[Sticky] USB-C Microphone (official topic)

Hello @marcdraco,

Thank you very much. Your advice was helpful. I will look into things now and update you when I have some findings. Thanks again!

- Nick Law

OK folks, for everyone who really wanted to make the Ultimate USB Microphone but was put off by all the fiddly bits (including mine) here's an alternative solution that uses a clever circuit (not mine). Keeping the a random JFET in a very stable biasing state (better quality audio) is difficult and the common-source method does produce more distortion because JFETs aren't good linear amplifiers.

This is how most simple capsules work which is obviously less than ideal.

A common drain on the other hand actually reduces the microphone's output slightly (about x0.9 to x0.95) which is almost useless unless you put it through a 60dB amp meant for dynamic mics and that comes with its own issues. While I've managed with 15-48V supplies to get a decent level out, that still meant a lot of fiddling around with extra gain. Varee does this, but also required at least 12V to get out of bed.

So what if we only have a fiddling amount of voltage (5V or LESS) probably throttled by a 2k2 resistor built into the mic circuit. Now we're into something a bit more tricky.

With a 20dB boost stuck on the end to bring the volume up to a reasonable level. I'm still testing this one but initial results are very promising. I used a portable audio recorder and while I had to jack the gain up (which was expected) it does produce a very pleasing sound and (using a cheap 25mm electret) didn't actually need a Faraday cage. Not that I'd recommend using it like that of course. The circuit is relatively simple to build on Vero/stripboard but I've crammed it all on a 32mm capsule adaptor that also fits the 25mm devices and will fit inside a donor body such as a BM800. The idea is to put one of those really small digitisers (based on the CMI chipsets) inside a BM800 body and you'll get USB out direct. And with the quality of these chips improving all the time, you can get decent performance from ones costing under £10 - and that's Amazon prices.

Details to follow when I've had time to do more tests but can say that it does work - not well (it's not LOUD) even down at 3V.

Initial tests suggest that it will work better from the USB power (rather than the fiddling amount of current the classic "mic" power does, presumably because it's got a load resistor in place and isn't expecting anything more than a JFET.

I expect (again, I'll have to verify this) that the USB noise will appear on the audio which happened on the early  Michelle boards - which is why I need to verify if it needs power conditioning. I left if off the first designs for simplicity and because it also means losing another volt or so due to the capacitor "multiplier". That's a circuit that takes a noisy supply, smooths it with capacitor (actually, a low-pass filter) and supplies that smooth voltage at much higher current. The problem with normal capacitors is they can't supply enough current unless we use a really large one and ... even today, the largest practical capacitor for this sort of tiny PCB is about 47uF - and even that's pushing it.

Hello everyone,

I've had this project among several in my saved bookmarks for a while now and finally decided to actually do it instead of pushing it off. As I currently don't have any real microphone (if you don't count built-in ones like in my phone or bt-earbuds), I'm making this with the goal of building something fun and having it as a main driver for vocal recording for songs (currently doing only instrumentals because of this reason).

Anyways, my goal is to build a version using this (or another) capsule, with XLR output, and support for 48V phantom power.

Why? I already have an audio interface (M-Audio M-Track Solo) for my guitar with an empty XLR input just waiting to be used. It includes a phantom power switch to power the preamp.

After reading a bunch of replies here I can't seem to find, or am yet to find exactly what I'm looking for. I'm relatively new to this stuff so I would need to start reading up on how everything works to go at it from scratch. If anyone could point the finger as to how I would approach this and/or readup somewhere how I would tackle it that would be appreciated, however I'm sure with the amount of work that's been put into this project it is somewhere here.

(apologies if any of this has been asked before, I've been reading for a while and my brain is fried haha, figured it would be simpler to ask, thanks in advance)

-Stefan

Hi Stefan, the Varee design which came out of this project works at 48V and has some tricked out circuitry to keep the quality higher than a traditional JFET interface. This will drive a normal, phantom powered 48V microphone input and is completely stand alone. The bit at the top can be snapped off or used to hang the adaptor inside a donor body (the Neewer BM800).

https://github.com/marcdraco/Varee2/

If you're in the UK, drop me a PM as I probably have a spare floating around I could let you have (it's Christmas after all).

You will probably have to make five posts first to get out of the moderation queue so if you can do that (keep them sensible obviously) and it's yours.

@marcdraco thanks for the quick reply!

So after reading up a bit what I'm getting at is this: get a cheap mic body (BM800), swap out the capsule for the JLI-2555BXZ3-GP, solder on the transistor, connect this to the Varee board, and hook up the xlr output jack. This sounds pretty straight forward so if I'm missing something please correct me.

Unfortunately I just moved out of the UK in july hahah, currently living in Germany, I do still have friends and a cousin living there so maybe you could send it there and they could send it to me. Anyways thanks for the offer!

Yeah, I can do that. Be happy to gift a 48V Varee to you - I'll stick it in with a Christmas card. 😉

I don't think it's massively expensive to send stuff to the EU. Might just have to go in really small card.

Varee has everything you need - including the JFET. You solder the board directly to the mic (any electret 25mm or 32mm form factor should work), put ground on the outer rim and power in/signal out comes out of the two little ports at the top. It's not the easiest thing to solder to, but with care it's pretty straightforward.

Matt and I spent quite a bit of time with these transistors - with the LSK170 coming out on top but the 2SK208 (fitted to the Varee) and controlled with this tricky self-regulating current loop, it produces incredibly clear sound.

It should have a 1G resistor soldered on but I've found that a little smudge of a soft pencil across the pads leaves enough of a trace of graphite to do the job. (A jfet needs a little bit of "back biasing" to work and this is why we need some resistance from gate to ground - for that piffling current to flow from the drain to ground via the "diode" junction.

Varee is primarily intended for 48V although you can use it with the existing pre-amp by desoldering the regulator.

Thanks for the board! Will look at getting those posts up haha!

Excuse the terrible paint-job, just wanted to make sure I got this right, so soldering goes like this:

And then I unfortunately don't get why and what we need the grafite on.

The graphite (just rub an HB pencil over it) is on Rg - or you can be fancy and put a 1G resistor in there. 🙂

Strictly there isn't a "negative" on a P48 system, just a Hot and a Cold which is In and Out of phase respectively. Both signal lines are driven to around 45-50V relative to the ground/shield but that's all good - don't worry about the painting - you should see my soldering. LOL

Aha, got it, thank you, I'll see what I end up doing, one would think rubbing the graphite on there wouldn't permanent no? I would probably have to open it up sometime again in the future and reapply some, so maybe for long term use I'd just solder a resistor on there for good measure.

On another note I went with the route you recommended, which is modifying a BM800 body. I ordered one off of AliExpress and it should arrive in the coming days, I got a good deal where it comes with some cheap boom arm which I hope is usable haha. Gonna order the capsule soon-ish as well, as soon as I find reasonable Germany shipping as on micbooster the shipping costs 5 more quid than the capsule itself, worst case scenario gonna end up buying that one. I'll keep updating in the coming days.

If you don’t mind a slight loss in quality (I can’t hear it personally) you can just use a 25mm electret or even go daft with the 34mm monsters. Varee drives all of them but those are available at Ali and eBay. Just make sure it’s electret or post a link Ans I’ll verify for you. 

Hi! I tried following the project along but got horribly stuck. My pre-amp just does not want to work. With or without the Capsule connected, it makes this horrible distorted noise and nothing else. Left is without capsule, right is with capsule.

I made my own pcb while basically copying the original schematic, except adding screw connectors to wire everything together and adding a second filter for high frequencies before the amplification. I have resoldered the entire pcb twice and it did the same thing both times. The voltage crashes a liiitle bit and I'm getting 4.9V instead of 5V at the source and after the booster it's -+14.5V. As an aside, I also looked at the input signal of the capsule with an oscilloscope and the positive and negative halves were separated, so I think the transistor is working fine too but the moment I connected the shielding, I got no signal at all. 

@yamithewitch Painful to hear that and if it makes you feel any better, this one turned out to be the most difficult boards I've ever designed. For reasons that I can only guess at, the original Vero design seems to have better performance than the schematic copied to a real PCB..

There are multiple potential issues we face but I have to wonder if Matt tested his boards with different PCs. I usually work in the analogue OR digital domains, working with both on the same board poses a whole bunch of problems. "But," one might argue, "this is analogue one!"

Only it's not, at least not in the way we'd normally think of as pure analogue.There's a lot of digital noise coming up that USB rail and this seems to be at the root of these issues. The noise spectrum here (shown from your sample) which I've boosted for 40db (x100) for clarity demonstrates this rather bluntly.

The fundamentals is appear at 1KHz and 2KHz with regular harmonics floating off into the sunset - and that's the whine you can hear. There are peaks at every harmonic which means this is likely a couple square waves (which only contain odd harmonics). Here's a pure square wave at 1KHz so you can see the effect. The other stuff is the background noise we'd expect to hear.

I'll admit this one drove me around the bend, up the road and right into the proverbial creek no one wants to swim in lest they come out smelling like a dung pile. And then I tried the board, by complete fluke, in a different system and ... no noise. Going back to the original (Lenovo) laptop I discovered that a surprising amount of noise was creeping up from the external switch-mode power supply. I suspect laptops are going to be worse in this regard because they usually don't have a ground/earth line, where desktops (and their metal-cased power supplies) do.

USB devices are (in the main) digital and the designers are expecting a lot of digital noise on those power lines.

But usually they don't have to regenerate that supply voltage to a much higher level - around 30V here. 29

Measuring with a DVM

A quick word about measuring voltages around your board though. A few tenths of a volt variation can be due to a number factors. (Cue Yorkshire accent) back when I were a nipper... we had little mirrors at the back of the meter's panel so you could be sure you were looking directly at the pointer. The idea is when the "needle" and its reflection line up, you're getting a true reading - assuming the people in calibration had done their job right of course.

Oddly enough, a lot of very high-accuracy equipment was made using gear like this (and some tricks like Wheatstone bridges) but we learned that in engineering there's a point where "it's good enough". This is unlike physics were a miniscule error in some calculations can create flusters like "faster than light neutrinos", a story from a few years back.

Remember that the scientists (men and women) who put the first men in space and later on the moon, used little more complicated that simple computer, slide rules (which are only slightly better than "I reckon") and log tables, books of pre-calculated values accurate to maybe six decimal places. Early aircraft from Boeing to Cessna and others were designed in much the same way. And yet they flew.

Calculators and digital measuring gear are very accurate these days and that can lull us into the idea that such accuracy matters when it often doesn't. Take a tenth of a volt on a five volt supply. That's just a 2% deviation which is not very much at. The spec says you can go around 1/2 volt under (11%) and 1/4 volt over (5%) - and that's assuming you have accurately calibrated equipment to check this.

A typical Chinese DVM might have a voltage (in)accuracy of 1% or 2% itself so your measured "drop" is not an issue and to be expected under a moderate load as presented by the NMA0515.

The voltage at output is also well within tolerance of the NMA0515's spec. It can go quite a bit higher (>20V per rail) when there's no load but the THAT151x does draw enough current to pull it down to something more reasonable when it's all running.

Your scope will show the noise on those USB lines and you're likely going to find there's all sorts of electronic "hash" all over your PCB. In the end, I went with an almost (*almost*) bullet-proof design which isolates the ground entirely. Matt is experimenting with those designs in the new year - as a final acid test. There is a smaller, simpler version without the isolation which I haven't had made up as yet but all of these boards are on my Github page: https://github.com/marcdraco/ where you can see the circuits, PCBs and so on. It's all Open Source Hardware so you can make any changes you see fit. They're KiCAD 8.x files and I haven't necessarily got all of the order numbers right so it's smart to check if you're ordering from JLCPCB.

The ultimate combination (Varee/Michelle) is quite a bit more expensive than a Vero and lose transistor design of the original but splitting it into multiple sections allows everyone to make the project they want at the budget that should be affordable.

I've done a few different capsule adaptors to suit various alternative designs. Sarah being is  simplest but she also requires a little knowledge so you get the right parts for your particular application (common source, common drain, "

balanced" out etc.). Sarah doesn't really do balanced in the strictest sense, but it supports the original DIY Perks design.

Jaime is the most recent one and sits between the Sarah and Varee in terms of complexity. It was suggested through discussions in this thread questioning if we needed a balanced output at all. Jaime runs from 5V (actually a little less) and should work from a "noise-free" USB supply quite well although the intention is to marry it to a USB mic/headset adaptor or even a standard USB Mic port - because those have to be noise free. Noise Free being an engineering term for sufficiently quiet not to be an issue. All circuits have some amount of noise - a resistor sitting on your bench generates noise without any external current flowing through it because it's above 0 kelvins. This noise (and noise from the active components like the FET and the THAT all contribute some amount of noise - and that's measurable too. The trick is to swamp it with the signal we want.

Varee 2 came out of the need to remove even more smush from the supplies when I found (much to my chagrin) that the USB and NMA0515 switching smush could travel all the way to the FET (impedance matching circuit) which means that any noise on the supply lines gets coupled back into the amplifier and... all you can heard is a horrible whine as above.

The balanced line is intended to eliminate mains hum and other signals that jump on the interconnect between the mic and amp - stuff that manages to get through the screen. In fact, twisted pairs are often used without Faraday screens because the noise rejection is often good enough. A lot of Ethernet cable, which is transmitting digital noise that can't be corrupted (error checking is there to check for occasional errors, not a compete mess) uses UTP - unshielded twisted pairs. And better yet, if you don't need to transmit power to the other end, you only need two wires.

Balanced lines are your friend.

This is where (and why) I isolated the power for the V2 board Michelle/Varee combo because that allowed me to screen the mic and FET with a Faraday cage that wasn't referenced b

ack to the computer's ground - which is where a LOT of noise is coming from. Doing this requires two NMA0515s. It could conceivably be done with one but the extra circuitry requires more current than these 1W devices can supply. The +/-5V ones might have worked but developing these things takes time and money. Perhaps someone else would take that one on?

If you're interested in how to do this, the idea is to isolate computer USB ground from the audio amps and then deliver a "balanced" pair to another differencing amplifier which is referenced to USB ground. This means getting four op amps (two duals or a quad) that work down to 5V, configuring them as a three-amp INA with a 1/2 rail bias at 2.5V to allow for the AC signal from the THAT. There are a couple of designs on my Github (one SMD the other through-hole) which perform this function if you want to see the circuits. You'll find them here: https://github.com/marcdraco/Quinn-Donna.

This USB supply doesn't have to be (quite) as smooth due to power-supply rejection in the Op Amps but it does need some.

I can't be sure what's causing your design to quit when you connect up the screen though, unless you've wired something wrong? The screen should be referenced at 0V with respect to circuit ground (USB ground) and the two signal lines carry +15 and -15 volts (ish) with respect to that.

This is where Matt's design (rather cleverly) gets around the usual way of doing "phantom" power which is always a positive voltage with respect to the shield. The reason for this goes back to this problem of ground isolation which is simpler when you're working with a power supply that is already galvanically isolated by the input transformer.

It's all a case of the devil hiding in the details.

As a complete aside (and a different project or variation of this) I've ordered a P48 compatible radio microphone adaptor from Ali Express. The Jayette C-01 [ https://www.aliexpress.com/item/1005007827004776.html ] which claims to supply 5 or 48V phantom from a 3.7V LiOn cell. The receiver is self-powered too which was a pleasant surprise IF it works with my designs - I don't know how quiet/noisy this thing us yet - it might be horrible, but there's only one way to find out.

The 5V option is the most interesting to me since Jaime, which has the performance enhancements of Varee (which is P48), runs at just 5V so that might be an option for a wireless mic. The output is (supposedly) compatible with a standard - balanced or unbalanced - mic input. I expect the actual output is unbalanced but the direct injection should remove that as an issue. This means it should also work with Matt's board in addition to mine. However, the question remains of how good it sounds. I'll run some tests over the next week or so.

EDIT: did a quick voltage test and (not to my surprise) the 48V phantom is nearer 28V and the 5V phantom is just 3.7 (which is a direct line to the LiOn cell. But I guess they don't expect people to check this stuff. It also hisses like a mutha too. I tried it with a professional grade microphone, an AKG perfection studio mic - and it does work, IF I scream at the thing, so as far as "real" mics go, that one is a total bust. Multiplying 5V to 48V (or even close) is very difficult indeed and seems to need VLSI to do effectively due to the sheer number of MOSFETs involved in the process. 3.7V is harder still, and the 6-7x multiplication is unlikely to be sufficiently low impedance to supply the requisite current (about 4-5mA). I try this stuff so you don't have to. 🙂

@marcdraco I think I didn't express myself quite right.

I have not wired up the capsule to a proper audio cable yet, I have left all the housing and mechanicals for last so I don't know how long it should be. I've left it at the braided cable signal wire combination with a capsule housing for now. Obviously, this picks up a lot of static. On the oscilloscope I can see Mains hum in the signal and I can see that positive/negative halves respectively, as the circuit is supposed to do, therefore, the transistor is working I believe. If I touch the braided shielding to a ground, the mains hum disappears and the entire signal drops out, leaving nothing except indiscernible noise at Mic+/Mic-. So I have no clue what the capsule is doing or why, I assume it's impossible to measure such small voltages anyways, so I am not too concerned about the capsule inputs, for now at least.

I was measuring the circuit as I was soldering it, notably, I checked the USB voltages and the boosted voltages. Before adding the THAT1512, I had 5.2V and -+15.11V on the lines which dropped to 4.9V -+14.5V after I soldered everything.

I don't think there is a lot of digital noise on my USB signal. Whenever I disconnect the preamp, the output is dead silent. Not even any noise. Interestingly, if I touch the input lead of the USB converter, I can hear mains hum in the recording, but still with all that horrible noise. I can try applying a pure sine wave to the input of the digital circuit to check and make sure but I really think the preamp, especially, because the noise persists without Mic+/Mic- connected to the preamp circuit.

Okie doke. I’ll run up a list of tests you can run with a multimeter. Your scope will show the output from an electret mic like this one but I doubt that is an issue. This only works with this type of capsule because there’s no FET in the way. 

Okay so after looking for capsules for quite some time, in germany I could only find the following:

https://www.micandmod.com/de/microphone-parts/microphone-capsule/tsb-2555-electret-capsule-premium/

https://www.thomann.de/de/the_tbone_capsule_for_sc_450.htm

The Thomann ones are for their own in house microphones and judging by reviews are received pretty well. They have multiple capsules depending on which model they're for.

The alternative to these would be ordering from the UK and paying for postage which at the moment seems like the less appealing choice.

Thoughts on these capsules?

Update:

I noticed that both of these have crazy shipping costs as well, meaning i'll probably just go ahead and order from the following link and pay that shipping fee.

https://micbooster.com/microphone-capsules/262-jli2555.html

Looks like a re-badged JLI2555 - not just from the similarity in model number, but it also has the same "error" on the datasheet because no one thought it was wrong.The specs are essentially identical as is the body construction and overall dimensions.

The giveaway seems to be this 10V max voltage. Now that could (emphasis on could) be the maximum voltage tolerance of the air-gap capacitor - that's the microphone itself, but from looking at the circuits you'll see that this always includes a JFET and typically lists a 2K2 output impedance. The output impedance of a capacitor mic is in the order of GIGA-ohms.

Here's a deeper dive if you're really curious.

The Weirdness of Impedance

I talk about impedance a lot without really explaining what I mean but it's far less complex than it sounds, so for everyone's sake, let's have a quick chat about that.

Impedance is inextricably linked to ohms law.

There's the output impedance of a circuit driving something and the input impedance of a circuit being driven.

But what does this really mean?

Impedance (which is sometime  "complex" because it has real and imaginary parts) is simply the "R" in Ohm's equation which means that you get it from  current and a voltage.

At a given voltage V, a current I flows through a resistance R.

The impedance of an electret is a wee bit tricky because it only causes current to flow when the diaphragm is moving. Further up the thread I discussed how electrets are permanently charged and "push or pull" electrons around due to the field holding them in place changing as the air gap gets smaller or larger. We can determine the charge if we know the capacitance which is around 40pF and its max voltage which I'll assume to be +/- 100mV at full extents. We can't stick a meter on an electret because the charge doesn't flow as it would in a normal capacitor. The field in "normal" capacitors is temporary and when when measure the voltage we're actually removing a small amount of charge. This is a very crude example of the Heisenberg's observer principal. The act of measuring the voltage actually removes some of the voltage.

The number of electrons for a 1 farad capacitor at 1V is one Coulomb = 6.25 x 10^18 electrons. (Imagine having to count that lot!)

This works out to 250 x 10^9 electrons for a 40pF cap and 250 x 10^9 at 100mV.

The current flowing is a bit more tricky and is most easily inferred using a FET with a gate load resistor. I know from experiment that it's in the order of nano amps, and more realistically in the order of a few hundred picoamperes to a couple of nanoamps

So we can GUESSstimate the output impedance of the capsule like this assuming 1nA:

100mV/1e^-9 amps = 100 megaohms.

It's convention to use a load resistor (for the next stage) of 10x the output impedance of the prior stage so if these calculations are correct we'd use a 10G load resistor in the gate-source circuit to avoid loading the capsule. We could use an even larger resistor but unsurprisingly there are limits even to this (not least in producing such eye-wateringly large resistances accurately!

The limit for a JFET is determined by the amount of reverse bias (that's current flowing from the more positive Drain to through the Gate to ground). This weirdness is a result of the way a JFET works by producing what is electrically a tiny capacitor at the diode junction formed where the gate terminal is mixed with the Drain-Source channel. In the self-biasing schemes we use, the amount of current is matters - we need enough (in the order of a few picoamperes) to cause a "capacitive field" to form. (Real diodes also produce a variable capacitance and can be designed to produce voltage controlled capacitances - which is how electronic tuning works.)

You'll find the specifications for the typical input capacitance and reverse currents on the spec sheets. For my favourite LSK170 that's about 20pF under typical conditions but this will change depending on the Drain-Gate potential and channel current. And this capacitance also loads the capsule to some degree.

(@YamiTheWitch you might find something of interest here too.)

The input impedance of our little JFET is essentially the same as a reverse-biased silicon diode - meaning that it's got a very high input impedance in its own right - usually assumed to be in the order of >100M. That's in parallel with the gate resistor so that about 90Meg total assuming a 1G resistor. So the actual impedance of this circuit is largely determined by the input impedance of the JFET.

If the gate impedance is too high (such as in the case where the input resistor is left off...) the JFET will not turn on. It's not entirely clear to me how a floating JFET soldered the way Matt did it, works. I have to assume that there is sufficient leakage caused by handing to provide just enough. I hit this snag when I did a very high quality PCB design which isolates the gate from everything else with a guard track.

When the board arrived (and I didn't have any 1G resistors) it worked for a few second and then just stopped and would remain like that until it was switched off and on again. In this case that 1G resistor (or the little bit of graphite) is essential. I think you asked about it wearing away Stefan - the answer is a small dab of the wife/girlfriend/sister/mom's clear nail polish.

There's a chance in fact that some people can't get the wretched thing to work for this precise reason. If that JFET doesn't self bias, the capsule won't work no matter how loud you shout at it!

A quick test is to use a few 1Meg resistors in a chain from gate to ground/shield. It will "work" (after a fashion) with such a small impedance but that will load the capsule down so you won't get the full signal. All three of the DIY Perks capsule carrier boards/adaptors have pads for the 1G resistor for this reason.

MOSFETs - which are the transistor of choice for almost all circuits these days (by component count) have a much higher input impedance (>1 terra-ohm) and don't need the fancy reverse bias. We don't use them in audio because they are noisy compared to the lowly JFET.

TL;DR

But all this talk of impedances really boils down to this:

Output impedance is the amount of current the circuit can supply at at given voltage.

Input impedance is the amount of current the circuit requires to produce a given voltage.

Batteries and other sources

Even a power supply like a battery has an output impedance. The impedance of battery (called it's "internal impedance")  changes as it wears out. You can see this on a meter if you measure the unloaded current (or the current with fixed load) and the voltage. Don't try this unless you have some idea of how much current your meter can manage without blowing a fuse.

A typical AA alkaline cell can deliver a peak 700mA under shorted conditions at 1.5V

This works out to an internal resistance of 1.7 / 0.7 = about 2 ohms.

Let's drain that cell down to 1.2 and it's now producing a short-circuit current of 70mA.

1.2/0.07 = now the internal resistance is 17 ohms.

And it will keep on rising as the battery runs down.

The internal resistance of a battery is equivalent to the output impedance of a power supply. All driving circuits work in precisely the same way with output impedances varying widely. A 90W laptop charger has an output impedance of around 4 ohms.

Negative Thinking

Things get a little weird when we account for negative feedback.

When we use negative feedback in an amplifier the calculated impedance in the order of tenths of an ohm or less but the reality is a bit more complex.

It would appear, given the impedance is so low, the device could supply huge amounts of current but we can infer that cannot be the case reading the spec sheet which for the 741 is 25mA. That's enough to drive a small load but the action of negative feedback means that capacitive loads - which look like shorts when things power up - throw everything to pot. If the amplifier can't charge the capacitor fast enough through its output stage, the rest of the loop is held up and that causes a phase shift which can throw the negative feedback so far out that it becomes positive feedback and what you've got is an oscillator.

@stefanotot Low cost mic capsule

I've gone through about a 1/2 dozen of these little fellas - mostly because they're a lot cheaper than the JLI2555. Ali also do the 34mm version which has a slightly better bass response at the expense of less high end. (Between us they're a good bit more sensitive too.)

I mean, I was wondering how the floating transistor was supposed to work. I guess I can try to add a resistor between gate and ground, but it's gonna be pretty painful to desolder the capsule assembly. Is there any other quick fixes to try?

Okay, so I added a resistor from ground to gate and tried recording again, this is the result. Inaudible noise. If you boost it by 50db you can hear it but as is, nothing. @marcdraco

@yamithewitch Ok well that's that potential problem eliminated. I have a Steam Deck to fix at the mo (just a new hard drive) when I've done that I'll get back do you with some DC tests.

OK @yamithewitch, let's see if we can see what's going on.

I'll assume you know to connect up your probes, but to be sure everything I'm going to ask you to measure is referenced to the 0V rail on the USB input. In Matt's design (and in Dorit) the ground is common to both the 5V USB input and the ground reference for the whole circuit. On a PCB this has the potential to create noise all over the ground and if you've painted with ground fill this can create real issues as the THAT1512 is AC referenced to that part of the circuit.

With that said, we can get the voltages around the THAT to get some idea of what's happening. Disconnect the capsule for these tests. Ideally, the two input lines should be grounded but that's fiddly because you don''t want to shunt the phantom power

On the DC range (20V or auto-range) so you can measure the power rails.

Pin 4 (bottom left) should be -15V (anything from 13V up is fine).

Then pin 7 should read almost exactly the same but positive.

Switching to a your lowest range (200mV typically) pin 5 should read 0V - it's unlikely you'd see any ground bounce with a DVM, however you can do that with your oscilloscope. This is worth looking at as any noise here IS going on the output.

Pin 6 will have a few mV DC when there's no input, this is normal. It's important to measure these directly from the DIL so you're not measuring via a capacitor.

Pins 1 and 8 are the gain resistor. You can measure across the resistor that but it's a fairly small voltage due to the tiny currents and small values of resistance at higher gains.

The voltages at the inputs should be + and - 15V. I think you've already done those but it's smart to check them again.

Let me know how you get on with these and we'll take it from there.

As an aside, if you don't mind sharing your PCB design, I'd be interested to see if anything pops out at me. You can just PM it - I have no reason to share it so you can be assured of privacy.

Before I start measuring, I seperated the 5v ground and +-15V ground. I assumed that was the intention with an isolated voltage booster.

Okay so, across gain it's 0.1mv. Pin2:13.11V, Pin3:-11.13V, Pin4:-0.4mV, Pin5:-13.11V, Pin6:-1.5mV, Pin7:-0.4mV. I measured against boosted voltage ground at the Mic output ground cause it's easy to access.

Can you double check those. From memory (I’m in my pit awake following the cat trying to flood the bathroom and my neighbour).

Pin 4 should be -13.5 ( around that) as it’s Vee and Pin 5 should be ground or very close. 

Pin 7 is Vcc so that should be around +13.5 (V+). 

if your measurements are correct it looks like the power rails are messed up. 

Oops, went clockwise last time, here are the actual numbers. Pin2:-0.6mV, Pin3:-0.5mV, Pin4:-13.81V, Pin5:-0.3mV, Pin6:-11.40V, Pin7:13.80V

Power rails look good but there are a couple of concerns I have. The output should (pin 6) should not be any higher than a few hundred mV.

Also, it looks like (from those measurements) that you have some latent noise on the ground - that might be ground bounce so you should check that with the scope to see what's going on.

There's a chance the THAT has gone to silicon heaven too if that output is hard over on the negative rail. I don't suppose you have a breadboard you could test it in isolation?

@yamithewitch Sorry, I missed the bit where you separated the grounds. You're correct that is the purpose of isolation but the circuit has to have a reference to a common ground for the audio output to reference. Ground isolation is a bit more complex when you are sending a signal back to something.This is where balanced lines come in again because they are self-referencing. The Michelle 2 design (you can see if on GitHub if you have KiCad 8) does has an balanced output with full ground isolation but a considerable extra cost.

I don't think the seperated grounds have much of an effect, if i short the 2 grounds together with a piece of wire nothing changes in the circuit/signal. I have some spare perf-board lying around, is there any particular test setup you want me to try?

Nothing comes immediately to mind - at this stage I'd probably go back to a breadboard but that's a lot of work when everything is soldered up.

Ground isolation won't work properly while there is still a ground path somewhere - which in this case is from the computer to the USB digitiser. I should have thought of that earlier but --- I didn't 😯. I was thinking of the way I did it with a balanced output using the two op amp buffers to send an phase and anti-phase signal independent of circuit ground.

A lot of what you can try depends on what sort of kit you have to hand. A decent (noise reduced) power supply with a 5V regulated output is very handy because that shouldn't carry any noise to the circuit which only has rudimentary ripple reduction. It's generally a smart move to have at least one inductor in the supply path to kill of the H/F rubbish that's riding on there. I've found that different PCs (I have a dozen or so lying around for different jobs) have vastly different levels of noise on the power line.

Have you tried scoping the INPUT 5V and the output "15V" lines? You will have to set the scope to AC to see these little signals and they are probably both fast >MHz and not that predictable so it's difficult to set a trigger point. The THAT does have power supply noise rejection but the FET (being just a FET) doesn't - so anything escaping onto those phantom rails is going to be a massive headache.

I'll try some tests over the next few days (time and Christmas lunacy allowing) to see if I can get some waveform dumps at various points. We can't ignore very high frequency stuff even though we'd love to, because they can fold back due to interference patterns. If you add two sine waves of the same magnitude but different frequencies, you end up with the sum and the difference of the two (a hetrodyne). This is how the intermediate frequency of most radio receivers work, by creating a much lower frequency than the incoming one, it's easier to amplify it. Apparently this is also used in optical signalling to but that's a bit outside of my field.

But the crucial point here is the difference signal. If the two frequencies are the right space apart the difference frequency can show up - and it's Sod's Law that it's going to infect your PCB.

It's a shame we can't easily analyse the power lines with an FFT (unless your scope has that function) but even then I can't really imagine that would tell us a whole lot. You should, however, (assuming your scope has it) run an FFT on that noise you're getting and compare it to my results. I've had cases in testing where the superb frequency response of the the THAT1512 has caused the device to saturate with noise - which, naturally, the Audiograbber can't digitise. When this happens the 1510/1512 get more than a little warm - about finger hot. Not enough to commit "Hari Silcone" but plenty sufficient to stop our desired signal coming through.