Skip to content
New experimental US...
 
Notifications
Clear all

New experimental USB C Mic design

5 Posts
1 Users
0 Reactions
413 Views
marcdraco
(@marcdraco)
Posts: 476
Member
Topic starter
 

I've just sent this to JLC for production (fingers crossed it works!) an experimental version of Matt's ultimate USB-C mic project [yes, another one!] so it's not really for the main thread elsewhere. Noise levels at this stage are impossible to accurately estimate but given it's 100% single-supply, 5V low-power, I don't anticipate it being amazing.

That's not to say amazing isn't possible, but budget and time constraints raise the old phrase, "impossible we do at once, miracles take a little longer". 

Without going into a long screed on how this all works, it's essentially an Op Amp (LM324, cheap, *potentially* a little noisy) configured as:

  • Differential input stage.
  • 5V power.
  • Mic Bias.
  • Baxendall volume control (either switched resistors or a linear pot).
  • El-cheapo headphone amp, TDA 2822.

If you know your parts you'll see this is far beneath (think BMW vs. rusty old Ford) what Matt's original used but there's a method to this madness, mostly in cost - about £10 each assembled in small quantities.

But as I learned the hard way, what works in the sim, don't always work on a PCB.

image


Take everything I say with a pinch of salt, I might be wrong!

 
Posted : 12/02/2024 11:54 am
marcdraco
(@marcdraco)
Posts: 476
Member
Topic starter
 

For anyone following along, I've hit a bump at JLC PCB but one of their wonderful engineers spotted a mistake (I blame the cat) on the board I didn't tell you about because it wasn't quite finished. I won't list the specifications at this stage, but this is now in an immutable condition and in production so you know that I haven't forgotten everyone.

My aim was (and remains) to do sufficient justice to Matt's creation by improving it in the requested areas and input from the beta testers (yes, I really have had a couple of beta testers).

And here it is - fits 25 or 34 millimetre capsules and powered from P48 phantom power but should work with Matt's board (with a small modification) as it's designed to work all the way down to 5V thanks to the shunt regulator.

For compatibility with "professional" equipment, I've returned to an "phantom" power design similar to what Matt used and based on what professional systems have used for multiple decades (this format was invented in 1966). 

This should (remember, it's a "release candidate") make it possible for anyone to convert those cheap Chinese condenser mics to something really special.

More details to follow, but here's a rendering from KiCAD but the whole design is attached - not this is a KiCAD 8.0 file and won't work with earlier versions.

 

image


Take everything I say with a pinch of salt, I might be wrong!

 
Posted : 16/02/2024 7:09 pm
marcdraco
(@marcdraco)
Posts: 476
Member
Topic starter
 

One of the wonderful engineers at JLCPCB spotted a cock-up (two actually) on my order and since I just assumed everything was OK, and didn't check my email (cough, cough) it held things up.

Which meant I manage to get this one on the same order. A big reveal later IF it works as advertised and the run I'm having of late, I've got everything including my legs and eyes crossed and touching a lot of wood.

While this looks stereo at first glace (this is just a 3D impression of it) it's actually a complete hi-fi quality microphone pre-amp with 48V phantom power, isolated microphone input and powerful headphone amp that *should* drive pretty much any headphone you throw at it. 

The jewel in this design is an updated version of the Jenson 990C discrete op amp: Simulation of the JE-990 OP Amp By Deane Jensen | diyAudio

The "C" version is the most update I could find, but for this "proof of concept design" I've replaced the [discontinued] LM394 with discrete transistors. 

The LM324 contained 100 transistors in its day, "randomly" connected to reduce noise and was an incredible performer in its day. The modern equivalent might be Analog's MAT12 MAT12 Datasheet and Product Info | Analog Devices

But as you can imagine, these things aren't cheap and are often not available as through-hole parts - the MAT12 (thankfully) is an exception and comes in a TO78 metal case.

Adding the MAT12 to this design isn't really practical as it's SMD and the test version, but the through-hole version will support it. I'm holding back on that until I can verify each section of this design and PCB works, I won't be ordering those. 

For techies, it uses MMBT3904 and MMBT3906 biased at 100uA which is the sweet spot for best performance of these low-noise transistors. Again, there are better ones but for choice, I'm doing the through-hole version with sockets or positions for "better parts, although this is probably over-specified. 

image

Here's a 3D rendering of the current state of the through hole version. I've used SMD components for non-critical parts and through-hole for all the semiconductors - this will support the MAT12 (matched pair for audio) and a more traditional NPN matched pair that you could select by measuring hFE from a box of parts and selecting two that are close.

The SMD version currently in production doesn't use matched parts so it'll be interesting to see what the CMMR is.  

This post was modified 4 months ago by marcdraco


Take everything I say with a pinch of salt, I might be wrong!

 
Posted : 20/02/2024 11:48 am
marcdraco
(@marcdraco)
Posts: 476
Member
Topic starter
 

Just a quickie for anyone wondering what the state is here. Well, behind the scenes I've developed several related designs (all at JLCPCB being manufactured right now). And for the love of Odin, I hope these work without issue (they are quite mature but even a small change holds the possibility of a "fatal" short)!

Everything here is provisional: subject to validation testing.

Without getting into the weeds (and I'll publish everything once I'm happy that I'm not leading anyone up the garden path) just yet, I've "forked" Matt's original design into two versions aimed at slightly different audiences.

The "ARTF" board, Michelle works at 5V from USB 2 or 3 but doesn't need any external power. The downside of that one is, that unless you trawl JLCPCB's extended catalogue for more modern devices, you're stuck with the LM324 quad amps which are not noted for their low-noise. It's not the show stopping WHOOSH from the incredibly nasty Chinese USB mics, but nowhere close to Matt's USB-C design (I never expect it would be). 

I've replaced the cheap SMD op amps with 8-pin DIL sockets allowing us all the choice of what to stick in there. It has a TDA2822 headphone amp which is surprisingly loud (to the point of pain!) even operating at 5V! I've slipped another little extra into that one for people who like their "blinkenlites". But all will be revealed when I've got it rigged up and verified, otherwise I'll be crying into my water for a month because all my beer money has gone. 😉 

Now this voltage stuff matters because we're working at very, very low voltage (compared with the professional 48V systems) and low-voltage op amps are more expensive if you need quality and/or low-noise. As it's split over three dual packages it should be possible to experiment starting with something exotic at the input and using jellybeans further down the signal path - specifically the Baxendall volume control which is easier, cheaper and more predictable to operate than changing the gain on an INA. Baxendall's design is simple and remarkably effective even with dual *linear* potentiometers.

I've left the option to change the INA gain on the board, however, for more advanced/experienced hobbyists.

The second one is more professional, also using DIP sockets for all the important components in the signal path which means you can start with the NE5532 or similar dual op amp. It will work with jellybean FETs like the TL072s but the noise levels will be considerably higher. A TI OPA series with Burr-Brown tech should give much better performance.

A class AB output stage ensures a lack of crossover distortion at the expense of a slightly higher power consumption. Crossover distortion is (according to those who know) thought to be the "transistor" sound so many audiophiles detest.

Negative feedback, rather incredibly, can remove a lot of that distortion - indeed many lower cost, but well respected op amps have class B stages that exhibit exactly this effect but it's rarely noticeable due to negative feedback. Amps like the 5532 (despite its advanced age) remain popular in high-end gear because  of the predictable feature an AB output stage, and with mains power we're less constrained. This has the side effect of drawing considerably more quiescent current than would be needed in a discrete stage because the designer had to allow for it to work every time.

The output stage is designed to push two amps through the headphones which is enough to drive all but the toughest headsets, so caution will likely be the order of the day.

There's also a secret one I dropped in there after getting eviscerated by an old hand at a specialist site for daring to use radio-frequency SMD components in something he admired. 🙂

This thing, which I've developed in two variations on the same board, is supposed to - again, emphasis on supposed to, knock the 5532 into a cocked hat in every metric. My version should level up quite well at the expense using considerably more real estate in your enclosure - it's slightly less than 40mm square.

The circuit originally formed the front end of an early incarnation of the "professional" board above until I ran out of space (I needed three of these on the input section alone)! The original, although best known for its use in high-end professional audio, that's high end by professional standards, had a very wide frequency response but I've limited that to something defined for audio without compromising much else.

The option to add a transistor (MAT12) that's so expensive I daren't even buy ONE yet to play with is the real game changer, again, so long as I got my math right. As is, at around a £10-£15 per board in small runs, it's not exactly "budget" unless you're Elon Musk. But it's all relative.

I believe in giving everyone the right to compromise where they think they can. For example, of the three boards required for the full complement, one could feature at MAT12 and the others just use their on-board MMBT3904 input stage.

And yes, I know about the different collector currents for best noise performance so that's adjustable by a single fixed, through-hole resistor. This is necessary because the 2N/MMBT390x series work best at a fiddling collector current of 100 *micro* amps whereas the MAT12 operates at around 10x that.

While not enormous they are rather more cumbersome than a classic DIP. Even though there is a standard of sorts for pins on the bottom of these fellas, I've kept it simpler with the option to wire these with interconnect wires - so you won't have to get a separate PCB made up just to play with them. I hope that Matt will be able to make something absolutely gorgeous with them, with a see-through case otherwise they should find a use in early higher-education physics or electronics courses as a more convenient way of tinkering with op-amps.

Output drive, like the 5532 is a pre-biased AB stage using 390x series transistors and even without the MAT12, I expect the performance should be impressive simply because the transistors are much larger (being discrete) and a larger die means less noise. Naturally, a specially designed, modern, matched pair like the MAT12 is going to make that look like a clapped out old motor, but the cost of MAT12s is prohibitive for me at any rate. I've set this one up for a through-hole MAT12 rather than DIP, although changing the board to fit alternatives, including a SOIC-8 isn't that difficult.

But I'm getting ahead of myself here. Keep everything crossed that my old brain didn't smeg up again or I'll be off to silicone heaven with all the pocket calculators.*

So there you have it - everything is moving, still at a glacial pace, but moving in the right direction. If anyone has questions, please ask them here. I will likely have some boards to sell at cost when this is done. But most people will probably find it more cost effective to buy a set and sell the ones they don't want on eBay at a profit.

This is not endorsed by DIY Perks of course, not until Matt says so!

 


*This is a joke from Red Dwarf, the cult sci-fi series that dominated some of my middle-age telly-watching.


Take everything I say with a pinch of salt, I might be wrong!

 
Posted : 01/03/2024 12:03 pm
marcdraco
(@marcdraco)
Posts: 476
Member
Topic starter
 

Just so you all know I'm still on this, I just received a bunch of new prototypes. Some of the experiments worked, others no so much so. More news as soon as I've analysed the goofs, slipped silkscreens, parts so close they will be breeding... and so on. I'm awaiting parts to complete a couple or complete testing of the others.

EDIT #1: First results are in and ... I goofed switching from LM324 (quads) to the OPA2134 (duals) and hooked one of the inputs to ground - when it should have gone to virtual ground. *sigh*. (This is why I miss having a simple simulator in KiCAD - second time a "quickie" swap out wiped out a whole set of boards. More test results follow. 🙂

This post was modified 3 months ago by marcdraco


Take everything I say with a pinch of salt, I might be wrong!

 
Posted : 09/03/2024 1:25 pm