LMH6643 amplifier

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hello all,

I'm beginning to build of my new headphone amplifier; it should be portable.

First of all I've seen the cMoy, but I don't like it too much because there's no output buffers; and my Sony MDR-V55 headphones are quite hard to drive.

So, I searched for a high current op-amp, and at the end I decided I wil ltry the LMH6643 one (even because I already have them available...).

A search for an already existing headphone amplifier utilizing such an op-amp got me discover the mini3,togheter with a lot of clones.

I really liked the simplicity, and the fact it uses the LMH6642, so I decided to give it a chance.

My idea was, however, to use a single lithium cell, with a charger/protection circuit already integrated.
This is because a, let's say, 1000/1500 mah mobile phone lithium battery should give me a long playtime.

Following are the schematics I would implement:

1. Without voltage splitter:
An externally hosted image should be here but it was not working when we last tested it.


2. With voltage splitter:
An externally hosted image should be here but it was not working when we last tested it.


3. The voltage splitter section:
An externally hosted image should be here but it was not working when we last tested it.


The difference with the mini3 is that I would prefer to use a double op-amp for each channel, even for the virtual ground.
With this solution I would loose the analog ground the mini3 has, but I don't think his would be a real problem, if the PCB will be routed correctly.

So, my questions:
1. Do you see anything wrong?
Any problem you can think of?

2. Would the circuit work fine with a single lithium cell as the only power source?
It will run at a voltage of about 3.6 volts, any problem you can see?

I would like to use only through-hole components, and to share the PCB layout with the community, so come on, help me!

Thank you very much indeed.
Ciao,
Giovanni
 
Well congratulations on your proposed design, you have some good stuff in there! :)

They way you have the output buffer arranged is jcx's favorite way of doing it, from what I can remember of various posts here and on Headwize. Probably a good way to go.

Both chips are rated for a minimum of 2.7Vdc so you would be OK with the minimum Lithium Ion protection circuit cutoff of 3.2Vdc or so.

Those are definitely the chips AMB specifies for the low power version of the mini^3. I've spent some time playing around with that circuit and have a bunch of both chips on hand. The output chip does have current limiting, which is important when using the 3.5mm TRS connectors that short on the way in and out.

One possible issue that pops to mind though is available output swing. Assume 3.2Vdc before the protection circuit cuts off in the battery. With a virtual ground that gives you a peak voltage of 3.2Vdc / 2 = 1.6Vdc. But there will be some voltage overhead in the chip's output section, worse with increasing load of course.

The data sheet lists how close you can get to the rails, but for 150R. Taking a wild guess here of 1Vdc for a 40R load given the paralleled chips, and it could well be (a lot) more, that would give 0.6Vdc peak or .42Vdc (rms) = 420mV (rms) for the maximum output voltage swing. With the 105dB/mW sensitivity of your phones that might be enough. Would take some more detailed math. My 114dB/mW AKG K550s only need about 50mV (rms) for full volume.

May also be good to add a small capacitor, like a 180pF MLCC, in parallel with the 47K feedback resistor to roll the stage gain off to unity around 100kHz or so. These are wideband video chips (take a look at the gain vs. freq. graph of fig 1 in the datasheet) and really need to have their wings clipped a bit to prevent high frequency ocillations. For audio work an amplifier has no business amplifying anything above 100kHz, IMHO.

I also like that your design does not have large electrolytics from the VG output to each rail. My current thinking is that essentially negates the whole idea of "active cancellation" by the VG, since these tiny chips can't push that much capacitance. Large caps effectively "pin" the VG. Instead try 0.1uF MLCC caps from VG to the rails to just bypass any high frequency (above audio) noise. I've found that effective in a test amp I made last year.

I see that TI has PSPICE models available for both chips. I may toss them into LT Spice for fun as time permits this weekend and see what the SPICE model has to say about maximum voltage swing with a 40R load.
 
Last edited:
Hello,

Thank you for your answer.

Actually I calculatedthe output on my needs, my headphones have an impedance of 32 Ohms, and a declared sensitivity of 105dB/mW, so I wouldn't need lot of voltage, but quite a good current.
So, for my needs, a singled battery should be enough.

Or I could use a 8.4V ni-mh battery, loosing however the possibility to charge the thing from a USB port.

And I could use a step-up converter to elevate the USB 5V to the required charging voltage for the 8.4V battery.

Actually I like the last solution; I could build a current-controlled step-up, and charge the battery with it.
After the battery is fully charged, I simply switch the charger off.
That way I would have about 7V before total discharge, meaning, let's say, 5Vpp, or about 1,78V rms (correct me if I'm wrong).
Plenty of power, also considering that, however, output section will be limited at 150 mA (75+75 mA).

Ciao
 
Sony claims 40 Ohm, 105 dB/mW

120 dB SPL before clipping is extreme headroom; requires +/-1.6 Vpk, +/-40 mApk with those specs

40 mA is more than many standard op amps are happy with - selecting one of the higher output op amps is one option, paralleling 2 lessor current rated op amps is another, buffering is a third option

LMH6643 is rated 75 mA – I would tempted to just use one per channel, not paralleled

the LMH6643 input range only goes to the negative rail – it is speced for 1 V below the positive supply rail
so you can't get near the full positive output swing in unity gain – this chip requires some V gain to let the output swing near the rails while keeping the input happy

for active supply splitting 2 paralleled in unity gain may be OK
but the mini3 project had some problems with fast op amps in the supply splitter (and don't even think about using "3-channel" gnd - all the reasons given for it are wrong)

2 other problems with the LMH6643 – 130 MHz is dangerously fast, Vnoise for audio is poor


130 MHz is dangerously fast – causes stability problems, needs very careful layout, bypassing, attention to impedance at several points in the circuit

the datasheet frequency response peaking at Av +2 is a clue that input C reacting with the feedback R is a problem – requires low source Z at the inputs to keep the added phase shift low

not shown but is a near certainty that at these speeds wiring, cable complex Z, Cload is an issue – you need to add some load isolating series Z – lossy ferrite is a good idea if you're trying to use all of the battery V



Vnoise for audio is poor – worse than a TL070 – not many choices for 3.3-3.6 V total supply but there are some – and a slower input op amp works well for global loop compensation with a really fast output buffer (which itself has to be operated with at least Av =2 as noted above)
LMV series, LMV751 has low noise, “slow “ at 4.5 MHz GBW – but makes a multiloop with LMH6643 fairly safe loop gain stability wise



my recommendation is to use the LMH6643 for output buffer with local gain of +2, ~500 Ohm feedback R, inside a LMV751 global loop, also gain of +2
 
Last edited:
[cont - missed this before edit timeout]

my recommendation is to use the LMH6643 with local gain of +2, ~500 Ohm feedback R, inside a LMV751 global loop, also gain of +2

use series lossy ferrite between the amp and output connector

the supply splitter probably wants local gain to overcompensate it too, + input needs low Z to both rails – at a guess ~ 100 Ohm, 100pf series “Zobels” forming a AC divider to the rails

power supply bypassing, layout are critical with 100+ MHz parts on the board

a 200+ MHz 'scope, 10+ MHz square wave source to see ringing, guess at stability
 
Wow,

What an answer!

Thank you very much, jcx.
I really underestimated the project.

I can see a can of worms, here, using a so fast op-amp.

What about my other idea, a not-so-fast op-amp (I have on hand LT1677), driving a discrete diamond buffer?

I simulated it last weekend, and for the moment abandoned the idea because, as you can imagine, the discrete output buffer cannot go near supply rails in any way, so output swing seemed to me quite low.
But I could increase the voltage (9V battery) so the diamond buffer returns.

i could use a third op-amp and a third diamond buffer for the virtual ground, too.

I would take the simulations done and play with them a little bit, tomorrow.

BTW, what does it means
(and don't even think about using "3-channel" gnd - all the reasons given for it are wrong)
?

What is a 3-channel grid? Please explain.

Ciao,
Giovanni
 
What is a 3-channel grid? Please explain.

I was hoping that jcx would jump in! :D

He means a certain configuration of AMB's beta22 headphone amp here: The β22 Stereo Amplifier . It can be wired up with one amp board for each channel with a "real" dual-rail power supply ground, or instead with a 3rd amp board acting as a virtual ground. That second config is called the 3-channel version.

Although your chips could run at 3.2Vdc minimum, they would likely be happier in terms of various distortion reduction and noise reduction when running closer to the mid or high(er) end of their supply voltage rating. Notice that the chip specifications in data sheets are never given just a volt or so away from either rail where the performance will start to go a bit south. Your 8.4Vdc option with the NiMH cell would probably measure better on a dScope or Audio Precision, although the difference vs. the 3.2Vdc version might not actually be audible.

The dc-dc converter and 8.4V NiMH cell are relatively big, physically, so the 3.7Vdc (nominal, the li-Ion cell) would be a lot smaller. And would be kind of cool to charge off USB. I would probably go with the 3.7Vdc version just for small/pocketability sake, but that is just me. The 8.4Vdc option should work just fine, especially like you plan on only using the (EMI noisy) DC-DC converter for charging overnight then actually run off the battery.

Yeah I would have to agree 100% with jcx with the trouble a fast op amp used for audio can bring. The problem with a 130mHz capable op amp is...well...it can do exactly what it is designed for and amplify up to 130mHz signals! There is just nothing good that can come from that in an audio circuit. Likely result is high frequency oscillation. Audio circuits just don't need that kind of bandwidth or that kind of slew rate found in the video chips. Video also doesn't usually need the kind of low distortion figures in chips that audio does. Like he says, very careful layout is needed. Then I would roll the gain off to unity at more audio-sane frequencies like 100Khz, 250kHz, etc.
 
Last edited:
So here are some LT Spice plots with the LMH6643 model from TI. :)

The first circuit and plot below is a single output stage in the buffer configuration. I ran the input voltage up to clipping to find out where the model predicted the chip would clip with the 40R load. With the gain resistors shown, that is around 0.28Vpeak input (0.2Vrms) to go up to about 1.4Vdc peak on the output. Not bad with effectively +/-1.6Vdc rails! Real life with the actual circuit probably would not be as good, but should be in the ballpark like 1.2Vpeak or so.

Note in this sim I'm using +/-1.6Vdc to equal the 3.2Vdc (minimum, when the protection circuit in the batt cuts off) single lithium cell with a virtual ground. I'm not showing the VG circuit yet to keep things simple at this point - and to help weed out effects later when the VG circuit is added.

Blue is output and green input, of course. The most interesting thing about this plot is the current through the 40R resistor (not shown for clarity), which is 30mA peak. 30mA is within the ability of just one half of the 75mA-per-half LM6643 chip, so in the second circuit and plot below I've left the buffer circuit out and just fed the 40R with a single section of the LMH6643. Almost no difference, as expected! Clipping went from 0.28 input with the buffer to 0.26 and clipped at 1.35Vpeak either way.

So the first take-away here is that for (just) a 40R load, no need to use the buffer configuration with the 75mA-per-section LMH6643 for the output circuit. A single section of the chip per channel does the job and stays well within power dissipation specs too at such a low voltage.

The second thing is that in this configuration, it has reduced to AMB's mini^3, for the output stages at least! Here The Mini³ Portable Stereo Headphone Amplifier . You could just get one of those PCBs and save a lot of work. AMB lists the LMH6643 as the alternate low power output chip for the mini^3, of course. AMB uses 6.2R output resistors which were apparently picked to stop oscillations with the "high performance" OPA6901 chip. I don't recall seeing anything in the original Headwize posts about their need (or not) for the LMH6643 stability. I tried the sim with 6.2R series resistors instead of 0.5R and it made very little difference. Slight reduction in output voltage, although damping factor would be increased given the low 40R load. You are better off with the lower 0.5R if you can get away with it (no oscillations).

The next plot and circuit is now two of the single section LMH6643 output stages with the virtual ground circuit added. Here I used croccodillo's VG circuit with the buffer rather than AMB's on the mini^3, given the current requirements involved. Since each output section produces a peak load current of 30mA, that means the VG would have to source/sink double that, 60mA, during times when the two channel's peak waveforms happened to overlap in time. To make things more interesting I picked source frequencies for the two channels that are just slightly different, 8kHz and 6kHz, so they will "beat" against each other in time (current phase relationships line up every so often), causing that 60mA VG load to happen periodically.

Since a single section of the LMH6643 is 75mA, that 60mA would be too close to the maximum given thermal de-rating and the heavy load on the chip. Hence the reason I used the buffer configuration here which would roughly double the VG current capability to 120mA, keeping the 60mA well within specs. In practical DIY terms this means you can use one of the dual section LMH6643 for the left and right channel outputs, then a second LMH6643 for the dual-section buffered VG circuit. You could use one of AMB's mini^3 boards just as it sits for the output LMH6643 circuit, then just make a new board for the buffered VG and use that in place of the mini^3 single chip VG for your specific application.

The third plot shows the 8kHz left channel input (green) and output (blue) on the left voltage scale, with the purple load current on the right scale. The blue output and load current overlap for much of the plot. The lower plot is the same thing with the 6kHz right channel, with blue (input), red (output), and grey load current on the left scale.

The final plot is really interesting. I've replaced the top left channel plot with a plot (green) of the current through the virtual ground (current though ferrite bead L3 in the schematic). I've left the bottom right channel plots just for a time reference. Note from the green VG current plot and left scale that sure enough, the peak current is 60mA, just as expected. The odd looking VG current waveform is what we expected with that 8kHz left channel voltage & load current wave "beating" in time (phase relatonship) with the 6kHz right channel wave. During the points in time that the left channel load current AND the right channel load current line up, it produces that 60mA VG current peak.

One more tip that I ran into last year when messing with a similar circuit. If you use diodes after the battery to make sure wrong polarity won't cook the circuit, like say a 1N5819 shottky, use then on both the positive and negative leads, not just one on the positive. This one was a real head-scratcher. :) Reduced distortion on the FFT. Finally came to realize that ground is no longer the battery negative terminal, it is the VG. So the battery leads are no longer "special" with respect to the ground reference, meaning essentially they need to look balanced looking back into the battery from the VG. The above-audio noise is now being bypassed back to both rails equally, and the impedance on each rail back to the battery needs to be the same. Or something like that.:)

Have fun with the DIY! If you build it please post the results and some pictures.

I'm gone for the next month or so on vacation and day job stuff and probably won't get much chance to log in again to follow up on anything. :) Happy holidays!
 

Attachments

  • LMH6643 ckt 1.png
    LMH6643 ckt 1.png
    36.2 KB · Views: 352
  • LMH6643 VG current plot.png
    LMH6643 VG current plot.png
    35.3 KB · Views: 100
  • LMH6643 with VG 40R load 8khz 6khz plot.png
    LMH6643 with VG 40R load 8khz 6khz plot.png
    42.2 KB · Views: 130
  • LMH6643 with VG ckt.png
    LMH6643 with VG ckt.png
    38.4 KB · Views: 211
  • LMH6643 8khz 0.,28vpk input 40R load plot 2.png
    LMH6643 8khz 0.,28vpk input 40R load plot 2.png
    22.9 KB · Views: 215
  • LMH6643 ckt 2.png
    LMH6643 ckt 2.png
    33.2 KB · Views: 251
  • LMH6643 8khz 0.,28vpk input 40R load plot 1.png
    LMH6643 8khz 0.,28vpk input 40R load plot 1.png
    23.4 KB · Views: 321
Last edited:
I don't want to discourage anyone - its not rocket science - the mini3 does work with chips as as fast

it just helps a lot to know what to look for, have the tools that can see the problems

you probably need a gain > ~ +3 to avoid the limited input Vcm range problem with split 3.3 V supply


I don't know if a few cuts and jumpers would fix the "3-channel" silliness of the mini3 pcb
 
I don't know if a few cuts and jumpers would fix the "3-channel" silliness of the mini3 pcb

Possibly! Photos attached :D

Last year after the O2 came out I got a burr under my saddle to see if I could improve the mini^3 virtual ground by replacing the OPA6901D chip with a 200mA OPA551UA (surface mount). Turns out with only a couple of cuts and jumper the OPA551 just fits right in the existing mini^3 PCB, as that photo shows.

In addition I converted the two point virtual ground (signal and output grounds separate in the mini^3) to a single point virtual ground, which turned out to be fairly easy too. In the photo I replaced the two electros pinning the VG with two 1uF mlcc caps. And on the back added that diode I wrote about above, on the negative battery lead, to make it a balanced circuit with respect to the new ground reference point (the VG).

Another benefit that came for free was that although the AD8397 output chip doesn't have current limiting, hence another reason AMB added those 6.2R output resistors to keep the TRS shorting from frying the thing (the LMH6643 does have current limitng, BTW, so 0.5R is OK), the OPA551 does have current limiting. And since the VG puts it in series with the AD8397, in theory at least the whole thing is now current limited against TRS shorts! I haven't tried to test that theory (yet) with multiple output shorts, but the OPA551 limits at around 220mA and the AD8397 is good for 310mA peak, so the potential is there at least.

Since the circuit was now current limited against output shorts, I also modified that PCB to have 1R output resistors instead of the standard 6.2R, to improve damping, and re-wired the feedback loop on both channels with a cut and jumper to take off before the 1R rather than after, so that the output resistor is no longer in the feedback loop.

All of this greatly cleaned up the simulation FFT (modified vs. original circuit), which was interesting. But when it came to real-world listening, of the modified version vs. a mini^3 I have here built the standard way, zero difference to my ears. Both sounded great. So although the mods probably reduced crosstalk and distortion, whatever amount it was is apparently below my threshold of hearing. And my philosophy is always along the lines of "if I can't hear it, then it doesn't matter". :)

AMB knows about my experiments.

I'm off for the holidays. Have fun! :hohoho:
 

Attachments

  • IMG_1200.JPG
    IMG_1200.JPG
    148.4 KB · Views: 194
  • IMG_1202.JPG
    IMG_1202.JPG
    141.8 KB · Views: 216
  • IMG_1194.JPG
    IMG_1194.JPG
    196.6 KB · Views: 208
Last edited:
Hello all, and thank you for your answer.

I was out for the weekend.
Last days I focused on the amplifier supply stage, and decided few tings:

1. I really need to charge it through USB: I have USB charger all around my house, my car, my office; even more, I have PCs all around me that can be used as battery chargers...
2. I want to use Li-ion batteries, they are the future, last forever (if correctly used), and you can find really cheap batteries from mobile phone world.
3. My instinct says me to do not use a VG, so I decided to go with REAL ground.

So, these are the solutions I have thinked about:

1. Real dual batteries, with dual battery charger.
You need two isolated 5V supply, two charger ICs, and a circuit to cuut-off everything if one of the two batteries disconnects itself from the circuit (I assume I would use self protected batteries).
This solution is quite complicated.
I know I can charge the batteries in series with a single charger, as for AMB X1 battery stage, but in that case you cannot balance your battery cells, which is really dangerous and will shorten their life.
For two batteries, you really need two chargers.
An externally hosted image should be here but it was not working when we last tested it.



2. Single battery, voltage doubler.
Quite simple, you only need a beefer battery (not a problem, nowadays).
An externally hosted image should be here but it was not working when we last tested it.


I would go with the second circuit, cheaper, simpler, smaller.

Third possibility is to develop a transformer coupled, dual output DC/DC converter from scratch.
Not too hard to make it wok, really hard to make it work fine with an audio amplifier.

What do you think?
 
Last edited:
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.