[Sticky] USB-C Microphone (official topic)

My scope does have a pretty good FFT and I have a reasonably good benchsupply. Here are the scoped images of 15V, -15V, 5V and Output lines in that order. The first 3 are from 0-500MHz, I couldn't increase the volt/div on the output line a lot so it only measured up to 5Mhz (though, on the output line, only 0-20Khz matter anyways. I am assuming 50Hz mains hum is just cause the PCB is unshielded. The THAT1512 does get a little warm and the preamp draws 250mA on it's own.

Siglent. Nice scope! I want a Tectronix for Christmas but Santa never listens to me.

Foldback harmonic interference will appear back in the audible range That's a real issue with mixed mode signals on the same board because square waves have such a rich harmonic spectra. For reference, here's a 50Hz pure sine wave in Audacity. Note that he peak energy is at 50Hz where we'd expect.

You'd expect some energy at 50Hz (or your mains frequency) when the board isn't shielded - how much is more difficult to gauge because everyone's home is set up differently.

A 50Hz square wave on the other hand... is rich in harmonics as you can see.

Now a pure 50Hz riding on some white noise - the peak is 50Hz as expected with lots of little peaks running all the way up the scale.

So if your issues were mains "hum" we'd expect to see some peaking there and indeed, in the pure sample (without correcting for amplitude) that's where the majority of the data seems to sit.

But normalising it, we can see there are peaks all up the spectrum all the way to the limit of what Audacity can work with. Some of this is probably due to aliasing but MP3 loses a lot of the HF detail, you usually need to save as floating point WAV to get the most useful data out.

Filtering off the L/F crud we see a more detailed picture of harmonics at popping up in places they have no right to be - note how the spectrum of pure white noise (the sort of unfiltered mush you get primarily from Johnson noise)

Looking at the wave (this is without the capsule) and removing the worst of the noise we can see a regular(ish) pattern:

There's a small amount of chaos in here (that's to say those peaks aren't entirely regular, but that's predictable because of the sources of noise all contribute in ways that are like the 3-body problem). What you have here is an oscillator with a fairly long time period and if we zoom in on that we can see (highlighted section) what's known as ringing. Note there's a large peak followed by the wave decaying slowly before the cycle repeats. This is a sure sign you've made a low-frequency oscillator and that is quite possibly a result of something pulling too much current from the NMA0515, shutting it down. As the excess load vanishes (because the circuitry has come back to life) the cycle repeats.

If this is the case you should be able to see that by scoping the power rails on a fairly slow sweep - the time period looks around 3/4 second for the cycle so it's too slow to be audible. Why this should happen, I can't be 100% certain. It might be that your USB side is dropping out at a peak current (it should be capable of 500mA, but if there are more things on the same hub drawing power, it'll go down faster). This circuit, by virtue of the split 30V supply is a bit of a power hog so there's no surprise if it's drawing a couple of hundred mA even though the NMA itself is only good for 33mA max (per rail). One option, although it will be a bind to modify the PCB but the one that worked for me is to put a capacitor multiplier on both rails. I used a Darlington but a single transistor will work.

The idea of a capacitor multiplier is to generate a very smooth DC with an emitter follower. Followers "follow" that is match the voltage on the base to the emitter minus a diode drop of the Base-Emitter junction. The follower allows the circuit to draw large amounts of current 10s of mA without the capacitor drooping because the capacitor is only loaded by the base.

For a typical small-signal transistor the beta (you can assume it to be 100) means that we can draw 100 times more current from the emitter than we could from simple low-pass filter formed by the 100R resistor and large electrolytic capacitors. There are some great demonstrations of this by people like David Jones of EEVBlog which go into the weeds a bit more and I won't repeat that here.

I've had a look at your PCB but it's not clear how you've routed everything. Might I ask what you used to design it? KiCAD which is the only software I fell justified in recommending to hobbyists like us is free and very powerful. I think you mentioned you'd dropped a clanger somewhere but KiCAD will stop that happening to a large degree because it can test (Design Rules Check) for more obvious blunders. That and simulation will get you a long way - and while the simulator can be tricky to get used to, it's very good indeed. Just be aware that SPICE is (a) only as good as its models and (b) quite a few things don't simulate as well as they might due to limitations of how SPICE works. This can lead you up some dark alleys in places like actual oscillators and in particular, op amp stability.

The problem is that in some cases capacitors can overload the output transistors and cause loop instability. The THAT is only guaranteed to be stable into 300pF which should only be an issue driving some coax. In NORMAL operation an external DC block shouldn't be required due to the balanced input - so any DC component should be removed by the action of the differential input stage. You could try a small resistor - say a few hundred ohms to no more than about 5K in series with that capacitor to prevent overloading the output. (Recall that a capacitor looks like a short circuit for a few fractions of a second until it starts to charge, and even then, the draw is considerable until the charge begins to rise significantly causing the current to drop off. This overload can trigger the amp's current limit to kick in.)

I can''t say any of these things are the issue - and it might be you've accidentally run a signal lane too close to a power lane and that's causing something else to go squiffy. I've gone overboard on the Michelle design using 4 layers everything as far apart as practical and giving the power lanes a ground plane to return on. Obviously this comes at a cost premium and .. well it worked on nasty Verboard didn't it...

I just realized, the output on the THAT1512 is -11V but the signal is a 50Hz 6V peak peak. I am hitting the maximum -15V that the THAT1512 can output. Is the THAT1512 working fine but just biased extremely negatively?

I was looking into making this microphone and the most confusing and not familiar part is making the preamp. I was struggling to find stripboards on the internet in the US. If anyone knows where to get the stripboard, capacitors, resistors, wire, ect please let me know, Thanks. Also where can I find a suitable rotary switch?

Edit: I found what appears to be the exact stripboard from the UK on ebay but the shipping is more than $30, so that makes it a bad option.

Stripboard or usually Verboard (which is the trademark of the Vero company, and generally of much better quality than the generic stuff) should be available on eBay or Amazon locally as will many of the parts.

However, you can get pretty much everything at the large electronics suppliers like Digikey and Mouser. The will also carry the multipole switch this one requires.

If you have trouble sourcing the JFET Matt used, there are a few alternates that work as well (or better, on paper anyway) including the 2SK208 and the LSK170.

Three years ago I saw the DYI mic and noticed many improvements could be made.
Low noise is one of the keys to a good mic as you know. Many different paths for noise to enter and different ways to minimize the noise and some other improvements.

Filtering:
1. DCDC converter: I don't see the need for the big RC filter (100Ω + 2200uF) with huge size electrolytic caps. The dcdc converter only has 33mA max output and the circuit current draw is very little. More importantly is low voltage ripple (from getting into the signal through the supply rails). The dcdc converter is running at 90KHz, that is the target freq needed to reduce ripple voltage. As a minimum a LC filter of 220uH and 1uF would be recommended. In effort to reduce voltage ripple even more, adding a second LC (making a PI filter) is one option. This should do much better for filtering noise from the switching boost converter and USB supply. For the input dcdc rail, low esr (X7R or X5R) 0.1uF and 22uF ceramic caps should be added.

2. IC Amp filtering: I would add caps placed as close as possible to the IC power pins. A low esr (X7R or X5R) 0.1uF + 1uF ceramic caps on each rail (pin #4 and #7).

3. Max capacitance: The dcdc converter datasheet shows "the maximum recommended output capacitance is 10μF", having a 2200uF with 100Ω might be alright, but I think the LC and/or PI filter, plus the rail caps close to the amp IC is a better solution.

Signal path improvements: Keeping all input analog signals short can help reduce noise.

1. The number one most sensitive signal is output of the mic, so keeping this signal as short as possible so placing the transistor as close as possible helps like in the video and away from the power supply.

2. The second most sensitive signal is outputs of the transistor, in the video, these wires are quite long and sort of shielded a portion of the distance in the video. I really like the idea of very short wires from the transistor to the amp (THAT1512), this can help with lowering the noise from long wires.

DCDC preload / LED's: The dcdc needs preload resistors that help regulate the dcdc output voltage. A single 4.5K resistor will work on each rail. As a preload option, you can add two LED's (one for each voltage rail) if desired. Preload current should be a minimum 10% of the output max current (33mA). so a minimum of 3.3mA is desired for preload.

If using the leds as a preload, R5 and R6 with 3.2K should give minimum 3.3mA with any color LED.

Resistors: I would personally use 1% on all resistors keep the circuit balanced.

Volume improvement: (only if you dont like the stepped volume in version #1)

1. A nice quality Alps or Bourns pot. For more info see THAT Design Application Note 138 for GAIN control - (THAT dn138) has additional info.

PCB improvements: These can help keep noise low.

1. Ground plane: A dual layer pcb with mostly all ground on the bottom (and using a SMD AMP ic) with stitched (many vias) to the top ground are common in high preformance low noise pcb's.

2. Short signal paths and routed away from power voltage rails can help.

3. Terminate components close to amp IC. (all input and output components placed close to the amp)

4. Distance the dcdc converter away from all signals and the AMP (as much as possible). The dcdc has
toroidal magnetics so it may not be an issue, but it is just good practice. Another option is mount the
dcdc on the bottom of the board and use the ground plane as a shield.

5. Direct mount Mic to PCB: Another option is design the pcb to allow the mic tabs to be soldered directly to the pcb and having zero cable (and then having Q1 mounted on the pcb).

Mechanical: I like the look of the mechanical design in the video, but with my long mechanical racing education/experience/tuning, the long tight O-ring suspension I believe is not optimal. It's too stiff is my guess. Normal high end mic's use many soft O-ring suspension for more compliance to external disturbances and a heavy weight mic assembly to help stability. So more mic mass helps this issue and placing both amp pcb and usb pcb in the same enclosure helps two birds with one stone (less noise and more mass, exactly like the brass does in the video design).

I think if I designed a pcb, it would be small enough (without all the large, unnecessary large and wrong value output caps) by using all ceramic X7R or X5R SMD caps. The pcb would fit right behind the mic element (maybe around 35mm diam), then you would not have any long run wires of the tiny sensitive signals and not have any chance to pick up noise. Depending on the usb audio input module size, it might fit behind the mic/amp pcb too. With this setup I would use a nice gaming mouse braided usb cable for flexibility, aesthetics and better preformance.

----------------------------------------------------------------------------------------------------------------------------------
UPDATE:
I started playing with easyEDA free schematic capture and pcb layout about a month ago for the first time, its pretty fun.

After playing with the mic schematic, I added all the things I suggested above. Learned from reading the datasheets, I changed the dcdc output capacitance, added the LC filter, added a option to run led's front or back of the pcb for the required dcdc minimum loads, simple resistors or both. Use only LED1 or LED3, and LED2 or LED4. I decided to add the second LC filter for the dcdc as this is a audio application. Its easy to bypass this option by shorting the inductor pads (L1 to L4).

Soldering the mic to the pcb shortens the critical signal length, so the signal coming out of the amp is much less susceptible to noise.

Volume adjustment: I prefer to have a pot instead of a stepped volume adjustment. Reading the APM IC datasheet refers to the application note "THAT corporation Design Note 138" shows many options for a volume pot and how to implement. The basic config is to use pins 1 and 2 on the volume pads and no R9. and ~5Ω value for R10.
Using a quality and the correct audio taper pot is recommended.

The J-FET (Q1) comes in three packages, two with metal cans (these are the best), but they can have different lead pinout spacing. So I designed the pcb to accept both, and in addition a surface mount package that should also fit in the Q1 pads (without using the case connection). I suggest getting the 2n4416A version, as it has slightly higher voltage (35V) verse the non "A" version that is rated at 30V and with +/- 15V rails, thats 30V.

Mic mounting options: The dcdc is 10mm high, Q1 is 6mm high with the metal can package, this shifted the mic mounting up to clear the dcdc instead of being centered. Also, the mic can be mounted on the back side of the board.

DC input power: Standard usb voltage or any 5V supply between 4.5v and 5.5v is fine. Added 22uF and 0.1uF for high freq switching noise coming out of the dcdc. Added a second set of 5V pads for aux circuit if needed.

PCB:

I ended up going with 45x45mm square pcb two layers with mounting holes spaced apart at 35mm square. this should fit in a 50x50mm or 2"x2" project box. And can panelize the pcb with four up on 100x100mm pcb with JLCPCB, so quantity of 20 is around $2 plus S/H. (I have never ordered from JLSCPV yet)

Copper poured top and bottom for lower noise and separated the input to dcdc from the ground planes.

Added a couple ground pads next to the mounting hole pads (GND_1 and GND_2) that can be soldered jumpered to connect the pcb ground to the case if desired.

USB Sound card options: Another option to gain preformance, but at what cost,,,
1. The low cost USB audio input is probably 16bit/24KHz or maybe 48Khz and great bang for buck,
2. The soundblaster AE-5 is 24bit/96KHz and around $60 to $70 used,
3. Then the price starts to up up quickly, 24/192, 32/192, 32/384
Would be nice to test the difference between a few of these.

USB: USB3 is not necessary unless you need more than 5V 500mA and this circuit does not. USB2 has plenty of data rate for this application.

Its not letting me attach the gebers/etc files, they can be found in a zip file here https://www.rcgroups.com/forums/showthread.php?4022321-DIY-Microphone-Improvements

My current version of the schematic and pcb:

Some interesting ideas there, we've already implemented these (and more) for the various "capsule" heads - many of which are described in the forum, but I expect you won't have caught those since this is a LONG thread... Still great minds and all! 🙂

But that's great thing about the community here. Curiously Matthew and I were literally just discussing the V2 a few hours ago and various options that have popped out of the iterative design. Including a P48 version that hooks up to professional phantom power with the low-impedance to drive long cables and some big/hairy PNP output drivers for reduced noise due to the much larger die.

All of mine (including some early cock-ups) are on my GitHub which includes a circular carrier complete with hooks to fit mount inside a BM800 microphone (Ali Express) for people like me who are all feet and no fingers when it comes to metalwork.

If you haven't already ordered yours, might I suggest you pop some "breather" holes in the back because without them the cardioid response is lost and you end up with an Omni which rather defeats the object of having such a sweet-sounding capsule. I'm just putting the finishing touches on something that converts the JLI2555 into a more conventional mic body with terminals for the source, gate and drain (uncommitted) for more advanced designs where there's more space to work.

The other thing that caught me out when I did the first "proper" PCB designs was the JFET won't self-bias as there's nowhere for the (albeit feeble) amount of reverse bias for the JFET. Cheap electrets don't have these, relying (I hear) on leakage on the circuit board but for such a simple biasing scheme, a 1G resistor is really the order of the day. I've found little graphite works if you can't find a 1G resistor - Mouser, etc. carry them but by the time you've paid S&H a single one will cost more than the rest of the BOM (if you get the rest from eBay)!

Couple of other things (some I learned the hard way) is that those little MLCC caps, while terrific for power smoothing are a bit of a no-no when it comes to critical circuits - partly that they are microphonic but mostly that the marked capacitance changes considerably when operated with a steady DC bias - which varies the impedance over the audio range, giving rise to distortion. When pushed "hard" this configuration starts to distort quite badly (several %) during peak program too. This is one area that modern electrolytics are actually the only real choice. Film (polyester, PTFE) caps are really the order of the day for a signal path but they're physically to large for a board that's meant to fit into such a cramped space and only available up to maybe 1uF which isn't nearly enough for the input load resistors. You can use larger load resistors but that means a huge increase in Johnson noise. I suspect the designs on the THAT's datasheets are intended for a much lower source impedance (<600R).

Input impedance is another gotcha on the original with the divider between the 2K2 load and 3K9 input resistors. Not a huge issue when you have 60dB of gain at hand fortunately.

My own experience of the NMA0515s was that they are interminably noisy despite running well beyond the audio range. Although that could (equally) be noise coming via the USB which I think you noted is noisy too. I found a combination of a small inductor (pretty much a requirement on USB) and a darlington-pair configured as a capacitor "multiplier" was most effective. YMMV of course.

Sourcing a THAT1512 has proven difficult for many people so I've done a version using Texas/Burr-Brown's OPA2134s as a three-amp INA.  Were you able to source it in SMD? I was only able to get it on a back order via JLC and it's not budget friendly. 😉 Once you add in cost of "free" assembly, "extended" components and so on they can work out quite costly.

The JFET (I'm sure you know was a radio part and is no longer in production like many JFETS) can be replaced with several more easily available ones. I prefer the LSK170 and its dual-version like the LSK389 and LSK489 (from Interfet) although after testing this thing to destruction (and for most uses, this is a vocal mic, not a measurement mic) I came to the conclusion that the benefits of the Interfet devices was far outweighed by the extra cost and a (albeit rather pedestrian) 2SK208 or even a J201 of all things does the job more than adequately.

The J201 can also be used in a "Feltzer" valve simulator too albeit with some modification.

Digitisers have proven an issue here too. Despite having RCA (yuck) sockets on the original device, I found out after assuming that it would use a standard Line in voltage, it's referenced to the 5V USB line which is ... AGGGH #&%*!. Makes sense but if CMI (and others) can made a very low-noise to USB device with a 10-100mV mic input, what's their excuse? And worse, push that digitiser over 5V (ish) for a fraction of a second and you can kiss goodbye to £15 (or fib to Amazon - "'t'was like that when I got it govn'r).

I don't know - but FWIW, UGreen do a USB-C digitizer with 192KHz sample rate and, of course, USB-C. It's not that USB 2.0's speed is the issue, it's the fact everyone's moving toward USB-C as a standard and while I don't expect USB2.0 to vanish any time soon, it's worthwhile keeping up with the trend.

I admire your work in EasyEDA. I've never managed to get much further than swearing at it. A LOT. KiCAD 8.x series that I'm using now is very good indeed and there are extensions so you can just hit a button in the PCB design and it does all the work for you (gerbers, BOM, correct POS files). It has a simulator too which can be handy but it's a bit short one models due to the Open Source nature.

@marcdraco

wow, a lot of good info.
Yes, I did not read 30 pages before I posted, and I haven't touched the mic project in 3+ years.
To be fair, watching the DYI Perks video, I just seen to many things I did not like, I wrote up some of the more glaring issues and sent it to him, zero response. 
Some time went by and I started playing with easyEDA, and made my first small pcb's for another project and had some fun, then I thought try something a little bigger, so I researched a little more, updated his schematic and layed out a board, I did it just for a exercise/learning experience and possibly make it a bit better. Also I sent it to him and same, zero response.

I just tossed the info on my rcgroups blog and forgot about it, until today I found electronoob made his mic video

and he copied a lot of the same issues and his video showed this forum, so I figured I would plop my crap over here, and wow, great feedback, thanks marcdraco! 

I would like to see the +/-15v rails come up with that much cap loading them.

Regarding the 22uF output cap, for the relatively small vocal range, is x7r really a problem? 

Thanks again!

You're more than welcome - I've got a lot of hours ahead of you that's why I'm further on (and that's ALL it is). I laugh because I come from the days when THT ruled and designing for SMD took some serious adjusting, I can tell you. Then cramming everything from a regulator to a matched impedance output (based on Wutke's 1950's original) was a hell of an experience.

My own experience of X5R add X7R hasn't been all that bad in all fairness. The microphonics is the bigger issue than the the capacitance flying off the reservation.The problem is that each of the two blocking capacitors from the 2K2 source and drain loads experience different voltages in a more "normal" configuration - single rail but the real PITA is the overload potential as these mics can exceed 100mV without really screaming at them. When I say normal, I mean a common source config with a 2K2 load resistor (and a capacitor). A better way is to use a common drain config (source follower) which is pretty much impossible to overload - at least with a capsule anyway. See the "Linkwitz mod" for details. An awful lot of low-cost JVC electrets have died in the pursuit of that particular one as it means cutting a trace on the back of the capsule. I've killed a couple myself trying to do it.

I've made some astonishing cockups (which I'd rather admit than pretend I'm smart enough to do it one go) everything from trying to run a class AB amp without a negative rail (stoopid boy) to dropping a CJ431 in place of a TL431 only to find out the hard (and expensive way) that the bloody thing has a different footprint despite being otherwise identical and with a 1% regulation for a few cents. If there was a way to screw it up, trust me, I've done it! 🙂 Most recently forgetting to swap out the 100R source load (for an LSK170 design) with a more reasonable value >1K for a 2SK208. Poor little thing is so weedy ... And on P48 which is what Varee was primarily designed for, you only have a 3-4mA to power everything from the regulator to the output stages.

Pretty much everything of any note is on my GitHub (github.com/marcdraco) so you can help yourself to pre-designed backplates for 25mm and 34mm capsules. The current "Pepper" (from pepperpot due the holes) is 25mm only but it's only because I had to go to bed or the cat was going to savage me.

That said, it's better to be open (all the stuff is open source) so other people don't step into the same bear traps. But those mistakes are the mother of invention. I just finishing a simple, dirt cheap (even a JLC prices) capsule back with the JFET mounted internally and then screened with plenty of copper fill. It's dotted with 2mm holes as breathers and exposes all three of the JFET terminals so we can hook that to more "adjustable" circuitry. I don't think this will work as well with a through hole TO-92 due to the small space at the back of the capsule. An SMD part does of course. Price on these is "just few bob" even in small numbers so I'll probably give it a go. It just means bending the wires really, really tight (without snapping them). "Please sir, may I have another JFET...)

With a FINE soldering tip anyone of modest skill (well below yours) should be able to solder the transistor and a whole range of JFETs can be used because it's not tied to a voltage range, which means ... shotguns and more are possible. Higher voltages and lower value load resistors make for less noise. JFET noise is mostly Johnson-Nyquist noise in the channel but there's also a considerable chunk coming from those 2k2 load resistors. Standard for a cracker-box electret. Lower is better and locking the FET current in a feedback loop (Varee2) is better still but - there's a gotcha. It won't work at low voltages - V2 works at 9.6V approximately on a 48V rail - not nearly enough to bring (say) and LSK170 into it's best performance.

Matt's not a big talker. We've become "work colleagues" partly because we only live about 50 miles apart (!) and partly because we share an interest in audio (specifically specialist microphones). I did a ribbon mic a few years back - 3D printed a corrugator and the housing but (unlike mein host) I'm not great at physical design. Plus the ribbon foil is so ludicrously thin, you have to pretty much work with a mask on, close and muffle all the doors and windows (I'm not even kidding) a 2mm motor ribbon is so light and so fragile, a mouse fart could blow it off the table. How the hell the old lads did it... it's beyond me. I'm going to revisit it some time later this year but we're looking at some really sweet new options based on the improvements.

I agree the original design did leave a lot to be desired but I know (from trying) that the blasted thing is super sensitive to any sort of crap on the power lines - including bleedthru from the digital switching inside a PC. For reasons I'll admit I don't fully understand, without the large caps (using the 47u max from the Murata datasheet) just left it screaming and me swearing so loud the neighbours thought I was about to murder someone. 🙂

I did hamper myself somewhat by trying to remain faithful as possible to the original. (I would have stuck with a more capable, lower voltage doubler like the NMA0505 because it's got more bang for your circuit (damn things are expensive though). I've also experimented with a fully isolated input and output which requires two boost converters but isolates PC ground from -- well, everything. It even outputs almost the full professional level balanced line output too which is nice. Rather over the top really, especially as it can be configured with links to run as +15 0 +15 (more like 14 volts after the capacitor multiplier drop) and +15 - 0 - 15 as Matt's original circuit. Almost all SMD with diode protections (see The Phantom Menace referred on on the THAT corp datasheet) and naturally the same input stage. Varee 3 is due back in a week or so with the incremental improvements so it's easier to configure. Looks nicer to with the new "grille".

Then there are Jamie and Jamie "Quicksilver". Both are single-rail, low-impedance (10R) output which swamps any mains crap and should interface quite well with the my board or yours without too much hassle. Although both are intended to work from 5V USB. Ultimately this is what most people will be more comfortable with, simply because it's way cheaper (no boosters) and it has a 20dB class A amp for a bit of extra sensitivity. The non-Quicksilver Jaime uses a simple source follower and precious little else so the distortion is well down.

Mind you, designing a reliable gain amplifier with just a handful of parts is ... challenging. With op-amps you know the open loop gain is mad so you can tame that to within a whisker of the desired gain with a simple voltage divider. Problem is they can't drive their way through a wet paper bag and the decent ones are... special order (again) if you want SMD.

And if you want double-sided ... that runs into serious money. You and I could acheive a lot more with a double-sided board but it does push the costs up - same with using the really, really small parts like 0203. Man those things make a the sharp end of a pin look big! Hence I've limited myself to 2 or 4 layer (4 is easier, naturally) and a single-side assembly.

Another little trick I picked up is a guard trace running around the Gate from the Source. That creates a field at the Source potential (and the same polarity) which blocks a lot of the magnetic fields running around the board messing with your nice clean signal. I should say this is really most effective with a source follower so I haven't included a loop on the Pepper as I didn't see much potential for improvement.

Beyond this level of tinkering there's the dual-monolithics (389 and 489) but watch out for that gate capacitance. This gives you either nice constant current cascode (so no Miller effect) or better yet, use a bipolar source and parallel them up for a significant drop in noise. You can do this with discretes but they need to be very, very well matched or one of them will take over and spoil the party for everyone.

I've done the layout and it can be done (just) but for a low-noise, balanced output that means going double-sided again (assuming everything is done at the head as it were). So as you might have guessed, my next iteration is to do just that on a Pepper and move everything else onto a separate board were we can have film caps in the signal path and, power using normal USB power we'd got current to spare - so no problem tweaking the bias and even a better gate bias circuit (self-bias only gets you so far). With those modifications I believe we can better the noise performance of the Rode NT1a. Hard to be 100% sure of course, and as Doug Ford (former head designer at Rode) points out we're designing a microphone that sounds nice, not one that necessarily has a very flat response. Many singers seem to prefer a little bit of peaking around 6K.

I don't know if you got a chance to look out the Fetlzer but that produces those even harmonics as it soft-clips which sound pleasant to our ears. I'm told (my hearing ain't what it used to be from standing too close to huge banks of very powerful speaker cabs in my "yoof") that this is why CD audio sounds cold - it's accurate. Vinyl (records) and valves in particular technically distort far more than a CD and decent transistorised system, especially Class A. But that purity comes at a price of sounding "cold". Human hearing is a result of millions of year of evolution and we're not machines. Sound purity is highly subjective. Transistor clipping is hash (lots and lots of odd harmonics) used by rock musicians as fuzz. Legend has it that the effect was discovered by chance when a transistor popped on stage - Hendrix is often pointed to as the "offender" and ... here we are!

The compliance of the moving head is an issue, certainly. Rode uses a rubber pin mount as I recall to dampen the vibrations. That's certainly better than springs because springs are, well, springy and they don't absorb a lot of the buffeting experienced on stage. If you have an ideas of how to make a decent mount, I'm (pardon the pun) all ears.

Speaking of compliance, I while Matt and I were chatting it dawned on me that the reason acoustic suspension speaker cabinets sound lovely and warm at higher volume is due to the soft clipping because the cone cannot reach it's design limit as it's restricted by the pressure inside the cab. Not musically accurate as such, but they sound great because of that simple trick.

I suspect we could have it 3D printed at JLC so long as they have some sort of elastomer (I don't). All these things go a little further down the rabbit hole. I'm sure you're aware, budget (and manufacturability) are key here. As I said to Matt earlier today, we *could* (you, me, Matt) safely make a pure valve-based mic without killing ourselves  - the same cannot be said of the general public sadly because 180-300V cathode bias really makes your teeth wobble. Not to mention the cost of the blighters - even the "cheap" ones are usually new old stock. Hence the Feltzer is a really sweet way to get the valve sound at respectable voltages. I know this from experience (as a tween, dozy harris grabbed hold of an old TV chassis - which sits a 240V mains potential. About the same time I got kicked in the chest by the biggest invisible horse I didn't see and ended up about six feet across the room. Which is better than being six-feet under, naturally, and that could have easily been the case. So no poking fingers into mains equipment. My lecturers (back in the early 80s) made damn sure that none of us made that sort of mistake. Safety first, safety always. Everything from isolation transformers to rubber souls... or better yet, stick the right side of 50V and you're pretty safe.

And I've prattled on quite enough - the sun will be up at this rate. Please feel free to fork my stuff and fix/improve anything with the understanding that much of it remains "experimental" - that is. it works in LTSpice (or PSpice if I'm not using op amps) but simulations are just that. I only rely on them to check my own maths and hopefully that I've got everything wired right. It's amazingly simple to do something "doop" by swapping out a component at the last minute or (cough) having to renumber a schematic - which often screws up the order of dual and quad op- amps and them put them back in the circuit BUT forget to check which way is "up". Got that tee-shirt too. :/

Hi, @marcdraco ,

Thank you very much for your amazing contribution to this thread, I’ve been reading it and decided to go with your Varee ver2 instead of the original design in the video because I have a P48 external soundcard. I ordered the capsule and PCB and got all the components for soldering. However, I have almost zero competence when it comes to theory and schemes, so please kindly allow me to ask you some dumb questions.

1. So the main question is, do I need to solder a 2N4416 fet to the capsule (I’ll use JLI-2555BXZ3-GP) before soldering it to the PCB? Looking at the scheme, I have a strong feeling that Q1 2sk208 does that job, so I do not need a separate fet between the capsule and PCB, but just in case want to check that.

2. The connection of the capsule to PCB is basically connecting (soldering?) the central screw of the capsule to the central mounting hole and soldering two ground plates to left and right GND holes directly, without wires? And the cable to the soundcard — it goes from the back side of the PCB hot and cold, and the shield connects to the ground ring on the back, right? I will attach a picture I saw a couple of pages earlier, which seems to be the right way.

3. Should I solder H1-4 mounting pins to the capsule itself? I have a feeling that if I leave then just touching the capsule, it would be quite flimsy… Or I can get them in plastic and glue them to the capsule to avoid electric contact.

4. I am going to connect the mic to the P48v sound card, so as far as I understood, I just need to not install the Ri (R14) resistor, right?

5. I will put the capsule with PCB into a round Faraday cage (made of brass mash). Since the PCB is bigger than the capsule, should l solder connect the ground of PCB to the cage?

6. And the last one — I was thinking about installing a button in the break of xlr cable to be able to mute the mic somewhere before the soundcard. Is there any possibility that this will worsen the sound?

Thank you again and sorry for so many questions!

Hi Ksardas , and thank you for your kind words.

Slightly awkward point here is Varee (while basically complete) is still experimental and there are a couple of things that I've had to alter.

I've refrained from publishing the final boards (although the GitHub ones are almost up to date). The main change is to the way the JFET is set up. I'll PM you some extra information regarding this,

OK, to your questions.

1. Varee (when assembled by JLC) - you can get a just 2 made and three "bare" boards quite cheaply if you use the slow-boat shipping. The board itself is complete and working when you get it. The 2N4416 is out of production now and not that easy to find. It's also a through-hole part, the SMD 2SK208 (I forget the SMD number) is in production and costs pennies (plus a couple of quid/dollars) extra on JLCPCB as it's not a "standard" part. The current version is V3 (I've remove the gain stage) and tightened the spacing for the mounting. While that version works, it's quiet due to a 100R resistor in on the BOM which should have been 1K - 2K2 for the 2SK208 due to it's feeble Gm. The latest version also has an option to run with P12 and potentially P24 too. I'm eternally grateful to the volunteers for testing and reporting back!
2. Connecting the board is simply a nut through the centre screw (you can solder it at your option) and solder one or both of the tabs to the ground connections at the right and left. The ground can be connected to any of the free copper (it's tinned when it comes from JLC). Varee also mounts to the larger 34 mm capsules which accounts for the larger diameter. (Also gave me a little more room to improve the circuitry.) This design (even with smaller output drive transistor just won't fit on a single-sided board and double-sided assembly is expensive. If there's any call for that, I'll design the option when the full series of "V2" adaptors is done. Varee R3 has a jumper to select Ri for P12 or P48. (It's the current limiting resistor for the regulator.)
3. The  four holes for "pins" are intended for suspension using rubber bands, springs etc. to reduce vibrations (hence springs are a poor option since they don't absorb the energy like a softer material). The top section can be clipped and removed if not required (I've re-designed them slightly so they are easier to clip off. It's there so you can "hang" the capsule adaptor inside a donor body from Neewer - the BM800 and similar bodies. It's not universal since I don't have the alternate spacing for the mounting screws.
4. There are a couple of new designs in the pipe which address the problems (including sourcing parts) on the original but they are still in development (see my GitHub page). They're not P48/P12 compatible so not relevant to your application. P48 like Varee was designed as much for professional use with a single supply as originated back in the heyday of powered capsules. The original circuit is due to Schoepes from the 1960s! (If it ain't broke, don't fix it!) Varee will, if wired correctly also drive an unbiased capsule but I'll leave details for that for another post, likely the detailed instructions of how it works as it's possible to blow everything up if done wrong!
5. Your Faraday cage can be soldered to the outer ring (It's there for soldering a brass/copper ring). It's not 100% necessary to solder the cage to ground (it can "float") it simply needs to surround the "high-impedance" section which is the capsule and rear electronics. High impedance inputs allow very low level signals floating around to impress a signal on the capsule. You can experiment with this if you have a small resistor around 100 ohms or less. Just touching the hot/cold input with a pinkie is enough to make your amplifier "humm" as it's driving a 6K8 resistor which is medium impedance. Shorting that to ground with a few hundred ohms is enough to "short" that to ground so that interference just gets shorted.

   Here's the background (you can skip this but as a neurodivergent old ****, I love to clear this stuff up for the rest of us). My teachers, who missed some of what I'm about to reveal made this very difficult to understand. It was one of those things you just use.

   The reason is this down to Ohm's law. A resistor connected to voltage at one end has the same voltage at the other end - which sounds  bizarre but it's true, because no current is flowing. Here's the circuit modelled in LTSpice.

   

   How is that possible? OK, this is complicated but voltage is more correctly called potential difference it's a mathematical idea that makes everything work. The word difference means at two different points, but (and this is the key point) there's no current flowing in this circuit. We can better think of voltage as being a form of energy. Mathematically (and in reality) we should ALWAYS model a voltage as an energy source in series with a resistor. High school physics often uses this model without explaining why we do it which is confusing, especially when we're using something like a battery because the "internal resistance" of a battery increases as it ages. In a battery this is because as a battery runs down, it's able to supply less and less current. We've all seen something like this when the bulb/led in a torch runs out of gas.

   This all seems very strange but you can verify what I've set up here with a modern multimeter Any old resistor from your workbench connected like this will measure the battery voltage at both terminals even if one isn't connected.

   If that doesn't seem to make a lick of sense, it will when you realise that energy (the thing that actually MOVES the electrons around in our equipment) is an electromagnetic (EM) field - and the field is in the air surrounding the conductors. Again, this sounds crazy, but it's exactly how radio works! Grasping this non-intuitive idea is key to understanding how circuits work at a very low level and allows you to design noise-rejecting circuits.

@marcdraco Thank you very much for reply!

Oh, now when I've checked your github again I've noticed that totally missed Varee v3 and went with ordering v2. Anyway, the PCBs are already printed and will arrive in a week or so... Would it be that big of a downgrade (I only care about the sound quality) to proceed with v2?

Regarding the 100ohm resistor (R2 I suppose) which makes the scheme too quite -- what about I just replace it with 1k or 2.2k? (and which one is better?)

And thank you also for the explanation on the interferences and Faraday cage, that’s very helpful!

P.S. regarding PM -- it says that I dont have enough posts to write ones, so I am not sure if I am able to receive PMs as well.

You can't reply, but you can read PM. I got logged out (damn you Google) and lost the end of the post but there's a long segue into how this all works and why we use the values and components necessary. I'll post it soon.

You have  a couple of options for R2 - (source resistor) 1K - 3k3 will give better performance. The 100R is perfect for the LSK170 buuuuttt...it's a high-gain JFET (gm around 14 mS) whereas the '208 is a rather modest gm of 1.2 so the gm multiplier (R1/Q2) simply won't work correctly. A workable fix is to leave Q2 out, short R1 (680R), and bump R2 to 2K2 (or something near like 1k8 to 2K5)

This reconfigures the '208 into a simpler, and perfectly acceptable, source follower with an increase in output around 4.5 to 5x. Source followers (like emitter followers) are "lossy" so in reality the mod just improves its performance by that factor, but in effect it also raises the overall output by the same amount which amounts to >10dB increase - more than enough to make an audible difference.

For greatly improved performance (albeit at increased cost) you could keep everything as is and replace the JFET with the LSK170 (in SOT-23 format - it could (in theory) be done with the TO-92 version but soldering a through-hole part on an SMD board is fraught with difficulties. I'll hold my hands up and admit this was entirely my cock-up for not checking the '208's gm... Unlike bipolar transistors which have a gain (hfe) around 100-300 for small-signal devices (so they are largely interchangeable in less demanding circuits), JFETs are very weird animals and have to be treated with rather more care. But worse, the choice of JFETs in current production is dropping and the number suitable for condenser mics has almost dwindled to a handful of usable parts.

Dimitri Danyuk, a researcher from the Ukraine, has used some relatively simple rules-of-thumb to create a circuit that behaves more like a valve/tube with that lovely warm sound desired by musicians. This is the subject of one of the new designs - the experimental version is on GitHub but it's not even been prototyped yet. Great things are in the pipe!

OK, I'm feeling a bit like I need to slap myself with a frozen haddock. Here's the recalculated (and simulation verified) transconductance raised sufficiently to create a follower with a performance similar to that of a BJT but with input impedance of a JFET.

This will produce a very sensitive microphone with excellent linearity by simply swapping out two of the resistors.

If you find the output levels are too high for your gear, you can increase the value of R6 all the way to 3k9 (where the transistor operates as a unity gain phase splitter.

@marcdraco Thank you very much, this is awesome!

I was actually thinking about replacing SK208 with LSK170 (its not that hard to find in Japan where I live), but if this problem can be solved by just changing resistors, that's would be much better. Thanks again!

There is one more point I wanted to check: when I was thinking about swapping sk208 to lsk170, i've noticed that their source and drain pins are mirrored (see attached screenshots from data sheets). Then I checked the PCB editor (see screenshot no.3) and it seems like Q1, which should be SK208, is actually pinned for LSK170 (if the Ground faces up, the Drain is on the left), when in case of SK208 when the Ground faces up, the Source is on the left. Or am I just reading the datasheets or PCB wrong?

Japan! Wow - you lucky duck! What a beautiful place to live (I have an interesting "anecdote" about Tokyo - specifically a small place just outside in the mountains) but that's for a PM because it''s my other hat. 😉

FETs (MOS and Junction) are weird little beasties. Really, really weird - that's what the BJT in the (revised) circuit does - it fixes many of their shortcomings. More *can* be done with this but not on such a tiny board.

One of the weird things (and one I love) is that FETs don't really care which end the source and drain are - the channel is ... well, a channel. You can (but you shouldn't) think of it a voltage controlled resistor. That's good enough for this discussion because the upshot of this is you can nominate either end of the channel as source and the other as drain. The one that matters is the gate because (in a JFET) that's what you have to supply some reverse current to. Since the JFET is a channel with a pin "stuck in" half way through, all the magic happens around the gate terminal. There are a few exceptions, but in general, FETs like the 2SK208 and LSK170 (J201, etc...) don't give a fig which way around the channel is wired.

The transconductance (gm) amplifier used in a follower like this is so efficient that it can (almost) turn our $0.02c JFET into a $5.00 FET at least where the follower's ability to, well, follow is concerned. The LSK170 is less noisy but experimentally, I haven't been able to notice any real difference. On paper, the 170 is certainly the better part though - and it can work without fancy gm amplifiers - hence my earlier cockup.

Distortion (technically) looks high on paper but it's almost all 2nd harmonic which is that warm sound you get with a valve. Oddly using a negative supply rail reduces distortion by an order of magnitude which is... something I need to swot up on because I can't figure out why. It could be something in Spice but I've tried a couple of different good quality simulators with different SPICE engines and got the same results.

The fact that JFETs can work either way around makes them awesome for voltage controlled volume and soft muting to name a couple of examples where this property shines - try and do that with a BJT and you're in for a world of pain and a LOT of parts.

I've just finished the basic circuit for the Jaime "Quicksilver" V3 with the same basic gm amp. but followed by a very high-quality audio amplifier and if it works on paper as well as it does in simulation, it's "bye-bye" to that THAT1512 and a lot of other circuitry - at least for cable runs of a metre or so. Long runs absolutely benefit from the fully-differential outputs of Varee 2 and 3, but I've (partly) developed a circuit based on Caprio's Quad which performs on a par with the THAT1512 and should cost less because the transistors are surprisingly low-cost and easily available. Good old mass-production!

Right, back to the grind, the cat is getting antsy again! 🙂

@marcdraco Got it! Sorry for false alarm then 😀

Then I will stick to the scheme you posted earlier with changing the resistors. I think I will receive capsule, pcb and components next week, so I hope I can update on the project soon.

Thanks again for great support!

No problem, getting caught out with footprints cost me dear more than once. I thiink I detailed that one in this thread by swapping out a TL431 for a CJ431. TL and CJ are maker codes (I used to know all the major ones, these days...) But you could usually be sure that the same part number and function would have the same pinout.

Imagine my surprise then... 😳 

Hello everyone!

So I've built the microphone like in the video with a couple of changes:

1. I used a cheaper microphone capsule from aliexpress

2. I added the diodes to prevent audio digitalizer damage

3. I left the type-a cable instead of making a type-c connector

4. I used a single resistor (10 ohm) instead of a rotary switch with variable resistance

So yesterday I completely finished the microphone and tried to test it. There was no sound until I flipped the THAT chip 180 degrees (initially it was installed as in the scheme). Then I added shielding and tried again and the mic sounded loud and almost perfect. But after some time (and some manipulations, I changed the gain resistance, it was 5 ohm during the successful tests) when I tested it again, there was no signal. Not even noise. I triple checked everything following the methodes described in the forum above, and the only thing that was not normal is that the voltage on the mic input is negative (around -4V). But the mic- is 0.7V more negative than mic+. I'm pretty sure, that mic- is connected to the source and mic+ to the drain (as shown in the video).

I have two assumptions on what went wrong here: 1 - something's wrong with the THAT chip. It's from aliexpress after all, althoug I have two of them (from one order) and neither works in any orientation anymore. 2 - I somehow damaged the transistor/the capsule. I have a replacement for the transistor, but for the capsule I'd have to wait until a new one arrives.

These are purely random guesses though, cause I'm an amateur and don't really understand what's going on here xD I would appreciate any helpful advice since I'm stuck at this point and need someone more experienced to help me

Here are some pics of the mic

My word, those are beefy resistors! 🙂 I haven't used anything that large since they stopped making them with the yellow background (the blue has jiggered my ability to read values by sight now due to my colour vision being off a bit. Thanks folks!

Anyway.

1. Capsule is fine - they're not as good as the JLI255 (by any stretch) but they are more than good enough for simple voice work. Oddly enough, they're quite a bit more sensitive than the JLI (quality comes at a price, you trade a little sensitivity for it).
2. The FET and the capsule are (almost) bullet-proof. (I know a song about that -
3. The THAT, on the other hand is egg-fragile and there's a chance that reversing it has sent it to silicon heaven.

All that said. Let's see what's going on.

First off, remove the THAT - it's pointless having it in if it's dead as it will throw all the measurements off, the output from the NMA0515 is weedy (33 mA per leg) and if something is loading it down, it's just going to take its ball and go home with a long face. Fortunately, it's also quite a tough little chap.

When you've done that connect your capsule up as normal and measure the voltage from GND (black lead to 0V) to each leg of the JFET - you might find it easier to do this where the mic comes in (particularly if you've screened it as Matt demonstrates. It's that beautiful design that triggered me into making it better.

FETs are unipolar - the channel (that's the bit between the source and drain) is like a very large resistor until the gate is connected to a ground via a resistor. Most of us get away without the resistor as leakage currents allow it to work, so while technically the resistor is mandator, you can get away with it. (My designs absolutely won't work without something because I've gone to some lengths to stop leakage messing everything up).

You should have a voltage around +14 to 15V on the drain (source if you have it backwards, but that's OK actually, JFETs are dead forgiving. The source (or drain) should be -14 to 15V. I would normally do this with the capsule out of circuit first to make sure the board is supplying the correct voltage to drive the JFET.

The other thing to try (unlikely you have this wrong) is to check you did trim the right leg of that JFET. Matt picked it as it's a quality part but it's really aimed at RF stuff so that extra pin assists like a Faraday cage around the actual device.

I can only apologise that the V2 isn't out yet - that's entirely my fault and while it's quite mature now, poor Matt hasn't got the final parts to re-make the project with the improved designs. (I did some hand-waving to get the parts "close" using a different JFET and it worked beautifully until I fitted a low-cost and easily available device - which broke everything for reasons that are a bit complex to explain, but it's down to the JFET's gain - even operating well, the 2SK208 can only manage a feeble gain of about 1.2 mS, compared to the much beefier LSK170 with a gm of 14 or better. So no amount of "close enough" works because it just sits there and drops the mic's output to about 1/10th of what it should be (and it's weak to start out with).

So, as the saying goes, I slapped myself senseless with a mackerel and went rummaging through my old textbooks for a solution (and a reminder of the equations...) the results are - on paper/simulation, quite remarkable compared with the original design (which considerably better than the usual "1v5 through a 2K2 resistor, in common source configuration that the capsule data sheets suggest.

And by that I mean orders of magnitude better - but the best bit is (when we find a distributor) they're not even that expensive. Being open source you will be able to order the one that best suits your needs and they're (relatively) easy to construct with most of the grunt work being done by the capsule adaptor which fits inside a modified version of the same head design. (If you wanted to go that route and work from my drawings, give it a few days until I've confirmed they work as advertised, and grab yourself a couple from JLC.

As you've already got the bits though, let's see if we can get you up and running. You can always upgrade at a later date - the last thing you need to be doing is throwing more money at the problem and getting nowhere! No one needs to be laying out the development costs - any cock-ups are mine alone and if it costs me £100s (and it has) that's on me too. The PCBs are a lot easier to work with than Vero/stripboard too with each part clearly marked on the silkscreen so even with my eyes, I can put one together in under an hour. And they are guaranteed to work (assuming they are made according to the design).

@marcdraco I measured he voltage with the capsule connected and it's +0.7V on the drain and -0.8 on the source. I'll have to desolder it to measure without it, can't do this at moment. So could this be the problem?

Unquestionably. [Edit - err... clearly someone had too much hooch... see the corrected post below]

That's so far out, it's half way to the mooooon :). I'm neurodivergent or madder than a box of frogs on acid according to some, so I hope my sense of humour isn't lost. Trust me, I've made more mistakes than most by this point so don't think for a moment I'm infallible.

BTW, welcome to our little community.

If you check Matt's original circuit you'll note that the +15 and -15 (nominal) voltage supplies are fed to the source and drain via 2k2 resistors.

A few posts earlier I mentioned that a "dangling" resistor (with no current flowing) has the same voltage at either end. You can check this with your meter if you doubt it (when you do the measurement a tiny current does flow but it's so small you'll see this effect).

If something at the capsule end is off (shorted, JFET wired wrong, etc.) that could easily "break everything". The advantage of phantom power arrangements like this is they are very current limited so it's unlikely to break anything. The maths come out to 6 mA per resistor (without the FET in the way) so there's insufficient current to damage anything.

With that sort of voltage dropped across the JFET, (1.5 volts) not much is going to happen.

I think (and this is off the top of my head) that I'd be checking the JFET is wired correctly. There's (effectively) a little diode junction in there that's supposed to be "reverse biased" but if you forward bias it by more than about 0.6V (obviously not much!) it will conduct and the JFET will saturate. At a guess (and that's all it is, this isn't me making a funny) you may have wired the JFET incorrectly. I'll have to check the exact part to see where that "spare" pin ends up. (But it's XBOX and chill time so it might be tomorrow before I can get back to you.)

I'll have a look in later tonight and see if you're any further on. If you have any JFETs lying around (probably not) you can run the tests with a few, but realistically that's a long shot - and you'd ideally need something like a J201 or the LSK170 to have a decent shot. Failing that, there's a design on my GitHub that mounts to the back of the capsule and ONLY has three connections out so it's ready to go. Cheap to - IF you are up for soldering an SMD J201 or similar. It's easily modified for a through-hole part so let me know if you fancy that route. It's a LOT less painful I can assure you - assuming the issue is at that end, of course.

1. I am pretty sure the jfet is a depletion style fet, in other words it's always on without a signal.so the voltage at the drain and source is close to ground without a signal.

Of course (thanks @oz, you are quite correct) - I was clearly having a senior moment 😳 (confusing my depletion with enhancement, and it's not like I don't know the difference!). Looks like I'll be defrosting the haddock again or failing that, I'll have my ex. wife beat me with a copy of The Art of Electronics 3rd Edition.

Here's a very crude simulation of what you have there.

The JFET is in full conduction [almost] at this point. Those voltages are within tolerance (I don''t have a 2N4416 to check it).
Definitely sounds like the THAT is the issue in "that" case. You can check your board's voltages against that earlier post (THAT removed) to see if it's receiving the correct voltages.

But I did manage to put a board together for people who want to use this particular animal.I'll pop it on GitHub for folks to try out, these designs are all deprecated now the new ones are in production (and that THAT has been put out to grass due to all the dubious ones floating around) but this will serve for anyone who wants to use the original transistor. Gerbers attached.

Sorry I haven't answered for a couple of days. As I said in the first post, I've tested the voltages without the module and they are in order. I guess I'll order a new one from ebay, since the price is not that much higher than aliexpress and the delivery is faster. Thanks for the help!

No sweat. Happy to help.

Hello, @marcdraco !

I’ve received all the components and soldered them on pcb with swapping resistors as you said above and everything works awesome! The gain is great too…. couldn’t be happier! Thank you very much!!

There is one thing I’ve noticed when already assembled the mic… I was too bound up in soldering and totally forgot to skip the Ri (R14) as it was said for 48v operation. (and I plug the mic in p48 preamp) The fact is that the mic works great, but I have no idea about possible long-standing consequences (can something burn?), so I am not sure whether I need to disassemble it to replace R14, or just leave it as it is?? Sorry for a little stupid question..

Great news, I'm delighted and it's not a stupid question, it's perfectly reasonable to ask so never feel afraid to ask me.

It's quite a challenging board so my hat is off to you! I expect most people will buy them pre-assembled. I get all mine done at JLC so there are no mistakes in the BOM (I caught one yesterday after I'd sent it over, so that's going to be a bit of a red-faced conversation with JLC support (who are excellent)! But this way anyone who buys one when they are officially released is guaranteed to get a working one per design and QC.

The rest of this post discusses the theory and takes a deep dive, so you can skip any bits that don't make sense.

Ri is the current limit for the TL0431 shunt regulator.

Given the amount of current limiting in the P48 system it  won't hurt, it's just putting more load than is strictly necessary for P48 operation. The pre-amp already has a 6K8 resistor in series with each Hot and Cold so the total current draw is limited to about 14 mA (seven per leg). Varee itself takes about 4mA and even flat out, the TL431 is only dumping 10 mA which is 10% of its max rated draw.

Shunt regulators work by forming a "voltage sensitive" voltage divider with a resistor in series. Here's a block diagram without voltage setting components for anyone who is learning this stuff. This applies to any shunt regulator from simple Zeners to devices like the 431.

The circuit on the left (ignoring the load) is a simple voltage divider which outputs 1/2 VDC at the output. If we put 10V DC in the top, 5V comes out at the output. Nothing too hard there. To work this using Ohm's law:

The total series resistance is 10K and we'll use 10V for ease.

5V / 10,000 ohms = 1mA

So the voltage across either resistor is:

5000 ohms at 1mA = 5V

According to the rules, Kirchoff's 1st says that 1mA must be flowing through both resistors. Kirchhoff always seems more difficult than it needs to be but it's just a case of what goes in, must come back out. Even if a circuit has millions of components, the amount of current going in is always the same as the current coming out. (Circuit comes from cycle after all, so the current is cycling.)

There are two problems with the circuit on the left - which has its  uses.

1. any variations in +VDC are instantly reflected at the output, so it's no use as a voltage regulator.
2. any current drawn by the load "steals" current from the lower resistor and here is the real issue. In effect, the load "looks" like a resistor in parallel with the lower resistor and since loads are measured in terms of fractions of an ohm through to a few 10s of ohms. Even without running the numbers for resistors in parallel, it's reasonable to see that a very small equivalent resistor in the bottom half of this circuit has an effect like this:

   5K in series with a load of 10R.

   10V = 5005R
   =1.99 mA

   So the voltage on the upper resistor is now:

   5000R * 2mA
   = 9.99V

   Leaving just 0.01 V for our load - which is clearly useless and this is why voltage dividers are terrible for voltage regulation.

Considering this, let's consider our "black box" in place of the lower resistor.

Whatever is in this box looks at its upper terminal and adjusts its internal resistance so that just the right amount current is drawn through the upper (current limiting) resistor to set Vout at the voltage we're interested in.

If the voltage at the input changes, the "resistance" of the black box changes at the same time.

Similarly, if the current drawn by the load changes (and it will for any active circuit) the black box varies its resistance to ensure the voltage drop across the upper resistor varies to maintain a constant voltage at the output.

In other words, the job of the upper resistor is to be able to pass sufficient current so that the load is always presented the design voltage, but that in lightly loaded conditions, it limits the current flowing through the black box.

In our case, the black box is a TL431 which can dump up to 100mA.

Varee (the load) draws about 4mA so the TL431 has to pass up to 96mA if the supply is capable of providing that, but I've  mentioned that the maximum P48 can supply is 14 mA, (the short circuit current) so it's very safe both for operators and equipment alike (except dynamic mics).

A regulator isn't absolutely needed for a simple P48 system since at 4mA draw the available voltage drops to ≈ 14V but more complex designs need a fixed voltage so the biasing works. Simple biasing in discrete circuits is most often performed using simple voltage dividers as above so the bus /supply voltage needs to be stable. Operational amplifiers, by contrast, don't have voltage regulators as such but can work from very wide supplies - often 5V to 30V and more. This is done using a centralised "current source" and lots of "current mirrors" all over the place that ensure the active devices. This is genius but it means a lot of extra - it's a trade-off between component count and design complexity.

Varee has simple current source limiting the current through the JFET as part of the transconductance amplifier - that's where I screwed up the values on the Varee V2 since the 2SK208 needs to run at a much lower current than the LSK170. The current is set by the resistor across the base-emitter junction of the PNP resistor (MMBT3906). We know that junction (silicone transistor) "drops" close to 0v7, so we can use ohm's law to set the JFET current. The latest versions have FET currents of a few 100 microamps (uA). I've done this to ensure any FET from the JLC part's bin can supply design current regardless. It's available in drain currents in a range from 300uA (max) to 6.5 mA (max) and although there are several rated versions, I've used the very lowest current (Id) to ensure it will work for any JFET in the 2SK208 series.

This is why I love the challenge of working with a JFET - and they're (almost) the only way to interface to a condenser microphone. Rather that being able to do some hand-waving (bipolar transistors are far more forgiving) it's necessary to get deep into the weeds. In order to get the distortion down even when the input voltage swing is so wide (the dynamic range of the microphone capsule) it's necessary to lock off the FET's current as tightly as possible.

There's a secondary effect, Gos, which affects JFETs at higher voltages (above 5 - 10V) and the combination pushes distortion levels on the original design into the single figures (THD) while these designs never pass the decimal point at the cost of (very slightly) increased noise. Although it's worth remembering that the capsule itself has distortion as high as a few percent THD (even amazing speakers distort sound to some degree but in both cases it's "soft" distortion so it's less noticeable to our ear. Electronic distortion, particularly in the bipolar stages can come on suddenly and hard (clipping) which produce a (theoretically) infinite amount of harmonics that sound horrible. The effect is often using in rock music and called "fuzz". While that's fine on an electric axe, a vocal mic needs to be as accurate as possible.

I'll mention here too (for anyone following) that I've just submitted some very low voltage designs to JLCPCB for testing. Matt's working on the rest of it but these things take time to get just right. What works in simulation often won't work in reality and there are things required on the PCB that aren't immediately obvious from the circuit.

The most nefarious little blighter (not the only one) is the high impedance point at the gate - in the order of 100 GIGAohms (more for a MOSFET).

Not only is this point highly sensitive to external interference, it's also affected by currents flowing through the traces nearby! One way to avoid this is to encase everything in a Faraday shield and keep a decent distance from supplies etc. Of course on a board this small, distance isn't an option.

So what's the issue?

Any current flowing through a wire produces a magnetic field (if you've read my earlier posts on this you'll recall that I've quoted "Hartley's Law" (my name) which states "the energy is in the fields".

When a changing magnetic field passes around a conductor a current flows in that conductor. Recall that electricity is generated using moving magnets and wires. You can show this at home with a few loops of copper wire, a neodymium supermagnet and your meter set to microamps. As the magnet moves you'll see the current register.

A little bit of current (a few millamps) moving through a little copper trace doesn't produce a very large magnetic field (Rick Hartley would smack me with the haddock for that. It's electro-magnetic field that causes the current, but enough of that because the fields move at the speed of light anyway so the distinction theoretical until we move into the radio frequencies and weird things like reflected power enter the fray).

So let's assume that this piffling amount of current produces a current in a "victim" trace (the gate terminal here) of a few picoamps, who cares?

Remember that input impedance at the gate and Ohm's law.

Even 1 picoampere on a the victim is 100mV which is more than the capsule is producing! At 0.001 pA (10 femtoamps) at 100G is fully 1mV and that should give you some idea of the scale of this problem!

The guard trace makes a low impedance loop (a few 1000ths of an ohm) which has a field fixed by the Source terminal. Since the Source "follows" (changes voltage at the same rate) the field in the guard trace isolates the high impedance from that interference elsewhere in the circuit. Currents caused by the magnetic field from the aggressor traces (power-supply variations and amplified signals which are at different phases from the input) are absorbed by that tiny impedance.

I don't know who figured this one out but it's genius whoever it was.

I hope the mic gives you many years of great service.

Hi! I’ve ordered the components for this device. While waiting for delivery, I realized that I forgot to order the "Audio Capture/Grabber Card Device - Analog Cassette to Digital MP3 Converter."

I wanted to ask if it’s possible to do without it and just use a regular 3.5mm output. If so, would I need to make any changes to the circuit?

Thanks in advance for the advice! The project is really cool, and I’d love to replicate it.

Big thanks to the author for their hard work!

The current pre-amp can output anything from 0V - muted - to about 13.5V (peak) so it can, in theory drive pretty much any analogue digitizer input. You have to be a little careful that you don't blow anything - there's a blocker elsewhere in the forum (just a couple of diodes and a small resistor) that will lock the output down to 1V, or a little less, which is a bit of a hack. (I know because 'twas I what did the deed after blowing a couple of digitiser boards).  The gotcha with many, but not all, digitisers is they expect to see a 5V peak signal (+/- 2.5V). It's a bit of a bind because they don't usually tell you! I've got some here that will take an input direct from a simple capsule, the ones you can pick up on eBay for a handful of salted peanuts. Such a simple setup only produces a few hundred mV - and a lot of distortion which is usually masked by the noise.

They actually sound fairly decent until you do a blind listening test and then the wheels come off.

I've tried it with a bunch of digitisers from a variety of manufacturers and they all work provided you don't push them into distortion. The active limit, while safe, does reduce the volume to about 20% of max before it starts to clip. This is by design but it''s not to everyone's taste.

You can get the Audiograbber, the original Matt used, for around £20/$25 from Amazon depending on where you live and its generally the best option having a very quiet input stage, especially at the price point. If you have any other questions feel free to give me a prod. This thing is perhaps one of Matt' most accessible and beautiful designs and it works like a charm, as beautiful as it looks.

But watch this space! I promise everyone it is worth the wait. Latest public developments are at beta stage and not really ready for the big time just yet. I don't encourage anyone to experiment with them just yet, except perhaps the two "breadboards" which I use myself. Lunar New Year has delayed the prototypes by a couple of weeks as the board house shut down.