DIY Micrphone

Woah! Made the beautiful build even more beautiful. Fantastic work. I really must pull my finger out and get the new pre-amps out to the PCB house.

@marcdraco Make sure the pre amps can fit in this enclosure. Lol

Stunning reconstruction! That's got to be the best by-the-letter project build I've seen. Well done! 😀 Mind if I Tweet it?

@diyperks Thank you Matt, that's really good to hear! Go right on ahead. Appreciated!

@midnight_vanisher That was a major consideration. The origmain board is within a mm or so of Matt's vero layout but I can get a lot more in the same space using surface mounted parts, including high/low pass filters, a headphone amp and a new volume control which is less twitchty that the THAT, and more power filtering.

I'm just re-doing the head unit with a three-wire design to retain compatibility with Matt's head unit. I've done three versions now using different FETs and lower noise design. On paper that is. I should have the first batches of the rev3 boards back in a couple of weeks.

The current batch have an extra power in but I figured this batch would be better using a sort of "phantom" power but that required a couple of unusual tricks and a different output stage - whic h is this. It only sends a single phase but the impedance is balanced so we still get excellent noise rejection.

This reduces the sensitivity by 6dB but these capsules are crazy sensitive anyway.

@marcdraco Excited to hear of this V2. Curious how it'll all look after making this board especially hearing all this spaceman language.

The main thread has some of my ideas rendered in 3D. I've actually done a small batch of "Cerberus" heads which was my original design using  pair of discrete FETs (LSK170). Performance is excellent - I've just received a THAT1512 over the weekend so I might be able to rig a preamp to compare to Matt's original design.

It's broadly similar to what Matt did (well, it is NOW) to retain backward compatibilty but I've had a brainfart over the weekend and decided I was stuck in a rut over the head. So I've designed no less than three completely new ones, all using different tricks and technologies. The simplest one (and the one least likely to work as is) is a simplified version of the OPA-Alice designed to operate at split 15V supplies.

The other two are both discrete FET designs using eitheir LSK389s (dual monolithic, super-low noise JFET)  or a full-realised SMD version based on an original layout I did for the discrete LSK170s.

These are all experimental at the present time but I'll drop the full board designs when I know they work. I've published a lot of my earlier ideas and while most work well enough, none have really matched what I managed with the separate supply.

I'll also write the whole thing up so that every last volt and milliamp is explained so everyone call follow along - without all the blasted maths. I love electronics, but I hate maths (I'm rubbish at it) so I'm using building blocks that are reliable and easy to describe.

There are several more ways of doing this but these are sufficiently different as to give us all a better idea of what works well and (more important) what doesn't. It's been quite the journey over the last few months - sitting on my hands waiting for the prototypes to get back from China and then working out if I could improve anything (Ii could, I did).

Right now we're up agains the limit of what's possible with an electret such as the JLI2555 capsule. Even Matt's original far outperforms what that capsule is capable of in terms of noise floor. Getting beyond this (and again, I already have in the lab) means an entirely different design using a 48V supply.

Not difficut but not what Matt imagined for a desk mic. I'm thinking something more like the slightly weird look of the Russian-made Sputnik mics - with Matt's head mounted atop a tube which contains the rest of the main electronics.

This should be right up your street since you have a lathe of course. I'm limited to a 3D printer and I can't afford to run it given the price of power in Europe now... Besides, there's the screening issue too.

But here's a rendering of the final (prototype) preamp with the improved power, headphone amp and all the trimmings that you can't do with through-hole parts.

Here's a close up of a batch of the heads. The problem with this design is two-fold. Firstly, I'd use a pair of diodes in place of a 1G load resistor (which, it transpires isn't needed for an electret!) and also the pad spacing for the LSK170s is so tight, they are very hard to solder. The diodes can be removed with a soldering iron making the board noticably quieter and I have a few of these left available at cost which is about £3 a piece plus shipping which is large letter in the UK/Europe if anyone really wants to get the jump. But be aware that this design *requires* external power which Matt's board doesn't supply. This was my mistake since I built these "blind" without reference to Matt's original design.

My bad, it was Soyuz, not Sputnik - (nice segue from your spaceman comment). 😉

Regardless my feelings for current troubles, there's no doubt these designs are at worst unusual and at best, artistically beautiful. The large metal tube contains all the electronics and the capsule (it's dual diagphram I believe in this one) is mounted in the head. I believe it's a 34mm capsule too so that will give you an idea of scale. There are some videos on YouTube that show the ladies and gents making these and it's really quite awe-inspiring. As an electronic-dude, it was weird to see people still using "wire wrapping" in the 21st century but I can see the reasoning.

@marcdraco Reminds me of a the Neat Microphone, King Bee 2. Very affective design for the capsules and the casing.

@marcdraco That's very pleasing to hear when the results come out even more promising. People are going to be stoked! It sounds like its capable of doing the trick. The last one almost worked in certain variations, but its all getting it down to a T. We might have to build this next. Lol  😉

I  finally worked out how to (unreliably) join several boards in KiCAD. I'm unsure if this is a bug in the 3D side or the plotter, but as is, I can get four different ideas onto a single board that can be chopped up with a hacksaw. 

Only one of these will be for the Perks microphone (the best one) but the others are experimental. I'll have to wait on Matt to let me know which way he wants to go. The one with the chip socket ("Kryten") is designed to be upgradeable as better ICs become available. It requires just a dual BiFET or MOS op amp (TI have an extensive range) to work. 

Power is always the issue with these things given the limited amount of current that phantom systems are designed to deliver. P48 delivers about 11-15 mA at 48V for example. My version of Matt's pre-amp should give about 15mA at +/- 10V. It's hard to pin it down as the voltage drops off significantly as the circuit draws more power, but most modern ICs of this type have a fairly low consumption.

The Rev. 3 "Cerberus" (bottom left) has been modified to deliver a more balanced output but still with paralleled LSK170s for greatly reduced noise. This one also delivers a small amount of pre-amplification so it's the most sensitive of the bunch.

The "Hannah" (top left) is the real beauty though, delivering noise performance on a par with the best of the bunch or better. This version uses the LSK389 dual monolithic FET but doesn't require the extra circuitry the discrete FETs (Cerberus) do.

The one I'm really playing with is top right, codename "Hydra". That has an array of four low-noise FETs again in a paralleled arrangement. The main advantage of that one though is there's nothing to solder in close quarters. 

They all fit right on the back of the JLI capsule and clamp/solder to the ground pins for excellent screening although a Faraday cage is still required since the circuitry is slightly exposed.

I have (experimentally) designed one that should self-screen but I don't see that one as having any real world use and the performance is only "m'eh" since it's only able to use a single FET due to space restrictions.

In the main thread I've posted the final (re) designs of the above. I literally had them uploaded to JLC PCB and about to go into production when I realise there was a fatal error in the design. (My fault for relying on memory rather than double-checking in SPICE).

Well, I say "fatal" it wasn't fatal but it was sloppy on my part, so I've spent the last couple of days cleaning up that mess and added some completely new designs to the quad, while simplifying Hannah a little. The one with the Op Amp allows for a fairly large swathe of off-the-peg dual amps to be used. They have to be FET input and low power (i.e. about 5 mA quiescent per amp, or 10 mA for the whole package) and capable of running at up to +/-15V, although that drops off sharply as the circuit demands more current.

The other two new designs use 2SK208s which are ostensibly cheaper versions of the LSk170s only they are available from my board shop and can be fitted by them which saves everyone the hassle of that.

I've got them on a rush order since I've been on this project for several months and people probably think I'll never do anything, but that's the thing with development like this, the turnaround time is measured in many weeks so I'm often sitting on my hands or tinkering while I wait to see what works and what doesn't.

Then it's back to the drawing board, calculator and SPICE simulator. 🙂

They've gotten dramatically better as I've been able to test out different designs but waiting for the board shop is tedious.

Assuming Matt's OK with this, I'll open source the final designs (complete with silkscreen logos) so everyone can buy their own, pre-made at JLC or a board shop of their own choice. On paper at least, these are the most performant I've made. A lot depends on the +/-15V power which is something I won't claim to have much experience of. It's a bit like having P48 but with a lot less room to move. The only advantage (and bizarrely, I missed this one which was right under my nose) is that you have a negative supply which reduces component count since the main circuit can be DC coupled.

Fantastic work! 😉  I have one just like it. 

@Midnight_Vanisher My word it's been a long road... I've done several versions of the pre-amp in the end, but all of them (except the educational version) should fit quite comfortably.

In all honesty, I've been so busy I haven't actually checked it for fit, I should probably donate you one so you can have a play.

I've just completed (waiting for prototypes) a variation with beefed up power supplies that might (or might not) work. They all work on paper/simulation but there are various gotchas that have troubled me so this time I've had crack at them.

I did add a super-accurate "Baxandall" volume control to the last one but there are issues with the large (2200 uF) capacitors that I didn't count on.

As a result I've (fingers crossed) done an experimental board that has an expansion for other low-power circuits using the OPA2134 and similar Op Amps or even just passive components to do things like a "normal" volume control (in addition to the switchable gain) an even switchable low pass (cut the high frequencies) and high pass (cut the hum and similar noise).

It measures about 50mm x 75mm. You can see the expansion connector here with (I really, really hope!) optional large capacitors. Beyond the pinheaders, the boards are pretty much ready to go with a variety of capsule designs ranging from Matt's original and (or a carrier board that does the same) right through to the more complex, but more performant heads like Varee.

With luck and a following wind I should have this lot back in a week. The real jewel in the crown remains Matt's design but with the Varee "capsule adaptor" you could (at your option) do away with the whole Michelle/Mary carry on and go all out on something like a Focusrite Scarlett - or ANY other professional mixing desk with a 48 volt phantom power but it will work with anything from 15 volts up (a little less actually but that discussion will have to wait until I can confirm everything works as it's supposed to).

The plus side to using the Focusrite is you can still make that wonderful stand and just feed the cable in via a standard socket BUT have the capsule held in the DIYPerks "Art Deco" design which I'm totally in love with.

Using an unbiased capsule is also possible but these designs are not optimised for it. I have another. 

OK *(for anyone following along). I've been beavering away at this even though there's little interest until there's a build video and I can't really ask Matt to do a video when the thing isn't 100% complete and working.

So the capsule adaptor (Varee) has had a major overhaul.

As promised, performance now matches or exceeds that of some the best commercial microphones. It uses a circuit like the one described in the Rode NT1a datasheet. I haven't copied Rode's design because (a) that's naughty and (b) I don't have a schematic anyway.

It uses a clever circuit (one that someone else came up) with that reduces FET distortion from a few 10ths even at low input levels by at least an order of magnitude. Here's a snip of the section. The addition of the PNP transistor keeps the FET working at a "programmed" current thus linearising the gm (I think, I've really got to go re-read that bit). It works quite brilliantly though.

Medium power output transistors with lower noise than low-power ones. due to the size of the wafer. And it works with P48 power in current form, but I've modified Matt's pre-amp to drive it. More of this when I do the instructions.

Frequency response is largely constrained by the capsule as this adaptor works with the 25mm and 34mm standard capsule sizes. (25mm preferred).

This version (sadly) doesn't support a fully 48V setup with an unbiased capsule because of costs. The actual job of converting isn't difficult and when I have a chance I may run up some instructions. Be interesting to see if the Ali Express 34mm unbiased capsules are any good. (In theory [!] they should be better than an electret but there's more to it than that, which is why the "proper" ones cost £100+ in English money and that's for a very basic one.

The pre-amp, base unit, call it what you will is at V1.4 after some spectacular cockups on my part, mostly missing things like decimal points in the ordering system so I got a 1.2 V Zener when I was expecting ... a 12V one. Oh darn.

It's a bit of a kitchen sink but I won't brag about specifications in case I jinx it. And of course, it's eaten into my beer money so badly that I'm having to drink Aldi's budget lager instead of coffee.

And another quick update and the resolution of something that's bugged me for (ever) since I started this. The issue of the 1G resistor. The big beastie, the one you absolutely can't run this circuit without.

For future readers, I "argued" some time back that a gate load resistor was essential to correctly bias the JFET.

And yet time after painful time my efforts to increase the effective impedance at the gate using feedback to keep the source and gate at the same "level" - it's a good technique but it caused a disruptive amount of noise.

So I folded and finally did some experiments that "proved" you don't need a resistor in that gate circuit and of course it just works.

Until one day, it doesn't.

Imagine, if you will, having created something of beauty, that breaks all the design goals (in some places smashing the into next week) and yet all of a sudden, the prototypes (with their improved noise rejection, ya-de-da) that are the cherry on the top?

And then it doesn't work.

Until you pick it up.

And then it works.

For a moment and then the LED fades. Slowly... eventually pulsing a sort of electronic death rattle.

It's all fine, no magic smoke as this circuit only pulls 1.2 - 1.5 mA - which is typical of these things.

Power down, power back up. Nothing. Tap the desk - light glows brightly then over a few seconds, it fades away.

Different capsule? No change?

Rip the capacitors off (in case it's some weird oscillation)? No change.

And then it's back to one of my own mentors, Rod Elliot of ESP. Great guy who can be found: https://sound-au.com/index.html

Maybe Rod has the answer as he's done quite a bit of work with JFETs in this application (audio pre-amps)? And sure enough there it was.

TL;DR

As I'd suspected all along but was unable to show experimentally (until I had a really cracking design) was that the gate's impedance was developed from what amounts to another, well capacitor in a way.

JFETs "self bias" because there's a tiny leakage current (pico amperes) through the diode forming the drain-gate junction and this current needs somewhere to go. (Everything is part of capacitor at some level of observation but you still need enough leakage current for the gate to operate.

Any large resistor will work here but unless it exceeds the impedance of the capsule at all frequencies, the output will be affected.

Posted by: @marcdraco

↑

In the main thread I've posted the final (re) designs of the above. I literally had them uploaded to JLC PCB and about to go into production when I realise there was a fatal error in the design. (My fault for relying on memory rather than double-checking in SPICE).

Well, I say "fatal" it wasn't fatal but it was sloppy on my part, so I've spent the last couple of days cleaning up that mess and added some completely new designs to the quad, while simplifying Hannah a little. The one with the Op Amp allows for a fairly large swathe of off-the-peg dual amps to be used. They have to be FET input and low power (i.e. about 5 mA quiescent per amp, or 10 mA for the whole package) and capable of running at up to +/-15V, although that drops off sharply as the circuit demands more current.

The other two new designs use 2SK208s which are ostensibly cheaper versions of the LSk170s only they are available from my board shop and can be fitted by them which saves everyone the hassle of that.

I've got them on a rush order since I've been on this project for several months and people probably think I'll never do anything, but that's the thing with development like this, the turnaround time is measured in many weeks so I'm often sitting on my hands or tinkering while I wait to see what works and what doesn't. https://eldfall-chronicles.com/eldfall-chronicles-northern-wind/

Then it's back to the drawing board, calculator and SPICE simulator. 🙂

They've gotten dramatically better as I've been able to test out different designs but waiting for the board shop is tedious. https://3dtrcek.com/en/tpu-filament

Assuming Matt's OK with this, I'll open source the final designs (complete with silkscreen logos) so everyone can buy their own, pre-made at JLC or a board shop of their own choice. On paper at least, these are the most performant I've made. A lot depends on the +/-15V power which is something I won't claim to have much experience of. It's a bit like having P48 but with a lot less room to move. The only advantage (and bizarrely, I missed this one which was right under my nose) is that you have a negative supply which reduces component count since the main circuit can be DC coupled.

All the open source design stuff is here - there are no real explanations of how things work or how to put it all together as that's really with Matt so he can do the video. I can only offer very basic help on these for now.

I've done a smaller version of Michelle (which doesn't have the isolated section) but it's not quite ready yet.

https://github.com/marcdraco/Sarah

https://github.com/marcdraco/Varee2

https://github.com/marcdraco/Michelle2