LM3886 chip power amplifier thermal design. is a very good primer by Tom. Whilst I consider 35v rails pushing luck, it shouldn't get that hot at a few watts though.
EDIT: not sure you are reading figure 34 right. How did you get 3C/w for one channel at 70v?
EDIT: not sure you are reading figure 34 right. How did you get 3C/w for one channel at 70v?
Last edited:
Correct. 70 ºC delta T seems to be the most common spec. This means 90 ºC heat sink temperature at 20 ºC room temp. That's *way* too hot for an external heat sink.
The heat sink manufacturer doesn't care one bit about the ambient temperature. All they care about is the temperature differential as that's the only parameter they can expect to have some say in or that will be meaningful to the customer.
I have no idea how Infineon came up with those, but they're dead wrong. I suggest having a look at a textbook, such as Sedra/Smith, for the derivation of the equations.
You can see the result of the derivations in the Thermal Design section of my Taming the LM3886 Chip Amp pages. The relevant equations about half way down on the page.
So you haven't read my website? Should I take that personally?
If any operating conditions.
Based on my glance at the app note, it looks like they're doing the math flat-out wrong.
The dissipation curves in the LM3886 data sheet (Fig. 35, 36) are calculated from the math I show on my website. The only difference is that I include the idle dissipation as well. The input and VAS stages of the LM3886 are operating in Class A, so no fancy math is needed to include the dissipation for those. It's simply Ibias * Vsupply. I set Ibias = Iquiescent. The only current I can't account for (because I don't know it) is the base current of the output devices. You could make some assumptions about output transistor beta and arrive at a guess - or you could live with the fact that you don't know everything and that the base current is still a relatively small fraction of the output current, thus the driver stage dissipation is significantly smaller than the output stage dissipation. It all depends on whether you take a mathematician's, a physicist's, or an engineer's approach to the thermal design. I take the engineering route.
For sine wave operation into a resistive load, I am able to reproduce the data sheet numbers and arrive at the expected heat sink temperature.
When you do the math, make sure you compensate for the temperature differential between the heat sink and ambient as you're probably not operating at the specified 70 ºC differential. I go through that math in the Thermal Design section of the Taming the LM3886 pages as well.
Finally, take into account the crest factor. The heat sink figures I recommend are for music reproduction and I am usually quite specific about that. I use a crest factor of 14 dB (see justification in the Thermal Design ...). Do note that an amp designed for a a crest factor higher than 3 dB will likely overheat during sine wave operation.
Tom
Last edited:
±35 V rails will cause the LM3886TF to overheat (4 Ω load, sine wave, worst case dissipation) when the heat sink temperature reaches about 60 ºC. ±35 V also requires 35/4 = 8.75 A of output current, which is beyond the 7 A the LM3886 is guaranteed to be able to deliver.
That's why I recommend ±28 to ±30 V for a single LM3886.
Tom
Once again Tom thanks for the clarification.
BTW, Infineon didn't come up with the apps note, it was written donkeys' years before Infineon bought IR (although maybe they ought to check its content).
One question though: why include the idle dissipation?
When the amp is ouputting a signal higher than the class A region the "opposite" transistor is off (zero dissipation due to zero current) and the transistor that is in the linear region is supplying all the load current (but no more).
Or have I missunderstood?
What I really wanted to include was the dissipation of the input and VAS stages. Since I have no idea what those currents are (I can guess based on my experience, but I honestly don't know) I figured worst case would be that all the idle dissipation went in the input stage and VAS. This is clearly a pessimistic estimate.
You are correct.
Tom
Are you referring back to AndrewT's post?
Yes, Figure 34
Go to right hand side and find 70V (= ±35Vdc) go up till you intersect the 8r0 line (they don't give data for 8ohms reactive as in a speaker load).
Go left until you reach the appropriate Ta. In Scotland Spring, Autumn & Winter are usually around 20°C, only Summer sometimes becomes a bit higher, as we generally don't use active cooling to destroy the ozone layer and create extra CO2.
Now look at Figure 36
the lines for all the supply voltages are crowded into that first 10W of output range.
Very difficult to extract what the dissipation is predicted to be when output is 1W and/or 2W into 8ohms and supplies are ±35Vdc.
So I built and measured it, with my finger. I got that surprise !
BTW,
a 60mA (max is specified @ 85mA) quiescent current with zero output power and zero output offset, will give 4.2W when Vsupply=70Vdc (two chipamps = 8.4W). Yet National don't show that on the plots. National seems to show going down to ZERO watts of dissipation !
Last edited:
I don't know how to calculate. That's different from I could look up Tom's webpage and follow through his example calculations and derive a prediction for Dave S to read, or just refer him to your example. Instead he looked up a different source and as you so kindly pointed out that is wrong as well !
I have no intention of trying to learn every new calculation/formulae I bump into in my researches. I'm not a professional any longer, I am retired. I'll leave the real learning for those that need to perform in their daily work.
BTW,
I have re-read most of your website, it was worth reading again. But you don't highlight any recent changes/additions.
Last edited:
That's why I refer to the National chipamps as "current crippled" and repeatedly try to persuade our Members to never use loadings of less than 8ohms (reactive).
I only use 8ohms or higher speakers at home. The only 4ohms speakers I have are in my cars and came as standard equipment.
The quiescent dissipation is significant and becomes the dominant heat loading when output signals are very low.
If one were listening to a pair of 8ohms, 90dB/W @ 1m speakers with an average signal power of ½W, would the quiescent dissipation be more or less than the dissipation due to output current? I don't know, but I could refer to Tom's work to find out.
Or just use my finger to measure and feel if there is a problem that needs attention.
Yeah that bit I think is wrong. As no way is PD only 30W, that is the max dissipation for that heatsink. If you look at figure 35 you'll see under use it may be worse. At the end of the day its a pretty graph and no substitute for running the numbers. Now Tom has already run the numbers, and done tests so the engineering has been done.
What I don't get is that you keep saying the LM is current crippled yet built one with 35v rails. Was that just a transformer you had lying around?
60W into 8r0 test load requires outputs of 31Vpk and 3.9Apk
Both those are available from a 3886 chipamp fed with supplies that are well inside the 84V limit
60W into 4ohms test load requires outputs of 22Vpk and 5.5Apk
Both these are available from a chipamp fed with supplies well inside the 84V limit.
But apply a factor to the current delivery for driving fast changing signals into reactive speakers and you find that the 4ohms rating takes the chipamp beyond what it was specified as. The 8ohms rating allows a much higher factor to drive reactive speaker loads.
Both those are available from a 3886 chipamp fed with supplies that are well inside the 84V limit
60W into 4ohms test load requires outputs of 22Vpk and 5.5Apk
Both these are available from a chipamp fed with supplies well inside the 84V limit.
But apply a factor to the current delivery for driving fast changing signals into reactive speakers and you find that the 4ohms rating takes the chipamp beyond what it was specified as. The 8ohms rating allows a much higher factor to drive reactive speaker loads.
look at Figure 36.
@ output power = 60W go up to intersect the Vs=±35V line. It intersects @ ~28W up the left side scale. That is pretty close to the 30W extracted from the Figure 34. table.
Going back to Figure 34.
The 30W of Pd and the 2.5C/W under the Ta=40°C gives a delta T of 75C degrees. Add that to 40°C and we get 115°C for Ts. They show a Tc=114°C.
I wonder if that should have been Ts and not Tc?
Check the Ta=25°C and use the 3C/W and again we come up with Ts~115°C
delta Tc-s should be around 30W*1C(Rthc-s) =30C degrees giving a Tc of ~145°C
Tj should be around 150°C, which they tell us is the design limit for highest temperature operation.
if you take a 50W into 4ohms rated 3886 operating from ±30Vdc.
Then using Figure35. we can see that Pd is ~45W @ that maximum power and the worst case Pd is ~47W when output power is 40W.
running the 3886 chipamp on ±30Vdc supplies into a reactive 4ohms load is more stressful than running on ±35Vdc supplies into a reactive 8ohms load.
I reckon that for equal stressing of the chipamp that you could run the 8ohms rated 3886 on supply rails ~ 8V higher than for a 4ohms rated version.
I have built a ±38Vdc supplied 3886 for 8ohms duty and it works very reliably. It never gets near the 84V rail limit with my upto 253Vac mains supply (I checked).
But the big downside to using these low supply rail chipamps is that they clip too often on ordinary music/audio, the audio becomes "harsh" compared to a true 100W amplfier that does not trigger protection circuits and does not transgress it's 40 to 42Vpk of maximum (transient) output.
Perhaps crucially my bigger amps are set up to deliver current when demanded. The chipamps cannot be coaxed into delivering more current than the internal circuit can provide.
Last edited:
There's no reason to recommend 8 Ω*only. What the recommendation should be is 7 A, peak, max. This means ±28 V rails max for 4 Ω operation.
Yes, but that's hardly an issue if the heat sink is designed for the maximum power dissipation. Who cares what the heat sink temperature is at idle? What matters is what the heat sink temperature will be under worst case conditions (or realistic music reproduction conditions depending on your design target).
That statement does not make sense. The power dissipated is the product of the voltage across the output transistor and the load current. "Dissipation due to output current" is meaningless.
We have these new things now called thermometers. A thermocouple thermometer is not that expensive. The many IR thermometers are an option too. I paid $50 for mine, used. They're not super accurate (typically within ±2 ºC) but they're a lot better than the finger method (though, I do use that in addition to my IR thermometer).
Tom
Sorry to drag ths out, but I am trying to understand RMS power...
It is clear that my method to calculate the RMS power drawn from the supply is wrong. I simply multiplied the RMS load current by the supply voltage: i.e. Ps = IoutRMS * Vcc
Looking at the equation on Tom's website: Ps = (2/pi) * (Voutpk/Rl) * Vcc
This could be rewritten as (2/pi) * (1.414 *VoutRMS/Rl) * Vcc
then 0.9 * (VoutRMS/Rl) * Vcc
so, Ps = 0.9 * IoutRMS * Vcc
That's 0.9 X what seems logically (at least to me) right. Can anyone explain what's gone wrong with my logic?
Got any measurements? The joy of digital sources is that you can set it up so you never clip unlike vinyl where you have somewhat variable headroom to deal with.
Many of us have limitations that mean we can't run at THX reference levels (esp with loudness wars popular music) but are generally around 75dBA for -20dBFS. With 85dB/W speakers that's 10W for 0dBFS at 2m. Adjust to taste for reactive loads and listening distance and seems well within the capabilities of the 3886. Now if you have a huge room, monster woofers, a linkwitz transform on the bass or 'bang head' a chipamp isn't for you.
In this case for a 'desktop active monitor' listening nearfield pushing the rail limits doesn't on first inspection seem the sensible way to go when dropping a dB in max SPL moves you into a much better space thermally.
Tom was prescient when he said on his website
and why I posted my result for a bigger than National heatsink (almost four times as big for the two chipamps) when the amplifiers were delivering only 1W and 2W into 8r0 resistive loads.
Now I could think again about a "realistic" music average signal, maybe 1W + 2W was too high.
The result was a surprise and judging by some of the replies, it's a surprise to other Members as well.
That makes the information worth posting.
It prompts discussion on worst case and reasonable testing procedures.
Last edited:
I don't have measurements.
I know that turning up the volume of a chipamp produces a harsh sound that does not happen with a discrete amp with much higher current capability, even though average levels are still very low.
My understanding is the interaction between driver and amp creates most of the harshness sound in solid state amplifiers, the interface between power supply and the amp is also critical. However, both relate with ground return design which could cause amplified effects. Tube amplifiers having output transformer isolates the interaction.