Solving USB audio interface noise

[ajay_m]

I know most of us have suffered from noise (typically an annoying buzz or whine) from USB audio interfaces and ground loops.
Today I tried a low-cost solution which worked extremely well. It's called an AV-Link 3.5mm jack ground isolator and is a stereo isolating transformer with a 3.5mm stereo jack on one side and a corresponding socket on the other.

It cost a massive £7.95 so I was understandably rather dubious. I connected up my Siglent scope and function generator and found that the frequency response is pretty flat from 20Hz through to 20KHz with the natural resonance of the transformers around 140KHz. Fed with a square wave there was obvious ringing on the output which was removed with a 1.2K resistor across the output, at which point although clearly the output was slew limited, and there was obviously a bass falloff towards 20Hz, nonetheless the performance was fairly decent.
With a load resistor the attenuation was around 0.75V out for 1V in, which is fairly acceptable. Crosstalk measured around 5mV for a 1V signal at 1KHz and did not seem to degrade significantly within the audio band.

When connected, the noise from the test audio interface (a Roland DuoCapture) was utterly eliminated. This seems like a pretty cost-effective fix and doesn't seem to cause significant signal degradation, but I'm sure below 35Hz or so the THD probably starts to climb; however, compared to noise prior to trying it, it's a huge improvement.

[Martin Walker]

Result!

And not a bad one either - folk would normally recommend a slightly more upmarket isolating transformer to avoid limiting the frequency response and adding some low end distortion, but as you've found sometimes the cheap end of the market works well enough (I bought a similar transformer just for its possible 'mojo' rather than breaking a ground loop, and that one worked better than I was expecting

Martin

[ef37a]

Result indeed. This, https://cpc.farnell.com/av-link/201-086 ... aoEALw_wcB
is probably the same transformers in a different guise and I 'specced' one out at work some years ago. So long as the signal is kept under about 2V rms (~+8dBu) the LF waveform was clean on a scope. That Roland AI seems to deliver about that but beefier sources might show some LF distortion. I actually re- boxed the transformers in a posher tin and send them to a guy on a forum who was having ground loop problems. He reported a total cure and no audio side effects.

The HF peak would be best curtailed with a series CR "Zobel" network but that would take some experimentation to get best results.

The transformers are so cheap that they should be in the 'gadget bag' of anyone dabbling with audio or/and live music.

Note however that a 'USB whine' is inherent in some first generation 16bit converters used mainly in budget USB mixers. This is not a ground loop problem. The problem can be largely solved by reducing the record 'gain' in Windows Sound Control. Normally set at 100%, drop it to 5% or even lower.

Dave.

[ajay_m]

I have now tested a second solution which also works extremely well. This is a USB isolator, using an ADUM3160 chip, which provides full USB isolation, albeit at only USB1 speeds (12Mbps). This is however more than adequate for a two channel USB audio interface - by my calculation 12Mbps is 1.5MBps and 44100 X 24 bit X 2 channels is approx 265KBps (or around 2.1Mbps).

In testing, this also completely eliminated noise from the audio interface and I could not measure any degradation in latency and overall performance.

Obviously the tradeoff is you will have some channel limitations, you might well get four channels over that and if you want higher sample rates that might constrain you back to a stereo pair.

Another consideration would be that nominal maximum output current is 400mA; at this load the measured output voltage was 4.9V which is well within spec.
(courtesy of my "X-Dragon" USB current and voltage meter, a handy cheap little device that I highly recommend).
However if your audio interface draws close to the USB 500mA current limit it might have a problem here, or if its inrush current was significant. The DuoCapture draws under 100mA according to Roland's spec sheet. Remember the isolator chip has to create a fresh, isolated 5V supply on the output side. Pretty clever stuff!.

That said, this is still a reasonably inexpensive solution at £17 and works entirely in the digital domain, so there is no risk of compromising the audio.

The devices are widely available; mine came from Amazon and is packaged in a transparent blue case (which made it nice and easy to see what the chip was).

There are also devices based on the ADUM4160. The only difference is this has a 5kV isolation rating instead of 'only' 2.5kV. Frankly in this application I rather doubt this is a parameter of much relevance!. Still it's reassuring to know you could have a piece of gear on one side at mains potential and the isolator would still protect the other side.

I'm not aware of any devices that can isolate at USB2 speeds. The required bandwidth would be way beyond what any optical isolator is currently capable of.

[ajay_m]

Just to round out this rather obsessive focus. I had already ordered a 'Feintech' isolating transformer from Amazon when I picked up the AV-Link unit from City Radio in Ealing where I work.
So tonight I had a chance to do some quick tests on this unit. It costs under a tenner and is much smaller than the AV-Link unit. Also, it has 3.5mm sockets and no captive leads, but does come with a male to male cable which is convenient.
Natural transformer resonance is much higher at over 200KHz and again the audio performance is pretty flat from 20Hz up to way beyond 40KHz, where some peaking starts to occur.
Below 30Hz there is some evidence of non-linearity though it is, in fact, better than the AV-Link unit. Crosstalk was also better, I could not measure any significant crosstalk at 1V in across the full frequency range.
Square wave slew rate was exceptional. The output slew rate was better than 1V/us and although, again, some ringing was evident, this was damped with a 1.2K load resistor. Since the ringing is essentially at the transformer resonant frequency it's way beyond the audio bandwidth anyway.
Again, as expected, this unit completely solved the noise problem. It's arguably better quality than the AV-Link unit, and certainly apart from some non-linearity below 30Hz, I can't see any reason it would introduce colouration into the signal.

[Martin Walker]

Just to round out this rather obsessive focus. I had already ordered a 'Feintech' isolating transformer from Amazon when I picked up the AV-Link unit from City Radio in Ealing where I work.
So tonight I had a chance to do some quick tests on this unit. It costs under a tenner and is much smaller than the AV-Link unit. Also, it has 3.5mm sockets and no captive leads, but does come with a male to male cable which is convenient.

Thanks for all the extra info ajay_m!

The tiny isolation transformer I bought cost me the princely sum of £4.21 for a stereo unit, and was like yours fitted with 3.5mm sockets and came with a short male to male lead.



As I mentioned, I was more interested in using this for some 'mojo' rather than for isolation purposes, and for this purpose it performed quite well - I don't have a scope, so tested it at 44.1kHz sample rate via PluginDoctor. It's 3dB down at about 10Hz (upper limit curtailed of course by Nyquist in this screenshot):

On the mojo front, with a nominal input level of -18dB its measured THD was -35.5dB overall, which equates to around 1.8%, so it proved useful for my purposes.

Very handy!

Martin

[ajay_m]

I'd bet that's the exact same unit, someone in China makes them and then they appear under dozens of random 'alphabet soup' brands on Amazon and elsewhere.

I'd be interested how well those very expensive boutique transformers perform relative to these units (i.e the products used by Neve, SSL etc). Are they significantly better, especially below 50Hz, or, in fact, given the limitations of copper wire and magnetic cores, are these cheap little gadgets actually very close in terms of performance?. I was very surprised by the excellent square wave performance, I would not have expected a transformer to have been able to do this.

In the studio I've gone with the USB isolator solution as it worked incredibly well, I had never heard of such devices until the other day; the idea of optically isolating the USB signals *and* regenerating an isolated 5V power supply on the other side is pretty amazing technology, and for my application, two channels is perfect.

The reason for the setup is that I have moved my KMI K-Mix across to my Akai Force and this is now my primary recording setup, it's so much more reliable and quick to interact with than the PC. Then I just want to bring the DAW outs across to two channels on the K-Mix. I can then record those on the Force. So I dug out the old DuoCapture and... that's where the problems started.

The K-Mix (unlike the Roland unit) of course is fully balanced, and because it supports external power (you can connect 5V to the MIDI expander port to power it), I never had problems with noise with it.

I also added a two way USB switch so that I can switch the K-Mix back to the DAW if I want. Everything analogue then stays connected as it was, and I use two mixer channels to route the Force back in so I can work 'in reverse' as it were. I then use one of the mixer aux busses to send outs back to the Force and I might now use a second bus pair to do the same back to the inputs on the DAW side. I just have to remember that Windows does not take kindly to having USB devices disconnected while they are open in the DAW; it'll often bluescreen if you switch away to the other channel.

The Force, on the other hand, behaves impeccably. It displays a dialogue indicating it's switched back to its internal audio but otherwise carries on fine.

[Hugh Robjohns]

I'd be interested how well those very expensive boutique transformers perform relative to these units (i.e the products used by Neve, SSL etc). Are they significantly better, especially below 50Hz

Yes, they can be significantly better, in terms of greatly reduced LF distortion when handling high level signals. Whether that is actually required or an advantage depends on the situation, obviously. Transformers capable of handling high levels of low frequencies tend to be both physically large and expensive.

One of the common issues with low cost transformers in a small box, such as the first model you tried, is magnetically-induced coupling between the two channel transformers resulting in crosstalk. Your measurement (5mV on the non-driven channel for a 1V signal on the driver channel) equates to around -46dB. While that is not particularly significant in the context of a stereo track, it could be a real problem if trying to record or playback to completely independent signals.

Clearly the different construction of the second device improved the crosstalk substantially.

Personally, I've settled on using ART DTI dual isolator units when required for general ground isolation duties, mainly because they are very versatile in their connectivity options -- balanced and unbalanced formats on a variety of connectors, and they use pretty decent transformers for the money so perform very well at normal levels.

For very high quality or challenging jobs I use an equivalent box made by Canford that uses Lundahl transformers which are truely superb... but very expensive!

[ajay_m]

That's very interesting. I have a background in electronics engineering but like most people, we treat inductors as 'black magic' because unlike other components they are very complex to model and every vendor's products differ. (hence, we would try and avoid them in most analogue circuits if possible, modelling them with gyrators where we need to simulate an inductor).

Capacitors can be manufactured these days to come much closer to an 'ideal component' (some of us remember paper capacitors, about which the least said the better!), resistors are available with extremely low noise and high tolerance (0.01% if you can afford it, I have a few as calibration references), along with good temperature stability. And active components are infinitely better than those available when I was young, which really opens up all sorts of design possibilities that weren't practical decades ago.

When it comes to transformers, there's no doubt that both of these low-cost devices struggle to handle significant amounts of signal below 50Hz or thereabouts, presumably due to core saturation or just material non-linearity, and the higher end transformers undoubtedly use more expensive, higher permeability materials, (there's not much you can do to improve copper wire, so all the magic has to go into the core material).

I did look at the ART units when thinking about solutions but at that point the cost seemed to be approaching the pain of just buying another 2 channel unit with balanced outs and hoping that would work, (after all, you can always ground just one end of the cable and that often does the trick), along with the ever-elusive dream that, this time, with a new interface, latency would be a little better. It seldom is, though, for some reason to do with the USB protocol, I suspect the setup time for transmitting a lot of small buffers is quite high compared to Firewire or Thunderbolt.

In the meantime - isn't this a great topic for an SoS article?. It comes up time and again and ground loops really baffle people. The best way I've been able to explain it is to imagine that your input signal is set to zero so there's no voltage between the input and ground. Now suppose you could 'wiggle' the ground potential up and down a little but leave the signal potential at zero.

Your audio interface still sees this as a signal because all it 'knows' is the difference between its 'ground' and the signal input. Now the ground wire is a very low impedance input - but it's still an input, so when you put 500mA of switched current across a USB connector which may have a ground resistance of 0.01 ohm that's 5mV which is, as you said, about -46dB relative to 1V and almost certainly audible.

With the input disconnected, though, the audio interface remains blissfully unaware that its 'ground' is wiggling up and down slightly as there's no signal input and hence no difference to amplify. Plug in a non-grounded signal source like a microphone, and with luck you still won't have any trouble.

But then you connect something that's grounded independently somewhere else. Probably via standard coaxial shielded cable.

The signal cable in the coax carries virtually no current. So the signal input on the interface is going to track the ground potential of the remote signal source pretty much perfectly.

But the shield of the signal cable, while quite thick, is compromised by connecting via typically two 6.5mm sockets, which were never designed to carry any kind of significant currents. So it's very likely that the ground resistance of this signal path is a good deal higher than the 0.01 ohm represented by the USB cable.

Consequently the ground on the audio interface is now 'wiggled' by the very low impedance voltage source represented by the USB cable and the signal input stays stuck at ground. Presto, noise!. You often notice, too, that rotating the 6.5mm jack plugs introduces additional crackles. This is because that ground current is trying to travel via the signal cable and those rather flakey sockets.

You can, in fact, often reduce this noise by providing a parallel path for the ground, but most interfaces don't have a point you can easily access to do this. On my old UA25, which had a metal case, I ran a very thick piece of speaker wire back to a ground point on the computer and this substantially improved the induced noise, thus demonstrating the cause.

You would think that balanced I/O would help but it sometimes does not. I surmise this is because the interference is actually at much higher frequencies than the audio range and is internally rectified and demodulated by the input circuitry due to non-linear effects (thus causing the audible 'whine'). And of course once it gets near the analogue input circuitry it can crawl all over the place via radiated energy as it's mostly RF.

So then despite theoretically the common mode rejection of the balanced input filtering out noise, this noise is so far outside the input circuitry's design range that any small differences in inductance and capacitance between the two balanced inputs, plus innate limitations in CMRR at high frequencies with opamps and discrete differential inputs, mean that the noise isn't mitigated as much as you'd hope.

Anyway, that's my explanation of something that seems to baffle people, as I said, perhaps time for (another) article on the topic, especially now we have so many great cures (like those USB isolators - that's real cleverness!)

[Hugh Robjohns]

In the meantime - isn't this a great topic for an SOS article?. It comes up time and again and ground loops really baffle people.

It certainly comes up on the forums regularly, and it does baffle people... but often the problem they assume is a 'ground-loop' isn't actually a loop at all! Increasingly -- due to the use of laptops, bus-powered interfaces, and class-2 mains-powered equipment -- the problem is actually ineffective screening due to the absence of solid grounding! In other words, a ground needs to be added, rather than removed!

Your audio interface still sees this as a signal because all it 'knows' is the difference between its 'ground' and the signal input.

Yes... but it shouldn't be if the noise current conveyed by the cable screen wasn't allowed to enter the audio circuitry ground reference. That's where most problems occur -- poor design and construction of the unit's internal grounding arrangements.

You would think that balanced I/O would help but it sometimes does not.

It usually does, and for two reasons. Firstly, the cable shield connecting the source and destination device grounds does not play a secondary role to convey the signal reference. The receiving amplifier is only looking for the voltage difference between the hot and cold wires, and doesn't care what's going on with the grounded screen...

That said, the amplifier itself will have it's own local ground reference so noise currents still have to kept away from that, which comes to the second reason: the majority of modern devices with balanced interfacing is designed up to a standard rather than down to a price, and the designers are generally aware of good grounding practices. This certainly isn't always the case, of course, and frequently wasn't the case with legacy equipment... but I very rarely experience ground-loop problems with the balanced equipment passing across my test bench.

I surmise this is because the interference is actually at much higher frequencies than the audio range and is internally rectified and demodulated by the input circuitry due to non-linear effects...

It's certainly a possibility, but that would imply inadequate bandwidth filtering at the input and/or ineffective shielding. So a design or construction failure rather than something implicitly wrong with the balanced interface concept.

[ef37a]

The reason why balanced operation does not always cure ground loop problems is
Because electronically balanced inputs have a much poorer CMRR than a good 10k 10k bridging transformer*. The common mode impedance of an op amp input might be a few tens of kOhms whereas said transformer could be over 50 meg!

There is a paper on this by the Transformer Man Bill Whitlock (I'm sure Hugh has it filed?) and he rather generously developed an op amp circuit that mimics this huge CM impedance. They are marketed by the 'THAT' corporation.

Just a bit of a gripe Ajay. USB poor latency? You have not been keeping up with current events friend! RME have made very low latency USB AIs for years and my NI KA6 is no slouch. Later generation gear such as the MOTU M2&4 are even better I think and I am deffo going to get one of those Minifuse M2s very soon (in the latest mag).

But ground loops were always with us. We called them "hum loops" and I am somewhat ashamed to say we did some very naughty things to fix them!
But, manufacturers did not help. Very common were gram decks where the pickup wiring went to a 3 pole tag strip, centre bolted to ttable metalwork. So, you either dissed the earth and took a chance or kept it safe and lived with hum!

Yes, VERY naughty but you have to remember that back then AC/DC radios and TVs were common. No earth and the chassis connected to one side of the mains. Since many homes had just 2 pin 2 and 5 amp outlets. Russian roulette!

*Yes, bit of a black art. The best transformers use sectioned 'pie wound' coils to improve coupling, reduce leakage inductance and reduce inter-winding capacitance. Naturally such involved construction costs. It also means less copper relative to spacing materials so for a given performance the transformer will be bigger. If testing LF performance remember that any significant source resistance will increase distortion. Some of the very cheap 1:1 isolating transformers are bifiliar wound and so the breakdown voltage twixt windings will not be in the kV region so do not rely on them for mains fault isolation.
Dave.

[Hugh Robjohns]

The reason why balanced operation does not always cure ground loop problems is because electronically balanced inputs have a much poorer CMRR than a good 10k 10k bridging transformer*.

Impedance imbalances in electronically balanced interfaces can certainly couple ground noise into the wanted audio, but I don't think that's as much of a problem as it once was --largely because of the widespread use of the excellent THAT interface chips which you mentioned, at least in the more professional products.

There is a paper on this by the Transformer Man Bill Whitlock (I'm sure Hugh has it filed?)

This one?

http://www.thatcorp.com/datashts/AES626 ... put_IC.pdf

[ef37a]

That's the paper Hugh! Thanks, I had it in the AES journal but it has got lost.

I should have said as well that the high end audio gear will have better front ends than the budget kit.

Dave.

[bkrivers]

After years of living with the “usb whine” problem, I finally solved it for $9.99 with clip-on ferrite coils! I bought a sampler to figure out what size it needed (mainly 5mm…). I put them on each end of an usb cords (I have a 6ft extension and a hub on my pedal board, I am a guitar player…) and the system is now dead quiet!

Here is the description: Ceedmon 20 Pieces Clip-on Ferrite Ring Core RFI EMI Noise Suppressor Cable Clip for 3mm/5mm/7mm/9mm/13mm Diameter Cable, Black

Here is a link: https://www.amazon.com/dp/B07CWCSNW9

The reason I tried this is I remembered one of my interfaces from long ago shipped with a usb cable that had these built in. I researched that type of usb cable and saw there was an after market clip-on ferrite ring which meant I wasn’t limited to cables with built-in ferrite shields - I could use my current high quality usb cables. Sometime taking a chance pays off!

I hope this helps someone…

[ajay_m]

I hadn't tried that option which, given I have a bunch of clipon ferrite suppressors, was stupid of me. (They were purchased to sort out an interference issue on a cable run on one of my 3d printers. And although it's on a different thread about buying a second hand reel to reel tape deck, trust me, 3d printing is an equally brilliant way of spending all your time tinkering).
Since the original post, I note that usb2 isolators are getting cheaper and the Topping unit looks promising. But the Roland duocapture has been replaced with a ka6 MK2 after an unbeatable deal from PMT during a sale and noise isn't an issue with that unit. I must try your trick with the Roland just to see if it fixes the issue. If so this tends to favour my theory that common mode RF interference is a possible cause

[Martin Walker]

After years of living with the “usb whine” problem, I finally solved it for $9.99 with clip-on ferrite coils! I bought a sampler to figure out what size it needed (mainly 5mm…). I put them on each end of an usb cords (I have a 6ft extension and a hub on my pedal board, I am a guitar player…) and the system is now dead quiet!

Here is the description: Ceedmon 20 Pieces Clip-on Ferrite Ring Core RFI EMI Noise Suppressor Cable Clip for 3mm/5mm/7mm/9mm/13mm Diameter Cable, Black

Here is a link: https://www.amazon.com/dp/B07CWCSNW9

About a year ago I bought an almost identical pack of assorted clip-on ferrite rings, but installed them one by one on selected mains leads, SMPS outputs & digital cables, to reduce RFI, listening out for any audio changes/improvements at each stage.

Apart from no more occasional click/crackling interference from kitchen refrigerator/freezers/microwave oven being audible on my studio audio, I also noticed significantly more tonal, room acoustic and reverb differences between various studio audio demo tracks, and any harshness in mixes showed up more clearly.

[sonics]

Apart from no more occasional click/crackling interference from kitchen refrigerator/freezers/microwave oven being audible on my studio audio, I also noticed significantly more tonal, room acoustic and reverb differences between various studio audio demo tracks, and any harshness in mixes showed up more clearly.

Do you recall which equipment it was most effective on?

[tea for two]

After years of living with the “usb whine” problem, I finally solved it for $9.99 with clip-on ferrite coils! I bought a sampler to figure out what size it needed (mainly 5mm…). I put them on each end of an usb cords (I have a 6ft extension and a hub on my pedal board, I am a guitar player…) and the system is now dead quiet!

Here is the description: Ceedmon 20 Pieces Clip-on Ferrite Ring Core RFI EMI Noise Suppressor Cable Clip for 3mm/5mm/7mm/9mm/13mm Diameter Cable, Black

Here is a link: https://www.amazon.com/dp/B07CWCSNW9

I hope this helps someone…

Appreciated BK. I shall get these.

I wonder if I place them on my ears whether they will work lol.

[Martin Walker]

Apart from no more occasional click/crackling interference from kitchen refrigerator/freezers/microwave oven being audible on my studio audio, I also noticed significantly more tonal, room acoustic and reverb differences between various studio audio demo tracks, and any harshness in mixes showed up more clearly.

Do you recall which equipment it was most effective on?

I didn't test after each ferrite was fitted, as I expected the differences to be tiny but accumulative, but I did spend some time deciding on the best positions before I fitted them all.

1. I placed four of the largest ferrites round the mains cables immediately before various distribution boards, to hopefully remove any interference already picked up on mains or from ethernet cables alongside.

2. I placed one large one on PC mains cable Immediately beyond PC (to prevent any RFI from PC getting into dis boards).

3. I placed four of the smaller ones on each of four Ethernet cables just before entry into router.

4. The remainder were placed on various external switched mode power supplies, as close as possible to their outputs, to once again hopefully prevent any RFI getting any further - I did this on keyboards/synths, an audio interface, and a few FX units.

I also have very detailed ATC loudspeakers and an excellent Lavry DA10 converter - you may not be able to discern audio improvements I could hear on a more modest system.

Hope this helps!

Martin

[sonics]

Thanks Martin.

I've being buying and using these things for over four decades; just wanted some input from a fellow "chokeaholic"!