Hi everyone, I'm building my first ever headphone amp very closely based on Chu Moy's pocket headphone amplifier but using the lme49723 opamp instead. When I test my circuit it was horribly distorted. Can hardly tell it's music! I then experimented with changing the resistor between input and ground, with a little change, but nowhere near right.
At the risk of sounding really stupid, I assume the input ground and output ground are to be tied to the virtual ground of the circuit, but I noticed that immediately upon removing either the input ground or the output ground I get nice clear sound for maybe a second. Then back to very distorted.
What am I doing wrong? Hopefully just a dumb mistake! A quick drawing of my particular schematic is attached. Thanks!
At the risk of sounding really stupid, I assume the input ground and output ground are to be tied to the virtual ground of the circuit, but I noticed that immediately upon removing either the input ground or the output ground I get nice clear sound for maybe a second. Then back to very distorted.
What am I doing wrong? Hopefully just a dumb mistake! A quick drawing of my particular schematic is attached. Thanks!
Attachments
Chu Moy has an error of sorts in the original article. The 10K in the feedback loop plus the 1K to ground on the inverting op amp input give you a gain of (1 + 10K/1K) = 11 which is just way way to much for anything. Try reducing the 10K to 1.5K for a gain of 2.5x.
The gain formula is here:
Operational amplifier applications - Wikipedia, the free encyclopedia
Desktop sources often only need a gain of 1x (just a current buffer, leave that resistor from inverting input to ground out altogether). Pocket sources like ipods may need the 2.5x to 4x or more. But not 11x! That only was needed in the movies, lol: Up to eleven - Wikipedia, the free encyclopedia
You have another problem too. Your power rails are +/-2.5Vdc and the absolute minimum for the LME49713 chip is +/-5Vdc. You need more voltage...
Another solution would be changing chips. The NJM4556AM is good down to +/-2Vdc power supply rails and the pinout of the surface mount version is exactly the same as your LME49723. It would drop right in. 73 cents at Mouser: 513-NJM4556AM.
Last edited:
Ah, thanks for the help! It was indeed a rookie mistake. I guess I read the datasheet wrong... I was planning on running this on USB power. Looks like I will be switching to that NJM chip you mentioned. How does it sound? In the small amount of browsing I've been doing, I don't remember seeing it mentioned.
The NJM4556A is the chip used in the O2 headphone amplifier in this forum here. There are probably several thousand O2s out there by now. It is also the chip used in Grado's RA1 headphone amplifier.
Please realize though that with only +/-2.5Vdc power rails on your CMOY there will be some limitations on the type of headphone that can be powered. You don't have a lot of voltage swing to work with there. The headphones will need to have a fairly high sensitivity rating, like 115dB/mW vs. something like 92.5dB/mW.
As for USB, you might get a kick out of reading through 00940's creation here to see where that can go on the more technical end of things.
A very good design. He is using a DC-DC converter module to take the 5V 500mA USB to +/-9V, filterting out the converter noise, then using a LME49713 gain op-amp wrapped around a current buffer op-amp for each channel.
Last edited:
The LME49723 is actually rated for 5..34 V of supply, so it should still work. Input common-mode range, however, is limited to within 2.0 V from both rails typ, so some gain would be needed to make the most out of output voltage swing, which is not quite as limited (about 1.0 V from each rail at limited loading). About a factor of 3 or so.
Very low-voltage circuits tend to employ inverting operation in order to avoid running into input common-mode range problems (as the inputs remain at a fixed potential then - and as an added benefit in single supply operation, the virtual ground is not being loaded).
I suspect that the OP either overdrove his circuit badly or made some mistake wiring up virtual and "real" grounds. This might already show on a complete schematic.
Ah, I read the first post again: I think the problem is that the virtual ground is tied to "real" ground, with the "+/-2.5 V" supply in fact being +5V/Gnd. A virtual ground does not simply turn a single supply into a split supply.
If so, things should be connected like this:
Input ground at the 9k4 goes to "real" ground.
The 100k goes to virtual ground.
The feedback network either connects to virtual ground, or (preferably) is capacitively coupled to "real" ground.
The output needs to be capacitively coupled and connected to "real" ground. (Virtual grounds usually cannot sustain load currents, or at least those typically used in a cMoy. Some circuits employ a dedicated "ground buffer" though. It means higher power consumption and some other potential pitfalls, but you can do away with the bulky and potentially rather non-ideal output coupling caps. Try squeezing coupling caps of 220 µF up into a tiny MP3 player. A generally even better approach involves generating your own negative supply with a charge pump on-chip.)
Very low-voltage circuits tend to employ inverting operation in order to avoid running into input common-mode range problems (as the inputs remain at a fixed potential then - and as an added benefit in single supply operation, the virtual ground is not being loaded).
I suspect that the OP either overdrove his circuit badly or made some mistake wiring up virtual and "real" grounds. This might already show on a complete schematic.
Ah, I read the first post again: I think the problem is that the virtual ground is tied to "real" ground, with the "+/-2.5 V" supply in fact being +5V/Gnd. A virtual ground does not simply turn a single supply into a split supply.
If so, things should be connected like this:
Input ground at the 9k4 goes to "real" ground.
The 100k goes to virtual ground.
The feedback network either connects to virtual ground, or (preferably) is capacitively coupled to "real" ground.
The output needs to be capacitively coupled and connected to "real" ground. (Virtual grounds usually cannot sustain load currents, or at least those typically used in a cMoy. Some circuits employ a dedicated "ground buffer" though. It means higher power consumption and some other potential pitfalls, but you can do away with the bulky and potentially rather non-ideal output coupling caps. Try squeezing coupling caps of 220 µF up into a tiny MP3 player. A generally even better approach involves generating your own negative supply with a charge pump on-chip.)
I was curious how this would simulate.
The first circuit and plot is with a 300R load while the second is with a 32R load. Green is output voltage and blue is input voltage on the left scale. Red is op amp output current on the right scale.I ran both of these up until just before clipping. For the 300R load with the 1V peak input that was around 1.5Vpeak. 1.6 gain = 16K/10K. For the 32R load with the 1V peak input the max output was around 0.75V, an attenuation, but still useful current buffering going on. Both transient plots are done at 20Hz to show low frequency performance given all the caps in the signal path.
I've added the cap in series with the pot on the inverting attenuator to block DC as per the good comments from 00940 in this this thread. As he mentions the 270K is in there just so the feedback loop doesn't go completely open loop when the pot's wiper eventually fails.
Attachments
Last edited:
Wow lots of stuff for me to read through already! Thanks guys.
agdr, thanks for the info on 00940's USB amp. I might have to try something like that. The headphones I'm planning on pairing this with are 32 ohm and 104 db/mW. That falls right in the middle of your given range but I think I may try the dc-dc convertor and buffer anyway (however, one step at a time). If I remember correctly, a buffer is recommended for low impedance headphones, right?
I may need you to explain your most recent post though... Are you showing that even with my low input voltage I still might have useful voltage swing at the output? Pardon my ignorance, I'm a mechanical engineering student, not electrical.
sgrossklass, I caught one mistake I made from your suggestions. The 9.4k resistor was indeed tied to virtual ground. That's fixed now. Capacitive coupling is something I also don't have. Can you explain the purpose of the coupling capacitor? If I remember, they will let AC current through to ground, but not DC current. I'm curious, this seems to me like the opposite of what we want to drive speakers...
Regardless, looks like I'll be making yet another digikey order!
agdr, thanks for the info on 00940's USB amp. I might have to try something like that. The headphones I'm planning on pairing this with are 32 ohm and 104 db/mW. That falls right in the middle of your given range but I think I may try the dc-dc convertor and buffer anyway (however, one step at a time). If I remember correctly, a buffer is recommended for low impedance headphones, right?
I may need you to explain your most recent post though... Are you showing that even with my low input voltage I still might have useful voltage swing at the output? Pardon my ignorance, I'm a mechanical engineering student, not electrical.
sgrossklass, I caught one mistake I made from your suggestions. The 9.4k resistor was indeed tied to virtual ground. That's fixed now. Capacitive coupling is something I also don't have. Can you explain the purpose of the coupling capacitor? If I remember, they will let AC current through to ground, but not DC current. I'm curious, this seems to me like the opposite of what we want to drive speakers...
Regardless, looks like I'll be making yet another digikey order!
Amazingly enough your headphones are sensitive enough that you should be able to hit very loud musical peaks just with that inverting CMOY circuit and 5Vdc USB power, unless I've missed something in the math.
Take a look at the spreadsheet below (PDF file). That is a headphone power calculator that was posted on Head-Fi last year. Enter the headphone impedance, sensitivity, and loudness in dB (Sound Pressure Level) that you are shooting for and it calculates the voltage swing (rms) and current requirements needed. Ignore the last spreadsheet section about how load the amp will drive your headphones - those are numbers left over from a different amp.
The inverting CMOY simulated at around 0.75V peak into 32 ohms. 0.75V peak is 0.53Vrms. Entering your 32R and 104dB in the spreadsheet and fiddling with loudness levels to hit 0.53Vrms comes up with 113.5dB. At that point the current requirement is 17mA(rms). The chip is good for around 25mA peak, which is around 17.5mA(rms), so that even works out. The power dissipation in the chip also looks OK if I've done the math on that correctly.
Take at look at the "hearing damage" chart at the bottom of that spreadsheet. 113.5dB SPL is pretty loud and should only happen in short peaks anyway. Average listening level will be way less. Musical peaks are often 2x to 3x the average voltage levels.
It also looks like I screwed up my previous description of 00940's amp too.
I looked at it a bit more today and it looks like he is using the LME49713 for the current buffer and another chip for the gain chip in a multiloop configuration. The gain chip is FET input so the thinking there may be that it wouldn't pull any significant current through the pot wiper, which can cause scratchiness and lead to pot failure. And it looks like he wound up with a +/-12Vdc converter rather than the +/-9Vdc in his first post. Which reminds me - putting 100uH chokes (coils) on the power leads coming from the USB in that inverting CMOY wouldn't be a bad idea to keep the PC digital noise out of the amp. All the resistors should be 1% metal film (if though hole) or 0.1% thin film if surface mount to keep noise down. That circuit could also be run off a 9V battery but the capacitor voltage would have to increase to 16Vdc which would make them a little larger physically. Those three 220uF caps should ideally be the new low ESR organic polymer electrolytics, like Mouser #667-6SVP220MX or Digikey #P16598CT-ND.
You are right, a "current buffer" is a chip or transistor circuit that is used mostly to source/sink current and not voltage. Low impedance headphone loads are the usual culprit for needing a lot of current, unless the sensitivity of the headphone's drivers are high enough where the voltage swing involved can be small, which is the case with your headphones. Remember ohm's law, voltage/resistance = current. If you are able to keep the voltage swing small then even a low resistance results in low(er) current flow.
Last edited:
Whoops - I left the "u" (micro) off the inverting CMOY output capacitor in the circuits above - 220 Farads, lol. With that fixed the organic polymer electrolytics have to go up to 470uF to be perfectly flat at 20Hz, and 3 have to be in parallel for the output, but the good news is those are only $1 each at Mouser these days with ESR=0.008R to boot.
I also forgot to capacitively load the output to see if the 4.99R isolation resistor is working. This one has 500pF in parallel with the 32R load. I've also split up the 10K input resistor to add an RF filter to the input. The 3.3uF film caps can be the cheaper 50Vdc/30Vac units since the voltage swings involved are small. 100uH coils on the USB inputs turn out to not be a good idea - they resonate. Ferrite beads would be better if EMI is a problem.
Here is a revised schematic with real-world (Mouser) part values filled in for everything.
sgrossklass - your idea is pretty interesting!
With that fixed the organic polymer electrolytics have to go up to 470uF to be perfectly flat at 20Hz, and 3 have to be in parallel for the output, but the good news is those are only $1 each at Mouser these days with ESR=0.008R to boot. I also forgot to capacitively load the output to see if the 4.99R isolation resistor is working. This one has 500pF in parallel with the 32R load. I've also split up the 10K input resistor to add an RF filter to the input. The 3.3uF film caps can be the cheaper 50Vdc/30Vac units since the voltage swings involved are small. 100uH coils on the USB inputs turn out to not be a good idea - they resonate. Ferrite beads would be better if EMI is a problem.
Here is a revised schematic with real-world (Mouser) part values filled in for everything.
sgrossklass - your idea is pretty interesting!
Attachments
Last edited:
You mean the coupling cap for the feedback network, right? This one just drops gain to unity at DC, in order to avoid amplifying DC offsets (which is more of a concern with DC-coupled outputs). It also allows using "real" ground, which tends to be quieter than virtual ground. The time constant would obviously be chosen high enough to cover the whole audible range at full gain, so f3dB <= 10 Hz typical.
The output cap should be self-explanatory - you've got +2.5 V on the output that you don't want to subject your headphone drivers to. Hence you either need the cap, or you need to hold headphone ground to +2.5 V as well with some oomph and very low output impedance behind it, or generate a negative supply, as described earlier.
Whoops - I left the "u" (micro) off the inverting CMOY output capacitor in the circuits above - 220 Farads, lol.
I also forgot to capacitively load the output to see if the 4.99R isolation resistor is working. This one has 500pF in parallel with the 32R load. I've also split up the 10K input resistor to add an RF filter to the input. The 3.3uF film caps can be the cheaper 50Vdc/30Vac units since the voltage swings involved are small. 100uH coils on the USB inputs turn out to not be a good idea - they resonate. Ferrite beads would be better if EMI is a problem.
A few comments if I may:
- the two 470uF are wasted in the voltage divider; they're in serie and that cuts the effective capacitance to 220uF. Since you only establish bias and don't drive the load with that divider, you can up the resistors values and reduce the caps values. You can now use a single 470uF from 5V to gnd for power and still have more power on reserve.
- I wonder if it wouldn't be better to use a higher value pot and a smaller resistor in // for feedback. With a linear pot and a carefully chosen // resistor, you might get usable attenuation curves while having extra safety if the pot goes wrong.
- 100uH coils are imo a better idea than ferrites as they start working earlier on but you need to size the cap at their output carefully to tame resonances; then a small R then 470uF (or 2X 470uF).
- the goal of being perfectly flat to 20hz seems to me a bit excessive. In practice, 470uF usually proves to be more than adequate for 32R headphones.
- I'd consider switching to something like an OPA2835 on such low voltages.
PS: about "my" usb amp. The reason I went with a fet input is both the ability to use it directly after a pot and the lack of worries about dc offset. Furthermore the opa134 is a relatively "slow" and behaved opamp. I can thus use it in multiloop with a faster opamp with gain, reducing distortion further. This might lead to a bit more noise than a two stages approach but since I designed this amp for my higher impedance headphones, it really doesn't matter. And the move from +/-9 to +/-12V is just that +/-9V didn't exist in that MTU2 serie and I wanted that size. One could use a +/-5V converter for lower impedance headphones. But this all OT.
Agreed. In addition to this, I would use a cap only in parallel with the lower (+2.5 V to ground) voltage divider resistor. A 47µ would be perfectly fine for 4k7 + 4k7 if there's nothing but the two 100ks connected.
If there is a cap in parallel with the upper resistor as well, you get a voltage divider of approximately cap1 impedance / (cap1 impedance + cap2 impedance) ~= 1/2 = -6 dB for AC noises on the supply, which then end up on your virtual ground. By contrast, with only 1 cap you get cap1 impedance / Rdivider/2, and the impedance of a 47µ at audio frequencies tends to be much smaller than 2k35, so attenuation of rail noise is much greater.
These are the small but decisive differences between single supply and split supply operation...
Hmm... so basically the voltage divider setup as suggested for the original single-9V cMoy is inherently flawed. Oops. (It's OK for a dual-9V one, though with a "real" ground in between the cells you can leave out the resistor divider altogether.) Then again, the virtual ground also has to support the feedback network there, so... For a 9 V cell it's probably alright (though power bandwidth in the bass may be a bit limited), but it's not very suitable for generally stable but "dirty" supplies such as USB power.
Last edited:
Agreed on the single cap.
I wouldn't be that harsh on the simple cmoy, it kind of works with higher impedance headphones and music (roughly a symmetrical signal). But, here you need the cap coupling anyway as USB ground is tied to earth. If connected as a line out, that could lead to shorts.
I wouldn't be that harsh on the simple cmoy, it kind of works with higher impedance headphones and music (roughly a symmetrical signal). But, here you need the cap coupling anyway as USB ground is tied to earth. If connected as a line out, that could lead to shorts.
00940 & sgrossklass - thanks guys for the feedback!
I agree about the series virtual ground caps. Changed to a single 47uF. Oddly enough Mouser doesn't stock those in 6.3V so went to 10V. The AC plot still looks just fine.
One question though about initial conditions. At t(0)+ when the plug is first inserted in the USB port the 47uF cap is a short and the non-inverting pin is at the bottom rail for an instant. My best guess is it shouldn't matter since the op amp rails are at zero at that instant and have no initial power, plus that non-inverting input has the 24.9K in series. I know some op amps are not happy with an input at the bottom rail, although it looks like that OPA2835 00940 suggested is rated for it. I've started an alternate op amp list on the schematic and added that one, along with the OP's LME49723.
Another problem with those original series virtual ground caps hit me this morning too. They would exceed the 10uF maximum direct capacitance on the power rails from the USB spec:
http://www.ti.com/lit/an/slyt118/slyt118.pdf
Problem solved by going with the single 47uF. The series pair would have glitched the USB bus on insertion. That would probably also limit being able to put one of the 470uF's across the rails, otherwise I would adopt that.
I've also added decoupling caps on the op-amp pins that I forgot. I realized that since the gain is variable anyway there is no benefit in carefully having the series input resistors add up to 10K. So I've gone back to the more standard 10K value.
I like the idea of using a linear pot resistor-loaded to simulate the log taper. Added. I know from some past math on pots that the 5x parallel resistor I had already slightly loads the tail end of the taper, sort of the absolute minimum to use with an audio taper. So I've just dropped that parallel resistor to 200K from 267K and switched to the linear pot.
Yeah with the 100uH coils the AC plot was coming up with a resonant dip in the audio band. Probably resonating with the two virtual ground caps, but I didn't spend any time with it. I tried going to 10uH which seemed to solve/reduce the problem, but with stuff like that I start wanting to actually build it and see the real results. Too many parasitic C and Ls around. I haven't done that much with USB. I 'm wondering if there is some standard coil or bead value(s) that folks use for isolation.
I agree about the series virtual ground caps. Changed to a single 47uF. Oddly enough Mouser doesn't stock those in 6.3V so went to 10V. The AC plot still looks just fine.
One question though about initial conditions. At t(0)+ when the plug is first inserted in the USB port the 47uF cap is a short and the non-inverting pin is at the bottom rail for an instant. My best guess is it shouldn't matter since the op amp rails are at zero at that instant and have no initial power, plus that non-inverting input has the 24.9K in series. I know some op amps are not happy with an input at the bottom rail, although it looks like that OPA2835 00940 suggested is rated for it. I've started an alternate op amp list on the schematic and added that one, along with the OP's LME49723.
Another problem with those original series virtual ground caps hit me this morning too. They would exceed the 10uF maximum direct capacitance on the power rails from the USB spec:
http://www.ti.com/lit/an/slyt118/slyt118.pdf
Problem solved by going with the single 47uF. The series pair would have glitched the USB bus on insertion. That would probably also limit being able to put one of the 470uF's across the rails, otherwise I would adopt that.
I've also added decoupling caps on the op-amp pins that I forgot. I realized that since the gain is variable anyway there is no benefit in carefully having the series input resistors add up to 10K. So I've gone back to the more standard 10K value.
I like the idea of using a linear pot resistor-loaded to simulate the log taper. Added. I know from some past math on pots that the 5x parallel resistor I had already slightly loads the tail end of the taper, sort of the absolute minimum to use with an audio taper. So I've just dropped that parallel resistor to 200K from 267K and switched to the linear pot.
Yeah with the 100uH coils the AC plot was coming up with a resonant dip in the audio band. Probably resonating with the two virtual ground caps, but I didn't spend any time with it. I tried going to 10uH which seemed to solve/reduce the problem, but with stuff like that I start wanting to actually build it and see the real results. Too many parasitic C and Ls around. I haven't done that much with USB. I 'm wondering if there is some standard coil or bead value(s) that folks use for isolation.
Attachments
Last edited:
I have a couple of questions:
1) What is the difference between "real" ground and virtual ground?
2) So using say a 10uF electrolytic in series with a 2K ohm resistor to "real" ground will help to lower DC offset when using bipolar and higher input bias current op amps?
Attachments
Lol! You are right, that is a screw-up.
For a single supply there should be just the one decoupling cap from V+ to (real) ground. Thanks for the find!00940 - you are right about the OPA8235 being a good 5V chip. It sims well, below. 0.85Vrms output before clipping vs. the 0.53Vrms with the LME49710. That hits 36.7mA peak into 32R. Looks like the chip is rated for 40mA vs. 25mA or so for the LME49710. I come up with around 90mW of total package dissipation, including the tiny quiescent, well within the 300mW or so for the SOIC8 package. Just $3.78 at Mouser!
Figure 66 in the datasheet led me to remove the resistor on the non-inverting input. Given the location of the pot the impedance looking out the inverting input will be all over the place with pot position. Kind of a puzzle to figure out what resistor to pick for the non-inverting input to best match impedance looking out that port. The two 4.7K's in parallel are probably good as any.
Attachments
Last edited:
Just got the OP's headphones to 120dB SPL!
I just found the AD8656. 2.7 to 5.5V, rail to rail input and output, 75mA output current and even has current limiting to deal with TRS jack insertions.
The sim below shows it going to 1.58V peak = 1.12Vrms before clipping (50mA rms at that point). Well as luck would have it the OP's headphone spec of 32R and 104dB/mW just happens to hit 120dB SPL at 1.12Vrms. Pretty slick to be able to do that with nothing more than USB input power.
The chip is a dual and the package power calculations come out to somewhere around 137mW of dissipation for both halves plus the quiescent. So just one chip should do the job for both channels. Still cheap - $3.25 at digikey.
Green and blue are output and input voltage on the left scale, red it chip output current on the right scale.
I just found the AD8656. 2.7 to 5.5V, rail to rail input and output, 75mA output current and even has current limiting to deal with TRS jack insertions.
The sim below shows it going to 1.58V peak = 1.12Vrms before clipping (50mA rms at that point). Well as luck would have it the OP's headphone spec of 32R and 104dB/mW just happens to hit 120dB SPL at 1.12Vrms.
Pretty slick to be able to do that with nothing more than USB input power.The chip is a dual and the package power calculations come out to somewhere around 137mW of dissipation for both halves plus the quiescent. So just one chip should do the job for both channels. Still cheap - $3.25 at digikey.
Green and blue are output and input voltage on the left scale, red it chip output current on the right scale.
Attachments
Last edited:
Don't take that app note too strictly. There are plenty of commercial products that don't respect it. Especially about the question of correct identification for 500ma draw. Very few USB ports will actually limit to 100ma.
Wrt capacitance, the 10uF is mostly about inrush current. Sim the inrush current of a 10uF ceramic cap and then you can sim a RC filter for a similar inrush, even with a 470uF cap at the output. Or at least 20uF oscon in // with 2.2uF x7r.
The 47uF in the bias network doesn't need to be polymer in my view.
Can't claim credit for the opa8235, I just remembered the part from the pupdac. See also the LMH6643. Btw, be careful with sims of output levels. Most spice models for opamps aren't very good about that (or anything related to their power pins).
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.