A version of an O2 Desktop Amp (ODA)

I'm guessing one problem may have been his 40.2K return resistors. With 4 op amps rather than two that input bias current would double and the voltage drop across the 40.2k would double. The DC offset of the O2 may have gone up to 6mV. :)

Thought about this some more, as I was sure I was just being dumb.
I was rather foolishly thinking about it from the perspective of the current being sourced from the pot/coupling cap.
Got it now. Double the bias current flowing out through the 40.2K = double the DC voltage at non-inverting input = worse offset. Does that sound about right?

Edit: cocked up my maths.
4 * 180nA typ. input bias current over 40.2K = 28.9mV. I see what you mean now, that's rather large.
 
Last edited:
weslito - sorry, I mis-read your post. Easy to do with text. :) Most of my posts are geared towards helping people just learning electronics and tend to be on the fundamental side. The forum here has many true audio pros posting and some real audio legends. Occasionally someone misses the intended audience and I get swatted. :p

The input bias current is what is flowing into (or out of) the input terminals to the op amp. The same amount may or may not flow through both the inverting and non-inverting pins depending on op amp design, but for sake of discussion I'll assume both are the same here. In the case of the NJM4556A chips in the O2 the data sheet shows a typical value of 50nA (0.05uA), with a max about 10 times that. Given the coupling capacitor in the O2 design, all of that current has to flow through the 40.2K ground return resistor since it has nowhere else to go (or come from).

So with two NJM4556A inputs tied together that would be a total of (2)(50nA) = 100nA = 0.1uA into (or out of) the two chip non-inverting inputs and through the one 40.2k resistor, for a voltage drop of 40200 ohms * 0.0000001 amps = 4mV. Multiplied by the 1x (current buffer) closed loop voltage gain of those stages that is still 4mV offset reflected to the op amp output.

The O2's I've built typically have 3mV of DC offset, so the result is in the ballpark. In addition there is the inherent DC input offset of the NJM4556A, which is listed at 0.5mV typical and 6mV maximum, times the closed loop voltage gain of 1 (buffer).

In one of the mods I posted in the O2 mod thread I cut the feedback PC traces (output to non-inverting input) on two of the NJM4556As and soldered in 40.2K resistors in the feedback loop to cancel out that input bias offset. That works since that same input bias current flows out of (or in to) both op amp inputs (assumption from above, which may or may not be valid). Adding a resistor in the feedback path of the same amount produces the same voltage drop, but now will get a polarity change at the op amp's differential input and cancel in the sum with the non-inverting input of the same DC voltage amount.

When I actually tried and measured it the result was the output offset voltage of that channel of the O2 amp dropped by about half, from 3mV to 1.5mV. I took the difference to be the input offset voltage of the NJM4556A or a difference in inverting and on-inverting bias currents. But as I write this I'm thinking I may have made a mistake on that one. :) Since the two halves of the NJM4556A are in parallel, and double the input current is flowing through that external 40.2K resistor to ground, the nulling resistors in the feeback loop of each op amp half may actually need to be 80.4K (82K standard value) to null properly when summed back together. Hmmm.... I'll have to ponder that one.
 
Last edited:
Yeah, that makes sense. Presumably the offset nulling effects of the feedback resistor would come at the price of rather high Johnson noise. 40.2K seems quite a large value to use there. 3mV offset isn't very harmful, is it worth the trade off?

Another question, if you don't mind me picking your brains. The 'typical' and 'max' input bias current often vary by quite a large margin, as you point out. Are the bias currents of non-inverting and inverting within the same device usually fairly well matched? If not, your feedback resistor would have to be carefully (and painfully) matched.
Also what about the two sides of a dual opamp (like the NJM4556)?
 
Last edited:
weslito - good questions! I'm still stuggling with the Johnson noise stuff myself with this ODA project so I'm not going to have any useful answers, unfortunately. In general though you are probably right, the resistor in the feedback loop (especially one that large) will probably be a trade-off of adding some Johnson noise in return for reduced output DC offset.

One of the slick things RocketScientist/NwAvGuy did with the O2 design was arrange things so that the pot and the large(r) 40.2K resistor were at the input of the stage with just 1x closed loop voltage gain. By putting more resistance at the input of a voltage gain stage, as with my ODA design here, any noise gets multiplied by the gain of the stage, up to 6x in this case.

But then again it may turn out that Johnson noise isn't the predominate source of noise in a particular design. I've been kind of amazed so far that I'm not getting more noise with the pot up front. In fact I just placed an order with Mouser for a 50K pot and intend to swap it out with the 5K and see what I can hear.

The data sheet for the OPA627 has an interesting graph on pp. 4 of source resistance vs. voltage noise.

http://www.ti.com/lit/ds/symlink/opa627.pdf (opens pdf)

The inflection point in op amp noise vs. source resistor noise comes at about 10k.
 
Very interesting, never noticed that graph before. It goes on to say "above a 2kΩ source resistance, the op amp contributes little additional noise. Below 1kΩ, op amp noise dominates over the resistor noise".

This can be seen on the graph also, as 2kOhms is the point at which resistor only noise is 4.5nV/√Hz; equal to that of the amplifier.

So as you say, 10K (it's actually ~8K if you zoom in on it) is the point at which the op amp contributes no considerable noise of it's own.

Up at 40K it's about 26nV/√Hz. Is it possible to convert this into say, dBV, for a given frequency? I've no idea how you'd go about that.
 
xnor - good link! Rod Elliot has has great stuff on that site. His power supply page here in section 5

Small Power Supplies

leaves little doubt that the power supply I have laid out for this project is utterly ridiculous in terms of noise removal overkill. :)

RocketScientist said he measured the O2 noise floor with his dScope and said the LM7x12s didn't make any difference. I know that opc has written the same about the LM3x7s and his Wire amp with the AP measurements. So me adding the LT regulators here is essentially just $15 in the wastebasket. But I've had a pair of them here for awhile and have wanted to try them out, and this is DIY audio after all (!), so in they go! :p But back OT, that is a helpful noise info page.

weslito - I haven't worked that part of the math out yet. I'm going to just swap in the 50K pot first and see if I hear background noise show up. If not I'll probably futz with the numbers a bit to figure out why. At 50K I should be well up on that "resistor over amp noise" curve. But even there it seems max noise will be at the middle of pot range. Lower in the range, which is where I usually run the pot due to "hot" sources, and both the signal and noise from the other half of the pot should get attenuated by the voltage divider. Or at least that is how it appears, I may have that mucked up. :)

Being able to use a higher resistance pot like that, if excessive noise doesn't show up, would make fuller use of the FET input on the OPA627. I've added a 500R series resistor too between the op amp and the pot wiper to help lower the DC change as the pot it turned from one end to the other vs. the parallel resistance combo of the feedback network out the inverting input.
 
Last edited:
The article linked seems to me (unless I skipped over some parts) not to address power supply self-noise. That being the load regulation - a headphone amp is going to be putting out peak currents in the hundreds of mA range so the power supply's output impedance is crucial.

<edit> When you examine the datasheets for the 78/79 series regulators you tend to find the output impedance plots look fairly decent, at least at LF. But then when you look closer they show the output impedance at high-ish output currents (like half an amp or so) - way beyond what's normal in opamp circuitry. At lower output currents the Zout gets considerably higher.
 
Last edited:
yep and the resonant pole in the impedance will be effected and moved around by the output caps and inductance. the same can be said for the LT, just because opamps have high PSRR doesnt mean you can forget designing a decent power supply. OPCs headphone amps have good PSRR, but they also have quite hardcore bruteforce local decoupling right on the device pins and a compact layout on the board and the power supply

just assuming you will have low noise operation because you used more expensive parts seems a bit against the grain...
 
Opamps do have high PSRR, but only for low frequencies. The PSRR always eventually falls at 20dB/decade above some corner freq which can be the open loop pole freq. Added to this, their output stages operate in classB meaning lots of harmonics on the rails. Good low-ESR HF decoupling is a must.
 
My 50k pot arrived today from Mouser, and I have to say the Hakko 808 is worth its weight in gold. I didn't have to go back and "unstick" even one pin. The old pot just lifted right out. Pretty amazing.

Equally amazing is no background noise/hiss at all on any gain setting or any rotation of the pot, with my old ears. I sure wouldn't have predicted that. Disclaimer: younger/better ears, better cans/IEMs, and certainly proper measuring equipment are likely to pick up noise and hiss. So YEMV (Your Ears May Vary). :p

I have even tried injecting noise with my finger and finally found a way. The metal shaft and faceplate on the Bourns pot is not grounded. I have a plastic knob on the pot, but by touching the shaft behind the knob or metal faceplate I can inject a small amount of 60 cycle hum with the gain on 6x and the volume pot all the way up. Even that goes away audibly on 4x gain. Touching anything else doesn't do it due to the grounding - gain switch, filter caps, bass boost switch.

So I'm going to call this arrangement of using the pot as both the input load and the op amp input bias current return resistor a big success, circuit below. I've eliminated one resistor over the more typical setup, as in the O2 headphone amp, of the pot feeding an input bias ground return resistor through a cap. The whole thing works because of the tiny 5 pico-amp input current of the FET op amp. With the 50K pot that translates into just 0.25uV of drop across the pot, worst case. Rotating the pot end to end still results in just a few micro-volts of DC offset variation at the output of the amp.

My big goal with this amp with to see if I could reduce the DC offset from that of the O2 amp. The ac parameters - distortion, crosstalk, etc - are likely a mess vs. the O2 due to the layout, buffer in the op amp loop, more op amps in parallel on the output, and other changes. This amp is just for fun to see if I could get the DC offset down. The amp sounds good to me, but that is purely subjective and the O2 was all about objectivity, so nothing here is going to give the O2 or other properly measured amps like the Wire any competition. :)
 

Attachments

  • ODA with 50K pot - circuit.jpg
    ODA with 50K pot - circuit.jpg
    141.5 KB · Views: 691
Last edited:
qusp and abraxalito - good thoughts! I should really be more specific with the power supply stuff and separate "ripple" comments from "noise". When I think ripple I think of correlated signals like mains frequency and harmonics, or current from the signal frequency at the moment multiplied by power rail wire resistance to form microvolt signal-frequency noise on the power rails. For noise I tend to think more about uncorrelated signals like diode wideband noise.

So you guys are right, the pre-regulator setup gets the ripple down to extremely small levels, at least the mains fundamental and harmonics, but uncorrelated noise is another issue. Those LT regulators may help in that department. I agree about decoupling caps! Very important.

qusp - I'm finally messing around with surface mount stuff. :) So far I'm keeping it to 1206 to make them solderable with bad eyesight, lol. I see why you like those z-foil resistors. I've spent some time looking at the specs now and those are pretty amazing.
 
Last edited:
agdr, using xnor's helpful link you can get an idea of how the Jonhnson noise is affected by your mods.

For your 40.2K feedback resistor:
Johnson noise = √(4k*T*R*f) = sqrt(4*(1.38e-23)*300*40200*20000) = 3.65µV

And to dBV:
20 * log (1V / 3.65µV) = 108.8dBV

In the absence of proper test equipment, this seems a much more solid approach than trying to measure noise with your ears :)
 
In the absence of proper test equipment, this seems a much more solid approach than trying to measure noise with your ears :)

Lol! :) But ears are the whole reason for audio equipment! :D In my case I really am more concerned in whether I can hear the background hiss or not, just myself personally, and not the absolute (measured) value. For my purposes, if I can't hear it, whatever level that is, then it really doesn't matter (to me).

But I fully agree, the math will then come up with a good approximation on what the absolute noise voltage is at that point, which will help attach a number to the result, which would be interesting.

There is a huge problem with my design though that didn't occur to me until last night. The wiper on the pot will eventually get bad over time - they all do - and when that happens bad things will happen with the output DC offset. So *don't build this* if anyone has been following along. It could damage headphones over time when the pot goes bad. If the pot can't be directly coupled then there isn't much justification for using the pricey FET input op amp.

I'm rethinking the whole thing. Might be better to stay with RocketScientist's original design for the O2, but with the extra paralleled NJM4556 chips, and then use a DC servo just on the output stage. The servo could be done with the much less expensive bipolar dual OPA211, which also has very low input offset voltage numbers, as long as the input resistances are kept about the same to cancel input bias currents.
 
Last edited:
This fun little DIY project is getting a complete reboot. :) Again, this amp isn't going to be distortion-measured, so for the real deal build OPCs new Wire amp version! That sounds like its going to be a real winner.

* The issue of the pot wiper wearing out is solved by dropping 100K resistors from the wiper to ground to insure a DC return path for the op amp input current when the wiper eventually nuts up. Same as AMB has done on the mini^3 which also has a direct-connected pot wiper.

* OPA827 now wrapped around a LME49600 on each channel instead of a OPA627 around parallel NJM4556As. Hard to justify the lower performance output buffer chips in parallel since it is exactly the same cost now to just use a LME49600, when the cost of the paralleling resistors and IC sockets are figured in. I'm learning that lots of OPA627 + BUF634 amps are out there already (thanks qusp!) so maybe this will kick things up a notch with the OPA827, although I'm sure lots of 827+LME49600s are out there too. It has a published distortion vs.output voltage graph in the datasheet that essentially slightly beats the same LME49600 graph at (heavy) 32R load. Having a slower (BW 22mHz) chip looped around a faster (110mHz) buffer now puts the cart back in front of the horse, from a stability standpoint, vs. the faster OPA627 around the slower (BW = 8mHz) NJM4556A chips.

* B5-080 case that is 1 inch wider than the O2 amp B2-080 case and 1/4 inch taller. The extra width allows more jacks and controls, plus the power supply stuff to be heat sinked in the back.

* Power supply included on the one single 80mm x 125mm PCB, rather than two 80x100 boards in the B3-160 case with amp in front and power supply in back. Still LM317/337 prereg feeding LT1963A and LT3015 low noise final voltage regulators. Power input connector and power switch on the back.

* LME49600s mounted on the bottom of the board with a large heatsink foil area tied to Vee (their tabs), along with the LT regulators as dpak also mounted under there. Since the LM pre-regulators burn up most of the incoming power, and they are heat sinked to the case, the LT regulators dissipate fairly little with only 1.25Vdc across them. The B5-080 case has 5.8mm free under the board while the LT and LME dpacks are 4.8mm. The PCB board can go up a notch in the case too if needed.

* RCA input connectors on the back of the case, along with 3.5mm input on the front, and 3.5mm and 1/4" output on the front. 4 position gain switch remains.

* Pads for attenuation resistor in series with the pot feeds.

More surface mount to make it all fit. The OPA827, LME49600, and LT regulators are all surface mount anyway.
 

Attachments

  • ODA circuit.png
    ODA circuit.png
    52.8 KB · Views: 595
Last edited:
* OPA827 now wrapped around a LME49600 on each channel instead of a OPA627 around parallel NJM4556As. Hard to justify the lower performance output buffer chips in parallel since it is exactly the same cost now to just use a LME49600, when the cost of the paralleling resistors and IC sockets are figured in. I'm learning that lots of OPA627 + BUF634 amps are out there already (thanks qusp!) so maybe this will kick things up a notch with the OPA827, although I'm sure lots of 827+LME49600s are out there too. It has a published distortion vs.output voltage graph in the datasheet that essentially slightly beats the same LME49600 graph at (heavy) 32R load.

I note you're following the examples in the LME49600 DS and having the opamp and buffer running from the same power supplies. I suggest not doing this as the heavy loads that get connected to the buffer will modulate the opamp's supply and the OPA827 has very poor PSRR on its negative rail (while its +ve PSRR is fairly good). Add extra filtering or a separate shunt regulated supply for the OPA827 to reduce the HF hash introduced on to the supplies from the buffer.