@ashleigh Yeah, whoops, I meant through-hole components. I'll look into trying that, then.

Songbird Michelle is considerably better than Matt's original.

You can't make it on stripboard because loads of the parts aren't available in through hole.

THT is going out which is an utter insult to people who are trying to learn as it creates a ridiculous barrier to entry. I already had a background (and I had a lot to learn) but there are numerous examples of what appear to be simple circuits that absolutely cannot be reproduced on strip.

The board itself could (in theory) be assembled at home but you need a hot air work station and a lot of patience.

I wouldn't and I designed the thing.

@marcdraco Ahhhh...I'd better not try it, then. Thanks for the warning. In what ways is Michelle most significantly better than the original?

Mostly noise (S/N) which is definitively better, particularly at larger gains.

It's also USB-C (running at 12 MBPS) so everything is on the board. It doesn't have a monitor because I never added one (cost overruns and such).

And now my Apple Watch has conked. How am I going to tell the time now?

@imp-inventor @wecan (I think that's the right handle, darn me and my blithering) hopefully should haves some field experience with it.

There's no such thing as noiseless in electronics (that would require physics to break and that's not on anyone's bingo card).

Michelle has a LOT of features that are available but simply aren't fully realised in this design (they're easy enough to add though).

The INA - (the thing that does what the THAT does) is similarly specced to the THAT1512.

There is space (and it's wired) for stereo but only one channel is available from the CMI108B. The I2S signals can be vampire tapped OR vectored off to another device (like a Pi Pico, STM32, ESP32 etc.)

The digitiser is capable of 192K which is twice the speed of the current offering (x 2 channels folks, I haven't forgotten how to count).

I guess the MAJOR change from Matt's (it's pretty much a ground up re-design) is the use of a very low drive voltage for the JFET. I could (conceivably) bump that back to 30v but that's even more cost.

Songbird uses a pair of chained LDOs (low-noise low-voltage regulators) to smooth out the USB voltage at the mic head until you'd be forgiven this was battery powered.

The original suffers not from poor design from on the board, but the utterly awful design of the USB power systems. These cables impress a LOT of noise on both rails and it's rich in harmonics - which makes it a nightmare to get rid of.

I was rewatching the instructional video on YouTube and forgot to solder the mic capsule ground to the copper insulation around the enamel coated wires. I've now done that. Do I also need to make sure I connect the copper insulation to the shield on my audio cable?

Not entirely sure what you mean by this.

Think of it this way. The mic has three connections in all, two carry the audio and the other is a "common" which is shared by both wires.

The third wire (the screen) connects to the brass ring/mesh at the head AND to the ground/0V connection at the board.

Yes sorry, I wasn't very clear. Here's an image where I've circled the bit that I think I need to connect to the copper shield. My plan is to test at this point too before I go further. I want to temporarily put a TR plug on it to test I get a signal and the capsule works. I plan on plugging it into my focusrite amp (pic below). It has both line/instrument input and the option for 48v phantom power. Has anyone attempted something similar? I'm happy to go a different route if others have had success with simpler tests. It doesn't have to sound good, I just want to test I get a signal and can record my voice into my pc.

Thanks!

I did a PCB that soldered to the back of the capsule and converted it into a P48 compatible head but for most of us that's OTT.

What you have here though, that won't work (at best) and will blow your JFET (at worst).

If you just want something "Quick and Dirty" you'll need a small resistor (about 2,200 ohms or 2k2), a capacitor, 10 uF to 100 uF and a small battery clip. You'll need to put power into it you see; phantom is a bit of a queer animal and to use that we'd need to add a small regulator.

And then you're going to hit the OTHER problem with this capsule - it absolutely has to be screened. It simply won't work if you're anywhere near to an electrical supply because that thing would pick up Taylor Swift (if Travis didn't get there first, and even then I'm not sure)! 😉

Songbird: Better Late Than ... 🤡 

Seems I didn't post the files (months ago) so here's the current project, including all my rude remarks to myself because I forget stuff. The KiCAD schematics needs a bit of a clean and there are some oddments in there that I reall should have thrown away but it's hot. That's my excuse, excuse me while I down a few cold ones.

Design Philosophy

Michelle was conceived as a practical, high-performance microphone front end that could be professionally manufactured without requiring firmware development or specialist programming tools. Throughout the design, emphasis was placed on measured analogue performance, straightforward manufacture and long-term maintainability rather than implementing every possible feature.

Every significant design decision represents a balance between performance, complexity, availability and cost. Rather than pursuing the highest specification in every area, the objective was to produce a reliable, well-understood platform that performs consistently and leaves scope for future development.

Design Decisions

Four-Layer Construction

Michelle is implemented on a four-layer PCB using controlled-impedance routing where appropriate.

A six-layer stack-up would have offered additional routing freedom, improved power-plane partitioning and greater flexibility for future expansion. However, the additional manufacturing cost was difficult to justify for the intended audience.

The four-layer implementation provides an effective compromise between analogue performance, signal integrity and manufacturing cost while remaining suitable for small-volume production.

Analogue Front End

The analogue performance of Michelle is centred around the Texas Instruments INA849 instrumentation amplifier.

The INA849 was selected because it offers exceptionally low input-referred noise, excellent linearity and high common-mode rejection, making it well suited to low-level microphone signals. Although significantly more expensive than conventional audio operational amplifiers, its performance justified the additional cost for the primary signal path.

Rather than attempting to compensate for a noisier front end elsewhere in the signal chain, the design philosophy was to preserve signal quality from the point where the microphone produces only a few millivolts. Once noise has been introduced at this stage it cannot be removed later. For this reason the JFET operates in a modified Common Source mode vs. Matt's original.

For cost reasons, the default build populates a single INA849, providing a high-performance mono input suitable for the majority of voice recording applications. The PCB layout, however, retains provision for a second INA849, allowing a stereo variant to be constructed if required.

The analogue circuitry surrounding the INA849 was developed through practical measurement and refinement. In several areas, the final implementation differs from more conventional reference designs because testing demonstrated measurable improvements in overall signal-to-noise performance.

The limited supply voltage rails (+/- 5 Volt) restricted choice further, and left the THAT151x series out of the picture. Not that these are NOT drop in replacements as they follow a non-standard layout. Gain equations can be found in the datasheet but it's currently "fixed" at 100 R  giving an overall gain of around 500 (54 dB). It's a slightly arbitrary value as I've found it's better to adjust the volume in the digital sections as an external resistor can pick up external interference.

USB Interface

The CM108B was selected as the USB interface for several practical reasons.

It is inexpensive, readily available and requires no firmware development, programming hardware or software toolchain. Once assembled, the board simply enumerates as a standard USB Audio Class device.

Although more capable microcontroller-based solutions exist, they inevitably increase both hardware and software complexity. Michelle was intentionally designed so that builders interested primarily in analogue electronics could assemble and use the board without first becoming embedded firmware developers.

The CM108B is also a mature device that is expected to remain available for the foreseeable future, reducing the likelihood of future redesigns caused by component obsolescence.

Clock Architecture

The PCM1809 ADC is capable of operating at significantly higher sample rates than are available in the default implementation.

In Michelle, however, the I²S clocks are generated by the CM108B. Consequently, the effective sample rate is determined by the USB interface rather than the converter itself.

For the intended applications of speech, streaming and general-purpose microphone recording this presents no practical disadvantage, whilst allowing the design to remain simple and robust.

Should a future revision require crystal-locked operation or higher sample rates, the modular architecture allows the digital interface to be replaced without redesigning the analogue circuitry.

Analogue Power Architecture

Although the CM108B provides internal regulated outputs, these are intentionally not used to power the analogue front end.

Instead, Michelle generates the microphone input bias using dedicated external low-noise regulators. A standalone 3.3 V LDO supplies a second 1.8 V regulator, producing a clean and stable bias voltage for the input JFET.

This approach deliberately separates the analogue input stage from the USB interface and its associated digital switching activity. Whilst it increases component count slightly, it significantly reduces the opportunity for USB-related noise to couple into the microphone signal path.

At the highest gain settings, a small amount of residual USB-related noise may still be measurable. Under normal operating conditions this remains well below the microphone signal and has no practical effect on recording quality.

Earlier prototypes investigated a fully differential input arrangement. Whilst this offered marginally improved common-mode rejection, practical testing demonstrated that the revised bias arrangement produced a better overall signal-to-noise ratio for the intended application. The final design therefore reflects measured performance rather than theoretical optimisation.

Analogue Supply Rails

The INA849 instrumentation amplifier is operated from dedicated bipolar analogue supply rails generated by the Texas Instruments LM27762.

Although Michelle is powered solely from the USB 5 V supply, the analogue front end was designed to avoid the compromises associated with single-supply operation or virtual-earth techniques. The LM27762 generates the positive and negative analogue supply rails required by the INA849, providing improved signal headroom and allowing the amplifier to operate around true ground.

These rails are used exclusively by the analogue amplification stages and remain isolated from the digital power domain wherever practical.

This arrangement complements the independent JFET bias network, which is generated separately using dedicated low-noise LDO regulators. The two systems perform different functions: the LDO chain establishes a clean, stable bias for the microphone input stage, whilst the LM27762 provides the bipolar operating supplies required by the instrumentation amplifier.

Separating these functions simplifies the analogue design and minimises the opportunity for digital switching noise from the USB interface to influence the microphone signal path.

Modularity

Michelle was intentionally partitioned into independent functional blocks comprising:

• Analogue microphone front end.
• PCM1809 analogue-to-digital converter.
• CM108B USB interface.

The interfaces between these sections remain accessible wherever practical.

The I²S bus is exposed through drilled test points designed to accept a modified 2.54 mm pitch header. By removing the associated link, the digital audio stream may be intercepted for development or experimental purposes.

The Master Clock is also available as a monitoring point. As this clock is generated by the CM108B, it is provided as a tap only and is not intended as an external clock input.

These facilities exist primarily to simplify development, measurement and experimentation. They are not intended as permanent expansion interfaces.

Future Development

The modular nature of the PCB makes it suitable as the basis for derivative projects.

Potential applications include:

• Stereo variants by populating the optional analogue channel.
• Alternative USB interfaces based on programmable microcontrollers.
• I²S-based audio processing platforms.
• Analogue-only microphone preamplifiers.
• Experimental digital audio interfaces.

These configurations fall outside the supported build but were considered during the original layout to ensure the PCB remained flexible without compromising the primary design.

Development Approach

Michelle was developed through practical measurement and iterative refinement rather than strict adherence to reference designs.

Where alternative implementations were investigated, the version adopted was selected on the basis of measured performance, manufacturability and overall usability rather than theoretical specifications alone.

The intention was not to produce the most complex microphone interface possible, but to produce one that performs reliably, is straightforward to manufacture professionally, and provides a solid platform for anyone wishing to understand, modify or extend the design.

The full KiCAD "dump" as a ZIP.

And the files you need to order.

Board "stackup" is JLC04161H7268 - 4 layer, impedance controlled. That's a no-cost extra but it's NOT an option. The USB needs an impedance control or it won't work with this design.

A Starting Point, Not the Destination

Michelle should not be viewed as the final word in microphone interface design. Instead, it is intended as a practical, well-engineered starting point for anyone wishing to explore high-quality analogue microphone electronics.

In many respects, Michelle is the spiritual successor to the original design, incorporating the experience gained during its development whilst taking a completely different technical approach. Although the architecture and component selection have changed substantially, the original objective remains the same: producing the cleanest practical microphone front end from readily available components.

Builders should not feel obliged to source expensive specialist capsules. Michelle was designed to perform exceptionally well with readily available electret microphone capsules, making the project accessible without compromising performance.

Although JLI do not currently manufacture a cardioid electret capsule suitable for this application, several other manufacturers offer high-quality cardioid alternatives that work well with the circuit. The microphone head geometry developed for the original project remains an excellent mechanical design and is well worth retaining (not to mention it just looks beautiful!)

Hopefully Michelle serves not only as a finished project, but also as a foundation from which others can experiment, learn and develop designs of their own.

@marcdraco FYI at around 1m 50s in this video he shows something similar to what I’m hoping to do. Just a quick and dirty test, audio quality isn’t a concern. Thanks

https://m.youtube.com/watch?v=_j5UO-791_4&t=124s&ra=m

Agh... I just posted this without checking and saw it as it was uploading. Draw me a simple circuit ChatGPT. Make my life a bit neater by making it look nice and clean and OOOOO NOOOOOO!

That doesn't work but this might (just a point to be careful with ChatGPT!) 

This is what I'm thinking of - it's the same idea but uses the P48 power. It's pretty hard to break P48 because the currents are severely limited. And I should say this is a hypothetical hack! 🙂

So I'm ordering the aforementioned circuit board, Songbird Michelle 3.5 from JLCPCB. This is my first time ordering something from them--it's my first time ordering any printed circuit board--so I'm quite out of my depth here. I ran into some things that brought up questions, and in order not to waste money, I thought I'd ask about them here. This question is mainly directed at @marcdraco but any help is welcome here. Here are the steps I went through (I also have a full screen recording of everything I did in case anyone wants it):

1. Uploaded .zip file.

2. Changed stackup to one listed here.

3. Changed "Impedance Control" from "No Requirement" to "+-10% (+-5 OHMs if value is less than or equal to 50 OHMs)", which was the only other option. This immediately jacked the price up about $33.

4. I'm not sure if the "Confirm Production File" setting needs to be selected as yes, especially as it adds an extra dollar to the price, but it was set to yes as default, so I left it.

5. I set PCB Assembly to "Assembly cost starting from $0 with coupon".

6. I changed the PCBA Quantity to 2 instead of 5, then clicked NEXT.

7. It took me to an image preview of the circuit board, and I clicked NEXT again.

8. It took me to a page with a "Add BOM file" upload field and a "Add CPL file" upload field. I stuck the bom.csv file in the BOM field, and the positions.csv file in the CPL field.

9. Weirdly it gave me an error when I clicked "Process BOM & CPL". It said "The below parts won't be assembled due to data missing. TP1, TP3, J1, TP2, TP4, J401, TP7, TP6, TP8, J405, J402 designators don't exist in the BOM file". I think this is not supposed to happen...? I clicked continue anyway. 

10. On the next page, at the very bottom of the BOM list, in a row labeled "R410,R420" it says 4 shortfall. Apparently there aren't any of these parts left in stock. I clicked next to see what would happen here, but I was pretty sure at this point something's gone wrong. 

11. I ended at the Quote & Order page, where it listed Charge Details: "PCB Price: $40.88", "Economic PCBA Price: $63.58", "TOTAL PRICE: $104.46". 

My first question is, was the Impedance Control setting in Step 3 the correct one? Second question, is the error in step 9 expected? Third question, should I bother waiting for them to restock in Step 10, or is it possible to choose a different part? Last question, why the bloody hell is it so expensive? I thought the price of each board was ~$30, and I would have to pay ~$60 for two of them. Is this assumption incorrect, or did I do something wrong during the ordering process? Does it have something to do with the number of Extended components?

Again, thanks for helping a beginner 😀 😓  

Hi @imp-inventor, I probably need to amend the instructions further up the thread to include the TP stuff. TP just means test-point and JP is jumper. They're not fitted by JLC due to cost considerations.

The main "gotcha" by the look of it was the impedance control

Posted by: @marcdraco

↑

Board "stackup" is JLC04161H7268 - 4 layer, impedance controlled. That's a no-cost extra but it's NOT an option. The USB needs an impedance control or it won't work with this design.

You can't just pick a random stack (and some of them are mighty expensive) - and JLC is weirdly complicated.

This is new since I designed the board. It just means that JLC will do the work I already did for you. But since I did it, you don't need to. 😉

Without going into a huge song and dance, at USB frequencies, we're dealing with radio frequency design and that means the designer has to adjust a the width and (usually spacing) of the USB data pair from the point where it enters at the plug to where it is terminated at the IC. It's quite the mathematical dance to achieve something we all take for granted.

TL;DR: Leave ths option alone. Select the correct board as shown and you're good.

The "shortfall" in the BOM is a it's a useful one to look at how to fix it (because this WILL happen from time to time).

Each part has a bunch of specs like "perfect" electrical value and then a bunch of assumptions where it will do what it says on the tin,

For a board like Songbird you can assume that everything is operating at 6.3 volts or better. I put as much headroom in as necessary without being "weird" and some resistors go as high as 75 V but that isn't an issue. 

What you have to watch out for with passives here is the size. The two pin everyday parts come in 0402, 0603, 0805, 1206 (and a lot of other sizes, but those are the "big four" currently).

Now this is where things get ... quirky. Not all resistors are created equally.

Two major types here are thick film and thin film.

• Thick film = everyday workhorse. Cheap.
• Thin film.  = specialist low noise work. Costly (sometimes exceptionally so).

This is where one choice you can make as the assembler that I can't make for you without forcing your hand, so to speak. These four resistors (2 x 2k2 and 2 x 100 R) are the input pair and this allows you to over-ride "my" values with your own. 

Most of you will want the BASIC type (what matters is the size 0603) and the value (2k2 and 100 ohms).

This is different from the setup Matt used and allows you to decide how you want the input stage to function (balanced or Common Source) and (this is another bit I left out) JFET operating current.

JFETs are odd fish - and the operate better (or worse) at specific currents and voltages - and you HAVE to trim them to do that. The 2N4416 works "better" at higher voltages than we have available. So these values can be adjusted at manufacture to match the JFET you have (if known) or the "impedance" if you only have the schematic from the mic supplier.

The other thing you can do here is replace the thick film "BASIC" set with something from the more exotic from the thin film "Extended" set. I don't have to run the numbers to guess how much difference thin film will make with a 10-12 dB boost provided at that stage. This is, as you might guess an advanced feature but as "Michelle" is designed for experimentation OR as a stand alone board for experimentation is with you.

When I did my own I just used the standard "thick film" type and the noise is all-but imperceptible - even in a no-signal condition with the gain shoved way up.

@wecan has one of the beta boards if he's still with us (assuming I didn't blow up his machine!) JK. 😉   

@marcdraco I did think the impedance thing was weird...thanks for setting that straight. However, I've also found some other weird things that I thought I'd ask about before clicking the final button, and I got some screenshots:

For this first one, while uploading the design and changing the settings, I noticed the stackup isn't quite the one you listed. The one I selected--JLC04161H7628--is the closest I could get to the one listed in your post, but I only realized it actually wasn't the same just now. Did you make a typo or is something else happening?

This second one I'm just asking about because I saw it in a video--is the highlighted setting necessary? It sounds important when you look at the description of the setting but you didn't mention it, and I assume you haven't made any polarity mistakes in the circuitry.

You mentioned the JFET. I have the 2N4416, and seeing as it would probably be easier to change the JFET than the preamp circuitry, do you have any suggestions for a replacement JFET?

Lastly, I suppose the missing items are back in stock now 🤣 Everything seemed to be in order with nothing missing when I went through the upload process today.