@counter culture, thanks. I didn't know TI had a better part--although it would be interesting to see which wins in real world testing rather than the "datasheet wars" (I have no idea which would). I know some really high-end uber-priced gear has used the Cirrus part. But both, and their related overhead, are off the menu for now. I've looked at some of Wakibaki's posts, etc. He seems like an interesting guy, but last I checked, he was banned here so I haven't been able to contact him.
Thanks for the tips, I'm just getting my feet wet in diy audio. I'm one of those that were enticed by the O2 and RS's detailed docs.
So the datasheet explains the components, but some things on the PCB I don't get.
The VR1 has 3 rows of 4 pads. I'm thinking a row for each channel. I also see that one column is not connected to anything, which I think is because you don't need that third pin for a pot. But I don't understand why are the 2 channels duplicated (if my assumption about one row per channel is correct).
Also looking at skkhai post #1254, I see that he's running a wire from the pot to the tab that is not connected to anything for both channels, am I missing something?
Also There are three pairs going to the 1/4" jack when there are only 3 tabs to connect to. So why run two wires to one tab?
May be I'm just going OCD on this, so apologies for that
The volume pot is explained in Circuit Board Construction. The 6 holes closest to the edge of the board are for the 15mm long pots. The 6 holes farthest from the edge are for the 20mm long pots. All 6 pins for each are connected and required, it's just that two have little "fingers" that connect the pads to the ground plane (it's called a thermal relief and is for soldering purposes). The two sets of 6 are electrically the same (duplicates).
If you're going to panel mount the pot, just wire each pin to either set of corresponding pins on the board. Each row of three (each channel) should be twisted together with the ground wire for that channel. or you could use 4 twisted pairs with 4 grounds. There are instructions for that in the article as well.
If you're going to panel mount the pot, just wire each pin to either set of corresponding pins on the board. Each row of three (each channel) should be twisted together with the ground wire for that channel. or you could use 4 twisted pairs with 4 grounds. There are instructions for that in the article as well.
As I am building this O2 (actually in the ordering Mouser stage) I can't get the dismisal of DC offset as unimportant. Take my grado for example. Typical listening is about .3V across the voice coil. Now put 5 mV of offset across the voicecoil. That is I/R=0.15mA. Now with big push-pull amp coils people get a bit anxious about a milliamp or so of current. With a mic input transformer this would be dead in the water. I think of a grado headphone voicecoil as being very susiptable to flux modulation. I mean where is it a given that hysteresis distotion is negligible with 5 mV's of offset? To put it into tangle terms that Poffset= 0.15mA * 5mV = .75mW. With a speaker of 98db/mW how can we say this is insignificant?
So with a typical Grado listening listening level of 0.3V, the DC component from the amp is only -18 dB down. I am not saying it harms the headphones, but I would be surprised if it doesn't significanly impact the driver response?
Do you think more emphasis should be placed on trying to eliminate the this DS offset? And I am not picking on the O2, as have wondered about ths with any DC coupled amp I have built, coincidentally I've yet to find a dc couped amp that I like with grados, hoping this O2 will do the trick.
Are you able to measure waterfall or basic frequency response of O2 driven headphones with the dscope ?
Maybe I am being anal, but -18dB down doesn't seem insignificant, then again maybe I am just pissed that Mouser zapped my O2 project off my account and I have to start over again
So with a typical Grado listening listening level of 0.3V, the DC component from the amp is only -18 dB down. I am not saying it harms the headphones, but I would be surprised if it doesn't significanly impact the driver response?
Do you think more emphasis should be placed on trying to eliminate the this DS offset? And I am not picking on the O2, as have wondered about ths with any DC coupled amp I have built, coincidentally I've yet to find a dc couped amp that I like with grados, hoping this O2 will do the trick.
Are you able to measure waterfall or basic frequency response of O2 driven headphones with the dscope ?
Maybe I am being anal, but -18dB down doesn't seem insignificant, then again maybe I am just pissed that Mouser zapped my O2 project off my account and I have to start over again
Last edited:
Your calculator sucks
yep microwatts, sorry for the typo (they really should have mu on keyboards
But the -18dB down figure is correct. The voltage ratio is .005/.3V = 0.0167= -18dB we have a DC "signal". Now I am not too hung up on it but it just seems significant, having read RS's awesome blog it is one thing that was a bit taken as as a given without an explaination or reference. I mean this is Objectists amp.
I also ejoyed his paragraph that I swear was directed at me regard powering my old discontinued AKG's
I laughed my a$$ off( at myself), RS says to buy another amp in that paragraph!, I wish he would measure an amp that gives 11Vrms of clean power like the O2 does at 7Vrms (hint they don't exist)Can't wait to get this little SS marvel in action though, I have a sneaking suspicion that 7Vrms on my old AKGs from the O2 will sound better than most other amps I have had my hand on.
Last edited:
Since you raise the issue of offset, perhaps you might find one or two of my observations illuminating
Caveat here - its been a while since I read through RS's awesome blog, so he might have updated things in the meantime. But as I recall the stated firgure for maximum output offset was wrong, by a factor of 2 if I recall correctly. The typical of course is considerably less. There's also a way to reduce the typical offset which is by balancing the impedances seen by the + and - input terminals of those output buffers. If you or anyone else is interested in output offset I'll be happy to share a few more details. The proviso is I haven't built an O2 myself so can't say that any mods I've thought up have been field tested. They'd need verification by at least one of the monster army of O2 devotees out there
I like this idea very much and I will follow this ODA/ODAC project.
Actually I drive your O2 over the uDAC-2's line out. I like this chain very much as it sounds much better in comparison with the uDAC-2 headphone out. I can even drive my K701 with this chain up to almost uncomfortable volume. I achive a quality which is - subjectively - similar to the RME Fireface UC (I have this magical device at work) with this new chain. Respect!
I'm curious how much more sound quality will be achievable by the ODA/ODAC project.
with respect, while the RME is a very nice piece of hardware for the studio, there are many DIY devices on these forums that beat out the RME analogue line and headphone outs. they are high quality, rugged designs with nice, but not exceptional mic pres, good quality clocking, excellent rock solid low latency drivers with a very flexible mixer, but the headphone performance and even really the dac and analogue stage performance at -96db THD+N its not setting the world on fire for DA; certainly not as good as some of the more highly engineered modern designs on here.. also K701 is a pretty innocuous ~50ohms thats not too low to be troublesome, but low enough that it only requires unity gain, with the main thing being a bit extra current, only a follower is needed.
and yes i owned a FF 400 for some time and still have an old HDSP9632 around for general ADC and measurement purposes. dont get me wrong i loved it and i'm still glad i have its little brother, but with enough attention you will find (when you do it) it can be handily beaten
methinks your disease has only just begun =)
and yes i owned a FF 400 for some time and still have an old HDSP9632 around for general ADC and measurement purposes. dont get me wrong i loved it and i'm still glad i have its little brother, but with enough attention you will find (when you do it) it can be handily beaten
methinks your disease has only just begun =)
May be ... I never thought the RME device is the best out there to drive my K701, but it is a pretty good one and the only one I had hands on in the last months. I have no other device to do comparisons, only the EMU0404, but this is definitely worse. And the mentioned uDAC-2, this also doesn't beat the level of the RME.
But normally I use the RME FF for other things in laboratory environment at work and it does a good job there ... due to its flexibility.
Yes I think this is true
... thats why I really can't wait for the next level ODA/ODAC.
Since you raise the issue of offset, perhaps you might find one or two of my observations illuminating
Caveat here - its been a while since I read through RS's awesome blog, so he might have updated things in the meantime. But as I recall the stated firgure for maximum output offset was wrong, by a factor of 2 if I recall correctly. The typical of course is considerably less. There's also a way to reduce the typical offset which is by balancing the impedances seen by the + and - input terminals of those output buffers. If you or anyone else is interested in output offset I'll be happy to share a few more details. The proviso is I haven't built an O2 myself so can't say that any mods I've thought up have been field tested. They'd need verification by at least one of the monster army of O2 devotees out there
Yes I would be interested, especially as I am ordering parts. Would it involve resistove compensation on the negative output opamp pins?
I already placed a 4.4uf Wima cap on the mouser build form and halved the R12/13 as an option to try to minimize dc offset but guess I don't like the idea of making the output stage a tougher (especially with the volume pot in between.) Would love to try a suggestion, this is just a hobby and I fully expect to blow up a few components, so don't let that hold you back
I'm sure the issue is nothing but I do have a long history of finding that Grados hate dc-offset.
Yep - you got it. Put resistors (I think from memory its 39k?) between the output (pins1,7) and the -ve inputs (pins2,6), not forgetting to cut the track between them first. You'll also want caps in parallel with those resistors - I suggest 220pF NP0 ceramic. Best choose 0805 SMT parts for both then you can solder direct to the IC pins close to the package.
The typical offset due to the bias current is 4mV (because two opamps share the same biassing resistor) - fitting the R in the feedback will typically reduce this substantially. I'd expect under 1mV in practice from the contribution of the offset voltage and bias current once the bias current is compensated for.
Yeah that's a much more expensive (and potentially lower performance too) way to skin the cat with the price of those Wima caps
<edit> I just realised I goofed in the resistor values - because we want to compensate for two bias currents the resistors you fit should be 82k (to the nearest E24 value).
Last edited:
Regal you bring up an interesting point and I would welcome some definitive data on the effects of tiny amounts of DC offset on headphone drivers. I have seen several "guidelines" stating up to around 20 mV is generally considered acceptable but none of them were terribly definitive.
I would expect a proper guideline should take headphone impedance and even sensitivity into account. One would expect headphones like the Grados, Denon D2000s, and IEMs likely more sensitive to offset than say HD600s.
At a typical listening level with relatively sensitive headphones of say 200 mV, a 5 mV DC "signal" is still 32 dB down. In terms of the electromagnetic force from the voice coil, I believe it's proportional to the square of the current (Maxwell's force equations, etc.). Which means for very small values of current--such as created by DC offset, the force produced (i.e. driver "offset") is genuinely miniscule--the manufacturing tolerances of the driver suspension likely create much greater variations in offset.
Likewise, I suspect (but can't say for certain) a tiny amount of DC current is not going to significantly change the operation of the driver. Drivers are basically an electromagnetic motor and all you're really concerned about is their ability to convert current into proportional force over the audio range.
It's worth pointing out my Benchmark DAC1's headphone amp has under 0.5 mV offset while my O2 amps are all around 3 mv - 4 mV. In the blind ABX testing, nobody with any of my headphones, including 21 ohm B.A. Ultimate Ears, low impedance Etymotics, and 25 Ohm Denon D2000s, could hear any difference. So in terms of being somehow audible, it certainly didn't show up in the tests.
I can use my lab microphone "head box" and/or "ear tube" to measure headphone performance, including such things as square wave response, distortion, frequency response, etc. on the dScope and plot the results (one the same graphs for comparison) for different amounts of DC offset. It might be an interesting topic for an article.
If anyone knows of any definitive research on this, please post a link here?
It's also worth pointing out lots of other headphone amps, like the Mini3, have offsets around 5 mV or even higher. If it really was a problem you would think someone would have identified it as such by now?.
As for minimizing the O2's offset, you're on the right track with lowering R12 and R13. You can lower them even more if you like (say 10K) as the low frequency roll off and phase shift will still remain better than lots of highly praised amps out there. If you go really low in value you'll start significantly changing the volume control curve but that may not be a bad thing. U1 mostly has to drive the 1.5K feedback resistor so lowering R12 and R13 have little effect on its loading and hence performance.
You might also want to consider buying extra 4556 op amps and hand selecting them for minimum offset. If you want to hack up the board, and can find the room (or somehow work with SMT parts which I can't really imagine without pads to solder them to), you can add compensation resistors. But I'm really not sure it's worth it. You might create higher distortion and/or stability problems. Without being able to measure the effect of the hack, I'd advise against it.
As stated in the design documents, the DC offset is a trade off with the O2. Lower offset op amps fall on their face trying to drive many headphones directly so that's not an option. Capacitor coupled outputs add significant distortion and LF roll off and phase shift (and there's no space for suitable caps). A DC servo would have meant an even larger case and PC board and shorter battery life. And there simply wasn't room for the compensation resistors and still route the high current traces properly to the 4556 op amps.
As for Abraxalito being concerned about the max offset, I'll re-check that number. It's possible the values of R12 and R13 changed from when I did it. In practice, I've measured a tube full of 4556s of different types (D and DD) and several different date codes. Every one has been under 5 mV in the O2. For anyone with a decent DMM it's easy enough to check if it's a concern and non-exotic op amps like the 4556 are very consistent and stable over time and temperature with respect to offset. So if your O2 measures 3 mV today you can be sure it will remain close to 3 mV. It's not like some fussy discrete design with JFETs glued together in the input stage trying to stop thermal drift.
But I still like the idea of measuring at what point does DC offset make a measurable difference on headphone performance. That threshold is still likely below the amount that would produce an audible change.
I would expect a proper guideline should take headphone impedance and even sensitivity into account. One would expect headphones like the Grados, Denon D2000s, and IEMs likely more sensitive to offset than say HD600s.
At a typical listening level with relatively sensitive headphones of say 200 mV, a 5 mV DC "signal" is still 32 dB down. In terms of the electromagnetic force from the voice coil, I believe it's proportional to the square of the current (Maxwell's force equations, etc.). Which means for very small values of current--such as created by DC offset, the force produced (i.e. driver "offset") is genuinely miniscule--the manufacturing tolerances of the driver suspension likely create much greater variations in offset.
Likewise, I suspect (but can't say for certain) a tiny amount of DC current is not going to significantly change the operation of the driver. Drivers are basically an electromagnetic motor and all you're really concerned about is their ability to convert current into proportional force over the audio range.
It's worth pointing out my Benchmark DAC1's headphone amp has under 0.5 mV offset while my O2 amps are all around 3 mv - 4 mV. In the blind ABX testing, nobody with any of my headphones, including 21 ohm B.A. Ultimate Ears, low impedance Etymotics, and 25 Ohm Denon D2000s, could hear any difference. So in terms of being somehow audible, it certainly didn't show up in the tests.
I can use my lab microphone "head box" and/or "ear tube" to measure headphone performance, including such things as square wave response, distortion, frequency response, etc. on the dScope and plot the results (one the same graphs for comparison) for different amounts of DC offset. It might be an interesting topic for an article.
If anyone knows of any definitive research on this, please post a link here?
It's also worth pointing out lots of other headphone amps, like the Mini3, have offsets around 5 mV or even higher. If it really was a problem you would think someone would have identified it as such by now?.
As for minimizing the O2's offset, you're on the right track with lowering R12 and R13. You can lower them even more if you like (say 10K) as the low frequency roll off and phase shift will still remain better than lots of highly praised amps out there. If you go really low in value you'll start significantly changing the volume control curve but that may not be a bad thing. U1 mostly has to drive the 1.5K feedback resistor so lowering R12 and R13 have little effect on its loading and hence performance.
You might also want to consider buying extra 4556 op amps and hand selecting them for minimum offset. If you want to hack up the board, and can find the room (or somehow work with SMT parts which I can't really imagine without pads to solder them to), you can add compensation resistors. But I'm really not sure it's worth it. You might create higher distortion and/or stability problems. Without being able to measure the effect of the hack, I'd advise against it.
As stated in the design documents, the DC offset is a trade off with the O2. Lower offset op amps fall on their face trying to drive many headphones directly so that's not an option. Capacitor coupled outputs add significant distortion and LF roll off and phase shift (and there's no space for suitable caps). A DC servo would have meant an even larger case and PC board and shorter battery life. And there simply wasn't room for the compensation resistors and still route the high current traces properly to the 4556 op amps.
As for Abraxalito being concerned about the max offset, I'll re-check that number. It's possible the values of R12 and R13 changed from when I did it. In practice, I've measured a tube full of 4556s of different types (D and DD) and several different date codes. Every one has been under 5 mV in the O2. For anyone with a decent DMM it's easy enough to check if it's a concern and non-exotic op amps like the 4556 are very consistent and stable over time and temperature with respect to offset. So if your O2 measures 3 mV today you can be sure it will remain close to 3 mV. It's not like some fussy discrete design with JFETs glued together in the input stage trying to stop thermal drift.
But I still like the idea of measuring at what point does DC offset make a measurable difference on headphone performance. That threshold is still likely below the amount that would produce an audible change.
Last edited:
FWIW I seem to recall a figure on your blog of 20mV. When I looked at the 4556 datasheet I came up with 46mV (40mV of this is the worst case input bias current through R12/13). With the compensation resistors I've suggested to regal, the worst case offset would I reckon decrease to 10mV and typically probably 0.5mV.
That pretty much concurs with my measurements over more than 20 different date codes. Worst case voltage offset I found was 2mV. Almost all your 5mV arises from the typical bias currents.
yep microwatts, sorry for the typo (they really should have mu on keyboards
If you're using Windows, try holding down Alt while typing 230 on your ten-key (numlock on). Let go of Alt and, voila!
Alt+230 = µ
Alt+234 = Ω
Alt+248 = °
Alt+241 = ±
There are many more. If you only have a laptop, you'll have to use the overlayed 10-key with numlock on. The strip of numbers over the keyboard doesn't work for this.
Also you can just open up the Character Map (charmap.exe) and copy the symbols from there.
@abraxalito, my 4556 datasheets list the worst case bias current as 500 nA. That means 20 mV worst case voltage across R12 or R13--not the 40 mV you suggest. V = I*R. 500 nA * 40K = 20 mV. The amps are unity gain so 20 mV offset at the input is 20 mV offset at the output (ignoring the voltage offset for the moment which you correctly identified as 6 mV worst case). Am I missing something?