That I like. I actually did a frequency sweep with a 2-way sealed box speaker connected. I didn't see any oddities that would indicate trouble. A speaker model would be nice, though. It certainly would be quieter... 🙂
BTW: Is anyone aware of a similar load for headphones? I know they exist, but haven't seen a schematic.
My amplifier torture experiments involve capacitive loads. I ran the MOD86 with 8 Ω in parallel with 1 nF .... 220 nF and looked at the transient response. As you'd expect, the resonance between the Thiele network inductor and the load capacitance causes some ringing. The amp behaves rather well even with a significant capacitive load.
~Tom
Peufeu, I assume you meant to write: "WIMA plastic CAPACITORS which have some INDUCTIVE properties at low frequencies"
Musings on Headphone Amplifier Output Impedance | InnerFidelity is all I have seen and I am not entirely convinced by the basic premise...
You went a lot further than I did. My reasoning was along the lines of
1. remote sense is usually proposed for power amplifiers by loons like DN Morecroft
2. Remote sense works really well for things that don't change quickly like ooh precision resistance measurements. For a complex AC signal it looks to be a recipe to really make your feedback hard to do.
oh and damn you. Just realised I can mount the amps on the back of the baffle...That's the rest of the evening gone.
Figured out a nice way to go from the Mod terminal block to the drivers with an uninterrupted cable run yet? 😀
There's quite a lot of activity around remote sense and virtual remote sense for microprocessor and USB power supplies with bandwidths well into the MHz. What makes power amplifier remote sense tricky isn't waveform complexity (processor loads tend to be a fair bit more aggro) but the required precision. 1% error is acceptable in supplies but no good for audio delivery.
An assembly question. the LM3886 is always shown perpendicular to the board. Having never held one in my hand. is there any way the leads can be bent for parallel mounting (like say a plate amp)?
yes, will require me to resolder the ribbon feeds, but that is easy. My planned heatsinking wont work tho. mind you the stands are hollow (well currently full of sand, so PSU can go in there. As I am being sneaky on the bass I can ramp down the power requirements. Some iteration required, but will at least make metalwork cheaper.
At the time (freshman year) was amazed how easily a simple ground mistake could lose you a LSB of performance. Mind you back then it was all pencil and head scratching as SPICE was, erm slow.
No rush. I'll be pushed to get the mods built and the apogees converted to open baffle before sprog the latest arrives and puts all projects on hold for a decade, so as long as you have it by 2020 that will be in time for me 🙂
I'm pretty sure National used to offer that type of horizontal package. I also seem to recall that it was nearly impossible to find at the distributors, so maybe it didn't move in enough volume for TI to care about it. It may have been "cost-reduced" - or I could be flat-out wrong in my recollection... 🙂
It should be possible to bend the leads on the LM3886, but you'll put an undue amount of stress on them in doing so. Personally, I'd just take a hunk of, say, 0.75x1.5" (15x30 mm) aluminum and make a heat spreader.
Another option would be to mount the MOD86 on a bracket and the bracket to a heat sink. I've built amplifier modules like that before. That tends to work rather well.
~Tom
Found it !
Analog Devices : Rarely Asked Questions (RAQs) : Where Good Signals Go When They Die
The red cubes have ESL in the 4-6 nH range (absolute minimum when mounted on test fixture with planes, it'll be more if traces are used) so they're perfectly fine for filtering in the audio band, like RIAA, EQ, crossover etc, or as signal coupling caps ; however they're completely useless as decoupling caps since above 1 MHz they're mostly inductors... can't use those on anything digital... besides, an electrolytic of same pin spacing has about the same inductance and much more capacitance, so might as well use that. And of course not recommended for any kind of EMI blocking, wether at the input of your amp or as a DAC output filter, since you want a shunt cap with low impedance...
If we are talking about how the impedance curve of a speaker interacts with the high output impedance of a tube amp, that is one thing. I agree that causes complex changes in the frequency response that would be specific to the particular speaker and might even need transmission lines to model accurately.
But this is not a thread about amps like that. We are talking about instrumentation grade solid-state stuff with practically zero output impedance. The amp's frequency response is unaffected by the speaker impedance, we just need to put some sort of worst-case reactive load on it to check that it can handle all of the peaks and dips of the speaker impedance without oscillating, going into protection or blowing up.
Douglas Self wrote several articles where he analysed output stage dissipation as a function of various reactive loads, and I chose my test load on the basis of those.
I think the spreader may the way to go. That way minimises fabrication. I have 200x150mm footprint to play with, but do have the ability to go down into the stands for things like the transformer. it's that or a die cast box bolted to the heatsink with holes machined for the 3886s.
I think, yes.
place the front row of leads on the top side of a double sided PCB and place the back row of leads on the bottom side of the PCB.
This arrangement would be very easy to dismantle if the chip needed off board checking.
There are many on this Forum that recommend plastic film capacitors for local decoupling.
You are showing how wrong they are !
One must use tiny pin pitch to achieve low inductance and actual capacitance into RF range for effective HF decoupling.
402 smd are about the smallest we see.
There are some that rotate the package shape for resistors.
Does any manufacture do this for smd capacitors. A wide short package should have a lower inductance.
Yes. A brief search on DigiKey will find them; scan down the "Package / Case" column looking for upside-down aspect ratios.
But they are functionally obsolete. Now SMD capacitors are available in ball grid array packages with even lower inductance, since they have many parallel leads (dividing inductance by #leads) AND since they interleave the terminals to get VERY tightly coupled mutual inductances. Remember Ldi/dt minus Mdi/dt.
BGA caps are used in and on the packages of 3GHz intel & AMD central processing unit chips, and on the motherboards too. Multilayer PCBs and packages are the norm in these applications.
Last edited:
Depends on what you consider to be "RF" and what you consider to be "HF".
In case of an LM3886, the supply impedance "seen" by the IC is actually limited by the bond wire and lead frame of the package. Even with multiple power pins and probably multiple bond wires per power pin, the lead inductance will dominate the supply impedance above 10 MHz. You can geek out all you want about 0402s soldered to the chip pins if that suits your fancy, but you won't impact the IC performance by doing so. All decoupling of frequencies above a few MHz is handled by the on-die decoupling capacitors.
If you could find a, say, 1 uF X7R dielectric 0402 cap that can handle the supply voltage of the LM3886, you could potentially drive the supply impedance down a few mΩ in the 1-10 MHz frequency range. That's over two orders of magnitude about the audible frequency range. It won't impact stability as this is already taken care of by proper on-board and on-die decoupling.
Personally, I think you'll get much better return on your investment of time by optimizing the PCB layout. That will impact the circuit performance.
If by "RF" and "HF" you mean frequencies above, say, 500 MHz, then you're right. You do need to use small SMD capacitors. However, that isn't relevant for the LM3886 for reasons outlined above.
You can check out my supply impedance simulations on my Taming the LM3886 - Supply Decoupling website.
One of the guys I work with can hand solder 01005 SMD components. I think he has to plan his caffeine intake around those soldering sessions, though... 🙂
~Tom
Last edited:
A couple of thoughts on recent posts: when I did my LM38xx, I analysed what the supply would do quite a bit, and the result was that I didn't use plastic caps for decoupling, just enough good quality electro's on power planes, the films would have added nothing; also, I didn't hesitate to heavily mutilate the legs of the chip amp, I reduced and bent them to the bare minimum to get the job done of connecting them into the circuit - there was no point to optimising elsewhere, to throw some of the benefit away by using the legs as supplied.
Yes, exactly. I was talking more about decoupling digital chips or the input anti-RF filter.
But TQFP packages and the like do the same thing as your LM3886. The bondwires add inductance, so if the chip has only 1 VCC pin, its inductance will dominate anyway. Of course a TQFP will have less lead inductance than a LM3886, but it may still be more than a single MLCC across the pins. So, stay Zen, no need to go on a capacitor binge. 1 or 2 MLCC and a good electrolytic is fine.
The job is to avoid adding a huge resonance peak, which is guaranteed to happen with thru-hole plastic caps 😀
It if has many more VCC/GND pins, then maybe. The ultra low ESL caps like X2Y, IDC and BGA types are really meant to decouple power/gnd planes for use with big fat BGAs with a huge number of supply pin pairs... they're not very interesting for decoupling audio stuff, unless you also want to exploit the fact that the low ESL makes a better lowpass filter and prevents noise current coming out of the chip from infiltrating the main PSU, but in this case, a proper pi filter is even better...
That's widely taken to be sufficient for the gain and phase margin kinds of questions which seem to be your primary interest. I'm just pointing out models typically used for that type of analysis aren't capable of representing aspects of speakers' impulse response which seem to be significant to the perceived sound quality of the amp+speaker system.
Well, proper component selection can mitigate it effectively. But, yeah, easier to choose not to have the problem in the first place.
It is a little annoying National chose to pin their TO-220-11 chipamp lineup with two or three supply pins and three no connects rather than making more full use of the package to get the inductance down---three V+ V- pairs is doable, though probably not optimal considering part of the loop is through the package's output pin and the amp's zobel. In practice it's not that much of an issue for standalone 2876, 3875, 3876, and 3886 performance. And the control device in composite amps like the Mod has enough loop gain to flatten it out pretty well. The Mod does have a number of optimizations in this area to minimize the error residual.