assembly. While the pumping effect of the cone's
movement does help to move the air around, in many cases there is actually nowhere for the air to go. Where the 'self cooling' effect
is designed well, this only works at low frequencies where there is significant cone movement - at higher frequencies the cone travel
is such that there is little or no pumping effect at all! Radiation will make the rest of the motor hot as well, but at least that has
enough area to get rid of some of the heat. Some manufacturers use aluminium baskets to support the speaker's motor components,
and this will act as a heatsink. One maker even has the 'basket' in front of the cone so it won't be trapped inside the box. Finned
magnet covers are fairly common now, and virtually all drivers that claim to be able to handle appreciable power use a vented pole
piece as shown.
But is this enough? How hard is it to dissipate heat into the surrounding atmosphere? What other options are there? Very few,
unfortunately, and this is a part of the overall problem. The problem is made worse with drivers with a low efficiency, because for a
given SPL, more power is needed right from the start.
Consider this ... Assume we have a loudspeaker rated at 90dB/W/m (a softspeaker?) versus another rated at a much more
respectable 100dB/W/m. With one Watt of electrical energy applied, one will have 10 times (10dB) the SPL of the other. While this
is insignificant if we are happy with 90dB SPL, if we try to obtain 110dB SPL at one metre, the efficient driver will do this with
only 10W, while the inefficient driver needs 100W. Another 10dB makes that 100W vs. 1000W - anyone want to guess which
speaker will last longer before the voicecoil melts?
There is another more insidious aspect to this. So far, we have assumed that the electrical input to SPL ratio is constant, but it most
certainly is not. As a voicecoil gets hot, its resistance rises. This increases the impedance of the speaker, so less electrical energy
goes in, and less acoustic energy comes out. The 1000W amp needed to drive our inefficient speaker will probably be delivering
half that (because of the increased impedance) by the time the voicecoil is ready to depart this world, so SPL is increased by only
7dB instead of the 10dB we expected when we applied 10dB more signal level. Meanwhile, the efficient driver only has to dissipate
100W, there is less heating, and consequently less relative drop in level as power is increased.
Welcome to the real world of 'power compression'. JBL [1] has performed tests showing that power compression can reduce output
by anything from 3dB to 7dB from the expected SPL at elevated temperatures. Seven Decibels! Remember that each 3dB means
double or half the power, so 7dB is more than 4 times. You use a 1kW amp on a speaker and expect it to be pretty loud (not an
altogether unrealistic expectation), but if another speaker can be just as loud with only 100W, then which one is preferable? Loaded
question ... of course the more efficient driver will be the better choice, all other things being equal.
Remember the bar radiator from the opening paragraphs? How long can a voicecoil survive with 500W or 1kW or more being
dissipated as heat? If the programme material has plenty of bass, the cone will move, and that will push air through the magnetic
polepiece gap and past the voicecoil. This will certainly help cool things down, but where does the hot air go? Into the cabinet, to be
sucked back through the gap next time the cone moves outwards? That's not very useful. It is fairly obvious that as a solution to

maintaining a sensible voicecoil temperature, this method sucks (pardon the pun ).
The nature of music helps us here. Music has (or should have) loud bits, soft bits and even silent bits. This diversity is called
dynamic range - a term that describes any signal that is not a continuous waveform. Dynamic range is simply another way to
describe the peak to average ratio.

Figure 2 - Typical Audio Waveform

The average power delivered to a system is where we get the SPL from - this is averaged over a period of time, and accounts for
asymmetrical waveforms, brief bursts of high levels followed by periods of lower levels, impulse signals, etc. It is not uncommon to
find the crest factor at around 3:1 - peaks are 3 times higher than the average level. This is approximately 10dB.
A 10dB in voltage or current is a ratio of 3.16:1, and 10dB with power is a ratio of 10:1. A change in voltage ratio of 3.16:1 causes a change in voltage, current
or power of 10dB. For those who do not fully understand the relationships of dB, I suggest you read Frequency, Amplitude & dB. In brief ...
 dB (Power) = 10 * log ( P1 / P2 ) = 10 * log ( 10W / 1W ) = 10 * log ( 10 ) = 10dB
dB (Volts) = 20 * log ( V1 / V2 ) = 20 * log ( 3.16V / 1V ) = 20 * log ( 3.16 ) = 10dB

This remains an area where people regularly become confused, but once you know the way dB is calculated it all falls into place.
Some recorded material has more dynamic range, some less, with typical values between 6dB and 20dB - the lower figure is more
likely on modern, highly compressed material.
To reproduce a signal with a 10dB crest factor cleanly (without clipping distortion) means that if your average level requires 10W,
the peaks will need 100W - a 100W (minimum) amplifier is needed to get 10W of clean undistorted average electrical energy. If you
use a low efficiency driver (such as the softspeaker described above), then these figures could be 100W and 1000W respectively!
To determine the level of power compression, it is necessary to know the thermal mass of the voicecoil assembly, the rate of heat
transfer, plus a whole swag of other things that the manufacturer does not tell us. Alternatively, it can be measured, albeit with some
risk to the driver itself. This is the quickest and most accurate way to figure out just how much thermal compression a given driver
exhibits. While we are at it, we'll also use an indirect method to measure the voicecoil temperature.

3.0 Temperature
Copper has a thermal coefficient of resistance of 3.95E-3 per °C. Therefore, if a voicecoil has a DC resistance of 6Ω at 25°C, at
200°C this will increase to ...
RT2 = RT1 * (1 + α(T2 - T1))
where T1 is the initial temperature, T2 is the final temperature, and α is the thermal coefficient of resistance. Substituting our values
in the above equation we get ...
R200 = R25 * (1 + 3.95E-3(200 - 25)) = 10.15Ω
There seems to be some discrepancy as to the actual coefficient of resistance for copper - figures found on the Net range from 3.9E-3
to 4.3E-3. I have adopted a middle ground, settling on 3.95E-3. Feel free to use the value with which you are most comfortable. Note
also that the coefficient of resistance does change depending on whether the copper is hard drawn or annealed.
If we know the change in resistance, then it is a relatively easy matter to calculate the temperature, provided we have a reference
resistance taken at a known temperature before the test.
ΔT = ΔR / (RT1 * α)
Where ΔT is change of temperature and ΔR is the change in resistance. For the previous example the change in resistance is 10.15 -
6 = 4.15 Ohms, so we get ...
ΔT = 4.15 / (6 * 3.95E-3) = 175.10°C
Since we started at 25°C, that gives a final temperature of 175.1 + 25 = 200°C (close enough). The only way to determine just how
long it will take for the voicecoil to reach this temperature is by measurement. Although it is possible to calculate it, this would
require far more information than you will be able to obtain from the maker, and far more maths than I am prepared to research and
pass on.
It is usually safe to assume that the temperature rise will take less than 30 seconds for most drivers - the thermal inertia of the
voicecoil assembly is normally quite low, but some low efficiency subwoofers may have relatively heavy coils and formers, thus
increasing the thermal inertia. This is not a bad thing for a driver that handles intermittent bursts of high power with long rest
periods in between, and it is the very nature of the programme material that allows many of these drivers to survive the power
applied.
The above assumes that resistance measurements are taken electrically - to obtain an accurate result requires that
you monitor the average voltage and current applied to the speaker. The voicecoil resistance may then be
determined using Ohm's law. Because of 'flux modulation' (see section 4), simply measuring the change in SPL is
an unreliable method for calculating the voicecoil temperature, and should not be used.

3.1 Thermal Compression

Let's say that we have a requirement for a continuous average SPL of 115dB at 1 metre. Such a system might be used in a movie
theatre (for example), and by the time the signal gets to the audience, the average SPL might be reduced by about 30dB (due to
"room loss" and distance) - around 85dB SPL within the theatre itself (this is the reference level for theatre systems).
From the above, it is quite obvious that as the voicecoil gets hotter, less power is delivered. Provided the signal is applied for long
enough and the heat can be removed at a rate that prevents meltdown of the voicecoil, the average SPL will be reduced - after
meltdown, SPL will be reduced to zero!. Based on what little information is available from manufacturers, it seems that the
loudspeaker voicecoil will reach thermal equilibrium within around 20-30 seconds. High frequency drivers will be faster (because
of the low thermal mass of the voicecoil assembly), and very large subwoofers will most likely be slower, as the voicecoil assembly
will be substantially heavier, and thus have a greater thermal mass.
If we work only within the constraints of the maths shown above, we can arrive at a good estimate of the final efficiency of a driver
that is being pushed to its limits. Consider the example of an 8 ohm driver, having a DC resistance (DCR) of 6 ohms at 25°C. The
ratio of nominal impedance (Z) to DCR is 8 / 6 = 1.33:1.
At a voicecoil temperature of 200°C, the resistance increases to 10.15 ohms, so the nominal impedance increases to ...
Z = DCR * ZRatio = 10.15 * 1.33 = 13.53 ohms
Neglecting (for the moment) any other effect, that means that if the nominal 8 ohm driver were to be supplied with a signal of 50V
RMS (average), that will work out to 312.5W at 25°C. After around 20 seconds that power will fall to ...
P = V² / Z = 50² / 13.53 = 184.8W
That represents a drop in power (and SPL) of 4.56dB, less than half the power you thought you had. A 90dB/W/m driver has fallen
to 85dB/W/m (close enough). Now, this is a simplification, but the actual power and SPL will be very close to the values calculated.
Where you expected the speaker to produce about 115dB SPL at 1 metre, after only a short period it will only give you a tad over
110dB SPL - a significant decrease. To achieve 115dB SPL, this driver now needs 1kW average (a 40dB increase over the
85dB/W/m effective sensitivity), but that will just cause the voicecoil to get hotter still, and it will fail - most likely without ever
achieving the target sound level on a long term basis.
Using the same expectation (115dB SPL), but substituting a 100dB/W/m driver, we can see that only 32W is needed to achieve the
115dB SPL required. With only 32W, there will be very little thermal compression - perhaps 0.5dB worst case - so power has to be
increased to a bit below 40W to compensate. Although the extra power will cause the voicecoil to get a little hotter (and so reduce
the actual power and SPL a little more), it is well within the capacity of a sensible sized amplifier to cope with.
Remember that we already established that the peak to average ratio is typically around 10dB, so the 90dB/W/m driver will need an
amp capable of 10dB more power than the average level (a rather daunting 3,000W!), while the 100dB/W/m unit will only need
10dB over and above 40W - namely 400W. It takes little imagination to realise that the lesser (and probably cheaper) speaker is no
bargain after all, and will almost certainly fail if a 3kW amplifier is used. All this, and it is still over 4.5dB shy of the 115dB SPL
expected of it. In my books, that represents a travesty, not a bargain. Looking at the data in a table helps to see the information at a
glance ...
Sensitivity Power Power Thermal SPL
(dB/W/m) (Average) (Peak) Compression (Actual)
90dB 300W 3,000W 4.56dB 110.4dB
100dB 40W 400W 0.5dB 114.5dB
Table 1 - Power Requirements for 115dB SPL at 1 Metre
This is simply a theoretical exercise, but the effects are both real and demonstrable, and the above is not at all unrealistic. Every
dynamic driver will suffer from thermal compression, because the voicecoil winding has no choice but to get hot when it is
dissipating a lot of power. Only a very few (mainly professional) loudspeaker manufacturers treat this as seriously as it deserves.
Covered in Design of High Quality Passive Crossovers (ESP website) is the effect of the changing impedance on a passive
crossover network. Thermal effects are wide ranging and rather insidious, changing the way the speaker system sounds depending
upon the power applied.
It is only by choosing a driver whose efficiency is matched to the requirements that the requirements have even the slightest chance
of being accomplished in practice. Quite obviously, a higher efficiency loudspeaker driver will need less power to achieve the result,
but not so obviously, high efficiency should be sought whenever possible - it will always give a better result (all other things being
equal).
I built the test voicecoil assembly (1.45 Ohm at 20°C) pictured below. This coil was subjected to a range of currents from 1 to 3A.
DC was used for all thermal tests so that there was no chance for inductance to influence the readings. The polarity was arranged so
that the voicecoil was pulled into the gap. Since the coil former was deliberately designed to sit against the rear plate, this removed
the necessity to clamp the coil in position. The purpose of the winding around the outside of the magnet will be explained in Section
4.
 Figure 3 - Test Motor Assembly
The test motor is not a powerhouse, but neither is it insignificant. It uses a 95mm diameter magnet, 15mm high. The front plate is
5mm thick, and the rear plate 4.5mm. The gap measures 1.65mm, having a centre pole of 25mm diameter. This driver would
originally have been rated for around 25W continuous power, with perhaps 100W peak power rating. Unfortunately, the exact
details are long gone, and the magnet assembly is all I have left of the speaker. By some 'standards' that seem to be applied today, it
is probable that my estimations are much too low - a small car loudspeaker I have lying around has a motor that is less than ½ the
size, but is rated to 80W peak power - in someone's dreams!
The measured values shown below represent a low power dissipation across the range. The voicecoil was in intimate contact with an
aluminium former which was in direct contact with the centre polepiece. This means that cooling was far better than would
normally be the case (although there can be no 'forced air' cooling because the voicecoil was not allowed to move during the tests),
yet the resistance range is considerable. I did try the coil withdrawn from the magnet assembly (so it had no cooling from the gap),
and it became extremely hot within a few seconds.
Using the formula from above we can determine how hot the coil actually became at each current ...
Current Voltage Power Resistance Temperature
0 0 0 1.45Ω 20°C
992 mA 1.54 V 1.528 W 1.55Ω 37°C
2.006 A 3.66 V 7.34 W 1.82Ω 85°C
2.91 A 6.68 V 19.44 W 2.295Ω 167°C
Table 2 - Voice Coil Resistance Vs. Power
Remember that these figures were easily reached at very low power - 20W is well within the normal range of a domestic system.
Based on these figures, we see a resistance change from 1.45 Ohms to 2.295 Ohms at just 20W - almost a 60% increase. An
impedance increase of 60% will certainly cause havoc with a passive crossover, and the power compression will be audible unless
all drivers have the same or very similar impedance increase. Even at an average power of only 1.5W there is an impedance increase
of almost 7%, which would take an 8 Ohm driver to 8.55 Ohms - enough to measure the effect on crossover performance, and it
may be audible with some material (that's about 0.57dB impedance change).

4 Magnetic Effects
Note that this section deals primarily with loudspeaker driver efficiency - distortion mechanisms abound, and the magnetic circuit
(flux modulation in particular) is a major contributor to the many different and exciting ways a dynamic loudspeaker can modify the
waveform, and thus introduce frequencies that were not present in the original signal (distortion). The distortion factors have
received more attention than the other effects discussed here, but this is still an area where most driver manufacturers would prefer
to remain silent. Nonetheless, a search for flux modulation reveals a surprisingly large number of documents, although not all are
useful, and too few link to loudspeaker manufacturer's websites. In nearly all cases, only the distortion mechanism is discussed, but
there's more ...
It is within the magnetic circuit that we see effects that are either ignored completely or glossed over. While most professional
loudspeaker manufacturers take the magnetic circuit seriously, the vast majority of general purpose driver makers do not - pressed
steel pole pieces that are unacceptably warped being a very common problem. (I have seen a rear plate (brand new) that had 0.5mm
of wobble when placed upon the magnet. That represents a significant air gap in the magnetic circuit. The front plate wasn't much
better.)
The essential parts of the motor's magnetic circuit were shown in Fig 1, but as noted this is a fairly generic arrangement. To make an
efficient loudspeaker means that the magnetic circuit must be optimised. An example of this optimisation is shown in the JBL paper
[1], and it is obvious that a great deal of thought and research has gone into the design of the magnetic circuit to ensure the flux
density across the gap is as high as possible.
Consider the effect of minute air gaps between the front and rear plates and the magnet itself. Any air gap (or anything else that has
low permeability to magnetic flux - e.g. adhesive) will reduce the effectiveness of the magnetic coupling between the magnet and
the plates, and hence to the gap itself. This weakens the flux across the gap, and in turn reduces efficiency. Of particular importance
is any part of the magnetic circuit that is saturated (a condition where the material will not accept any more magnetic flux) - an
excellent example can be seen at the Infolytica website, where saturation is shown in the rear plate of a loudspeaker motor.
Now, consider the effect when there is current in the voicecoil. There are now three different sources of flux in the gap and the
magnetic circuit as a whole ...
1. The static flux from the magnet assembly
2. The flux from the voicecoil - varies with the voicecoil current
3. Flux generated by eddy currents in the pole pieces - depends on voicecoil current and position
The force produced by the current in the voicecoil creates a magnetic field that uses the static flux in the gap as its only means of
propulsion. The static flux is not a solid! It will bend when an opposing magnetic force is applied, and the amount by which it bends
is determined by the static flux density and the voicecoil current. Likewise, the magnet flux will be modulated - this isn't solid
either, and that's what the coil around the magnet in Figure 3 was used for.
The first thing I had to do before useful tests could be continued was rewind the voicecoil - the thermal tests (and some initial
magnetic tests) had damaged the wire insulation causing shorted turns. The second version had a measured impedance of 2.11Ω
when installed in the gap.
The test motor was driven with varying levels at 400Hz, and the modulation of magnet flux density was monitored by the outside
winding. The coil was locked for all tests. As the voicecoil is driven, it is logical that the flux produced by the coil must traverse the
magnetic circuit, and the external coil picks up the variation. In an ideal situation, the induced signal should be a smaller replica of
the modulating voltage (and current). Smaller because the outside of the magnet is a good distance from the mean magnetic path, so
the variation of field strength can be expected to be a lot less than that within the magnet itself. A replica means that there should be
no additional distortion, and the voltage change on the outer coil should be directly related to the drive voltage.
Two interesting things were revealed by this test, and while one was pretty much expected, I didn't notice the other until I had run a
number of tests. That a small signal would be picked up was completely expected, and likewise I figured that it may not be linear.
The part that I almost missed was the distortion of the flux at higher drive levels - the waveform is shown in Figure 4. Why did I
almost miss it? Simply because it was obscured by the distortion in the applied waveform, and I was concentrating on measuring the
amplitude. The test procedure was amended after I saw this - the original drive signal was 50Hz, derived from a Variac. I then
switched to using a 400Hz signal, amplified with a P68 subwoofer amp to get the voltage and current (with minimum noise and
external distortion) needed to repeat the original tests with a clean sinewave. The coil is a difficult load, having an impedance of just
above 2 ohms.
Tests were also performed on a pair of real loudspeakers - the test motor is one thing, but to be meaningful a magnetic assembly
needs to be tested in a working condition. Figure 4 shows the waveform obtained from a Vifa M13MH-08 driver (now out of
production), with 15V RMS at 400Hz applied. Although I collected test results for the other driver (a car speaker, rather
optimistically rated for 80W peak), the results were very similar to the Vifa and test unit.
 Figure 4 - M13MH-08 Motor Outer Coil Distortion
The distortion shown above started (in less severe form) at a relatively modest drive level. Distortion was measurable at very low
levels, and started to become clearly visible at about 7V. The waveform in Figure 4 is a screen capture of that seen at a drive level of
15V. The signal was applied in very short bursts to eliminate thermal compression, and while the waveform could be captured, I
could not measure distortion accurately. I did capture the levels using the FFT capability of my PC based oscilloscope, and the
second harmonic of the Figure 4 waveform is about 20dB below the fundamental (an exact figure is difficult because of the burst
waveform). The distortion is quite visible - note that the positive and negative peaks are of different amplitudes. If you look very
carefully, you will see that the level (look closely at the peaks) is starting to fall after only 8 cycles. This is thermal compression
starting to show !
15V RMS is equivalent to only 28W (based on the nominal impedance), so one doesn't really expect 'bad' things to happen. There
was measurable distortion even with less than 1V drive in all motors tested. It is certain that distortion started earlier than this, but
the signal level from the outer coil was too low to get an accurate reading. Where I was able to measure the distortion, this is shown
in Table 3. Measured distortion in the outer coil was almost all third harmonic, until the waveform started to become asymmetrical.
The applied signal has a distortion of about 0.04%.
The effect of changing the flux levels depending on drive level is often lumped together with other magnetic effects and collectively
called 'flux modulation', and it works alongside power (thermal) compression to reduce the efficiency of the driver at high power
levels. There are several AES papers that discuss the magnetic circuit, but unfortunately they are not available except at
considerable cost - while not expensive per se, they are expensive if it proves that they do not contain any information you need.
One thing that is very apparent (from examination of the data available from the few magnetic simulation tool suppliers world
wide), is that the traditional motor assembly depicted in Figure 1 is flawed. In nearly all cases, the back plate will saturate, reducing
the available flux at the gap and (probably) causing asymmetry of the voicecoil induced flux.
Voicecoil (V RMS) Outer Coil (mV RMS) Distortion
1 6.57 0.8%
2 13.3 1.1%
3 21.3 1.3%
4 30.8 1.6%
5 39.3 1.9%
6 48.3 2.1%
7 56.7 Visible
14 126 Moderate
15 135 See Figure 4
Table 3 - Test Motor Results
The distortion figures in Table 3 show what I was able to measure - using the test motor rather than a real speaker. Above 6V, the
coil became too hot too quickly for a distortion reading, and even the figures shown for coil voltages above 3V are lower than they
should be, because thermal compression had already reduced the coil current noticeably. At 7V, the onset of distortion was visible
on the oscilloscope, and at 14V saturation was noticeable. At 15V it appeared as shown in Figure 4 - this is a sure sign that there is
magnetic saturation because the waveform is asymmetrical. All three motors tested this way showed identical behaviour.
Unlike thermal compression, the flux modulation (or flux compression) effect is not limited by time - it is instantaneous. If a single
short duration burst is applied at 10W, we may measure a peak instantaneous SPL of 100dB (for example). Increase the power by
10dB (100W), and instead of 110dB SPL, we may measure 109dB - an instantaneous loss of 1dB. Sustained power will then heat
the voicecoil so thermal compression adds to the problem. After perhaps 10 seconds, the SPL may fall by a further 5dB, giving a
total loss of ~ 6dB SPL. The speaker has effectively become 6dB less efficient than expected, and requires 4 times as much power
as we thought we'd need. This additional power will only cause more of the same effects, and the process is a vicious cycle - the
more power we apply to overcome the magnetic and thermal losses, the greater the magnetic and thermal losses become, so
requiring even more power. This can continue until the smoke is released from the voicecoil, at which point we have an ex-
loudspeaker.
Thermal compression was a problem during magnetic circuit tests too, and this was another reason I could only use short bursts of
signal for testing. With an applied voltage of 5V RMS, the outer coil voltage fell from an initial value of 39.3mV to 31.8mV over a
period of about 5 seconds. That's a 1.8dB fall at an initial power of 11.8W (falling to 9.5W in 5 seconds).
For every action, there is an equal and opposite reaction. This applies in all areas of physics, and the loudspeaker magnetic circuit is
no exception. Fig 6 shows Fleming's 'left hand rule' applied to a voicecoil in a loudspeaker motor. From this we see that the
voicecoil exerts a magnetic force against the static field set up across the gap. It is not sensible to assume that the static field is
unaffected.
 Figure 6 - Left Hand Rule and the Loudspeaker Motor
Based on the information in the references, it would seem that unless one goes to fairly extreme lengths to get it right, flux modulation can have a
profound effect on the instantaneous efficiency of a loudspeaker. Saturation of the pole-pieces in particular should be avoided, but few loudspeaker
motors are designed to prevent it.
The modulation can be reduced only by reducing the voicecoil current or by increasing the static flux density - thereby increasing its resistance to
bending forces. In addition, the magnetic circuit must have sufficient reserve capacity to ensure that it never saturates. In a cycle of audio signal, the
voicecoil current ...
• Reinforces the static field in the top plate and opposes that in the centre pole during one half-cycle, and
• Reinforces the static field in the centre pole and opposes that in the top plate on the other.
If the pole piece(s) are already close to saturation (where they cannot sustain any further magnetic 'lines of force'), the field strength cannot be
increased and decreased by the same amount in the centre pole and top plate, so the waveform will be distorted and will also lose some efficiency.
Although I have a magnetometer, unfortunately it is not only uncalibrated, but as I discovered during tests designed to prove the above, it is also non-
linear at high field strengths as used in a loudspeaker. This made any tests based on its use rather pointless - hence the coil around the outside of the
magnet.

Figure 7 - Principle of Motor Action

The figure above is from one of my old textbooks [4] showing the exact parameters that affect the operation of a loudspeaker. The wire diagrams
show a + to indicate that the current is moving away from you (conventional current flow, from positive to negative), and the dot means it is flowing
towards you.
A quote from the text ...
The basic principle of the conversion of electrical energy to mechanical energy in a motor rests upon the fact that when a current-carrying conductor
is placed in a region occupied by a magnetic field (unless the direction of current and the direction of the magnetic field are parallel), a reaction is
set up that tends to move the conductor out of the field. This principle is illustrated in the diagrams of Fig. 14.1. (Figure 7 above)
The effect of any external field on a static magnetic field must cause the static field to be deformed. This deformation is a part of 'flux modulation',
and a considerable amount of the effect will be found in the gap - the magnet itself will normally be relatively immune from demagnetisation caused
by voicecoil current (although Alnico magnets have apparently been known to have been demagnetised by excess voicecoil current), but the magnetic
path itself is another matter altogether. This is very difficult to measure, and I was unable to detect any variation in actual magnet 'strength' during my
tests. Highly specialised equipment is needed for these tests, and some further information is available from Reference 5.
As the voicecoil moves, there is distortion of the static flux because instantaneous movement of the coil, former and cone is not possible. Movement
is impeded by inertia and the loudspeaker's suspension - there is also air loading on the cone, but this is comparatively insignificant. To minimise
distortion of the static flux, moving mass and suspension stiffness must be reduced to the minimum - these factors are a matter of compromise, based
on the end use of the driver.
With low flux density across the gap and/or a wide gap, it is logical that this makes it easier for the voicecoil flux to force it 'out of the way'. Many of
the current crop of subwoofer speakers will have this problem - a heavy cone means that by the time it has started to move, the signal induced flux
will have distorted the static flux significantly.
This concept is easily demonstrated. Take a pair of magnets, and align them so that they oppose each other. At low field strengths (magnets a fair
distance apart) it is easy to push them closer together. As they get closer, the forces become greater, and far more effort is needed to move them that
last millimetre than was needed when they are further apart. The same thing happens (but the other way around) with the magnets attracting. It's easy
to keep them apart when they are some distance from each other, but when they get close ... snap! (Be very careful if you use neodymium magnets -
they can really hurt if you get skin caught between them.)
That this form of flux modulation will reduce (instantaneous) efficiency should be quite apparent, and the wide voicecoil gaps favoured by modern
speaker manufacturers will make the situation many times worse (magnets further apart). The wide gaps are used because this makes the speaker
cheaper to make, having no close tolerances and thus requiring no skilled assembly workers.
It is also worth noting that because magnetic flux is not solid, the voicecoil may not be fully immersed in the magnetic field well before it actually
leaves the gap. This contributes distortion and loss of efficiency, since the total flux through the coil is not constant, and varies with applied current.
This may occur before the coil even starts to move.

5 Loudspeaker Tests
The speakers I experimented with are a mixed bag, with some fairly well known drivers and some that are rather less well known. They are also of
rather different vintages, ranging from very recent to many years old. This doesn't change anything, since so few manufacturers have ever published
power compression figures, even fewer have examined flux modulation, and just as few have tried to do anything about it.
All drivers were measured free field (without a baffle), and only the instantaneous compression levels were measured - some of the drivers I have in
my workshop are on loan, and I certainly never wanted to blow any of them. The test was arranged as follows ...
• Set the amplifier power such that a peak of 80dB SPL was measured with a tone-burst impulse, having 20 cycles at 877Hz (22.8ms), followed
by a rest period of 584ms. Mic distance was ~100mm to avoid reflections and keep power within reasonable limits.
• Once the reference level was set, increase the voltage by ~20dB (10 times the voltage, 100 times the power). This resulted in a typical peak
power of a bit under 20W for all drivers (hardly stressful one would think). Both the amplifier output and mic output were monitored closely
to ensure that neither distorted during the test.
• Measure the peak microphone output - it should increase by 19.35dB (the actual voltage change of the tone burst generator).
Not one driver I tested managed the 19.35dB increase, and remember that this test was specifically designed to keep the voicecoil cool enough to
prevent thermal compression effects - the first high power burst only was measured, and that was not long enough to allow voicecoil heating at a
level that would skew the results. I verified that heating was minimal by allowing the power test to run for some time, and no significant thermal
compression was noticed.
The driver measurements are shown in the following table. Note that all voltage levels are peak-to-peak (P/P). The measurements were taken using a
Philips PM382A Analogue/Digital oscilloscope. The signal processing capabilities of this 'scope make it ideal for detailed measurements at this level.
The microphone was an ESP measurement mic, and was powered from a phantom feed / preamp combination. All microphone voltages listed are at
the output of the preamp, which was not changed for the duration of the tests. Drivers were nominally 8 ohms (except #1 - 4 ohms), although the
actual impedance at the test frequency was not measured.
Test # Amp V80dB Mic V80dB Amp V100dB Mic V100dB Amp Change Mic Change dB Loss
1 (4Ω) 2.41V 811mV 23.30V 7.50V 19.707dB 19.321dB 0.386dB
2 2.41V 579mV 23.10V 5.47V 19.632dB 19.506dB 0.126dB
3 2.40V 516mV 23.10V 4.79V 19.668dB 19.354dB 0.314dB
4 2.43V 488mV 23.50V 4.51V 19.709dB 19.315dB 0.394dB
5 2.43V 448mV 23.30V 4.04V 19.635dB 19.102dB 0.533dB
6 2.43V 404mV 23.30V 3.72V 19.635dB 19.283dB 0.352dB
Table 4 - Driver Test Results (Ranked by Efficiency - Note that #1 is 4Ω)
In case you are wondering, the tone burst frequency of 877Hz was selected for two reasons. Firstly, the relatively high frequency ensures that there is
minimal cone excursion, so we can be certain that the voicecoil remains centred in the gap regardless of power level. Secondly, that happens to be the
frequency my tone burst generator provides (it is a fixed frequency type), so I had to choose between modifying it, building a new one that used an
external signal generator, or using what I had. Not a difficult choice given that I don't have as much time as I'd like for pure research.
Be aware that there is inevitably some margin for error in all acoustic measurements, especially at the lower (80dB reference) level.
I would not expect the error to exceed 0.1dB, and I made every effort to get the results as close as possible.

The likely magnitude of measurement error is seen in the small variation in measured difference of amplifier output level - the
maximum difference between all measurements is 0.077dB. Some of that is caused by differing driver impedance interacting with
amplifier output impedance and cable resistance. There are also limitations imposed by the digital oscilloscope.
Overall, most of the results are not too disappointing - Driver #2 was the best, 'losing' only 0.126dB. Driver #5 is the worst, with over 0.5dB loss. The
point to note is that all drivers lost some of the expected increase in level. It is very obvious indeed that had the peak power been increased to (say)
100W or so, these results would have been much worse.
Although there are a couple of exceptions, the higher the efficiency of the speaker, the less 'magnetic compression' is seen. Be aware though - I did
not test for distortion, and this can be quite high as a direct result of flux modulation. Also, because my test equipment (and environment) is not
optimised for loudspeaker testing, there is invariably some influence despite the mic being close to the loudspeaker. Although I took pains to ensure
that reflections were minimised, these effects cannot be eliminated completely without a fully anechoic test environment.
While it may have been nice to have been able to drive the speakers a lot harder to obtain a better indication, I didn't have the luxury of expendable
drivers or a suitable soundproof enclosure (the instantaneous SPL was high enough as it was, and I wore hearing protection).
6 Conclusions
Power compression comes in two distinct forms - thermal (long term) compression, and instantaneous flux modulation compression. Thermal
compression upsets the tonal balance of a multi-way system, as the driver with the greatest compression becomes softer with respect to the other
drivers in the system. The change in voicecoil impedance with varying temperature will also affect any passive crossover network, with high order
networks being the worst affected (they are very critical with respect to load impedance).
Flux modulation effects are instantaneous, and can affect any driver, although tweeters are less likely to suffer because the power is relatively low
even at high volume levels. The use of ferro-fluid may be of great assistance in this respect, although I was unable to test this. Flux modulation
causes dynamics to suffer and distortion is created, because the magnetic circuit cannot sustain the maximum flux across the gap as the signal level
varies. One of the big problems with flux modulation is that most people are oblivious to its existence, and the details shown here are almost never
discussed or published by manufacturers. It is possible that some of the driver manufacturers are unaware of the problem, let alone what needs to be
done to minimise it.
Even the limited tests I was able to perform show that flux modulation (magnetic compression) is quite capable of 'squashing' transients to some
degree. In an extreme case (assuming low efficiency drivers and considerable amplifier power), where transients should jump out at you, they may
blend into the overall mix, losing impact and removing some of the life from your music. The many owners of low powered Class-A amplifiers are
forced to use high efficiency drivers to get an acceptable sound level in their listening room. Although the amplifier is often cited as the reason the
systems sound good, one of the likely reasons should now be obvious - with no (or little) power compression of either form, high efficiency systems
will give much better transient (impulse) response and dynamics. There can be no doubt that these systems will have dynamics that are very difficult
to match by systems that require hundreds of Watts to achieve the same in-room SPL. Having said this, please bear in mind that at this stage there
appears to be little or no evidence to suggest that these effects are actually audible. They are measurable, even with relatively primitive techniques,
but it is quite possible that a blind A-B test would not reveal any problems at a sensible SPL. Other effects may be present which are audible, but not
related to the problem. This is a very difficult area, because it is very hard to isolate the effects as they are somewhat interdependent with other driver
parameters, and there does not seem to be any way the various effects can be isolated.
A general solution seems easy, if rather expensive and very limiting with most modern loudspeakers ... use high efficiency drivers. The lower the
power that is needed for a given SPL, the less compression the driver will create - be it thermal or due to flux modulation. Because flux modulation
effects are comparatively small - at least for domestic reproduction - thermal compression is by far the most dominant factor. For very low frequency
drivers, high efficiency is not possible - the moving mass usually needs to be fairly large to obtain a low resonant frequency, and this will always
have an adverse effect on efficiency. Despite the limitations, there seems no good reason that any driver should have an efficiency of less than
90dB/W/m - anything lower means that amplifier power has to be increased, and the problems then become apparent.
In some cases it may be possible to use multiple drivers to increase the effective SPL by creating a small array, which improves the on-axis effective
efficiency. By using more than one driver, the power needed by each is reduced for the same SPL, so the effects of both thermal compression and flux
modulation are reduced. There are many who claim that their arrays sound exceptional, and this may be part of the reason. Another alternative is
horns (loved and hated by a roughly equal number of people). Having very high efficiency, all power compression effects are reduced to a fraction of
that of direct radiating loudspeaker systems. Both approaches come at the expense of power response within the listening space, however few
loudspeaker systems have a flat power response anyway so this may not be as great a problem as may be expected.
Yet another possible solution is to use electrostatic drivers, since these have no magnetic circuit and are renowned for their dynamics. They are not to
everyone's taste though, and have a much smaller 'sweet spot' than most other speaker systems. Planar drivers rely on a much more distributed
magnetic circuit, so may also be an improvement, but I have no information to support this so it remains conjecture.
One thing that is readily apparent for dynamic (moving coil) drivers, is that the static field strength should be as high as possible. Typical flux
densities for (half decent) loudspeakers range from around 1 Tesla (10,000 Gauss) up to around 2.4T, and I would suggest that anything less than 1T
is next to useless. Very few drivers use magnetic materials that will provide much more than 1.8T across the gap - it seems to be accepted that mild
steel (as used by most of the cheaper drivers and many not-so-cheap drivers too) is unable to provide a gap flux density of more than 1.8T, regardless
of magnet strength - the now common use of dual magnets on subwoofers should be seen for what it is - a marketing ploy!. To obtain higher field
strengths requires the use of specialised alloys that are optimised for magnetic circuits. It is important that the static flux is many times stronger than
the dynamic (voicecoil) flux to obtain maximum performance.
Finally however, much as we may make a fuss about the theory and reality of magnetic compression effects (flux modulation), there is little or no
data to suggest that the dynamic sound quality of even low efficiency loudspeakers is considered lacking by the majority of listeners. Even the
distortion components that are introduced are barely audible [6] - if at all. Thermal compression is another matter altogether, and it should be obvious
to anyone that low efficiency is the curse of dynamics and reliability in high power systems. Although it will not often be a problem for domestic
systems, there are plenty of people who have experienced it first hand as the result of a party or similar gathering. The result is usually measured by
the number of speaker drivers that have failed. While it is commonly believed that the mere act of an amplifier clipping is the major cause, see the
article Why do Tweeters Blow when Amps Distort?, the real reason is sustained power, (and thus excessive heat) and all of the drivers in your system
are vulnerable.
Please note that the many references to JBL in this article are not intended to be an advertisement for the company or its products. While I do have
considerable respect for JBL, it just so happens that they provide more detailed information than anyone else - to not refer to the extensive data would
be to diminish the value of this article considerably.

Footnote - Passive Crossovers

Although it has been mentioned above and in many other ESP articles, passive crossover networks are affected by the driver impedance, and if the
impedance changes due to temperature, the crossover network is no longer accurate. Although this is not usually a major issue with domestic systems
used at relatively low levels, passive networks in conjunction with (very) low efficiency drivers can have a dramatic effect on the sound of the
loudspeaker system.
Passive networks have another issue as well - the coils used have resistance, and at moderate to high power levels, the effective series resistance can
increase further. This will not only affect the network's frequency accuracy, but may further diminish the damping factor for woofers. All commercial
products have to balance the component cost against the final selling price, and use of economical (rather than ideal) coils is not at all uncommon.
Consider a coil with a resistance of 1 ohm (fairly typical of a reasonably good quality inductor of around 3mH), in series with a woofer. The damping
factor is already reduced to a maximum of 8 (for an 8 ohm driver) because of the resistance, but if the driver is also low efficiency, you need more
power for the desired SPL. If the average power is (say) 20W, then about 2.5W is dissipated in the woofer's series crossover coil. That's not much, but
it has very little cooling - the crossover network is often underneath damping material, so airflow is almost zero. Even 2.5W will cause the coil to get
rather hot with no effective cooling, so the resistance goes up. While damping factor will probably remain about the same (because the woofer's
impedance has also increased), the crossover network's parameters have changed even more than first expected. The network has a different load
impedance, and has a different series resistance.
The problem gets worse as crossover frequency is reduced (bigger inductor, more resistance) and/or driver efficiency is reduced, needing more
power. This is one of the more compelling reasons to use active crossovers and separate amps. We can't do much about the speaker's impedance
change as it gets hot, but to compound the errors by using passive crossovers is not high fidelity. Again, using high efficiency drivers mitigates many
of the problems for normal listening levels.
The use of passive crossover networks in high power systems is strongly discouraged. There is absolutely no benefit to be gained, but they cause a
great many problems. Making a passive crossover network that has adequately sized inductors (series resistance must be as low as possible) becomes
very expensive - just for the copper wire alone! The money is far better spent on building an active crossover and separate power amplifiers for each
speaker. For optimum performance and freedom from crossover interactions, my philosophy is that any speaker rated at or above 50W programme
material should be active. The benefits far outweigh the disadvantages. It does make auditioning a stand-alone amplifier rather difficult, but I can live
with that.