Debugging noisy headphone amp

[il_mix]

I bought a Rolls PM351 (personal monitor/mixer, headphone amp) several weeks ago. Turned out that, connecting actual in-ear phones (didn't have any when the device arrived) instead of normal headphones, the device is quite noisy. More precisely:
- Beyerdynamics DT770 (80 Ohm); clean sound
- Beyerdynamics Byron (23 Ohm); medium noise
- Shure SE425 (22Ohm); medium noise
- KZ AS10 (14 Ohm); quite loud noise
- KZ CA CCA10 (32 Ohm); quite loud noise
KZ are the most noisy, and the noise amount seems almost independent to the headphone impedance. One can complain about china products, here. But I can hear the same noise from Shure, too; not that loud, but still annoying. Also, on another (non technical) forum one said an interesting thing: "Passive headphones cannot create hiss". Therefore, the noise must be generated by the device. Debatable, but also quite resonable.
The noise can be heard with all input volumes to 0, and increasing the volumes doesn't increase the noise. Is a "noise floor". Have to notice that not all volumes when set to 0 "moves" the signal to ground; some move to VBias, that can be noisy.
Also, to reject any problem regarding the DC power, I've tested with an Agilent lab power supply; exact same noise.
Given that, I suppose the culprit is the output amp (or the companion electronics)
Here is the schematic



Any first impression?

I've added some green point/letters on the schematic. My idea is to take one pin of the highlighted components (tipically resistor) and move it to ground, and see (hear) what changes on the output. I'll work on a signle output channels, so I can use the second one as "reference noise". I will start from point A through E. I suppose something will happen when playing with points C-E (points A, B will practically connect the output to ground, so I doubt I can't hear anything; or, at least, I have to amplify the output to hear something. Uhm... can be an actual test to do).

Do you think this can be a good workflow to detect where the problem lies?


Thanks in advance for your hints!

[garcho]

are the op amp ps bypass caps ceramic?

what op amps are you using?

did you buy expensive transformers?

[anotherjim]

The 4556 driving the phones should have power decoupling caps right by the power pins. The scheme shows 1000uF in one cap on the supply, but it would be much better for that amount of uF to be split up and distributed around the power traces in 100uf chunks. Additionally 100nF ceramics at least on the DC power input and the 4556 power pins.

As phones resistance decreases, the phones amp may be unstable with the inductive load. I'd have though Zobel networks should be fitted. I'm not 100% certain, but I think 4R7 with 220nF caps are usually used.

Nothing to do with your problem, but I don't like that "balanced" XLR input much.

[PRR]

> in-ear phones... instead of normal headphones, the device is quite noisy.

Some of those in-ear phones are VERY sensitive. More sound per Volt. This is because they descend from hearing-aid phones which have tight limits on power and battery life.

Yes, some ROLLS stuff is marginal for hiss, but this one is not bad.

_I_ think the "fix" is to put a large resistor in series with this type earphones. Cut the electrical power way down.

If you would *only* use this box with these phones, we "could" redesign the box for a weaker output, better efficiency than the series-resistor technique. But that could be a major project and then it would not do as well for your other phones.

Get a male-female extender cable. Cut it. Try 100 Ohms in series with each hot conductor. Idle hiss will be very significantly lower. Gain knob will have to be advanced to get the same listening level, but I suspect it has reserve gain, especially with these hot phones. Bread-tie this resistor cable to the in-ear phones, you use it any time a standard phone jack is too hot and hissy for these phones.

A different approach would be to tack 33k across R9 and R13, 100K, to make it more like 24k, which makes the output amps unity-gain instead of gain-of-4.5. However that may lame the performance for standard phones.

[il_mix]

are the op amp ps bypass caps ceramic?

Uhm... looks like there are no actual bypass caps between VCC and the op amp (and GND and opamp, if needed)

what op amps are you using?
did you buy expensive transformers?

I'm using the components listed in the schematic. I didn't "replicate" the device; I've bought the original one. And it kinda sucks in terms of noise...

The 4556 driving the phones should have power decoupling caps right by the power pins. The scheme shows 1000uF in one cap on the supply, but it would be much better for that amount of uF to be split up and distributed around the power traces in 100uf chunks. Additionally 100nF ceramics at least on the DC power input and the 4556 power pins.

So I need a decouplig caps between VCC and the amp V+. Better insert it on all the op amps? (input ones, too) What about V- and GND?
Also, can you better explain the concept "split up and distributed around the power traces"?

As phones resistance decreases, the phones amp may be unstable with the inductive load. I'd have though Zobel networks should be fitted. I'm not 100% certain, but I think 4R7 with 220nF caps are usually used.

Will have to read something about this Zobel networks...

Nothing to do with your problem, but I don't like that "balanced" XLR input much.

Nothing to do with my problem, but please elaborate! I'm interested.

> in-ear phones... instead of normal headphones, the device is quite noisy.

Some of those in-ear phones are VERY sensitive. More sound per Volt. This is because they descend from hearing-aid phones which have tight limits on power and battery life.

Yes, some ROLLS stuff is marginal for hiss, but this one is not bad.

_I_ think the "fix" is to put a large resistor in series with this type earphones. Cut the electrical power way down.

Quite funny... I thought I needed to put a SMALL resistor in PARALLEL with the output resistors, to lower the output impedance (not that the two output series resistors are actually defining the output impedance, though...) and have the 1/8 source/load impedance "rule".

If you would *only* use this box with these phones, we "could" redesign the box for a weaker output, better efficiency than the series-resistor technique. But that could be a major project and then it would not do as well for your other phones.

Well, yeah, I think I will always use it with in-ear phones for live monitoring.

Get a male-female extender cable. Cut it. Try 100 Ohms in series with each hot conductor. Idle hiss will be very significantly lower. Gain knob will have to be advanced to get the same listening level, but I suspect it has reserve gain, especially with these hot phones. Bread-tie this resistor cable to the in-ear phones, you use it any time a standard phone jack is too hot and hissy for these phones.

When you talked about a LARGE series resistor I thought you were talking about kOhm range...

A different approach would be to tack 33k across R9 and R13, 100K, to make it more like 24k, which makes the output amps unity-gain instead of gain-of-4.5. However that may lame the performance for standard phones.

I can safely remove the original 100K and put 24K ones instead of adding a parallel resistor.
Is this the metioned "redesign", so a "permanent" alternative for the output series large capacitor?

Well, guys. Many thanks for the advices!
I think that I will do these 2 things first:
- add decoupling caps: first, I need a couple more advices about this (and regarding adding one to other amps). Then hear if something changes
- exchange feedback resistors with lower values one. Again, hear the difference. I can do this last tweak on a single channel, to check the actual difference with the original one

-- EDIT --
Uhm... I'm afraid that I'll have to cut some traces to actually add decoupling capacitors. Better to test PRR proposed tweak first.

[il_mix]

Or maybe it's enough to put a 100nF capacitor between V+ and V- pins of the amp (or, again, the ampS) since there is no negative voltage?

[garcho]

^ it is indeed, you can just "straddle" the IC with the cap leads from pin 4 to 8

[bool]

The PRR's suggestion of using a series resistor (he said 100R) is the way to deal with the background noise levels when using in-ear and low-impedance headphones. You don't need it with 80, 64 or 50-ohm cans.

The simplest way would be to just make a short jack-jack "extention cable" and solder these resistors to the socket. My suggestion is to use a little lower resistance; somewhere at around 68-82R; perhaps you will be fine with just 47R? You decide. 0'6W metal resistors will be fine.

[il_mix]

Some preliminary feedbacks.

I've installed the decoupling capacitor on the output amp. Nothing changed. Actually, I didn't expect much from this, since powering with an ultra-stable power supply didn't change the noise at all. Anyway, it can't be harmful, so I think I'll leave it there (and maybe add one to the input op-amps, too).

I've installed the parallel 33k resistor on R9 (to test a single channel for comparison). The background noise become way lower. Awfully, I don't have something to input to the device at the moment, so I don't know what will happen when I turn on the volumes to actually hear the various sound sources.

Noticed that a second advice to add a series resistor appeared (thanks bool). I will test this hack, too (with and without altering the amp feedback resistor).
Quick question about this.
At the moment, the amp output goes to a 10 Ohm resistor and then a 47uF capacitor. From your suggestion, I have to connect another (47, 68, 100, whatever) resistor on the output; so, connect the resistor after the capacitor. Will I obtain the same effect by replacing the 10 Ohm resistor (so connecting the resistor BEFORE the capacitor)?
Now for a technical/practical question. It looks like that, by adding this series resistor, what I am doing is simply adding, say, 68 Ohm to the in-ear impedance. So I'm simply "converting" my 14 Ohm impedance earplugs to a 82 Ohm impedance earplugs. Is it what this hack does?

[anotherjim]

Yes, replace the 10R already there with the larger value.

Depends on the pcb traces & distances whether just one bypass cap on the power input is sufficient. The bypass works best close to the loads, mostly because it prevents each circuits AC currents from inter-mixing in the power traces. They would not need 1000uF each either, a few 100uF close to each amp would probably do it.

In the XLR inputs, those series 47uF caps have no polarizing reference and the neg plates in between are floating. 22uF non polarized would be better, with some resistance (10k) to ground on the input side to stop it floating. Otherwise, you can get loud pops on the outputs if you plug something in - and that isn't nice in headphones!

[PRR]

> So I'm simply "converting" my 14 Ohm impedance earplugs to a 82 Ohm impedance earplugs.

AND 1/6th the sensitivity! Say 1 Volt test tone. 0.82V appears as silent heat. 0.17V goes toward transducer sound.

It is "wasteful". The waste is really small.

[il_mix]

Late night update.

I've compared left (normal) and right (reduced feedback resistor) channels with an actual line input. I didn't notice that much difference in volume between the two channels. So, given the noticeable background noise reduction, I've permanently changed these resistors with 22k ones.
Now the device is becoming usable!

I've also added the decoupling capacitors on the various op-amps V+/V-, just in case.

I've made a quick test with the output series resistor, before changing the feedback resistor on the left channel. I didn't notice any significant benefit in terms of background noise. But, as said, it was a quick test. I will perform some more accurate test with various resistor values using alligator clips (again, PRR advice from other thread).

[il_mix]

In the XLR inputs, those series 47uF caps have no polarizing reference and the neg plates in between are floating. 22uF non polarized would be better, with some resistance (10k) to ground on the input side to stop it floating. Otherwise, you can get loud pops on the outputs if you plug something in - and that isn't nice in headphones!

The douple caps are indeed weird... I didn't notice them. I was focused on the output stage. Makes sense to exchange with a single 22uF one (two 47uF in series == one 22uF). Or does someone have any explaination for such caps configuration?
Regarding the resistor to ground, you are suggesting something like this?



The pop is probably not a big deal. When I have nothing connected I'll have the MIC volume to 0, so I think I won't "feel" the pop. Have to say that I've just discovered yesterday that when the line in volume is set to 0, I can still hear a faint signal (I've connected an MP3 player). Maybe this will happen with the MIC in, too. But I suppose the pop will be really low. ...as soon as I won't forgot the MIC volume to any level >0 ...
Will this resistor to ground affect the input sound in any way? I smell a little filtering...

[bluebunny]

The douple caps are indeed weird... I didn't notice them. I was focused on the output stage. Makes sense to exchange with a single 22uF one (two 47uF in series == one 22uF). Or does someone have any explaination for such caps configuration?

Two 47uF in series is indeed (almost) the same as one 22uF.  But the two 47uF are back to back.  So they appear to be a non-polar 22uF (as you drew them).  If you have NP 22uF caps, then use them.

[anotherjim]

Yes, the input resistors go there.
22uF is still a generous size -  goes down to 7Hz. In this case, because it's a single supply circuit and Vref is positive from ground, the input caps can be ordinary polarised electro's with negative plate toward the input; though non-polarized are nice if we have them as there's always a chance of a DC leakage fault from anything plugged in such as a self-powered condenser mic.
The non-inverting isn't really balanced with inverting one anyway. The inverting input is 1k impedance, the non-inverting is 1k+49.9k. The treatment of the non-inverting with an attenuating voltage divider is an attempt to solve the problem of the op-amp non-inverting gain always adding x1 signal {(Rf/Rin)+1}


In this partial scheme, the gain from XLR pin3 would be x2 while from pin2 it's only x1, so the attenuator halves the signal going in from pin3 which brings it out at x1 after amplification. The difference in input impedance is still there, 100k for pin2 and 200k for pin3. The scheme recognises this by halving the input cap to maintain the same low cut off frequency. However, it could have used 2x 47k for pin3 input which would have been near enough 100k input impedance and the input cap could be the same.

[PRR]

> The douple caps are indeed weird...

That's fine. Don't change it.

It "needs" to be non-polar in case Phantom voltage is applied. It needs to be dozens of uFd for decent performance with microphones. At a ROLLS price-point, multiple polar caps back to back is a fine way to do it. The "lack of bias" at the center point is not any problem.

[il_mix]

I logged in just to ask "what about caps polarity and phantom power?".
I got the answer before asking the question... magic!

[il_mix]

I finally had time to test the output series resistor trick.

As suggested, I've created a custom extention (5cm) cable. Again, to appreciate the improvements, I've simply wired one channel, while the second one was floating, to be connected using different series resistors values. I had 4 values at hand (47, 56, 82, 100). Eventually I've just tested the 47 Ohm one, and didn't move on because I was already happy with it.

As said before, I've already applied the "smallest amp feedback resistor" trick on both  channels. Had a noticeable floor noise reduction without significantly losing output signal.
When I've connected my custom extention cable with the 47 Ohm series resistor on one channel, I've noticed a further background noise reduction (I would remind you that this background noise can be heard with all output volumes to zero), that now was barely noticeable. However, when I connected a line input the volume difference between the two channels became way more obvious. So I thought 47 Ohm is a good trade-off between pros and cons, and didn't move further.

Since the device will be used as wired in-ear monitor system, and I will use an extention cable anyway, this time I will modify the extention cable instead of hacking the device itself.

I think I will stop here. First of all, I have to actually test the device in real case scenario; line in from mixer as main monitor, mic line with instrument input, hear everything during reharsal and, more import, in an actual live. After that I'll decide if I have to dig further. Or maybe take some steps back, like removing (or reducing) the output series resistor because the sensitivity became too low, while discovering that the background noise is a minor issue in an already acoustically noisy environment (doubt, since I'll hopefully be quite isolated from the background noise with IEM. Well, it will never be a 100% isolation. Awfully, way less...)

Thank you all for your precious suggestion!
Hope this thread will be useful for some other users.